142 article(s) will be saved.
Record: 1- A New Livestream Retail Analytics Framework to Assess the Sales Impact of Emotional Displays. By: Bharadwaj, Neeraj; Ballings, Michel; Naik, Prasad A.; Moore, Miller; Arat, Mustafa Murat. Journal of Marketing. Jan2022, Vol. 86 Issue 1, p27-47. 21p. 2 Diagrams, 5 Charts, 7 Graphs. DOI: 10.1177/00222429211013042.
- Database:
- Business Source Complete
A New Livestream Retail Analytics Framework to Assess the Sales Impact of Emotional Displays
At the intersection of technology and marketing, this study develops a framework to unobtrusively detect salespeople's faces and simultaneously extract six emotions: happiness, sadness, surprise, anger, fear, and disgust. The authors analyze 99,451 sales pitches on a livestream retailing platform and match them with actual sales transactions. Results reveal that each emotional display, including happiness, uniformly exhibits a negative U-shaped effect on sales over time. The maximum sales resistance appears in the middle rather than at the beginning or end of sales pitches. Taken together, the results show that in one-to-many screen-mediated communications, salespeople should sell with a straight face. In addition, the authors derive closed-form formulae for the optimal allocation of the presence of a face and emotional displays over the presentation span. In contrast to the U-shaped effects, the optimal face presence wanes at the start, gradually builds to a crescendo, and eventually ebbs. Finally, the study shows how to objectively rank salespeople and circumvent biases in performance appraisals, thereby making novel contributions to people analytics. This research integrates new types of data and methods, key theoretical insights, and important managerial implications to inform the expanding opportunity that livestream e-commerce presents to marketers to create, communicate, deliver, and capture value.
Keywords: deep learning; emotions; face detection; livestream e-commerce; salesperson effectiveness
Livestream retailing augments traditional go-to-market strategies by reaching consumers via screen-mediated sales presentations for a variety of products. Amazon Live, Facebook Live, Taobao Live, and QVC serve as prominent exemplars. This type of retailing blends technology and marketing: the technology integrates video stream broadcast platforms, electronic payment systems, and forward and reverse logistics for efficient delivery and hassle-free returns, and the marketing combines entertainment and retailing, enhances reach via influencer marketing, shortens the purchase journey to the duration of the sales presentation, and permits value capture via innovative payment plans. In a typical video sales pitch, a host (salesperson) nudges a prospective customer through the purchase funnel by building awareness of an item's features, benefits, price, and discounts, as well as instilling urgency to buy. For example, on Taobao Live, which reaches over 37 million Chinese viewers monthly, top influencer Wei Ya helped Procter & Gamble accelerate its customers' journey from awareness to purchase and Tesla generate customer leads ([19]).
The continuous stream of sales presentation videos can be captured using advanced computing capabilities ([40]; [59]). Using video footage, marketing scientists can apply artificial intelligence technologies to automatically detect a salesperson's face in each frame, extract emotional expressions, and relate them to actual customers' behavioral data —all unobtrusively at a large scale— to generate novel insights ([34]), thereby augmenting the sparse knowledge on the business impact of emotional displays.
Recent studies have investigated the effects of emotional displays on various marketing metrics. For example, [50] offer the first study to automatically extract the emotions of joy and surprise that viewers experience when watching television commercials and then relate these emotions to attention and ad avoidance behavior. Similarly, [32] examine the impact of emotional displays on sales using the Facial Action Coding System ([13]) to categorize a set of emotions (e.g., happiness, surprise, disgust) based on facial expressions of viewers watching movie trailers, compute the average watching intention for each movie trailer, and relate it to box-office revenues. They find that viewers' emotional displays of happiness positively influence both watching intentions and box-office revenues. However, although [32] incorporate prospective consumers' emotional displays into their study, the role of emotional displays on the seller side of the exchange dyad remains an unexplored cue that shapes consumers' purchases. Indeed, in their survey of the literature on emotions, [ 3], p. 184) emphasize that "much of what we do know is confined to consumer behavior, as opposed to the behavior of salespeople or marketing managers."
To our knowledge, the current study is the first to assess the sales impact of product, price, sales force, advertising, and promotion in the presence of a salesperson's face and emotional displays. Specifically, we address the following knowledge gaps:
- Do salespeople's emotions in livestream presentations impact sales? If so, to what extent?
- How do the effects of emotions vary over an item's presentation span?
- What is the optimal allocation of face presence and emotions over the presentation span?
To answer these and related questions, we collaborate with a livestream retailer that broadcasts television shows 24 hours each day of the week, deploys salespeople to deliver live sales presentations of hedonic products, receives payments by credit cards, and ships orders by mail. A typical show lasts one hour and presents about eight items. We analyze the video data consisting of 62.32 million frames over two years. To put this scale in perspective, this footage exceeds two million 30-second TV ads. Then, we apply two machine-learning algorithms: real-time face detection ([56]) and real-time emotion classification ([ 1]). Specifically, the face detection algorithm discovers the presence or absence of a face in every frame of the video, while the emotions classifier (based on a convolutional neural network with rectified linear unit activation) assigns probabilities to the facial expressions in each frame with a face. Thus, we unobtrusively extract the display of emotions of each salesperson in one-to-many screen-mediated marketing communications in consumer markets.
Next, across 99,451 sales pitches, we match the salespeople's six emotional displays —happiness, sadness, surprise, anger, fear, and disgust— to how long each item was shown, the product category to which it belongs, the number of units sold, the price charged, and whether shipping fees were waived. Finally, we extend marketing mix models on two frontiers: the inclusion of emotional displays and salespeople's effectiveness. We emphasize that the literature on marketing mix models is vast, as is the literature on emotions; however, they do not overlap. This study bridges the two distinct domains.
Our analysis of large-scale video data shows that salespeople's emotions negatively impact sales across all six emotions, including happiness. The magnitude of sales decline across all the emotions is.47%, which is more than double the free-shipping effect (.20%). Happiness constitutes more than one-third of the total sales decline. Thus, we uncover a new maxim: sell with a straight face (i.e., reduce facial expressions).
Furthermore, the level of the optimal face presence reduces over the initial 10% span, then gradually increases as the presentation progresses, and subsequently tapers down in the last 15% span. Finally, most marketing mix models ignore the role of the sales force (e.g., [ 2]; [39]), and when they do include it (e.g., [16]), the sales force variable is operationalized at an aggregate level (e.g., number of salespeople, number of calls). Consequently, companies are limited in their insights into an individual salesperson's effectiveness. By contrast, the proposed framework uses person-specific dummy variables to estimate individual salesperson's impact, yielding valuable information on salespeople's performance, which circumvents managers' cognitive biases (e.g., homophily) in recognizing excellence and identifying candidates for retraining, thus contributing to people analytics in a sales setting. Next, we describe the conceptual background needed to interpret the empirical results.
Facial expressions are social displays that a sender strategically deploys to elicit a desired response from a receiver ([11]; [15]). In other words, facial expressions are "declarations that signify our trajectory in a given social interaction, that is ... what we would like the other to do" ([15], p. 130). They are social communicative moves that serve as "tools for social influence" ([11], p. 393).
A sender may mask true intentions. The onus is, therefore, on the receiver to decipher the sender's intent. The Emotions as Social Information (EASI) model ([51], [52]; [53]) asserts that buyers scrutinize the seller's expressions in commercial exchanges in an attempt to discern the seller's strategic intentions. For example, [57] show that sellers sporting a broad smile during an encounter are perceived as less competent and that perceived incompetence is more likely to be evident among prevention-focused customers and in high-risk consumption settings. Furthermore, their field study in a crowdfunding context reveals that a project creator with a broad smile is perceived as less competent, which reduces the total amount of money pledged for a project, the total number of large-scale donations made by backers, and the average amount of money pledged per backer. Similarly, [10] find that displays of intense happiness (e.g., a broad smile) by customer-facing employees can undermine trust and reduce satisfaction with the product.
[44], [45], [46]) provides a theoretical explanation for why receivers are likely to draw certain inferences with respect to a sender's facial expression, and the EASI model offers clues on how receivers are likely to react to the cue. Salespeople's expressions elicit customers' inferences about sellers' characteristics such as competence, trustworthiness, and persuasive intent. Such inferences, in turn, impact customers' purchasing behaviors. Drawing on extant theory, Table 1 presents the seller's facial expressions, intent, consumers' inferences about sellers, and consumers' action tendency with examples. According to [53], consumers' action tendencies are to ( 1) move toward (i.e., consumers experience a positive emotional reaction toward the influence attempt and thereby seek to cooperate with the seller); ( 2) move away (i.e., consumers experience a slightly negative emotional reaction toward the influence attempt and thereby seek to temporarily ignore or avoid the sender); or ( 3) move against (i.e., consumers experience a highly negative emotional reaction toward the influence attempt and thereby seek to terminate the interaction). Thus, Table 1 explains how salespeople's emotional displays trigger buyers' inferences in one-to-many broadcast communications.
Graph
Table 1. The Implications of a Sender's Strategic Social Communicative Moves.
| Seller's Facial Expressiona,b | Seller's Intenta | Consumers' Inference about Seller | Consumer Action Tendency | Illustration of Consumer Action Tendency |
|---|
| Happiness (smile) | Influence consumer to affiliate | A seller's happiness may be taken as a sign that the seller is gaining in the negotiation at the target's expense (Van Kleef et al. 2010). | Move against | An entrepreneur displaying a broad (slight) smile in a profile photo on a website is likely to receive less (more) financial backing for a crowdfunded project (Wang et al. 2017). |
| Sadness (pouting) | Influence consumer to provide support | A seller's sadness may be taken as a sign that the seller is recruiting succor (Scarantino 2017a) and trying to get consumers to lower their guard. | Move away or against | A service provider displaying intense sadness during a cell phone purchase is likely to result in the customer registering lower satisfaction with the product received and service (Cheshin, Amin, and Van Kleef 2018). |
| Surprise (startled) | Influence consumer to engage | A seller's surprise may be taken as a sign that the seller is trying to garner attention/liking in an attempt to make them more amenable to persuasion. | Move toward or away | A host on livestream video displaying surprise to customers is likely to garner greater interest in an offering. Alternatively, the surprise may serve as an unwelcome distraction and cause the customer to lose interest in the item. |
| Anger (scowling) | Influence consumer to submit | A seller's anger may be taken as a sign that the seller is losing in the negotiation and engaging in aggressive action to reassert dominance (Fridlund 1994). | Move toward or away | A manager displaying anger in response to employees' competence-based violations diminishes perceptions of leader effectiveness (Wang et al. 2018). |
| Fear (gasping) | Influence consumer to help and protect | A seller's fear may be taken as a sign that the agent is recruiting empathy and trying to get consumers to lower their guard. | Move away or against | A fear appeal in a television ad (e.g., Nationwide Insurance's "Make Safe Happen" 2015 Super Bowl Ad featuring a young boy who is no longer alive) led to negative social media posts and reduced liking of the ad (Bharadwaj, Ballings, and Naik 2020). |
| Disgust (nose wrinkling) | Influence consumer to reject current situation | A seller's disgust may be taken as a sign that the seller is seeking to violate behavioral norms in the negotiation (Heerdink et al. 2019). | Move away or against | A third-party observer of scenario featuring disgust (vs. neutral expression) finds that it triggers inferences that a norm has been violated (Heerdink et al. 2019). |
1 a[11].
2 b[53].
More specifically, Table 1 provides theoretical bases to interpret our empirical results. First, positive facial expression (i.e., happiness) invokes the action tendency to move against. In a competitive buyer–seller exchange setting, a seller's happiness expression engenders consumers' inference that the seller is gaining an advantage, thereby reducing the seller's trustworthiness as well as consumers' purchasing tendency ([51], [52]; [53]) — a finding consistent with [57] that a seller's broad smile results in the inference of low competence and the action tendency of reduced buying activity. Second, negative facial expressions (i.e., sadness, anger, fear, and disgust) invoke the action tendency to move away. For instance, a seller's sadness expression in a selling context invokes the consumers' inference of garnering empathy as an attempt to lower their guard, which can be off-putting to consumers, who might then either ignore or avoid the seller. Consistent with this expectation, [10]) show that the display of sadness by frontline employees undermines trust and lowers satisfaction. Last, surprise can be either a positive or negative facial expression: consumers can infer that the seller is trying to garner attention, which invokes the action tendency to either move toward or move against, depending on whether consumers believe the expression is appropriate.
The proposed framework circumvents limitations such as simulated interactions in laboratory studies, survey-based studies relying on self-reports, manual observation and coding of facial displays, small sample sizes, limited set of emotional displays, and lack of business metrics as the response variable. For example, [25] investigate the impact of service representatives' happiness expressions on subjects in lab settings (using trained student actors). [17] conduct surveys with airline passengers' assessments of flight attendants, and [27] rely on surveys completed by sales managers self-reporting their own ability to perceive their salespeople's emotions. [42] manually evaluates a small sample of bank employees' smiles in salesperson–customer service encounters. These studies lack the full spectrum of emotional displays and sales performance as the response. [26], a notable exception, seek to understand the content effects on sales. Specifically, they analyze a small sample of 275 sales pitches from Home Shopping Network and incorporate minute-by-minute cumulative sales; however, they ignore the role of facial expressions. To circumvent the aforementioned limitations, we develop a framework that unobtrusively collects nonsimulated market interactions, does not rely on self-reports, does not require manual observation or coding of facial displays, involves a large sample size, covers a broad spectrum of six emotional displays, and, most importantly, uses sales transactions as the response variable. Figure 1 presents the ten-step framework to capture and analyze the structured and unstructured data from livestream retailing.
Graph: Figure 1. Livestream retail analytics framework.
The first two pennants in Figure 1 list three steps that pertain to data capture. Transaction data contain structured information such as quantity of items sold, prices of items, duration of display, shipping cost, and product category. Video footage of salespeople's presentations offers the unstructured data. We process the video footage as follows. Each video frame is a colored image with a resolution of 480 360 pixels. For every second of the video footage, we select a frame, convert it to grayscale, and present it to a pretrained OpenCV frontal face detection model based on the Haar cascade algorithm ([56]). For each detected face, the grayscale frame bounded by the face's region of interest is forwarded to an emotion classification model to infer the emotional state of the salesperson by producing probabilities for happiness, sadness, surprise, anger, fear, and disgust. We classify emotional displays using a pretrained mini-Xception model developed by [ 1]). Thus, we unobtrusively extract data on whether a face exists in 62.32 million frames and the probabilities of emotional displays. The third pennant in Figure 1 uses time stamps (i.e., the start and end times of item displays) to compute the display duration. We match 25,565 distinct items across 6,065 shows to the accounting data on orders placed, selling prices, and free shipping waivers.
The fourth pennant in Figure 1 consists of dynamic time warping (step 4), dimension reduction (step 5), feature engineering (step 6), and mixed models (step 7). To understand dynamic time warping, let denote the time pattern of the emotional display where for an item i in show s. Data analysis commonly uses variable transformation such as squaring, which alters the magnitude of but keeps its x values (time) unaltered. In contrast, we transform by shifting, stretching, or shrinking the time argument, denoted by t in , but keeping its magnitude (y values) the same. This shifting, stretching, or shrinking applies to each item i in show s and emotion k. For example, Sin(t) and Cos(t) are different curves, yet if we replace t in Cos(t) by ( , we shift the cosine curve along the time dimension to overlap it with the sine curve exactly. The function is called a "warping" function that performs shifting; however, more generally, it can stretch or shrink time, differentially at various instants, to align curves more closely with each other with respect to their landmarks such as peaks, valleys, and inflection points. This process of dynamic time warping is also known as curve registration or landmark alignment. The resulting aligned curves serve as inputs for analysis rather than the raw curves . Subsequently, we present the empirical results with and without curve alignment to understand the benefits of this optional step.
Dimension reduction enables us to capture the dynamic effects of emotional displays on quantity sold. Specifically, for each item-show, the time t in spans over 30 epochs, which are defined as 1/30th of the total duration of item i displayed in show s. Consequently, we have 180 additional variables (i.e., 30 epochs for six emotions). Furthermore, we generate an additional 180 variables by including its quadratic terms and yet another 360 variables by interacting them with price and promotion. To maintain parsimony and mitigate collinearity, we extract the principal component to capture "happiness" (say, when ) via the principal scores , where is the principal eigenvector that reduces the dimensionality from 30 epochs to the scalar . Then, we regress the quantity sold on the principal scores to estimate the trajectory of sales impact of an emotional display together with its confidence intervals. For details, see the Web Appendix.
Feature engineering supplements the new features to represent the quadratic and interaction effects. To this end, we included for each emotion k and interaction terms such as , where denotes a moderating variable of interest (e.g., price). To generate the outputs listed in the fifth pennant in Figure 1, we formulate a set of mixed models.
Figure 2 illustrates how marketing mix — product, price, sales force, display duration (advertising), and free shipping (promotion) — together with face presence and emotional displays (happiness, sadness, surprise, anger, fear, and disgust) affect the focal outcome representing customers' purchase behavior (i.e., sales). In Model 1, we formulate the marketing mix model with marketing mix variables, time effects, and random effects for items and shows. Model 2 extends Model 1 by incorporating the face presence and six emotional displays. Model 3 adds the quadratic effects of emotional displays. Model 4 further augments Model 3 with interactions of price and promotion with the emotional displays.
Graph: Figure 2. Modeling framework.
We investigate how the number of units of an item sold on a given show varies with the item's price, its duration of display, whether free shipping was waived, the product category to which it belongs, the salesperson who presented it, and the time effects (day effect, week effect, and year). The model specification is as follows:
Graph
( 1)
where denotes the quantity sold of an item i in the show s with ( ; represents the item's price in dollars; is the display duration in seconds; captures free shipping ( or not ; and indicate one of the two types of products (whose names are not disclosed for confidentiality). These five variables represent the proxies for the traditional marketing mix variables: product, price, sales force, marketing communications (i.e., length of ad), and promotion (i.e., free shipping). Given the log-log specification, the parameters ( respectively yield the price and duration elasticity, which quantifies the percentage sales impact associated with a 1% increase in price or duration. The parameters ( measure the percentage change in sales due to free shipping and product category. In addition, is a dummy vector, with unity for the element j and zero elsewhere, that identifies individual salesperson j hosting the show s. The corresponding furnishes the sales lift due to various salespeople, , relative to the baseline salesperson 23. A single salesperson owns the entire show in our data. The parameters represent the fixed intercept, random intercept for items, random intercept for shows, and the usual zero mean and constant variance normal error term, respectively. The random effects parsimoniously reflect the variability about the intercept due to heterogeneous impact of items and shows.
Time flows across 62.32 million seconds of the video footage in our analysis, and it exhibits periodicity for the seconds across days and for the days across weeks. To clarify, consider time in seconds since midnight. A total of 86,400 seconds elapse by the midnight of the next day, and then the clock resets to zero (00:00:00 hours). At 23:59:59 hours, the elapsed time is 86,399 seconds, and it is 5 seconds at 00:00:05 hour. Although the instants 23:59:59 and 00:00:05 differ by just 6 seconds, these two instants would be represented as if they are 86,394 seconds apart under a linear scale. Thus, to account for periodicity of the days and weeks, sine and cosine terms should be used as follows. Let represent the seconds of a day when an item i in show s is displayed, where the full day of 86,400 seconds equals 360˚ or radians. Then the two periodic regressors for the day effect are . Similarly, let represent the day of a week when an item i in show s is displayed, where the full week equals seven days. Then the two periodic regressors for the week effect are . Because calendar years are not periodic, unlike seconds or days, let the dummy variable indicate the years. Thus, in Equation 1 includes these five regressors with the conformable parameter vector that constitutes the year effect, the week effect, and the day effect on item sales.
The proposed livestream retail analytics framework provides the fraction of the frames containing a face when item i was displayed in show s, which we denote by . In addition, when item i was displayed in show s, it furnishes the principal score for happiness ( ), sadness ( ), surprise ), anger ), fear ), and disgust ( ). Incorporating them in Model 2, we extend the right-hand side (RHS) of Model 1 as follows:
Graph
( 2)
where are the effects of face presence and six emotional displays on sales. Because the score , the sales impact exhibits the trajectory over epochs.
The effects of emotional displays may wax and wane. For example, moderate happiness may be effective, but limited or excessive happiness display may not be. To investigate such intensity effects, we extend Model 2 by incorporating the quadratic effects of facetime and emotional displays. Then, Model 3 is given by
Graph
( 3)
where , and represent the quadratic effects for the face presence and emotional displays, respectively. Equation 3 also includes the simple effects of face presence ( and emotions ( , the marketing mix effects, and salesperson's effectiveness, time, and fixed and random intercepts via Model 1.
We derive the optimal face presence and emotional displays over time in the Web Appendix, which shows that the optimal number of frames to devote to face presence and each emotion k in every epoch t is given by
Graph
( 4)
Thus, for every epoch t in the presentation span T, the optimal allocation of face presence is ; and the optimal allocation for each emotional display k is . To gain intuition, observe that and in Equation 4, which reveals that face and emotions allocation are proportional to the eigenvector weights: the larger the weight, the greater the intensity of face presence or emotional expressions.
Various factors can moderate the impact of sellers' affective displays on customers' attitudinal and behavioral outcomes. For example, some studies investigate boundary conditions from perceivers' characteristics, such as emotional receptivity ([30]) and epistemic motivation ([57]). Others examine the moderating roles of the selling context, such as store busyness (e.g., [18]). To complement, we explore the moderating role of factors under managers' control such as price and promotion. Specifically, we augment Model 3 as follows:
Graph
( 5)
In Equation 5, the free shipping effect equals which depends on the level of emotional display, . Similarly, each emotion moderates the price elasticity . The preceding discussion completes the inclusion of emotions in marketing mix models.
A livestream retailer, whose identity remains confidential, broadcasts shows 24 hours a day, seven days a week, on its own television channel and sells exclusive hedonic products in multiple product categories. The salesperson hosting the show presents information on products and encourages viewers to place orders by telephone. Each show lasts for 60 or 120 minutes, is planned weeks in advance, and contains live sales pitches of items (i.e., not scripted or prerecorded). Besides selling, the salespeople attempt to build parasocial relationships with viewers so that they feel a bond with virtual personalities analogous to those with television celebrities or news anchors ([49]).
Our direct-to-consumer retailer sells items from two product categories using 23 hosts as salespeople. The salesperson pitches multiple items during a show, and the item appears throughout the presentation span. We observe 99,451 sales pitches at an item-show level on salespeople's presence of face, their facial expressions, item prices, duration of display, shipping fee waivers, and, most importantly, actual sales as the dependent variable. Table 2 presents the descriptive statistics, and Tables 3 and 4 contain the estimation results for Models 1–4 obtained via the R package lme4.
Graph
Table 2. Descriptive Statistics.
| Variable | Description | Mean | SD | Range |
|---|
| Dependent Variable | | | | |
| Quantity, | Number of items sold in a show | 69.64 | 93.21 | [1, 2024] |
| Marketing Mix | | | | |
| Price, | Price of the item | 110.48 | 196.22 | [6.29, 8,000] |
| Duration, | Display duration in seconds | 487.1 | 333.09 | [61, 2,886] |
| Free shipping, | Dummy variable for free shipping | .21 | .29 | {0,1} |
| Product category, | Dummy variable for two categories | .04 | .19 | {0,1} |
| Emotional Displays | | | | |
| Face Presence, | Fraction of the frames with a face | .19 | .12 | [0, 1] |
| Happiness, | Grand mean of probability of happiness display in all the frames with a face | .23 | .21 | [0, 1] |
| Sadness, | Grand mean probability of sadness display in all the frames with a face | .10 | .06 | [0,.62] |
| Surprise, | Grand mean probability of surprise display in all the frames with a face | .08 | .12 | [0,.78] |
| Anger, | Grand mean probability of anger display in all the frames with a face | .09 | .07 | [0,.76] |
| Fear, | Grand mean probability of fear display in all the frames with a face | .11 | .06 | [0,.58] |
| Disgust, | Grand mean probability of disgust display in all the frames with a face | .02 | .03 | [0,.52] |
| Sales Force | | | | |
| Salesperson 1, | Dummy variable for Salesperson 1 | .004 | .065 | {0,1} |
| Salesperson 2, | Dummy variable for Salesperson 2 | .004 | .065 | {0,1} |
| Salesperson 3, | Dummy variable for Salesperson 3 | .034 | .182 | {0,1} |
| Salesperson 4, | Dummy variable for Salesperson 4 | .002 | .044 | {0,1} |
| Salesperson 5, | Dummy variable for Salesperson 5 | .047 | .212 | {0,1} |
| Salesperson 6, | Dummy variable for Salesperson 6 | .036 | .187 | {0,1} |
| Salesperson 7, | Dummy variable for Salesperson 7 | .072 | .258 | {0,1} |
| Salesperson 8, | Dummy variable for Salesperson 8 | .079 | .270 | {0,1} |
| Salesperson 9, | Dummy variable for Salesperson 9 | .008 | .091 | {0,1} |
| Salesperson 10, | Dummy variable for Salesperson 10 | .000 | .020 | {0,1} |
| Salesperson 11, | Dummy variable for Salesperson 11 | .078 | .268 | {0,1} |
| Salesperson 12, | Dummy variable for Salesperson 12 | .082 | .274 | {0,1} |
| Salesperson 13, | Dummy variable for Salesperson 13 | .012 | .108 | {0,1} |
| Salesperson 14, | Dummy variable for Salesperson 14 | .126 | .332 | {0,1} |
| Salesperson 15, | Dummy variable for Salesperson 15 | .055 | .228 | {0,1} |
| Salesperson 16, | Dummy variable for Salesperson 16 | .054 | .226 | {0,1} |
| Salesperson 17, | Dummy variable for Salesperson 17 | .070 | .255 | {0,1} |
| Salesperson 18, | Dummy variable for Salesperson 18 | .052 | .223 | {0,1} |
| Salesperson 19, | Dummy variable for Salesperson 19 | .028 | .166 | {0,1} |
| Salesperson 20, | Dummy variable for Salesperson 20 | .041 | .198 | {0,1} |
| Salesperson 21, | Dummy variable for Salesperson 21 | .002 | .039 | {0,1} |
| Salesperson 22, | Dummy variable for Salesperson 22 | .040 | .196 | {0,1} |
| Salesperson 23, | Dummy variable for Salesperson 23 | .072 | .259 | {0,1} |
| Time Effects | | | | |
| Year, | Year of the display (annual) | 2018 | .50 | [2017, 2019] |
| Day of week, | Day of the display (weekly) | 3.98 | 2.0 | [1, 7] |
| Time of day, | Seconds since midnight | 41508 | 25614 | [3.16, 86400] |
Graph
Table 3. Sales Impact of Marketing Mix, Sales Force, and Time Effects.
| Model 1 | Model 2 | Model 3 | Model 4 |
|---|
| Est. | t-val. | Est. | t-val. | Est. | t-val. | Est. | t-val. |
|---|
| Marketing Mix | | | | | | | | |
| Ln(Price), | –.756 | –142.28 | –.765 | –148.54 | –.769 | –151.28 | –.769 | –150.43 |
| Ln(Duration), | .552 | 127.77 | .626 | 132.10 | .671 | 136.02 | .671 | 135.98 |
| Free shipping, | .218 | 20.36 | .198 | 18.78 | .190 | 18.20 | .189 | 17.97 |
| Product category, | .414 | 5.91 | .386 | 5.47 | .360 | 5.11 | .357 | 5.07 |
| Sales Force | | | | | | | | |
| Salesperson 15, | .438 | 14.92 | .418 | 13.82 | .427 | 14.19 | .426 | 14.16 |
| Salesperson 20, | .412 | 6.59 | .398 | 6.18 | .385 | 6.01 | .385 | 6.01 |
| Salesperson 18, | .351 | 11.37 | .388 | 12.20 | .358 | 11.32 | .358 | 11.34 |
| Salesperson 12, | .313 | 11.61 | .237 | 8.50 | .238 | 8.57 | .237 | 8.57 |
| Salesperson 8, | .311 | 11.01 | .264 | 9.04 | .232 | 7.97 | .231 | 7.96 |
| Salesperson 7, | .288 | 9.92 | .321 | 10.69 | .299 | 10.01 | .300 | 10.04 |
| Salesperson 11, | .270 | 9.40 | .262 | 8.83 | .241 | 8.16 | .241 | 8.18 |
| Salesperson 19, | .255 | 6.48 | .299 | 7.35 | .267 | 6.60 | .267 | 6.60 |
| Salesperson 22, | .249 | 7.34 | .308 | 8.81 | .303 | 8.70 | .303 | 8.71 |
| Salesperson 3, | .227 | 6.07 | .145 | 3.75 | .173 | 4.49 | .173 | 4.52 |
| Salesperson 17, | .169 | 5.57 | .171 | 5.45 | .146 | 4.70 | .147 | 4.73 |
| Salesperson 21, | .140 | .84 | .085 | .49 | .098 | .57 | .096 | .56 |
| Salesperson 6, | .083 | 2.37 | .075 | 2.07 | .092 | 2.55 | .090 | 2.52 |
| Salesperson 16, | –.036 | –1.15 | .064 | 1.95 | .055 | 1.68 | .054 | 1.68 |
| Salesperson 5, | –.099 | –2.95 | –.103 | –2.97 | –.057 | –1.67 | –.058 | –1.67 |
| Salesperson 13, | –.168 | –3.04 | –.200 | –3.50 | –.187 | –3.29 | –.187 | –3.30 |
| Salesperson 23, | –.183 | –6.11 | –.122 | –3.93 | –.090 | –2.92 | –.090 | –2.92 |
| Salesperson 14, | –.212 | –7.90 | –.237 | –8.57 | –.201 | –7.27 | –.200 | –7.25 |
| Salesperson 2, | –.312 | –3.17 | –.186 | –1.83 | –.179 | –1.77 | –.181 | –1.80 |
| Salesperson 1, | –.488 | –6.17 | –.495 | –6.08 | –.496 | –6.14 | –.496 | –6.14 |
| Salesperson 4, | –.539 | –4.22 | –.662 | –5.02 | –.663 | –5.06 | –.662 | –5.05 |
| Salesperson 10, | –.679 | –2.25 | –.697 | –2.23 | –.690 | –2.22 | –.688 | –2.22 |
| Salesperson 9, | –.790 | –11.16 | –.735 | –10.03 | –.749 | –10.29 | –.749 | –10.29 |
| Time Effects | | | | | | | | |
| Year Effect, | –.204 | –16.81 | –.180 | –14.63 | –.167 | –13.60 | –.167 | –13.63 |
| Week Effect (Sin), | –.113 | –13.38 | –.115 | –13.48 | –.112 | –13.24 | –.112 | –13.22 |
| Week Effect (Cos), | .096 | 11.44 | .099 | 11.67 | .105 | 12.36 | .105 | 12.37 |
| Day Effect (Sin), | –.292 | –32.77 | –.302 | –33.52 | –.314 | –34.95 | –.314 | –34.97 |
| Day Effect (Cos), | –.541 | –57.35 | –.563 | –58.98 | –.569 | –59.95 | –.570 | –59.99 |
| Intercept | 414.298 | 16.94 | 347.097 | 13.78 | 320.512 | 12.80 | 321.780 | 12.86 |
| VIF Median (Max) | 1.443 (5.110) | 1.304 (5.172) | 1.604 (9.721) | 1.616 (9.724) |
| Distinct Items | 25,565 | 25,565 | 25,565 | 25,565 |
| Unique Shows | 6,065 | 6,065 | 6,065 | 6,065 |
Graph
Table 4. Sales Impact of Face and Emotional Displays.
| Model 2 | Model 3 | Model 4 |
|---|
| Estimate | t-values | Estimate | t-values | Estimate | t-values |
|---|
| Main Effects | | | | | | |
| Face Presence, | .338 | 12.66 | 3.138 | 43.35 | 3.140 | 43.37 |
| Happiness, | –.033 | –40.39 | –.025 | –25.26 | –.024 | –20.23 |
| Sadness, | –.003 | –3.56 | –.001 | –.90 | .001 | .62 |
| Surprise, | –.001 | –1.07 | .007 | 7.32 | .008 | 7.09 |
| Anger, | –.033 | –45.82 | –.031 | –29.70 | –.031 | –26.03 |
| Fear, | –.005 | –6.58 | –.003 | –3.13 | –.003 | –2.92 |
| Disgust, | –.012 | –16.63 | –.011 | –10.47 | –.010 | –8.52 |
| Quadratic Effects (Estimate) | | | | | | |
| Face Presence, | | | –4,641 | –41.66 | –4,644 | –41.69 |
| Happiness, | | | –.769 | –6.79 | –.764 | –6.73 |
| Sadness, | | | –.296 | –4.31 | –.303 | –4.41 |
| Surprise, | | | –.613 | –7.80 | –.629 | –7.99 |
| Anger, | | | .010 | .17 | .017 | .28 |
| Fear, | | | –.442 | –6.20 | –.440 | –6.17 |
| Disgust, | | | –.060 | –1.93 | –.061 | –1.97 |
| Moderation Effects (Estimate) | | | | | | |
| Happiness Price, | | | | | –.009 | –2.04 |
| Sadness Price, | | | | | –.012 | –3.55 |
| Surprise Price, | | | | | –.012 | –4.10 |
| Anger Price, | | | | | .005 | 1.27 |
| Fear Price, | | | | | .001 | .32 |
| Disgust Price, | | | | | –.001 | –.27 |
| Happiness Shipping, | | | | | 2.248 | .84 |
| Sadness , | | | | | –.411 | –.18 |
| Surprise Shipping, | | | | | 3.699 | 1.80 |
| Anger Shipping, | | | | | –1.671 | –.76 |
| Fear Shipping, | | | | | 1.144 | .51 |
| Disgust Shipping, | | | | | –1.882 | –.82 |
[31]) apply convolutional neural networks to detect the presence of a person's face in a Kickstarter crowdfunding video and show empirically that the presence of a human face makes a difference in shaping the desired funding outcomes. But does it impact sales? If so, to what extent? Our study answers these questions. The estimate of.338 in Table 4 (Model 2) means that sales increase by.34% when a face is present, an effect common to all the hosts. For a specific salesperson, say salesperson 15, the impact of sales pitch is , which means sales increase by.76%. This magnitude explains why the livestream retailer prefers live broadcasts even when items could have been posted on the internet in a faceless manner.
Table 4 for Model 2 partially presents the sales impact of happiness, sadness, anger, fear, and disgust. The estimates for happiness (–.033), sadness (–.003), surprise (–.001), anger (–.033), fear (–.005), and disgust (–.012) are uniformly negative and statistically significant for all emotions except surprise. Thus, we conclude that emotional displays decrease sales.
We present the dynamic pattern of emotional displays with (see Figure 3) and without (see Figure 4) dynamic time warping. These dynamic patterns emerge from the elements of the eigenvector across the 30 epochs . For clarity, Figures 3 and 4 present the epochs on the unit interval. The elements of the eigenvector , together with the estimates , yield the total sales impact of emotions given by . Summing across all the epochs, the sales impact of emotional displays are as follows: happiness (–.18%), sadness (–.02%), surprise (–.004%), anger (–.18%), fear (–.03%), and disgust (–.06%). Happiness and anger induce the largest sales decline; surprise the smallest. Summing across these emotions, the magnitude of total sales decline (.47%) is more than twice the free-shipping effect (.198%). Happiness contributes more than one-third to the total sales decline.
Graph: Figure 3. Time-varying sales impact of emotional displays with dynamic time warping.
Graph: Figure 4. Time-varying sales impact of emotional displays without dynamic time warping.
What accounts for the negative sales impact? As discussed in the "Conceptual Background" section, sellers' emotional displays trigger buyers' inference and action tendencies. Specifically, Table 1 shows that positive facial expressions such as happiness negatively impact sales because consumers infer that the seller is gaining an advantage at their expense, thereby reducing sellers' trustworthiness and, in turn, buyers' tendency to purchase ([51], [52]; [53]). This expectation corroborates [57]) findings, which show that a seller's broad smile results in a buyer's inference of a seller's low competence and reduces buying activity. A similar situation occurs with politicians sporting a "permasmile" (i.e., maintaining a smile for an extended period of time); they are not perceived as genuine, which induces distrust and leads to lost votes ([60]). As for negative facial expressions (i.e., sadness, anger, fear, and disgust), they invoke the action tendency to move away, which corroborates [10]) finding that frontline employees' displays of sadness undermine trust and reduce satisfaction. Last, surprise can be either positive or negative, and it results in an insignificant effect on sales. Thus, this large-scale evidence supports recent studies ([10]; [57]), and so we caution that emotional displays are bad for livestream retailing business.
Over an item's presentation span, the magnitude of sales impact builds up, attains a maximum in the middle, and recedes toward the end. Across the six emotions, this dynamic pattern holds uniformly. Are the U-shaped patterns significant? Using the expressions in the Web Appendix, we plot the confidence intervals in Figures 3 and 4. We conclude that because zero does not belong in it, except for surprise, the rest of the emotional displays, including happiness, exert significantly negative effects on sales.
What accounts for the U-shaped dynamics? The literature on advertising repetition (e.g., [ 5]; [ 9]; [38]; [41]) provides a plausible interpretation. As the sales pitch progresses, the repetitiveness of facial expressions exacerbates the negative sales impact (i.e., becomes more negative) and drives it to the lowest level. After that, often due to the tedium ([ 5]) of a protracted sales pitch, viewers' attention drifts from the message-related thoughts to their own thoughts ([ 9]) of purchase consideration, namely, balancing the benefits and costs of the presented item and deciding whether to buy. Consequently, the negative effect ameliorates during purchase consideration. [38] find a similar U-shaped pattern for the effectiveness of television commercials (see their Figures 4 and 9 for chocolate and cereal brands, respectively).
Graph: Figure 9. Optimal face allocation.
Model 3 specifies the quadratic effects to explore whether emotional displays can be optimized. Table 4 shows that the conditions and are not satisfied by happiness, sadness, anger, fear, and disgust. Consequently, their resulting optimal level according to Equation 4. Although surprise satisfies the conditions and , the salesperson cannot express only surprise throughout the presentation in the absence of other emotions; thus, this corner solution does not seem practically useful. In contrast, the face presence satisfies the conditions and , and the optimal .34. For comparison, the average face presence in Table 2 is.19. Thus, face presence should be increased from 19% to 34% to maximize sales.
Model 4 specifies the interactions of emotional displays with free shipping and price. Table 4 shows that the estimated are not significant for all k. Hence, the main effects of emotional displays hold regardless of the shipping fee waiver. Similarly, the price interaction effects of fear ( ), anger ( ), and disgust ( ) are not significant, thereby generalizing their main effects across various prices.
In contrast, the interaction effects of happiness ( ), sadness ( ), and surprise ( ) are significant and negative. They moderate price elasticity: . Substituting for happiness from Table 4 and the average price of $110.48 from Table 2, we get price elasticity , which means viewers become more price sensitive as the intensity of sellers' happiness increases. Why? Because the buyers suspect that the seller is gaining at their expense ([53]), and they exhibit "move against" tendencies (see Table 1). The qualitatively similar results hold for sadness and surprise. These interaction effects generalize our previous findings: emotional displays are bad for business.
According to the log-linear specification, the estimated coefficient of.386 (see Model 2 in Table 3) means that, ceteris paribus, a product from category 1 sells 1.47 ( times more than a product from category 2. The estimated price elasticity equals –.765 (see Model 2 in Table 3), which means a 10% increase in price corresponds to a 7.65% decrease in sales. Similarly, the estimated display duration elasticity equals.626 (see Model 2 in Table 3), which means a 10% increase in display duration corresponds to a 6.26% increase in sales, which is about 2 to 6 times larger than advertising elasticity (see [47]). The free shipping increases the quantity sold by.198% (see Model 2 in Table 3). Using the average price of $110.48 and the average quantity of 69.64 (see Table 2), the shipping waiver increases revenues by $15.23 and is profitable when the shipping costs are below $16.
Human biases affect the performance appraisal process (e.g., pay, bonus, advancement rate, prestige). We propose that to mitigate these biases, salespeople should be ranked on their individual effectiveness (objective attribution) rather than average sales (naïve attribution). Model 1 facilitates the estimation of the effectiveness of an individual salesperson by controlling for prices, duration, free shipping, time of day, and week. Table 3 reports the estimates of percentage sales increase for an individual salesperson relative to the group average based on effects coding of the dummy variables (see [22]). Consider the estimate of –.488 for salesperson 1 from Model 1 in Table 3. That estimate means salesperson 1's performance is.488% below the group average. Similarly, salesperson 6's performance is.083% above the group average. These estimated effects are not affected by human cognitive biases.
Graph: Figure 5. Salesperson performance appraisal.
We compare the salesperson's performance rank based on the naïve versus objective attributions. Panel A in Figure 5 shows the ranking of 23 salespeople based on the average sales, which ignores the effects of prices, duration, free shipping, and time of day and week. In contrast, Panel B shows the ranking of the same 23 salespeople based on their individual effectiveness. The top and the bottom three salespeople remain the same under both metrics, thereby showing that the ranking attains convergent validity by identifying the same set of best and worst performers. However, the majority of salespeople (∼75%) reside in the middle, where the rank ordering differs across metrics. Thus, the objective attribution based on salesperson's effectiveness after controlling for prices, duration, free shipping, and time of day and week should guide supervisors in more objectively selecting salespeople for rewards, recognition, and retraining.
Figures 6 and 7 depict the sales impact of emotional displays based on the full model (Model 4). In these figures, a low (high) price refers to the 25th (75th) percentile of the price distribution. First, emotional displays decrease sales. This finding holds uniformly for negative and positive emotions. Because serves as the dependent variable, the marginal change in it equals , which represents "percentage change in sales." Thus, a marginal increase in emotional display corresponds to a sales decline that ranges from.004% to.18%. Across the six emotions, the magnitude of the total sales decline (.47%) is more than double the free-shipping effect (.20%). Second, because the tangent to the curves in Figures 6 and 7 becomes steeper as the intensity of emotional display increases, the sales decline accelerates. In other words, the sales decline increases at an increasing rate. Thus, not displaying emotions emerges as the optimal course of action. So, salespeople should sell with a neutral face, although how consumers interpret "neutral" depends on the sellers' gender, facial morphology, and contextual factors (e.g., [23]). Finally, the parallel curves in Figures 6 and 7 reveal the modest magnitude of moderation effects: sales decrease as price increases or promotion decreases (see the dashed curves).
Graph: Figure 6. Emotional displays by price interactions.
Graph: Figure 7. Emotional display by free shipping interactions.
In the week effect, sine and cosine variables jointly identify the sales variations across days of the week. The cosine variable differentiates the first half of the week (Monday to noon Thursday) from the second half of the week (noon Thursday to Sunday). The sine variable differentiates the middle of the week (9 p.m. Tuesday to 9 p.m. Friday) from the end of the week (9 p.m. Friday to 9 a.m. Tuesday). Similarly, in the day effect, the sine and cosine variables identify the sales variations across hours of the day. Specifically, the cosine variable differentiates post meridiem (p.m.) from ante meridiem (a.m.), while the sine variable captures the rhythms across the late evening (6 p.m. to midnight) through the night hours (midnight to 6 a.m.) to the morning hours (6 a.m. to noon) and the afternoon (noon to 6 p.m.). Because the empirical results indicate that the cosine variable is less important than the sine variable, the a.m./p.m. distinction is not critical. As expected, sales occur 24 hours a day, including the nights; peak during the day; and are larger during the weekends.
Figure 8 presents the relative contribution of marketing mix and nonmarketing variables: the former contributes 71%, whereas the latter accounts for 29% of the total . The time of day, the day of the week, and the week of the year explain 20%. Emotional displays and face presence further explain 9% of the explained variance. Thus, nonmarketing variables boost explanatory power.
Graph: Figure 8. Variable importance.
A glance across the columns in Table 3 indicates a remarkable robustness. The columns reveal the estimated marketing mix effects in the presence of various operationalizations of emotional displays. For example, across Models 1–4 the price elasticity varies from –.76 to –.77, and the duration elasticity ranges from.55 to.67. Likewise, shipping and product estimates are (.22,.41), (.20,.39), (.19,.36), and (.19,.36) across Models 1–4, respectively. Salesforce effectiveness across the four models is also stable; for example, the percentage sales increase due to salesperson 15 hovers around.42. Even the rhythms of daily and weekly sales deviate only marginally. These results hold even when we replaced the static face variable in Equation 3 with the dynamic component across the epochs . Furthermore, we tested for heterogeneous effects of emotional displays and found that the effects were homogeneous across the two product categories (see Figures 6 and 7). Thus, the broad robustness —for all the variables and across all the models— enhances confidence in these results.
We also analyzed data by splitting the presentation span of an item i displayed in a show s into three time segments. We discovered V-shaped effects across the beginning, middle, and end of the presentation span for all the six emotions, including happiness. We then extended this analysis tenfold to 30 epochs and found that not only does the parameter stability hold in both analyses, but also qualitatively similar results persist: negative U-shaped effects of emotions over the presentation span. Furthermore, the average variance inflation factor across all independent variables was 1.66, ranging from 1.01 to 5.17, which is far below 10 and thus rules out multicollinearity concerns.
To our knowledge, this study marks the first time dynamic time warping appears in marketing. To further assess robustness, we reestimated Model 2 without dynamic time warping. As mentioned previously, dynamic time warping aligns landmarks such as the peaks, valleys, and inflections of the raw emotional curves . Such landmark alignment homogenizes the timing of peaks, valleys, and inflections in . Consequently, the estimated trajectories, , in Figure 3 are smoother than those in Figure 4 without landmark alignments. More importantly, the overall pattern remains the same: the sales impact is negative, U-shaped, and similar across the six emotions. In summary, the U-shaped patterns as well as other results are robust. We close this section by comparing the performance of models on multiple metrics.
Which one of the four models is the best? Although the adjusted of about 80% is remarkable, especially given 99,451 sales pitches, it does not discriminate among the four models as the log-likelihood, Akaike information criterion, and Bayesian information criterion do. Therefore, we compared the models using these metrics and present the results in Table 5. Specifically, Models 3 and 4 dominate Models 1 and 2 on all the metrics. The Bayesian information criterion selects Model 3, whereas both the other metrics (log-likelihood and Akaike information criterion) indicate that Model 4 outperforms the rest. We used Model 4 to plot Figures 6 and 7. We next discuss the implications of these findings.
Graph
Table 5. Models Comparison.
| Model 1 | Model 2 | Model 3 | Model 4 |
|---|
| R2 | 79.10% | 79.7% | 80.0% | 80.0% |
|---|
| Log Likelihood | –110,560 | –108,447 | –107,472 | –107,456 |
| AIC | 221,184 | 216,972 | 215,036 | 215,027 |
| BIC | 221,488 | 217,343 | 215,473 | 215,579 |
| Fixed Parameters | 32 | 39 | 46 | 58 |
| Observations | 99,451 | 99,451 | 99,451 | 99,451 |
This research offers important insights into livestream retailing by addressing two foundational questions identified in this special issue dedicated to understanding the interface of technology and marketing: ( 1) How can managers use new types of data to improve marketing decision making? and ( 2) What new methods can deliver the best consumer insights to improve marketing strategy? To address the first question, the second pennant in Figure 1 captures the new type of data available from streaming videos of sales presentations, which can identify a large-scale, unobtrusive, and comprehensive set of emotions. To address the second question, the fourth pennant in Figure 1 contributes the new methods to create six emotional trajectories via functional principal components analysis and dynamic time warping to align them. Incorporating them as quadratic and moderating variables, we then assess the value of emotional displays. Building on Models 1–4, we discuss the following theoretical and managerial contributions.
Livestream e-commerce, which features hosts promoting and selling goods and services in real time via screen-mediated sales presentations, represents an emerging opportunity for marketers to create, deliver, and communicate content so as to monetize in ways not possible previously. Specifically, marketers can, first, reach customers via new channels such as social messaging apps (Facebook, WeChat), livestreaming services (e.g., Twitch), and internet platforms (e.g., Taobao Live) that integrate shopping and entertainment. Second, these technology platforms facilitate purchases from wherever and whenever customers are seeking to buy. Third, they shorten the purchase funnel by demonstrating a product and describing why it is a must-have item; conveying that only limited quantities are available; counting down the time remaining on the item before the next item is to be introduced; and injecting such calls to action as "grab it before its gone." Finally, technology allows value capture in formats not possible previously: noncash payments (e.g., PayPal, Venmo), installment payments (e.g., Klarna unsecured loans), and barter payments (e.g., BarterOnly.com, which provides exchanges of used products). Marketers need to imagine how they can integrate such value creation and value capture opportunities made possible by technological advances.
Earlier studies used human intervention to collect data on emotional displays at a small scale (e.g., [ 8]; [30]; [42]). In contrast, applying artificial intelligence (see, e.g., [33]; [35]), we extract face presence and facial expressions from 62.32 million frames of streaming video sales presentations automatically and unobtrusively, thereby responding to calls to harness machine learning to generate meaning from big data (e.g., [ 4]; [29]; [40]; [59]).
Earlier studies consider either a single (e.g., [32]; [57]; [58]) or a select few emotions (e.g., [10]; [50]; [54]; [55]), potentially resulting in biased estimates due to omitted variables. Hence, Model 2 specifies a comprehensive set of six emotional displays simultaneously. Our focus on salespeople's emotional displays is also responsive to an earlier call to devote greater attention to emotions on the seller side of the exchange dyad (see [ 3]).
Earlier studies focus on static episodic expressions (e.g., [10]; [55]). Our Model 2 permits capturing the dynamic trajectories of emotions at a more granular level, thereby revealing time-varying patterns of sales impact (see Figure 3). More importantly, the Web Appendix makes original contributions to the theory of inference on the effects of functional principal components.
Virtually all studies on emotions have used customer mindset metrics as the dependent variables (e.g., [54]; [57]; [58]). While [32] use box-office revenues, they specify happiness to monotonically affect sales, ruling out the possibility of that an optimal level of emotions exists. Given the nonmonotonic effects in Models 3 and 4, the theoretical existence of the optimal mix arises. Equation 4 presents the optimal emotions to display so as to maximize sales. These results not only make original contributions to the extant literature but also offer guidance to design technology-inspired service agents (e.g., avatars, virtual news anchors) to be more humanlike ([11]; [37]). They also inform the discussion about technology and marketing in that artificial intelligence can be used to monitor the seller's facial activity, provide real-time coaching, and thus assist in training salespeople to improve business outcomes ([20]; [33]).
A marketplace increasingly characterized by greater technological connectivity and interactivity has prompted calls to investigate the business impact of a seller's facial expressions in screen-mediated commercial interactions ([ 7]). [28], for instance, underscore the need to evaluate whether their results from in-person, face-to-face customer encounters involving "emotionally calibrated" salespeople will hold in digital exchanges. The authors contend that this type of salesperson exhibits calmness and that exuding calmness builds rapport, which in turn drives favorable sales performance outcomes. Our theorizing, which is steeped in EASI's predictions about the inferences that buyers draw about a seller's facial expressions in a competitive exchange (see Table 1), and findings from a one-to-many livestream broadcast setting reaffirm the importance of reducing emotional displays in driving sales effectiveness. We thereby contribute to understanding the communicative role of facial expressions in screen-mediated exchanges and elaborate further in the following section on "selling with a straight face."
How should face presence be allocated over an item's presentation span? Should frames containing a face be displayed uniformly or in chunks? If the latter, should they be concentrated in the middle, when sales resistance is highest? To this end, we evaluate Equation 4 and present the optimal percentage allocation of the total number of frames with a face over an item's presentation span in Figure 9. We observe that the optimal allocation is neither uniformly displayed nor chunked in the middle. Rather, the optimal number of face frames wanes at the start, gradually builds to a crescendo, and eventually ebbs. Specifically, the optimal allocation decreases on the initial 10% span, then gradually increases as the presentation progresses, and finally decreases in the last 15% span. Remarkably, this optimal allocation conforms to the three-part structure of stories: the beginning, the middle, and the end (see [36]).
Figures 6 and 7 uncover the novel and provocative findings that, first, positive emotional displays reduce sales. Second, the greater the intensity, the larger the decline. To mitigate the negative effect, salespeople should consider toning down their facial expressions. To mitigate the quadratic effects of intensity, they can abate their exaggerated expressions. Together, these findings indicate a new maxim: sell with a straight face. Consistent with this maxim, [12], p. 82) advocates that direct marketers should use a "journalist approach" to answer the "who, what, why, when, where of a product. Whom is the product for? What does it do? Why is it beneficial? When can it be used? Where can it be bought?" In other words, livestream salespeople should broadcast their pitch with a stoic expression akin to that of news anchors, though we acknowledge that this implication may not generalize to face-to-face communications in business markets.
Figures 3 and 4 can be interpreted as the sales resistance curve. The maximum sales resistance is near the middle of an item's presentation; the least sales resistance is at the beginning and end of presentations. This U-shaped sales resistance curve provides actionable guidelines to practitioners. Emotional displays at the beginning and end of presentations help engage consumers and build rapport. However, during the livestream show, hosts should monitor the frequency and intensity of their emotional expressions. Because genuine interactions involve less emotional and more neutral expressions, salespeople can make emotional connections with the audience with neutral expressions and lessen the insidious effects of sales resistance. Although hosts cannot completely avoid emotions, they should take advantage of livestreaming platforms to emotionally connect the brands with customers.
A naïve assessment of a salesperson's performance is based on the actual quantity sold. However, this quantity depends on factors such as prices, duration, free shipping, and time of day and week. In other words, a salesperson can validly object that another salesperson's larger actual sales are not reflective of his/her performance alone because price, duration, free shipping, and time of day and week also impacted sales. Cognitive biases further compound such performance appraisals. Prominent factors driving cognitive biases include the fundamental attribution error, halo effects, the leniency bias, the recency bias, selective perception, the self-serving bias, and the similarity bias (e.g., [48]). Fundamental attribution error refers to supervisors underestimating the influence of external factors and overestimating the influence of internal factors when judging a salesperson's performance; halo effects arise when a general impression of a salesperson overshadows the relevant metrics; leniency bias refers to a supervisor's tendency to rate all salespeople positively (or negatively), reducing the difference between top and bottom performers; recency bias creeps in when recent events (e.g., bumper sales, sharp declines) influence supervisors' judgments; selective perception refers to the supervisor's tendency to notice certain metrics and filter out others; self-serving bias emerges when a salesperson attributes own successes to internal factors and failures to external factors; and similarity bias (i.e., homophily) shapes the evaluation when supervisors reward a salesperson similar to themselves.
Using Table 3, managers can objectively rank salespeople (see Panel C in Figure 5) to circumvent the effects of the aforementioned biases. Indeed, the CEO and senior leadership team of the livestream retailer we examined found our proposed framework valuable to recognize excellence and identify candidates for retraining. Both recognition and training, in turn, help improve future sales performance ([61]). Thus, the framework in Figure 1 unlocks the power of data and contributes to sales performance analytics.
[26]) recent study suggests that information content can matter. Specifically, they analyze 275 sales pitches from the Home Shopping Network, manually code the minute-by-minute content on the cumulative sales thus far during an item's presentation span, and show that the intermittent availability of this information increases item sales by.084 units at the onset and decreases linearly to.015 units at the end for a unit increase in the displayed cumulative sales. We encourage future researchers to automate such content analysis to extract and incorporate facial expressions.
Our empirical study pertains to one-to-many screen-mediated competitive exchanges, and it shows that the salesperson's emotional expressions evoke negative inferences by viewers about the salesperson's intentions. One explanation may be the absence of social interaction. When the salesperson smiles, the viewers may not reciprocate because the salesperson's emotions are not targeted to a specific viewer. Such differences provide the impetus to study screen-mediated face-to-face interactions in the presence of social others. For example, [21] contend that the customer purchase journey involves traveling with social others, which necessitates investigations into the various influences that members of the social network can have on buyers' appraisals, intentions, and actions. In a livestream e-commerce setting, the host becomes an important social other. Viewers can readily communicate with the influencer via live chat texts, emojis, voice, and/or video and further enhance their sense of connection with that celebrity (i.e., parasocial relationship). Does the host's verbal and nonverbal communications influence viewers' behavior in such communal, two-way screen-mediated exchanges? Do purchases by social others induce "fear of missing out"? [43] suggests conversion rates of 30% in livestream shopping versus 3% in traditional marketing. To incorporate such two-way communications in the models, researchers should augment the regressors with the characteristics of not only the items and sellers (as in this study), but also the network of social others and the hosts. We encourage further research to shed light on two-way communications in livestream shopping.
[25] manipulate authenticity (i.e., surface or deep acting) and emotional intensity in simulated service encounters (i.e., actors played the role of employees) with 223 consumers to understand the effects on customer satisfaction, customer–employee rapport, and loyalty intentions. They show that authenticity rather than intensity influences customers' reactions (for similar results, see [57]). We encourage future researchers to design emotion recognition algorithms that can classify facial expressions on the basis of emotional authenticity in addition to intensity.
Previous studies predominantly focus on marketing mix effects on sales because when they were conducted, machine learning technology was not available to detect faces and extract emotions at scale. This study combines machine learning technology and marketing. Specifically, we develop the retail analytics engine (see Figure 1) to unobtrusively collect data on face presence and emotional displays. Applying this technology to livestream retail data, we found that facial expressions, including happiness, adversely impact sales. This counterintuitive and provocative finding suggests that salespeople should sell with a straight face. These negative effects exhibit U-shaped dynamics over an item's presentation span, uniformly across six emotions, revealing that the largest sales resistance occurs during the middle of the presentation. Furthermore, the presence of a face matters because it impacts sales positively; therefore, it should be present more than is currently the case. Yet, its optimal allocation over time should be reduced over the initial 10% span, then gradually increased as the presentation progresses, and subsequently tapered down in the last 15% span. Finally, the retail analytics engine empowers managers to more objectively assess the effectiveness of each individual salesperson (see Figure 5), thereby circumventing cognitive biases in performance appraisals.
This study highlights the importance of monitoring and managing facial expressions. One implication is to train new salespeople. The firm can analyze the video footage, much like sports teams watch films of critical moments in previous games to learn what individual players did well and not so well, and sales coaches can help discern the extent to which they displayed emotions and the proportion of each emotion expressed. The feedback from such debriefing sessions could be used to modify sales pitches. Another implication is to retrain experienced sales professionals. The firm can compare each salesperson with the top performer (see Figure 5) and identify which emotions the salesperson ought to tackle. Happiness is the first one that should be addressed. While previous research advocates "service with a smile," we suggest selling with a straight face. Smiling may be off-putting because it lacks authenticity ([25]), reducing trust in the seller ([10]). Subsequently, salespeople should address displays of anger, then fear, and other negative emotions. Last, this study has implications for bot marketing. As technology advances, bots will more closely mimic human facial expressions and supersede humans in monitoring and managing facial expressions. Chat bots, like humans, provide voice assistance to customers. Similarly, three-dimensional audiovisual bots, like salespeople, can engage with customers. For example, HSBC Bank in Northern California employs Pepper, a social humanoid robot ([14]). Further technological advances will bestow bots with the ability to express and reciprocate emotions, thereby assisting livestream retailers to nudge prospective customers through the purchase funnel by explaining features and benefits, instilling urgency to buy, and entertaining them along the way.
sj-pdf-1-jmx-10.1177_00222429211013042 - Supplemental material for A New Livestream Retail Analytics Framework to Assess the Sales Impact of Emotional Displays
Supplemental material, sj-pdf-1-jmx-10.1177_00222429211013042 for A New Livestream Retail Analytics Framework to Assess the Sales Impact of Emotional Displays by Neeraj Bharadwaj, Michel Ballings, Prasad A. Naik, Miller Moore and Mustafa Murat Arat in Journal of Marketing
Footnotes 1 Michael Ahearne
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received the following financial support for the research, authorship, and/or publication of this article: The first two authors received research support from the Haslam College of Business, and the third author acknowledges funding received from UC Davis research grants and the Class of 1989 Endowment.
4 Online supplement:https://doi.org/10.1177/00222429211013042
References Arriaga Octavio , Ploger Paul , Valdenegro-Toro Matias. (2017), "Image Captioning of Classification of Dangerous Situations," working paper , https://arxiv.org/abs/1711.02578.
Ataman M. B. , Mela Carl F. , Van Heerde Harald J.. (2008), " Building Brands ," Marketing Science , 27 (6), 1036 – 54.
Bagozzi Richard P. , Gopinath Mahesh , Nyer Prashanth U.. (1999), " The Role of Emotions in Marketing ," Journal of the Academy of Marketing Science , 27 (2), 184 – 206.
Balducci Bitty , Marinova Detelina. (2018), " Unstructured Data in Marketing ," Journal of the Academy of Marketing Science , 46 (4), 557 – 90.
5 Berlyne D. E. (1970), " Novelty, Complexity, and Hedonic Value ," Perception and Psychophysics , 8 (5), 279 – 86.
6 Bharadwaj Neeraj , Ballings Michel , Naik Prasad A.. (2020), " Cross-Media Consumption: Insights from Super Bowl Advertising ," Journal of Interactive Marketing , 50 (2), 17 – 31.
7 Bharadwaj Neeraj , Shipley Garrett M.. (2020), " Salesperson Communication Effectiveness in a Digital Sales Interaction ," Industrial Marketing Management , 90 (7), 106 – 12.
8 Brown Carolyn S. , Sulzer-Azaroff Beth. (1994), " An Assessment of the Relationship Between Customer Satisfaction and Service Friendliness ," Journal of Organizational Behavior Management , 14 (2), 55 – 75.
9 Calder Bobby J. , Sternthal Brian. (1980), " Television Commercial Wearout: An Information Processing View ," Journal of Marketing Research , 17 (2), 173 – 86.
Cheshin Arik , Amit Adi , Van Kleef Gerben A.. (2018), " The Interpersonal Effects of Emotion Intensity in Customer Service: Perceived Appropriateness and Authenticity of Attendants' Emotional Displays Shape Customer Trust and Satisfaction ," Organizational Behavior and Human Decision Processes , 144 , 97 – 111.
Crivelli Carlos , Fridlund Alan J.. (2018), " Facial Displays Are Tools for Social Influence ," Trends in Cognitive Science , 22 (5), 388 – 99.
Eicoff Al. (1995), Direct Marketing Through Broadcast Media. Chicago : NTC Publishing Group.
Ekman Paul , Friesen Wallace V.. (1978), Facial Action Coding System: A Technique for the Measurement of Facial Movement. Palo Alto, CA : Consulting Psychologists Press.
Eng Ilene. (2019), "Bank Branch Unveils Its Social Humanoid Robot for Customer Service in Silicon Valley," (accessed February 26, 2021) , https://www.ntd.com/bank-branch-unveils-its-first-social-humanoid-robot-for-customer-service-in-silicon-valley%5f369330.html.
Fridlund Alan J. (1994), Human Facial Expression: An Evolutionary View. San Diego, CA : Academic Press.
Gatignon Hubert , Hanssens Dominique M.. (1987), " Modeling Marketing Interactions with Application to Salesforce Effectiveness ," Journal of Marketing Research , 24 (3), 247 – 57.
Gountas Sandra , Ewing Michael T. , Gountas John I.. (2007), " Testing Airline Passengers' Responses to Flight Attendants' Expressive Displays: The Effect of Positive Affect ," Journal of Business Research , 60 (1), 81 – 3.
Grandey Alicia , Goldberg Lori S. , Douglas Pugh S.. (2011), " Why and When Do Stores with Satisfied Employees Have Satisfied Customers? The Roles of Responsiveness and Store Busyness ," Journal of Service Research , 14 (4), 397 – 409.
Greenwald Michelle. (2020), "Live Streaming E-Commerce Is the Rage in China. Is the U.S. Next?" Forbes (December 10) , https://www.forbes.com/sites/michellegreenwald/2020/12/10/live-streaming-e-commerce-is-the-rage-in-china-is-the-us-next/?sh=7a9cf92b6535.
Grewal Dhruv , Hulland John , Kopalle Praveen K. , Karahanna Elena. (2020), " The Future of Technology and Marketing: A Multidisciplinary Perspective ," Journal of the Academy of Marketing Science , 48 (1), 1 – 8.
Hamilton Ryan , Ferraro Rosellina , Haws Kelly L. , Mukhopadhyay Anirban. (2020), " Traveling with Companions: The Social Journey ," Journal of Marketing , 85 (1), 68 – 92.
Hardy Melissa A. (1993), " Alternative Coding Schemes for Dummy Variables ," in Regression with Dummy Variables , Hardy Melissa A. , ed. Newbury Park, CA : SAGE Publications , 64 – 74.
Hareli Shlomo , Shomrat Noga , Hess Ursula. (2009), " Emotional Versus Neutral Expressions and Perceptions of Social Dominance and Submissiveness ," Emotion , 9 (3), 378 – 84.
Heerdink Marc W. , Koning Lukas F. , Van Doorn Evert A. , Van Kleef Gerben A.. (2019), " Emotions as Guardians of Group Norms: Expressions of Anger and Disgust Drive Inferences About Autonomy and Purity Violations ," Cognition and Emotion , 33 (3), 563 – 78.
Hennig-Thurau Thorsten , Groth Markus , Paul Michael , Gremler Dwayne D.. (2006), " Are All Smiles Created Equal? How Emotional Contagion and Emotional Labor Affect Service Relationships ," Journal of Marketing , 70 (3), 58 – 73.
Hu Ye , Wang Kitty , Chen Ming , Hui Sam. (2021), " Herding Among Retail Shoppers: The Case of Television Shopping Network ," Customer Needs and Solutions , 8 , 27 – 40.
Kadic-Maglajlic Selma , Micevski Milena , Arslanagic-Kalajdzic Maja , Lee Nick. (2017), " Customer and Selling Orientations of Retail Salespeople and the Sales Manager's Ability-to-Perceive Emotions: A Multi-Level Approach ," Journal of Business Research , 80 (11), 53 – 62.
Kidwell Blair , Hasford Jonathan , Turner Broderick , Hardesty David M. , Zablah Alex. (2021), " Emotional Calibration and Salesperson Performance ," Journal of Marketing, forthcoming , DOI: 10.1177/0022242921999603.
Lam Son K. , Sleep Stefan , Hennig-Thurau Thorsten , Sridhar Srihari , Saboo Alok. (2017), " Leveraging Frontline Employees' Small Data and Firm-Level Big Data in Frontline Management: An Absorptive Capacity Perspective ," Journal of Service Research , 20 (1), 12 – 28.
Lee Yih Hwai , Lim Elison Ai Ching. (2010), "When Good Cheer Goes Unrequited: How Emotional Receptivity Affects Evaluation of Expressed Emotion , " Journal of Marketing Research , 47 (6), 1151 – 61.
Li Xi , Shi Mengze , Wang Xin (Shane). (2019), " Video Mining: Measuring Visual Information Using Automatic Methods ," International Journal of Research in Marketing , 36 (2), 216 – 31.
Liu Xuan , Shi Savannah Wei , Teixeira Thales , Wedel Michel. (2018), " Video Content Marketing: The Making of Clips ," Journal of Marketing , 82 (4), 86 – 101.
Luo Xueming , Qin Marco Shaojun , Fang Zheng , Qu Zhe. (2021), " Artificial Intelligence Coaches for Sales Agents: Caveats and Solutions ," Journal of Marketing , 85 (2), 14 – 32.
Marinova Detelina , Singh Sunil K. , Singh Jagdip. (2018), " Frontline Problem-Solving Effectiveness: A Dynamic Analysis of Verbal and Nonverbal Cues ," Journal of Marketing Research , 55 (2), 178 – 92.
McDuff Daniel , Berger Jonah. (2020), " Why Do Some Advertisements Get Shared More than Others? Quantifying Facial Expressions to Gain New Insights ," Journal of Advertising Research , 60 (4), 370 – 80.
McKee Robert. (1997), Story: Substance, Structure, Style, and the Principles of Screenwriting. New York : Harper Collins.
Miao Fred , Kozlenkova Irina V. , Wang Haizhong , Xie Tao , Palmatier Robert W.. (2022), " An Emerging Theory of Avatar Marketing ," Journal of Marketing , 86 (1), 67 – 90.
Naik Prasad A. , Mantrala Murali K. , Sawyer Alan. (1998), " Planning Pulsing Media Schedules in the Presence of Dynamic Advertising Quality ," Marketing Science , 17 (2), 214 – 35.
Naik Prasad A. , Raman Kalyan , Winer Russell S.. (2005), " Planning Marketing-Mix Strategies in the Presence of Interaction Effects ," Marketing Science , 24 (1), 25 – 34.
Oblander Elliot Shin , Gupta Sunil , Mela Carl. F. , Winer Russell S. , Lehmann Donald R.. (2020), " The Past, Present, and Future Customer Management ," Marketing Letters , 31 , 125 – 36.
Pechmann Cornelia , Stewart David W.. (1988), " Advertising Repetition: A Critical Review of Wearin and Wearout ," Current Issues and Research in Advertising , 11 (1/2), 285 – 329.
Pugh S. Douglas. (2001), " Service with a Smile: Emotional Contagion in the Service Encounter ," Academy of Management Journal , 44 (5), 1018 – 1027.
Rockwater (2020), "Can the US Replicate China's $63B Livestream Shopping Industry?" (accessed February 7, 2021) , https://www.wearerockwater.com/blog/livestream-selling-china-us.
Scarantino Andrea (2017a), " How to Do Things with Emotional Expressions: The Theory of Affective Pragmatics ," Psychological Inquiry , 28 (2/3), 165 – 85.
Scarantino Andrea (2017b), " Twelve Questions for the Theory of Affective Pragmatics ," Psychological Inquiry , 28 (2/3), 217 – 32.
Scarantino Andrea. (2018), " Emotional Expressions as Speech Act Analogs ," Philosophy of Science , 85 (5), 1038 – 1053.
Sethuraman Raj , Tellis Gerard J. , Briesch Richard. (2011), " How Well Does Advertising Work? Generalizations from Meta-Analysis of Brand Advertising Elasticities ," Journal of Marketing Research , 48 (3), 457 – 71.
Siders Mark A. , George Gerard , Dharwadkar Ravi. (2001), " The Relationship of Internal and External Commitment Foci to Objective Performance Measures ," Academy of Management Journal , 44 (3), 570 – 79.
Stephens Debra Lynn , Hill Ronald Paul , Bergman Karyn. (1996), " Enhancing the Consumer–Product Relationship: Lessons from the QVC Home Shopping Channel ," Journal of Business Research , 37 (3), 193 – 200.
Teixeira Thales , Wedel Michel , Pieters Rik. (2012), " Emotion-Induced Engagement in Internet Video Advertisements ," Journal of Marketing Research , 49 (2), 144 – 59.
Van Kleef Gerben A.. (2009), " How Emotions Regulate Social Life: The Emotions as Social Information (EASI) Model ," Current Directions in Psychological Science , 18 (3), 184 – 88.
Van Kleef Gerben A.. (2016), The Interpersonal Dynamics of Emotion: Toward an Integrative Theory of Emotions as Social Information. Cambridge, UK : Cambridge University Press.
Van Kleef Gerben A. , De Dreu Cartsen K.W. , Manstead Anthony S.R.. (2010), " An Interpersonal Approach to Emotion in Social Decision Making: The Emotions as Social Information (EASI) Model ," Advances in Experimental Social Psychology , 42 , 45 – 94.
Van Kleef Gerben A. , Homan Astrid C. , Beersma Bianca , van Knippenberg Daan , van Knippenberg Barbara , Damen Frederic. (2009), " Searing Sentiment or Cold Calculation? The Effects of Leader Emotional Displays on Team Performance Depend on Follower Epistemic Motivation ," Academy of Management Journal , 52 (3), 562 – 80.
Van Kleef Gerben A. , Van den Berg Helma , Heerdink Marc. W.. (2015), " The Persuasive Power of Emotions: Effects of Emotional Expressions on Attitude Formation and Change ," Journal of Applied Psychology , 100 (4), 1124 – 42.
Viola Paul , Jones Michael. (2001), "Rapid Object Detection Using a Boosted Cascade of Simple Features," paper presented at 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition , Kauai, HI, DOI: 10.1109/CVPR.2001.990517.
Wang Ze , Mao Huifang , Li Yexin Jessica , Liu Fan. (2017), " Smile Big or Not? Effects of Smile Intensity on Perceptions of Warmth and Competence ," Journal of Consumer Research , 43 (5), 787 – 805.
Wang Lu , Restubog Simon , Shao Bo , Lu Vinh , Van Kleef Gerben A.. (2018), " Does Anger Expression Help or Harm Leader Effectiveness? The Role of Competence-Based Versus Integrity-Based Violations and Abusive Supervision ," Academy of Management Journal , 61 (3), 1050 – 1072.
Wedel Michel , Kannan P. K.. (2016), " Marketing Analytics for Data-Rich Environments ," Journal of Marketing , 80 (6), 97 – 121.
Zetlin Minda. (2017), "Do You Smile Too Much? The Answer Is Probably Yes. Here's Why that's Bad," Inc. (May 31), https://www.inc.com/minda-zetlin/do-you-smile-too-much-the-answer-is-probably-yes-heres-why-thats-bad.html.
Zoltners Andris A. , Sinha Prabhakant , Lorimer Sally E.. (2008), " Sales Force Effectiveness: A Framework for Researchers and Practitioners ," Journal of Personal Selling and Sales Management , 28 (2), 115 – 31.
~~~~~~~~
By Neeraj Bharadwaj; Michel Ballings; Prasad A. Naik; Miller Moore and Mustafa Murat Arat
Reported by Author; Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 2- A Practice Perspective on Market Evolution: How Craft and Commercial Coffee Firms Expand Practices and Develop Markets. By: Dolbec, Pierre-Yann; Arsel, Zeynep; Aboelenien, Aya. Journal of Marketing. May2022, p1. DOI: 10.1177/00222429221093624.
Ahead of Print- Database:
- Business Source Complete
Record: 3- Acknowledgments. Journal of Marketing. Mar2022, Vol. 86 Issue 2, p1-4. 4p. DOI: 10.1177/00222429221078918.
- Database:
- Business Source Complete
Acknowledgments
Thank you to the following ad hoc reviewers who reviewed papers for the Journal of Marketing during 2021. The editors greatly appreciate their expert and constructive reviews.
Nidhi Agrawal, University of Washington
Joe Alba, University of Florida
Paulo Albuquerque, INSEAD
Rene Algesheimer, University of Zurich
B.J. Allen, Brigham Young University
Asim Ansari, Columbia University
Gil Appel, George Washington University
Deep Ashana, Duke University
Berk Ataman, Koç University
Ana Babić Rosario, University of Denver
Michelle Barnhart, Oregon State University
Yakov Bart, Northeastern University
Maren Becker, University of Cologne
Silvia Bellezza, Columbia University
Jonathan Berman, London Business School
Len Berry, Texas A&M University
Marco Bertini, ESADE Business School
Yashoda Bhagwat, Texas Christian University
Amit Bhattacharjee, INSEAD
Sean Blair, Georgetown University
Alexander Bleier, Frankfurt School of Finance and Management
Jeff Boichuk, University of Virginia
Raghu Bommaraju, Indian School of Business–Mohali
Samuel Bond, Georgia Institute of Technology
Andrea Bonezzi, New York University
Torsten Bornemann, Goethe University Frankfurt
Eric Boyd, University of Central Florida
Eric Bradlow, University of Pennsylvania
Josko Brakus, University of Leeds
Adam Brasel, Boston College
Els Breugelmans, KU Leuven
Aaron Brough, Utah State University
Alexander Brown, Texas A&M University
Tom Brown, Oklahoma State University
Melanie Brucks, Columbia University
Hernan Bruno, University of Cologne
Eva Buechel, University of Southern California
Jim Burroughs, University of Virginia
Romain Cadario, Erasmus University Rotterdam
Nuno Camacho, Erasmus University Rotterdam
Robin Canniford, University of Melbourne
Francois Carrillat, HEC Montréal
Steve Carson, University of Utah
Noah Castelo, University of Alberta
Lisa Cavanaugh, University of British Columbia
Deepa Chandrasekaran, University of Texas at San Antonio
Yubo Chen, Tsinghua University
Zoey Chen, University of Miami
Yimin Cheng, Monash University
Andrew Ching, Johns Hopkins University
Hana Choi, University of Rochester
Jeonghye Choi, Yonsei School of Business
Junhong Chu, NUS Business School
Luca Cian, University of Virginia
Danny Claro, Insper
Bart Claus, IESEG School of Management
Anatoli Colicev, Universita Bocconi
Yann Cornil, University of British Columbia
Gokcen Coskuner-Balli, Chapman University
Dave, Crockett, Moore School of Business
Tony Cui, University of Minnesota
Amy Dalton, Hong Kong University of Science and Technology
Derick Davis, University of Virginia
Arne De Keyser, EDHEC Business School
Bart de Langhe, ESADE
Yiting Deng, University College London
Berkeley Dietvorst, University of Chicago
Isaac Dinner, University of North Carolina
Beibei Dong, Lehigh University
Grant Donnelly, Ohio State University
Matilda Dorotic, BI Norwegian Business School
Rex Du, University of Texas at Austin
Shuili Du, University of New Hampshire
David Dubois, INSEAD
Adam Duhachek, University of Illinois at Chicago
Christine Eckert, University of Technology Sydney
Alexander Edeling, University of Cologne
Martin Eisend, European University Viadrina
Ryan Elder, Brigham Young University
Jenny Escalas, Vanderbilt University
Zachary Estes, City University of London
Jordan Etkin, Duke University
Theodoros Evgeniou, INSEAD
Peter Fader, University of Pennsylvania
Tingting Fan, The Chinese University of Hong Kong
Er Fang, Lehigh University
David Faro, London Business School
Philip Fernbach, University of Colorado at Boulder
Nathan Fong, Rutgers The State University of New Jersey
Bram Foubert, Maastricht University
Andrey Fradkin, Boston University
Rossella Gambetti, Università Cattolica del Sacro Cuore
Dinesh Gauri, University of Arkansas Fayetteville
Gary Gebhardt, HEC Montréal
Karen Gedenk, University of Hamburg
Katja Gelbrich, Catholic University Eichstätt-Ingolstadt
Sarah Gelper, Eindhoven University of Technology
Andrew Gershoff, University of Texas at Austin
Maggie Geuens, Ghent University
Mrinal Ghosh, University of Arizona
Michael Giebelhausen, Clemson University
Maarten Gijsenberg, University of Groningen
Ayelet Gneezy, University of California San Diego
Kelly Goldsmith, Vanderbilt University
Gabriel Gonzalez, San Diego State University
Shyam Gopinath, Indiana University
Dwayne Gremler, Bowling Green State University
Lauren Grewal, Dartmouth College
Rajdeep Grewal, University of North Carolina
Abhijit Guha, University of South Carolina
Kelley Gullo Wight, Indiana University
Tong Guo, Duke University
Aditya Gupta, Texas State University
Jonne Guyt, University of Amsterdam
Johannes Habel, University of Houston
Hanna Halaburda, New York University
Zachary Hall, Texas Christian University
Ryan Hamilton, Emory University
Kyuhong Han, Korea University
Mike Hanssens, University of California Los Angeles
David Hardisty, University of British Columbia
Jochen Hartmann, University of Hamburg
Gerald Häubl, University of Alberta
Mark Heitmann, University of Hamburg
Conor Henderson, University of Oregon
Dennis Herhausen, VU Amsterdam
Manuel Hermosilla, Johns Hopkins University
Christian Hildebrand, University of St. Gallen
JoAndrea Hoegg, University of British Columbia
Fei Fei Huang, Hong Kong Polytechnic University
Szu-Chi Huang, Stanford University
Christian Hughes, University of Notre Dame
Sam Hui, University of Houston
Tomas Hult, Michigan State University
Chris Hydock, California Polytechnic State University
Caglar Irmak, University of Miami
Masakazu Ishihara, New York University
Chris Janiszewski, University of Florida
Kamel Jedidi, Columbia University
He (Michael) Jia, University of Hong Kong
Niket Jindal, Indiana University Bloomington
Pranav Jindal, University of North Carolina at Chapel Hill
Leslie John, Harvard University
Eric Johnson, Columbia University
Jeff Johnson, University of Missouri-Kansas City
Mingyu Joo, University of California Riverside
Brett Josephson, George Mason University
Annamma Joy, University of British Columbia
Manish Kacker, McMaster University
Bernadette Kamleitner, Vienna University of Economics and Business
Diwas KC, Emory University
Kevin Keller, Dartmouth College
Kristopher Keller, University of North Carolina at Chapel Hill
Robert Kent, University of Delaware
Mansur Khamitov, Indiana University Bloomington
Sungjin Kim, University of Hawai'i at Manoa
TI Tongil Kim, University of Texas at Dallas
Anne Klesse, RSM Erasmus University
Ceren Kolsarici, Queen's University
Manfred Krafft, University of Münster
Raoul Kübler, University of Münster
V. Kumar, St. John's University
Ann-Kristin Kupfer, University of Muenster
Daniella Kupor, Boston University
Vardit Landsman, Erasmus University Rotterdam
Jan Landwehr, Goethe University Frankfurt
Andreas Lanz, HEC Paris
Kathryn LaTour, Cornell University
Justin Lawrence, Oklahoma State University Stillwater
Jeff Lee, American University
Ju-Yeon Lee, Iowa State University
Saerom Lee, University of Texas at San Antonio
Hongshuang (Alice) Li, Ohio State University
Jing Li, Nanjing University
Krista Li, Indiana University
Xi Li, University of Hong Kong
Yang Li, Cheung Kong Graduate School of Business
Yexin Li, University of Kansas
Noah Lim, National University of Singapore Business School
Hongju Liu, Peking University
Qiang Liu, Purdue University
Wendy Liu, University of California San Diego
Yan Liu, Texas A&M University
Jennifer Logg, Georgetown University
Chiara Longoni, Boston University
Carlos Lourenço, University of Lisbon
Mitch Lovett, University of Rochester
Michael Lowe, Georgia Institute of Technology
Tina Lowrey, HEC Paris
Shasha Lu, Fudan University
Andrea Luangrath, University of Iowa
Michael Luchs, College of William and Mary
Xueming Luo, Temple University
Nick Lurie, University of Connecticut
Richard Lutz, University of Florida
John Lynch, University of Colorado Boulder
Mike Lynn, Cornell University
Yu Ma, McGill University
Rhiannon, MacDonnell, University of Lethbridge
Andre Maciel, University of Nebraska–Lincoln
Shilpa Madan, Virginia Tech University
Durairaj Maheswaran, New York University
Ashwin Malshe, University of Texas at San Antonio
Alan Malter, University of Illinois at Chicago
Puneet Manchanda, University of Michigan
Murali Mantrala, University of Missouri
Greg Marshall, Rollins College
Charlotte Mason, University of Georgia
Marlys Mason, Oklahoma State University
Daniel McCarthy, Emory University
Ann McGill, University of Chicago
Ofer Mintz, University of Technology Sydney
Arul Mishra, University of Utah
Natalie Mizik, University of Washington
Daniel Mochon, Tulane University
Bhavya Mohan, University of San Francisco
Jakki Mohr, University of Montana
Erik Mooi, University of Melbourne
Jihwan Moon, University of New South Wales
Sangkil Moon, University of North Carolina at Charlotte
Sarah Moore, University of Alberta
Carey Morewedge, Boston University
Vishal Narayan, University of Connecticut
Sriram Narayanan, Michigan State University
Pravin Nath, Clemson University
Martin Natter, University of Zürich
Gideon Nave, University of Pennsylvania
Gergana Nenkov, Boston College
Nico Neumann, University of Melbourne
Sharon Ng, Nanyang Technological University
Hristina Nikolova, Boston College
Charles Noble, University of Tennessee Knoxville
Stephanie Noble, University of Tennessee Knoxville
Erica Okada, Hitotsubashi University
Jenny Olson, Indiana University Bloomington
Davide Orazi, Monash University
Yesim Orhun, University of Michigan
Ernst Osinga, Singapore Management University
Eunho Park, California State University Long Beach
Jeffrey Parker, University of Illinois at Chicago
Adithya Pattabhiramaiah, Georgia Institute of Technology
Joann Peck, University of Wisconsin-Madison
Andrew Perkins, Washington State University
Kirk Plangger, King's College London
Hilke Plassmann, INSEAD
Ruth Pogacar, University of Cincinnati
Evan Polman, University of Wisconsin–Madison
Linda Price, University of Wyoming
Davide Proserpio, University of Southern California
Andrea Prothero, University College Dublin
Marina Puzakova, Lehigh University
Yi Qian, University of British Columbia
William Rand, North Carolina State University
Anita Rao, University of Chicago
Vithala Rao, Cornell University
Thomas Reutterer, WU Vienna University
Daniel Ringel, University of North Carolina at Chapel Hill
Nicole Robitaille, Queen's University
Cristel Russel, Pepperdine University
Linda Salisbury, Boston College
Adriana Samper, Arizona State University
Gulen Sarial-Abi, Copenhagen Business School
Rom Schrift, Indiana University Bloomington
Janet Schwartz, Duke University
Sydney Scott, Washington University in Saint Louis
Seethu Seetharaman, Washington University in St. Louis
Kathleen Seiders, Boston College
Raj Sethuraman, Southern Methodist University
Julio Sevilla, University of Georgia
Franklin Shaddy, University of California Los Angeles
Avni Shah, University of Toronto Scarborough
Marissa Sharif, University of Pennsylvania
Amalesh Sharma, Texas A&M University
Eesha Sharma, San Diego State University
Edlira Shehu, Copenhagen Business School
Savannah Shi, Santa Clara University
Scott Shriver, University of Colorado at Boulder
Suzanne Shu, University of California Los Angeles
Sunil Singh, University of Nebraska–Lincoln
Vishal Singh, New York University
Robert Smith, Tilburg University
Jelena Spanjol, University of Munich
Stephen Spiller, University of California Los Angeles
Joydeep Srivastava, Temple University
Samuel Stäbler, Tilburg University
Rick Staelin, Duke University
Florian Stahl, University of Mannheim
Laurel Steinfield, Bentley University
Lena Steinhoff, University of Rostock
Jacob Suher, Portland State University
Harish Sujan, Tulane University
Yacheng, Sun, Tsinghua University
Sarang Sunder, Texas Christian University
Abby Sussman, University of Chicago
Courtney Szocs, Louisiana State University
Thales Teixeira, Decoupling.co
Artem Timoshenko, Northwestern University
Vilma Todri, Emory University
Gabriela Tonietto, Rutgers University
Rima Touré-Tillery, Northwestern University
Remi Trudel, Boston University
Yanping Tu, University of Florida
Anna Tuchman, Northwestern University
Catherine Tucker, Massachusetts Institute of Technology
Oleg Urminsky, University of Chicago
Demetrios Vakratsas, McGill University
Ana Valenzuela, Baruch College
Francesca Valsesia, University of Southern California
Bram Van den Bergh, Erasmus University Rotterdam
Ellis Van Den Hende, Delft University of Technology
Tom van Laer, University of Sydney
Noah Van Bergen, University of Cincinnati
Eric VanEpps, University of Utah
Rajan Varadarajan, Texas A&M University
Xavi Vidal-Berastain, Brandeis University
Scott Wallace, University of Washington
Melanie Wallendorf, University of Arizona
Echo Wan, University of Hong Kong
Jessie Wang, Miami University
Pengyuan Wang, University of Georgia
Yang Wang, Temple University
Morgan Ward, Emory University
Luc Wathieu, Georgetown University
George Watson, Portland State University
Klaus Wertenbroch, INSEAD
Hauke Wetzel, Massey University
Kim Whitler, University of Virginia
Jaap Wieringa, University of Groningen
Mike Wiles, Arizona State University
Russ Winer, New York University
David Woisetschläger, Technical University of Braunschweig
Kaitlin Woolley, Cornell University
David Wooten, University of Michigan
Stefan Worm, BI Norwegian Business School
Chunhua Wu, University of British Columbia
Freeman Wu, Vanderbilt University
Ying Xie, University of Texas Dallas
Alison Xu, University of Minnesota
Jennifer J. Xu, Bentley University
Adelle Yang, National University of Singapore
Nathan Yang, Cornell University
Yang Yang, University of Florida
Song Yao, Washington University in Saint Louis
Michael Yeomans, Harvard Business School
Gokhan Yildirim, Imperial College Business School
Sunyee Yoon, University at Buffalo
Tae Jung Yoon, Korea Advanced Institute of Science and Technology
Jonathan Zhang, Colorado State University
Jurui Zhang, University of Massachusetts-Boston
Kuangjie Zhang, Nanyang Technological University
Lingling Zhang, University of Maryland
Shunyuan Zhang, Harvard Business School
Ying Zhao, Hong Kong University of Science and Technology
Rongrong Zhou, Hong Kong University of Science and Technology
Meng Zhu, Johns Hopkins University
Ting Zhu, Purdue University
Peter Zubcsek, Tel Aviv University
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 4- An Emerging Theory of Avatar Marketing. By: Miao, Fred; Kozlenkova, Irina V.; Wang, Haizhong; Xie, Tao; Palmatier, Robert W. Journal of Marketing. Jan2022, Vol. 86 Issue 1, p67-90. 24p. 3 Diagrams, 3 Charts. DOI: 10.1177/0022242921996646.
- Database:
- Business Source Complete
An Emerging Theory of Avatar Marketing
Avatars are becoming increasingly popular in contemporary marketing strategies, but their effectiveness for achieving performance outcomes (e.g., purchase likelihood) varies widely in practice. Related academic literature is fragmented, lacking both definitional consistency and conceptual clarity. This article makes three main contributions to avatar theory and managerial practice. First, to address ambiguity with respect to its definition, this study identifies and critically evaluates key conceptual elements of the term avatar, offers a definition derived from this analysis, and provides a typology of avatars' design elements. Second, the proposed 2 × 2 avatar taxonomy suggests that the alignment of an avatar's form realism and behavioral realism, across different contingencies, provides a parsimonious explanation for avatar effectiveness. Third, the authors develop an emerging theory of avatar marketing by triangulating insights from fundamental elements of avatars, a synthesis of extant research, and business practices. This framework integrates key theoretical insights, research propositions, and important managerial implications for this expanding area of marketing strategy. Lastly, the authors outline a research program to test the propositions and insights as well as advance future research.
Keywords: avatar; artificial intelligence; behavioral realism; chatbot; form realism; human–computer interaction
Advances in computer technology have supported the proliferation of virtual characters, broadly known as avatars, which we define as digital entities with anthropomorphic appearance, controlled by a human or software, that have an ability to interact. Companies are heavily investing in avatars to engage and serve their customers better, and the use of avatars is predicted to increase by 241% in the travel and hospitality industry and by 187% for consumer goods ([92]). In the banking industry, 87% of companies already use some form of an avatar or plan to implement one within two years ([93]).
Even as the use of avatars continues to rise, their effectiveness varies significantly across firms. For example, Progressive Insurance's avatar, Flo, serves many customers successfully on Facebook Messenger ([13]), but IKEA withdrew its avatar, Anna, from its online website following its unsatisfactory performance ([11]). No guidelines exist for effective design or use of avatars ([10]), and academic research often lags behind business practices. In addition, it is difficult to integrate extant research on avatars to establish a strong foundation because the literature in this domain is very fragmented and lacks definitional and conceptual precision, as is typical for an emerging research area. With this article, we aim to provide an integrated theoretical framework of avatars by establishing definitional and conceptual clarity, synthetizing academic research and business practices, and offering propositions that include both managerial insights and future research opportunities.
This article aims to make three main contributions to avatar theory and managerial practice. First, because extant research remains ambiguous with respect to defining and creating a taxonomy of avatars, it also remains difficult for researchers to compare empirical results and draw meaningful conclusions across studies. We offer an overview, in which we summarize the various ways the term avatar has been defined, identify and critically evaluate key conceptual elements of these definitions, and propose a definition on this basis. Then, drawing on this definition, we present a typology of avatar design to isolate elements that academics and managers can leverage to ensure avatars' effectiveness for achieving specific goals (e.g., providing standard vs. personalized solutions). This typology represents an overall organizing framework for thinking about the uses of avatars, making relevant design/implementation decisions, and identifying research gaps.
Second, applying our proposed avatar typology, we synthesize findings from prior literature and business practices to produce a 2 × 2 taxonomy composed of two dimensions: avatars' form realism and behavioral realism. This taxonomy enables us to generate specific research propositions to advance our understanding of and improve effectiveness of avatars in marketing. The level of alignment between an avatar's form and behavioral realism across different contingencies provides a parsimonious explanation for when an avatar is most effective, which we highlight with examples of successful and failed avatars in business practices. In particular, avatars with high form realism but low behavioral realism, which we call superficial avatars, can hinder customer experiences in high-risk transactions (e.g., stock purchases) because the misalignment between an anthropomorphic appearance and low intelligence leads to negative disconfirmations of expectations (e.g., Nordnet's Amelia exhibits realistic human appearance but does not offer good stock tips). In contrast, an intelligent unrealistic avatar (i.e., low form realism, high behavioral realism) may produce positive disconfirmations in socially complex interactions that require exchanges of sensitive personal information (e.g., Ellie, an online avatar, helps assess people's depression and posttraumatic stress disorder [PTSD] symptoms); because the avatar's unrealistic human appearance cannot be mistaken for a real human, people are more likely to provide responses that are free from a social desirability bias.
Third, by triangulating insights from the avatar fundamentals and our synthesis of the extant research and business practices, we develop an integrative framework of avatar performance and offer key theoretical insights, research propositions, and managerial implications for this expanding area of marketing strategy. This framework sheds new light on the underlying mechanisms of effective avatar design and implementation strategies, as well as potential contingencies to these effects. In turn, we offer managerial guidelines regarding avatar deployment, specific decision rules, and insights into how and why specific avatar strategies may be effective. This foundation for a contemporary theory of avatars may spur future research; the identified academic–practice gaps and propositions in particular point to promising research directions.
The popularity of avatars is fueled by two macroenvironmental factors. First, advancements in computer/digital technologies (e.g., artificial intelligence [AI]) enabled the development of more complex avatars, and they often appear in three-dimensional (3D) forms, imbued with seemingly distinctive personalities, appearances, and behavioral patterns, and are overall more appealing than the previous, simple versions ([ 3]; [40]). Second, increase in the use of avatars reflects the growing importance of online service experiences, such as education, gaming, banking, and shopping ([40]; [56]), which firms want to make as convenient and hassle-free for customers as possible ([58]). For example, online customers often express frustration when they cannot find relevant information on a website quickly and easily; avatars can effectively and efficiently provide a solution to this issue. Today's customers also expect faster communication from companies, but immediate responses tend to be difficult and expensive for firms to achieve through traditional channels (e.g., face-to-face, telephone) ([54]). Shopping in the online environment also reduces customers' sense of social interaction and personal consultation, a concern that avatars can help address ([45]). Finally, recent technology developments enable avatars to offer expanded benefits to firms, in that "avatars capable of having complex and interactive conversations with customers will exponentially increase the amount of data businesses can access. Avatars can potentially know if [customers] were bored or happy in real time and know the exact moment someone became disengaged" ([37]).
Although we can readily list the benefits of avatars, no strong consensus exists regarding their precise definition (Table 1). Furthermore, academics have used multiple terms interchangeably to refer to avatars, such as automated shopping assistants ([ 4]), chatbots ([44]), virtual customer service agents ([96]), embodied conversational agents ([ 7]; [59]; [86]), or virtual/digital assistants ([18]; [38]). The ambiguity surrounding the definition of avatars makes it difficult for researchers to compare empirical results or draw meaningful conclusions across studies ([74]). To advance scientific knowledge, we need a precise definition that clearly delineates the boundaries of the construct. In this section, we review various ways avatars have been defined, identify and critically evaluate some key definitional elements, and offer a new definition derived from this analysis.
Graph
Table 1. Avatar Definitional Elements in Empirical Research.
| Illustrative Research | Labels/Aliases | Definitions | Definitional Element | Avatar by Our Definition? |
|---|
| Digital | Anthropomorphic Appearance | Interactivity | Controlling Entity |
|---|
| Percentage of element | 78% | 70% | 78% | 90% | 51% |
| Ho, Hancock, and Miner (2018) | Chatbot | "Computer programs that can simulate human-human conversation" (p. 712) | ✓ | | ✓ | Human | No |
| Holzwarth, Janiszewski, and Neumann (2006) | Avatar | "General graphic representations that are personified by means of computer technology" (p. 20) | ✓ | ✓ | ✓ | Software | Yes |
| Jin (2009) | Avatar | "Artificial, computer-animated representations of human interlocutors" or "pictorial representations of humans in a chat environment" (p. 234) | ✓ | ✓ | | Software | No |
| Kang and Watt (2013) | Avatar | "Digital models of people that either look or behave like the people they represent." (p. 1170) | ✓ | ✓ | ✓ | Human | Yes |
| Kim, Chen, and Zhang (2016) | Anthropomorphized helper / digital assistant | Entities that "are often imbued with humanlike features and characteristics" (p. 283) | ✓ | | ✓ | Software | No |
| Köhler, Rohm, and De Ruyter (2011) | Socialization agent/online agent | "Computer mediated personas that possess the capability to involve customers in rich interactive conversations, rather than discrete, basic exchanges, and that have the ability to apply past interaction content to current interactions" (p. 96) | | | ✓ | Software | No |
| Nunamaker et al. (2011) | Embodied conversational agent | "Virtual, three-dimensional human likenesses that are displayed on computer screens...and interact with people through natural speech" (p. 21) | ✓ | ✓ | ✓ | Software | Yes |
| Schuetzler et al. (2018) | Conversational agent | "Systems that mimic human-to-human communication using natural language processing, machine learning, and/or artificial intelligence" (p. 94) | ✓ | ✓ | ✓ | Software | Yes |
| Sivaramakrishnan, Wan, and Tang (2010) | Anthropomorphic information agent | "A humanlike chatbot that acts as an interactive online information provider" (p. 60) | ✓ | | ✓ | Software | No |
| Touré-Tillery and McGill (2015) | Anthropomorphized agent (partial human) | "Nonhuman entities that deliver message content across a variety of media (e.g., print, online, television), are typically imbued with various combinations of human characteristics, such as human form (e.g., human-like faces, arms, and legs), and the apparent ability to speak and think" (p. 94) | ✓ | | | Software | No |
| Verhagen et al. (2014) | Virtual customer service agent | "Computer-generated characters that are able to interact with customers and simulate behavior of human company representatives through artificial intelligence" (p. 530) | ✓ | ✓ | ✓ | Software | Yes |
| Wang et al. (2007) | Virtual character | "Avatar with some type of combination of 4 online social cues: language, human voice, interactivity, and social role" (pp. 143–44) | ✓ | ✓ | ✓ | Software | Yes |
| This research | Avatar | Digital entities with anthropomorphic appearance, controlled by a human or software, that have an ability to interact | ✓ | ✓ | ✓ | Software or human | |
One aspect on which there is no consensus when it comes to defining avatars is whether avatars need to have an anthropomorphic appearance ([74]). Anthropomorphism refers to "the extent to which an image looks human" ([72], p. 154). In prior academic research, 70% of articles identify an anthropomorphic or humanlike appearance as a necessary condition of a conceptual definition of an avatar. This element is important because the degree to which an avatar is anthropomorphic provides cues of its social presence ([69]). Research shows that the more anthropomorphic an avatar is perceived to be, the more credible and competent it seems ([99]), such that "a person may be represented by a highly accurate and lifelike avatar of a fir tree. Although this avatar is realistic, other users may be less likely to attribute social potential to it—and less likely to communicate with it—because it is not anthropomorphic" ([74], p. 37). Research shows that how anthropomorphic we perceive something to be impacts our expectations of certain behaviors and our willingness to interact; people treat something with a human appearance differently than they do inanimate objects ([36]). For example, Neytiri in the film Avatar is not a human, but because she has an anthropomorphic appearance, other characters interact with her the same way they would with a human.
Knowledge about how to deal with other humans generally is learned early in life and is more detailed and readily accessible in people's memory than knowledge about how to interact with inanimate objects ([30]). According to the computers-as-social-actors (CASA) paradigm ([68]; [70]; [84]), people tend to treat computer technology that exhibits humanlike characteristics as a social actor and apply the same social rules to it during interactions, despite being fully aware that they are dealing with a machine ([45]). The presence of an anthropomorphic appearance triggers people's simplistic social scripts (e.g., politeness, reciprocity), which in turn induce cognitive, affective, and social responses during interactions with technology ([98]). Thus, we regard an anthropomorphic appearance as an important, required element of the conceptual definition of digital avatars, because people interact differently with something they perceive as more "human." This requirement, thus, would exclude inanimate objects and brands as well as voice-only digital assistants that lack an anthropomorphic appearance.
Interactivity refers to "the extent to which individuals perceive that the communication allows them to feel in control as if they can communicate synchronously and reciprocally with the communicator" ([18], p. 317). In defining interactivity as another critical requirement for digital avatars, we refer specifically to the ability to engage in two-way interactions, which may be verbal (voice) or nonverbal (text, animation). Prior research has established three dimensions of interactivity: ( 1) the user's active control, or ability to participate and influence communication; ( 2) bilateral interactions; and ( 3) synchronicity ([32]; [61]). Approximately 78% of the research we reviewed include interactivity as one of the elements of the conceptual definition of avatars. For example, in defining avatars as "virtual characters that can be used as company representatives in online stores," [60], p. 2) allow them to be noninteractive and capable of only one-way conversation, such as welcoming users, introducing the company, or describing available products. Some other researchers similarly do not consider interactivity as a necessary element of digital avatars (e.g., [48]). Yet most researchers focus on interactive avatars and find that they can increase customers' satisfaction with a website or product, credibility, or patronage intentions ([17]; [45]).
However, designing a truly interactive avatar that can engage in synchronous communication is not an easy task: "Natural language dialogue in chatbots suggests a low threshold for users to access data and services. However, whereas conversational interfaces are truly intuitive when applied to interactions between people, conversations between humans and automated conversational agents are more challenging" ([11], p. 41). The interactivity requirement would exclude entities such as "self-avatars" in clothing stores, which do not offer bidirectional communication, as well as any instances of asynchronous content, such as a lecture delivered as a prerecorded video of the instructor, or a standard greeting from a chatbot that cannot offer personalized interactions with each user. However, when there is true bidirectional interactivity, it can satisfy customers' hedonic (e.g., having fun while shopping on a website) and utilitarian (e.g., efficiently finding a solution to a problem on a website) needs ([60]). Thus, we include it as a requirement for digital avatars.
Researchers also have different perspectives on the controlling entity, which refers to whether the control over an avatar involves a human operator or an automated computer program ([74]). Some researchers make this distinction explicit and refer to anything controlled by technology as an agent or bot, while referring to anything controlled by humans as an avatar ([74]). However, in business practices, due to cost considerations, digital avatars appear almost exclusively enabled by AI (e.g., Sophie, the Air New Zealand customer service rep). Yet we have no theoretical reason to limit the conceptual definition of avatars to only those which are enabled by AI, because it seems that consumers want a perception of an avatar having some level of intelligence but often cannot tell precisely who or what controls it ([57]), as there are typically no solid clues available. According to the modality–agency–interactivity–navigability model and its applications in virtual environments, if agency cues are present in an interface, they influence users' perceptions by prompting their cognitive heuristics about the nature and content of the interaction ([91]). Users' perceptions and behaviors thus differ if they learn that they are interacting with an AI-backed avatar versus one controlled by a person, reflecting the different heuristics that are evoked by machine versus human counterparts ([36]; [42]). Thus, "identity cues suggesting that the user is chatting with a human agent or machine agent can trigger human or machine heuristics respectively and accordingly affect the criteria by which they evaluate the quality of the interaction" ([42], p. 305).
In summary, we define avatars as digital entities with anthropomorphic appearance, controlled by a human or software, that are able to interact. Among the academic papers we reviewed, about half (51%) include all of these elements in their conceptual definitions of avatars; the inclusion rates for each specific definitional element vary between 70% and 90%.
Drawing on this derived definition of avatars, we propose a typology of avatar design. This typology allows academics and managers to isolate elements that make an avatar more or less effective for specific goals, such as providing product information, answering customers' process questions, and so forth. Furthermore, this typology provides an overall organizing framework for thinking about, making design/implementation decisions regarding, and researching avatars.
Different design elements cause avatars to vary in their visual appearances and behaviors during interactions with humans. All of the design elements affect avatars' form realism and behavioral realism. Form realism refers to the extent to which the avatar's shape appears human, while behavioral realism captures the degree to which it behaves as a human would in the physical world ([ 5]; [ 9]; [36]). Some researchers argue that behavioral realism is more important than form realism ([ 9]), but both form and behavioral realism are associated with greater avatar usefulness in most contexts ([39]; [51]; [104]). Figure 1 provides an overview of all the design elements, with examples, that managers can use to understand and influence the degree of avatars' form and behavioral realism.
Graph: Figure 1. Typology of avatar design.aAlthough these specific avatar examples are used to illustrate a specific characteristic, they also satisfy all of the other requirements of our conceptual definition of avatars.b[15], p. 9.Notes: For live links to the URLs noted herein, see the Supplemental Material link.
Higher form realism may lead users to develop social expectations for their subsequent interactions with avatars ([73]). Managers can impact the degree of form realism of an avatar through the design elements such as spatial dimensions (2D vs. 3D avatars), ability to have movement in the face or body (visually static vs. dynamic avatars), and other characteristics that enhance the perception of "humanness" of avatars, such as signals of gender, race, age, or names.
Avatars can be 2D or 3D. In the sample of articles we reviewed, 52% focused on 2D avatars and 48% on 3D avatars. Research indicates that 3D avatars are perceived as more compelling and impactful, relative to 2D versions ([ 5]; [36]; [81]).
Both technological advancements and customer expectations have prompted the development of more realistic, visually dynamic avatars that can move their bodies and faces ([106]). In our sample of reviewed studies, 38% of studies examined static avatars, while 62% considered visually dynamic avatars. For example, Amelia is a visually dynamic avatar that is capable of facial expressions and movement and has been used, with modifications, in various industries such as banking, insurance, and health care ([46]). Visually dynamic avatars with ability for facial expressions can convey emotions, which is especially helpful for customers from different cultural backgrounds: "avatars with high intensity expression and dynamics allow both the local and global audiences to achieve approximately equal levels in subject identification and emotion perception" ([106], p. 21). Thus, visually dynamic avatars may be more effective for global corporations. Additional evidence indicates that the greater an avatar's ability to exhibit facial expressions, the less perceived human agency is required to exert social influence ([ 5]).
To enhance form realism, avatars can be designed to include additional "human" elements, such as an identifiable name, gender, race, and age. In the studies we reviewed, out of the aforementioned set of characteristics, gender and age are the most commonly studied ones, followed by name and race. For example, the Air New Zealand virtual customer service avatar is female, named Sophie. Research shows that characteristics such as gender can increase the effectiveness of avatars ([71]).
Avatars' behavioral realism can facilitate more natural interactions with users ([ 9]), and managers can manipulate the degree of avatars' behavioral realism using design elements associated with avatars' interactivity and controlling entity. Specifically, relevant design elements that managers can use to impact avatar interactivity are communication modality (avatars' ability to communicate verbally, nonverbally, or through a combination of both), response type (whether avatars' responses are scripted or natural), and presence of social content (whether avatars can engage in interactions about social and personal matters in addition to task-oriented communications). In terms of controlling entity, avatars can be controlled by a computer program or algorithm or a human, with the latter, predictably, increasing avatars' behavioral realism.
Avatars differ in the modalities of communication they use. Nonverbal avatar communication can be represented by text (speech-to-text avatars), gestures, or facial expressions; verbal avatars communicate via speech; and both nonverbal and verbal avatars can communicate using a combination of these modalities. The latter category would be the highest in behavioral realism from a communication modality perspective. Research investigating the communication modality of avatars has focused primarily on avatars that are capable of both verbal and nonverbal interactions, accounting for 68% of reviewed articles, followed by research on non-verbal avatars in 26% of articles, with verbal avatars attracting the least academic attention (6%).
In addition, managers might increase behavioral realism by enabling avatars to recognize the nonverbal behaviors of users, such as their facial expressions, prompting more appropriate responses. For example, Microsoft's XiaoIce can interpret users' photos and make relevant inferences and comments ([28]). However, even with significant advances in AI, creating an avatar capable of correctly identifying and responding to users' various emotions and contexts remains a challenge because "large, interpersonal variability exists in how people express emotions. Humans also have diverse preferences for how an agent [avatar] responds to them" ([64], p. 76).
Managers can design avatars with an ability to converse in a way that feels natural to users. Previous research has mostly (60% of articles) focused on avatars that are capable of merely selecting a response from a set of preexisting, predetermined, scripted responses. For example, HSBC's virtual assistant, Amy, currently is able to select and provide users with only predetermined responses about a limited number of the bank's products. Avatars with a capability for natural responses instead can have a "relatively free-flowing conversation, using accepted vocabulary and grammar, and with the ability to track the context of the conversation and make appropriate responses" ([15], p. 9). For example, the skincare company SK-II's YUMI understands users expressing themselves in their own words and responds in an organic, conversational manner. The ability to have a conversation that feels natural is also highly correlated with perceived agency type; avatars controlled by humans would have a natural response, whereas software-controlled avatars tend to rely on scripted responses.
Another design element that can increase avatars' interactivity is their ability to provide some social content during interactions with users, as opposed to purely task-oriented communication (e.g., providing product information). For example, Microsoft's XiaoIce is an AI assistant that also attempts to function like a friend, checking on users after a relationship breakup or asking about the physical recovery of a user who posted a photo of a bruised leg. Since its launch in China in 2014, XiaoIce has gained great popularity, due to its emotional intelligence: "The real key takeaway is that we've focused on emotional intelligence.... We call this an empathetic computing framework, [designed to] have conversations with humans naturally, which can build a social and emotional connection. It's a good friend. As a result, they can better participate and help out in human society" ([28]). In the research we reviewed, 34% of avatars could offer some social content during their interactions.
Research shows that consumers interacting with an avatar that they perceive to be controlled by a human behave differently from customers who believe the avatar is controlled by software. According to a meta-analysis, avatars controlled by humans elicit more presence and a stronger social influence than do computer-controlled avatars ([36]). Therefore, for firms that rely on software-controlled avatars (the majority of avatars in practice), reinforcing "human" elements can be very effective. They can leverage the design elements we have discussed, such as more natural speech programmed for software-controlled avatars, non-verbal behaviors and emotions, and the ability to provide social content. However, perceptions of human agency might not be desirable in all settings, as research shows that people perform worse on certain tasks when they recognize that they are interacting with a human-controlled avatar rather than software, due to social inhibition, social desirability bias, and perceptions of reduced autonomy ([56]; [105]). For example, avatars have become popular in health care, and when it comes to disclosing sensitive information such as drinking habits, users are more comfortable revealing information if they perceive less human agency ([86]).
In the proposed typology, all of the design elements serve to increase or decrease the form and behavioral realism of avatars. Most prior research investigates only a narrow set of design elements in piecemeal fashion, independent of other elements, or at an aggregate level, without considering the granularity or interactions of specific elements. In the following sections, we synthesize academic literature and business practices related to avatars to provide insights and identify research gaps. We also derive propositions to help advance theory, inform business practices, and guide future research.
Because research on avatars is diverse and cross-disciplinary in nature, our goal is to provide a representative rather than exhaustive literature review, covering a variety of research disciplines and empirical settings. Using "avatar" and related terms[ 5] as keywords, we searched article titles and abstracts in electronic databases (Academic Search Complete, Business Source Complete, Science Direct, and Google Scholar) to find studies published during the 1990–2020 period. To identify research involving avatars that are consistent with our definition, we first excluded articles that did not provide detailed information about avatar definitional elements or scope. In addition, we excluded research on topics that fall outside our definitional boundary (e.g., consumers' self-avatars). Purely technical articles (e.g., programming of avatar) and studies that do not address real-time consumer–avatar interactions were removed as well. To strike a good balance between research diversity and quality, we sampled only reputable journals across various disciplines (i.e., impact factor of at least 3).[ 6] In total, we compiled 98 empirical research articles (the full list of reviewed research is available in the Web Appendix). Table 2 provides a summary of select illustrative research.
Graph
Table 2. Avatars in Empirical Research.
| Illustrative Research | Original Label | Context | Theoretical Perspective | Mediator Variables | Moderator Variables | Key Findings |
|---|
| Al-Natour, Benbasat, and Cenfetelli (2011) | Automated shopping assistant | Online shopping for a laptop computer | Computers as social actors (CASA) framework; similarity-attraction hypothesis | Perceived decision process similarity | None | Perceived decision process similarity mediates the effect of perceived personality similarity on several beliefs (enjoyment, social presence, trust, ease of use, and usefulness). |
| Bickmore et al. (2016) | Embodied conversational agent | Cancer patients identifying and learning about clinical trials on the internet | Not specified | None | None | Patients were more satisfied with the conversational agent compared to the conventional web form-based interface, and patients with low health literacy had a higher success rate in finding relevant trials. |
| Brave, Nass, and Hutchinson (2005) | Embodied computer agent | Casino-style blackjack game | CASA framework | None | None | Empathic emotion of the agent leads to greater user-rated likeability, trustworthiness, perceived caring, and perceived support. |
| Chattaraman et al. (2019) | Digital assistant | Online purchase of athletic shoes by older consumers | Social response theory | Trust in online store; information overload; perceived self-efficacy; ease of use; usefulness | Internet competency | Users' internet competency interacts with the digital assistant's conversational style (social- vs. task-oriented) in affecting social, functional, and behavioral intention outcomes. |
| Chattaraman, Kwon, and Gilbert (2012) | Virtual agent | Online purchase of apparel by older consumers | Social response theory; CASA framework | Perceived social support; trust in online store; perceived risk | None | Virtual agents can increase older users' patronage intentions by enhancing perceived social support and trust in the online store while reducing perceived risk. |
| Derrick and Ligon (2014) | Embodied conversational agent | An avatar-based screening checkpoint experiment | CASA framework | None | Gender | Self-promoting agents are perceived as more powerful, more trustworthy, and more expert, whereas ingratiating agents are perceived as more attractive. Ingratiation impression management techniques are viewed less (more) favorably by females (males) than self-promotion techniques. |
| D'Mello, Graesser, and King (2010) | AutoTutor | Computer literacy learning with a fully automated computer tutor | Social agency theory | None | None | Students who interacted with the AutoTutor through a spoken dialogue used more cognitive resources and completed more problems than students who had to type. |
| Go and Sundar (2019) | Chatbot | Online digital camera purchase | CASA framework | Social presence; homophily | Anthropomorphic visual cues; agency | Chatbot's message interactivity has positive effects on customers' attitude toward the website and return intention mediated by perceived social presence and homophily; anthropomorphic visual cues and agency moderate these effects. |
| Holzwarth, Janiszewski, and Neumann (2006) | Avatar | Online purchase of shoes that are customizable via online consultation | Social response theory | Entertainment value; information value; likeability of avatar; credibility of avatar | Product involvement | Use of an avatar sales agent increases satisfaction with the retailer, attitude toward the product, and purchase intentions, mediated by perceived entertainment and information value; an attractive avatar is more effective at moderate levels of product involvement, mediated by likeability of the avatar, whereas an expert avatar is more effective at high levels of product involvement, mediated by credibility of the avatar. |
| Keeling, McGoldrick, and Beatty (2010) | Avatar | Online experiments of retail websites selling books/CDs and travel insurance | CASA framework | Trust perception | Goods/services high in credence vs. search qualities | Avatars' social orientation and task orientation increase customers' trust perception, which subsequently has a positive effect on patronage intention. Effects of task- (social-) oriented communications are stronger for search (credence) goods/services. |
| Kim, Hong, and Cho (2007) | Intelligent conversational agent | Online electronic product (e.g., cellular phone) information search | Not specified | None | None | Agents capable of the probabilistic inference and the semantic inference show superior performance in providing suitable responses to user inquiries with only a few interactions. |
| Lee and Choi (2017) | Conversational agent | Interactive movie recommendation system | CASA framework; media equation theory; uncertainty reduction theory | Intimacy; trust; interactional enjoyment | None | Self-disclosure and reciprocity of the conversational agent have positive impacts on user satisfaction and intentions to use, mediated by intimacy, trust, and interactional enjoyment. |
| Mimoun and Poncin (2015) | Embodied conversational agent | Furniture purchase with Anna on IKEA's website | Technology acceptance model | Utilitarian value;hedonic value | None | Anna increased consumers' satisfaction and behavioral intentions through utilitarian and hedonic value. |
| Nunamaker et al. (2011) | Embodied conversational agent | Automated kiosk-based interviews | Not specified | None | None | Male-embodied agents are perceived as more powerful, more trustworthy, and more expert than female ones; however, the latter are more likeable. Avatars with neutral expressions are perceived as more powerful, whereas smiling avatars are more likeable. |
| Von der Pütten et al. (2010) | Embodied conversational agent | Interactions involving personal questions | Threshold model of social influence; Ethopoeia concept | None | None | Beliefs about whether a participant is interacting with a human-controlled or a computer-controlled agent lead to almost no differences in the evaluation of the virtual character or its behavioral reactions; higher behavioral realism affected both. |
| Qiu and Benbasat (2009) | Anthropomorphic interface agent | Online recommendation system for complex and attribute-intensive digital cameras | CASA framework; social agency theory | Social presence | None | Humanoid appearance and human voice-based communication of avatars significantly increase participants' perceived social presence, which has a positive effect on trust, perceived usefulness, perceived enjoyment, and the decision to use the avatar as a decision aid. |
| Schuetzler et al. (2018) | Conversational agent | Responses to sensitive questions to a person vs. a conversational agent vs. online survey | Self-disclosure; social desirability; social presence theories | None | None | Conversational agents with better conversational abilities prompt more socially desirable responses from participants, with no significant effect found for embodiment. |
| Verhagen et al. (2014) | Virtual customer service agent | Inquiries about online mobile phone service | Social response; implicit personality; primitive emotional contagion; social interaction theories | Social presence; personalization | Communication style (socially vs. task-oriented); anthropomorphism | Friendliness and expertise have positive effects on participants' service encounter satisfaction mediated by social presence and personalization; the effect of friendliness on personalization is stronger for socially oriented agents than for task-oriented agents, as is the effect of expertise on social presence. |
| Wang et al. (2007) | Virtual character | Online travel information service | Social response theory; stimulus-organism-response framework; cognitive mediation theory | Arousal; pleasure; flow; hedonic and utilitarian values | Product involvement | Social cues from interacting with the avatar increase perceptions of website socialness, feelings of arousal, pleasure, and flow, leading to greater hedonic and utilitarian values, which then increase patronage intentions. The effect of arousal on pleasure is stronger when product involvement is high; the influence of arousal on hedonic value is stronger for women. Flow does not lead to pleasure for older consumers, and pleasure has a much weaker impact on utilitarian value for those consumers. |
| Yokotani, Takagi, and Wakashima (2018) | Virtual agent | Mental health interviews | Threshold model of social influence | None | None | Participants revealed more sex-related symptoms to the virtual agent than to a human expert, whereas they disclosed mood and anxiety symptoms more often to the human expert than to the virtual agent. |
The overarching theoretical framework that guides empirical studies of human–avatar interactions is social response theory, which is sometimes referred to as the CASA paradigm ([68]; [70]; [84]). It suggests that anthropomorphic characteristics of avatars elicit consumers' socialness perceptions, often in an automatic, spontaneous, mindless process that induces varying degrees of cognitive, affective, and social responses to avatars ([ 4]; [45]; [96]; [98]). Although the theory suggests that an anthropomorphic appearance positively affects customer outcomes, empirical results indicate some mixed effects. In various situations, lower or higher levels of form realism appeared more effective, but in other cases, no differences emerged. For example, visually static, cartoonish avatars with very low form realism increased satisfaction with a retailer, attitude toward products, and purchase intentions in some studies ([31]; [45]). However, [83] find that avatars with more realistic, humanlike appearance increased customers' perceptions of social presence, leading to higher usage intentions. [96] found no significant differences between avatars that are low or high in form realism in terms of service satisfaction. Similarly, [86] reported that a more anthropomorphic appearance of an avatar had no effect on participants' disclosure of information about sensitive topics, such as drinking behaviors.
Two factors may help explain these inconsistent effects. First, prior studies have not investigated all of the design elements identified in our typology that help establish avatars' form realism (e.g., visually static vs. dynamic avatars, avatars' age, gender) or how these underlying elements might induce specific effects. By focusing on only a subset of visual characteristics and, thus, failing to account for the totality of the elements that establish form realism of an avatar, these studies may have produced biased estimates. Second, we propose that avatars' form and behavior must be considered simultaneously, because form realism is meaningful only in the context of behavioral realism ([ 5]); however, few studies have done so.
Extant research consistently shows positive effects of greater behavioral realism. For example, [98] report that an avatar's scripted text or spoken communications can enhance customers' hedonic and utilitarian benefits when shopping online, as well as increasing their patronage intentions. Similarly, [59] find that when users interact with an avatar that has a high degree of behavioral realism, trust between the parties is higher. Other studies offer similar results, noting that avatars' behaviors, such as decision-making style ([ 4]) and socially oriented communication ([96]), significantly affect avatar trustworthiness and the overall customer experience ([12]; [18]).
Advancements in AI technology have enabled avatars to exhibit higher levels of cognitive and emotional intelligence. For example, they can engage in autonomous conversations by analyzing and responding to customers' requests in real time, thereby significantly increasing customers' trust in them ([64]). Using avatar interviewers equipped with video, audio sensors, and advanced analytical software, [75] demonstrate how intelligent avatars can detect, interpret, and respond to human interviewees' emotions, cognitive effort, and potential deceptions. [97] examine the effects of an avatar that can collect and analyze a human's voice and upper-body movement and coordinate its own responses accordingly. Results show that the avatar's intelligent behavior led to positive evaluations of the avatar, regardless of whether it was controlled by a human or a software.
Even in light of these consistent findings related to avatar behavioral realism, some important research issues remain unresolved. First, few studies have investigated the underlying behavioral realism elements identified in our typology of avatar design (e.g., communication modality, social content, response type) to determine which are most critical or how they might interact with other form or behavioral realism elements. For example, [ 7] found that an avatar nurse incorporating social content in its scripted conversations produced better patient experiences, but [86] reported that a scripted, task-focused avatar interviewer elicits more socially biased responses. Second, a few studies revealed some unexpected negative effects of behavioral realism (e.g., [86]), but research has yet to identify the conditions in which detrimental effects are more likely or design strategies to address them.
Our review of extant literature thus reveals a key limitation: lack of consideration of the alignment between form and behavioral realism of avatars. If the levels of form and behavioral realism are mismatched, the consequences for avatars' effectiveness may be profound and can help explain inconsistent past findings. Yet some misaligned avatars (e.g., the REA avatar is high in behavioral realism but low in form realism) seem equally as effective as well-aligned avatars (e.g., SK-II's skincare advisor YUMI is very high in both behavioral and form realism). However, other misaligned avatars have failed (e.g., Nordnet's Amelia, with high form realism but low behavioral realism). A systematic analysis of avatar effectiveness thus seems warranted and requires identifying different categories of avatars along the form and behavioral realism dimensions as a first step.
We suggest that avatars can be parsimoniously grouped into a 2 × 2 taxonomy, according to their form and behavioral realism (Figure 2). This taxonomy provides a foundation for predicting the success or failure of avatars in business practices and can inform avatar design strategies. We identify four distinct categories of avatars: simplistic, superficial, intelligent unrealistic, and digital human avatars. A simplistic avatar has an unrealistic human appearance (e.g., 2D, visually static, cartoonish image) and engages in low intelligence behaviors (e.g., scripted, only task-specific communication). For example, in the Netherlands, ING Bank uses a 2D, cartoonish-looking avatar, Inge, to provide responses to simple customer inquiries with a set of predetermined answers. In contrast, a superficial avatar has a realistic anthropomorphic appearance (e.g., 3D, visually dynamic, photorealistic image), such as Natwest Bank's Cora, but low behavioral realism, in that it is only able to offer preprogrammed answers to specific questions. An intelligent unrealistic avatar (e.g., REA) is characterized by humanlike cognitive and emotional intelligence but exhibits an unrealistic (e.g., cartoonish) human image. These avatars can engage customers in real-time, complex transactions without being mistaken for real human agents. Finally, a digital human avatar such as SK-II's YUMI is the most advanced category of avatars, characterized by both a highly realistic anthropomorphic appearance and humanlike cognitive and emotional intelligence, and is designed to provide the highest degree of realism during interactions with human users.
Graph: Figure 2. Form realism versus behavioral realism taxonomy.
To advance extant literature, we propose the need to consider the interrelationship of form and behavioral realism. Visual information (i.e., what an avatar looks like) often gets processed automatically and almost immediately, requiring minimum cognitive resources ([65]). This visual appearance then becomes the basis for probabilistic consistency inferences, where consumers form an expectation of some unknown attribute, based on a known attribute with which it is believed to be correlated ([27]). For example, consumers often make inferences about an unfamiliar brand's quality by using price as a signal of quality, in the belief that the two are correlated. Similarly, when an avatar looks more like a human, consumers may expect it to also behave like a human. Thus, the visual characteristics of an avatar may influence consumers' judgments of its behavioral competence, even before an interaction takes place ([73]). A more realistic anthropomorphic appearance suggests a higher level of behavioral realism, leading to a greater expectation that the avatar will behave like a real human might.
- Proposition 1: As the form realism of an avatar increases, so do customers' expectations for its behavioral realism.
Customers will use an avatar's form realism as a frame of reference for forming initial expectations about its behavioral realism. The expected level of behavioral realism will then serve as a benchmark, against which consumers will form comparative judgments of their subsequent experience. According to expectation confirmation theory ([76]), when the actual outcome is worse than expected, consumers experience a negative disconfirmation, leading to decreased overall satisfaction. A better-than-expected outcome instead results in a positive disconfirmation, which increases customers' overall satisfaction ([33]; [100]).
Consistent with this theory, if an avatar's behavioral realism exceeds the consumer's initial expectations, which were based on the avatar's form realism, a positive disconfirmation likely occurs, and the consumer should perceive the avatar as more credible and attractive, as well as feel increased trust or confidence in it ([ 2]). Ellie, an avatar that assesses depression and PTSD symptoms in veterans, serves as a good illustration of a positive disconfirmation: her cartoonish appearance paired with highly intelligent, humanlike behavior has proven highly effective with vulnerable individuals ([85]). Conversely, if the avatar's behavioral competency falls short of the expectations that users formed on the basis of the avatar's form realism, it may lead to a negative disconfirmation and dampen customers' satisfaction ([65]). Nordnet's Amelia exhibited minimal competence in providing customized advice for high-risk transactions (e.g., stock purchase), which proved disappointing. Amelia's realistic anthropomorphic appearance might have led customers to develop high behavioral expectations, which Amelia could not deliver, giving rise to a negative disconfirmation. Therefore, we expect asymmetric effects of misaligned avatar form and behavioral realism.
- Proposition 2: Differences between the avatar's form and behavioral realism have asymmetric effects, such that customers experience positive (negative) disconfirmation when an avatar's behavioral realism is greater (less) than its form realism.
Integration of evidence across multiple research streams suggests that avatars affect performance outcomes (e.g., customers' likelihood to purchase a product) indirectly through customers' cognitive, affective, and social responses, depending on the context ([45]; [59]). Customers form cognitive responses to avatars according to the avatars' informativeness or competence in helping them make well-informed decisions ([45]; [98]). Cognitive trust, or willingness to rely on another entity's help to achieve goals in uncertain situations, is another key dimension of customers' cognitive response ([53]; [63]).
Interactions with avatars can also evoke affective responses in customers, such as by providing them with unique entertainment experiences ([45]). Avatars can deliver pleasurable experiences independent of their ability to facilitate a specific functional goal, such as a shopping task, by offering entertainment and emotional value during the shopping process ([98]). Human–avatar interactions are also social in nature. Avatars can enhance customers' perceived social presence (i.e., the feeling of being with another person) and create feelings of human contact or connection ([18]; [83]). Moreover, the use of an avatar can provide a sense of personalization, so customers receive information that appears tailor-made to their specific needs ([96]; [98]). The CASA framework argues that avatars can also induce feelings of reciprocity in human–computer interactions, which can strengthen perceived rapport with the avatar and enhance the users' social experience ([18]; [59]).
When the levels of an avatar's form realism and behavioral realism are aligned, customers' behavioral expectations tend to be confirmed. This simple confirmation, together with high initial behavioral expectations induced by form realism, may have a strong, positive, additive effect on performance outcomes, such as customers' purchase likelihood ([76]; [90]). We expect that avatars that are aligned in their form and behavioral realism can affect customer performance outcomes through all three types of mediating responses: cognitive, affective, and social. However, when the levels of form and behavioral realism are misaligned, the outcomes might be mediated through different responses. Consider the misalignment that occurs when form realism exceeds behavioral realism. In this situation, customers may find the avatar to be especially entertaining, because its realistic anthropomorphic appearance and characteristics can also serve as hedonic elements that can increase perceived entertainment, which often is intrinsically enjoyable in its own right, regardless of performance outcomes ([24]; [103]). To the extent that perceived enjoyment creates a pleasant mood, the hedonic aspect of high form realism can improve performance outcomes such as impulsive online purchases ([78]). Yet customers also might experience negative disconfirmation, stemming from the disappointment with an avatar's cognitive and social capabilities, resulting in weakened customer performance outcomes.
Alternatively, when misalignment arises because the avatar's behavioral realism exceeds its form realism, the positive effects on customer outcomes might be mediated primarily by cognitive and social responses. Although typically this avatar's lower form realism is unlikely to provide much entertainment for the customer,[ 7] the positive disconfirmation of avatar's behavioral competence may significantly boost customers' confidence in the avatar's overall ability to provide valid information, offer personalized service, or build customer rapport.
- Proposition 3: When an avatar's form realism exceeds its behavioral realism, it has a positive effect on performance outcomes (e.g., purchase likelihood), through customers' (a) affective responses, but a negative effect on performance outcomes through customers' (b) cognitive and (c) social responses.
- Proposition 4: When an avatar's behavioral realism exceeds its form realism, it has a positive effect on performance outcomes (e.g., purchase likelihood) through customers' (a) cognitive and (b) social responses.
As we noted previously, many companies are adopting avatars to humanize their brands with a scalable personalized human touch, but managers lack guidance for how to design these avatars to ensure their effectiveness ([10]; [101]). In this section, we use business examples to illustrate the avatar categories from our taxonomy (Figure 2) and thereby clarify which factors and conditions make them effective. We also generate theoretical insights and managerial implications.
An obvious benefit of using avatars is the firm's improved efficiency and scalable customer service. For example, ING Bank's cartoonish avatar, Marie, answers common debit and credit card questions with preprogrammed information and solutions (www.ing.com). In the Los Angeles Superior Court, an animated cartoon avatar, Gina, which speaks multiple languages, successfully handles 1.2 million new traffic citations a year ([62]). A start-up company called TwentyBN has introduced an animated cartoon sales avatar, Millie, which can understand and answer simple questions while presenting various products ([49]) and seems to be especially effective in promoting low-ticket items, such as sunglasses. Simplistic avatars thus seem most effective in providing hassle-free, convenient options for completing quick, specific tasks (e.g., information inquiries), especially when relatively little risk is involved (e.g., inexpensive online purchases).
The use of superficial avatars in various industries shows more mixed results. Among the successes, HSBC Hong Kong's Amy, a photorealistic avatar that handles routine customer inquiries similar to ING's Marie, was well-received by customers ([93]). A very realistic-looking 3D avatar, Cora of Natwest Bank in the United Kingdom, can answer 200 basic questions, such as how to open an account or complete a mortgage application ([79]). In the insurance industry, Lemonade Insurance's avatar Maya and Progressive's Flo, both very humanlike, are programmed to provide category-specific information and handle simple transactions, such as onboarding customers and giving online quotes ([13]; [82]). However, other superficial avatars have been less effective. The Swedish bank Nordnet had to discontinue its realistic-looking avatar Amelia, presumably due to her failure to provide intelligent stock purchasing advice. At IKEA, the decision to eliminate its avatar Anna stemmed from a recognition that her realistic anthropomorphic appearance led to complex customer questions, which required responses beyond the predetermined set available in the avatar's programming ([11]; [87]). Overall, superficial avatars can entertain customers while enhancing efficiency in low-risk transactions (e.g., bank account information inquiries), but they also can produce detrimental effects for customers seeking high-risk or complex transactions (e.g., financial investments) because these avatars lack the level of intelligence that their realistic anthropomorphic appearance leads users to expect.
This type of avatar is relatively rare but generally successful. For example, the REA avatar has been effective in providing virtual showings of homes for sale; Ellie, an avatar therapist, has been useful in detecting PTSD and depression symptoms in military veterans. With its humanlike intelligence, Ellie can engage in context-appropriate, autonomous conversations and build rapport with subtle, supportive, and sympathetic gestures when listening to a veteran's sensitive story. In turn, veterans disclose significantly more PTSD symptoms to her than to a human therapist ([43]). Thus, intelligent unrealistic avatars seem especially effective for complex relational transactions involving sensitive personal information (e.g., finances, health) as they can provide a sense of nonjudgment because customers recognize that these avatars are not human but are still competent in their tasks.
With advanced digital and computing technologies, pioneer avatar companies such as Soul Machines are breaking new ground for digital human avatars in marketing applications (www.soulmachines.com). For example, skincare brand SK-II uses an incredibly realistic looking and behaving avatar, YUMI, whose AI-powered digital brain enables advanced cognitive and emotional intelligence. In addition, YUMI can recognize users' gestures and features, such as eye color; communicate via speech or text; and deliver customized tips with credible, highly personalized beauty advice ([14]). Another digital human avatar, modeled after Daniel Kalt, the chief economist of UBS investment bank, can forecast financial data and present investment advice to high-wealth customers (www.nanalyze.com). Overall, digital human avatars seem most effective for building long-term customer relationships in contexts that feature substantial complexity or risk (e.g., financial investments), where users prioritize realistic, trustworthy, and personalized service.
Observations from business practices suggest that avatars' effectiveness may be highly contingent on the level of perceived uncertainty users experience during their interactions with avatars. This uncertainty might arise from contextual factors, such as functional risk, financial risk, or price ([25]). Functional risk refers to the concern that the product/service may fall short of performance expectations. Financial risk refers to a possible loss of money, independent of purchase price, due to a poor decision (e.g., stock performance) ([25]). In addition, as the purchase price increases, the need for information about the quality of products or services also becomes more important to manage perceived risk ([102]). Overall, we predict that when customers feel greater uncertainty, they develop heightened expectations that an avatar that offers a realistic anthropomorphic appearance will also have a comparable level of behavioral realism, because they rely on the avatar's informativeness and ability to provide personalized advice to reduce the perceived risks associated with the purchase.
- Proposition 5: The positive effect of an avatar's form realism on customers' behavioral realism expectations is stronger when (a) functional risk is higher, (b) financial risk is higher, and (c) the product is more expensive.
We previously posited that when form realism exceeds behavioral realism, the avatar can have both positive (via customer's affective responses) and negative (via customer's cognitive and social responses) effects on performance outcomes. These mediated effects also might be moderated by perceived uncertainty. When perceived uncertainty is high, the avatar's entertainment value may become less salient to the consumer because an entertaining avatar cannot overcome the perceived risks associated with the purchase. The negative disconfirmation induced by the avatar's lack of behavioral competence should become more salient and detrimental to consumers' cognitive and social responses, resulting in weaker performance outcomes. For example, whereas the realistic, anthropomorphic appearance of HSBC's Amy seems effective because she was programmed to provide basic information to routine customer questions that do not involve high-risk transactions, the very realistic appearance of Nordnet's Amelia could not compensate for her lack of competence to offer stock advice (i.e., high financial risk transactions). Thus, we expect the following moderation effects:
- Proposition 6: When an avatar's form realism exceeds its behavioral realism, its positive effect on customers' affective responses is weakened if ( 1) functional risk is higher, ( 2) financial risk is higher, and ( 3) the product is more expensive.
- Proposition 7: When an avatar's form realism exceeds its behavioral realism, its negative effects on customers' cognitive and social responses are strengthened if ( 1) functional risk is higher, ( 2) financial risk is higher, and ( 3) the product is more expensive.
Conversely, when an avatar's behavioral realism exceeds its form realism, it may have a stronger positive effect on customer performance outcomes under high perceived uncertainty. This is because an avatar's form realism may induce a positive disconfirmation regarding its behavioral competence, which can reassure customers of the usefulness of the information and personalized service provided by the avatar, thereby boosting customers' confidence in their risky purchase decisions. Another pertinent risk in today's world is that of privacy violations ([25]). If an avatar's behavioral realism exceeds its form realism, the avatar provides reassurance that customers are dealing with an intelligent, nonhuman entity that will not judge them, so it should mitigate social unease and embarrassment. The PTSD therapist Ellie works with highly private matters, involving patients disclosing emotionally and psychologically sensitive information. Such situations might evoke greater concerns if patients cannot tell whether they are dealing with a real human therapist behind the screen.
- Proposition 8: When an avatar's behavioral realism exceeds its form realism, its positive effects on customers' cognitive and social responses are stronger if (a) functional risk is higher, (b) financial risk is higher, (c) the product is more expensive, and (d) privacy risk is higher.
As avatars are also increasingly used in mobile apps (e.g., ING's Inge, Progressive's Flo, Bank ABC's Fatema), the choice of mobile or fixed devices as platforms on which the firm decides to provide an avatar also may be pertinent. Compared with fixed devices (e.g., desktops), mobile devices are particularly entertaining as hedonic information technologies (e.g., video games, MP3 players) have made their way into these portable devices ([103]). Moreover, consumers spend more time in online communities when they use mobile rather than fixed devices ([66]), suggesting that mobile devices are the preferred channel for online social experiences. Thus, avatars that are entertaining and capable of establishing personalized social connections with customers may be especially effective on mobile devices.
- Proposition 9: As form realism increases, relative to behavioral realism, the use of mobile devices (compared with fixed devices) strengthens the positive effect of avatars on customers' affective responses.
- Proposition 10: As behavioral realism increases, relative to form realism, the use of mobile devices (compared with fixed devices) strengthens the positive effect of avatars on customers' social responses.
Finally, the effects of customers' cognitive, affective, and social responses on performance outcomes may depend on the consumer relationship phase, which refers to the relational trajectory of a customer–seller exchange ([29]; [77]). During consumer–avatar interactions, three relationship phases are particularly important: exploration, build-up, and maturity ([47]). During the exploration phase, a consumer is primarily concerned with the potential value and benefits of dealing with the seller, so the perceived informativeness of the avatar becomes especially critical. During this initial stage, entertainment or social engagement provided by the avatar might even detract from the consumer's task objectives and compromise customer outcomes. In this stage, avatar design should allow for behavioral competence, so that the avatar can provide accurate, task-specific, customized information rather than focusing on designing highly attractive or socially engaging avatars. Such an approach should produce a positive disconfirmation for customers' cognitive experience. As the relationship proceeds to the build-up phase, the consumer has experienced some benefits from interactions with the firm's avatars, so socialization processes (e.g., building rapport) become more important. Engaging and fulfilling social experiences can help customers develop long-term commitment to the firm. A positive disconfirmation about the avatar's ability to deepen social bonds might prove especially effective. Finally, during the maturity phase, a consumer satisfied with the cognitive and social benefits of interacting with an avatar may focus less on these factors and instead seek more entertainment value, prioritizing affective responses.
- Proposition 11: The effects of customers' cognitive, affective, and social responses on performance outcomes are moderated by the relationship phase, such that (a) the effects of cognitive responses are amplified in the exploration phase but suppressed in the build-up and maturity phases, (b) the effects of affective responses are enhanced in the maturity phase but suppressed in the exploration and build-up phases, and (c) the effects of social responses are strengthened in the build-up phase but weakened in the exploration and maturity phases.
To synthesize these insights, we offer an integrated framework of avatar performance (Figure 3). This framework provides a visual summary of key insights from our review of extant research and business practices, as well as our development of the avatar taxonomy and propositions. With this framework, we work to advance thought in this emerging, contemporary marketing area by integrating three mediation mechanisms (customers' cognitive, affective, and social responses) together with theory-driven moderators to test theory as well as offer managerial implications.
Graph: Figure 3. An integrated framework of avatar performance.
Our integrative analyses of academic research and business practices generate practical implications and research directions for avatar-based marketing, which we group into five key managerially relevant areas: ( 1) when to deploy avatars, ( 2) avatar form realism, ( 3) avatar behavioral realism, ( 4) form–behavioral realism alignment, and ( 5) avatar contingency effects (Table 3).
Graph
Table 3. Managerial Implications and Research Directions for Avatar-Based Marketing.
| Key Issues/Decisions | Implications | Directions for Future Research |
|---|
| Avatar deployment | Avatars can be used to humanize a brand with scalable, cost-effective, responsive (24/7), humanlike interactions.
| How can avatar–human collaborations be optimized?
|
Avatars can be used when the scale of service requirements and customer inquiries overwhelm company employees (e.g., financial, travel, telecom services). The employees can use their new free time for more complex issues.
| When the customer encounters a problem with the avatar, which type of "exit ramp" to a human employee is most effective: avatar-initiated, employee-initiated, or customer-initiated, and when during an interaction should it be deployed?
|
Avatars can be used to enhance customer engagement and relationship building through emotional connection, personalization, and service consistency.
| In the event of an avatar service failure, what service recovery strategies should be employed?
|
Avatars can be used to offer multichannel flexibility based on segment preferences (e.g., mobile social media, company website, dedicated apps).
| To what extent can a successful/failed service recovery experience shape customers' (dis)confirmation of the avatar's effectiveness?
|
| Avatar form realism | Avatars' anthropomorphic appearance is a double-edged sword. A more humanlike appearance appeals to consumers, due to enhanced entertainment value, but it also raises consumers' expectations of the avatar's behavioral competence.
| Which dimension of the avatar's anthropomorphic appearance (i.e., spatial dimension, movement, and other human characteristics) has the greatest impact on consumers' expectations for the avatar's behavioral realism?
|
If behavioral realism falls short of expectations, a negative disconfirmation will be produced, resulting in lower levels of customers' cognitive and social responses to the avatar.
| When might a digital assistant without a visual representation (e.g., Amazon's Alexa) outperform an avatar with an anthropomorphic appearance?
|
Avatars' form realism should not exceed the level of its behavioral competence to avoid unfavorable customer experiences.
| Which avatar form realism elements create the most entertaining avatar experience?
|
| Avatar behavioral realism | An avatar's behavioral competence is a more impactful design factor than its appearance. In case of a budget constraint, more resources should be allocated to improving avatar behavioral competence than to enhancing its visual appearance.
| What is the role of avatar emotional intelligence, relative to its cognitive abilities, in shaping customers' expectations and overall experience?
|
The higher the avatar's behavioral competence relative to its appearance, the more favorable the customer's cognitive and social experiences, due to a positive disconfirmation.
| What corrective actions can be taken to redress a negative disconfirmation stemming from avatar's behavioral realism?
|
As a caveat, high levels of behavioral competence may produce negative effects (e.g., social desirability bias) when the avatar also has a very realistic anthropomorphic appearance.
| How might other types of avatars (e.g., customers' self-avatars) facilitate social media-based marketing campaigns?
|
| Which behavioral realism elements have the greatest impact on customers' expectations?
|
| Form realism–behavioral realism alignment | High form realism induces high behavioral realism expectations, which will then be confirmed.
| What are the unique benefits, challenges, and risks associated with using high-realism avatars in brand campaigns?
|
The additive nature of high expected behavioral realism and its subsequent confirmation produces satisfactory customer experience with the avatar.
| When would avatar virtual influencers (e.g., Lil Miquela) be more effective than human endorsers in brand promotion?
|
Alignment of high form realism–high behavioral realism results in high levels of customers' affective, cognitive, and social experiences, which subsequently increase firm performance.
| How should hyper-realistic avatars be deployed in marketing campaigns?
|
| Avatar contingency effects | Consumers' expectations that the avatar's anthropomorphic appearance reflects a comparable level of behavioral competence will be more pronounced when consumers' perceived uncertainty (e.g., the product's functional performance, financial risk) is high.
| How might customer segmentation strategies (e.g., psychographics, benefits sought) inform effective avatar designs?
|
Behavioral realism should account for more weight in avatar design decisions than form realism, especially when customers' perceived uncertainty is high.
| What form realism–behavioral realism elements are most relevant and impactful, given a specific customer segment profile?
|
When the avatar is designed with a low level of behavioral competence, companies should manage customers' expectations by giving the avatar a less realistic, less humanlike appearance (i.e., simplistic avatar). Avoiding a design with high form realism–low behavioral realism (i.e., superficial avatar) becomes even more important when customers' perceived uncertainty is high.
| How do avatar mediation mechanisms differ across customers in different segments?
|
When the exchange entails privacy concerns (e.g., mental health), an avatar characterized by low form realism–high behavioral realism (i.e., intelligent unrealistic avatar) may be more effective than an avatar with high form realism–high behavioral realism (i.e., digital human avatar), because it reassures customers that they will not be judged and promotes more honest responses.
| In which circumstances will avatars likely distract from, rather than contribute to, customers' experience?
|
Use of mobile devices (e.g., smartphones), compared with fixed devices (e.g., desktops), can enhance avatars' impact on consumers' affective and social experiences.
| What contextual factors determine the relative effectiveness of avatars vs. other digital entities (e.g., anthropomorphized products, brand mascots) in online shopping experience?
|
Avatar designs should account for the consumer relationship phase. During the exploration phase, avatar behavioral realism should focus on providing the best cognitive experience; during the build-up phase, avatar design should be directed at enhancing consumers' social experience (e.g., rapport) to promote commitment; during the maturity phase, emphasizing the entertainment value of the avatar (e.g., funny, attractive) may prove more effective in sustaining the established relationship.
| |
In terms of when and where avatars should be used, avatars seem to be most effective in service-oriented industries (e.g., financial, travel, telecom services), in which the sheer scale of service requirements and customer inquiries can easily overwhelm a company's employees. Avatars can free up employees' time, so that the employees can focus on complex customer needs and offer more value-added services, with greater productivity. Avatars can provide consistent, personalized service and help build emotional connections between the firm and its customers ([22]; [52]). For companies that serve a large portfolio of customers, avatars can also make it feasible to launch a segmented, multichannel strategy. By offering customers opportunities to engage with avatars through different channels (e.g., social media, company websites, dedicated apps), the firm ensures that it meets each customer's unique needs at the right time and in the right place.
After confirming that an avatar should be implemented, the firm should determine the design of avatar's appearance, with the clear recognition that its form realism is a double-edged sword. On the one hand, a more realistic, anthropomorphic appearance appeals to consumers, because it offers greater entertainment value ([78]). On the other hand, it elevates customers' expectations of the avatar's behavioral competence, which is much more difficult and costly to develop. If the avatar's behavioral competence falls short of customers' expectations, they experience a negative disconfirmation, which can decrease their satisfaction. To avoid negative performance outcomes, an avatar's anthropomorphic appearance should not exceed the level of its behavioral competence.
Having decided on the avatar's appearance, companies can design the avatar's behavioral competence. If companies lack the resources to develop high form–high behavioral realism avatars (i.e., digital human avatars), they should allocate more resources to the avatar's behavioral intelligence than to its appearance. A positive disconfirmation of the avatar's cognitive and social competence likely produces an above-average level of customer satisfaction or even customer delight ([34]), which can increase firm performance.
Managers should account for form realism–behavioral realism alignment too. If an avatar has high levels of both form and behavioral realism, customers' high initial expectations about the avatar's behavioral performance will be confirmed. Because customer satisfaction is an additive function of ( 1) initial expectations of the avatar's behavioral competence and ( 2) subsequent confirmation or disconfirmation of this expectation ([76]), the high form realism–high behavioral realism alignment will likely produce high levels of affective, cognitive, and social responses in consumers, as well as better outcomes overall. Additional research is needed to determine the "zone of tolerance" and the precision required when aligning form and behavioral realism of avatars.
Finally, in addition to considering uncertainty factors and media channel choice, design efforts should take the customer relationship phase into account, because the relative effects of customers' cognitive, affective, and social responses differ across relationship stages. For example, during the exploration phase, a positive confirmation regarding an avatar's behavioral realism can ensure good cognitive experiences (e.g., cognitive trust), but during the maturity phase, a more entertaining avatar (e.g., funny, attractive appearance) may prove more effective for sustaining the relationship. Future research that takes a lifecycle approach to avatar design and use could determine avatar effectiveness at each stage and the strategies needed to adapt in accordance with customers' dynamic needs.
Our analysis of avatar design strategies indicates some promising research opportunities. First, propositions derived from our conceptual framework provide opportunities for empirical research, which can be tested using different methods. For example, researchers might collaborate with an avatar design company to manipulate form and behavioral realism in a 2 × 2 full-factorial experiment, consistent with our taxonomy in Figure 2. These results would provide evidence of the effects of form realism on expectations for behavioral realism (P1), as well as (dis)confirmations induced by any (mis)alignment (P2). Moderation tests also might be carried out with lab experiments that allow for manipulations of uncertainty factors (e.g., risk, price, privacy) or channels (e.g., mobile app vs. desktop) (P5–P10). To test the mediation effects, researchers might use a difference-in-differences field experiment to examine the aggregate main effect. Using a low form realism–low behavioral realism avatar as a baseline (control group), researchers could increase form (treatment 1 group) and behavioral (treatment 2 group) realism. Any significant differences in daily sales across the treatment groups and control group, between the pre- and posttreatment periods, would indicate the external validity of the asymmetric effects of form realism–behavioral realism misalignment (P3, P4). To confirm the distinct mediation effects, customers also could be surveyed regarding their cognitive, affective, and social responses, shortly after the treatments, followed by tests of the effects on performance outcomes (e.g., purchase intentions, word of mouth). Alternatively, a lab experiment can be used for differential mediation tests. As for the test of the relationship phase (P11), a cross-sectional survey posted on a retail website that uses an avatar might enable researchers to conduct subgroup analyses or moderated regressions to detect differential effects of mediators across relational phases. A more demanding and rigorous approach would secure panel data from a collaborating retailer that uses avatars and agrees to let the researchers track customers' relational trajectories through longitudinal surveys, while also granting them access to objective customer sales data.
Research opportunities also exist in areas that our framework does not cover. Avatars are designed to enhance the productivity of company employees, rather than replace them altogether, so continued research might investigate how to optimize avatar–human collaborations, especially if problems arise. Various approaches allow customers to switch to a human representative when necessary, using avatar-initiated, employee-initiated, or customer-initiated "exit ramps." These approaches can differ in their nature (proactive vs. reactive) and timing (early vs. late in the interaction). Thus, determining how and when human intervention is introduced could provide significant insights for achieving service recovery and ensuring customers' overall satisfaction with and commitment to the firm. In this article, we have highlighted the influence of an avatar's form realism on the overall customer experience. Additional research should uncover which specific dimensions of the avatar's anthropomorphic appearance exert the strongest impacts on customers' behavioral realism expectations, which form realism elements create the most entertaining avatar experience, or when a digital assistant without an anthropomorphic appearance (e.g., Amazon's Alexa) would perform better than an avatar. For example, if the avatar's form realism exceeds its behavioral realism, which subset of behavioral elements is most likely to lead to customers' expectation disconfirmation? What corrective actions would be effective in addressing negative disconfirmations? Research also could delve deeper into the effects of avatars' emotional intelligence, relative to their cognitive intelligence, in shaping customers' expectancy (dis)confirmations and overall experience.
Another type of avatar that is growing in popularity is customers' self-avatars, created with virtual model technology ([89]). To inform research into these applications, the typology we developed would need to be modified to reflect the unique characteristics of self-avatars (e.g., resemblance to self). Some designer brands (e.g., Gucci) have successfully engaged customers to dress their self-avatars in branded products, then share them on social media ([16]). The limited research on self-avatars has focused primarily on enclosed online environments (e.g., retailer's website), not social media ([20]; [35]), indicating the pressing need for insights into how, why, and when self-avatars could perform in social media marketing campaigns.
Brands also have turned to virtual influencers (3D, computer-generated personalities) instead of or in addition to human influencers for online marketing campaigns. Powered by advanced AI, these avatars can attract significant followers on social media platforms. For example, with almost 3 million Instagram followers, Lil Miquela has endorsed brands such as Prada and Calvin Klein ([ 1]; [ 6]). This avatar actively replies to social media comments, appears in publications like Vogue, and even participates in live media interviews ([19]). While virtual influencers offer unique benefits to brands such as content control and versatility, they also create potential risks. For example, 61% of consumers assert that authentic, relatable content is the primary appeal of human influencers ([80]), but only 15% of followers of virtual influencers describe them as credible ([21]). Moreover, because the virtual influencer avatar is not a human, the brand it endorses ultimately is held responsible for its actions. Academic research has yet to investigate the unique benefits, risks, and operational mechanisms associated with avatar-based virtual influencer marketing.
Future research might also explore avatar-based targeting strategies. Demographics, psychographics, and benefits sought are widely used customer segmentation bases ([95]); they also might be used to predict which customers will be best served by a given type of avatar. For example, customer's demographic traits might interact with the avatar's demographic attributes or behavioral elements to influence customer's cognitive, affective, and social responses to the online experience. Creating a personality or decision-making style for an avatar that matches those of the customer might be an effective, psychographics-based design strategy ([ 4]). Segmenting markets on the basis of in-depth analyses of the motives that lead certain people to interact with avatars could also inform benefits-based avatar deployment strategies ([11]). Distinct mediation mechanisms could be uncovered across different customer segments to inform idiosyncratic avatar designs. Research is also needed to determine when avatars may distract from, rather than contribute to, the customer's experience, as well as find strategies to address these challenges. Finally, to extend beyond our focus on the relative effects of the different types of avatars established by our taxonomy, future research might compare the effects of avatars with the impacts of other digital representations such as emoji, anthropomorphized products, brand mascots, or voice-only digital assistants. Understanding when and how avatars, versus these alternative marketing tools, are more effective in influencing online shopping experience and performance outcomes demands further scientific inquiry.
Rapid increases in the use of avatars have been fueled by two main factors: advances in digital technologies and increasing reliance on online experiences among both consumers and firms. The use of avatars is projected to grow by 35% annually ([41]). However, the effectiveness of avatars continues to be uncertain, so we offer an integrated theoretical framework to establish definitional and conceptual clarity, synthetize academic research and business practices, and offer insights and propositions that provide managerial implications and an agenda for future research. The proposed definition of avatars, as digital entities with anthropomorphic appearance, controlled by a human or software, with an ability to interact, helps us establish a design typology, which in turn gives academics and managers insights into how to isolate elements that make avatars more or less effective for specific goals.
We synthesize academic literature and business practices by offering a 2 × 2 form realism–behavioral realism taxonomy, which in turn enables us to derive propositions regarding the effectiveness of avatars in marketing. The level of alignment between an avatar's form realism and behavioral realism, according to several contingencies, can provide a parsimonious explanation of when an avatar will be most effective. With insights gained from our investigation of fundamental avatar elements, extant research, and business practices, we develop an integrative framework of avatar performance that offers theoretical insights, research propositions, managerial implications, and future research directions.
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921996646 - An Emerging Theory of Avatar Marketing
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921996646 for An Emerging Theory of Avatar Marketing by Fred Miao, Irina V. Kozlenkova, Haizhong Wang, Tao Xie and Robert W. Palmatier in Journal of Marketing
Supplemental Material, sj-pptx-1-jmx-10.1177_0022242921996646 - An Emerging Theory of Avatar Marketing
Supplemental Material, sj-pptx-1-jmx-10.1177_0022242921996646 for An Emerging Theory of Avatar Marketing by Fred Miao, Irina V. Kozlenkova, Haizhong Wang, Tao Xie and Robert W. Palmatier in Journal of Marketing
Footnotes 1 Cait Lamberton
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The authors acknowledge the financial support of the National Natural Science Foundation of China (grant #71832015; 72072191)
4 Online supplement: https://doi.org/10.1177/0022242921996646
5 Related terms included animated agent, anthropomorphic agent, artificial agent, chatbot, conversational agent, digital assistant, electronic shopping agent, embodied agent, embodied virtual agent, nonhuman agent, spokes-avatar, spokes-character, virtual agent, virtual assistant, and virtual character.
6 The academic disciplines included were (1) marketing (e.g., Journal of Marketing), (2) computer science (e.g., Computers in Human Behavior), (3) information systems (e.g., Journal of Management Information Systems), (4) communications (e.g., Human Communication Research), (5) education (e.g., Computers & Education), (6) health care (e.g., Journal of Medical Internet Research), and (7) general business (e.g., Journal of Business Research).
7 We acknowledge there may be exceptions in which a very "cute" avatar, although low in form realism, might provide strong hedonic value. We thank an anonymous reviewer for this insight.
References Abad Mario. (2019), " People Are Slamming Bella Hadid Kissing a Female Influencer as 'Queer-Baiting' in Calvin Klein Ad ," Yahoo! Life (May 17), www.yahoo.com/lifestyle/people-slamming-bella-hadid-kissing-205745785.html.
Afifi Walid A. , Burgoon Judee K.. (2000), " The Impact of Violations on Uncertainty and the Consequences for Attractiveness ," Human Communication Research , 26 (2), 203 – 33.
Ahn Sun Joo , Fox Jesse , Bailenson Jeremy N.. (2012), " Avatars ," in Leadership in Science and Technology: A Reference Handbook , Bainbridge William Sims , ed. Thousand Oaks, CA : Sage Publications , 695 – 702.
Al-Natour Sameh , Benbasat Izak , Cenfetelli Ron. (2011), " The Adoption of Online Shopping Assistants: Perceived Similarity as an Antecedent to Evaluative Beliefs ," Journal of the Association for Information Systems , 12 (5), 347 – 74.
Bailenson Jeremy N. , Yee Nick , Blascovich Jim , Guadagno Rosanna E.. (2008), " Transformed Social Interaction in Mediated Interpersonal Communication ," in Mediated Interpersonal Communication , Konijn Elly A. , Utz Sonja , Tanis Martin , Barnes Susan B. , eds. New York : Routledge , 77 – 99.
Bezamat Bia. (2018), " Prada Enlists Computer-Generated Influencer to Promote Fall 18 Show ," Current Daily (February 27), https://thecurrentdaily.com/2018/02/27/prada-enlists-computer-generated-influencer-ss18-show/.
Bickmore Timothy W. , Pfeifer Laura M. , Jack Brian W.. (2009), " Taking the Time to Care: Empowering Low Health Literacy Hospital Patients with Virtual Nurse Agents ," in Proceedings of the SIGCHI Conference on Human Factors in Computing Systems. New York : Association for Computing Machinery , 1265 – 74.
8 Bickmore Timothy W. , Utami Dina , Matsuyama Robin , Paasche-Orlow Michael K.. (2016), " Improving Access to Online Health Information with Conversational Agents: A Randomized Controlled Experiment ," Journal of Medical Internet Research , 18 (1), e1.
9 Blascovich Jim , Loomis Jack , Beall Andrew C. , Swinth Kimberly R. , Hoyt Crystal L. , Bailenson Jeremy N.. (2002), " Immersive Virtual Environment Technology as a Methodological Tool for Social Psychology ," Psychological Inquiry , 13 (2), 103 – 24.
Bradbury Danny. (2018), " The Changing Faces of AI ," Workflow (August 27), https://workflow.servicenow.com/customer-experience/build-ai-avatar/.
Brandtzaeg Petter Bae , Følstad Asbjørn. (2018), " Chatbots: Changing User Needs and Motivations ," Interactions , 25 (5), 38 – 43.
Brave Scott , Nass Clifford , Hutchinson Kevin. (2005), " Computers That Care: Investigating the Effects of Orientation of Emotion Exhibited by an Embodied Computer Agent ," International Journal of Human-Computer Studies , 62 (2), 161 – 87.
Briggs Bill. (2018), " Guess Who Wants to Talk! How Flo and Her Fellow Chatbots Engage Customers ," Microsoft (March 12), https://news.microsoft.com/transform/flo-rise-ai-chatbots-progressive-sabre-ups/.
Brunsman Barrett J.. (2019), " P&G Introduces Virtual Ambassador 'Obsessed' over Skin Care (Video) ," Cincinnati Business Courier (June 26), https://www.bizjournals.com/cincinnati/news/2019/06/26/p-g-introduces-virtualambassador-obsessed-over.html.
Burden David , Savin-Baden Maggi. (2019), Virtual Humans: Today and Tomorrow. New York : CRC Press.
Carson Biz. (2019), " Billionaires Jim Breyer and Thomas Tull Lead $15 Million Bet That Genies' Avatars Will Be Next Big Thing in Social ," Forbes (June 11), www.forbes.com/sites/bizcarson/2019/06/11/jim-breyer-thomas-tull-genies-funding/#7468ec1b600a.
Chattaraman Veena , Kwon Wi-Suk , Gilbert Juan E.. (2012), " Virtual Agents in Retail Web Sites: Benefits of Simulated Social Interaction for Older Users ," Computers in Human Behavior , 28 (6), 2055 – 66.
Chattaraman Veena , Kwon Wi-Suk , Gilbert Juan E. , Ross Kassandra. (2019), " Should AI-Based, Conversational Digital Assistants Employ Social- or Task-Oriented Interaction Style? A Task-Competency and Reciprocity Perspective for Older Adults ," Computers in Human Behavior , 90 (January) , 315 – 30.
Chichioco Aaron. (2019), " Virtual Influencers: The Significance of Influencer Chatbots to Your Brand Strategy ," Chatbots Magazine (May 23), https://chatbotsmagazine.com/virtual-influencers-the-significance-of-influencer-chatbots-to-your-brand-strategy-f6206c48adea.
Cho Hyejeung , Schwarz Norbert. (2012), " I Like Your Product When I Like My Photo: Misattribution Using Interactive Virtual Mirrors ," Journal of Interactive Marketing , 26 (4), 235 – 43.
Chowdhary Mukta. (2019), " How the Humans Behind CGI Influencers Need to Adapt to Consumer Needs: Lil Miquela Isn't Making as Good of an Impression as Real People ," Adweek (March 13), www.adweek.com/digital/how-the-humans-behind-cgi-influencers-need-to-adapt-to-consumer-needs/.
Corner Stuart. (2018), " Vodafone to Deploy Digital Humans for Customer Service ," (accessed October 2, 2019), www.computerworld.com/article/3478861/vodafone-to-deploy-digital-humans-for-customer-service.html.
D'Mello Sidney K. , Graesser Art , King Brandon. (2010), " Toward Spoken Human–Computer Tutorial Dialogues ," Human-Computer Interaction , 25 (4), 289 – 323.
Davis Fred. D. , Bagozzi Richard P. , Warshaw Paul R.. (1992), " Extrinsic and Intrinsic Motivation to Use Computers in the Workplace ," Journal of Applied Social Psychology , 22 (14), 1111 – 32.
De Haan Evert , Kannan P.K. , Verhoef Peter C. , Wiesel Thorsten. (2018), " Device Switching in Online Purchasing: Examining the Strategic Contingencies ," Journal of Marketing , 82 (5), 1 – 19.
Derrick Douglas C. , Ligon Gina Scott. (2014), " The Affective Outcomes of Using Influence Tactics in Embodied Conversational Agents ," Computers in Human Behavior , 33 (April) , 39 – 48.
Dick Alan , Chakravarti Dipankar , Biehal Gabriel. (1990), " Memory-Based Inferences During Consumer Choice ," Journal of Consumer Research , 17 (1), 82 – 93.
Dormehl Luke. (2018), " Microsoft's Friendly Xiaoice A.I Can Figure Out What You Want—Before You Ask ," DigitalTrends (November 18), www.digitaltrends.com/cool-tech/xiaoice-microsoft-future-of-ai-assistants/.
Dwyer F. Robert , Schurr Paul H. , Oh Sejo. (1987), " Developing Buyer-Seller Relationships ," Journal of Marketing , 51 (2), 11 – 27.
Epley Nicholas , Waytz Adam , Cacioppo John T.. (2007), " On Seeing Human: A Three-Factor Theory of Anthropomorphism ," Psychological Review , 114 (4), 864 – 86.
Etemad-Sajadi Reza. (2014), " The Influence of a Virtual Agent on Web-Users' Desire to Visit the Company ," International Journal of Quality & Reliability Management , 31 (4), 419 – 34.
Etemad-Sajadi Reza. (2016), " The Impact of Online Real-Time Interactivity on Patronage Intention: The Use of Avatars ," Computers in Human Behavior , 61 (August) , 227 – 32.
Evangelidis Ioannis , Osselaer Stijn M.J. Van. (2018), " Points of (Dis)parity: Expectation Disconfirmation from Common Attributes in Consumer Choice ," Journal of Marketing Research , 55 (1), 1 – 13.
Finn Adam. (2012), " Customer Delight: Distinct Construct or Zone of Nonlinear Response to Customer Satisfaction? " Journal of Service Research , 15 (1), 99 – 110.
Fiore Ann Marie , Kim Jihyun , Lee Hyun-Hwa. (2005), " Effect of Image Interactivity Technology on Consumer Responses Toward the Online Retailer ," Journal of Interactive Marketing , 19 (3), 38 – 53.
Fox Jesse , Ahn Sun Joo , Janssen Joris H. , Yeykelis Leo , Segovia Kathryn Y. , Bailenson Jeremy N.. (2015), " Avatars Versus Agents: A Meta-Analysis Quantifying the Effect of Agency on Social Influence ," Human-Computer Interaction , 30 (5), 401 – 32.
Frank Aaron. (2019), " The Rise of a New Generation of AI Avatars ," SingularityHub (January 15), https://singularityhub.com/2019/01/15/the-rise-of-a-new-generation-of-ai-avatars/.
Freeman C. , Beaver I.. (2018), " The Effect of Response Complexity and Media on User Restatement with Multimodal Virtual Assistants ," International Journal of Human-Computer Studies , 119 (November) , 12 – 27.
Garau Maia , Slater Mel , Vinayagamoorthy Vinoba , Brogni Andrea , Steed Anthony , Angela Sasse M.. (2003), " The Impact of Avatar Realism and Eye Gaze Control on Perceived Quality of Communication in a Shared Immersive Virtual Environment ," in Proceedings of the SIGCHI Conference on Human Factors in Computing Systems. New York : Association for Computing Machinery , 529 – 36.
Garnier Marion , Poncin Ingrid. (2013), " The Avatar in Marketing: Synthesis, Integrative Framework and Perspectives ," Recherche et Applications en Marketing (English Edition) , 28 (1), 85 – 115.
Globe Newswire (2019), " Chatbot Market-Growth, Trends, and Forecast (2019-2024) ," Research Report No. 4622740ResearchAndMarkets.com.
Go Eun , Sundar S. Shyam. (2019), " Humanizing Chatbots: The Effects of Visual, Identity and Conversational Cues on Humanness Perceptions ," Computers in Human Behavior , 97 (August) , 304 – 16.
Gonzalez Robbie. (2017), " Virtual Therapists Help Veteran Open Up About PTSD ," Wired (October 17), www.wired.com/story/virtual-therapists-help-veterans-open-up-about-ptsd/.
Ho Annabell , Hancock Jeff , Miner Adam S.. (2018), " Psychological, Relational, and Emotional Effects of Self-Disclosure after Conversations with a Chatbot ," Journal of Communication , 68 (4), 712 – 33.
Holzwarth Martin , Janiszewski Chris , Neumann Marcus M.. (2006), " The Influence of Avatars on Online Consumer Shopping Behavior ," Journal of Marketing , 70 (4), 19 – 36.
Ipsoft (2018), " Amelia in Action ," (January) https://www.ipsoft.com/wp-content/uploads/2016/10/Amelia-in-Action.pdf.
Jap Sandy D. , Ganesan Shankar. (2000), " Control Mechanisms and the Relationship Life Cycle: Implications for Safeguarding Specific Investments and Developing Commitment ," Journal of Marketing Research , 37 (2), 227 – 45.
Jin Seung-A. Annie. (2009), " The Roles of Modality Richness and Involvement in Shopping Behavior in 3D Virtual Stores ," Journal of Interactive Marketing , 23 (3), 234 – 46.
Kahn Jeremy. (2018), " Meet 'Millie' the Avatar. She'd Like to Sell You a Pair of Sunglasses ," Bloomberg (December 15), www.bloomberg.com/news/articles/2018-12-15/meet-millie-the-avatar-she-d-like-to-sell-you-a-pair-of-sunglasses.
Kang Sin-Hwa , Watt James H.. (2013), " The Impact of Avatar Realism and Anonymity on Effective Communication Via Mobile Devices ," Computers in Human Behavior , 29 (3), 1169 – 81.
Kang Sin-Hwa , Watt James H. , Ala Sasi Kanth. (2008), " Communicators' Perceptions of Social Presence as a Function of Avatar Realism in Small Display Mobile Communication Devices ," in Proceedings of the 41st Annual Hawaii International Conference on System Sciences (HICSS 2008). Washington, DC : IEEE Computer Society , 147 – 56.
Kannan P.V. , Bernoff Josh. (2019), " Does Your Company Really Need a Chatbot? " Harvard Business Review (May 21), https://hbr.org/2019/05/does-your-company-really-need-a-chatbot.
Keeling Kathleen , McGoldrick Peter , Beatty Susan. (2010), " Avatars as Salespeople: Communication Style, Trust, and Intentions ," Journal of Business Research , 63 (8) , 793 – 800.
Kilens Mark. (2019), " 2019 State of Conversational Marketing [Free Report] ," (September 14), www.drift.com/blog/state-of-conversational-marketing/.
Kim Kyoung-Min , Hong Jin-Hyuk , Cho Sung-Bae. (2007), " A Semantic Bayesian Network Approach to Retrieving Information with Intelligent Conversational Agents ," Information Processing & Management , 43 (1), 225 – 36.
Kim Sara , Chen Rocky Peng , Zhang Ke. (2016), " Anthropomorphized Helpers Undermine Autonomy and Enjoyment in Computer Games ," Journal of Consumer Research , 43 (2), 282 – 302.
Kim Youjeong , Shyam Sundar S.. (2012), " Anthropomorphism of Computers: Is It Mindful or Mindless? " Computers in Human Behavior , 28 (1), 241 – 50.
Köhler Clemens F. , Rohm Andrew J. , de Ruyter Ko , Wetzels Martin. (2011), " Return on Interactivity: The Impact of Online Agents on Newcomer Adjustment ," Journal of Marketing , 75 (2), 93 – 108.
Lee Seo Young , Choi Junho. (2017), " Enhancing User Experience with Conversational Agent for Movie Recommendation: Effects of Self-Disclosure and Reciprocity ," International Journal of Human-Computer Studies , 103 (July) , 95 – 105.
Liew Tze Wei , Tan Su-Mae , Ismail Hishamuddin. (2017), " Exploring the Effects of a Non-Interactive Talking Avatar on Social Presence, Credibility, Trust, and Patronage Intention in an E-Commerce Website ," Human-Centric Computing and Information Sciences , 7 (1), 1 – 21.
Liu Yuping , Shrum Lawrence J.. (2002), " What Is Interactivity and Is It Always Such a Good Thing? Implications of Definition, Person, and Situation for the Influence of Interactivity on Advertising Effectiveness ," Journal of Advertising , 31 (4), 53 – 64.
Llop Cristina. (2016), " Gina—LA's Online Traffic Avatar Radically Changes Customer Experiences ," Self-Represented Litigation Network (accessed April 20, 2021) , www.srln.org/node/1186/gina-las-online-traffic-avatar-radically-changes-customer-experience-news-2016.
Martin Kelly D. , Borah Abhishek , Palmatier Robert W.. (2017), " Data Privacy: Effects on Customer and Firm Performance ," Journal of Marketing , 81 (1), 36 – 58.
McDuff Daniel , Czerwinski Mary. (2018), " Designing Emotionally Sentient Agents ," Communications of the ACM , 61 (12), 74 – 83.
McGloin Rory , Nowak Kristine L. , Stiffano Stephen C. , Flynn Gretta M.. (2009), " The Effect of Avatar Perception on Attributions of Source and Text Credibility ," in Proceedings of ISPR 2009, International Society for Presence Research Annual Conference. Philadelphia : Temple University Press , 1 – 9.
Melumad Shiri , Inman J. Jeffrey , Pham Michel Tuan. (2019), " Selectively Emotional: How Smartphone Use Changes User-Generated Content ," Journal of Marketing Research , 56 (2), 259 – 75.
Mimoun Mohammed Slim Ben , Poncin Ingrid. (2015), " A Valued Agent: How ECAs Affect Website Customers' Satisfaction and Behaviors ," Journal of Retailing and Consumer Services , 26 (September) , 70 – 82.
Moon Youngme. (2000), " Intimate Exchanges: Using Computers to Elicit Self-Disclosure from Consumers ," Journal of Consumer Research , 26 (4), 323 – 39.
Nass Clifford , Moon Youngme. (2000), " Machines and Mindlessness: Social Responses to Computers ," Journal of Social Issues , 56 (1), 81 – 103.
Nass Clifford , Moon Youngme , Fogg Brian Jeffrey , Reeves Byron , Dryer D. Christopher. (1995), " Can Computer Personalities Be Human Personalities? " International Journal of Human-Computer Studies , 43 (2), 223 – 39.
Nass Clifford , Yen Corina. (2010), The Man Who Lied to His Laptop: What Machines Teach Us About Human Relationships. New York : Current.
Nowak Kristine L. , Rauh Christian. (2006), " The Influence of the Avatar on Online Perceptions of Anthropomorphism, Androgyny, Credibility, Homophily, and Attraction ," Journal of Computer-Mediated Communication , 11 (1), 153 – 78.
Nowak Kristine L. , Biocca Frank. (2003), " The Effect of the Agency and Anthropomorphism on Users' Sense of Telepresence, Copresence, and Social Presence in Virtual Environments ," Presence: Teleoperators & Virtual Environments , 12 (5), 481 – 94.
Nowak Kristine L. , Fox Jesse. (2018), " Avatars and Computer-Mediated Communication: A Review of the Definitions, Uses, and Effects of Digital Representations ," Review of Communication Research , 6 , 30 – 53.
Nunamaker Jay F. , Derrick Douglas C. , Elkins Aaron C. , Burgoon Judee K. , Patton Mark W.. (2011), " Embodied Conversational Agent-Based Kiosk for Automated Interviewing ," Journal of Management Information Systems , 28 (1), 17 – 48.
Oliver Richard L.. (1980), " A Cognitive Model of the Antecedents and Consequences of Satisfaction Decisions ," Journal of Marketing Research , 17 (4), 460 – 69.
Palmatier Robert W. , Houston Mark B. , Dant Rajiv P. , Grewal Dhruv. (2013), " Relationship Velocity: Toward a Theory of Relationship Dynamics ," Journal of Marketing , 77 (1), 13 – 30.
Parboteeah D. Veena , Valacich Joseph S. , Wells John D.. (2009), " The Influence of Website Characteristics on a Consumer's Urge to Buy Impulsively ," Information Systems Research , 20 (1), 60 – 78.
Peddie Bryan. (2018), " Are Virtual Tellers the Future of AI in the Banking Sector? ," NCR (April 13), www.ncr.com/company/blogs/financial/are-virtual-tellers-the-future-of-ai-in-the-banking-sector
Penny Sarah. (2019), " Virtual Influencers Might Be Easier to Mould but They're Not Necessarily a Safer Option ," (March 27), https://phvntom.com/virtual-influencers-might-be-easier-to-mould-but-theyre-not-necessarily-a-safer-option/.
Persky Susan , Blascovich Jim. (2007), " Immersive Virtual Environments versus Traditional Platforms: Effects of Violent and Nonviolent Video Game Play ," Media Psychology , 10 (1), 135 – 56.
Phaneuf Alicia. (2020), " 7 Real Examples of Brands and Businesses Using Chatbots to Gain an Edge ," Business Insider (February 12), www.businessinsider.com/business-chatbot-examples
Qiu Lingyun , Benbasat Izak. (2009), " Evaluating Anthropomorphic Product Recommendation Agents: A Social Relationship Perspective to Designing Information Systems ," Journal of Management Information Systems , 25 (4), 145 – 82.
Reeves Byron , Nass Clifford Ivar. (1996), The Media Equation: How People Treat Computers, Television, and New Media Like Real People and Places. New York : Cambridge University Press.
Robinson Ann. (2015), " Meet Ellie, the Machine That Can Detect Depression ," The Guardian (September 17), www.theguardian.com/sustainable-business/2015/sep/17/ellie-machine-that-can-detect-depression
Schuetzler Ryan , Giboney Justin Scott , Grimes G. Mark , Nunamaker Jay F. Jr. (2018), " The Influence of Conversational Agent Embodiment and Conversational Relevance on Socially Desirable Responding ," Decision Support Systems , 114 , 94 – 102.
Scott David M.. (2008), " Anna from IKEA Is Intellectually Challenged (but She Has a Sense of Humor) ," blog post (August 20), www.davidmeermanscott.com/blog/2008/08/anna-from-ikea.html
Sivaramakrishnan Subramanian , Wan Fang , Tang Zaiyong. (2010), " Giving an 'E-Human Touch' to E-Tailing: The Moderating Roles of Static Information Quantity and Consumption Motive in the Effectiveness of an Anthropomorphic Information Agent ," Journal of Interactive Marketing , 21 (1), 60 – 75.
Smith Stephen P. , Johnston Robert B. , Howard Steve. (2011), " Putting Yourself in the Picture: An Evaluation of Virtual Model Technology as an Online Shopping Tool ," Information Systems Research , 22 (3), 640 – 59.
Stayman Douglas M. , Alden Dana L. , Smith Karen H.. (1992), " Some Effects of Schematic Processing on Consumer Expectations and Disconfirmation Judgements ," Journal of Consumer Research , 19 (2), 240 – 55.
Sundar S. Shyam. (2008), " The MAIN Model: A Heuristic Approach to Understanding Technology Effects on Credibility ," in Digital Media, Youth, and Credibility , Metzger Miriam J. , Flanagin Andrew J. , eds. Boston : The MIT Press , 73 – 100.
Sweezey Mathew. (2019), " Consumer Preference for Chatbots Is Challenging Brands to Think 'Bot First' ," Forbes (August 16), www.forbes.com/sites/forbescommunicationscouncil/2019/08/16/consumer-preference-for-chatbots-is-challenging-brands-to-think-bot-first/#4407c60c10f8
Torresin Veronica. (2019), " How Chatbots Improve User Experience in Online Banking ," Ergomania (February 7), https://ergomania.eu/how-chatbots-improve-user-experience-in-online-banking/.
Touré-Tillery Maferima , McGill Ann L.. (2015), " Who or What to Believe: Trust and the Differential Persuasiveness of Human and Anthropomorphized Messengers ," Journal of Marketing , 79 (4), 94 – 110.
Tynan A. Caroline , Drayton Jennifer. (1987), " Market Segmentation ," Journal of Marketing Management , 2 (3), 301 – 35.
Verhagen Tibert , van Nes Jaap , Feldberg Frans , van Dolen Willemijn. (2014), " Virtual Customer Service Agents: Using Social Presence and Personalization to Shape Online Service Encounters ," Journal of Computer-Mediated Communication , 19 (3), 529 – 45.
Von der Pütten Astrid M. , Krämer Nicole C. , Gratch Jonathan , Kang Sin-Hwa. (2010), " 'It Doesn't Matter What You Are!' Explaining Social Effects of Agents and Avatars ," Computers in Human Behavior , 26 (6), 1641 – 50.
Wang Liz C. , Baker Julie , Wagner Judy A. , Wakefield Kirk. (2007), " Can a Retail Web Site Be Social? " Journal of Marketing , 71 (3), 143 – 57.
Westerman David , Tamborini Ron , Bowman Nicholas David. (2015), " The Effects of Static Avatars on Impression Formation Across Different Contexts on Social Networking Sites ," Computers in Human Behavior , 53 (December) , 111 – 17.
White Susan S. , Schneider Benjamin. (2000), " Climbing the Commitment Ladder: The Role of Expectations Disconfirmation on Customers' Behavioral Intentions ," Journal of Service Research , 2 (3), 240 – 53.
Wooler Brodie. (2019), " We Need to Chat About Chatbots ," LinkedIn (accessed May 7), www.linkedin.com/pulse/we-need-chat-chatbots-brodie-wooler/
Wu Jen-Her , Wang Shu-Ching. (2005), " What Drives Mobile Commerce? An Empirical Evaluation of the Revised Technology Acceptance Model ," Information & Management , 42 (5), 719 – 29.
Xu Jingjun D. , Abdinnour Sue , Chaparro Barbara. (2017), " An Integrated Temporal Model of Belief and Attitude Change: An Empirical Test with the iPad ," Journal of the Association for Information Systems , 18 (2), 113 – 40.
Yee Nick , Bailenson Jeremy N. , Rickertsen Kathryn. (2007), " A Meta-Analysis of the Impact of the Inclusion and Realism of Human-Like Faces on User Experiences in Interfaces ," in Proceedings of the SIGCHI Conference on Human Factors in Computing Systems. New York : Association for Computing Machinery , 1 – 10.
Yokotani Kenji , Takagi Gen , Wakashima Kobun. (2018), " Advantages of Virtual Agents over Clinical Psychologists During Comprehensive Mental Health Interviews Using a Mixed Methods Design ," Computers in Human Behavior , 85 (August) , 135 – 45.
Yun Chang , Deng Zhigang , Hiscock Merrill. (2009), " Can Local Avatars Satisfy a Global Audience? A Case Study of High-Fidelity 3D Facial Avatar Animation in Subject Identification and Emotion Perception by US and International Groups ," Computers in Entertainment , 7 (2), 1 – 26.
~~~~~~~~
By Fred Miao; Irina V. Kozlenkova; Haizhong Wang; Tao Xie and Robert W. Palmatier
Reported by Author; Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 5- Analyzing the Cultural Contradictions of Authenticity: Theoretical and Managerial Insights from the Market Logic of Conscious Capitalism. By: Thompson, Craig J.; Kumar, Ankita. Journal of Marketing. Sep2022, Vol. 86 Issue 5, p21-41. 21p. 1 Color Photograph, 2 Diagrams, 1 Chart. DOI: 10.1177/00222429221087987.
- Database:
- Business Source Complete
Analyzing the Cultural Contradictions of Authenticity: Theoretical and Managerial Insights from the Market Logic of Conscious Capitalism
This research analyzes the cultural contradictions of authenticity as they pertain to the actions of consumers and marketers. The authors' conceptualization diverges from the conventional assumption that the ambiguity manifest in the concept of authenticity can be resolved by identifying an essential set of defining attributes or by conceptualizing it as a continuum. Using a semiotic approach, the authors identify a general system of structural relationships and ambiguous classifications that organize the meanings through which authenticity is understood and contested in a given market context. They demonstrate the contextually adaptable nature of this framework by analyzing the authenticity contradictions generated by the cultural tensions between "conscious capitalism"—a market logic that encompasses both global brands and small independent businesses, such as a farm-to-table restaurant or an organic food co-op—and the elitist critique. The Slow Food movement provides a case study for analyzing how consumers, producers, and entrepreneurs who identify with conscious capitalist ideals understand these disauthenticating, elitist associations and the strategies they use to counter them. The authors conclude by discussing implications of the analysis for theories of authenticity and for managing the authenticity challenges facing conscious capitalist brands.
Keywords: authenticity; brand image; conscious capitalism; ethical consumerism; consumer identity; market logics; semiotics
Consumers crave authenticity—so much so that their quest for authenticity is considered "one of the cornerstones of contemporary marketing" ([11], p. 21). This has created an enormous challenge for the field, considering that marketing itself is typically considered inherently inauthentic. —[67], p. 1)
In the field of marketing, little doubt exits that "authenticity" is highly desired by consumers and thereby is a crucially important strategic resource for marketing management. Consumers are more likely to form stronger emotional attachments to a brand, business, or tourist site they perceive as being authentic ([20]; [30]; [57]; [86]) and to incorporate these market resources into their identities ([ 7]; [11]; [45]). On the managerial side, [31], p. 610) conclude that authenticity is "the most rare and coveted asset in the contemporary branding landscape." Their assertion is supported by an array of studies indicating that authenticity is integral to the enhancement of brand equity ([60]), effective brand extensions ([82]), persuasive marketing communications ([ 5]), success in relationship marketing ([22]), and emotionally engaging person and celebrity brands ([31]; [87]).
Although there is a clear consensus that authenticity profoundly matters to both consumers and marketers, the marketing literature also presents a recurrent concern that authenticity is a nebulous concept that has eluded precise definition ([ 5]). [67], p. 2) proclaim that this conceptual ambiguity poses a significant barrier to creating "a coherent theory of authenticity." Accordingly, they aim to redress this dilemma by presenting a general definition of authenticity based on six key perceptual components. In contrast, [81], p. 3) proposes that "authenticity is a polysemous and multilayered concept" and thus "it might not [emphasis added] be helpful to compress the wealth of disparate meanings associated with the concept into a single definition."
As Södergren further notes in his meta-analysis, "the majority of the research [on authenticity] has focused on characteristics that distinguish the 'real thing' from the fake" (p. 11). To further elaborate on this conceptual tendency, marketers' efforts to define authenticity almost invariably invoke some variant of genuineness, such as brands (via their management teams) staying true to ideals of timeless tradition, heritage, craftsmanship, and quality (see also [ 6]; [82]). In this spirit, [55], p. 30) propose that the authenticity of a brand's social media communications hinges on perceptions of honesty, sincerity, and being "real." [ 7] similarly contend that the core cultural meanings of authenticity are truth, genuineness, and reality. [67] comprehensive definition of authenticity also incorporates a series of veracity-oriented constructs, such as originality (i.e., not being a copy), accuracy (i.e., being true to others), and integrity (i.e., being true to oneself).
While the analytic goal of distinguishing the authentic from the inauthentic makes intuitive sense, it is a Sisyphean undertaking that attempts to specify an ambiguous cultural category by referring to other semantic terms whose meanings are also contextually contingent and malleable (i.e., honesty, sincerity, originality, genuineness, and truthfulness). Furthermore, informing marketing managers that their brand lacks authenticity because consumers see it as being unoriginal, insincere, or dishonest offers little guidance on how to resolve the deeper cultural tensions that drive these unfavorable perceptions. Rather than a checklist of definitional attributes, we argue that marketing managers need an analytic approach that can enable them to answer questions such as ( 1) why is their brand or business susceptible to certain kinds of authenticity challenges?, ( 2) what cultural meanings and contradictions underlie those challenges?, and ( 3) what responses could they take to mitigate the disauthenticating associations that ensue from these tensions?
Returning to our opening vignette, we can reframe [67], p. 1) statement that marketers face an "enormous challenge" because their profession is "typically seen as inauthentic" (see also [ 5]) as a realization that marketing, as a business practice, also occupies an ambiguous cultural position. On the one hand, marketing aims to advocate for the needs (and voices) of customers ([41]) and, yet, it is also means for companies to enhance their profits and market share. This tension readily gives rise to concerns that short-term (and potentially exploitive) profitability goals might take priority over serving customers' best interests. Accordingly, consumers are inundated with cultural narratives (ranging from journalistic reports about deceptive marketing tactics to portrayals of unscrupulous marketers by entertainment media) that encourage cynicism and distrust toward marketers' branding claims and persuasive communications ([32]; [45]; [66]).
However, the specific cultural meanings and associations that lead to perceptions of authenticity or inauthenticity vary across brands and markets. For example, consumers are likely to deploy different configurations of meanings, beliefs, and evaluative norms when judging the authenticity of a high-fashion retailer ([23]), a café owner who promotes their establishment as a home-away-from-home ([20]) or a global brand that positions itself as an advocate for environmental justice (e.g., Patagonia; [49]).
In this article, we explain and demonstrate how the semiotic square ([40]) can be used to systematically analyze such culturally heterogeneous authenticity contradictions and to develop contextually appropriate responses to the specific authenticity challenges that arise in a given market. The semiotic square is an analytic tool that has often been used to delineate cultural meanings and semantic contradictions that are manifest in both consumer perceptions and marketing strategies ([29]; [36]; [50]; [51]; [54]; [68]; [69]). From a semiotic perspective, the cultural categories of the authentic and the inauthentic are not just contrasting or oppositional terms. Rather, they are anchor points in a broader network of relationships through which the authenticity of a given brand, business, brand ambassador, social media influencer, and the like is culturally constructed and potentially contested.
Our market context is conscious capitalism, which refers to a "way of thinking about capitalism and business that better reflects where we are in the human journey, the state of our world today, and the innate potential of business to make a positive impact on the world" ([62], p. 273). Conscious capitalism is particularly vulnerable to the broader authenticity–inauthenticity tension that all marketers confront to varying degrees. Therefore, it serves as a very relevant and informative context for our analysis.
Conscious capitalism's key premise is that capitalism's societal purpose has, historically, been defined too narrowly (i.e., maximizing shareholder wealth and optimizing consumers' market choices) and, accordingly, its society-enhancing potential remains greatly underutilized. Rather than grafting a social mission onto a traditional profit-maximization model, as per conventional corporate social responsibility approaches, proponents of conscious capitalism contend that businesses should place value-driven goals and social consciousness at the core of their institutional missions ([62]).
By aiming to redefine the nature and function of capitalism, conscious capitalism can be analyzed as a market logic that transcends its iconic brands (e.g., Patagonia, Starbucks, TOMS, Whole Foods) or socially conscious businesses (e.g., a cooperatively owned, fair trade, local coffee shop). As discussed by [27], pp. 40–42), a market logic is an integrated network of meanings, values, and norms that provide ( 1) principles that can guide thoughts, actions, and preferences; ( 2) vocabularies of motivation and justification; and ( 3) material and symbolic resources for constructing an identity (such as being an ethical consumer or a purpose-driven business owner).
Conscious capitalism organizes a constellation of ideologically aligned brands and an even larger network of businesses that have different scales of operation and serve different roles in the supply chain. Thus, consumers who support this array of brands and enterprises have access to a set of normative principles to guide their purchase choices (e.g., locally sourced materials are preferred over imported ones, plastic product packaging should be avoided); they learn an intricate system of terms and codes (e.g., "postconsumer recycled content," third-party certification labels such as the Rainforest Alliance or Certified Carbon Neutral); and they can express their socially conscious sensibilities through an array of consumption practices—wearing a Patagonia fleece, driving an electric car, shopping at a farmers' market, brandishing a reusable Whole Foods' canvas tote bag, buying fair trade chocolate, or supporting a farm-to-table restaurant.
In the general public discourse, however, the authenticity of conscious capitalist brands and businesses, and their consumer supporters, is frequently called into question. These authenticity challenges are sufficiently problematic that leading proponents of conscious capitalism feel compelled to address them:
There is a growing network of people building their companies based on the idea that business is about more than making a profit. It's about higher purpose ... and the innate potential of business to make a positive impact on the world.... But one of the most predictable responses we get from people when we mention the idea of conscious capitalism is, "That's an oxymoron!" ([61])
Conscious capitalism's authenticity challenges hinge on a cultural tension between the profit-maximizing ethos of capitalism and the ennobling idea that capitalist enterprises can serve higher societal and moral purposes that supersede commercial interests ([ 3]; [33]). In this vein, critics often suggest that conscious capitalism deploys the language of sustainability and other socially beneficial goals for the instrumental purpose of catering to higher-income consumers who will pay a premium to imbue their consumption practices with an aura of moral virtue:
Thus, the rise of social enterprises [i.e., conscious capitalist enterprises] has been met with hostility, particularly toward its authenticity and its sustainable impact. If their goods and services continue to be priced as they are, is the sustainable movement only for the demographic that can afford it? ([14])
[70], p. 73) similarly argue that conscious capitalism is more promotional hyperbole than a viable business reality and, further, add this reservation: "It is important to note that the firms associated with the Conscious Capitalism movement are far from a random sample of American businesses: In fact, a great many sell relatively expensive products to relatively affluent, socially- or health-conscious consumers."
This incredulous and, at times, adversarial public response to conscious capitalism has not arisen ex nihilo. Rather, it draws from a cultural narrative that we characterize as the "elitist critique." As historian [34] elaborates, the political charge of elitism has evolved from its classic populist roots, which railed against the undue power wielded by the captains of industry and affluent political insiders, to an antipathy toward the intellectual class (who may not be unduly wealthy or politically powerful). Through this shift, the charge of elitism was distanced from its origins in economic conflicts between the working class and the owners of capital (and their management intermediaries) and became repositioned in a culture war rift whereby "the 'elite' could be identified by its liberal ideas, coastal real estate, and highbrow consumer preferences" ([34], emphasis added).
We investigate how the specific authenticity challenges posed by the elitist critique of conscious capitalism are negotiated by consumers and producers in the context of the Slow Food movement ([89]). Slow Food encompasses an array of ideologically aligned brands, enterprises (farm-to-table restaurants, artisan producers, and organic and free-range farmers), consumption practices (e.g., shopping at a farmers' market or a local co-op), and goods and services (e.g., an heirloom tomato, grass-fed beef, a class in fermentation techniques). Slow Food's signature issues and social change goals are grounded in the market logic of conscious capitalism, including local sourcing, fair wages for workers, sustainable modes of production, environmental awareness and habitat protection, and a broader project of redressing societal ills through the coordinated actions of socially conscious businesses and consumers (see [73], [74]). The elitist critique has also become part and parcel of Slow Food's brand image, and it poses salient authenticity challenges for Slow Food's producers, entrepreneurs, and consumers.
In the following sections, we first discuss the key analytic premises of the semiotic square. Next, we develop a semiotic conceptualization of authenticity that maps out its structural contradictions (and ambiguous classifications). We use this analytic framework to explicate the ways in which the elitist critique gives a particular cultural form to the authenticity contradictions plaguing the market logic of conscious capitalism. We then profile the authenticating strategies that Slow Food advocates (consumers, producers, and restauranteurs) use to counter these disauthenticating elitist associations. We conclude by discussing the implications of this analysis for theories of authenticity and for managing the authenticity challenges facing conscious capitalist enterprises.
From a semiotic perspective ([40]), the meaning and categorical boundaries of a given concept are defined through relations to what it is not. For example, the cultural meanings of masculinity have been historically established through contrasts to those that have defined femininity and the related nexus of ever-changing ideals, values, and practices through which this binary contrast has been culturally articulated and transformed over time ([50]). These structural relations give rise to ambiguous categories whose associated cultural meanings can become points of contestation and debate, such as in the cases of "metrosexuals" ([78]), stay-at-home dads ([19]), or the ongoing controversies sparked by the category of transgender athletes ([12]).
The binary opposition between authenticity and inauthenticity presents a similar arrangement of contradictions and ambiguous classifications. Consequently, we propose that authenticity is not a set of discrete properties that distinguish the genuine from the fake—but, rather, an ongoing process of managing a network of contingent relationships. In some markets, for some brands and enterprises, these contingencies may be more stable, whereas in others, they may become more culturally contested and, thus, unstable. We suggest that conscious capitalist brands and businesses, owing to the elitist critique, exemplify this latter and more managerially challenging case.
Figure 1 presents a semiotic square representation of authenticity. In this article, we use the "contradictions of authenticity" as an integrative term that encompasses the structural relations among the semiotic 3Cs (contrariety, complementarity, and contradictory relations).
Graph: Figure 1. A semiotic model of the authenticity–inauthenticity opposition.
The horizontal arrows represent contrariety relations. These relations are roughly analogous to the standard binary oppositions that anchor semantic differential scales. However, relations of contrariety further indicate that the meaning of a term is defined through a relationship to its binary contrast (e.g., good is understood relative to evil). Accordingly, the meaning of authenticity is always contingent on the operative meaning of inauthenticity, and vice versa. We refer to the authentic ↔ inauthentic contrariety as the primary contrariety relation because it represents the dominant tension that, in turn, sets the complementary terms for the secondary contrariety relation (i.e., not inauthentic ↔ not authentic).
The vertical arrows represent complementarity relations. Such conceptual pairings are compatible and noncontradictory (but are not necessarily synonymous or interchangeable). For example, "not inauthentic" is congruent with the dominant term, authentic. However, this classification also harbors other connotations and, thus, ambiguous meanings. For example, imagine a painting created by a famous artist, say Picasso, who at the time was a fledgling beginner, imitating the style of another painter. Because the painting does not evince Picasso's quintessential artistic motifs, its authenticity becomes ambiguous (and debatable)—that is, at what point in his career does a painting by Picasso truly become a "Picasso"? The term "not inauthentic" conveys this type of ambiguity.
The diagonal arrows represent contradictory relations. These relations indicate that any entity or action deemed to be authentic (or inauthentic) will harbor some qualities that can be judged as contradicting such an assessment. As an illustration, let us again consider the idea of artistic authenticity. From a conventional standpoint, the authenticity of an artist, even a renowned one, can always be challenged on the grounds that their creations exhibit properties that are derivative of other genres, styles, or artistic predecessors (authentic ↔ not authentic). Conversely, the art world's postmodern movement disavows the idea of artistic originality and, instead, celebrates that all artistic productions are, in some sense, a reworking of something prior. As exemplified by Andy Warhol's replications of iconic cultural images (Coca-Cola bottles, Campbell Soup cans, the face of Marilyn Monroe), postmodern art is also heralded for its capacity to surprise and inspire revelatory aesthetic experiences through its creative (and often ironic) uses of repetition, collage, assemblage, montage, and bricolage (inauthentic ↔ not inauthentic) ([43]).[ 4]
In Figure 1, the cloud-like drawings represent the specific cultural meanings that give contextual form to the contradictions of authenticity. For purposes of our analysis, the relevant meaning systems are the market logic of conscious capitalism and the elitist critique. This system of semiotic relationships gives rise to four emergent (and ambiguous) classifications, each harboring latent contradictions. In discussing these ambiguous categories, we first illustrate them in more general terms and then address their manifestations in the context of conscious capitalism and the elitist critique.
This complementarity relation corresponds to what [39] discuss as indexical authenticity. In this usage, an index refers to a given object or behavior—for example, the actions of a whitewater raft guide, handprints in front of Grauman's Chinese Theater, a painting, or a branded good. Indexes are classified as authentic when they are believed to possess a factual and spatiotemporal connection to some validating condition. For example, consumers will judge the actions of a whitewater raft guide as authentic if they are believed to reflect an inner passion for the outdoors (rather than being a calculated performance done for remunerative purposes; [ 2]). Similarly, consumers will typically deem a branded good to be authentic when they believe its design, production, and quality certification has proceeded under the auspices of those who own or manage the brand of note.
As these examples suggest, perceptions of indexical authenticity can be more or less certain. As an example of higher certainty, Prada certifies the genuineness of its handbags by assigning each a unique and traceable serial number that is documented on an authenticity card. On the less certain side, customers have to infer the indexical authenticity of a whitewater raft guide's passion for the outdoors or a retail associate's expressions of friendliness and interpersonal concern. In these cases, consumers' judgements about the authenticity (or inauthenticity) of a marketer's actions (or the actions of other consumers) depend on their inferences about underlying motivations and intent.
The elitist critique provides a constellation of culturally shared meanings and rationales that support disconfirming suppositions about the indexical authenticity of conscious capitalist enterprises and their consumer followers. These disauthenticating associations directly correspond to the ambiguous categories emerging from the primary contrariety (authentic ↔ inauthentic), the secondary contrariety (not inauthentic ↔ not authentic), and the complementarity relation of inauthentic ↔ not authentic relations.
This ambiguous classification corresponds to seemingly oxymoronic constructions such as authentic reproductions—or, in semiotic vernacular, "iconic authenticity" ([39]). In this usage, the icon is an object that is a known facsimile of an original referent and that is appreciated for its mimetic properties, as in the case of a comedian doing an uncanny impression of a celebrity. For the category of iconic authenticity, the ensuing goal is to present a compelling sense of verisimilitude through a meticulous recreation of the original referents' characteristics. Iconic authenticity is pursued by, among others, members of the cosplay community ([80]) and consumers who perform in historical recreations, such as Civil War reenactments ([17]). In a different market application, iconic authenticity would also be highly relevant to a budget-conscious consumer who wants to buy a convincing counterfeit version of an expensive designer brand.
When situated in the context of the elitist critique, the "authentic + inauthentic" category assumes less favorable meanings of moral pretentiousness and hypocrisy. In this disauthenticating cultural frame, affluent (and typically left-leaning) consumers use conscious capitalist brands and goods to distinguish themselves from the price-conscious mainstream and their socioeconomic peers who display affluence through more ostentatious lifestyle choices ([24]). By claiming the mantle of moral virtue, such consumers can pursue social distinction in an otherwise orthodox manner—that is, through material displays of refined tastes ([ 9]; [44])—while appearing to disavow materialism and status consciousness.
For example, during its heyday as a cultural icon, the Toyota Prius inspired oppositional brand communities (Muñiz and O'Guinn 2001) who referred to the vehicle (and its drivers) as "the pious." This epithet suggested that Prius drivers evinced a self-aggrandizing "holier than thou" stance that amplified the moral merits of their automotive preferences relative to those who made different choices ([59]). In a similar cultural vein, [13] note that ecofriendly consumers who purchase organic foods, drive electric luxury cars, and use natural cleaning products typically lead lifestyles that carry a much higher carbon footprint than lower-income consumers who live in smaller housing units, rely on public transport, and seldom fly. Seen in this critical light, such ecoconscious (affluent) consumers are virtue signaling ([24]; [42]; [90]), but their pretense of moral superiority is not warranted by these symbolic acts.
This ambiguous classification highlights that perceptions of genuineness (often taken as the sine qua non of authenticity) are a necessary but not insufficient condition for ascribing this honorific appellation to an object or action. That is, an entity or action may be deemed as genuine (i.e., not fake) but lack the perceived aesthetic or moral virtues needed to be classified as "authentic," or, conversely, to have its authenticity challenged. Thus, we can have marketplace conditions where authenticity, in its full moral and aestheticized sense, is not a relevant cultural category.
To illustrate, barring extenuating circumstances, consumers seldom venerate conventional mass-produced goods (e.g., a Big Mac, a Gillette disposable razor) for their authenticity because they lack potent associations with rarefied aesthetic ideals. Conversely, such items are not typically classified as inauthentic either (assuming that they are not knock-off products), with companies often promoting the standardized nature of their branded offerings—and the resulting performance consistency—as value-added benefits.
In the context of the elitist critique, the "not inauthentic + not authentic" classification suggests that middle-class consumers who support conscious capitalist brands and enterprises are, owing to their class privileges, inherently "not authentic." This disauthenticating implication hinges not on conscious intent but on the systemic advantages afforded by their relatively privileged socioeconomic position. Rather than being hypocritical per se (i.e., authentic + inauthentic), the implication is that such consumers may genuinely believe that conscious capitalism is a viable means to create a more equitable and just society. However, their genuine belief is an ideological one, steeped in their internalized class interests. Middle-class consumers' ideological affinity for the market logic of conscious capitalism allows them to lead a materially privileged lifestyle in a guilt-free manner ([92]). By purchasing brands and goods that signify a heightened social consciousness (e.g., fair trade coffee; TOMS shoes; organic, locally sourced foods), they can feel symbolically absolved from culpability in the perpetuation of socioeconomic inequalities. Consequently, their habituated class predilections also create an ideological blind spot toward the exclusionary signals that conscious capitalist ideals and values convey to those who lack the economic and cultural resources needed to fully participate in a middle-class lifestyle ([47]).
This ambiguous classification suggests that conscious capitalists' products, services, and brands are, to use[66] term, "gimmicks" that always promise more than they can deliver. [21], p. 668) further discuss issues relevant to this classification in their typology of transactions. Among their designations of deceitful transactions (i.e., scams), they list "fraud" and "confidence games." In the former condition, a disingenuous party misrepresents their intentions to an unsuspecting partner; in the latter condition, the scammer actively enrolls their target in the ruse, such as in catfishing and pyramid schemes.
The inauthentic + not authentic classification implies a manipulative opportunism whereby an unethical agent feigns genuineness to extract ill-gotten gains from another. In the context of conscious capitalism, this disauthenticating association is most germane to the marketer side of the exchange. As one well-known example, the business ethics journalist Jon Entine accused the pioneering conscious capitalist brand The Body Shop, and its founder Anita Riddick, of fraudulent misrepresentation. According to [25], Riddick stole the brand concept from a local entrepreneur and fabricated an authenticating origin story about traveling the world searching for natural skin care and hair treatments. His exposé further contended that The Body Shop significantly overstated the percentage of profits that it donated to philanthropic causes. Though Riddick formally denied these charges, the authenticity challenges posed by these accusations, as well as others that subsequently followed, continued to plague the brand ([76]). After years of underperformance, relative to the brand's prescandal pinnacle, The Body Shop undertook a revitalizing strategy that its management characterized as an activist revamp ([77]).
The "inauthentic + not authentic" classification can also cast more nuanced doubts on the authenticity of conscious capitalist entrepreneurs. Although such conscious capitalist entrepreneurs would not be committing overt acts of fraud (i.e., they are not lying about their business practices per se), the disauthenticating implication is that they are cynically espousing higher-order civic ideals to serve commercial ends, such as charging a premium to their consumers or driving higher stock valuations. This disauthenticating association can arise, for example, when the founder/chief executive of a privately owned conscious capitalist brand sells its rights to a larger corporate entity. Such a backlash arose when Gene Kahn—the founder of Cascadian Farms—sold his business to General Mills. Many leading voices in the organic food community lambasted Kahn's integrity, condemning him as a Boomer sellout and warning that the brand's corporate ownership would not stay true to the higher-order values that originally galvanized the organic food movement (see [75]).
To investigate how Slow Food advocates negotiate the elitist critique of conscious capitalism and its disauthenticating connotations, we recruited informants from a Slow Food chapter located in a metropolitan area of the Midwestern United States using informational flyers, contacts made at local chapter meetings, and snowballing referrals. We conducted interviews at public locations such as coffee shops or at Slow Food–sponsored events, with exception of two that respectively occurred in these participants' domestic residence and private work office. Interviewees were paid $20 in appreciation for their time. Interviews were audiotaped and ranged from one to four hours in duration, yielding 830 double-spaced pages of verbatim text. All participant names are pseudonyms.
Of our 19 interviews, 8 were conducted with chapter organizers, 5 with Slow Food producers and entrepreneurs, and 6 with Slow Food advocates who had volunteered their time to different outreach activities (for our participants' profiles, see Table 1). Most of our Slow Food organizers and consumer advocates are college graduates employed in professional occupations and hail from middle- and upper-middle-class families. Among the entrepreneurs, Dave, Leslie, Maggie, and Tom are also college graduates. This demographic profile matches the membership ranks of Slow Food USA, which skews toward middle-class professionals ([16]).
Graph
Table 1. Participant Profiles.
| Pseudonym | Gender | Age (Years) | Education | Occupation | Slow Food Role |
|---|
| Aaron | M | 28 | Ph.D. | Research scientist | Middle-class advocate |
| Alex | M | 28 | Some college | Artisan cheese maker | Entrepreneur |
| Amanda | F | 31 | MBA | Small business owner | Chapter organizer |
| Ben | M | 61 | M.S. | Pursuing Ph.D. | Middle-class advocate |
| Brenda | F | 44 | B.A. | Former organic farmer, at-home mom | Middle-class advocate |
| Caroline | F | 56 | B.A. | Government employee | Chapter organizer |
| Chris | M | 47 | B.A | Software designer | Chapter organizer |
| Christina | F | 44 | M.S. | Nurse practitioner | Chapter organizer |
| Dave | M | 30 | B.A. | Organic farmer | Entrepreneur |
| Erin | F | 29 | M.S. | E-commerce food entrepreneur | Chapter organizer |
| Hank | M | 59 | M.S. | Interior designer, former chef | Middle-class advocate |
| Heidi | F | 58 | B.A. | Professional chef | Middle-class advocate |
| Jane | F | 68 | Associate's degree | Holistic medicine practitioner | Middle-class advocate |
| Kevin | M | 42 | B.S. | Software designer | Chapter organizer |
| Leslie | F | 23 | B.S | Organic farmer | Entrepreneur |
| Maggie | F | 34 | B.S. | Free range farmer | Entrepreneur |
| Paula | F | 49 | M.S. | Retail buyer and jewelry maker | Chapter organizer |
| Richard | M | 34 | B.S. | Computer systems analyst | Chapter organizer |
| Tom | M | 35 | M.A. | Farm-to-table restauranteur | Entrepreneur |
1 Notes: M = male; F = female; B.A. = bachelor of arts; B.S. = bachelor of science; M.A. = master of arts; MBA = master of business administration; M.S. = master of science; Ph.D. = doctor of philosophy.
Following the conventions of in-depth phenomenological interviewing ([85]), our participants largely determined the course of the dialogue. The interviewer relied on follow-up probes to elicit more detailed accounts of the informants' experiences and viewpoints and to ensure that various aspects of food production, distribution, and consumption were covered. Procedurally, our interpretation developed through an iterative process of creating, challenging, and reworking provisional understandings by tacking back and forth between individual transcripts and the broader data set ([84]). We then pivoted to another level of hermeneutic tacking that entailed iterations between these emic narratives and theoretical concepts, which led us to the application of the semiotic square and our resulting focus on the elitist critique and corresponding strategies for countering the authenticity challenges posed by the cultural contradictions manifest in this market system.
As an institutional entity, Slow Food is a transnational organization encompassing 1,500 local chapters plus numerous subsidiary organizations. Beyond its formal institutional boundaries, Slow Food's culinary practices, values, and activist goals organize ideological and economic alliances among a globally diffused network of food writers (such as Mark Bittman and Michael Pollan), consumers, producers, merchants, and restauranteurs (including celebrity chefs Alice Waters and Jamie Oliver). As [18], p. 131) writes, "The phrase 'slow food' strikes a chord among the public not because it is the name of an organization but because it reflects a series of desires, interests and concerns."
Slow Food discourses valorize meals that are traditionally prepared with fresh ingredients as unique sources of pleasurable experiences that can mobilize consumers to resist the industrialized system of food production. Over the years, the Slow Food movement has embraced a broader conscious capitalist agenda that advocates for sustainable production, environmental protection, and social justice (i.e., fighting hunger, advocating for living wages for agricultural workers; see [16]; [89]).
Like other conscious capitalist exemplars, Slow Food has also been plagued by charges of elitism from its inception in 1986 when its founder, Carlo Petrini, organized a series of public protests over the opening of a McDonald's in the heart of Rome (see [89]). This ignominious view of Slow Food finds ready expression in both academic analyses (e.g., Guthman 2007; [56]) as well as journalistic accounts, such as [38], who states that "none of the aggressive, judgmental pitches of the movement have ever been proven. The power of its association with the economic elite has."
From this skeptical standpoint, Slow Food's exalted rhetoric of sustainable diets, biodiversity, and socially conscious eating (see https://www.slowfood.com/about-us/our-philosophy/) is a guise for privileging upper-middle-class tastes over the dietary practices of less affluent (and lower-cultural-capital) consumer segments (see [53]; [56]). Even Slow Food's ardent proponents, such as food writer Annie Levy, concede that a tacit elitism has hindered the cultural diffusion of its core principles:
—The revered Alice Waters once said, "when we eat food that is fast, cheap, and easy, we digest those very values." What are the judgments contained in this kind of statement? She intends, I believe, to critique the values of a food system that doesn't care about its conditions or effects on people and the environment. But the words suggest that if you eat fast, cheap, and easy you become fast, cheap, and easy—language many women might recognize as shaming. Isn't this how it really sounds to someone who enjoys such food, or is caught in situations in which it might seem the best available option? ([58])
Slow Food encourages consumers to shift their culinary tastes away from fast food and industrialized fare (including the oft-demonized category of junk food) ([16]; [83]). Such admonitions can imply a moralistic hectoring and an invidious comparison with those whose food tastes and practices are more orthodox. These elitist associations often cross into other sociocultural spheres, such as the controversy sparked by former First Lady Michelle Obama's school lunch initiative, which was institutionalized through the Healthy, Hunger-Free Kids Act of 2010. By explicitly disavowing fast food and processed foods, the revised school lunch guidelines dovetailed with Slow Food's mobilizing agenda—an alliance that Slow Food USA was eager to promote (see Figure 2).
Graph: Figure 2. Fodder for the elitist critique: Slow Food's controversial political alliances.
Once these Slow Food–friendly standards went into effect, news (and social) media began to feature anecdotal reports of children refusing to eat these presumably unpalatable lunches and skyrocketing food waste ([28]), with some critics characterizing the program as "gastro-fascism" ([71]). The elitist charge became integral to this cultural (and political) backlash:
Michelle Obama thinks she knows what your children should eat. She's adamant about promoting her nutrition policies for kids, even the new and disastrous school meal standards implementing the "Healthy, Hunger-Free Kids Act."... But attending Ivy-League schools doesn't magically make someone better parent material than an individual who attended a public university, or, dare it be said, someone who didn't attend college. ([ 4])
Like other market exemplars of conscious capitalism, Slow Food's aesthetic and experiential arguments have become strongly associated with an unwarranted moralism (i.e., the authentic + inauthentic classification; primary relation of contrariety). Slow Food advocates frequently argue that fast food is a debased cuisine that deprives humanity of meaningful and rewarding experiences of eating and sociability ([73]; [83]). For the many consumers who have warm memories of enjoyable fast-food meals with friends and family (and maybe look forward to such treats), Slow Food's moralizing pronouncements seem to emanate from an elitist taste bubble that is disconnected from everyday pleasures and real-world practicality. Similarly, Slow Food's veneration of locally sourced ingredients, heirloom vegetables and grains, artisan-crafted foods, and seasonal cuisine also seems to assert an unwarranted claim to moral virtue. Rather than sacrificing for a greater societal good, such rarefied culinary objects seem more attuned to signaling that one possesses the discretionary resources of time and money to treat food and cooking as a self-actualizing identity practice. Accordingly, Slow Food is easily, via the elitist critique, decried as an aggrandized form of cultural snobbery (e.g., [53]).
Figure 3 represents the correspondences between Slow Food's contextualized authenticity contradictions, the disauthenticating association that ensues from each contradiction, and the strategies through which Slow Food advocates seek to negate these authenticity challenges. In this representation, indexical authenticity (authentic ↔ not inauthentic) is the contested ideal that our Slow Food advocates are seeking to defend.
Graph: Figure 3. Authenticity contradictions and authenticating strategies in the Slow Food market.
Our Slow Food consumers place the most emphasis on the reflexive strategy, which they use to counter the authentic + inauthentic contradiction (primary relation of contrariety) and its disauthenticating association of virtue signaling and moral pretentiousness. Rather than rejecting the elitist critique outright, Slow Food advocates interpret it as a warning sign that the Slow Food market has become a gentrified facsimile of the movement's origins in the everyday cuisines of rural Italians (i.e., a disparaging version of iconic authenticity). Accordingly, our participants revere practices that seem to resurrect Slow Food's agrarian values and democratizing goals.
The humanistic rebel strategy redresses the not inauthentic + not authentic contradiction (secondary relation of contrariety) and its disauthenticating association of social exclusion. In the context of the elitist critique, this contradiction holds that individuals whose lives have been shaped by class privilege may be blithely unaware of their own internalized elitist predispositions. From this standpoint, Slow Food advocates may have a genuine interest in making the world a better place (i.e., they are not consciously "faking it"; rather, they are being "not inauthentic"). However, they are largely oblivious to how their viewpoint on these problems and solutions has been shaped by a life of class privilege and their habituated, middle-class (bourgeoisie) sensibilities. This disauthenticating association renders Slow Food consumers as being somewhat akin to the proverbial fish in water. Rather than not realizing they are wet, however, the analogical implication is that they cannot comprehend that other terrestrial animals lack the requisite resources to enjoy life in the water, as they do.
To negate this authenticity challenge, our participants drew from humanistic rationales, such as the idea that certain kinds of experiences and social connections have magical and transformative qualities that transcend social differences ([ 2]). Importantly, this strategy combines a humanistic ethos with the idea of rebelling against a deleterious marketing and cultural status quo and, thereby, creates a distinction to the complicit, part-of-the-problem connotations of liberal elitism (see [92]).
The perfective strategy corresponds to the inauthentic + not authentic contradiction (relation of complementarity). This strategy is most relevant to those positioned on the entrepreneurial/production side of this market system. It aims to negate the disauthenticating association of commercialism (i.e., conscious capitalist enterprises are profit-seeking marketing ploys). In response, our Slow Food producers and entrepreneurs draw from the bohemian ideal of the artist who refuses to compromise their artistic vision, despite market incentives to "sell out" (i.e., betraying one's artistic integrity in return for financial reward) ([10]; [87]). Accordingly, they present themselves as being intrinsically committed to perfecting their Slow Food craft and pursuing conscious capitalist values and ideals, rather than doing it for the money.
Slow Food advocates use the reflexive strategy to negate the authenticity challenge of moral pretentiousness. The implication is that Slow Food assigns an unwarranted degree of moral virtue to those who have the economic wherewithal to buy rarefied ingredients, spend time on complex meal preparations, and dine at expensive farm-to-table restaurants while casting those who lack such resources as less virtuous consumers. In response, our participants interpret Slow Food's cultural associations with affluent foodies and elite taste practices as a regrettable, but correctible, market distortion of the movement's authentic values and practices.
While acknowledging that market upscaling has imbued Slow Food with an elitist aura, our participants reiterate that expensive, epicurean cuisine need not be and, indeed should not be, regarded as the quintessential expressions of Slow Food:
I think one of the things is this perception that if you shop at farmers' markets or at the co-op, it's a lot more expensive. And there is a little bit of this Slow Food bent into cooking elaborate meals, and I think some people perceive that as being elitist because it's sort of this educated way of thinking about food. I don't think of it as being elitist because a lot of times, recipes can be super expensive to buy all the ingredients for, but they don't have to be. I don't think that enjoying your food should be something that is thought of as elitist.... Like, I buy what's not super expensive at the co-op and I cook pretty simply.... What I really like about Slow Food in particular is the aspect of enjoyment and that good food is for all—what good, fair, clean food means for the farm worker to the people who are consuming the food. (Erin)
Erin does not summarily dismiss the elitist charge. Rather, she takes a more ambivalent stance by first conceding that Slow Food values and ideals are often enacted in ways that can be read as elitist, such as cooking elaborate meals using expensive ingredients. In her authenticating interpretation, she counters that Slow Food values are better expressed through fundamentally simple meals that do not require extensive preparation time or costly ingredients. Her emphasis on there being many affordable options at her food co-op (which is, of course, a relative judgment) and on "cooking simply" (another relative assessment) convey that she is staying true to Slow Food's core principles rather than trying to place superficial, foodie predilections on a higher moral plane.
Erin further counters this aspect of the elitist critique by incorporating the economic interests of farmers into her inclusive interpretation of Slow Food stakeholders. This interpretation creates a rhetorical contrast between Slow Food's foundational discourse of economic populism (emphasizing fair wages for agricultural workers) ([73]) and the elitist condemnation that higher prices are merely a means for affluent consumers to mark status distinctions.
Paula's narrative exhibits a similar authenticating logic to that expressed by Erin:
Slow Food has often come under fire for being elitist. I don't actually think that's true.... The beginnings of Slow Food were about people eating good food, and those were not necessarily rich people. We are talking about people who might have had very little money.... When most people think about amazing Italian cuisine, they were eating very basic foods. So, the whole idea of eating good food to me doesn't seem elitist at all.... Slow Food in the United States, yes, we do certain things that might be seen as elitist—the restaurant dinners and stuff like that. But again, you are still educating people. You are still getting more people involved. And the more people who know about local farming, sustainable farming, eating seasonally, making sure that farm workers are protected and paid properly, that spreads out. And we do projects with a variety of different populations, and we are trying to do more of that.... Slow Food does a lot of work in all its chapters to help with urban gardens or school gardens.... In the long run, our goal is that all people have access to this kind of food.... We are working toward passing that power on to more people. So, I don't think wanting children and families in need to have high-quality food is elitist. (Paula)
In this vignette, Paula first differentiates Slow Food practices from elitist pretentions by invoking its historical connections to rustic Italian foodways. She asserts that Slow Food enjoins a pleasurable, resourceful, and fundamental relationship to food that should be accessible to people from all walks of life, rather than being an exclusive province of affluent consumers. However, Paula also recognizes that her inclusive rendering of Slow Food is contradicted by the realities of socioeconomic stratification. From Paula's viewpoint, Slow Food's community outreach efforts can play a pivotal role in democratizing these forms of culinary cultural capital so that consumers from less privileged backgrounds can acquire the skills and knowledge needed to incorporate Slow Food practices and ideals into their culinary routines.
When utilizing this reflexive strategy, Slow Food advocates routinely assert that cultural capital ([ 9]), rather than a lack of economic resources per se, is the primary barrier that keeps consumers from integrating Slow Food ideals and practices into their everyday lives. Christina echoes this rationale when discussing how low-income consumers could enact Slow Food practices if they had more knowledge about utilizing the fresh produce and bulk goods that often go to waste in the local food pantry where she volunteers:
Some people who are in Slow Food are foodies. However, it does not cost a lot of money to eat right. There are food pantries who throw away produce because people who come to the pantry don't know what to do with it and they don't take it.... Fresh produce going to waste.... No one wants it because they don't know what to do with it. It's really unfortunate. So, people have more access than they think. There are bulk aisles at grocery stores that you can get food for less money. It is actually a lot less expensive to buy bulk rice or bulk oats or whatever else, than to buy the bagged, boxed stuff that's like creamy preprocessed. I think the real lack of resource is education, not so much money. (Christina)
Slow Food advocates often draw an authenticating contrast between foodie-ism—which fetishizes highly aestheticized meals prepared with exotic (and typically expensive) ingredients (see [52])—and the Slow Food ethos of preserving traditional foodways and skills ([75]). This distinction is quite salient to Slow Food chapter leader Kevin. He posits that Slow Food's culinary values and ideals have become misconstrued in their translation to the consumerist, status-conscious culture of the United States. Kevin's goal is to reclaim Slow Food's original ethos from its commercial appropriation by high-end retailers and restaurants:
To buy imported cheese, organic wine, and all these kinds of things, I don't think those are meant to be the most obvious expressions of Slow Food values. And I think this is where the cultural translation from Italy to the United States went wrong, is that it got tied up with those folks [i.e., affluent foodies]. In Italy, it's much more about cooking at home. It's much more about preserving grandma's recipes. It's much more about celebrating the seasons and the tradition and preserving home ways of life than it is about eating in restaurants that do everything right. And you know, like anything else, capitalism wants to subsume this revolution.... That's a schism that I am personally trying to address and maybe lead by example. I don't think we should be cooking like a Michelin-starred restaurant at home. I think we should be cooking like our grandparents and great-grandparents. And I think we can learn a lot from traditional cultures and indigenous people—to the extent that there still are any indigenous people—how to eat well, and you know, a lot of those foods have become an affectation in restaurants. They'll have poutine, but it's made with truffles, and confit duck and elaborate things. I have realized that we are all attracted to the comfort foods and the simple foods, like tacos, and they are easy and fun to make. (Kevin)
For Kevin, Slow Food should be accessible, basic, fun, and easy—characteristics that diverge from associations with rarity, cosmopolitan sophistication, aesthetic refinement, and technical proficiency that mark elite tastes ([44]). If read in a more critical light, Kevin's narrative reiterates the nostalgic glossing of preindustrial foodways that critics of the Slow Food movement assail for ignoring the harsh realities of scarcity and subsistence endured by those who had to survive on "traditional diets" ([56]).
While romanticizing images of a bucolic culinary past have considerable appeal to our Slow Food advocates, the idea of rekindling a premodern utopia is not that central to the reflexive strategy's authenticating function. Rather, these homages to a bygone era, when people lived close to the land and prepared food in traditional ways, symbolically link Slow Food practices to agrarian and/or rural lifestyles far removed from elite pretensions:
I did an internship through Worldwide Working on Organic Farms.... I went to Italy and I milked sheep and goats for a couple of months. And it was very rural. It was very low-tech. We milked in buckets by hand, sheep and goats, and we kind of went out with big sticks and sheepdogs and herded them.... Slow Food originated in Bra, Italy, and that was only like an hour and a half away from the farm. So, I think that's kind of how Marco [the farmer] was involved in Slow Food. He made cheese that was very well regarded, and he went to cheese festivals and stuff. But I mean the whole day was slow. Like wake up kind of late, drink your espresso, milk leisurely, walk the mount, you know. Dinner took a really long time, but that was kind of okay. And just kind of do the same things over and over again. So, we all cooked together. They also did some kind of agro-tourism. They'd have people from the city come out and we would cook with them the food we either grew or found.... That was fun. (Leslie)
Leslie's narrative validates a nexus of Slow Food ideals regarding small-scale production and the slower pace of agrarian life. While this rural setting has some trappings of a staged performance—most notably Marco's side business in agro-tourism—it places Slow Food in a symbolic sphere far removed from the Whole Foods brandscape, expensive farm-to-table restaurants, and other consumption domains invocative of foodie affectations (see [52]). For Leslie, her story of interning on a rustic Italian farm affirms that she has actually lived the conscious capitalist values she endorses through her Slow Food advocacy (and thus is not a hypocritical moralizer). More generally, consumer narratives that link Slow Food practices to traditional modes of food production, down-home family meals, and simpler ways of living—rather than exorbitantly priced gourmet dishes—express a rhetorical parry to the elitist charge of moral pretentiousness.
Our Slow Food advocates use the humanistic rebel strategy to redress the authenticity challenge posed by the elitist critique's connotation of social exclusion and the related sociological argument that consumers' social class backgrounds structurally predetermine their taste affinities ([44]). From this critical viewpoint, Slow Food advocates may not consciously intend to be elitists, but their preferences for goods that convey meanings of sustainability, locavorism, and artisanship betray a host of class advantages that distinguish them from consumers whose lives are marked by conditions of necessity ([47]). While Slow Food advocates may be well intentioned (i.e., they are being "not inauthentic"), they are also complicit in a system of institutionalized class-based inequities.
To illustrate this tension, let us reassess Leslie's preceding vignette in relation to this association with social exclusion (rather than moral pretentiousness). On the one hand, working for room and board on a small, rural farm clearly diverges from conventional notions of status posturing. However, a sociological counterpoint is that Leslie—as a college-educated young adult engaging in an exploratory experience—is building a reservoir of life stories and cultural capital that can afford career and status advantages later in life (see [91]). Seen in this sociological light, Leslie is enacting her class privilege by having the economic and social latitude to intern on a rustic, Italian farm before transitioning into more conventional middle-class occupational pursuits, such as attending graduate school.
As a chapter leader, Kevin has oriented his local chapter's activities toward the goal of making Slow Food a more inclusive organization that does not merely cater to the interests of middle-class consumers. When implementing these outreach projects, however, Kevin recognizes that many Slow Food practices are simply incompatible with the situational demands that lower-income consumers have to negotiate on a daily basis:
I have amazing privilege,... like being an American to a middle-aged white guy who has very marketable skills.... But I realize that that is a privilege, and this is the biggest thing for us in Slow Food to grapple with—that a single mother who has three jobs and two kids doesn't have the luxury of deciding, "I think I would like to work less and spend more time in my garden".... So, you have to be very careful and sensitive about it. (Kevin)
Kevin believes that this class chasm can be bridged if his team of middle-class volunteers presents Slow Food's approach to food provisioning, cooking, and eating in a manner that is "sensitive" (i.e., adapted) to the lifestyle constraints faced by less well-resourced consumers. His optimistic viewpoint is based on the authenticating assumption that Slow Food's experiential and social benefits have an inherent appeal to consumers, regardless of their class position, because they tap into fundamental human desires and needs.
The humanistic rebel strategy takes this inclusive rationale further by suggesting that a confluence of technological and commercial forces have locked individuals into an accelerating pace of life. Consequently, experiences of social connection, spontaneity, and everyday small pleasures are sacrificed to demands for efficiency, convenience, and the seductive (and ultimately alienating) effects of social media and digital communications.
Aaron echoes this humanistic rebel mantra in his commentary on the inherent importance of eating and cooking with others:
You have the social component of Slow Food as cooking and eating together and taking time to reflect and to connect and to develop community. That's something that we lose when we have things like drive-through or microwave dinners, which aren't to be destroyed or demonized altogether. I take advantage of these services of society. But for them to be the baseline means that we are losing what ... enriches our social systems a lot more than people eating alone and interacting through screens.... So, I think food, when it's jointly cooked and eaten, serves as a very natural medium for connection and idea generation and creativity. (Aaron)
Aaron characterizes Slow Food as a needed corrective to societal transformations in the practices of cooking and eating that have resulted in a loss of sociability, communal bonding, and creative interactions. By interpreting Slow Food as a medium for enjoining meaningful social interactions, Aaron expands its cultural province well beyond the predilections of affluent consumers. Aaron's caveat that he has partaken in the conveniences afforded by fast food and microwave meals is another rhetorical means for distancing his Slow Food preferences from elitist connotations. His narrative signals a conscious disavowal of snobbery by acknowledging that there is a legitimate time and place for "fast food." Aaron then specifies the problem as the cultural ubiquity of fast food, which, in turn, he links to dehumanizing consequences.
When expressing the humanistic rebel strategy, our participants often couched the communal experiences of preparing or eating food as magical moments that affirm Slow Food's class-transcendent qualities:
I ran a cheese-making class, and that was really fun because I've always been interested in cheese-making, just for the fun of it. There were seven or eight people there. And one of the members still is making cheese today. And it was really fun to be able to share that magic with people. That this is how this cheese actually comes into being, and it's totally doable by yourself at home. So, that's really exciting, that he got so inspired. It's fun to see somebody get really interested in something. (Amanda)
Amanda emphasizes the magic of cheese-making and the inspiration and rewarding personal experiences of self-sufficiency it can enjoin (i.e., "totally doable by yourself at home"). Her narrative expresses a cosmological view of nature as a magical, life-transforming force ([ 2]) whose authenticating properties are not inherently tied to class-shaped tastes. In keeping with this formulation, Amanda interprets the sharing of her Slow Food skills as a means to help others experience these inspiring connections to nature, which, in turn, implies a revelatory contrast to the alienated experiences of industrialized fast food.
Maggie similarly interprets her self-taught Slow Food skills as a means to help people create a sense of communal togetherness and to experience new sensory pleasures and magical connections to the land:
I think of Slow Food as taking your time to respect the ingredients and preparing them from scratch and enjoying food. And that, to me, resonates. And bringing back the social aspect of eating. Like, you take time to prepare this meal, you sit down, you share it with people who care about the same things that you do. And it's also, creating another community of people who value these things whether they're growing, or cooking, or eating; having that kind of common thread, I think is really satisfying. (Maggie)
Maggie venerates Slow Food as being inherently conducive to experiences of sociability and community and as a way to recognize the meaningfulness of food (a normative orientation that reads as a general human value, rather than a class-interested practice). However, Maggie is a meat producer who, like other Slow Food entrepreneurs, faces an additional task of negating contradictions that derive from the "inauthentic + not authentic" category.
The perfective strategy seeks to negate Slow Food's disauthenticating association with commercialism. This aspect of the elitist critique casts Slow Food producers and entrepreneurs as disingenuous actors who are enrolling consumers into an inauthentic market relationship (akin to a gimmick or a confidence game) to serve their own economic interests. In response, our Slow Food entrepreneurs strive to authenticate their actions by signaling that they would never compromise their Slow Food ideals for the sake of profit, such as by recounting the copious amounts of time and energy they invest into perfecting their Slow Food enterprises
In this spirit, Tom, a farm-to-table restauranteur, views his business as a way to enact his passionate commitment to producing food in a more meaningful and socially beneficial way:
When you go to a fast-food restaurant, you have no idea of who actually made that food and the process of where it came from is not known to you. The taste and flavor are mostly engineered to play off the cheap sensory sensations. So, it's fatty and salty and sweet and so, yeah, on a certain level, it might be gratifying, but it's a cheap way to do it that is less meaningful. Slow Food is like, "We're going to do things in a way that is process oriented!" I talk about process a lot.... We [Tom and his restaurant staff] were really structured around learning, and so it was a process where we feel like we've excelled and learned a lot and we'll keep pursuing that.... [With Slow Food,] you have this process where people are eating and making something and understanding where it came from and how it works. Eating is such an important part of our lives, and it can have a really important impact on our community and environment. So, the more you understand about it, hopefully you'll make better decisions. The basic motto of Slow Food is clean, fair, and good food. I can totally get behind those values. (Tom)
During his interview, Tom extensively discussed "the process" aspects of his restaurant and how he views it less as a business than a means to cultivate and diffuse knowledge about the complex interrelationships among food, cooking, sensory enjoyment, ecology, and societal well-being. His quote also reiterates Slow Food's argument that the experiences of these simple culinary pleasures can mobilize consumers to resist the industrialization of the food system (thus echoing the humanistic rebel strategy). For Slow Food entrepreneurs, however, it serves the additional function of associating their enterprises with civic goals that stand distinct from conventional commercial aims.
Returning to Maggie, she raises pasture-fed rabbits for sale to farm-to-table restaurants and consumers. In developing her production techniques, Maggie has constantly experimented with different procedures and equipment designs. Through this long trial-and-error process, Maggie believes she has developed an innovative method that better simulates the lives her rabbits would enjoy outside of captivity:
Maggie: Daniel Salatin is the son of Joel Salatin, who is the owner of Polyface Farms, and he is the person who is raising rabbits in this system that he has devised and calls the Hare Pen system. So essentially, you still have your does in cages.... You put them in a glorified cage that you then put on grass.... I've copied their system exactly, and I was very unsatisfied with the results that I got. [Maggie then provides an extensive description of her alternative and labor-intensive system and how she developed it]... I don't know why I kept doing it. But I finally have a system that is really effective.... It just was a lot of observation of the rabbits on pasture, making so many mistakes and then incorporating what I had learned.
Interviewer: Did you have any economic incentives?
Maggie: No! It has to be a personal belief that there might be a better way to do things.... It's kind of like what makes an artist a good artist. If they all hold the brush the same way and they are using the same colors, but they create vastly different things, and one appeals to you, and one doesn't appeal to you. So, what makes that one piece of art recognized by the vast majority of people as superior?... I have this wonderful platform to invest energy and creativity, and it's nice. And so, I feel in some ways really lucky.
Invoking the image of the passionate artist, Maggie distinguishes her efforts to perfect an ecologically appropriate system for raising rabbits from crass commercial and economic interests. Maggie's closing sentiment expresses her authenticating belief that such actions can make things better, rather than being driven by instrumental aims. Through storytelling, and by showing how her system works to customers who visit her farm, Maggie deploys narrative and material resources to negate disauthenticating concerns that her Slow Food affinities are merely an instrumental means to charge higher prices. Her personal investment in learning about rabbits' natural habitats/behaviors and inventing a complex ecosystem for raising them further signals that she is not likely to compromise her Slow Food principles in the interest of commercial expediency.
Prior research has treated authenticity as a perceptual value or quality that consumers attribute to a brand ([11]; [55]), person-brand ([87]), product ([57]), or performance ([ 5]; [39]). In contrast, we have reconceptualized authenticity as an ongoing process through which consumers and marketers negotiate a contextualized system of cultural contradictions and ambiguous classifications. We suggest that our semiotic framework can better analyze the authenticity contestations that arise in a given market or sociocultural context than conventional theories that assume authenticity perceptions operate on a continuum or selectively draw from an essential set of defining attributes.
The conceptualization of authenticity as a relative point along an authentic-to-inauthentic continuum ([22]; [65]) can depict a zone of ambiguity where the authenticity or inauthenticity of a market actor is perceived as being uncertain and, thus, debatable. However, this conceptualization does not offer a means to specifically analyze the cultural meanings (and the interrelationships among them) that generate these ambiguous perceptions. Accordingly, it offers limited theoretical discrimination and managerial guidance.
For example, [65] argue that quality commitment, heritage, and sincerity are the primary perceptual cues of authenticity. They then propose that brands should differentially leverage these cues depending on whether consumers perceive them as having low, moderate, or high levels of authenticity. In their normative framework, brands with low perceived authenticity should emphasize sincerity, brands with moderate perceived authenticity should emphasize quality and heritage, and brands with a high level of perceived authenticity should emphasize all three authenticity cues.
Such recommendations presume that brands falling into the lower and middle sectors of this proposed continuum have a shortfall of perceived quality commitment, sincerity, or heritage that is rectifiable through compensatory signaling. However, such contested brands are often plagued by contradictory meanings that undermine their promoted claims to authenticity ([37]; [86]). Furthermore, more complex, disauthenticating narratives, such as the elitist critique, can cast doubt on the very credibility of such authenticating cues when used by a contested brand or actors in a market system.
Turning to combinatory definitions, [67] have offered a comprehensive theorization of authenticity (as understood from the consumer's perspective) that warrants comparison to our approach. They identify six subdimensions of authenticity (accuracy, connectedness, integrity, legitimacy, originality, and proficiency) and then trace out the relative impact of those dimensions across different market categories and on consumers' behavioral intentions. Rather than a continuum, Nunes, Ordanini, and Giambastiani argue for a family resemblance explanation in which "a concept (authenticity, in this case) may be qualified by different subsets of its dimensions across different contexts, and not always by all of them in the same way" (p. 16).
Like a continuum, [67] family resemblance logic is limited to the explanation that authenticity is a multidimensional construct whose subcomponents may be more or less important in a given market or consumption context. In contrast, a semiotic framework shifts attention from correlational premises (e.g., this authenticity subdimension seems more important for hedonic products than utilitarian ones) to the cultural meanings, and underlying structural contradictions, relevant to a particular judgment regarding the authenticity of a given product, brand, or market action. For example, the authentic ↔ inauthentic tension elevates the importance of authenticity's moral dimensions in ways that traverse product category distinctions, such as hedonic or utilitarian.
To illustrate, a hamburger would typically be classified as a hedonic good. [67], pp. 3–4) find that judgments of "legitimacy"—which they define as "the extent to which a product or service adheres to shared norms, standards, rules, or traditions present in the market ... appear to matter for utilitarian but not hedonic products." However, if we examine this consumer choice in the context of the Slow Food market, then legitimacy becomes a far more important issue. From this standpoint, an "authentic burger" would need to exhibit fidelity to various aesthetic and moral norms—grass-fed beef, local sourcing, traditional preparation techniques, and so on—and, its perceived authenticity would be understood and legitimated through a contrast to fast-food burgers. That authenticating contrast (the fast-food burger vs. a Slow Food burger) could then become subject to the elitist critique, which, in turn, would provide motivation for Slow Food advocates to negate these disauthenticating associations.
In summary, we have argued that authenticity is culturally constructed (and contested) in a network of structural relations (rather than being a discrete set of essential properties attributed to a brand, person-brand, market performance, or market relationship). Consumers and marketers alike covet indexical authenticity (i.e., the abstract ideal of authenticity) because it can confer cultural legitimacy ([51]), moral authority ([59]), and identity validation ([ 7]; [86]) all of which, can be converted into micro-celebrity status ([79]) and a branding asset ([31]; [45]). However, this authenticity ideal is structurally linked to contradictory meanings and ambiguous classifications. When consumers' or marketers' authenticity claims are challenged by these cultural contradictions, they have pressing incentives to distinguish their actions and identities from the invoked disauthenticating associations. In the following subsection, we discuss how this authenticating goal can be enacted by negating associations that flow along the contradictory path of deception and promoting those that follow the contradictory path of redemption.
As [49] have argued, marketing managers often find it difficult to redress brand image problems because they are unable to effectively decipher the cultural meanings contributing to those dilemmas. Our semiotic framework can help redress this managerial shortfall. It offers a tangible means for marketing managers to systematically analyze the cultural contradictions of authenticity that emerge in a given market and then to identify strategies for authenticating their brands in the face of these challenges.
As a general heuristic, we propose that marketers can be successful in authenticating their brands and/or other strategic assets when they are able to accomplish two complementary goals. The first is to leverage cultural meanings that negate the disauthenticating associations that flow along the contradictory of deception path (authentic → not authentic; see Figure 1). When consumers follow this perceptual path, they experience a glaring contradiction between a prevailing ideal of authenticity and its market manifestation in a brand or marketing practice (authentic ↔ not authentic), which then leads to an association of inauthenticity via the complementary relation of not authentic → inauthentic. In response, marketers should try to provide consumers with compelling and emotionally resonant meanings and rationales that discount the credibility, relevance, or importance of the disauthenticating associations that have gained cultural currency in their respective market.
As one illustration, Patagonia confronted a path of deception authenticity challenge soon after it began campaigning against the Trump administration's executive order to reduce the size of Utah's Bears Ears National Monument by two million acres. On December 4, 2017, Patagonia featured this message on the front page of its website: "The President Stole Your Land: In an illegal move, the president just reduced the size of Bears Ears and Grand Staircase-Escalante National Monuments. This is the largest elimination of protected land in American history." This web page then directed consumers to various information sources and encouraged consumers to contact their elected officials and to also take the protest to social media, using the hashtag #MonumentalMistakes (see [ 1]).
However, defenders of the administration's policy change were quick to denigrate Patagonia's activism as a deceptive marketing ploy. Interior Secretary Ryan Zinke condemned Patagonia as a dishonest "special interest" and proclaimed it was "shameful and appalling that they would blatantly lie in order to put money in their coffers." Utah Representative Bob Bishop, then chairman of the House Natural Resources Committee, also evoked the elitist critique in his tweet proclaiming that "Patagonia is Lying to You... A corporate giant hijacking our public lands debate to sell more products to wealthy elitist urban dwellers from New York to San Francisco" (quoted in [35]).
In terms of our model, the Trump administration's response challenged the authenticity of Patagonia's mobilizing campaign by impugning its motivations, thereby reframing an ostensibly authentic (conscious capitalist) action as a disingenuous public relations stunt designed to extract more profits from elite consumers (authentic → inauthentic), which, in turn, triggers the complementary association to inauthenticity. In response, Patagonia joined as a coplaintiff with five Native American tribes and several nonprofit groups in a lawsuit aiming to halt the policy change ([35]; see also https://www.patagonia.com/stories/hey-hows-that-lawsuit-against-the-president-going/story-72248.html). Patagonia also continued to be a vocal critic of the Trump administration's environmental policies and, in a politically and ideologically related vein, donated all its tax savings from the Trump-backed corporate tax cut to environmental groups while condemning the new corporate tax rates as being irresponsible ([63]). Through these responses, Patagonia signaled a deeper commitment to its conscious capitalist values and gave consumers reasons to doubt or dismiss the disauthenticating associations of greed and deception that were being cast on it. In response to Patagonia's uncompromising stance, Inc. offered the following commentary on its 2018 Company of the Year finalist:
For Patagonia and its fans, that purpose is doing whatever they can to try to save the planet. In 2018, Patagonia proved that it will not only preach that mission, it will do so with a much louder voice than most other companies. And—so far, anyway—it's only further burnished the Patagonia brand. ([ 8])
The second marketing goal is to create conditions in which consumers interpret a brand or business along the contradictory of redemption path (inauthentic → not inauthentic) (see Figure 1). This redemptive chain of associations begins with the widespread cultural view of marketers as inauthentic and, thus, untrustworthy actors ([45]; [66]; [67]). Accordingly, we can assume that consumers will typically harbor varying degrees of skepticism and suspicion toward the authenticity of marketing and branding claims. Redemptive meanings encourage consumers to believe that a given brand or business is operating in ways that favorably diverge from the marketing status quo (i.e., not inauthentic) which, in turn, leads to the complementary relation of not inauthentic → authentic.
Volkswagen's (VW's) "Hello Light" advertisement, which launched its new line of electric vehicles (circa 2019), takes viewers on a journey that follows a path of redemption arc (see https://www.youtube.com/watch?v=qEvNL6oEr0U). The ad begins with a silhouetted figure entering a dark and seemingly abandoned production facility, while a news report about VW's emission scandal, or "Dieselgate," blares in the background. In a seemingly counterproductive marketing communications move, the ad explicitly reminds its viewers of all the inauthentic associations (VW as liar, deceiver) that arose from those "dark" days. The protagonist is revealed to be a despondent engineer struggling to design a new VW model, against the musical backdrop of Simon and Garfunkel's 1960s anthem, "The Sound of Silence." Desperate for inspiration, our engineer scours the company archives and finds his creative muse—an image of the iconic VW Van (aka the "Love Bus").
Through the choice of song and reference to this totem of the 1960s counterculture, the ad recalls VW's countercultural legacy as an authentic symbol of antimaterialist values and a rebuke to status consciousness and marketing hype ([46]). The ad's message is that VW, despite having lost its way, still possesses a latent essential "goodness" that is "not inauthentic." As "The Sound of Silence" reaches its crescendo, lights go on, puncturing the darkness. We observe the production facility come to life and give metaphorical rebirth to the VW brand in the form of an electric van (which also places "The Sound of Silence" on a different, ecofriendly cultural register). Thus, the ad's narrative follows the redemptive path of "inauthentic" (VW's Dieselgate) to "not inauthentic" (VW's 1960's countercultural heyday) to "authentic" (signifying that VW has rekindled its socially conscious roots).
Our discussion of the three authenticating strategies used by Slow Food consumers and entrepreneurs has emphasized their function as a defensive means to disavow or negate the disauthenticating associations that flow along the contradictory of deception path. However, these same strategies also promote affirmative meanings and associations that operate along the contradictory of redemption path. For example, the perfective strategy does more than negate the disauthenticating association with commercialism. It also magnifies authenticating differences to fast food or industrialized food production and thereby encourages consumers to interpret Slow Food enterprises in a manner compatible with the contradictory of redemption path (even though they may be aware of some disauthenticating associations). This redemptive associative chain takes the following form: Slow Food entrepreneurs are suspected to be "inauthentic" due to their commercial motivations → Slow Food entrepreneurs are seen as being "not inauthentic" because their deep commitment to artisan ideals and noncommercial values differentiates them from conventional fast-food establishments and industrialized modes of commercial food production → the signification of "not inauthentic" supports a broader conclusion that the Slow Food entrepreneur is an authentic actor.
Given their shared ideological affinities, the authenticating strategies used by Slow Food advocates should also have a high degree of applicability to brands espousing conscious capitalist goals and ideals. Of the three authenticating strategies, we find numerous examples of conscious capitalist brands that have enacted some version of the perfective strategy, which aims to negate associations with commercial opportunism and foster interpretations compatible with the contradictory of redemption path. While less commonplace, we can also find branding campaigns that align with the reflexive and humanistic rebel strategies. In the following discussion, we use these various exemplars to illustrate how these authenticating strategies can be implemented by conscious capitalist brands and the authenticity contradictions they potentially redress.
Brands using the perfective strategy engage in unconventional actions that demonstrate a deep commitment to activist causes that supersede profit motives. Over the years, Patagonia has made frequent use of this authenticating strategy to signal that proenvironmental values were central to its corporate mission, even when such acts could mean sacrificing sales, such as its iconic "Don't Buy" promotion ([49]) or their "Give a Damn" holiday messaging ([72]). REI has also enacted a perfective strategy in its #OptOutside campaign, whereby the retailer closes its stores on Black Friday and encourages consumers to engage in a range of proenvironmental, outdoor activities. Like Patagonia, REI's campaign builds on (and authenticates) the brand's history of supporting environmental causes and promoting a heightened concern for habitat protection and environmental conservation. Last but not least, Clif Bar illustrated a fairly novel implementation of the perfective strategy when its founder and chief executive officer, Gary Erickson, published an "advertorial" in the New York Times, offering to donate ten tons of organic ingredients to his main competitor Kind Bars. This advertorial further promised to share his company's knowledge about organic sourcing and production so that the two companies could collectively "lay the foundation for a healthier, more just and sustainable food system" ([26]).
Owing to their status as commercial enterprises, whose existence depends on profitability, the perfective strategies of conscious capitalist brands can always be reframed as yet another kind of commercial deception. However, such brands can lessen the cultural viability of such recursive challenges by further signaling that their passionate commitment to the supported causes takes precedence over profit motives. Though addressing a different context, [20] offer evidence that supports this strategic approach. They find that customers attribute the quality of authenticity to third-place establishments (e.g., cafes, coffee shops, restaurants) when they believe the respective proprietors are aiming to create meaningful social connections rather than merely trying to make a profit. As they write, "The authenticity perceived in treasured commercial places is based on exchanges that go beyond mere commercial aspects.... Although being business operators, proprietors invite the consumer to engage in activities that are not undertaken purely for profit" ([20], p. 913).
Accordingly, we propose that conscious capitalist brands are more likely to be perceived as authentic when they provide tangible means for consumers to participate in their social change mission but do so in ways that are not dependent on purchases. From this standpoint, The Body Shop's repositioning of its stores as activist hubs ([77])—where consumers can listen to speakers discuss environmental and social justice issues, sign petitions, and join activist organizations—is an enactment of the perfective strategy and a culturally viable means to reestablish the authenticity of its conscious capitalist branding claims.
This strategy aims to negate the charge that a market actor is evincing a "holier than thou" stance for actions that are either hypocritical (e.g., "do as I say, not as I do") or overstate the positive impact of the self-proclaimed act of conscious capitalist rectitude. For conscious capitalist brands, this authenticating logic most readily translates into a reformist agenda. As one prominent example, Chipotle's "Back to the Start" campaign (circa 2012) rallied a diverse assemblage of activist groups that shared a commitment to transforming the corporate-controlled system of food production and who saw the fast-food sector as exemplifying its presumed ills (see [48]). The two-minute short film, which ran across multiple media platforms, shows an increasingly disenchanted farmer witnessing the steady industrialization of his enterprise, replete with enclosed animals, the heavy use of antibiotics, and food being transformed into nondescript goo-like substances. Against the backdrop of Willie Nelson's plaintive version of Coldplay's "The Scientist," the farmer triumphantly decides to go "back to the start" by raising free-range animals, using traditional farming techniques, and selling his preindustrial goods to Chipotle.
Some relevant insights into this campaign and its authenticating effects can be gleaned from a Fast Company interview with Jesse Coulter, co–chief creative officer of Creative Artists Agency Marketing, which worked with Chipotle's management team in developing this campaign:
We were tasked to find new ways to tell Chipotle's Food with Integrity story.... The first issue Chipotle wanted to address was industrial farming.... Chipotle shared many stories of family farmers who have turned their farms into factory farms and have subsequently grown to regret it.... It was provocative because it took a stab at Big Agriculture. Chipotle is a bold company, who has the courage to really stand up for what they believe in.... At the end of the film, a title card appears letting people know that they can download the song on iTunes, and the proceeds benefit the Chipotle Cultivate Foundation, which is dedicated to creating a sustainable, healthy, and equitable food future. People responded and the song reached number one on the iTunes Country chart. ([15])
When interpreted through the lens of the reflexive strategy, Chipotle's "bold" move was to accept and amplify activist groups' criticisms of the fast-food industry's sourcing and production practices, rather than attempting to deny, rationalize, or obscure these problems through conventional images of consumers enjoying their fast-food meals. The campaign thereby draws an ideological distinction between Chipotle (as a reformist enterprise returning to more humane and sustainable agricultural practices) and the broader fast-food industry that is portrayed as having debased the time-honored practice of farming in the name of speed, efficiency, and cost reduction. In this way, Chipotle aligned itself not with the interests of the fast-food industry at large, but with activist groups who are seeking to reform the broader system of industrialized food production. Chipotle further reinforced the credibility of their reflexive strategy by investing additional resources to support the broader cause (i.e., creating a synergy between the reflexive and the perfective strategy).
This strategy promotes the brand as a means for reconstituting meaningful social connections and breaking down societal boundaries that artificially separate people. To avoid being just another nostalgic marketing ode, this strategy should take a critical stance toward selected status quo consumption and marketing practices. The intended message is that the conscious capitalist brand is enabling consumers to resist or escape the dehumanizing and/or isolating influences of materialism, status consciousness, and upward-ratcheting lifestyle competitions.
IKEA has run numerous campaigns that align with the humanistic rebel strategy. These campaigns embed its conscious capitalist commitments to sustainability and support of social justice issues, such as gender equity and LGBTQ rights, in a home-as-haven brand narrative. In these ads, the IKEA-furnished home represents a therapeutic space where people can, at least temporarily, unplug from the stresses and distractions of the "networked life" ([88], p. 17) and experience meaningful human connections and emotional fulfillment.
More than just a haven, however, IKEA often portrays the home as an active force that keeps at bay the outside forces that would interfere with the pleasures of slow living. In an ad titled "Home Is a Haven," we see a father and daughter running to their house during a rainstorm. As they approach the front door, the child's teddy bears spring to life as human-sized entities (whose muscular physiques resemble bouncers at a club). The bears rearrange the house into an open play area and protect the dad from intrusive calls and other outside distractions. We watch as father and daughter play dress-up and numerous other games, eventually falling asleep after their fully engaged bonding time (https://www.youtube.com/watch?v=wGgcYNlH02g).
IKEA's "Let's Relax" commercial presents a pointedly critical take on the performative, competitive affectations of Instagram micro-influencers, for whom everyday social activities are treated as an instrumental means to garner likes and followers. In the ad, we first observe an eighteenth-century family about to begin formal dinner in a very well-appointed dining room. Suddenly, the father halts the proceedings so that an artist can paint a portrait of the meal, which is immediately transported across the town in a horse-drawn carriage so that affirmative thumbs up gestures from the populace can be tabulated. The scene then suddenly shifts to a modern-day kitchen table, where the same father meticulously photographs the family meal, while his wife and children begrudgingly wait for this documenting ritual to end. The dad sheepishly retires his camera, and the family begins their more enjoyable and authentic social interactions, all framed by the closing caption: "Relax: It's a meal, not a competition" (https://www.youtube.com/watch?v=2BXRGzjo1%5fQ).
Drawing on our semiotic framework, we anticipate that conscious capitalist brands would gain the most authenticating benefit from the humanistic rebel strategy when they present their brands as ideological allies ([48]; [49]) of consumers who are sensitized to the psychological and social costs of careerism, exclusionary status hierarchies, and the calculated practices of social media self-presentations. In this way, the humanistic rebel strategy undercuts the elitist critique by suggesting that a conscious capitalist brand enables consumers to tap into more basic and rewarding emotional and sensory experiences. It further emphasizes that helping people, from all walks of life, feel genuinely connected to each other is an important and accessible way to make the world a better place.
Drawing from structural semiotics ([40]), we have developed a conceptual framework that can be used to analyze the cultural contradictions of authenticity, as they emerge in a given market context, and then to identify strategies for combatting their disauthenticating associations. Our analytic approach recognizes that perceptions of authenticity are constructed and contested in a dynamic cultural system. When negotiating such dynamism, marketing managers need to identify strategically significant patterns in the flux of cultural change and to adroitly react to cultural flash points, competitive shifts, and other exogenous shocks that could undermine the credibility of their existing authenticity claims. Whether undertaken in the context of conscious capitalist brands, status-marketing luxury goods, price-driven big-box retailers, or sharing-economy enterprises such as Uber or Airbnb, marketing managers can use our semiotic approach to more effectively negotiate the sociocultural complexity inherent to the process of authenticating their strategic assets.
Footnotes 1 Rob Kozinets
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 For purposes of expositional clarity, our analysis will not explicitly highlight these two paths of contradictory relations. However, they are implicit to the operation of the contrariety and complementarity relations and, therefore, are part and parcel of what we more generally characterize as the contradictions of authenticity. In our closing discussion, we address the implications that the deception and redemption paths hold for marketing management.
References Andrews Travis. (2017), "The President Stole Your Land': Patagonia, REI Blast Trump on National Monument Rollbacks, The Washington Post (December 5), https://www.washingtonpost.com/news/morning-mix/wp/2017/12/05/the-president-stole-your-land-patagonia-rei-blast-trump-on-national-monument-rollbacks/.
Arnould Eric J. , Price Linda L.. (1993), " River Magic: Extraordinary Experience and the Extended Service Encounter ," Journal of Consumer Research , 20 (1), 24 – 45.
Aschoff Nicole. (2017), "Whole Foods Represents the Failures of 'Conscious Capitalism,' The Guardian (May 29), https://www.theguardian.com/commentisfree/2017/may/29/whole-foods-failures-conscious-capitalism.
Bakst Darren. (2014), "Why Michelle Obama Is Wrong on School Lunches," The Heritage Foundation (June 24), https://www.heritage.org/public-health/commentary/why-michelle-obama-wrong-school-lunches.
5 Becker Maren , Wiegand Nico , Reinartz Werner J.. (2019), " Does It Pay to Be Real? Understanding Authenticity in TV Advertising ," Journal of Marketing , 83 (1), 24 – 50.
6 Beverland Michael B.. (2005), " Crafting Brand Authenticity: The Case of Luxury Wines ," Journal of Management Studies , 42 (5), 1003 – 29.
7 Beverland Michael B. , Farrelly Francis J.. (2010), " The Quest for Authenticity in Consumption: Consumers' Purposive Choice of Authentic Cues to Shape Experienced Outcomes ," Journal of Consumer Research , 36 (5), 838 – 56.
8 Blakely Lindsay. (2018), "Company of the Year: Patagonia's Unapologetically Political Strategy and the Massive Business It Has Built," Inc. (accessed March 30, 2022), https://www.inc.com/lindsay-blakely/patagonia-2018-company-of-the-year-nominee.html.
9 Bourdieu Pierre. (1986), " Forms of Capital ," in Handbook of Theory and Research for the Sociology of Education , Richardson John , ed., translated by Richard Nice. New York : Greenwood , 241 – 58.
Bradshaw Alan , Holbrook Morris B.. (2007), " Remembering Chet: Theorizing the Mythology of the Self-Destructive Bohemian Artist as Self-Producer and Self-Consumer in the Market for Romanticism ," Marketing Theory , 7 (2), 115 – 36.
Brown Stephen , Kozinets Robert , Sherry John F. Jr.. (2003), " Teaching Old Brands New Tricks: Retro Branding and the Revival of Brand Meaning ," Journal of Marketing , 67 (3), 19 – 33.
Buzuvis Erin. (2021), " Law, Policy, and the Participation of Transgender Athletes in the United States ," Sport Management Review , 24 (3), 439 – 51.
Carfagna Lindsey B. , Dubois Emilie A. , Fitzmaurice Connor , Ouimette Monique Y. , Schor Juliet B. , Willis Margaret , et al. (2014), " An Emerging Eco-Habitus: The Reconfiguration of High Cultural Capital Practices Among Ethical Consumers ," Journal of Consumer Culture , 14 (2), 158 – 78.
Cayco Anna. (2019), "Is Sustainability for the Elite?" OffCrowd (June 6), https://www.offcrowd.com/stories/economy/is-sustainability-for-the-elite/.
Champagne Christine. (2013), "How to Make a Cannes Contender: Chipotle's 'Back to the Start'," Fast Company (September 13), https://www.fastcompany.com/1680942/how-to-make-a-cannes-contender-chipotles-back-to-the-start.
Chaudhury Sarita Ray , Albinsson Pia A.. (2015), " Citizen-Consumer Oriented Practices in Naturalistic Foodways: The Case of the Slow Food Movement ," Journal of Macromarketing , 35 (1), 36 – 52.
Chronis Athinodoros. (2008), " Co-Constructing the Narrative Experience: Staging and Consuming the American Civil War at Gettysburg ," Journal of Marketing Management , 24 (1/2), 5 – 27.
Chrzan Janet. (2004), " Slow Food: What, Why and to Where? " Food, Culture and Society , 7 (2), 117 – 32.
Coskuner-Balli Gokçen , Thompson Craig J.. (2013), " The Status Costs of Subordinate Cultural Capital: At-Home Fathers' Collective Pursuit of Cultural Legitimacy Through Capitalizing Consumption Practices ," Journal of Consumer Research , 40 (1), 19 – 41.
Debenedetti Alain , Oppewal Harmen , Arsel Zeynep. (2014), " Place Attachment in Commercial Settings: A Gift Economy Perspective ," Journal of Consumer Research , 40 (5), 904 – 23.
Deighton John , Grayson Kent. (1995), " Marketing and Seduction: Building Exchange Relationships by Managing Social Consensus ," Journal of Consumer Research , 21 (4), 660 – 76.
Dickinson J. Barry. (2011), " The Role of Authenticity in Relationship Marketing ," Journal of Management and Marketing Research , 8 (September), 1 – 12.
Dion Delphine , Borraz Stéphane. (2017), " Managing Status: How Luxury Brands Shape Class Subjectivities in the Service Encounter ," Journal of Marketing , 81 (5), 67 – 85.
Elliott Rebecca. (2013), " The Taste for Green: The Possibilities and Dynamics of Status Differentiation Through 'Green' Consumption ," Poetics , 41 (3), 294 – 322.
Entine Jon. (1994), " Shattered Image ," Business Ethics: The Magazine of Corporate Responsibility , 8 (5), 23 – 28.
Erickson Gary. (2019), "An Open Invitation to Kind Bar from Clif Bar," The New York Times (March 6), reprinted at https://www.clifbar.com/article/an-open-invitation-to-kind-bar-from-clif-bar.
Ertimur Burçak , Coskuner-Balli Gokçen. (2015), " Navigating the Institutional Logics of Markets: Implications for Strategic Brand Management ," Journal of Marketing , 79 (2), 40 – 61.
Ferdman Roberto. (2014), "School Kids Are Blaming Michelle Obama for Their 'Gross' School Lunches," The Washington Post (November 24), https://www.washingtonpost.com/news/wonk/wp/2014/11/24/students-are-blaming-michelle-obama-for-their-gross-school-lunches/.
Floch Jean-Marie. (1988), " The Contribution of Structural Semiotics to the Design of a Hypermarket ," International Journal of Research in Marketing , 4 (3), 233 – 52.
Fournier Susan. (1998), " Consumers and Their Brands: Developing Relationship Theory in Consumer Research ," Journal of Consumer Research , 24 (4), 343 – 73.
Fournier Susan , Eckhardt Giana M.. (2019), " Putting the Person Back in Person-Brands: Understanding and Managing the Two-Bodied Brand ," Journal of Marketing Research , 56 (4), 602 – 19.
Friestad Marian , Wright Peter. (1994), " The Persuasion Knowledge Model: How People Cope with Persuasion Attempts ," Journal of Consumer Research , 21 (1), 1 – 31.
Fyke Jeremy P. , Buzzanell Patrice M.. (2013), " The Ethics of Conscious Capitalism: Wicked Problems in Leading Change and Changing Leaders ," Human Relations , 66 (12), 1619 – 43.
Gage Beverly. (2017), "How 'Elites' Became One of the Nastiest Epithets in American Politics, New York Times Magazine (January 3), https://www.nytimes.com/2017/01/03/magazine/how-elites-became-one-of-the-nastiest-epithets-in-american-politics.html.
Gelles David. (2018), "Patagonia v. Trump," The New York Times (May 5), https://www.nytimes.com/2018/05/05/business/patagonia-trump-bears-ears.html.
Giesler Markus. (2008), " Conflict and Compromise: Drama in Marketplace Evolution ," Journal of Consumer Research , 34 (6), 739 – 54.
Giesler Markus. (2012), " How Doppelgänger Brand Images Influence the Market Creation Process: Longitudinal Insights from the Rise of Botox Cosmetic ," Journal of Marketing , 76 (6), 55 – 68.
Gillison Samantha. (2018), "'Clean Eating' Has Become Such a Sham That Fast Food Chains Are Pushing It," NBC (February 6), https://www.nbcnews.com/think/opinion/clean-eating-has-become-such-sham-fast-food-chains-are-ncna845081.
Grayson Kent , Martinec Radan. (2004), " Consumer Perceptions of Iconicity and Indexicality and Their Influence on Assessments of Authentic Market Offerings ," Journal of Consumer Research , 31 (2), 296 – 312.
Greimas Algirdas Julien. (1987), On Meaning: Selected Writings in Semiotic Theory. Minneapolis : University of Minnesota Press.
Griffin Abbie , Hauser John. (1993), " The Voice of the Customer ," Marketing Science , 12 (Winter), 1 – 27.
Griskevicius Vlad , Tybur Joshua M. , Van Den Bergh Bram. (2010), " Going Green to Be Seen: Status, Reputation, and Conspicuous Consumption ," Journal of Personality and Social Psychology , 98 (3), 392 – 404.
Heartney Eleanor. (2001), Postmodernism (Movements in Modern Art). New York : Cambridge.
Holt Douglas B.. (1998), " Does Cultural Capital Structure American Consumption? " Journal of Consumer Research , 25 (1), 1 – 26.
Holt Douglas B.. (2002), " Why Do Brands Cause Trouble? A Dialectical Theory of Consumer Culture and Branding ," Journal of Consumer Research , 29 (1), 70 – 90.
Holt Douglas B.. (2004), How Brands Become Icons: The Principles of Cultural Branding. Cambridge , MA : Harvard Business Press.
Holt Douglas B.. (2014), " Why the Sustainable Economy Movement Hasn't Scaled: Toward a Strategy That Empowers Mainstreet ," in Sustainable Lifestyles and the Quest for Plenitude: Case Studies in the New Economy , Schor Juliet B. , Thompson Craig J. , eds. New Haven , CT : Yale University Press , 202 – 32.
Holt Douglas B.. (2016), " Branding in the Age of Social Media ," Harvard Business Review , 94 (3), 41 – 50.
Holt Douglas B. , Cameron Douglas. (2010), Cultural Strategy: Using Innovative Ideologies to Build Breakthrough Brands. New York : Oxford University Press.
Holt Douglas B. , Thompson Craig J.. (2004), " Man-of-Action Heroes: The Pursuit of Heroic Masculinity in Everyday Consumption ," Journal of Consumer Research , 31 (2), 425 – 40.
Humphreys Ashlee. (2010), " Semiotic Structure and the Legitimation of Consumption Practices: The Case of Casino Gambling ," Journal of Consumer Research , 37 (3), 490 – 510.
Johnston Josée , Baumann Shyon. (2015), Foodies: Democracy and Distinction in the Gourmet Foodscape, 2nd ed. New York : Routledge.
Kliman Todd , Laudan Rachel. (2015), "How Michael Pollan, Alice Waters, and Slow Food Theorists Got It All Wrong: A Conversation with Food Historian (and Contrarian) Rachel Laudan," The Washingtonian (May 29), https://www.washingtonian.com/2015/05/29/rachel-lauden-how-michael-pollan-alice-waters-got-everything-wrong/.
Kozinets Robert. (2008), " Technology/Ideology: How Ideological Fields Influence Consumers' Technology Narratives ," Journal of Consumer Research , 34 (6), 865 – 81.
Kupfer Ann-Kristin , Pähler vor der Holte Nora , Kübler Raoul V. , Hennig-Thurau Thorsten. (2018), " The Role of the Partner Brand's Social Media Power in Brand Alliances ," Journal of Marketing , 82 (3), 25 – 44.
Laudan Rachel. (2010), " A Plea for Culinary Modernism: Why We Should Love New, Fast, Processed Food ," in The Gastronomica Reader , Goldstein Darra , ed. Berkeley : University of California Press , 280 – 92.
Leigh Thomas W. , Peters Cara , Shelton Jeremy. (2006), " The Consumer Quest for Authenticity: The Multiplicity of Meanings Within the MG Subculture of Consumption ," Journal of the Academy of Marketing Science , 31 (4), 1 – 13.
Levy Annie. (2019), "Does the Food Movement's Elitism Hinder Our Progress?" Comestible , Issue 4, https://www.comestiblejournal.com/blog/does-the-food-movements-elitism-hinder-our-progress.
Luedicke Marius K. , Thompson Craig J. , Giesler Markus. (2009), " Consumer Identity Work as Moral Protagonism: How Myth and Ideology Animate a Brand-Mediated Moral Conflict ," Journal of Consumer Research , 36 (6), 1016 – 32.
Luffarelli Jonathan , Mukesh Mudra , Mahmood Ammara. (2019), " Let the Logo Do the Talking: The Influence of Logo Descriptiveness on Brand Equity ," Journal of Marketing Research , 56 (5), 862 – 78.
Mackey John , Sisodia Rajendra. (2013), "Conscious Capitalism Is Not an Oxymoron," Harvard Business Review (January 14), https://hbr.org/2013/01/cultivating-a-higher-conscious.
Mackey John , Sisodia Rajendra. (2014), Conscious Capitalism: Liberating the Heroic Spirit of Business. Cambridge , MA : Harvard University Press.
Miller Ryan W.. (2018), "Patagonia Plans to Donate $10 Mllion Saved from Trump Tax Cuts to Environmental Groups," USA Today (November 28), https://www.usatoday.com/story/money/2018/11/28/patagonia-money-saved-trump-tax-cut-environmental-cause/2143733002/.
Muñiz Albert , O'Guinn Thomas C.. (2001), " Brand Communities ," Journal of Consumer Research , 27 (4), 412 – 32.
Napoli Julie , Dickinson-Delaporte Sonia , Beverland Michael B.. (2016), " The Brand Authenticity Continuum: Strategic Approaches for Building Value ," Journal of Marketing Management , 32 (13/14), 1201 – 29.
Ngai Sianne. (2020), Theory of the Gimmick: Aesthetic Judgment and Capitalist Form. Cambridge , MA : Harvard University Press.
Nunes Joseph C. , Ordanini Andrea , Giambastiani Gaia. (2021), " The Concept of Authenticity: What It Means to Consumers ," Journal of Marketing , 85 (4), 1 – 20.
Østergaard Per , Hermansen Judy , Fitchett James. (2015), " Structures of Brand and Anti-Brand Meaning: A Semiotic Square Analysis of Reflexive Consumption ," Journal of Brand Management , 22 (1), 60 – 77.
Oswald Laura. (2015), Creating Value: The Theory and Practice of Marketing Semiotics Research , Oxford, UK : Oxford University Press.
O'Toole James , Vogel David. (2011), " Two and a Half Cheers for Conscious Capitalism ," California Management Review , 53 (3), 60 – 76.
Parker Kathleen. (2014), "First Lady's Lunch Plan Is a Failed Recipe," Chicago Tribune (June 4), https://www.chicagotribune.com/opinion/ct-xpm-2014-06-04-ct-michelle-obama-nutrition-children-lunch-gop-ope-20140604-story.html.
Patagonia (@patagonia) (2021), " Consider how and what you give this holiday season. Repair and pass along cherished gear, share experiences and knowledge, or give to good causes. If you buy new, look for gear that'll last, support Fair Trade factories and choose organic or recycled materials," Twitter (November 18), https://twitter.com/patagonia/status/1461521095324880899.
Petrini Carlo. (2001), Slow Food: The Case for Taste. New York : Columbia University Press.
Petrini Carlo. (2007), Slow Food Nation: Why Our Food Should be Good, Clean, and Fair. New York : Rizzoli Ex Libris.
Pollan Michael. (2006), The Omnivore's Dilemma: A Natural History of Four Meals. New York : Penguin.
Purkayastha Debapratim , Fernando Rajiv. (2017), " The Body Shop: Social Responsibility or Sustained Greenwashing? " in Case Studies in Sustainability Management and Strategy , Hamschmidt Jost , ed. New York : Routledge , 226 – 51.
Rao Priya. (2019), "The Body Shop Is Tapping Its Activist Roots with Updated Product and Store Initiatives," Glossy (February 26), https://www.glossy.co/beauty/the-body-shop-is-tapping-its-activist-roots-with-updated-product-and-store-initiatives/.
Rinallo Diego. (2007), " Metro/Fashion/Tribes of Men: Negotiating the Boundaries of Men's Legitimate Consumption ," in Consumer Tribes , Cova Bernard , Kozinets Robert V. , Shankar Avi , eds. Burlington , MA : Butterworth-Heinemann , 76 – 92.
Rokka Joonas , Canniford Robin. (2016), " Heterotopian Selfies: How Social Media Destabilizes Brand Assemblages ," European Journal of Marketing , 50 (9/10), 1789 –1 813.
Seregina Anastasia , Weijo Henri A.. (2017), " Play at Any Cost: How Cosplayers Produce and Sustain Their Ludic Communal Consumption Experiences ," Journal of Consumer Research , 44 (1), 139 – 59.
Södergren Jonatan. (2021), " Brand Authenticity: 25 Years of Research ," International Journal of Consumer Studies , 45 (4), 645 – 63.
Spiggle Susan , Nguyen Hang T. , Caravella Mary. (2012), " More Than Fit: Brand Extension Authenticity ," Journal of Marketing Research , 49 (6), 967 – 83.
Stokes Ashli Quesinberry. (2013), " You Are What You Eat: Slow Food USA's Constitutive Public Relations ," Journal of Public Relations Research , 25 (1), 68 – 90.
Thompson Craig J.. (1997), " Interpreting Consumers: A Hermeneutical Framework for Deriving Marketing Insights from the Texts of Consumers' Consumption Stories ," Journal of Marketing Research , 34 (4), 438 – 55.
Thompson Craig J. , Locander William B. , Pollio Howard R.. (1989), " Putting Consumer Experience Back into Consumer Research: The Philosophy and Method of Existential-Phenomenology ," Journal of Consumer Research , 16 (2), 133 – 46.
Thompson Craig J. , Rindfleisch Aric , Arsel Zeynep. (2006), " Emotional Branding and the Strategic Value of the Doppelgänger Brand Image ," Journal of Marketing , 70 (1), 50 – 64.
Thomson Matthew. (2006), " Human Brands: Investigating Antecedents to Consumers' Strong Attachments to Celebrities ," Journal of Marketing , 70 (3), 104 – 19.
Turkle Sherry. (2012), Alone Together: Why We Expect More from Technology and Less from Each Other. New York : Basic Books.
Van Bommel Koen , Spicer André. (2011), " Hail the Snail: Hegemonic Struggles in the Slow Food Movement ," Organization Studies , 32 (12), 1717 – 44.
Wallace Elaine , Buil Isabel , De Chernatony Leslie. (2020), " 'Consuming Good' on Social Media: What Can Conspicuous Virtue Signalling on Facebook Tell Us About Prosocial and Unethical Intentions? " Journal of Business Ethics , 162 (3), 577 – 92.
Weinberger Michelle , Zavisca Jane , Silva Jennifer. (2017), " Consuming for an Imagined Future: Middle-Class Consumer Lifestyle and Exploratory Experiences in the Transition to Adulthood ," Journal of Consumer Research , 44 (2), 332 – 60.
Žižek Slavoj. (2011), Living in the End Times. New York : Verso.
~~~~~~~~
By Craig J. Thompson and Ankita Kumar
Reported by Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 6- Artificial Intelligence Coaches for Sales Agents: Caveats and Solutions. By: Xueming Luo; Shaojun Qin, Marco; Zheng Fang; Zhe Qu. Journal of Marketing. Mar2021, Vol. 85 Issue 2, p14-32. 19p. 2 Diagrams, 5 Charts, 5 Graphs. DOI: 10.1177/0022242920956676.
- Database:
- Business Source Complete
Record: 7- Artificial Intelligence in Utilitarian vs. Hedonic Contexts: The "Word-of-Machine" Effect. By: Longoni, Chiara; Cian, Luca. Journal of Marketing. Jan2022, Vol. 86 Issue 1, p91-108. 18p. 2 Diagrams, 4 Graphs. DOI: 10.1177/0022242920957347.
- Database:
- Business Source Complete
Artificial Intelligence in Utilitarian vs. Hedonic Contexts: The "Word-of-Machine" Effect
Rapid development and adoption of AI, machine learning, and natural language processing applications challenge managers and policy makers to harness these transformative technologies. In this context, the authors provide evidence of a novel "word-of-machine" effect, the phenomenon by which utilitarian/hedonic attribute trade-offs determine preference for, or resistance to, AI-based recommendations compared with traditional word of mouth, or human-based recommendations. The word-of-machine effect stems from a lay belief that AI recommenders are more competent than human recommenders in the utilitarian realm and less competent than human recommenders in the hedonic realm. As a consequence, importance or salience of utilitarian attributes determine preference for AI recommenders over human ones, and importance or salience of hedonic attributes determine resistance to AI recommenders over human ones (Studies 1–4). The word-of machine effect is robust to attribute complexity, number of options considered, and transaction costs. The word-of-machine effect reverses for utilitarian goals if a recommendation needs matching to a person's unique preferences (Study 5) and is eliminated in the case of human–AI hybrid decision making (i.e., augmented rather than artificial intelligence; Study 6). An intervention based on the consider-the-opposite protocol attenuates the word-of-machine effect (Studies 7a–b).
Keywords: algorithms; artificial intelligence; augmented intelligence; hedonic and utilitarian consumption; recommendations; technology
Recommendations driven by artificial intelligence (AI) are pervasive in today's marketplace. Ten years ago, Amazon introduced its innovative item-based collaborative filtering algorithm, which generates recommendations by scanning through a person's past purchased or rated items and pairing them to similar items. Since then, more and more companies are leveraging advances in AI, machine learning, and natural language processing capabilities to provide relevant and in-the-moment recommendations. For example, Netflix and Spotify use AI and deep learning to monitor a user's choices and provide recommendations of movies or music. Beauty brands such as Proven, Curology, and Function of Beauty use AI to make recommendations about skincare, haircare, and makeup. Real estate services such as OJO Labs, REX Real Estate, and Roof.ai have replaced human real estate agents with chatbots powered by AI. AI-driven recommendations are also pervading the public sector. For example, the New York City Department of Social Services uses AI to give citizens recommendations about disability benefits, food assistance, and health insurance.
In response to the proliferation of AI-enabled recommendations and building on long-standing research on actuarial judgments ([12]; [17]; [30]), recent marketing research has focused on whether consumers will be receptive to algorithmic advice in various domains ([ 9]; [14]; [24]; [26]; [27]). However, no prior empirical investigation has systematically explored if hedonic/utilitarian trade-offs in decision making determine preference for, or resistance to, AI-based (vs. human-based) recommendations.
We focus our investigation on hedonic/utilitarian attribute trade-offs because of their influence on both consumer choice and attitudes ([ 6]; [11]). Specifically, we examine when and why hedonic/utilitarian attribute trade-offs in decision making influence whether people prefer or resist AI recommenders. This question is of pivotal importance for managers operating in both the private and public sectors who are looking to harness the potential of AI-driven recommendations.
Across nine studies and using a broad array of both attitudinal and behavioral measures, we provide evidence of a "word-of-machine" effect. We define "word of machine" as the phenomenon by which hedonic/utilitarian attribute trade-offs determine preference for, or resistance to, AI-based recommendations compared with traditional word of mouth, or human-based recommendations. We suggest that the word-of-machine effect stems from a lay belief about differential competence perceptions regarding AI and human recommenders. Specifically, we show that people believe AI recommenders are more competent than human recommenders to assess utilitarian attribute value and generate utilitarian-focused recommendations. By contrast, people believe that AI recommenders are less competent than human recommenders to assess hedonic attribute value and generate hedonic-focused recommendations. As a consequence, and as compared with human recommenders, individuals are more (less) likely to choose AI recommenders when utilitarian (hedonic) attributes are important or salient, such as when a utilitarian (hedonic) goal is activated.
Our research is both theoretically novel and substantively impactful. A first set of theoretical contributions relates to research on the psychology of automation and on human–technology interaction ([12]; [17]; [30]). The pervasiveness of AI-driven recommendations has led to a burgeoning body of research examining whether consumers are receptive to the advice of algorithms, statistical models, and artificial intelligence ([14]; [24]; [27]). With respect to this literature, we make three novel contributions. First, we extend it by addressing the previously unexplored question of when and why hedonic/utilitarian trade-offs in decision making influence preference for or resistance to AI recommenders. Second, we show under what circumstances AI-driven recommendations are preferred to, and therefore more effective, than human ones: when utilitarian attributes are relatively more important or salient than hedonic ones. These results are especially noteworthy, as most research in this area has documented a robust and generalized resistance to algorithmic advice (for exceptions, see [ 9]; [15]; [26]). Third, we explore under what circumstances consumers will be amenable to AI recommenders in the context of human–AI partnerships: when AI supports rather than replaces a human. These results are also novel as researchers have just begun devising AI systems capable of deciding when to defer (vs. not defer) to a human ([18]), and empirical investigations are yet to examine if consumers will embrace such hybrid human–AI decision making.
Our research makes a second theoretical contribution to the literature on hedonic and utilitarian consumption ([ 1]; [22]; [31]; [44]). Prior research in this area has examined how the evaluation of hedonic and utilitarian products depends on characteristics of the task, locus of choice, and justifiability of choice (e.g., [ 5]; [ 8]; [34]). However, research in this area has not addressed the question of whether shifts in hedonic/utilitarian trade-offs in decision making determine preference for the source of a recommendation (e.g., an AI vs. a human recommender). Recent developments of AI have brought this question to the fore, making it of critical importance for companies seeking to leverage the potential of AI-driven recommendations.
From a managerial perspective, our results are useful for companies in both the private and public sectors that are looking to leverage AI recommenders to better reach their customers. As we investigate when consumers prefer AI over human recommenders, our findings are useful for companies debating if and how to effectively leverage AI-based recommendation systems. Our findings have implications for a host of marketing decisions. For instance, our results indicate that a shift away from hedonic attributes and toward utilitarian attributes leads to consumers preferring AI recommenders. Accordingly, AI recommenders may be more aligned with functional positioning strategies than experiential ones. In addition, emphasizing utilitarian benefits may be relatively more impactful with an AI-based system than emphasizing hedonic benefits. Taken together, our research and findings provide actionable insights for managers looking for ways to leverage AI to orchestrate consumer journeys so as to successfully move customers through the funnel, increase the likelihood of successful transactions, and, overall, optimize the customer experience at each phase of the journey.
Although consumption involves both hedonic and utilitarian considerations, consumers tend to view products as either predominantly hedonic or utilitarian (for a review, see [23]). Hedonic consumption is primarily affectively driven, based on sensory or experiential pleasure, reflects affective benefits, and is assessed on the basis of the degree to which a product is rewarding in itself ([ 8]; [11]; [21]). Utilitarian consumption is instead more cognitively driven, based on functional and instrumental goals, reflects functional benefits, and is assessed on the basis of the degree to which a product is a means to an end ([ 8]; [11]; [21]).
Prior research on hedonic/utilitarian consumption has focused on the effect of characteristics of the task on product judgments. For instance, choice tasks tend to favor utilitarian options, whereas rating tasks tend to favor hedonic options ([ 5]; [34]), and forfeiture increases the relative salience of hedonic attributes compared to acquisition ([13]). Justifiability leads people to assign greater weight to utilitarian (vs. hedonic) options ([34]), and hedonic (vs. utilitarian) choices are associated with greater perceived personal causality ([ 8]).
Although spanning over a decade, research on hedonic/utilitarian consumption has not yet addressed the question of whether hedonic and utilitarian trade-offs influence preference for the source of a recommendation (AI vs. human). This question has come to the fore given its importance for managers looking to leverage the potential of algorithmic recommendations. We discuss prior research on algorithmic recommendations in the next section.
Ever since seminal work on statistical and actuarial predictive models was published ([12]; [17]; [30]), a large body of research has documented how statistical/actuarial models outperform clinical/human judgments in predicting a host of events, such as students' and employees' performance ([12]) and market demand ([37]). Despite the superior accuracy of algorithmic models, people tend to eschew them. With only a few exceptions ([ 9]; [15]; [26]), most of the extant literature has shown that people resist the advice of a statistical algorithm. For instance, recent research in the medical domain has shown that consumers may be more reluctant to utilize medical care delivered by AI providers than by comparable human providers ([27]; [28]). Corporate settings show similar patterns, with recruiters ([19]) and auditors ([ 7]) trusting their judgment and predictions more than algorithms.
There are numerous reasons why people resist algorithmic recommendations. People (erroneously) believe that algorithms are unable to learn and improve ([12], [19]) and therefore lose confidence in algorithms when they see them err ([14]). People also believe that algorithms assume the world to be orderly, rigid, and stable and therefore cannot take into consideration uncertainty ([17]) and a person's uniqueness ([27]). Resistance to algorithmic advice may also be borne out of generalized concerns, such as people's fear of being reduced to "mere numbers" ([12]) and mistrust of algorithms' lack of empathy ([17]).
We extend this literature and show circumstances under which people prefer (and not just resist) algorithmic recommendations. Specifically, we examine how and why hedonic/utilitarian trade-offs determine preference for, or resistance to, AI recommenders, as articulated in the next section.
We hypothesize a word-of-machine effect, whereby hedonic and utilitarian trade-offs determine preference for or resistance to AI recommenders compared to human ones. We suggest that the word-of-machine effect stems from consumers' differing competence perceptions of AI and human recommenders in assessing attribute value and generating recommendations. Specifically, we suggest that people believe AI recommenders to be more (less) competent to assess utilitarian (hedonic) attribute value and generate utilitarian-focused (hedonic-focused) recommendations than human recommenders.
These predictions rest on the assumption that people believe hedonic and utilitarian attribute value assessment to require different evaluation competences. Hedonic value assessment should map onto criteria on the basis of experiential, emotional, and sensory evaluative dimensions. By contrast, utilitarian value assessment should map onto criteria on the basis of factual, rational, and logical evaluative dimensions. This assumption is rooted in the very definition of hedonic and utilitarian value. Hedonic value is conceptualized as reflecting experiential affect associated with a product, sensory enjoyment, and emotions ([ 4]; [20]). Indeed, hedonic consumption tends to be affectively rich and emotionally driven ([ 8]). By contrast, utilitarian value is conceptualized as reflecting instrumentality, functionality, nonsensory attributes, and rationality ([ 4]; [20]). Overall, utilitarian consumption is cognitively driven ([ 8]).
How do different types of recommenders (AI vs. human) then fare with respect to assessing hedonic and utilitarian attribute value? We suggest that people believe AI recommenders are more competent to assess utilitarian attribute value than human recommenders and less competent to assess hedonic attribute value than human recommenders. We attribute this lay belief to differing associations people have about how AI (vs. human) recommenders process and evaluate information. Lay beliefs are developed either directly through personal experience ([36]) or indirectly from the environment ([32]). Throughout childhood we learn firsthand that, as humans, we are able to perceive and connect with the outside world through our affective experiences. By contrast, we learn that AI, computers, and robots are rational and logical, and lack the ability to have affective, experiential interactions with the world. These associations are reflected in idioms such as "thinking like a robot," which refers to thinking logically without taking into consideration more "human" aspects of a situation such as sensations and emotions. Thus, whereas AI and computers are associated with rationality and logic, humans are associated with emotions and experiential abilities. These associations are also echoed in books, songs, and movies. For example, in the Star Trek universe, the artificially intelligent form of life named Data has superior intellective abilities but is unable to experience emotions. Popular movies like Her, Ex Machina, and Terminator further reinforce these associations.
Accordingly, we suggest that people believe AI recommenders are more competent than human recommenders when assessing information because they use criteria that rely relatively more on facts, rationality, logic, and, overall, cognitive evaluative dimensions. By contrast, we propose that people believe human recommenders are more competent than AI recommenders when assessing information because they use criteria that rely relatively more on sensory experience, emotions, intuition, and, overall, affective evaluative dimensions.
Because people perceive AI and humans to have different competency levels when assessing information, and because assessment of utilitarian and hedonic attribute value underscore different evaluative foci, it follows that people perceive AI and humans to have different competency levels with respect to assessing utilitarian and hedonic attributes. This lay belief about competence perceptions forms the basis for the proposed word-of-machine effect. In summary, we predict that if utilitarian (hedonic) attributes are more important or salient, such as when a utilitarian (hedonic) goal is activated, people will be more (less) likely to choose AI recommenders than human recommenders.
A final note warrants mention. As competence perceptions driving the word-of-machine effect are based on a lay belief, they are embedded in the cultural context. That is, humans are not necessarily less competent than AI at assessing and evaluating utilitarian attributes. Vice versa, AI is not necessarily less competent than humans at assessing and evaluating hedonic attributes. Indeed, AI selects flower arrangements for 1-800-Flowers and creates new flavors for food companies such as McCormick, Starbucks, and Coca-Cola ([41]).
Studies 1a–b focus on product choice in field settings and show the main word-of-machine effect: that AI (human) recommenders lead to greater choice likelihood when a utilitarian (hedonic) goal is activated. Study 2 shows different perceptions that result from the two recommendation sources: AI (human) recommenders lead to higher evaluation of utilitarian (hedonic) attributes upon consumption. Study 3 shows that when a utilitarian (hedonic) attribute is considered important, consumers prefer AI (human) recommenders. Study 4 uses an analysis of mediation to corroborate the role of competence perceptions in explaining the word-of-machine effect while ruling out attribute complexity as alternative explanation. Studies 5–7 explore the scope of the word-of-machine effect by identifying boundary conditions. Study 5 shows that the effect is reversed for utilitarian goals when the recommendation needs to match to a person's unique preferences, a type of task people view AI as unfit to do. Study 6 shows that the effect is eliminated when AI is framed as "augmented" intelligence rather than artificial intelligence, that is, when AI enhances and supports a person rather than replacing them. Finally, Studies 7a–b test an intervention using the consider-the-opposite protocol to moderate the word-of-machine effect.
Studies 1a–b focus on the word-of-machine effect on actual product choice in field settings as a function of an activated utilitarian or hedonic goal. We first activated either a utilitarian or a hedonic goal and then, in an incentive-compatible setting, measured choice as a function of recommender.
Two hundred passersby in a city in northeast United States participated in Study 1a on a voluntary basis. We handed willing passersby a leaflet explaining that we were conducting a blind test for products in the haircare industry and, specifically, for hair masks—a leave-in treatment for hair and scalp. Passersby read that for the purpose of the market test, we wanted them to select one of two hair mask samples solely on the basis of the instructions in the leaflet. These instructions activated, in a two-cell between-subjects design, either a hedonic or a utilitarian goal:
[Hedonic] For the purpose of this blind test, it is very important that you set aside all thoughts you might already have about hair masks. Instead, we would like you to focus only on the following. Imagine that you have a "hedonic" goal. We would like you to imagine that the only things that you care about in a hair mask are hedonic characteristics, like how indulgent it is to use, its scent, and the spa-like vibe it gives you. When you make the next choice, imagine that there are no other things that are important for you in a hair mask.
[Utilitarian] For the purpose of this blind test, it is very important that you set aside all thoughts you might already have about hair masks. Instead, we would like you to focus only on the following. Imagine that you have a "utilitarian" goal. We would like you to imagine that the only things that you care about in a hair mask are utilitarian characteristics, like how practical it is to use, its objective performance, and the chemical composition. When you make the next choice, imagine that there are no other things that are important for you in a hair mask.
The leaflet further explained that there were two hair mask options from which they could choose. One option had been recommended by a person, and the other option had been recommended by an algorithm. The leaflet specified that the person and the algorithm had the same haircare expertise and that the pots of hair masks, available for pickup on a desk, all contained the same amount of fluid ounces. The pots were identical except for a marking of "P" if selected by a person or "A" if selected by an algorithm (stimuli in Web Appendix A). The key dependent variable was whether passersby chose the product selected by the person or by the algorithm.
To assess product choice, we compared the proportion of people who chose the product recommended by the algorithm with the proportion of people who chose the product recommended by the person depending on the activated goal (utilitarian vs. hedonic). The two proportions differed significantly (χ2( 1, N = 200) = 12.60, p =.001). As predicted, when a utilitarian goal was activated, more people chose the product recommended by the algorithm (67%) than by the person (33%; z = 4.81, p <.001). When a hedonic goal was activated, more people chose the product recommended by the person (58%) than by the algorithm (42%; z = 2.26, p =.024).
Study 1b was a field study conducted over four consecutive days in Cortina, a resort town in northeast Italy. We selected this town because in 2026 it will host the Olympic games and is likely to experience a boom in its real estate market, which is the domain of the study. We secured the use of a centrally located bar and set up the study as follows. We placed an ad (translated to Italian) promoting a local real estate agency at the bar entrance. The ad headline reminded people of the opportunity to make fruitful real estate investments due to the upcoming Olympic games. In a two-cell, between-subjects design, we alternated the text in the ad to focus people on a hedonic or utilitarian goal:
[Hedonic] With the Olympic games coming up, it is really important that you look for a real estate investment that is fun, enjoyable, and speaks to your emotions. You want a place that pleases your senses considering all the changes that will affect [name of town] in the next few years.
[Utilitarian] With the Olympic games coming up, it is really important that you look for a real estate investment that is functional, useful, and speaks to your rationality. You want a place that is practical considering all the changes that will affect [name of town] in the next few years.
At the bottom of the ad there were two envelopes described as containing a curated selection of available properties in Cortina that could fit with the opportunity in the ad (i.e., one of the activated goals). One property selection had been (ostensibly) curated by a person (the respective envelope read: "one of [name of agency]'s agents has selected these properties") and the other by an algorithm (the respective envelope read: "[name of agency]'s proprietary algorithm has selected these properties"). The ad invited people to pick up only one of the two envelopes given the limited quantity of promotional materials (stimuli in Web Appendix B). The key dependent variable was whether people chose the selection made by the agent or by the algorithm. A waiter ensured that participants took only one of the two envelopes, and we excluded two participants who picked up two (final N = 229).
We compared the proportion of people who chose the selection made by the algorithm with the proportion of people who chose the selection made by the agent depending on the activated goal (utilitarian vs. hedonic). The two proportions differed significantly (χ2( 1, N = 229) = 29.33, p <.001). When the goal was utilitarian, more people chose the selection made by the algorithm (59.8%) than by the agent (40.2%; z = 3.07, p =.002), whereas when the goal was hedonic, more people chose the selection made by the agent (75.7%) than by the algorithm (24.3%; z = 7.52, p <.001).
Together, Studies 1a–b show that when a utilitarian goal is activated, people are more likely to choose an AI recommender than a human recommender. When a hedonic goal is activated, people are less likely to choose an AI recommender than a human recommender.
Study 2 examines the word-of-machine effect upon consumption. As conceptual information such as expectations affects food consumption experiences (e.g., [ 2]; [42]), we predicted that the type of recommender would affect perceptions of hedonic and utilitarian attributes upon actual consumption of a product (a chocolate cake).
One hundred forty-four participants from a paid subject pool (open to students and nonstudents) at the University of Virginia completed this study (Mage = 27.5 years, SD = 9.5; 60.4% female). We told participants that we were testing chocolate cake recipes on behalf of a local bakery (stimuli in Web Appendix C). We told participants that the bakery had two options for chocolate cake recipes: one created using the ingredient selection of an AI chocolatier and one created using the ingredient selection of a human chocolatier. We specified that both the human and AI chocolatier had access to the same recipe database. We invited participants to look at the two chocolate cakes on top of a podium in a pop-up bakery/classroom desk. The two types of cake looked (and were) identical. We told participants that the two chocolate cakes, although based on different recipes, looked the same because the bakery did not want them to be influenced by the shape or the color. In a two-cell between-subjects design, we asked participants to consume either the chocolate cake whose recipe was selected by the human chocolatier or the one selected by the AI chocolatier. After consuming the cake, we measured hedonic/utilitarian attribute perceptions by asking participants to rate the cake on two hedonic items (indulgent taste and aromas; pleasantness to the senses [vision, touch, smell, etc.]) and two utilitarian items (beneficial chemical properties [antioxidants]; healthiness [micro/macro nutrients, etc.]) on seven-point scales anchored at 1 = "very low" and 7 = "very high." The order of hedonic and utilitarian items was randomized.
A one-way analysis of variance (ANOVA) on the average of the two hedonic items (r =.87, p <.001) revealed that, upon consumption, participants rated the chocolate cake as having lower hedonic value when based on the recommendation of an AI chocolatier than a human one (MAI = 4.57, SD = 1.38; MH = 6.17, SD = 1.03; F( 1, 142) = 61.33, p <.001).
A one-way ANOVA on the index of the two utilitarian items (r =.84, p <.001) revealed that, upon consumption, participants rated the chocolate cake as having higher utilitarian value when based on the recommendation of an AI chocolatier than a human one (MAI = 5.48, SD = 1.21; MH = 5.02, SD = 1.35; F( 1, 142) = 61.33, p =.034).
Thus, Study 2 shows that the word-of-machine effect extends to actual consumption and that the type of recommender influences people's perceptions of hedonic/utilitarian trade-offs. AI recommenders led participants to perceive greater utilitarian attribute value and lower hedonic attribute value compared to human recommenders.
Study 3 further tests the word-of-machine effect. Instead of activating hedonic/utilitarian goals as in Studies 1a–b, we measured the importance given to hedonic/utilitarian attributes with respect to a specific product category (winter coats). Then, we assessed relative preference for a human or an AI recommender. We expected people to prefer AI to human recommenders when utilitarian attributes were more important to them, and to prefer human over AI recommenders when hedonic attributes were more important to them. We benchmarked these hypotheses with a condition in which people chose between two human recommenders, wherein we expected recommender preference to be uncorrelated with importance assigned to hedonic/utilitarian attributes.
Three hundred three respondents (Mage = 38.0 years, SD = 11.1; 49.5% female) recruited on Amazon Mechanical Turk participated in exchange for monetary compensation. Participants imagined that they were planning to purchase a new winter coat (as it was the winter season) and were looking for recommendations. Participants read that winter coats have functional/utilitarian aspects ("Winter coats have functional or utilitarian aspects, such as insulating power, breathability, and the degree to which the coat is rain and wind proof") and sensory/hedonic aspects ("Winter coats have sensory or hedonic aspects, such as the color and other aesthetics, the way the fabric feels to the touch, and the degree to which the coat fits well"). Then, to measure the importance of hedonic/utilitarian attributes, participants rated the extent to which, in general, they cared about sensory/hedonic and functional/utilitarian aspects in winter coats (1 = "mostly care about functional/utilitarian aspects," and 7 = "mostly care about sensory/hedonic aspects").
Participants then read that to get recommendations about winter coats, they could rely on one of two shopping assistants, X or Y. We specified that both assistants had access to the same type and size of database, would charge the same fees, would generate recommendations autonomously, and were trained to serve users well and to the best of their capacity. To control for the possibility that different recommenders would be associated with different service quality perceptions, we also specified that the two shopping assistants had the same rating of 4.9/5.0 stars provided by 687 consumers that had used their services in the past. To manipulate choice set, half of the participants chose between two human shopping assistants (both X and Y were people and were described as two different sales associates at that particular retailer), and the other half chose between a human assistant, X, and an AI assistant, Y. Thus, whereas X was always human, Y was either human or AI depending on the condition. Finally, participants indicated their preference for one of the assistants (1 = "definitely shopping assistant X," 4 = "indifferent," and 7 = "definitely shopping assistant Y").
We regressed recommender preference on choice set (human–human vs. human–AI), hedonic/utilitarian attribute importance, and their interaction. This analysis revealed significant main effects of choice set (b =.85, t(299) = 5.49, p <.001) and hedonic/utilitarian attribute importance (b =.32, t(299) = 7.46, p <.001), as well as a significant two-way interaction (b = −.29, t(299) = −6.91, p <.001). As hedonic/utilitarian attribute importance was continuous, we explored the interaction using the Johnson–Neyman floodlight technique ([39]), which revealed a significant effect of recommender preference in human–AI choice set for levels of hedonic/utilitarian attribute importance lower than 2.35 (bJN =.15, SE =.08, p =.050) and higher than 3.36 (bJN = −.14, SE =.07, p =.050). That is, the more participants cared about utilitarian attributes (values lower than 2.35 on the seven-point scale), the more they preferred an AI assistant over a human one. Conversely, the more participants cared about hedonic attributes (values higher than 3.36 on the seven-point scale), the more they preferred a human assistant over an AI one. As predicted, in the human–human choice set, which served as the control condition, participants were indifferent between the two assistants (M = 3.98, SD =.34) and recommender preference was uncorrelated with hedonic/utilitarian attribute importance (r =.116, p =.162; see Figure 1).
Graph: Figure 1. Results of Study 3: Preference for AI (human) recommenders when utilitarian (hedonic) attributes are more important.Notes: The y-axis represents preference for recommender measured on a seven-point scale anchored at 1 = "definitely shopping assistant X," and 7 = "definitely shopping assistant Y." The x-axis represents importance of hedonic/utilitarian attributes, measured on a seven-point scale anchored at 1 = "mostly care about functional/utilitarian aspects," and 7 = "mostly care about sensory/hedonic aspects." The shaded region represents area of significance.
These results provided correlational evidence that hedonic/utilitarian attribute importance predicts preference between human and AI recommenders. The next study utilizes an analysis of mediation to test competence perceptions as drivers of the word-of-machine effect.
Study 4 uses an analysis of mediation to measure competence perceptions as lay beliefs underlying the word-of-machine effect. In addition, this study tests attribute complexity as an alternative explanation: a belief that AI recommenders are better capable to process more complex attribute information than human recommenders. One could argue that utilitarian attributes seem more complex to evaluate than hedonic attributes. If this argument is accurate, preference for AI recommenders when utilitarian attributes are more salient could be explained by a lay belief about the recommender's ability, higher for AI recommenders, to deal with complexity.[ 7] We tested this alternative explanation by manipulating attribute complexity orthogonally to recommender type (human, AI) and activated goal (hedonic, utilitarian). We manipulated attribute complexity by way of number of product attributes, which is consistent with prior research ([25]; [40]).
Four hundred two participants (Mage = 38.5 years, SD = 12.6; 46% female) from Amazon Mechanical Turk participated in exchange for monetary compensation in a 2 (complexity: low, high) × 2 (goal: hedonic, utilitarian) × 2 (recommender: human, AI) between-subjects design.
Participants read about the beta testing of a new app created to give recommendations of chocolate varieties by relying on one of two sources: a human or an AI master chocolatier (i.e., a computer algorithm). We told participants that the human and AI recommenders relied on the same database of chocolate varieties and operated autonomously. The app had the same cost regardless of recommender. Participants saw screenshots of the app (Figure 2).
Graph: Figure 2. Stimuli of Study 4.
We specified that the ratings of the chocolate varieties in the data set were not based on personal experience but rather that they had been rated by consumers and manufacturers in terms of certain dimensions that varied by complexity condition. In the high complexity condition, we described the chocolate varieties as being rated on eight attributes, four of which were hedonic (sensory pleasure, taste, fun factor, and pairing combinations) and four of which were utilitarian (chemical profile, nutritional index, digestibility profile, and health factor). In the low complexity condition, we described the chocolate varieties as being rated on two attributes, one of which was hedonic (sensory pleasure) and one of which was utilitarian (chemical profile).We then activated either a hedonic or a utilitarian goal by asking participants to set aside all thoughts they might already have had about chocolate and instead imagine that they wanted a recommendation based only on ( 1) sensory pleasure, taste, fun factor, and pairing combinations (hedonic/high complexity); ( 2) sensory pleasure (hedonic/low complexity); ( 3) chemical profile, nutritional index, digestibility profile, and health factor (utilitarian/high complexity); or ( 4) chemical profile (utilitarian/low complexity). Finally, we manipulated recommender in a two-cell (recommender: human, AI) between-subjects design by telling participants that in the version of the app they were considering, it was either the human or the AI master chocolatier that would give them a recommendation.
As a behavioral dependent variable, we asked participants if they wanted to download the chocolate recommendation at the end of the survey (yes, no), specifying that payment would not be conditional on electing to download the recommendation (which is consistent with previous research; see [10]). We then measured the hypothesized mediator (competence perceptions) by asking participants to rate the extent to which they thought the human (AI) recommender ( 1) was competent to recommend the type of chocolate they were looking for and ( 2) could do a good job recommending the type of chocolate they were looking for (1 = "strongly disagree," and 7 = "strongly agree"; r =.89, p <.001).[ 8] At the very end of the survey, participants who elected to download the recommendation were automatically directed to a downloadable PDF document with information about the chocolate (a relatively more indulgent hazelnut-based chocolate called "gianduiotti" in the hedonic condition or a relatively healthier chocolate toasted at low temperature called "crudista" in the utilitarian condition).
We assessed behavior (i.e., the proportion of participants who decided to download vs. not download the recommendation) by using a logistic regression with complexity, goal, recommender, and their two-way and three-way interactions as independent variables (all contrast coded) and download (1 = yes, 0 = no) as dependent variable. We found no significant main effect of complexity (B = −.04, Wald =.09, 1 d.f., p =.77) or goal (B =.03, Wald =.06, 1 d.f., p =.81), and we found a marginally significant main effect of recommender (B =.25, Wald = 3.75, 1 d.f., p =.053). The three-way goal × recommender × complexity interaction was not significant (B = −.11, Wald =.80, 1 d.f., p =.37), ruling out the role of complexity. In terms of two-way interactions, complexity did not interact with goal (B = −.13, Wald = 1.04, 1 d.f., p =.31) nor with recommender (B = −.18, Wald = 1.99, 1 d.f., p =.16). Replicating previous results, the two-way goal × recommender interaction was significant (B =.75, Wald = 34.60, 1 d.f., p <.001). The AI recommender led to more downloads than the human recommender when the goal was utilitarian (MAI = 82%, MH = 63%; z = 3.10, p =.002) and fewer downloads when the goal was hedonic (MAI = 52%, MH = 88%; z = −5.63, p <.001).
A 2 × 2 × 2 ANOVA on competence perceptions revealed no significant main effect of complexity (F( 1, 394) = 1.24, p =.27) and significant main effects of goal (F( 1, 394) = 8.99, p =.003) and recommender (F( 1, 394) = 19.81, p <.001). The three-way complexity × goal × recommender interaction was not significant (F( 1, 394) =.64, p =.44), ruling out complexity. In terms of two-way interactions, complexity did not interact with goal (F( 1, 394) =.61, p =.44), nor with recommender (F( 1, 394) =.36, p =.55). Importantly, the two-way goal × recommender interaction was significant (F( 1, 394) = 57.63, p <.001). Planned contrasts revealed that participants perceived the AI recommender as more competent than the human recommender in the case of a utilitarian goal (MAI = 5.92, SDAI = 1.10; MH = 5.50, SDH = 1.38; F( 1, 394) = 4.90, p =.027) and less competent in the case of a hedonic goal (MAI = 4.51, SDAI = 1.77; MH = 6.13, SDH =.96; F( 1, 394) = 73.04, p <.001).
We ran a moderated mediation model using PROCESS Model 8 ( 5,000 resamples; Hayes 2018). In this model, the moderating effect of goal takes place before the mediator (competence perceptions). The interaction between recommender and goal was significant (95% CI =.38 to.64) in the path between the independent variable and the mediator but not in the path between the independent variable and the dependent variable (95% CI = −.08 to.63). As predicted, the indirect effect recommender → competence perceptions → download was significant but in the opposite direction conditionally on the moderator (hedonic: 95% CI = 1.19 to 2.40; utilitarian: 95% CI = −.88 to −.06).
These results provide evidence for the hypothesized role of competence perceptions as drivers of the word-of-machine effect. Participants rated AI recommenders as more (less) competent in the case of utilitarian (hedonic) goals. Differential competence perceptions explained higher choice likelihood for the AI's recommendation than the human's if a utilitarian goal had been activated and lower choice likelihood for the AI's recommendation than the human's if a hedonic goal had been activated. Furthermore, we did not find evidence that the word-of-machine effect was moderated by complexity. The next three studies tested the scope of the word-of-machine effect by identifying boundary conditions.
Study 5 explores a circumstance under which the word-of-machine effect might reverse: when consumers want a recommendation that matches their unique needs and preferences.[ 9] Matching a recommendation to one's preferences is valued and might even be expected ([16]). In this study, we tested the hypothesis that consumers view the task of matching a recommendation to one's unique preferences as being better performed by a person than by AI.[10] This argument is in line with recent research in the medical domain showing that consumers perceive AI as less able than a human physician to tailor a medical recommendation to their unique characteristics and circumstances ([27]). Thus, we expected people to choose AI recommenders at a lower rate and, conversely, choose human recommenders at a higher rate if matching to unique preferences was salient, even in the case of an activated utilitarian goal. In other words, if matching to unique preferences was salient, we expected people to prefer a human recommender for both hedonic and utilitarian goals. We tested this possibility by manipulating whether participants' desire to have a recommendation matched to their unique needs and preferences was salient and then measuring their choice of recommender.
Five hundred forty-five respondents (Mage = 39.0 years, SD = 12.9; 46.6% female) from Amazon Mechanical Turk participated in exchange for monetary compensation in a 2 (goal: hedonic, utilitarian) x 2 (matching: unique preferences, control) between-subjects design. Participants read information about the beta testing of a new smartphone app offered by a real estate service. The app would allow users to chat with a Realtor to find properties to buy or rent. Participants further read that there were two versions of this app. In one version of the app, users would interact with a human Realtor, and in the other version, users would interact with an AI Realtor (i.e., a computer algorithm). Participants saw screenshots of the app (Figure 3) and read about how the app would work: the users would indicate what attributes they were looking for in a property (square footage, number of rooms, budget) and the [Realtor/AI Realtor] would use [their/its] training and knowledge to make apartment recommendations. We specified that both the human and AI Realtors had access to the same number and type of property listings. We then activated either a hedonic or a utilitarian goal by asking participants to set aside all thoughts they might already have had about apartments and instead imagine that they wanted a recommendation based only on: ( 1) how trendy the neighborhood is, the apartment views, aesthetics (hedonic goal condition) or ( 2) distance to their workplace, proximity to public transport, functionality (utilitarian goal condition; based on [ 6]). Finally, to make unique preference matching salient, we told half of the participants that it was very important for them to get a recommendation that would be matched to their unique needs and personal preferences. Participants in the control condition were not focused on unique preference matching. As a dependent variable, we measured choice of recommender by asking participants if, given the circumstances described, they wanted to chat with the human or the AI Realtor.
Graph: Figure 3. Stimuli (top) and results (bottom) of Study 5: The word-of-machine effect is reversed for utilitarian goals if the recommendation needed to match participants' unique preferences.Notes: The y-axis represents the proportion of participants who chose to chat with the human versus AI realtor.
We assessed choice on the basis of the proportion of participants who decided to chat with the human versus AI Realtor by using a logistic regression with goal, matching, and their two-way interaction as independent variables (all contrast coded) and choice (0 = human, 1 = AI) as a dependent variable. We found significant effects of goal (B = 1.75, Wald = 95.70, 1 d.f., p <.000) and matching (B =.54, Wald = 24.30, 1 d.f., p <.000). More importantly, goal interacted with matching (B =.25, Wald = 5.33, 1 d.f., p =.021). Results in the control condition (when unique preference matching was not salient) replicated prior results: in the case of an activated utilitarian goal, a greater proportion of participants chose the AI Realtor (76.8%) over the human Realtor (23.2%; z = 8.91, p <.001), and when a hedonic goal was activated, a lower proportion of participants chose the AI (18.8%) over the human Realtor (81.2%; z = 10.35, p <.001). However, making unique preference matching salient reversed the word-of-machine effect in the case of an activated utilitarian goal: choice of the AI Realtor decreased to 40.3% (from 76.8% in the control; z = 6.17, p <.001). That is, making unique preference matching salient turned preference for the AI Realtor into resistance despite the activated utilitarian goal, with most participants choosing the human over the AI Realtor. In the case of an activated hedonic goal, making unique preference matching salient further strengthened participants' choice of the human Realtor, which increased to 88.5% from 81.2% in the control, although the effect was marginal, possibly due to a ceiling effect (z = 1.66, p =.097).
Overall, whereas the word-of-machine effect replicated in the control condition when unique preference matching was salient, participants preferred the human Realtor over the AI recommender both in the hedonic goal conditions (human = 88.5%, AI = 11.5%; z = 12.40, p <.001) and in the utilitarian goal conditions (human = 59.7%, AI = 40.3%; z = 3.24, p =.001; Figure 3), corroborating the notion that people view AI as unfit to perform the task of matching a recommendation to one's unique preferences.
These results show that preference matching is a boundary condition of the word-of-machine effect, which reversed in the case of a utilitarian goal when people had a salient goal to get recommendations matched to their unique preferences and needs. The next study tests another boundary condition.
Study 6 explores under what circumstances the word-of-machine effect is eliminated, and it tests the role of AI as boundary condition. Studies 1–5 tested cases in which the role of AI was to replace human recommenders. Study 6 explores the case in which AI is leveraged to assist and augment human intelligence. "Augmented intelligence" involves AI's assistive role in enhancing and amplifying human intelligence instead of replacing it ([ 3]). So far, we have showed that consumers resist AI recommenders when a hedonic goal is activated. In Study 6, we tested the hypothesis that consumers will be more receptive to AI recommenders, even in the case of a hedonic goal, if the AI recommender assists and amplifies a human recommender who retains the role of ultimate decision maker. In this case, we expected people to believe that the human decision maker would compensate for the AI's relative perceived incompetence in the hedonic realm. We expected the reverse effect in the case of a utilitarian goal. In other words, we expected that augmented intelligence—a human–AI hybrid decision making model— would help bolster AI to the level of humans for hedonic decision making and help bolster humans to the level of AI for utilitarian decision making. In addition, we added a control condition in Study 6 in which neither recommender was mentioned to serve as a baseline measure of participants' perceptions of hedonic and utilitarian attributes.
Four hundred four respondents (Mage = 40.2 years, SD = 12.5; 48.9% female) from Amazon Mechanical Turk participated in exchange for monetary compensation in a three-cell (recommender: human, artificial intelligence, augmented intelligence) between-subjects design. A fourth control condition contained no recommender manipulation and served as the baseline.
The stimuli and procedure were identical to those of Study 4. Participants read about the beta testing of a new app created to give recommendations of chocolate varieties by relying on one of two sources: a human or an AI master chocolatier. Participants read that human and AI recommenders relied on the same database, which comprised a large number of chocolate varieties that had been rated by consumers and manufacturers. Participants read that the app had the same cost regardless of the type of recommender it relied on. Finally, participants read that the app would suggest a curated selection of five chocolate bars.
We then manipulated recommender by randomly assigning participants to ( 1) a human condition, in which a human chocolatier would curate the chocolate section; ( 2) an artificial intelligence condition, in which an AI chocolatier (i.e., a computer algorithm) would curate the chocolate section; or ( 3) an augmented intelligence condition, in which the AI chocolatier would assist the human chocolatier in the curation of the chocolate selection. Specifically, participants read:
[Human condition] In the version of the app we are testing today, it is the human chocolatier that curates a selection of chocolate bars. This selection contains five chocolate bars selected by the human chocolatier. That is, it is a person who selects chocolate bars. This version of the app is technically called "human intelligence," because it uses what human intelligence can do.
[Artificial intelligence condition] In the version of the app we are testing today, it is the A.I. chocolatier that curates a selection of chocolate bars. This selection contains five chocolate bars selected by the A.I. chocolatier. That is, it is a computer algorithm that selects chocolate bars. This version of the app is technically called "artificial intelligence," because it uses a computer algorithm to substitute and replace what human intelligence can do.
[Augmented intelligence condition] In the version of the app we are testing today, it is the A.I. chocolatier that curates a selection of chocolate bars. This selection contains five chocolate bars selected by the A.I. chocolatier. That is, it is a computer algorithm that selects chocolate bars. The computer algorithm makes the initial selection and assists a human chocolatier, who will make the final decision about which chocolate bars to recommend. This version of the app is technically called "augmented intelligence," because it uses a computer algorithm to enhance and augment what human intelligence can do.
The control condition entailed no recommender manipulation; instead, it merely included a description of the app and no information about the source of the chocolate bar recommendation. As a dependent variable, we measured hedonic attribute perceptions with two items (indulgent taste and aromas; pleasantness to the senses [vision, touch, smell, etc.]) and utilitarian attribute perceptions with two items (beneficial chemical properties [antioxidants, etc.]; healthiness [micro/macro nutrients, etc.]), all on seven-point scales anchored at 1 = "very low," and 7 = "very high." The order of items was randomized.
The one-way ANOVA on the average of the two items measuring hedonic attribute perceptions (r =.79, p <.001) was significant (F( 1, 436) = 48.92, p <.001). In line with previous results, and replicating the word-of-machine effect, participants reported higher hedonic attribute perceptions when the recommender was human (MH = 6.00; SD = 1.06) than when the recommender was AI (Martificial_intelligence = 4.15, SD = 1.64; F( 1, 436) = 125.55, p <.001). However, when the AI recommender was augmenting human intelligence, the word-of-machine effect was eliminated: participants reported the same hedonic perceptions (Maugmented_intelligence = 5.74, SD = 1.11) as they did when the recommender was human (F( 1, 436) = 2.31, p =.129) and higher hedonic perceptions than when the recommender was AI alone (F( 1, 436) = 84.73, p <.001). Participants in the control condition reported lower hedonic perceptions (Mcontrol = 5.62, SD = 1.09) than participants in the human condition (F( 1, 436) = 5.32, p =.022) and higher hedonic perceptions than participants in the AI condition (F( 1, 436) = 77.92, p <.001). Control condition and augmented intelligence condition did not differ (F( 1, 436) < 1, p =.49).
The one-way ANOVA on the average of the two items measuring utilitarian attribute perceptions (r =.75, p <.001) was significant (F( 1, 436) = 6.60, p <.001). In line with previous results, and replicating the word-of-machine effect, participants reported higher utilitarian attribute perceptions when the recommender was AI (Martificial_intelligence = 5.24; SD = 1.41) than when the recommender was human (MH = 4.75, SD = 1.57; F( 1, 436) = 6.40, p =.012). However, when the AI recommender was augmenting human intelligence, the word-of-machine effect was eliminated: participants reported the same utilitarian perceptions (Maugmented_intelligence = 5.44, SD = 1.32) as they did when the recommender was AI alone (F( 1, 436) =.99, p =.321) and higher utilitarian perceptions than when the recommender was human (F( 1, 436) = 11.87, p <.001). Participants in the control condition reported the same utilitarian perceptions (Mcontrol = 4.70, SD = 1.56) as participants in the human condition (F( 1, 436) =.05, p =.820) and lower utilitarian perceptions than participants in both the AI (F( 1, 436) = 7.47, p =.007) and augmented intelligence conditions (F( 1, 436) = 13.22, p <.001; Figure 4).
Graph: Figure 4. Results of Study 6: The word-of-machine effect is eliminated in the case of augmented intelligence (human–AI hybrid decision making).Notes: The y-axis represents hedonic attribute perceptions and utilitarian attribute perceptions measured on seven-point scales anchored at 1 = "very low," and 7 = "very high." Error bars represent standard errors. The solid-line pairwise comparisons represent the word-of-machine effect. The dashed-line pairwise comparisons represent moderation by augmented intelligence: A human–AI hybrid decision making model bolsters AI to the level of humans for hedonic decision making, and humans to the level of AI for utilitarian decision making. Details of all pairwise comparisons are reported subsequently.Hedonic Attribute PerceptionsWord-of-machine effect: Human versus AI: F( 1, 436) = 125.55, p =.000Moderation by augmented intelligence (H + AI hybrid decision making bolsters AI to the level of humans for hedonic decision making): Human versus H + AI: F( 1, 436) = 2.31, p =.129AI versus H + AI: F( 1, 436) = 84.73, p =.000Control versus H: F( 1, 436) = 5.32, p =.022Control versus AI: F( 1, 436) = 77.92, p =.000Control versus H + AI: F( 1, 436) =.49, p =.486Utilitarian Attribute PerceptionsWord-of-machine effect: Human versus AI: F( 1, 436) = 6.40, p =.012Moderation by augmented intelligence (H + AI hybrid decision making bolsters H to the level of AI for utilitarian decision making): AI versus H + AI: F( 1, 436) =.99, p =.321H versus H + AI: F( 1, 436) = 11.87, p =.001Control versus H: F( 1, 436) =.05, p =.820Control versus AI: F( 1, 436) = 7.47, p =.007Control versus H + AI: F( 1, 436) = 13.22, p =.000
These results delineate the scope of the word-of-machine effect and show a circumstance under which the effect is eliminated. Even when a hedonic goal was activated, AI recommenders fared as well as human recommenders as long as they were in a hybrid decision-making model in partnership with a human.
Studies 7a and 7b test an intervention to attenuate the lay belief underlying the word-of-machine effect—that AI recommenders are less (more) competent than human recommenders in assessing hedonic (utilitarian) value. We used a protocol called "consider-the-opposite," in which people are prompted to consider the opposite of what they initially believe to be true and take into account evidence that is inconsistent with one's initial beliefs. This protocol has been effectively used to correct biased beliefs in judgment, such as the explanation bias ([29]), confirmatory hypothesis testing ([43]), anchoring ([33]) and halo effects in marketing claims ([35]). Study 7a tests this intervention following the original protocol (i.e., [33]), and Study 7b tests a protocol that is relatively easier to implement and scale by embedding the intervention in a real chatbot.
Three hundred sixty-eight respondents (Mage = 39.8 years, SD = 12.5; 49.2% female) from Amazon Mechanical Turk participated in exchange for monetary compensation in a 2 (recommender: human, AI) × 2 (intervention: consider the opposite, control) between-subjects design.
The stimuli and procedure were identical to those of Studies 4 and 6: participants read about a new app created to give chocolate recommendations by relying on either a human or an AI master chocolatier. We manipulated recommender between subjects by telling participants that, in the version of the app they were considering, it was either the human or the AI chocolatier that would suggest a curated selection of five chocolate bars. We also implemented the intervention between subjects by prompting half of the participants to "consider the opposite": consider the ways in which they could be wrong about what they expected the [human/AI] recommender to be good at (based on [33]):
Think for a moment about what you expect the [human/AI] chocolatier to be good at when selecting chocolate bars. Before you rate the chocolate selection, we would like you to consider the opposite. Can your expectations about what the human chocolatier is good at when selecting chocolates be wrong? Imagine that you were trying to be as unbiased as possible in evaluating this chocolate selection—consider yourself to be in the same role as a judge or juror. Could the [human/AI] chocolatier be good at the opposite of what you expect them to be good at? Please write down some ways in which you could be wrong in terms of your expectations about what the [human/AI] chocolatier is good at when selecting chocolates.
This prompt was absent for participants in the control condition. As a dependent variable, participants reported their perceptions of hedonic/utilitarian attributes of the curated selection of chocolate bars, measured on a seven-point scale ranging from 1 = "sensory pleasure (taste, aromas, etc.)" to 7 = "healthy chemical properties (antioxidants, micro/macro nutrients, etc.)." Thus, lower numbers indicated higher hedonic value.
A 2 × 2 ANOVA on hedonic/utilitarian attribute perceptions revealed no significant main effect of intervention (F( 1, 364) =.25, p =.62), a significant main effect of recommender (F( 1, 364) = 65.17, p <.001), and a significant two-way recommender × intervention interaction (F( 1, 364) = 12.11, p =.001). Planned contrasts revealed that the word-of-machine effect replicated both in the control and intervention conditions, with lower hedonic perceptions (or, conversely, higher utilitarian perceptions) for AI recommenders than human recommenders (control conditions: MAI_control = 4.51, SD = 1.84, MH_control = 2.49, SD = 1.48; F( 1, 364) = 81.48, p <.001; intervention conditions: MAI_intervention = 3.99, SD = 1.78, MH_intervention = 3.18, SD = 1.44; F( 1, 364) = 8.93, p =.003; higher numbers indicate higher utilitarian/lower hedonic perceptions). More importantly, the intervention attenuated the word-of-machine effect and led to participants perceiving the AI's recommendation as having higher hedonic value compared with the control condition (MAI_intervention = 3.99, SD = 1.78, MAI_control = 4.51, SD = 1.84; F( 1, 364) = 7.66, p =.006) and the human recommendation as having higher utilitarian value compared to the control condition (MH_intervention = 3.18, SD = 1.44; MH_control = 2.49, SD = 1.48, F( 1, 364) = 4.59, p =.033; higher numbers indicate higher utilitarian/lower hedonic perceptions; Figure 5).
Graph: Figure 5. Results of Study 7a: Prompting people to consider the opposite attenuated the word-of-machine effect.Notes: The y-axis represents perceived hedonic/utilitarian attribute value measured on a seven-point scale anchored at 1 = "sensory pleasure (taste, aromas, etc.), and 7 = "healthy chemical properties (antioxidants, micro/macro nutrients, etc.)"; therefore, higher numbers indicate higher utilitarian value/lower hedonic value. Error bars represent standard errors.
Thus, these results provide evidence for a potential intervention that alleviates initial beliefs about, and therefore resistance to, AI recommenders: prompting people to consider the opposite.
Study 7b builds on the original consider-the-opposite protocol and the results of Study 7a to test an intervention better suited for implementation and scalability in a real-world setting. To do so, we created a real chatbot that participants could interact with and that delivered the intervention.
Two hundred nighty-nine respondents (Mage = 40.4 years, SD = 12.6; 43.1% female) from Amazon Mechanical Turk participated in exchange for monetary compensation in a two-cell (intervention: consider the opposite, control) between-subjects design. Participants read about an app called "Cucina" that would rely on AI to give recipe recommendations. The app worked by giving users the chance to chat with the AI Chef and ask for recipe suggestions and recommendations. Participants further read that they could try out the AI Chef by chatting with it in a web browser window. We created a chatbot ad hoc for this experiment by embedding a JavaScript in the Qualtrics survey (Figure 6). The chatbot was programmed to first introduce itself: "Hello I am an A.I. Chef at Cucina! Thank you for trying out our app! What is your name?" Participants could then reply to the chatbot using a text box. We programmed the chatbot's next response to differ depending on the intervention condition:
[Intervention: consider the opposite] "Hi [participant's name]! I am here to suggest a recipe for you to try! Some people might think that an Artificial Intelligence Chef is not competent to give food suggestions...but this is a misjudgment. For a moment, set aside your expectations about me. When it comes to making food suggestions, could you consider the idea that I could be good at things you do not expect me to be good at? Okay, let's chat about food. How can I help you?"
Graph: Figure 6.Stimuli of Study 7b.
[Intervention: control] "Hi [participant's name]! I am here to suggest a recipe for you to try! Okay, let's chat about food. How can I help you?"
As a dependent variable, we measured hedonic/utilitarian attribute perceptions of the recipes suggested by the AI chatbot, as measured on a seven-point scale ranging from 1 = "mostly based on sensory pleasure (taste, aromas, etc.)" to 7 = "mostly based on healthy chemical properties (antioxidants, micro/macro nutrients, etc.)."
A one-way ANOVA on hedonic/utilitarian attribute perceptions revealed that the intervention attenuated the word-of-machine effect and led to higher hedonic perceptions compared to the control condition (Mintervention = 3.75, SD = 1.46, Mcontrol = 4.25, SD = 1.37; F( 1, 297) = 9.15, p =.003; lower numbers indicate higher hedonic perceptions). These results corroborate those of Study 7a and provide evidence for a practical and relatively easier-to-implement intervention for managers looking to attenuate the lay belief underlying the word-of-machine effect.
As companies in the private and public sectors assess how to harness the potential of AI-driven recommendations, the question of how trade-offs in decision making influence preference for AI recommenders is of great importance. We address this question across nine studies and show a word-of-machine effect: the phenomenon by which hedonic and utilitarian trade-offs determine preference for (or resistance to) AI-driven recommendations. Studies 1a–1b show that a utilitarian (hedonic) goal makes people more (less) likely to choose AI recommenders than human ones. Study 2 shows that AI (human) recommenders lead to higher perceptions of utilitarian (hedonic) attributes upon consumption. Study 3 shows that people prefer AI (human) recommenders when utilitarian (hedonic) attributes are more important. Study 4 shows that differing competence perceptions underlie the word-of-machine effect and rule out complexity. Studies 5 and 6 identify boundary conditions: Study 5 shows that the word-of-machine effect is reversed for utilitarian goals if the recommendation needs to match a person's unique preferences, and Study 6 shows that the effect is eliminated when AI is framed as "augmented" rather than "artificial" intelligence, that is, in human–AI hybrid decision making. Finally, Studies 7a–7b tested an intervention to attenuate the word-of-machine effect.
Our research makes several important theoretical contributions. A first set of contributions speaks to research on the psychology of automation and on human–technology interactions ([12]; [17]; [30]). First, we extend this literature by addressing the question of whether hedonic/utilitarian trade-offs in decision making drive preference for or resistance to AI recommenders. This question is novel, as prior research has not relied on differences inherent to hedonic/utilitarian consumption to predict people's reactions to receiving advice from automated systems.
Second, we show under what circumstances AI-driven recommendations are preferred to, and therefore more effective, than human ones: when utilitarian attributes are relatively more important or salient than hedonic ones. Research in this area has largely focused on consumers' resistance to automated systems. For example, in the domain of performance forecasts, people are less likely to rely on the input of an algorithm than a person to make predictions about student performance, an effect that is due to the belief that algorithms, unlike people, cannot learn from their mistakes ([14]). In the domain of health care utilization, people are less likely to rely on an automated medical provider if a human provider is available, even when the two providers have the same accuracy ([27], [28]).
Limited research has identified under what circumstances resistance to algorithmic advice is attenuated: if people have the opportunity to modify algorithms and thus exert control over them ([15]), if the human likeness of algorithms is increased ([ 9]), if the task entails a numeric estimate of a target ([26]), and if the algorithm is described as tailoring a recommendation to a person's unique case ([27], [28]). We extend this literature by showing circumstances in which consumers' resistance to AI may be reversed and by showing cases in which consumers even prefer automated systems: when they assign greater importance to utilitarian attributes or when a utilitarian goal is activated.
Third, we explore under what circumstances consumers will be amenable to AI recommenders in the context of human–AI partnerships. We show that augmented intelligence helps bolster AI to the level of humans for hedonic decision making and helps bolster humans to the level of AI for utilitarian decision making. This contribution is important because it represents the first empirical test of augmented intelligence as an alternative conceptualization of artificial intelligence that focuses on AI's assistive role in advancing human capabilities. We hope that this contribution will prioritize new research focused on understanding the potential of AI in conjunction with humans rather than in contraposition, as this seems to be the advocated way forward by many practitioners ([ 3]; [18]).
We also contribute to the literature on hedonic and utilitarian consumption ([ 1]; [22]; [31]; [44]). Literature in this area has identified the factors that influence evaluation of hedonic and utilitarian product dimensions. We extend this literature by investigating how hedonic/utilitarian attribute trade-offs influence the effectiveness of a source of a product recommendation (i.e., a human vs. an AI recommender; Studies 1a, 1b, 3–5) and how the source of a product recommendation influences hedonic/utilitarian perceptions (Studies 2, 6–7b).
The current speed of development and adoption of AI, machine learning, and natural language processing algorithms challenge managers to harness these transformative technologies to optimize the customer experience. Our findings are insightful for managers as they navigate the remarkable technology-enabled opportunities that are growing in today's marketplace. These new technologies are also experiencing a renewed prominence in public discourse. For instance, the U.S. government has established the National Artificial Intelligence Research and Development Strategy to address economic and social implications of AI.
Our findings provide useful insights for both companies and public policy organizations debating if and how to effectively automate their recommendations systems. A company like Sephora relies both on human-based recommendations from sales associates and its customer base and AI-based recommendations through its Visual Artist app, a conversational bot that interacts with prospective shoppers. Our results suggest cases in which AI-based recommendations would be more effective (i.e., when utilitarian attributes are more salient or important, such as grooming products) and when they would be less effective (i.e., when hedonic attributes are more salient or important, such as fragrances).
Our results are insightful for strategic and tactical marketing decisions. Marketers could prioritize functional positioning strategies over experiential ones in the case of AI-based recommendations for target segments for whom utilitarian attributes are more important. For instance, a company in the hospitality industry such as TripAdvisor should emphasize AI-based recommendations for business travel services and deemphasize AI-based recommendations for leisure travel services. Our results also apply to a host of tactical decisions such as marketing communications. Managers could communicate to their customers in a way that is aligned with a target segment's goal (i.e., hedonic vs. utilitarian) and emphasize the most effective points of parity/difference with competing brands or across different products in the portfolio. Companies like Netflix and YouTube could emphasize AI-based recommendations when utilitarian attributes are relatively more important (e.g., documentaries) and human-based recommendations ("similar users") when hedonic attributes are relatively more important (e.g., horror movies).
This research also highlights boundary conditions that may prove useful for practitioners. Study 5 indicated that when consumers want recommendations that are matched to their unique preferences, they resist AI recommenders and instead prefer human recommenders, regardless of hedonic or utilitarian goals. These results suggest that companies whose customers are known to be satisfied with "one size fits all" recommendations, or who are not in need of a high level of customization, may rely on AI systems. However, companies whose customers are known to desire personalized recommendations should rely on humans. Some companies, such as Amazon, seem to be implementing a similar strategy. Even though most of Amazon's recommendations are based on algorithms, the company has recently started offering an additional service for an added fee called "personal shopper." This service relies on human shopping assistants to give clothing recommendations rather than on algorithms. Our results indicate that more companies, especially those in markets that are relatively more hedonic, should follow Amazon's example.
Study 6 provides another managerially relevant boundary condition: augmented intelligence. The results of this study indicate that consumers are more receptive to AI recommenders, even in the case of hedonic goals, if the AI recommender does not replace a human recommender but instead assists a human recommender who retains the role of ultimate decision maker. These results are important for practitioners managing relatively more hedonic products or services. For instance, in a personal conversation with the authors, a Walmart marketing manager noted how the top two most frequently ignored recommendations on the company's website are those for alcoholic beverages and food items—arguably products for which hedonic attributes tend to be more salient and important. In these circumstances, practitioners could leverage our results and utilize AI systems to generate an initial recommendation on which a human then "signs off."
Finally, in Studies 7a–7b we tested an intervention that practitioners managing relatively more hedonic products and relying on AI systems may execute. Building on the consider-the-opposite protocol, we created a realistic chatbot that interacted with participants and nudged them to consider that the AI recommender could be good at things that participants did not expect it to be good at. The intervention was successful in both studies, suggesting that practitioners may utilize this technique if hedonic attributes are important.
Despite the robustness of the word-of-machine effect, our research has limitations that offer several opportunities for future research. First, there is the possibility that drawing attention to the source of a recommendation primed study participants. AI recommenders might have primed utilitarian attributes or made utilitarian goals more salient, and it was the associated increased activation of these concepts, rather than competence perceptions, that gave rise to the word-of-machine effect. Although possible, this alternative explanation based on priming is unlikely given the results of a study we report in Web Appendix D. In this study (N = 230), we first primed participants with either human or AI-related concepts by drawing their attention to either a human or an AI recommender, thus approximating the kind of priming that could have occurred in our studies. To assess whether the AI recommender primed utilitarian concepts, we then measured perceptions of utilitarian and hedonic attributes of a stimulus in a domain unrelated to one in which the priming manipulation occurred. This stimulus was pretested to be neutral (i.e., perceived to be equally utilitarian and hedonic). The results indicate that the stimulus was perceived to be equally utilitarian and hedonic regardless of the priming manipulation. Although these results offer preliminary evidence that priming does not account for the word-of-machine effect, the inferences one can draw from a null effect are limited. More broadly, the question of whether AI-based recommendations activate specific constructs that might be influential on decision making is a worthy avenue for future research.
Second, even though we tested the word-of-machine effect across multiple domains, there remains the possibility that the effect is stronger or weaker in certain categories. For instance, the effect might be stronger in categories (e.g., a chocolate cake) in which discerning hedonic attributes (e.g., how tasty or how indulgent it is) is easier than discerning utilitarian attributes (e.g., how many macronutrients it contains, or how healthy it is). Future research could more systematically investigate what dimensions of different product categories strengthen versus weaken the word-of-machine effect.
Third, the lay beliefs underlying the word-of-machine effect may be transitional. As competence perceptions driving the word-of-machine effect are based on a lay belief, they are embedded in a cultural view that may change over time. The lay belief about differential competence perceptions may already be inaccurate, as AI is already utilized in domains that are relatively more hedonic. For instance, AI curates flower arrangements on the basis of customers' past transactions and inferred preferences (1-800-Flowers) and creates new flavors for food companies such as McCormick, Starbucks, and Coca-Cola ([41]).
Our research also suggests opportunities for future exploration of this area. First, the word-of-machine effect may have interesting downstream consequences on other responses. For instance, relying on an AI recommender may lead consumers to compensate by adjusting their own choices. Given the belief that AI-based recommendations excel on utilitarian attributes and are weaker on hedonic attributes, consumers may choose from a set of options by paying closer attention to the hedonic attributes of the options, assuming that the options are satisfactory in terms of utilitarian attributes. This "second-step choice" is an interesting question to consider in the future.
Second, in Studies 7a–b we show preliminary evidence of how lay beliefs toward AI systems could be successfully alleviated through a protocol utilized in the decision making literature. Future research could identify other real-world variables that might have similar attenuating effects, such as domain expertise, involvement, time spent making decisions, or familiarity/repeated use of AI systems. A third fruitful research opportunity would be to explore whether consumers can be persuaded to trust AI systems, even more than humans, in the eventuality that AI systems are sufficiently sophisticated to pass the Turing test. In this vein, future research could identify conditions under which the word-of-machine effect reverses, with AI recommenders being more persuasive than humans for hedonic products.
As research on the psychology of automation expands to include developments such as AI, we hope that our findings (especially those of Study 6) will spur further research prioritizing the understanding of the vast potential of AI operating in partnership with humans. More research is also necessary to map out the impact of AI systems across consumption settings. AI-powered technologies will be instrumental in optimizing the customer experience at each phase of the consumer journey by offering products of increasing personalization ([41]). New technologies like image, text, and voice recognition, together with large-scale A/B testing will provide managers with the data necessary for a complete, AI-driven customization of the journey ([41]) and will allow researchers to gather the consumer signals that are produced as a by-product of consumer activities ([38]). We hope that future research will focus on how to harness this great potential of AI for managers and researchers alike.
Overall, understanding when consumers will be amenable to and when they will resist AI-driven recommendations is a pressing and complex endeavor for researchers and firms alike. We hope that our research will spur further exploration of this important topic.
Supplemental Material, WOM_Web_Appendix_PDF - Artificial Intelligence in Utilitarian vs. Hedonic Contexts: The "Word-of-Machine" Effect
Supplemental Material, WOM_Web_Appendix_PDF for Artificial Intelligence in Utilitarian vs. Hedonic Contexts: The "Word-of-Machine" Effect by Chiara Longoni and Luca Cian in Journal of Marketing
Footnotes 1 All authors contributed equally.
2 Connie Pechmann
3 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
4 The author(s) received no financial support for the research, authorship, and/or publication of this article.
5 Luca Cian https://orcid.org/0000-0002-8051-1366
6 https://doi.org/10.1177/0022242920957347
7 We thank the associate editor and two anonymous reviewers for this suggestion.
8 We added manipulation checks at the end of the survey. These manipulation checks were of recommender and goal, and participants indicated the recommender (human, AI) of the app they considered and the goal they had (hedonic, utilitarian). The recommender manipulation check was answered correctly by 93.0% of the participants, and the goal manipulation check was answered correctly by 90.8% of the participants. Statistical conclusions did not differ when restricting the analysis to participants who passed either manipulation check.
9 We thank an anonymous reviewer for this suggestion.
We validated this hypothesis by asking respondents from Amazon Mechanical Turk (N = 95) the extent to which they would expect a property selected by [a human/an AI] Realtor to match their unique preferences and needs (1 = "not at all," and 7 = "very much"). Indeed, participants expected the human Realtor to be more able than the AI Realtor to match a property recommendation to their unique preferences and needs (MH = 5.85, SD = 0.82, MAI = 4.70, SD = 1.30; F(1, 93) = 26.69, p <.001).
References Alba Joseph W. , Williams Elanor F.. (2013), " Pleasure Principles: A Review of Research on Hedonic Consumption ," Journal of Consumer Psychology , 23 (1), 2 – 18.
Allison Ralph I. , Uhl Kenneth P.. (1964), " Influence of Beer Brand Identification on Taste Perception ," Journal of Marketing Research , 1 (3), 36 – 39.
Araya Daniel. (2019), "3 Things You Need to Know About Augmented Intelligence," Forbes (January 22) , https://www.forbes.com/sites/danielaraya/2019/01/22/3-things-you-need-to-know-about-augmented-intelligence/.
Batra Rajeev , Ahtola Olli T.. (1991), " Measuring the Hedonic and Utilitarian Sources of Consumer Attitudes ," Marketing Letters , 2 (4), 159 – 70.
Bazerman Max H. , Tenbrunsel Ann E. , Wade-Benzoni Kimberly. (1998), " Negotiating with Yourself and Losing: Understanding and Managing Conflicting Internal Preferences ," Academy of Management Review , 23 (2), 225 – 41.
Bhargave Rajesh , Chakravarti Amitav , Guha Abhijit. (2015), " Two-Stage Decisions Increase Preference for Hedonic Options ," Organizational Behavior and Human Decision Processes , 130 , 123 – 35.
Boatsman James , Moeckel Cindy , Pei Buck K.W.. (1997), " The Effects of Decision Consequences on Auditors' Reliance on Decision Aids in Audit Planning ," Organizational Behavior and Human Decision Processes , 71 (2), 211 – 47.
Botti Simona , McGill Ann L.. (2011), " Locus of Choice: Personal Causality and Satisfaction with Hedonic and Utilitarian Decisions ," Journal of Consumer Research , 37 (4), 1065 – 78.
Castelo Noah , Bos Maarten W. , Lehman Donald R.. (2019), " Task-Dependent Algorithm Aversion ," Journal of Marketing Research , 56 (5), 809 – 25.
Cian Luca , Longoni Chiara , Krishna Aradhna. (2020), " Advertising a Desired Change: When Process Simulation Fosters (vs. Hinders) Credibility and Persuasion ," Journal of Marketing Research , 57 (3), 489 – 508.
Crowley Ayn E. , Spangenberg Eric R. , Hughes Kevin R.. (1991), " Measuring the Hedonic and Utilitarian Dimensions of Attitudes Toward Product Categories ," Marketing Letters , 3 (3), 239 – 49.
Dawes Robyn M.. (1979), " The Robust Beauty of Improper Linear Models in Decision Making ," American Psychologist , 34 (7), 571 – 82.
Dhar Ravi , Wertenbroch Klaus. (2000), " Consumer Choice Between Hedonic and Utilitarian Goods ," Journal of Marketing Research , 37 (1), 60 – 71.
Dietvorst Berkeley J. , Simmons Joseph P. , Massey Cade. (2014), " Algorithm Aversion: People Erroneously Avoid Algorithms After Seeing Them Err ," Journal of Experimental Psychology: General , 144 (1), 114 – 26.
Dietvorst Berkeley J. , Simmons Joseph P. , Massey Cade. (2016), " Overcoming Algorithm Aversion: People Will Use Imperfect Algorithms If They Can (Even Slightly) Modify Them ," Management Science , 64 (3), 1155 – 70.
Franke Nikolaus , Keinz Peter , Steger Christoph J.. (2009), " Testing the Value of Customization: When Do Customers Really Prefer Products Tailored to Their Preferences? " Journal of Marketing , 73 (5), 103 – 21.
Grove William M. , Meehl Paul E.. (1996), " Comparative Efficiency of Informal (Subjective, Impressionistic) and Formal (Mechanical, Algorithmic) Prediction Procedures: The Clinical-Statistical Controversy ," Psychology, Public Policy, and Law , 2 (2), 293 – 323.
Hao Karen. (2020), " AI Is Learning When It Should and Shouldn't Defer to a Human ," MIT Review (August 5), https://www.technologyreview.com/2020/08/05/1006003/ai-machine-learning-defer-to-human-expert/.
Highhouse Scott. (2008) " Stubborn Reliance on Intuition and Subjectivity in Employee Selection ," Industrial and Organizational Psychology: Perspectives on Science and Practice , 1 (3), 333 – 42.
Hirschman Elizabeth C. , Holbrook Morris B.. (1982), " Hedonic Consumption: Emerging Concepts, Methods, and Propositions ," Journal of Marketing , 46 (3), 92 – 101.
Holbrook Morris B.. (1994), " The Nature of Customer Value ," in Service Quality: New Directions in Theory and Practice , Rust Roland T. , Oliver Richard L. , eds. Thousand Oaks, CA : SAGE Publications , 21 – 71.
Khan Uzma , Dhar Ravi. (2010), " Price-Framing Effects on the Purchase of Hedonic and Utilitarian Bundles ," Journal of Marketing Research , 47 (6), 1090 – 99.
Khan Uzma , Dhar Ravi , Wertenbroch Klaus. (2005), " A Behavioral Decision Theory Perspective on Hedonic and Utilitarian Choice ," in Inside Consumption: Frontiers of Research on Consumer Motives, Goals, & Desires , Ratneshwar S. , Mick David Glen , eds. Abingdon-on-Thames, UK : Routledge , 144 – 65.
Leung Eugina , Paolacci Gabriele , Puntoni Stefano. (2019), " Man Versus Machine: Resisting Automation in Identity-Based Consumer Behavior ," Journal of Marketing Research , 55 (6), 818 – 31.
Littrell Mary A. , Miller Nancy J.. (2001), " Marketing Across Cultures: Consumers' Perceptions of Product Complexity, Familiarity, and Compatibility ," Journal of Global Marketing , 15 (1), 67 – 86.
Logg Jennifer M. , Minson Julia , Moore Dan A.. (2019), " Algorithm Appreciation: People Prefer Algorithmic to Human Judgment ," Organizational Behavior and Human Decision Processes , 151 , 90 – 103.
Longoni Chiara , Bonezzi Andrea , Morewedge Carey K.. (2019), " Resistance to Medical Artificial Intelligence ," Journal of Consumer Research , 46 (4), 629 – 50.
Longoni Chiara , Bonezzi Andrea , Morewedge Carey K.. (2020), " Resistance to Medical Artificial Intelligence is an Attribute in a Compensatory Decision Process: Response to Pezzo and Beckstead ," Judgment and Decision Making , 15 (3), 446 – 48.
Lord Charles G. , Lepper Mark R. , Preston Elizabeth. (1984), " Considering the Opposite: A Corrective Strategy for Social Judgment ," Journal of Personality and Social Psychology , 47 (6), 1231 – 43.
Meehl Paul. (1954), Clinical Versus Statistical Prediction: A Theoretical Analysis and Review of the Literature. Minneapolis : University of Minnesota Press.
Moreau C. Page , Herd Kelly B.. (2009), " To Each His Own? How Comparisons with Others Influence Consumers' Evaluations of Their Self-Designed Products ," Journal of Consumer Research , 36 (5), 806 – 19.
Morris Michael W. , Menon Tanya , Ames Daniel R.. (2001), " Culturally Conferred Conception of Agency: A Key to Social Perception of Persona, Groups, and Other Actors ," Journal of Personality and Social Psychology Review , 5 (2), 169 – 82.
Musselweiler Thomas , Strack Fritz , Pfeiffer Tim. (2000), " Overcoming the Inevitable Anchoring Effect: Considering the Opposite Compensates for Selective Accessibility ," Personality and Social Psychology Bulletin , 26 (9), 1142 – 50.
Okada Erica M.. (2005), " Justification Effects on Consumer Choice of Hedonic and Utilitarian Goods ," Journal of Marketing Research , 42 (2), 43 – 53.
Ordabayeva Nailya , Chandon Pierre. (2016), " In the Eye of the Beholder: Visual Biases in Package and Portion Size Perceptions ," Appetite , 103 , 450 – 57.
Ross Lee , Nisbett Richard E.. (1991), The Person and the Situation: Perspectives of Social Psychology. New York : McGraw-Hill.
Sanders Nada R. , Manrodt Karl B.. (2003), " The Efficacy of Using Judgmental Versus Quantitative Forecasting Methods in Practice ," The International Journal of Management Science , 31 (6), 511 – 22.
Schweidel David , Bart Yakov , Inman Jeff , Stephen Andrew , Libai Barak , Andrews Michelle , et al. (2020), " In the Zone: How Technology Is Reshaping the Customer Journey ," working paper.
Spiller Stephen A , Fitzsimons Gavan J. , Lynch John G. Jr. , McClelland Gary H.. (2013), " Spotlights, Floodlights, and the Magic Number Zero: Simple Effects Tests in Moderated Regression ," Journal of Marketing Research , 50 (2), 277 – 88.
Timmermans Danielle. (1993), " The Impact of Task Complexity on Information Use in Multi-Attribute Decision Making ," Journal of Behavioral Decision Making , 6 (2), 95 – 111.
Venkatesan Rajkumar , Lecinski Jim. (2020), The AI Marketing Canvas: A Five Stage Roadmap to Implementing Artificial Intelligence in Marketing. Redwood City, CA : Stanford University Press.
Wardle Jane , Solomons Wendy. (1994), " Naughty but Nice: A Laboratory Study of Health Information and Food Preferences ," Health Psychology , 13 (2), 180 – 83.
Wason P.C. , Golding Evelyn. (1974), " The Language of Inconsistency ," British Journal of Psychology , 65 (4), 537 – 46.
Whitley Sara C. , Trudel Remi , Kurt Didem. (2018), " The Influence of Purchase Motivation on Perceived Preference Uniqueness and Assortment Size Choice ," Journal of Consumer Research , 45 (4), 710 – 24.
~~~~~~~~
By Chiara Longoni and Luca Cian
Reported by Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 8- Attribute Embedding: Learning Hierarchical Representations of Product Attributes from Consumer Reviews. By: Wang, Xin (Shane); He, Jiaxiu; Curry, David J.; Ryoo, Jun Hyun (Joseph). Journal of Marketing. Nov2021, p1. DOI: 10.1177/00222429211047822.
Ahead of Print- Database:
- Business Source Complete
Record: 9- Augmented Reality in Retail and Its Impact on Sales. By: Tan, Yong-Chin; Chandukala, Sandeep R.; Reddy, Srinivas K. Journal of Marketing. Jan2022, Vol. 86 Issue 1, p48-66. 19p. 1 Diagram, 8 Charts. DOI: 10.1177/0022242921995449.
- Database:
- Business Source Complete
Augmented Reality in Retail and Its Impact on Sales
The rise of augmented reality (AR) technology presents marketers with promising opportunities to engage customers and transform their brand experience. Although firms are keen to invest in AR, research documenting its tangible impact in real-world contexts is sparse. In this article, the authors outline four broad uses of the technology in retail settings. They then focus specifically on the use of AR to facilitate product evaluation prior to purchase and empirically investigate its impact on sales in online retail. Using data obtained from an international cosmetics retailer, they find that AR usage on the retailer's mobile app is associated with higher sales for brands that are less popular, products with narrower appeal, and products that are more expensive. In addition, the effect of AR is stronger for customers who are new to the online channel or product category, suggesting that the sales increase is coming from online channel adoption and category expansion. These findings provide converging evidence that AR is most effective when product-related uncertainty is high, demonstrating the technology's potential to increase sales by reducing uncertainty and instilling purchase confidence. To encourage more impactful research in this area, the authors conclude with a research agenda for AR in marketing.
Keywords: augmented reality; mobile app; online retail; product uncertainty; virtual product experience
"At some point, we're going to look back and think, how did
we not have a digital layer on the physical world?" – Greg Jones, Director of VR and AR at Google
Augmented reality (AR) is a technology that superimposes virtual objects onto a live view of physical environments, helping users visualize how these objects would fit into their physical world. Even though AR is in its early stages of growth, leaders in the field such as Apple's CEO, Tim Cook, and Google's Director of Virtual Reality (VR) and AR, Greg Jones, have lauded its potential to transform the retail experience ([ 3]; [23]). With the launch of AR toolkits by technology giants Apple and Google, it is now easier for companies to develop their own AR-enabled mobile apps. Jumping on the bandwagon, Facebook recently introduced AR-enabled display advertisements for their News Feed ([11]), making the technology even more accessible to companies.
Augmented reality (AR) is a technology that superimposes virtual objects onto a live view of physical environments, helping users visualize how these objects would fit into their physical world. Even though AR is in its early stages of growth, leaders in the field such as Apple's CEO, Tim Cook, and Google's Director of Virtual Reality (VR) and AR, Greg Jones, have lauded its potential to transform the retail experience ([ 3]; [23]). With the launch of AR toolkits by technology giants Apple and Google, it is now easier for companies to develop their own AR-enabled mobile apps. Jumping on the bandwagon, Facebook recently introduced AR-enabled display advertisements for their News Feed ([11]), making the technology even more accessible to companies.
From a retail perspective, a promising application of AR is to facilitate product evaluation by letting customers experience products virtually prior to purchase. Although research has emphasized the importance of direct product experiences to help customers learn about product benefits and assess product fit ([ 6]; [12]), offering direct product experiences can be a logistical challenge, especially in online retail. The introduction of AR has made it possible for shoppers to experience products virtually in the absence of physical products, managing their expectations and instilling purchase confidence ([44]). For example, Amazon and IKEA are using this technology to help customers determine if products or furniture pieces offered online are compatible with their existing room décor, and L'Oréal and Sephora are using AR to show customers how different cosmetic products would alter their appearance. Some of these applications are illustrated in Web Appendix A.
Despite the keen interest in AR, there has been limited research demonstrating its tangible impact in real-world contexts. Understanding the potential for AR to increase revenues is important for justifying investments in this new technology. However, the impact of AR on actual product sales is still ambiguous. By helping customers visualize products in their consumption contexts, AR could reduce product fit uncertainty, resulting in more sales. Conversely, AR may also discourage purchases if it leads to perceptions that the products may not fit well. As the technology is unable to convey experiential product attributes that could be important in purchase decisions (e.g., product texture or scent), the impact of AR on sales could also be insignificant. This uncertainty surrounding the impact of AR has been cited as one of the main reasons why companies are still hesitant to embrace the technology, even though most recognize the exciting opportunities it offers ([ 5]). Echoing this lack of clarity, a recent CNN article regarding applications of AR in the cosmetics industry expressed that "virtual lipsticks and smokey eye shadows are popular in apps, but are they translating into more makeup sales? Hard data isn't easy to come by" ([38]).
Furthermore, whether and how the impact of AR varies across different products or customer segments is also unclear. Having a more nuanced understanding of how AR affects sales would help marketing managers determine when it would be most appropriate to deploy the technology. Conceivably, if AR increases sales by reducing uncertainty, its impact may depend on product and customer characteristics that influence uncertainty in purchase decisions, such as brand popularity, product appeal, and customers' familiarity with the retail channel or category. Accordingly, the present research adopts the retailers' perspective to examine the following questions:
- How does the use of AR to facilitate product evaluation impact product sales?
- How does the sales impact of AR usage differ across product characteristics, such as brand popularity, product appeal, rating, and price?
- How do customers' prior experiences with the online channel and product category influence the sales impact of AR usage?
Given that AR is predominantly available on mobile apps ([43]), we focus on the mobile app platform for our analyses. We obtained data from an international cosmetics retailer that incorporated AR into its mobile app to help customers realistically visualize how they would look when they are using different cosmetic products (e.g., eyeshadows, lipsticks). The data contain sales records for 2,300 products, as well as browsing and purchase histories for 160,400 customers, allowing us to investigate how the sales impact of AR varies by product and customer characteristics. In addition, introduction of the AR feature for two product categories during the observation period provided us with a quasi-experimental setting to examine the impact of AR introduction on category sales.
Findings from our research provide preliminary evidence that AR usage has a positive impact on product sales. The overall impact appears to be small, but certain products are more likely to benefit from the technology than others. In particular, the impact of AR is stronger for brands that are less popular and products with narrower appeal, suggesting that AR could level the playing field for niche brands or products (sometimes referred to as products in the "long tail" of the product sales distribution; e.g., [ 8]). The increase in sales is also greater for products that are more expensive, indicating that AR could increase overall revenues for retailers. In addition, customers who are new to the online channel or product category are more likely to purchase after using AR, suggesting that AR has the potential to promote online channel adoption and category expansion. These findings provide converging evidence that AR is most effective when product-related uncertainty is high, implying that uncertainty reduction could be a possible mechanism by which AR could improve sales.
This article is one of the first to empirically demonstrate the impact of AR on sales and how it varies across product and customer characteristics using real-world data. In doing so, it extends prior studies on AR in the marketing field and represents an initial step in understanding what AR means for marketers and retailers. Beyond influencing sales, AR could transform the way brands reach out to and connect with customers at different stages of the customer journey. In the following section, we provide an overview of AR and elaborate on four ways the technology can be incorporated into brands' marketing strategies to reshape the customer retail experience. Then, we focus specifically on how the use of AR to facilitate product evaluation prior to purchase impacts sales in online retail. To encourage marketing academics to further engage in impactful and managerially relevant research in this area, we conclude with a research agenda that we developed in consultation with industry experts and marketing practitioners.
Augmented reality integrates virtual elements into real-world environments to create alternate perceptions of reality. Using sensors and object recognition capabilities from input devices such as cameras, AR technology scans the physical environment, identifies features in the environment, and superimposes virtual objects (e.g., two- or three-dimensional images or animations, text, sounds) on top of a live view of the real world. By blending virtual elements into physical environments in real time, AR enriches users' visual and auditory perceptions of reality. In most cases, the virtual elements are also responsive to movements or gestures, creating an interactive experience for users.
Although AR is often classified together with VR, the two technologies are distinct, both in how they function and the way they are experienced. Unlike AR, which receives input from the real world and adds virtual elements to it, VR immerses users in a completely digital and artificial environment, shutting them out from their surroundings. Due to the disorienting experience of being entirely isolated from the real world and the expensive headsets required ([19]), the appeal of VR has largely been limited to industries with products high in simulated content, such as gaming and entertainment ([13]). In contrast, AR allows users to experience virtual elements without the vulnerability of being blind to the real world. In addition, AR can be experienced directly from handheld devices that users already own (e.g., tablets or smartphones). Thus, AR is rapidly gaining prominence, and close to 100 million U.S. consumers are expected to use the technology regularly by 2022 ([43]).
The unique capabilities of AR present marketers with new opportunities to engage customers and transform the brand experience. Drawing on an extensive review of current applications of AR, we identified four broad uses of the technology in retail settings: to ( 1) entertain and ( 2) educate customers, help them ( 3) evaluate product fit, and ( 4) enhance the postpurchase consumption experience. These uses loosely correspond to customers' journey from awareness to interest, consideration, purchase, and consumption, and they may not be mutually exclusive. Next, we elaborate on these four uses and provide a summary with relevant examples in Table 1.[ 6]
Graph
Table 1. Uses of AR in Retail.
| Uses of AR | Role of AR | Illustrative Use Cases |
|---|
| Entertain customers | Create novel and engaging experiences for customers Build brand interest Drive foot traffic to physical stores
| Walmart collaborated with DC Comics and Marvel to bring exclusive superhero-themed AR experiences to selected outlets. Starbucks Reserve Roastery in Shanghai uses AR to offer customers a digital tour of their massive roasting facility.
|
| Educate customers | Deliver content and information in an interactive and visually appealing manner Help customers understand complex mechanisms and better appreciate the value of products
| Walgreen's and Lowe's use AR in their in-store navigation apps to guide users to product locations and notify them if there are special promotions along the way. Toyota and Hyundai use AR to demonstrate key features and innovative technologies in their new car models.
|
| Help customers evaluate product fit | Help customers visualize products in their actual consumption contexts Increase customers' confidence in their purchase decisions in the absence of physical products Accommodate wide product assortments and customization without the need for physical inventory
| IKEA's Place app uses AR to help customers determine if products fit with their existing room décor. L'Oréal's Virtual Try-On feature and Sephora's Virtual Artist app use AR to show customers how different cosmetic products would look on them. Uniqlo and Topshop use AR to offer a more convenient way of trying on different outfits in their physical stores. BMW and Audi use AR to give customers a preview of cars based on customizable features such as paint color, wheel design, and interior aesthetics.
|
| Enhance customers' postpurchase consumption experience | Offer new ways of enjoying products after they are purchased Deliver additional information while the products are being used or consumed
| LEGO's Hidden Side sets are specially designed to be played together with the companion AR app. McDonald's used AR to let customers discover the origins of ingredients in the food they purchased. Hyundai's Virtual Guide app uses AR to teach car owners how to perform basic maintenance.
|
1 Note: URL links to these examples are provided in Web Appendix B.
AR's ability to transform static objects into interactive and animated three-dimensional objects offers new ways for marketers to create fresh experiences to captivate and entertain customers. Besides generating hype and interest, marketers have also used AR-enabled experiences to drive traffic to their physical locations. For example, Walmart collaborated with media companies such as DC Comics and Marvel to bring exclusive superhero-themed AR experiences to their stores by placing special thematic displays in selected outlets. In addition to creating novel and engaging experiences for customers, it also encouraged them to explore different areas within the stores.
Due to its interactive and immersive format, AR is also an effective medium for delivering content and information to customers. For instance, to help customers better appreciate their new car models, Toyota and Hyundai have utilized AR to demonstrate key features and innovative technologies in a vivid and visually appealing manner. Retailers can also use AR to help customers navigate stores or highlight relevant product information to influence their in-store purchase decisions. Companies such as Walgreen's and Lowe's have developed in-store navigation apps that overlay directional signals onto a live view of the path in front of users to guide them to product locations and notify them if there are special promotions along the way.
By retaining the physical environment as a backdrop to virtual elements, AR also helps users visualize how products would appear in their actual consumption contexts, allowing them to more accurately assess product fit prior to purchase. For example, IKEA's Place app uses AR to give customers a preview of different furniture pieces in their homes by overlaying true-to-scale, three-dimensional models of products onto a live view of the room. Customers can easily determine if a product fits in a given space without the hassle of taking measurements. Fashion retailers Uniqlo and Topshop have also deployed the same technology in their physical stores, offering customers greater convenience by reducing the need for them to change in and out of different outfits. An added advantage of AR is its ability to accommodate a wide assortment of products. By replacing tangible product displays with lifelike virtual previews of products, retailers can overcome the constraints of physical space while still offering customers the opportunity to explore different product options. This capability is particularly useful for made-to-order or bulky products. Car manufacturers BMW and Audi have used AR to provide customers with true-to-scale, three-dimensional visual representations of car models based on customizable features such as paint color, wheel design, and interior aesthetics. These cases exemplify AR's huge potential to increase customers' confidence in their purchase decisions for a variety of products.
Lastly, AR can be used to enhance and redefine the way products are experienced or consumed after they have been purchased. For example, LEGO recently launched several brick sets that are specially designed to combine physical and virtual gameplay. Through the companion AR app, animated LEGO characters spring to life and interact with the physical LEGO sets, creating a whole new playing experience. In a bid to address skepticism about the quality of its food ingredients, McDonald's has also used AR to let customers discover the origins of ingredients in the food they purchased via storytelling and three-dimensional animations.
The present research focuses on the use of AR to help customers evaluate products prior to purchase. Specifically, we explore the possibility of leveraging AR to reduce product-related uncertainty in online purchase decisions. To extend prior research on AR in retail (summarized in Table 2), we use real-world data to examine how customers' use of AR to try products (for brevity, we refer to this as "AR usage" for the rest of the article) affects product and brand sales. In the following section, we present our conceptual framework and develop hypotheses for the impact of AR usage on sales.
Graph
Table 2. Selected Literature on AR in Retail.
| Article | Methodology | Context | Key Outcome Variables | Key Findings |
|---|
| Hilken et al. (2017) | Experimental | Using situated cognition theory to understand AR's potential to enhance online experiences | Value perceptions of online experiences; decision comfort; purchase and word-of-mouth intentions | The combination of simulated physical control and environmental embedding offered by AR creates a feeling of spatial presence. As a result, AR enhances consumers' perceptions of online experiences, decision comfort, and behavioral intentions.
|
| Yim, Chu, and Sauer (2017) | Experimental | Comparing AR versus web-based product presentations | Attitude toward AR and purchase intentions | Compared to web-based displays, AR is more immersive due to its interactive and vivid nature. As a result, AR is perceived to be more useful and enjoyable, leading to positive consumer attitudes and purchase intentions.
|
| Brengman, Willems, and Van Kerrebroeck (2019) | Experimental | Impact of AR on perceived ownership | Perceived ownership and purchase intentions | Compared to other touch and nontouch interfaces, mobile-enabled AR creates higher feelings of perceived ownership, positively affecting consumers' attitudes and purchase intentions.
|
| Heller et al. (2019a) | Experimental | Using mental imagery theory to understand how AR influences word of mouth | Word-of-mouth intentions | AR improves processing fluency by facilitating imagery generation and transformation, leading to higher consumer decision comfort and word-of-mouth intentions.
|
| Heller et al. (2019b) | Experimental | Comparing touch versus voice control modalities in multisensory AR | Decision comfort and willingness to pay | Touch control (vs. voice control) reduces mental intangibility, leading to higher consumer decision comfort and willingness to pay.
|
| Hilken et al. (2020) | Experimental | Shared decision making using social AR | Decision makers' product choice; spillover effects to recommenders | AR empowers recommenders by allowing them to take the point of view of the decision makers. AR stimulates recommenders' desire for products, leading to positive behavioral intentions.
|
| Current article | Instrumental variable estimation and quasi-experiment using real-world data | Examining the impact of AR usage on sales and the moderating impact of product and customer characteristics | Product and category sales | AR has a positive impact on sales for brands that are less popular, products with narrower appeal, and products that are more expensive. AR has a stronger impact for customers who are new to the retailer's online channel or product category.
|
Because customers cannot perfectly predict the consequences of their purchase decisions, uncertainty is inherent in market exchanges ([ 4]). However, it is especially pronounced in online environments due to the spatial separation between buyers and sellers as well as the temporal separation between payment and product fulfillment ([ 9]; [41]). Unlike in traditional retail settings, customers are unable to physically inspect or evaluate products before making a purchase, resulting in greater uncertainty that the products would be able to deliver the expected level of performance or benefits ([ 6]; [17]; [36]).
Researchers have broadly distinguished between two types of product uncertainty in online markets: product performance uncertainty and product fit uncertainty. Product performance uncertainty occurs when customers are unable to evaluate or predict product performance due to imperfect knowledge ([17]). In contrast, product fit uncertainty occurs when customers are unable to determine if the product matches their needs ([ 6]; [32]). The latter form of uncertainty is typically higher for products with experience attributes (i.e., attributes that can only be evaluated after the product has been experienced; [32]), such as apparel or beauty products.
Several mechanisms to reduce product performance uncertainty in online retail have been suggested. For example, retailers could lower information asymmetry by providing diagnostic product descriptions or by including credibility signals such as third-party product assurances, warranties, or customer reviews ([17]; [51]). In contrast, product fit uncertainty typically requires direct product experience to resolve, as it is idiosyncratic in nature and varies from individual to individual. Although some retailers have adopted try-before-you-buy programs (e.g., Warby Parker's home try-on program; [ 6]) or lenient product return policies ([24]; [53]) to provide opportunities for direct product experiences, these measures are notoriously costly for retailers due to the additional shipping and handling costs and risks of product damage ([22]). Furthermore, direct product experiences may not be viable or appropriate for certain products, such as products that are customized (e.g., engagement rings), products that require assembly (e.g., furniture), or personal care products (e.g., cosmetics).
The introduction of AR has made it possible to substitute direct product experiences with virtual product experiences to facilitate product evaluation and reduce product fit uncertainty. Using a situated cognition perspective, [29] propose that the value of AR lies in its ability to help customers visually integrate virtual products into the real-world environment (i.e., "environmental embedding") and use bodily movements and physical actions to control how products are presented (i.e., "simulated physical control"). The unique combination of these two properties induces perceptions that the virtual products are physically present in the real world, creating realistic product experiences. Consequently, customers are able to evaluate products as if they are actually interacting with the real products, resulting in reduced product fit uncertainty. In line with this, prior research finds that vivid images and greater control over the presentation of information are effective ways to alleviate uncertainty in online environments ([51]). By helping customers visualize products in their consumption contexts and reducing product fit uncertainty, AR-enabled product experiences increase the level of ease customers feel in the decision-making process, translating to positive behavioral intentions ([26]; [29]).
However, although AR communicates visual information about products, it is unable to convey other experiential product attributes (e.g., product texture, scent). For example, even though customers may use AR to visualize an IKEA sofa in a room, they are unable to assess how comfortable it is. Similarly, users trying on cosmetic products via AR are unable to evaluate other product attributes such as the texture and consistency of the product, which may affect ease of application and the way the product feels on the skin. According to [34], if customers do not perceive trial experiences as accurately representing actual consumption experiences, they may discount those trial experiences when they form judgments about the product. Thus, the extent to which virtual product experiences involving AR could influence online purchases is unclear. Nevertheless, as prior research has demonstrated the positive effects of providing fit information in online retail (e.g., [21]; [35]), we expect AR usage to have a positive impact on product sales because the technology could convey visual information that may reduce product fit uncertainty in online purchase decisions. Therefore, we predict the following:
- H1: AR usage has a positive impact on sales.
Building on the proposition that AR usage increases sales by reducing product fit uncertainty, we further hypothesize that AR would have a stronger impact when customers experience higher levels of uncertainty. In particular, the level of uncertainty experienced in a purchase decision could depend on product characteristics such as brand popularity, product appeal, and ratings. The level of uncertainty may also influence the price that customers are willing to pay for the product. Thus, the relationship between AR usage and sales may differ across these product characteristics. In addition, customers also vary in their need to reduce product fit uncertainty before making a purchase ([ 6]). This need to reduce uncertainty could depend on customers' familiarity with the online channel and product category. As a result, the impact of AR may also vary across these customer characteristics. Accordingly, we develop hypotheses for the moderating effects of product and customer characteristics in the following sections. Our conceptual framework is presented in Figure 1.
Graph: Figure 1. Conceptual framework.Note: Signs in parentheses represent the hypothesized effects.
Prior research has shown that consumers are more cautious when they purchase from lesser-known brands, as they anticipate feeling more regret if the product turns out to be inferior ([45]). Consistent with this, [18] find that cultures high in uncertainty avoidance place greater emphasis on brand credibility. In online environments, brand signals are even more important because consumers are not able to inspect products before purchasing ([16]). However, [31] demonstrates that when additional information is available to facilitate decision making, consumers rely less on brand signals. As a result, less-established brands benefit more from the increased availability of information. In the same vein, by communicating visual information to help customers assess product fit, AR may reduce uncertainty in online purchase decisions. Consequently, AR may decrease customers' reliance on brand signals and inadvertently increase preference for brands that are less popular. We use the term "popular" in a general sense to refer to brands that are more widely adopted. Therefore, we hypothesize the following:
- H2a: The impact of AR usage on sales is stronger for brands that are less popular.
Within the same category or brand, products may also have different levels of appeal due to the alignment between their inherent characteristics and general consumer preferences. For example, a red lipstick is more mainstream and has broader appeal than a blue lipstick. We draw a distinction between brand popularity and product appeal in that the latter depends on intrinsic properties of the product and could be independent of the brand. Thus, a red lipstick from an unknown brand could have broad appeal but low brand popularity, whereas a blue lipstick from a well-known brand could have limited appeal despite having high brand popularity. As products with broad appeal cater to the masses, they are more likely to match the needs of the general consumer. Conversely, because products with narrower appeal serve a niche segment, there is a higher probability that they will not match the preferences of the general consumer and will thus carry greater product fit uncertainty. Nevertheless, [ 8] demonstrate that in online contexts, search and discovery features such as search tools or recommendation engines can shift consumers' preferences to niche products by lowering the cost of acquiring product information. Consistent with this, [50] find that products with narrower appeal benefit more from greater information availability. By visually conveying product information to help customers assess product fit in an effortless and risk-free environment, AR could have a stronger impact for products with narrower appeal due to the higher product fit uncertainty associated with these products. Therefore, we hypothesize the following:
- H2b: The impact of AR usage on sales is stronger for products with narrower appeal.
Customers often turn to online ratings or reviews as a source of information to resolve uncertainty about product quality and fit ([14]). In line with this, [37] find that consumers from countries that are high in uncertainty avoidance are more sensitive to both the valence and volume of product ratings. However, as consumers tend to overrate direct experiences with products ([30]), the ability to evaluate products and resolve uncertainty via firsthand experiences with those products on AR platforms may reduce customers' reliance on online ratings. Thus, by enabling customers to learn about product benefits and assess product fit through their own virtual experiences, AR could diminish the role of online ratings in purchase decisions. As a result, customers may be more amenable to purchasing products despite their lower ratings if they are able to try these products using AR. Therefore, we predict the followsing:
- H2c: The impact of AR usage on sales is stronger for products with lower ratings.
When customers experience product uncertainty, they are not able to accurately assess the benefits the products offer. As a result, customers are more likely to undervalue the products and are less willing to pay a premium ([17]). Consistent with this, [36] find that customers who are familiar with online shopping are still hesitant to purchase expensive products through the internet when there is a high degree of product uncertainty because they could suffer greater financial losses if these products do not fit them well. By facilitating product evaluation prior to purchase, AR helps customers ascertain if products match their needs and preferences. Consequently, customers may experience less uncertainty and feel more comfortable purchasing products that are more expensive. In line with this, [27] find that AR usage improves decision comfort, leading to higher willingness to pay. Therefore, we predict the following:
- H2d: The impact of AR usage on sales is stronger for more expensive products.
According to [36], customers who are familiar with online shopping are more inclined to purchase products with a higher degree of uncertainty because their cumulative online shopping experiences help them develop the ability to assess products when limited information is available. Thus, customers who have purchased from a retailer's online channel in the past may feel more comfortable making subsequent online purchases despite experiencing product uncertainty, potentially making them less dependent on AR to make their purchase decisions. In contrast, customers who are new to the retailer's online channel (but have made prior purchases at the retailer's offline channel) are not accustomed to making purchases in the absence of actual products. As a result, they may experience greater product fit uncertainty and may be deterred from purchasing online due to the inability to assess product fit. Because AR simulates the in-store experience of trying products, it may help reduce product fit uncertainty for customers who are new to the online channel. These customers may derive greater value from the ability to evaluate products virtually, potentially making them more likely to purchase online after using AR. Therefore, we predict the following:
- H3a: The impact of AR usage on sales is stronger for customers who are new to the retailer's online channel.
Besides channel experience, customers' familiarity with the product category also affects their level of product fit uncertainty ([32]). Customers who are familiar with a product category can draw on their prior experiences as an information source to form judgments about products ([46]). As a result, they may rely less on AR in their purchase decisions. Conversely, customers who are unfamiliar with a product category lack the necessary category knowledge to evaluate product attributes and, at the same time, may not be aware of their own preferences ([32]). Consequently, these customers will have more difficulty assessing whether a product's attributes match their preferences, resulting in greater product fit uncertainty. By helping customers visualize how products would appear in their actual consumption contexts, AR could reduce product fit uncertainty and increase purchase confidence for customers who are new to the product category. As a result, AR usage may have a stronger impact on the purchase decisions for these customers. Therefore, we predict the following:
- H3b: The impact of AR usage on sales is stronger for customers who are new to the product category.
To summarize, we propose that AR usage will positively impact sales by reducing product uncertainty. Following this line of reasoning, we developed several predictions about which products would be more likely to benefit from AR and which customers would be more likely to respond to AR.
As AR is predominantly available through mobile apps ([43]), we focus our analyses on the mobile app platform. To test our hypotheses, we obtained data from an international cosmetics retailer with both an online and offline presence. Leveraging AR technology, the retailer integrated a new feature on its existing mobile app that allowed customers to virtually try on makeup products (e.g., eyeshadows, lipsticks). The AR technology detected customers' facial features via their smartphone cameras and superimposed the shade of chosen products onto a live view of their face in real time, giving them a realistic visual representation of their appearance when they are using the products. The brand, product name, and price were displayed at the top of the screen. Figure A3 in Web Appendix A provides a visual example of a customer trying on a lipstick using the AR feature. For comparison, the corresponding product detail page view (i.e., the conventional way of conveying product-related information on mobile retail apps) is also provided. Prior to the start of our observation period in December 2017, the AR feature was only available for lip categories (i.e., lipstick and lip gloss) and was subsequently introduced for eye categories (i.e., eyeshadow and eyeliner) in March 2018. Figure A4 in Web Appendix A provides a visual overview of AR availability for the different categories.
We obtained two separate data sets from the retailer for one of its key markets in Asia Pacific. The first data set contained information about browsing activities on the mobile app, including specific products customers tried using the AR feature, and covered a 19-month period from December 2017 to June 2019. The second data set contained transaction records from June 2017 to June 2019 for all retail channels, including mobile app, website, and offline stores. We merged the two using customers' loyalty card number, which allowed us to match AR usage and product purchases at a disaggregate level.
During the 19-month period, a total of 160,407 shoppers browsed products from the lip and eye categories across 806,029 sessions, 20.8% of which involved AR usage. Customers who used AR during the session spent 20.7% more time browsing (Mwith_AR = 16.6 minutes, Mwithout_AR = 13.8 minutes, p <.01) and browsed 1.28 times more products (Mwith_AR = 53.9, Mwithout_AR = 42.2, p <.01). The purchase rate for sessions with AR usage was 19.8% higher than for sessions without AR usage (3.15% with AR vs. 2.63% without AR, p <.01), providing preliminary indication of the positive impact of AR on sales.
We divide our analyses into three sections. In the first section, we perform the analysis at the product level to examine the moderating effects of brand popularity, product appeal, rating, and price. To minimize selection bias arising from availability of the AR feature, we focus on lipsticks and lip glosses, as the feature was available for > 96% of products in each of these categories. In the second section, we take advantage of the introduction of AR for two eye categories (i.e., eyeshadow and eyeliner) to examine the effect of AR introduction on category sales using a quasi-experimental differences-in-differences-in-differences (DDD) approach. Finally, we investigate how the impact of AR varies at the customer level. As all customers had no knowledge that the AR feature would be introduced for the two eye categories prior to the introduction, the event provided us with a clean setting for examining how customers' channel and category experience (prior to the introduction) would moderate the impact of AR usage on purchase probability.
As product color is an important factor in cosmetic purchases, we considered each shade/color of retail merchandise as a unique product. In total, we had 2,334 products in the lipstick and lip gloss categories (1,984 products across 41 brands for lipstick; 350 products across 28 brands for lip gloss). Our empirical strategy was to relate the number of customers using AR to try each product during a particular time period with sales volume for that product during the same time period. We estimated the model at the monthly product level, giving us a total of 44,346 observations (2,334 products × 19 months from December 2017 to June 2019). As one of our objectives was to examine the moderating effect of product ratings, we included products with a rating in the main analysis and replicated the analysis for all products as a robustness check. Our final sample for the main analysis consisted of 29,345 observations.
For each product i, we modeled how the volume of AR usage in month t, AR Usageit, influenced the number of products sold in month t, Product Salesit. As Product Salesit was a count variable with significant over-dispersion (M =.46, SD = 1.73) and over 80% of observations were 0, we used a zero-inflated negative binomial model for the estimation. The vector of covariates in the regression is given by the following equation:
Graph
1
In Equation 1, we measured AR Usageit as the number of customers using AR to try product i during month t. As brands that are more widely adopted should have higher sales, and because the web and app channels are both online and carry identical products, we used total brand sales (within the category) from the web channel during the same period as a proxy for brand popularity, Brand Popularityit. Following prior research using product sales as an indicator of mass or niche appeal (e.g., [ 8]), we used total product sales from the web channel during the same period to reflect product i's breadth of appeal, Appealit. Ratingit and Priceit were the rating and price of product i at time t, respectively. To examine how the impact of AR was influenced by brand popularity, product appeal, rating, and price, we included the corresponding interactions in the model. In addition, we included Categoryi (1 = lipstick, 0 = lip gloss) and a series of dummy variables, Montht, (for t = 1,..., T months) to control for category and month effects. Table 3 provides a summary of how the variables were operationalized and their descriptive statistics, and we provide their correlations in Web Appendix C. All the correlations were low, and the variance inflation factors were below 1.62, indicating that multicollinearity was not an issue. To prevent overestimation of effects due to the panel structure of the data, we clustered standard errors at the product level (e.g., [49]).
Graph
Table 3. Variable Operationalization and Descriptive Statistics for Product Model.
| Variable | Operationalization | Mean | SD | Min | Median | Max |
|---|
| Product Sales | Total product sales from mobile app (in units) | .46 | 1.73 | .00 | .00 | 64.00 |
| AR Usage | Number of customers using AR to try the product in focal country | 13.90 | 22.67 | .00 | 6.00 | 611.00 |
| AR UsageAlt | Number of times the product was tried using AR in focal country | 14.45 | 23.92 | .00 | 7.00 | 620.00 |
| Brand Popularity | Total brand sales from website (in thousands of units) | .04 | .07 | .00 | .02 | .51 |
| Brand PopularityAlt | Number of customers buying the brand from website (in thousands) | .03 | .04 | .00 | .01 | .42 |
| Appeal | Total product sales from website (in units) | .25 | .97 | .00 | .00 | 32.00 |
| AppealAlt | Number of customers buying the product from website | .25 | .95 | .00 | .00 | 31.00 |
| Rating | Product rating at time t (on a five-point scale) | 4.14 | .67 | .50 | 4.25 | 5.00 |
| Price | Product price at time t | 31.80 | 11.17 | 5.00 | 32.00 | 77.00 |
| AR UsageCountry_A | Number of customers using AR to try the product in Country A | .59 | 1.43 | .00 | .00 | 39.00 |
| AR UsageCountry_B | Number of customers using AR to try the product in Country B | .28 | .71 | .00 | .00 | 11.00 |
Our objective was to understand how the volume of AR usage for product i during month t, AR Usageit, influenced product sales, Product Salesit. However, AR Usageit could be endogenous, as customers may have been more inclined to use AR to try products they were already interested in purchasing. To account for this endogeneity, we used the two-stage residual inclusion method ([48]), which has been used in recent research when both the endogenous and dependent variables are nonlinear (e.g., [ 2]; [15]).
Following the two-stage residual inclusion method, we first regressed the endogenous variable, AR Usageit, on all other covariates in Equation 1. Residuals from this first stage were then included to estimate Product Salesit. Similar to the control function approach ([42]), the included residuals controlled for the portion of the endogenous variable that would otherwise correlate with the error term in Equation 1. According to [48], we needed to include instruments in the first stage estimation to resolve the identification problem in Equation 1. These instruments should ( 1) be strongly related to the endogenous variable and ( 2) not be correlated with the error term in Equation 1. In other words, the instruments should only have an indirect relationship with the outcome variable, Product Salesit, through their association with the endogenous variable, AR Usageit. As realizations of the same variable from different markets can serve as suitable instruments ([40], p. 601), we used the volume of AR usage in two other countries for the same product during the same month as our instruments (i.e., AR UsageitCountry_A and AR UsageitCountry_B, respectively). Underlying this choice of instruments is the assumption that customer preferences are similar across markets and that product-specific factors affecting customers' interest in trying products using the AR feature should be constant in all markets, satisfying the first condition. However, the number of customers using the AR feature to try products in other markets should have no bearing on customers' purchase decisions in the focal market, satisfying the second requirement. We also used lagged values of AR Usageit as an alternative instrument (e.g., [15]) and discuss this further in the robustness analyses section.
Because AR Usageit is a count variable with significant over-dispersion (M = 13.9, SD = 22.7), we used a negative binomial model for the first stage estimation. As predicted residuals from the first stage were used in the estimation of Equation 1, standard errors needed to be corrected to account for this additional source of variation ([42]). We implemented the cluster bootstrapping method ([10], p.327) to approximate the correct standard errors using 1000 bootstrap samples.
From the first stage estimation (provided in Web Appendix D), coefficients for the instruments are positive and significant (.414 for AR UsageCountry_A and.301 for AR UsageCountry_B, p <.01 for both). Furthermore, the instruments are highly correlated with AR Usage (.75 for AR UsageCountry_A and.64 for AR UsageCountry_B, p <.01 for both), and the F-statistic of excluded instruments in the first stage regression is 5,520, which is well above the recommended cutoff of 10 ([ 1]). These results indicate that the instruments are strongly related with the endogenous variable. To assess validity of the instruments, we performed the Hansen J-test for over-identifying restrictions. Results from the test fail to reject the null hypothesis that the instruments are uncorrelated with the second stage error term (χ2 ( 1) =.699, p =.40), providing additional support for the choice of instruments.
To examine the main effect of AR usage in H1, we estimated the second stage model without interaction terms. Results for this model are presented in Table 4, Column 1. The coefficient for AR Usage is significantly positive (.006, p <.01), suggesting a small but positive relationship between the number of customers using AR to try the product and sales for that product during the same month. Thus, H1 is supported. The coefficients for other variables are largely in line with common intuition. For example, brand popularity (.894, p <.05), breadth of product appeal (.385, p <.01), and product rating (.094, p <.05) are positively associated with product sales, whereas price (−.005, p <.10) has a negative relationship with product sales. The coefficient for the residual correction term, which is equivalent to the Hausman test for the presence of endogeneity ([40]), is significant (.071, p <.01), indicating that the endogeneity-corrected estimates are preferred. Thus, we focus on results from the two-stage model and provide results for the uncorrected model in Web Appendix D.
Graph
Table 4. Product Model: Impact of AR Usage on Product Sales and Moderating Effects of Product Characteristics.
| Column 1Second Stage(WithoutInteractions) | Column 2Second Stage(Full Model) | Column 3Alternative Instrument for AR Usage | Column 4Including Products Without Ratings |
|---|
| AR Usage (H1) | .006 (.001) *** | −.002 (.006) | −.003 (.007) | .007 (.006) |
| Brand Popularity | .894 (.364) ** | 1.482 (.356) *** | 1.796 (.396) *** | 1.675 (.333) *** |
| Appeal | .385 (.029) *** | .416 (.023) *** | .419 (.027) *** | .473 (.023) *** |
| Rating | .094 (.042) ** | .052 (.054) | .062 (.056) | – |
| Price | −.005 (.003) * | −.009 (.004) ** | −.008 (.004) ** | −.010 (.003) *** |
| AR Usage × Brand Popularity (H2a) | – | −.022 (.009) ** | −.025 (.008) *** | −.028 (.010) *** |
| AR Usage × Appeal (H2b) | – | −.001 (.000) *** | −.001 (.000) ** | −.001 (.000) *** |
| AR Usage × Rating (H2c) | – | .001 (.001) | .001 (.001) | – |
| AR Usage × Price (H2d) | – | .000 (.000) * | .000 (.000) * | .000 (.000) * |
| Correction term | .071 (.027) *** | .050 (.027) * | .102 (.053) * | .063 (.026) ** |
| Constant | −1.114 (.223) *** | −.913 (.264) *** | −1.185 (.280) *** | −.687 (.162) *** |
| Category dummy | Included | Included | Included | Included |
| Month dummies | Included | Included | Included | Included |
| Observations | 29,345 | 29,345 | 28,305 | 44,346 |
| Log likelihood | −18,573 | −18,517 | −17,630 | −25,310 |
- 2 *p ≤.10; **p ≤.05; ***p ≤.01.
- 3 Note: Standard errors (clustered at product level) are in parentheses.
Results for the full second stage model are presented in Table 4, Column 2. In support of H2a and H2b, the interactions between AR Usage and Brand Popularity (−.022, p <.05) and Appeal (−.001, p <.01) are significantly negative, indicating that the sales impact of AR usage is stronger for brands that are less popular and products with narrower appeal. The interaction between AR Usage and Price is significantly positive (.000, p <.10), suggesting that the sales impact of AR usage is stronger for products that are more expensive. Thus, H2d is also supported. However, the results do not provide support for H2c, as the interaction between AR Usage and Rating is not significant (.001, p >.10).
We performed several analyses to ensure that our findings are robust to different assumptions and model specifications. First, following prior research, which has used lagged values of endogenous variables as instruments (e.g., [15]), we used the volume of AR usage for product i in the past one month as an alternative instrument. As app activity data prior to the first month (i.e., December 2017) was unavailable, we excluded observations for the first month. Results for this model are presented in Table 4, Column 3, and the findings are consistent. Because we were interested in the moderating effect of ratings, we focused on products that had a rating in the main analysis. Since the coefficient for Rating was not significant, we excluded it in the model specification and replicated the analysis for all products. Results for this model are also consistent with the main findings and are presented in Table 4, Column 4.
We also explored alternative operationalizations for AR Usage, Brand Popularity, and Appeal. Instead of operationalizing AR Usage as the number of customers using AR to try product i, we used the number of times product i was tried using AR to account for repeated AR usage from the same customer. We also operationalized Brand Popularity and Appeal as the number of customers purchasing the brand and product, respectively. Results for these models are reported in Web Appendix E. Across all robustness analyses, results are generally consistent with the main model, providing further validation for our findings.
The introduction of AR for two eye categories (i.e., eyeshadow and eyeliner) in mid-March 2018 presented a unique opportunity to examine the impact of AR introduction on sales. Using a quasi-experimental approach, we regarded AR introduction as a treatment and examined its impact by comparing differences in sales for products with and without the AR feature, before and after the feature was introduced. Because the AR feature was only available for eyeshadows and eyeliners, a potential comparison could be between these categories and other eye categories that did not have the feature (i.e., eyebrows, mascaras, and eye palettes). This between-category comparison relies on the crucial assumption that sales trends across different eye categories would be parallel in the absence of AR introduction. As cosmetic products are often used concurrently, sales for products targeting the same facial feature should generally move in the same direction. Since the AR feature was only available on the mobile app, an alternative comparison could be between the app and web channels. This approach avoids the assumption that trends across different eye categories are similar, but it requires a separate assumption that without AR introduction, sales trends in the two online channels would be parallel.
A more robust approach that does not require either of these assumptions is the DDD approach ([ 1], p. 181; [54], p.150), which combines both comparisons. Specifically, the DDD analysis measures differences between app and web sales for eyeshadows and eyeliners before and after AR introduction, relative to the same differences for other eye categories that do not have the AR feature. Thus, the DDD approach controls for both channel and category trends that could potentially confound the effect, and it relies on the more relaxed assumption that in the absence of AR introduction, sales trends in the two online channels would be parallel for products in the same category. Following [33] and [20], we conducted two falsification tests using data from the pre–AR introduction period, and the results provide support that this assumption holds in our study. Details and results for these falsification tests are included in Web Appendix F.
Accordingly, we examined changes in weekly sales for the five product categories (i.e., eyeshadow, eyeliner, eyebrows, mascara, and eye palettes) across two channels (i.e., app and web) before and after AR introduction. Our sample covers a duration of 108 weeks (i.e., 42 weeks for the pre–AR introduction period and 66 weeks for the post–AR introduction period), giving us a total of 1,080 observations (5 × 2 × 108 = 1,080).
The outcome variable of interest was sales for category j on channel k during week t, Category Salesjkt. As Category Salesjkt was a count variable with significant overdispersion (M = 64.8, SD = 78.0), we used a negative binomial model for the estimation. Following [54], p.150), the vector of covariates in the regression is given by the following:
Graph
2
In Equation 2, AR Introt is a dummy variable with a value of 1 if week t was in the post–AR introduction period and 0 otherwise. Appk is a dummy variable with a value of 1 for the mobile app and 0 for the website, and AR Featurej is a dummy variable with a value of 1 for eye categories with the AR feature (i.e., eyeshadow and eyeliner) and 0 for other eye categories. The key coefficient of interest was β4, which captured the three-way interaction between AR introduction, retail channel, and categories that have the AR feature. Thus, β4 represents the additional change in mobile app sales post–AR introduction for eyeshadow and eyeliner, after accounting for channel and category-related changes over the same period (captured by β5 and β6 respectively). We included all lower-order interactions in the model, as well as a series of dummy variables, Categoryj (for j = 1,..., J categories) and Weekt, (for t = 1,..., T weeks) to control for category and week effects. Because Categoryj was perfectly collinear with AR Featurej and Weekt was perfectly collinear with AR Introt, we excluded dummy variables for an additional category and week. To account for the panel nature of the data, we clustered standard errors at the category-channel level, allowing errors for observations from the same category within each channel to correlate.
Before discussing results for the DDD analysis, we present the basic pre-post model in Table 5, Column 1. We regressed weekly mobile app sales for eyeshadow and eyeliner on AR Introt and the vector of dummies. The coefficient for AR Introt is significantly positive (.611, p <.05), providing preliminary evidence that sales increased after AR was introduced. Results for the DDD analysis are presented in Table 5, Column 2. The coefficient for the three-way interaction between AR introduction, app, and categories with the AR feature is marginally significant (.449, p <.10), providing some evidence that sales for eyeshadows and eyeliners increased on the app channel after AR was introduced.
Graph
Table 5. DDD Analysis: Impact of AR Introduction on Category Sales.
| Column 1Basic Pre-PostModel | Column 2DDD Analysis | Column 3Including Trends | Column 4Excluding Sale Events |
|---|
| AR Intro | .611 (.245) ** | .265 (.214) | .771 (.445) * | .720 (.429) * |
| App | – | .125 (.058) ** | 39.187 (9.72) *** | 31.953 (8.95) *** |
| AR Feature | – | .068 (.146) | 5.794 (8.44) | 3.720 (7.57) |
| AR Intro × App × AR Feature | – | .449 (.262) * | .441 (.249) * | .465 (.264) * |
| AR Intro × App | – | −.601 (.237) ** | .243 (.154) | .036 (.169) |
| AR Intro × AR Feature | – | −.155 (.177) | −.066 (.209) | −.113 (.200) |
| App × AR Feature | – | −.034 (.088) | −.027 (.056) | −.036 (.048) |
| Constant | 2.043 (.040) *** | 2.604 (.156) *** | 2.403 (.226) *** | 2.464 (.201) *** |
| Category dummies | Included | Included | Included | Included |
| Week dummies | Included | Included | Included | Included |
| Category trend | Not included | Not included | Included | Included |
| Channel trend | Not included | Not included | Included | Included |
| Observations | 216 | 1,080 | 1,080 | 940 |
| Log likelihood | −897 | −5,143 | −5,095 | −4,179 |
- 4 *p ≤.10; **p ≤.05; ***p ≤.01.
- 5 Note: Standard errors (clustered at category-channel level) are in parentheses.
To check the DDD identification strategy, we included channel and category trends in Equation 2 ([ 1], p. 178). Results of this alternative model are presented in Table 5, Column 3, and the coefficient of the three-way interaction of interest is similar in direction, magnitude, and significance with the main model. Although the weekly fixed effects controlled for variations in overall sales between weeks, they did not account for time-varying confounding effects that were specific to the channel-category. Thus, if there were more app-exclusive sale events for the eyeshadow and eyeliner categories in the post–AR introduction period relative to the pre–AR introduction period, the effect of AR Introduction on app sales in these two categories would be overstated. As a robustness check, we removed weeks that coincided with sale events from the analysis and present the results in Table 5, Column 4. We also split AR Featurej into the two eye categories with the AR feature, Eyeshadowj and Eyelinerj, and the coefficients of both three-way interactions are marginally significant, providing convergent validity for our results. Furthermore, results from the Wald test for equality of coefficients fail to reject the null hypothesis that the coefficients are the same (p =.76), indicating that the effect of AR introduction on sales is not category-specific. Lastly, we estimated the same model using a Poisson regression. Results of these additional analyses are provided in Web Appendix G. Across all robustness analyses, the direction, magnitude, and significance of coefficients are similar to the main model.
Overall, the results provide additional support for H1 and demonstrate that the positive impact of AR generalizes to other product categories. We note that because the retailer did not announce the introduction of AR for the eye categories, usage of the feature was low. On average, the weekly number of customers using AR to try products from the eye categories was 6.4 times lower than the number for lip categories (Meyes = 271.14 vs. Mlips = 1,737.00). Thus, our result is a conservative estimate of the impact of AR introduction, and we speculate that the effect could have been larger if the retailer had advertised the feature. To establish a direct relationship between AR usage and purchase, and to further examine the moderating effects of customers' channel and category experience, we next turn our attention to the customer level.
We focused on the sample of active customers (i.e., those who made a purchase in the past one year) who browsed products in the eyeshadow or eyeliner categories during the 12-month period[ 7] after the retailer introduced AR for these two categories (i.e., mid-March 2018 to mid-March 2019). In total, our sample included 42,493 customers. At the time of AR introduction, 40.2% of these customers had never purchased online before (i.e., new to the online channel) and 43.4% had never purchased eyeshadow or eyeliner before (i.e., new to the categories). During the 12-month period after the retailer introduced the AR feature for the two categories, 13.9% of customers used the feature to try eyeshadows and eyeliners, and 15.0% purchased at least one product from these categories using the app. Accordingly, we modeled how AR usage influenced customers' probability of purchasing products from these two categories in the focal period.
The dependent variable of interest, P(Purchaseieyes), was customer i's probability of purchasing at least one eyeshadow or eyeliner on the app within 12 months of AR introduction for the two categories. As the dependent variable was binary, we used a probit model with the following specification:
Graph
3
In Equation 3, Φ denotes the standard probit link function. AR Usageieyes represents the focal independent variable and takes a value of 1 if customer i used the AR feature to try eyeshadows or eyeliners during the period and 0 otherwise. New Channeli and New Categoryi are both indicator variables representing customers' (lack of) prior experience with the channel and category. New Channeli takes a value of 1 if customer i is new to the online channel and 0 otherwise, and New Categoryi takes a value of 1 if customer i is new to the two eye categories and 0 otherwise. To examine how these two variables moderate the effect of AR usage on purchase, we included interactions between the variables and AR Usageieyes. We also included a vector, Browsingi, to control for customers' browsing behavior before and during the focal period to account for customer interest and engagement. Because the browsing activity data set starts at December 2017 (i.e., three months prior to the introduction of AR for the eye categories), we used a three-month window for past browsing behavior. Lastly, we included a vector, Past Purchasei, to control for customers' purchase history in the 12 months prior to AR introduction for the eye categories to account for customer loyalty. Table 6 provides a summary of how the variables are operationalized and their descriptive statistics. The correlations are provided in Web Appendix H, and the variance inflation factors are below 1.75, indicating that multicollinearity is not an issue.
Graph
Table 6. Variable Operationalization and Descriptive Statistics for Customer Model.
| Variable | Operationalization | Mean | SD | Min | Median | Max |
|---|
| Purchaseeyes (1/0) | 1 if customer bought eye products in the focal period, 0 otherwise | .15 | .36 | .00 | .00 | 1.00 |
| AR Usageeyes (1/0) | 1 if AR was used to try eye products in the focal period, 0 otherwise | .14 | .35 | .00 | .00 | 1.00 |
| New Channel (1/0) | 1 if customer had never purchased from the retailer's online channel before the focal period, 0 otherwise | .40 | .49 | .00 | .00 | 1.00 |
| New Category (1/0) | 1 if customer had never purchased eye products from the retailer before the focal period, 0 otherwise | .43 | .50 | .00 | .00 | 1.00 |
| Browsing controls | | | | | | |
| Past Duration | Total browsing duration in the past three months (in minutes) | 11.56 | 25.60 | .00 | .00 | 150.18 |
| Past Pageseyes | Number of eye product pages viewed in the past three months | .77 | 2.90 | .00 | .00 | 23.00 |
| Past AR Usagelips (1/0) | 1 if customer had used AR for lip products in the past three months, 0 otherwise | .02 | .15 | .00 | .00 | 1.00 |
| Duration | Total browsing duration in the focal period (in minutes) | 153.17 | 148.32 | 2.82 | 106.57 | 1054.25 |
| Pageseyes | Number of eye product pages viewed in the focal period | 19.21 | 24.98 | 1.00 | 9.00 | 128.00 |
| Purchase controls | | | | | | |
| Past Order Number | Number of transactions in the past one year | 6.92 | 5.83 | 1.00 | 5.00 | 37.00 |
| Past Order Value | Average value of transactions in the past one year | 81.88 | 50.49 | .00 | 70.68 | 547.00 |
| Past Eye Purchases | Number of eye products purchased in the past one year | .89 | 1.50 | .00 | .00 | 11.00 |
| Recent Order | Number of months from the most recent transaction | 2.64 | 2.49 | .03 | 2.03 | 12.13 |
| First Order | Number of months from the first transaction | 18.92 | 7.73 | .03 | 22.37 | 26.80 |
6 Notes: "Eye products" refers to eyeshadow and eyeliner, the two categories of interest. The focal period is 12 months after the retailer introduced AR for eye products (i.e., March 15, 2018, to March 15, 2019).
As customers who already intend to purchase products may be more likely to try them using the AR feature, we used the two-stage residual inclusion method to account for this self-selection bias. We used customers' past AR usage for lip products (prior to AR introduction for the eye categories) as the instrument. Customers who had used AR to try lip products in the past were already aware of the feature and could have been more likely to use it again to try eye products. Conversely, customers who had never used the feature to try lip products in the past may have been unaware of it. Because the retailer did not announce the AR introduction for the eye categories, these customers could still have been unaware of the feature. As a result, they would have been less likely to use it to try eye products. Furthermore, because lip and eye products target different areas of the face, past usage of the AR feature to try lip products should not have directly affected the probability of purchasing eye products during the focal period. Thus, we included Past AR Usageilips as an instrument in the first stage to estimate customer i's likelihood of using AR in the focal period. The variable was coded as 1 if customer i used the AR feature to try lip products in the three months before AR was introduced for the eye categories and 0 otherwise. Residuals from the first-stage estimation were then included in Equation 3 to estimate P(Purchaseieyes). Similar to the product model, we bootstrapped 1,000 samples to obtain the proper standard errors. To examine if the findings are robust to alternative identification strategies, we also adopted the propensity score weighting approach, which does not rely on instruments. We discuss this further in the robustness analyses section.
The coefficient of the instrument in the first stage estimation (provided in Web Appendix I) is positive and significant (.176, p <.01), and the F-statistic of excluded instrument in the first stage regression is 15.8, providing evidence for the strength of the instrument. However, the coefficient for the residual correction term, which is equivalent to the Hausman test for the presence of endogeneity ([40]), is not significant, suggesting that endogeneity may not be a concern. We also used the Heckman selection method ([25]) as an alternative identification strategy, and the inverse Mills ratio is similarly not significant. Therefore, we report estimates for the uncorrected model in the results section and provide the full result for both the two-stage residual inclusion and Heckman selection methods in Web Appendix I. We note that across all models, the substantive findings of interest remain consistent.
Table 7, Column 1, displays the results for the model without interactions, representing factors influencing the purchase of eyeshadows or eyeliners during the 12 months after AR introduction for these categories. The coefficient of AR Usageeyes is positive and significant (.046, p <.05), providing further evidence for H1. The coefficients of other variables are largely in line with expectations. For example, the coefficient for New Channel (−.329, p <.01) and New Category (−.120, p <.01) are significantly negative, indicating that customers who are new to the online channel or product category are less likely to make a purchase. The number of orders (.007, p <.01), average order value (.002, p <.01), and number of eye products purchased in the past (.080, p <.01) are positively related to probability of purchasing eye products. Furthermore, total browsing duration (.000, p <.01) and number of eye product pages viewed (.007, p <.01) are also positively related to the purchase of eye products.
Graph
Table 7. Customer Model: Impact of AR Usage on Probability of Purchase and Moderating Effects of Customer Characteristics.
| Column 112-Month Model (Without Interactions) | Column 212-Month Model(Full Model) | Column 36-Month Model(Full Model) | Column 43-MonthModel(Full Model) |
|---|
| AR Usageeyes (H1) | .046 (.022) ** | −.015 (.030) | .004 (.047) | −.036 (.084) |
| New Channel | −.329 (.018) *** | −.344 (.019) *** | −.323 (.029) *** | −.296 (.040) *** |
| New Category | −.121 (.020) *** | −.134 (.021) *** | −.060 (.032) * | −.122 (.042) *** |
| AR Usageeyes × New Channel (H3a) | – | .091 (.046) ** | .144 (.070) ** | .091 (.138) |
| AR Usageeyes × New Category (H3b) | – | .082 (.045) * | .075 (.068) | −.004 (.135) |
| Past Duration | −.002 (.000) *** | −.002 (.000) *** | −.002 (.000) *** | −.002 (.000) *** |
| Past Pageseyes | −.003 (.003) | −.003 (.003) | .000 (.002) | −.004 (.002) ** |
| Duration | .000 (.000) *** | .000 (.000) *** | .001 (.000) *** | .001 (.000) *** |
| Pageseyes | .007 (.000) *** | .007 (.000) *** | .008 (.000) *** | .017 (.001) *** |
| Past Order Number | .007 (.002) *** | .007 (.002) *** | .005 (.003) | .003 (.004) |
| Past Order Value | .002 (.000) *** | .002 (.000) *** | .002 (.000) *** | .002 (.000) *** |
| Past Eye Purchases | .080 (.006) *** | .081 (.006) *** | .103 (.012) *** | .062 (.011) *** |
| Recent Order | −.019 (.004) *** | −.020 (.004) *** | −.020 (.006) *** | −.016 (.008) * |
| First Order | −.001 (.001) | −.001 (.001) | −.002 (.002) | .005 (.002) * |
| Constant | −1.285 (.035) *** | −1.276 (.035) *** | −1.567 (.054) *** | −1.771 (.077) *** |
| Observations | 42,493 | 42,493 | 24,147 | 13,434 |
| Log likelihood | −16,614 | −16,609 | −7,423 | −3,773 |
- 7 *p ≤.10; **p ≤.05; ***p ≤.01
- 8 Note: Standard errors are in parentheses.
Table 7, Column 2, provides results for the 12-month model, including interactions. The interaction between AR Usageeyes and New Channel is positive and significant (.091, p <.05), suggesting that AR has a stronger effect among customers who had never purchased online in the past. The average marginal effect of AR usage for customers who are new to the online channel is significantly positive (.018, p <.01), but this effect is not significant for existing online customers (.004, p =.59). Thus, H3a is supported. While the interaction between AR Usageeyes and New Category is marginally significant (.082, p <.10), the average marginal effect of AR usage is significantly positive for customers who are new to the product category (.019, p <.01) and not significant for existing category customers (.003, p =.65), providing support for H3b as well.
To understand how the impact of AR changes over time, we repeated the same analysis using a six-month and three-month window, presented in Columns 3 and 4 of Table 7. We find that the interactions between AR Usageeyes and both New Channel and New Category become stronger over time. Although both interactions (as well as the average marginal effects) are insignificant in the 3-month period, the interaction with New Channel becomes significant in the 6- and 12-month period, and the interaction with New Category becomes marginally significant in the 12-month period. Similar to the 12-month model, the average marginal effects in the 6-month model are significantly positive for customers who are new to the online channel (.022, p <.01 vs..006, p =.44 for existing online customers) and product category (.019, p <.05 vs..007, p =.31 for existing category customers). These results suggest that customers may require some time to become comfortable with the technology before using it to make purchase decisions. In addition, the results also imply that the impact of AR does not wear out over time, which rules out novelty effects as an alternative explanation.
Results for the two-stage residual inclusion and Heckman selection methods for all 3-, 6-, and 12-month periods are provided in Web Appendix I. To further examine if the findings are robust to alternative identification strategies, we applied the propensity score weighting approach. We used the first stage equation to calculate customers' propensity for using AR in the focal period and include this as weights in the estimation of Equation 3, following [ 6]. The results are consistent with the main model and are also reported in Web Appendix I.
We also examined if the findings are robust to alternative variable operationalizations. First, instead of the probability of purchasing eye products, we used the number of eye products purchased during the focal period as an alternative dependent variable. Second, we replaced the binary AR Usage variable with the number of sessions involving AR usage during the focal period. Third, as alternative measures of channel and category experience, we used the number of online transactions and number of eye products purchased prior to AR introduction for the eye categories, respectively. Findings from these models are consistent, and the results are presented in Web Appendix J.
Although firms are keen to invest in AR, research demonstrating its impact in real-world contexts is limited. The present research provides some preliminary confirmation that both the availability and usage of AR have a small but positive impact on sales. Taken together, our findings provide converging evidence that AR is most effective when product-related uncertainty is high, indicating that uncertainty reduction could be a possible mechanism through which AR could improve sales. Nevertheless, we do not find a significant moderating effect for product ratings, suggesting that even though AR may reduce product fit uncertainty, it may still be unable to compensate for the higher performance uncertainty associated with products that have lower ratings.[ 8] Although we have adopted instrumental variable and quasi-experimental approaches to address endogeneity that is inherent in observational data, we acknowledge that these findings should be viewed as evidence based on correlations, with attempts to come close to causality.
Complementing past research that has explored how website features drive sales for niche products (e.g., [ 8]; [50]), we show that AR can increase preference for products or brands that are less popular. Thus, retailers carrying wide product assortments can use AR to stimulate demand for products in the long tail of the sales distribution. AR may also help level the playing field for less popular brands. With the launch of AR-enabled display ads on advertising platforms such as Facebook and YouTube, less-established brands could consider investing in this new ad format, as they stand to benefit most from this technology. Retailers selling premium products may also leverage AR to improve decision comfort and reduce customers' hesitation in the purchase process.
We find that the impact of AR is stronger for customers who are new to the product category, suggesting that AR could increase sales via category expansion. However, because AR seems to be most effective when the level of uncertainty is high, its impact may diminish over time as customers become more familiar with the product category and experience less uncertainty.[ 9] Nevertheless, the finding that AR has a stronger impact for products that are more expensive suggests that, beyond increasing unit sales, AR can also improve category revenues by encouraging customers to purchase products with wider margins. Thus, investments in deploying AR in retail could pay off in the long run.
Compared with customers who are already familiar with purchasing online, we find that AR has a stronger effect for customers who are new to the online channel. As prior research has shown that multichannel customers are more profitable ([39]), omnichannel retailers can use AR to encourage their offline customers to adopt the online channel. Given that AR increases online sales among customers who are new to the channel, a potential concern is that AR could lead to cannibalization of sales from offline channels. To understand if the increase in app purchases that we observed was happening at the expense of other sales channels, we ran the same model in Equation 3 but replaced the dependent variable with the probability of purchasing eye products in the web and offline channels (results reported in Web Appendix K). We did not find evidence to indicate that offline customers who use AR (on the app) are more likely to purchase from the web, suggesting that the impact of AR is specific to the app platform. Interestingly, we find that offline customers who use AR are more likely to purchase from the offline channel in the three-month model but not in the six and 12-month model. Thus, contrary to our expectations, the results suggest that AR could have a positive spillover effect to the offline channel, at least in the short run.
Complementing prior research, which has predominantly studied AR from a consumer perspective, our research extends the literature by examining what AR means for retailers. To encourage the academic community to produce more impactful research in this nascent field, we developed a research agenda for AR in marketing, with an emphasis on identifying research topics that have strong managerial relevance for industry practitioners. Drawing on a review of the academic literature (e.g., [52]) and recent advancements in AR technology, we generated a list of potential research topics and synthesized these topics into five themes. Next, we consulted two senior marketing practitioners and two academics with expertise in this area to review the research themes and associated topics, and we refined the list according to their feedback.
To determine the practical importance of each research theme, we conducted an online survey with 36 marketing practitioners from companies that were using (or planning to use) AR in their marketing, advertising, or retailing activities. Survey respondents first independently rated each research theme in terms of importance to business performance (see [47]) before ranking the five research themes from most to least important. To avoid primacy and recency effects, the order of research themes was randomized across respondents. The mean rating (ranging from 5.1 to 5.8 on a seven-point scale) and ranking scores (from 1 to 5; lower number reflects higher importance) are inversely proportional, demonstrating internal consistency. Web Appendix L provides details for the survey, including survey design, respondent recruitment, and background of respondents.
Table 8 presents the research agenda for AR in marketing, comprising the five research themes (ordered by practical importance) and potential topics that could be explored under each theme. Given the novelty of the technology, marketers were primarily concerned with how different design features could be configured to create more effective AR experiences for consumers. For example, greater clarity is needed regarding factors that affect AR experiences, such as fidelity (i.e., how closely virtual objects resemble real objects), motion (i.e., static vs. animated virtual objects), spatial presence (i.e., the feeling that virtual objects exist in a physical space), and embodiment (i.e., the ability to use bodily movements to control virtual objects), and how these can be delivered on AR interfaces. Beyond visual and auditory senses, how haptic feedback (e.g., emission of vibrations on devices to stimulate the sense of touch) influences AR experiences is also of interest.
Graph
Table 8. Research Agenda for AR in Marketing.
| Research Themes | Potential Research Topics |
|---|
| Designing effective AR experiencesRating: 5.81Ranking: 2.14 | How factors such as fidelity (i.e., how closely virtual objects resemble real objects), motion (i.e., static vs. animated virtual objects), spatial presence (i.e., the feeling that virtual objects exist in a physical space), and embodiment (i.e., the ability to use bodily movements to control virtual objects) affect AR experiences. How the incorporation of senses such as haptic feedback (e.g., emission of vibrations) influences AR experiences. How content and other elements in the virtual environment could be personalized to enhance AR experiences and influence behavior.
|
| AR and marketing strategyRating: 5.50Ranking: 2.78 | How marketers could use AR more effectively at different stages of the customer journey to increase brand engagement and improve relationships with customers. Synergy between AR and other elements in the marketing communications mix (e.g., advertising, sales promotions). Effectiveness of product placements and pop-up stores in AR-enabled virtual environments, as well as their potential to complement or replace physical stores. How AR could be deployed in service industries (e.g., tourism and hospitality, food and beverage retail).
|
| AR and consumer behaviorRating: 5.22Ranking: 3.06 | How AR experiences affect sensory perceptions and cognitive functions (e.g., attention, information processing, learning, and memory). How AR experiences affect rational decision making (e.g., product selection strategies, relative importance of attributes) and irrational tendencies (e.g., psychological ownership). The role of AR experiences in attitude formation and brand perceptions. How AR experiences affect purchase behaviors and postpurchase product evaluations.
|
| Promoting AR adoptionRating: 5.22Ranking: 3.42 | Exploring barriers to consumers' use of AR technologies (e.g., awkwardness of using it in public, privacy and security concerns, lack of realism in virtual environments) and how marketers can overcome these barriers to encourage wider adoption. How delivery of the AR experience and advancements in high-tech devices (e.g., 3D depth camera technology, wearable AR glasses) influence consumers' acceptance and usage of the technology. How offline contextual factors (e.g., distance to physical stores, private vs. public space) affect AR usage.
|
| AR as a marketing intelligence toolRating: 5.11Ranking: 3.61 | How AR experiences could be used to generate insights for new product development, assortment planning, and store layout/design. Identifying new behavioral data (e.g., motion, interactions within virtual environments) that could be obtained from AR platforms, as well as how such data could be used to measure/predict consumers' responses or decision-making processes. Privacy and security concerns regarding behavioral data collected on AR platforms and what marketers can do to reduce these concerns.
|
9 Notes: Research themes are ordered by importance on the basis of surveys with 36 marketing practitioners. Details of the survey are provided in Web Appendix L. "Rating" refers to the mean importance rating score (on a seven-point scale). "Ranking" refers to the mean importance ranking (from 1 to 5; lower number reflects higher importance).
Another important consideration is how AR fits into companies' overall marketing strategy. Specifically, marketers would like to know how they can better integrate AR at different stages of the customer journey to increase brand engagement, build emotional connections, and improve relationships with customers. There is also ambiguity regarding the synergy between AR and other elements of the marketing communications mix (e.g., advertising, sales promotions), as well as the effectiveness of product placements and pop-up stores in AR-enabled virtual environments. In particular, the potential for this new technology to complement or replace existing communication and retail channels is still uncertain. As most recent applications of AR have focused on consumer products, marketers also need more guidance on how AR can be appropriately deployed in service industries, such as the tourism and hospitality sector.
Besides these two key areas, other worthwhile avenues to explore include the impact of AR on consumer behavior (e.g., cognitive functions, rational decision making, and brand perceptions), how marketers can promote wider adoption of AR, and how the technology can be used to generate valuable marketing insights. Although our research agenda focuses on AR, we note that the research themes could be broadened to encompass other extended realities (i.e., virtual reality and mixed reality).
In conclusion, we believe that the marketing community would benefit from a deeper investigation of virtual experiences and their role in marketing. We are excited about where this field is heading, and we look forward to more insightful research to reinforce our understanding of the profound impacts of these new technologies in the marketing domain.
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921995449 - Augmented Reality in Retail and Its Impact on Sales
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921995449 for Augmented Reality in Retail and Its Impact on Sales by Yong-Chin Tan, Sandeep R. Chandukala and Srinivas K. Reddy in Journal of Marketing
Footnotes 1 Benedict Dellaert
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research received financial support from the Singapore Economic Development Board (Grant Number: EDB-EDAS-2019-1).
4 Yong-Chin Tan https://orcid.org/0000-0002-6244-6564
5 Online supplement: https://doi.org/10.1177/0022242921995449
6 URL links to these examples are provided in Web Appendix B.
7 We also repeated the analysis for the three- and six-month periods and discuss insights from these analyses in the "Results" section.
8 We thank an anonymous reviewer for suggesting this possible explanation for the lack of a significant result that supports H2c.
9 We thank an anonymous reviewer for highlighting this possibility, and we encourage future research to explore the dynamic effects of AR usage.
References Angrist Joshua D. , Pischke Jörn-Steffen. (2009), Mostly Harmless Econometrics: An Empiricist's Companion. Princeton, NJ : Princeton University Press.
Arora Sandeep , Hofstede Frenkel ter , Mahajan Vijay. (2017), " The Implications of Offering Free Versions for the Performance of Paid Mobile Apps ," Journal of Marketing , 81 (6), 62 – 78.
Arthur Rachel. (2017), " Augmented Reality Is Set to Transform Fashion and Retail ," Forbes (October 31), https://www.forbes.com/sites/rachelarthur/2017/10/31/augmented-reality-is-set-to-transform-fashion-and-retail/.
Bauer Raymond A.. (1960), " Consumer Behavior as Risk Taking ," in Dynamic Marketing for a Changing World , Hancock Robert S. , ed. Chicago : American Marketing Association , 389 – 98.
BCG (2018), " Augmented Reality: Is the Camera the Next Big Thing in Advertising? " (April 3), https://www.bcg.com/publications/2018/augmented-reality-is-camera-next-big-thing-advertising.
Bell David R. , Gallino Santiago , Moreno Antonio. (2018), " Offline Showrooms in Omnichannel Retail: Demand and Operational Benefits ," Management Science , 64 (4), 1629 – 51.
Brengman Malaika , Willems Kim , Kerrebroeck Helena van. (2019), " Can't Touch This: The Impact of Augmented Reality Versus Touch and Non-Touch Interfaces on Perceived Ownership ," Virtual Reality , 23 (3), 269 – 80.
Brynjolfsson Erik , Yu (Jeffrey) Hu , Simester Duncan. (2011), " Goodbye Pareto Principle, Hello Long Tail: The Effect of Search Costs on the Concentration of Product Sales ," Management Science , 57 (8), 1373 – 86.
Burke Raymond R.. (2002), " Technology and the Customer Interface: What Consumers Want in the Physical and Virtual Store ," Journal of the Academy of Marketing Science , 30 (4), 411 – 32.
Cameron A. Colin , Miller Douglas L.. (2015), " A Practitioner's Guide to Cluster-Robust Inference ," Journal of Human Resources , 50 (2), 317 – 72.
Carnahan Daniel. (2019), " Facebook Has Made AR Ads Available to All Marketers Through Its Ad Manager," Business Insider (December 12), https://www.businessinsider.com/facebook-experiments-with-augmented-reality-advertising-2019-12.
Chandukala Sandeep R. , Dotson Jeffrey P. , Liu Qing. (2017), " An Assessment of When, Where, and Under What Conditions In-Store Sampling Is Most Effective ," Journal of Retailing , 93 (4), 493 – 506.
Chaykowski Kathleen. (2018), " Inside Facebook's Bet on an Augmented Reality Future ," Forbes (March 8), https://www.forbes.com/sites/kathleenchaykowski/2018/03/08/inside-facebooks-bet-on-an-augmented-reality-future/#6973829b4d56.
Chen Yubo , Xie Jinhong. (2008), " Online Consumer Review: Word-of-Mouth as a New Element of Marketing Communication Mix ," Management Science , 54 (3), 477 – 91.
Danaher Peter J. , Danaher Tracey S. , Smith Michael Stanley , Loaiza-Maya Ruben. (2020), " Advertising Effectiveness for Multiple Retailer-Brands in a Multimedia and Multichannel Environment ," Journal of Marketing Research , 57 (3), 445 – 67.
Danaher Peter J. , Wilson Isaac W. , Davis Robert A.. (2003), " A Comparison of Online and Offline Consumer Brand Loyalty ," Marketing Science , 22 (4), 461 – 76.
Dimoka Angelika , Hong Yili , Pavlou Paul A.. (2012), " On Product Uncertainty in Online Markets: Theory and Evidence ," MIS Quarterly , 36 (2), 395 – 426.
Erdem Tülin , Swait Joffre , Valenzuela Ana. (2006), " Brands as Signals: A Cross-Country Validation Study ," Journal of Marketing , 70 (1), 34 – 49.
Ericsson (2017), " Merged Reality: Understanding How Virtual and Augmented Realities Could Transform Everyday Reality ," (accessed June 26, 2020), https://www.ericsson.com/en/reports-and-papers/consumerlab/reports/merged-reality.
Fisher Marshall L. , Gallino Santiago , Joseph Jiaqi Xu. (2019), " The Value of Rapid Delivery in Omnichannel Retailing ," Journal of Marketing Research , 56 (5), 732 – 48.
Gallino Santiago , Moreno Antonio. (2018), " The Value of Fit Information in Online Retail: Evidence from a Randomized Field Experiment ," Manufacturing & Service Operations Management , 20 (4), 767 – 87.
Gray Alistair. (2019), " Retailers Grapple with $100bn Returns Problem ," Financial Times (December 27), https://www.ft.com/content/5bafd9c0-235f-11ea-92da-f0c92e957a96.
Griffin Andrew. (2017), " Apple's Tim Cook on iPhones, Augmented Reality, and How He Plans to Change Your World ," The Independent (October 12), https://www.independent.co.uk/life-style/gadgets-and-tech/features/apple-iphone-tim-cook-interview-features-new-augmented-reality-ar-arkit-a7993566.html.
Gu Zheyin (Jane) , Tayi Giri K.. (2015), " Consumer Mending and Online Retailer Fit-Uncertainty Mitigating Strategies ," Quantitative Marketing and Economics , 13 (3), 251 – 82.
Heckman James J.. (1979), " Sample Selection Bias as a Specification Error ," Econometrica , 47 (1), 153 – 62.
Heller Jonas , Chylinski Mathew , de Ruyter Ko , Mahr Dominik , Keeling Debbie I.. (2019 a), " Let Me Imagine That for You: Transforming the Retail Frontline Through Augmenting Customer Mental Imagery Ability ," Journal of Retailing , 95 (2), 94 – 114.
Heller Jonas , Chylinski Mathew , de Ruyter Ko , Mahr Dominik , Keeling Debbie I.. (2019 b), " Touching the Untouchable: Exploring Multi-Sensory Augmented Reality in the Context of Online Retailing ," Journal of Retailing , 95 (4), 219 – 34.
Hilken Tim , Keeling Debbie I. , de Ruyter Ko , Mahr Dominik , Chylinski Mathew. (2020), " Seeing Eye to Eye: Social Augmented Reality and Shared Decision Making in the Marketplace ," Journal of the Academy of Marketing Science , 48 (2), 143 – 64.
Hilken Tim , de Ruyter Ko , Chylinski Mathew , Mahr Dominik , Keeling Debbie I.. (2017), " Augmenting the Eye of the Beholder: Exploring the Strategic Potential of Augmented Reality to Enhance Online Service Experiences ," Journal of the Academy of Marketing Science , 45 (6), 884 – 905.
Hoch Stephen J.. (2002) " Product Experience Is Seductive ," Journal of Consumer Research , 29 (3), 448 – 54.
Hollenbeck Brett. (2018), " Online Reputation Mechanisms and the Decreasing Value of Chain Affiliation ," Journal of Marketing Research , 55 (5), 636 – 54.
Hong Yili , Pavlou Paul A.. (2014), " Product Fit Uncertainty in Online Markets: Nature, Effects, and Antecedents ," Information Systems Research , 25 (2), 328 – 44.
Janakiraman Ramkumar , Lim Joon Ho , Rishika Rishika. (2018), " The Effect of a Data Breach Announcement on Customer Behavior: Evidence from a Multichannel Retailer ," Journal of Marketing , 82 (2), 85 – 105.
Kempf Deanna S. , Smith Robert E.. (1998), " Consumer Processing of Product Trial and the Influence of Prior Advertising: A Structural Modeling Approach ," Journal of Marketing Research , 35 (3), 325 – 39.
Kim Jiyeon , Forsythe Sandra. (2008), " Adoption of Virtual Try-On Technology for Online Apparel Shopping ," Journal of Interactive Marketing , 22 (2), 45 – 59.
Kim Youngsoo , Krishnan Ramayya. (2015), " On Product-Level Uncertainty and Online Purchase Behavior: An Empirical Analysis ," Management Science , 61 (10), 2449 – 67.
Kübler Raoul , Pauwels Koen , Yildirim Gökhan , Fandrich Thomas. (2018), " App Popularity: Where in the World Are Consumers Most Sensitive to Price and User Ratings? " Journal of Marketing , 82 (5), 20 – 44.
Metz Rachel. (2019), " Virtual Makeovers Are Better than Ever. Beauty Companies Are Trying to Cash In," CNN Business (February 19), https://edition.cnn.com/2019/02/19/tech/augmented-reality-makeup/index.html.
Montaguti Elisa , Neslin Scott A. , Valentini Sara. (2016), " Can Marketing Campaigns Induce Multichannel Buying and More Profitable Customers? A Field Experiment ," Marketing Science , 35 (2), 201 – 17.
Papies Dominik , Ebbes Peter , Heerde Harald J. van. (2017), " Addressing Endogeneity in Marketing Models ," in Advanced Methods for Modeling Markets , Leeflang Peter S. H. , Wieringa Jaap E. , Bijmolt Tammo H. A. , Pauwels Koen H. , eds. Cham, Switzerland : Springer , 581 – 627.
Pavlou Paul. A. , Liang Huigang , Xue Yajiong. (2007), " Understanding and Mitigating Uncertainty in Online Exchange Relationships: A Principal-Agent Perspective ," MIS Quarterly , 31 (1), 105 – 36.
Petrin Amil , Train Kenneth. (2010), " A Control Function Approach to Endogeneity in Consumer Choice Models ," Journal of Marketing Research , 47 (1), 3 – 13.
Petrock Victoria. (2020), " US Virtual and Augmented Reality Users 2020," eMarketer (April 7), https://www.emarketer.com/content/us-virtual-and-augmented-reality-users-2020.
Porter Michael E. , Heppelmann James E.. (2017), " Why Every Organization Needs an Augmented Reality Strategy ," Harvard Business Review , 95 (6), 46 – 57.
Simonson Itamar. (1992), " The Influence of Anticipating Regret and Responsibility on Purchase Decisions ," Journal of Consumer Research , 19 (1), 105 – 18.
Smith Robert E. , Swinyard William R.. (1982), " Information Response Models: An Integrated Approach ," Journal of Marketing , 46 (1), 81 – 93.
Stremersch Stefan , Dyck Walter van. (2009), " Marketing of the Life Sciences: A New Framework and Research Agenda for a Nascent Field ," Journal of Marketing , 73 (4), 4 – 30.
Terza Joseph V. , Basu Anirban , Rathouz Paul J.. (2008), " Two-Stage Residual Inclusion Estimation: Addressing Endogeneity in Health Econometric Modeling ," Journal of Health Economics , 27 (3), 531 – 43.
Tucker Catherine. (2014), " Social Networks, Personalized Advertising, and Privacy Controls ," Journal of Marketing Research , 51 (5), 546 – 62.
Tucker Catherine , Zhang Juanjuan. (2011), " How Does Popularity Information Affect Choices? A Field Experiment ," Management Science , 57 (5), 828 – 42.
Weathers Danny , Sharma Subhash , Wood Stacy L.. (2007), " Effects of Online Communication Practices on Consumer Perceptions of Performance Uncertainty for Search and Experience Goods ," Journal of Retailing , 83 (4), 393 – 401.
Wedel Michel , Bigné Enrique , Zhang Jie. (2020), " Virtual and Augmented Reality: Advancing Research in Consumer Marketing ," International Journal of Research in Marketing , 37 (3), 443 – 465.
Wood Stacy L.. (2001), " Remote Purchase Environments: The Influence of Return Policy Leniency on Two-Stage Decision Processes ," Journal of Marketing Research , 38 (2), 157 – 69.
Wooldridge Jeffrey M.. (2010), Econometric Analysis of Cross Section and Panel Data , 2nd ed. Cambridge, MA : MIT Press.
Yim Mark Yi-Cheon , Chu Shu-Chuan , Sauer Paul L.. (2017), " Is Augmented Reality Technology an Effective Tool for E-Commerce? An Interactivity and Vividness Perspective ," Journal of Interactive Marketing , 39 , 89 – 103.
~~~~~~~~
By Yong-Chin Tan; Sandeep R. Chandukala and Srinivas K. Reddy
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 10- Bad News? Send an AI. Good News? Send a Human. By: Garvey, Aaron M.; Kim, TaeWoo; Duhachek, Adam. Journal of Marketing. Feb2022, p1. DOI: 10.1177/00222429211066972.
Ahead of Print- Database:
- Business Source Complete
Record: 11- Befriending the Enemy: The Effects of Observing Brand-to-Brand Praise on Consumer Evaluations and Choices. By: Zhou, Lingrui; Du, Katherine M.; Cutright, Keisha M. Journal of Marketing. Jul2022, Vol. 86 Issue 4, p57-72. 16p. 1 Diagram, 1 Graph. DOI: 10.1177/00222429211053002.
- Database:
- Business Source Complete
Befriending the Enemy: The Effects of Observing Brand-to-Brand Praise on Consumer Evaluations and Choices
Consumers have grown increasingly skeptical of brands, leaving managers in a dire search for novel ways to connect. The authors suggest that focusing on one's relationships with competitors is a valuable, albeit unexpected, way for brands to do so. More specifically, the present research demonstrates that praising one's competitor—via "brand-to-brand praise"—often heightens preference for the praiser more so than other common forms of communication, such as self-promotion or benevolent information. This is because brand-to-brand praise increases perceptions of brand warmth, which leads to enhanced brand evaluations and choice. The authors support this theory with seven studies conducted in the lab, online, and in the field that feature multiple managerially relevant outcomes, including brand attitudes, social media and advertising engagement, brand choice, and purchase behavior, in a variety of product and service contexts. The authors also identify key boundary conditions and rule out alternative explanations, further elucidating the underlying mechanism and important implementation insights. This work contributes to the understanding of brand perception and warmth, providing a novel way for brands to connect to consumers by connecting with each other.
Keywords: praise; brand relationships; brand communication; brand evaluations; warmth and competence
In 2017, a popular video gaming brand, Xbox, openly congratulated its competitor, Nintendo, on the launch of its new Switch gaming system ([53]). A few months later, The New York Times encouraged readers to read other news sources such as The Wall Street Journal ([20]). And, in responding to a playful challenge from Kit Kat, Oreo disarmed the brand by communicating how truly irresistible Kit Kat is ([62]). Conventional wisdom fervently advises that "even mentioning your competition is a bad idea" ([51]), so why have these brands not only mentioned their competitors but praised them?
In a world where brands are trying hard to connect with consumers, many of whom have grown increasingly skeptical of marketers' intentions ([21]), it may be that praising the competition provides unexpected benefits. Our research explores the consequences of brand-to-brand praise—when a brand communicates positively and publicly about another brand. We argue that consumers who observe a brand praising a competitor will believe that the brand has positive intentions toward others, also known as brand warmth, which heightens consumer evaluations and interest in the brand giving the praise.
As an introductory illustration of this idea, we scraped data from the Twitter pages of Nintendo and its fiercest competitors, Xbox and PlayStation, around the time of the Nintendo Switch launch in 2017. We found a greater number of likes and retweets (in fact, over ten times as many), as well as more positive sentiment among consumers' comments, when Xbox and PlayStation praised Nintendo for the launch compared with all other types of messages (Web Appendix A). Such preliminary field data motivate our exploration into whether, when, and why brand-to-brand praise affects consumer reactions to brands.
Across seven studies, we examine the effects of brand-to-brand praise on consumer attitudes and behavior versus more common forms of brand messaging (e.g., self-promotion or providing helpful information to consumers) and identify important boundaries. In doing so, this research offers several contributions.
First, we contribute to the brand perception literature by demonstrating how consumers' perceptions of brands are affected by a brand's interactions with other brands. Prior work has focused on how brand-to-consumer interactions affect consumer perceptions (e.g., [42]) but has not yet explored how observing brand-to-brand interactions do so. Furthermore, we contribute to research on the fundamental dimensions by which people judge other people and brands: warmth and competence (e.g., [ 1]). We identify brand-to-brand praise as a novel antecedent that often leads consumers to perceive brands that praise their competitors as warmer. We show that praising a competitor is viewed as a costly action that does not obviously benefit the praiser, thus making it a credible signal. In doing so, we demonstrate the role of costliness in signaling warmth that effectively combats consumer skepticism, a major barrier to warmth identified in prior research ([18]). We also introduce two moderators—organization type and individual differences in skepticism—to identify when costly displays of warmth are most important.
Second, we further contribute to the literature on warmth and competence by introducing a context in which brand-to-brand praise increases warmth without damaging perceptions of competence. Prior work suggests that warmth and competence are often negatively correlated, particularly in contexts in which people are considering two or more entities ([32]). Leveraging consumers' lay theories about the characteristics of brands that would be willing to praise their competitors, brand-to-brand praise provides an opportunity for brands to communicate warmth while maintaining perceptions of competence.
Third, while prior work has examined brand communication in which the competitive brand is ultimately shown to be inferior, such as in comparative advertising and two-sided messaging campaigns (e.g., [ 4]; [17]; [40]), our research identifies how directing attention away from one's own brand and toward the competition in a purely positive light affects brand evaluations and choice.
Finally, we add to the current literature on praise by identifying brand-to-brand communication as a viable and distinct form of praise, noting that observers respond more favorably to praise in a brand-to-brand context than is typically observed in traditional person-to-person contexts. In doing so, we also highlight the understudied effects of praise in competitive relationships. In what follows, we review literature on brand relationships, praise, and brand warmth and present seven studies that test our hypotheses.
A great deal of research has investigated how brands establish and build relationships with consumers, identifying influential factors such as the personalities, motives, and communication styles of both consumers and brands (for a review, see MacInnis, Park, and Priester [2009]). Surprisingly, researchers interested in brand relationships have not yet explored how brands' public communication with other brands affects their relationships with consumers. Of course, researchers have explored how strategic partnerships, such as brand collaborations and alliances, affect brand outcomes (e.g., [38]; [57]). However, these brand interactions occur largely behind the scenes and reflect formal business arrangements as opposed to informal, public communication that makes consumers privy to how a brand treats its competitors. We suggest that consumers will infer important information about a brand based on the way it communicates with other brands. Specifically, we posit that consumers who observe a brand praising its competition will perceive the praiser brand as having positive intentions toward others, known as brand warmth. Subsequently, consumers will develop more positive evaluations of and interest in the praiser brand.
Prior research has established that people judge others—individuals, groups, cultures, countries—using two fundamental dimensions, often referred to as warmth and competence ([25]; [32]).[ 5] Warmth is the degree to which one has positive intentions toward others and includes perceptions of thoughtfulness, kindness, honesty, and trustworthiness ([ 1]; [25]). Evolutionarily speaking, it allowed our ancestors to quickly distinguish friend from foe and prepare to fight or flee. Warmth judgments are therefore formed more quickly and generally have the greatest impact on attitudes toward individuals ([64]). Conversely, competence reflects the degree to which one is able to enact one's intentions.
Given that people relate to brands similarly to how they relate to people in many ways ([26]), warmth and competence are important traits for firms to consider ([33]). Surprisingly, scant literature has explored the drivers of warmth and competence for brands ([ 1]). The research that does exist identifies factors such as a firm's profit focus ([ 2]), social responsibility ([ 8]), racial dynamics ([ 6]), and expression/communication style ([61]; [65]) as affecting consumers' perceptions of warmth and competence. Significantly more work has explored the consequences of warmth and competence. As examples, research has shown that warmth and competence perceptions affect consumer emotions ([ 1]), product evaluations and interest ([10]; [14]; [34]; [36]; [39]; [58]), and word of mouth ([ 7]; [54]). Importantly, while the precise contribution of warmth versus competence to different downstream consequences varies (e.g., [37]), brands generally aspire to be strong on both dimensions and occupy the coveted "golden quadrant" ([ 1]).
In this research, we suggest that brand-to-brand praise affects consumers' reactions primarily through perceptions of warmth. Prior research has theorized that warmth is established through signals of cooperation (vs. competition) and actions that appear to serve others as opposed to the self (i.e., actions that are "other-profitable"; [16]; [52]). We suggest that offering praise to a competitor provides a strong illustration of such cooperative, "other-profitable" activity. Consumers will therefore perceive a brand that praises competitive brands as warmer (relative to a brand that engages in other types of common brand messaging). This then leads to positive downstream consequences, such as increased purchases.
In hypothesizing the positive effect of brand-to-brand praise on warmth, it is important to note that high warmth is more difficult to establish and maintain than high competence, as people are often skeptical of others' motives. That is, warm behavior is often discounted, considered to be driven by ulterior motives and easier to fake than competence ([18]; [56]). Such skepticism has also been identified as an issue in research on praise more specifically. For example, in many brand-to-consumer exchanges, such as salesclerk-to-consumer interactions, praise generates suspicion from the consumer; consumers often suspect ulterior motives behind a salesclerk's compliment and perceive the salesclerk to be insincere when the praise occurs before a purchase ([11]; [44]). Further, observers who witness praise happening among others tend to be more skeptical of the praiser and do not react as positively as recipients of the praise ([13]; [29]; [60]). This then leads to the question: When and why might brand-to-brand praise surmount the skepticism associated with praise and other displays of warmth?
We suggest that brand-to-brand praise operates uniquely from the aforementioned person-to-person and brand-to-consumer displays of warmth, particularly when the praise is directed toward competitors, due to its costliness. Consumers assume that complimenting a competitor is a costly action that does not directly benefit the complimenting brand. Research across disciplines suggests that the costliness of an act is the key component of whether the act is perceived as a meaningful signal of an underlying trait rather than an uninformative act motivated by devious or ulterior motives, also known as "cheap talk" (e.g., [59]; [67]). A common example from the natural world is the male peacock's tail. The costliness of this large and colorful tail, such as how it can handicap the bird's ability to escape from predators, is what makes it a credible signal of fitness to potential mates. Only those who truly possess the focal underlying trait would or could incur such costs. Linking this principle to the present context, only brands that are truly warm would incur the real or potential costs of praising the competition. Consumers should therefore not be as suspicious of brands that praise competitors as compared with brands that engage in less costly messaging. This costliness is why brand-to-brand praise, when directed toward competitors, is a strong signal of warmth, differentiating it from other common types of praise (e.g., salesclerk-to-consumer), surmounting consumer skepticism, and driving its positive influence on consumer attitudes and reactions.
Although the focus thus far has been on perceptions of warmth, one might wonder whether brands sacrifice competence when enhancing perceptions of warmth via brand-to-brand praise. This is a reasonable concern, as prior research suggests that in comparative contexts, for people and brands alike, when warmth (competence) is relatively high for a given entity, perceptions of that entity's competence (warmth) suffer (e.g., [ 2]; [28]; [36]; [54]). We suggest that praising a competitor offers brands unique advantages that allow them to maintain perceptions of competence in the face of increasing warmth. In particular, we suggest that consumers hold a lay intuition that a brand that is willing to praise its competitors must be fairly confident in its own abilities, which then allows consumers to maintain, if not increase, positive perceptions of its competence. This is reminiscent of the lay belief—confirmed by academic research—that people who are secure in their identity are the most willing to be kind and compliment others (e.g., [ 9]; [22]; [41]; [63]). Still, while competence may play a role in driving the effects of brand-to-brand praise, we do not expect it to be the primary driver of increased interest in the praiser brand, as perceptions of competence should be most strongly driven by cues of status versus the cues of cooperation that are focal in brand-to-brand praise ([24]).
In summary, we argue that brand-to-brand praise often promotes positive brand evaluations and choice of the praiser. Specifically, we predict that consumers will evaluate a brand more favorably and show more interest in the brand when observing brand-to-brand praise compared with observing traditional self-focused messages or even other benevolent messages (H1). We theorize that this is because such praise increases perceptions of the praiser brand's warmth (H2).
We expect that this effect exists when the praise is costly (H3), such as when the brand is praising a competitor, so as to provide a meaningful signal of warmth to the consumer. We reason that the costly signal of warmth indicates to the consumer that the brand truly has positive intentions toward others, even when it is not in the best interest of the brand. As prior work suggests, such warmth leads consumers to be more interested in identifying with, using, and sharing the brand ([ 7]; [37]).
As additional evidence for the role of warmth, we expect the effect of brand-to-brand praise to be stronger for brands that are typically associated with lower levels of warmth than those already endowed with high levels of warmth; brands with high levels of warmth, such as nonprofits, have little to gain from increasing warmth even further (and thus less to gain from brand-to-brand praise). This suggests that for-profit brands (which are lower in warmth than nonprofit brands; [ 2]) will benefit more from brand-to-brand praise (H4).
Finally, we posit that brand-to-brand praise allows brands to manage consumer skepticism often associated with displays of warmth. We introduce an important moderator to illustrate this point: individual differences in consumers' skepticism toward brand messaging. We predict that the effect of brand-to-brand communication will be strongest among individuals who are highly skeptical of brands. This is because the costliness associated with praising a competitor minimizes the extent to which persuasion knowledge concerns are activated among this group of consumers as compared with more traditional brand communication (H5). We summarize these predictions in the conceptual model in Figure 1 and discuss the boundary predictions further in the study introductions relevant to each.
Graph: Figure 1. Conceptual model.
We first demonstrate the positive effects of brand-to-brand praise (vs. more traditional messages) across three field and lab studies involving real, consequential behaviors (Studies 1a, 1b, and 2; H1). Next, we show that this effect is specific to praise that is aimed toward a brand's competitor and thus deemed costly (Study 3; H3). We then explore warmth as the key mechanism, demonstrating that it mediates the effect of brand-to-brand praise (Study 4; H2) and that the effect is strongest for for-profit brands that have greater need to enhance perceptions of warmth (Study 5; H4). Finally, we demonstrate that brand-to-brand praise has the largest effect on individuals with high levels of skepticism toward traditional brand communication (Study 6; H5). Throughout these studies, we also test for alternative explanations, including that praise confers benefits due to novelty or authenticity, or that the effect is driven by negative reactions to self-promotion (i.e., bragging) rather than positive reactions to praise. Finally, we also examine the role of competence in several studies (Studies 4, 5, and 6). Evidence suggests that, while perceptions of brand warmth are the primary mechanism underlying this effect, brand-to-brand praise does not harm and can even boost perceptions of competence, which can influence desired outcomes.
Together, these studies provide insight into when and why brand-to-brand praise is beneficial for brands. Notably, however, we demonstrate (in Studies 3, 5, and 6) and explain (in the "General Discussion" section) when and among whom brand-to-brand praise might not be as beneficial or might even be less beneficial.
Studies 1a and 1b provide initial causal evidence for our hypothesis by testing the effects of various types of brand messaging on advertisement click-through rate and brand choice in a field and lab study, respectively. We compare the effects of brand-to-brand praise with a traditional self-promotion message from the brand. In addition, in Study 1a, we utilize another type of control message in the form of an endorsement from another organization in the industry to show that praising a competitor enhances evaluations more than receiving an endorsement from others. Further, the endorsement condition serves as a separate point of comparison, ensuring that the hypothesized difference between the brand-to-brand praise condition and the self-promotion condition can be interpreted as a boost from brand-to-brand praise and not merely a negative reaction to self-promotion.
Participants and design. This preregistered study utilized a three-cell between-subjects design: Facebook users saw an advertisement on Facebook featuring self-promotion, an external endorsement, or brand-to-brand praise.[ 6]
Procedure. We created a fictitious car wash brand called Precision Car Wash and launched three advertisements for the brand on Facebook. The self-promotion message stated, "Precision Car Wash is proud to receive the Industry Best 2020 Award." The external endorsement consisted of a message from a fictitious organization, The Industry Best 2020 Award Committee, announcing Precision Car Wash as the year's award recipient. Finally, the brand-to-brand praise ad consisted of a message from Precision Car Wash congratulating another fictitious car wash business, LikeNew Car Wash, on winning the Industry Best 2020 Award (for study stimuli, see Web Appendix B). Facebook users who saw the ad could click on the message to be taken to the Facebook page of Precision Car Wash for more information. We measured the number of impressions and clicks, which allowed us to compare click-through rates (clicks as a percent of impressions; CTRs), for each ad. We ran the ads over the course of five days and used Facebook's recommended advertising algorithm to reach 80% power in testing the ads for a final sample of 13,719 impressions.
We also conducted two separate supplemental tests of these stimuli. First, we conducted a manipulation check for the ads, assigning participants to one of the three ad conditions and asking them to indicate the extent to which the brand was praising its competition (1 = "definitely NOT praising the competition," and 7 = "definitely praising the competition"; N = 105). Second, given variations in the text used to communicate the focal messages (e.g., number of words, fonts), we conducted a posttest to assess the "graphic design" (three items: quality, clarity, graphic appeal) of the assigned ad (1 = "not done very well," and 7 = "done very well"; α = .87; N = 105). These items allowed us to be sure that our focal praise condition was not (unintentionally) more aesthetically appealing and more likely to be clicked for that reason.
Manipulation check and aesthetic appeal. The separate manipulation check showed a significant difference across conditions (F( 2, 101) = 21.47, p < .001, = .30). Those who viewed the brand-to-brand praise message perceived it to praise the competition (M = 4.82) more than the self-promotion (M = 1.96; p < .001) and external endorsement (M = 2.43; p < .001) ads. The self-promotion and external endorsement conditions did not differ (p = .30). This manipulation check was conducted for and confirmed the manipulations in each of the remaining studies (see Web Appendix C).
In addition, the separate test designed to assess the aesthetic appeal of the different messages indicated that the focal praise ad was not advantaged aesthetically. We find a significant difference across conditions (F( 2, 102) = 5.50, p = .005, = .10) such that those who viewed the brand-to-brand praise ad perceived it to be lower in aesthetic appeal (M = 3.83) than the self-promotion ad (M = 5.03; p = .001) and not statistically different from the external endorsement ad (M = 4.42; p = .11). The self-promotion and external endorsement conditions did not significantly differ (p = .10). However, because the praise ad was rated to be lower in aesthetic appeal, any positive benefits of the praise ad should not be a result of viewers liking the appearance of the praise ad more.
CTR. A chi-squared analysis revealed that the CTR differed across conditions (χ2( 2, N = 13,719) = 91.59, p < .001). The percentage of those who clicked on the ad was greater for the brand-to-brand praise condition (5.4% of 4,392 impressions) compared with the self-promotion (3.3% of 4,075 impressions; χ2( 1, N = 8,467) = 21.42, p < .001) and external endorsement (1.8% of 5,252 impressions; χ2( 1, N = 9,644) = 90.45, p < .001) conditions.
Participants and design. One hundred fifty-four members of the local community (general public, students, and staff) were recruited through a business school's behavioral lab. Participants were randomly assigned to one of two conditions: self-promotion or praise.
Procedure. First, participants learned that they would give their opinions about snack brands that the business school was considering for its café and vending machines. Next, they were informed that they would be able to choose a sample snack to take home after the study.
Participants then saw two advertisements from real local popcorn shops in the Raleigh-Durham, North Carolina, area. The first ad was from the recipient brand, Carolina Popcorn Shoppe, and stated, "Come check out our FIVE newest flavors! In-store or online." Everyone saw this ad. The second ad was from the focal brand of the study, The Mad Popper, and differed by condition. The praise condition ad read, "We love good popcorn. Big shout-out to Carolina Popcorn Shoppe on their FIVE new flavors!" The self-promotion ad read, "We love good popcorn. Come explore our FIVE brand new flavors! In-store or online" (Web Appendix D).
Next, participants were reminded that both brands were being considered for use at the business school. They were then asked to choose Carolina Popcorn Shoppe or The Mad Popper as the brand they would prefer to sample, which they received at the conclusion of the study.[ 7] Finally, participants responded to a stimuli believability measure, assessing their willingness to suspend disbelief and assume that the experimental stimuli were real ("How believable was this advertisement by The Mad Popper?"; 1 = "not at all," and 7 = "very"). We asked this believability question consistently across experiments (except Study 1a, where we could not) because of the low prominence of brand-to-brand praise currently in the market. Although brand-to-brand praise is not yet commonly practiced in the real world and thus may not yet be highly believable (i.e., seem real) for some consumers, we are examining what could happen if consumers witnessed brand-to-brand praise, meaning controlling for variation in whether the stimuli were believed to be real.[ 8] We use believability as a covariate across each subsequent experiment to enhance experimental power ([45]). Details on this measure appear in Web Appendix E, which includes a summary of key results with and without the covariate.[ 9]
Brand choice. We conducted a binary logistic regression with condition (1 = praise, 0 = self-promotion) as the key predictor and believability as a continuous covariate on brand choice (1 = The Mad Popper, 0 = Carolina Popcorn Shoppe). Conceptually replicating the results of Study 1a, those in the praise condition were significantly more likely to choose the focal brand, The Mad Popper, compared with those in the self-promotion condition (β = .91, SE = .43, Wald χ2( 1, N = 154) = 4.47, p = .03). In percentages, 27.63% chose the focal brand in the self-promotion condition, and 34.62% did so in the praise condition.[10] We also find that believability does not act as a moderator (β = −.02, SE = .25, Wald χ2( 1, N = 154) = −.08, p = .93); thus, we use it as a covariate in our remaining studies (for moderation by believability results for the remaining studies, see Web Appendix E).
Studies 1a and 1b provide initial experimental evidence that brand-to-brand praise can have positive consequences for the praiser on real behavior, including advertisement CTR and brand choice. The comparison of brand-to-brand praise to an external endorsement in Study 1a suggests that the key effect is not because consumers dislike the other control condition—the self-promotion message—but is instead driven by a boost from observing brand-to-brand praise. Furthermore, by presenting participants with a forced choice between two brands in Study 1b, this data begins to suggest that praise benefits the praiser more than the praised. We further explore this in Study 2.
In Study 2, we build on the findings of Studies 1a and 1b in assessing consumers' real, behavioral reactions to brand-to-brand praise, but we do so with some important changes. First, we assess a longer-term reaction to brand-to-brand praise by investigating purchase behavior for popular national brands (Kit Kat and Twix) 11 days after people were exposed to the brand messaging. Second, unlike Study 1b (but similar to Study 1a), we only expose participants to the competitor brand in the brand-to-brand praise condition. This enables us to further assess the potential costs of making competitive brands more top-of-mind (via praise) than they might otherwise be. This study also allows us to assess behavioral reactions to both the focal and competitive brands by gauging consumers' purchase behavior toward both, as opposed to forcing a choice between them (Study 1b) or tracking reactions to only the focal brand (Study 1a). Finally, this design offers an opportunity to see the proposed effect of praise on behaviors of greater financial consequence for the consumer and the brand: purchases.
This preregistered study had a two-cell between-subjects design (control, praise) and was conducted in two stages on Prolific.[11]
In the first stage of this study, participants recruited from Prolific (N = 1,502; 49.6% female) viewed an image of Kit Kat's Twitter page. In the control condition, participants read a tweet that said, "Start your day off with a tasty treat!" In the praise condition, participants read a tweet that said, "@twix, Competitor or not, congrats on your 54 years in business! Even we can admit—Twix are delicious" (Web Appendix F). Unlike Study 1b (but similar to Study 1a), participants did not see a separate introduction to the nonfocal brand, Twix. Thus, Twix would presumably not be as top-of-mind unless they read the praise tweet. Afterward, we measured participants' attitudes toward both Kit Kat and Twix using the attitude measure from [46] ("negative/positive," "dislikeable/likeable," and "unfavorable/favorable" on seven-point scales; α = .95; for details on this measure, see Web Appendix F).
From the initial sample of 1,502 participants, we excluded those who indicated that they have not bought chocolate candy for themselves within the past six months, those with dietary restrictions that prevent them from buying chocolate candy, and those who indicated that they were not interested in completing a follow-up survey, leaving us with a sample of 1,298 potential participants for the second stage. Approximately 11 days after participants completed the first stage, we sent a follow-up survey to the 1,298 participants. Of these participants, 772 participants completed the second-stage survey. Attrition rates were similar across conditions (control = 40.64%, praise = 40.40%, χ2( 1, N = 1,298) = .008, p = .93).
In the second-stage survey, participants indicated whether they had purchased any Kit Kats (yes/no) or Twix (yes/no) since they took the first portion of the survey. Finally, we asked participants in an open-ended question what they could recall about the tweet they saw in the first stage of the study, bringing the tweet to their attention so that we could measure the believability of the tweet using an adapted version of the measure in the prior study.
We conducted a binary logistic regression with condition (1 = praise, 0 = control) as the key predictor and believability as a continuous covariate on purchase behavior (1 = purchased Kit Kat, 0 = did not purchase Kit Kat). Those in the praise condition were significantly more likely to purchase Kit Kat compared with those in the control condition (β = .34, SE = .16, Wald χ2( 1, N = 772) = 4.17, p = .04). In raw percentages, 23.77% in the control condition purchased Kit Kat versus 31.95% in the praise condition. We conducted the same analysis for purchase behavior for Twix and find that there was no difference between the praise and control conditions (β = −.25, SE = .19, Wald χ2( 1, N = 772) = 1.68, p = .20)
Study 2 extends our prior findings, demonstrating the effect of brand-to-brand praise on actual purchase behavior over a longer time horizon and relative to competitors' purchase outcomes. In doing so, we also show that even if the competitor brand becomes more salient than it would have normally been as a result of brand-to-brand praise, such praise still primarily provides positive benefits to the focal brand.
Notably, in a supplemental study, we replicate the current findings using a paradigm with Subway and Jimmy John's (Web Appendix G). We find that when Subway brings Jimmy John's into the consideration set through brand-to-brand praise, Subway gains a boost in brand attitude compared with the self-promotion condition; Jimmy John's, however, does not gain a boost from receiving the praise.
In the remaining studies, we identify when the beneficial effects of praise are mitigated and the underlying mechanism of the effect. In doing so, we offer additional understanding of the psychological processes driving consumer responses to praise as well as practical insights for choosing the appropriate recipients of and circumstances for praise.
In Study 3, we expand on the findings of the prior studies by examining the effects of brand-to-brand praise toward both a direct competitor and a noncompetitor. We expect that when a brand compliments a relevant direct competitor, which we define as a brand that competes in the same category as the focal brand, consumers view that compliment as relatively costly or risky because a brand has more to lose when bringing positive attention to such competition. It is such costliness that credibly signals that the brand must truly have warm intentions. We do not expect our effects to hold when a brand compliments an irrelevant noncompetitor, which we define as a brand that does not compete in the same category as the focal brand. In this scenario, the brand incurs lower cost by bringing attention to the irrelevant noncompetitor because there are less obvious repercussions, and thus the compliment is a less credible signal of a brand's warmth. This study also aims to demonstrate that the effects of brand-to-brand praise are not driven solely by the perceived novelty of the message or by negative reactions to self-promotional messages.
Three hundred ninety-nine participants (50.4% female) recruited from Amazon Mechanical Turk took part in this four-cell between-subjects design (helpful control, self-promotion, costly praise, noncostly praise).
We created two eyeglasses brands for this study, Franklin's Frames and Lazlo's Lenses, and introduced them to participants as competitors. Participants then viewed an ostensibly recent series of three tweets from the focal brand, Franklin's Frames—one manipulated tweet that varied by condition and two filler tweets. The manipulated tweet in the praise condition read, "@lazloslenses, Wow! Your new frames are looking good!" with an image of a pair of glasses. In the self-promotion condition, the manipulated tweet read, "Wow! Our new frames are looking good!" with an image of a pair of glasses identical to the praise condition. In the helpful control condition, the manipulated tweet read, "How to clean your frames:" with a screen shot from a video showing how to clean eyeglasses. Participants in the noncostly praise condition saw a burger brand that is clearly not a direct competitor to Franklin's Frames, called Ben's Burgers. The focal tweet for those in the noncostly praise condition said, "@bensburgers, Wow! Your new burgers are looking good!" with an image of a burger attached (Web Appendix H).
After reading the scenario, participants completed the measures of brand attitude (α = .95) noted in Study 2. In addition, we measured the perceived costliness of the tweets with four items on seven-point scales (α = .75; Web Appendix C). Finally, participants answered the same believability question as in prior studies as well as a measure of perceived novelty ("How novel was this tweet by Franklin's Frames?"; 1 = "not novel," and 7 = "very novel") of the focal tweets.
A one-way analysis of covariance (ANCOVA) showed significant differences across conditions on perceived costliness of the focal tweet (F( 3, 394) = 37.66, p < .001, = .22).[12] As we expected, contrasts revealed that perceived costliness was significantly greater for the costly praise condition (M = 3.82) than the self-promotion (M = 2.21, p < .001), control (M = 2.37, p < .001), and noncostly praise (M = 2.52, p < .001) conditions. Perceived costliness for the noncostly praise condition was greater than the self-promotion condition (p = .05) but not significantly different from the control condition (p = .34). Lastly, the self-promotion and control conditions did not differ (p = .32). Notably, we measure perceived costliness of the stimuli in all of our studies and find the same pattern of results; praise toward a direct competitor is perceived as more costly than other types of messages (Web Appendix C).
A one-way ANCOVA revealed an effect of condition on brand attitude toward the focal brand, Franklin's Frames (F( 3, 394) = 4.83, p = .003, = .04). Contrasts revealed that brand attitude was significantly greater for the costly praise condition (M = 6.05) than the control (M = 5.43, p < .001) and self-promotion (M = 5.46, p < .001) conditions. Brand attitudes were also significantly greater for costly praise than noncostly praise (M = 5.65, p = .02). Brand attitude for those in the control, self-promotion, and noncostly praise conditions were not significantly different (all ps > .18), suggesting that the results are driven by a boost from the praise message and not a dislike of the self-promotion message.
Unsurprisingly, a one-way ANCOVA revealed significant differences across conditions on how novel the tweet seemed (F( 3, 394) = 8.80, p < .001, = .06). Contrasts revealed that both the noncostly praise (M = 4.30) and the costly praise (M = 4.56) conditions were perceived to be more novel than the control (M = 3.48) and self-promotion (M = 3.43, all ps ≤ .001) conditions. Crucially, however, there was no difference in perceived novelty between the noncostly praise and the costly praise conditions (p = .31), suggesting that the differences between these conditions on brand attitude were not driven purely by differences in novelty.[13]
In Study 3, we replicate the findings of prior studies, demonstrating that brand-to-brand praise between competitors boosts brand evaluations compared with other messages, including self-promotion and a helpful control message. Furthermore, we find that noncostly praise (i.e., praise toward irrelevant noncompetitors) did not give the same boost. This result suggests that simply speaking positively about another brand is not enough; the consumer must view the compliment as costly to the brand for it to have positive effects. Although we find costliness to be necessary to show the benefits of praise, it is not the underlying driver of the effects (mediation comparing the self-promotion and costly praise conditions: ab = .06, 95% confidence interval [CI] = [−.045,.191]; PROCESS Model 4, 5,000 bootstrap samples; [31]). This study also begins to cast doubt on novelty as a primary driver of the effect, given that the noncostly praise was perceived to be as novel (but not evaluated as positively) as the costly praise. We further test the role of novelty in subsequent studies. In addition, we again find that the effects on brand attitude do not occur because consumers dislike self-promotion messages but are instead driven by the boost from seeing the costly praise message, replicating the findings of Studies 1a and 2.
To begin to understand why brand-to-brand praise increases brand evaluations and choice, we conducted a qualitative pretest gauging people's natural reactions (i.e., inferences, attributions) to observing a brand complimenting a competitor. Participants (N = 150) on Prolific saw a tweet from PlayStation congratulating Nintendo on its Nintendo Switch launch and then listed five adjectives to describe PlayStation. We find that observers more frequently attribute warmth-related words (e.g., friendly, supportive, kind; 55.73% of the adjectives), rather than competence-related words (e.g., confident, successful, intelligent; 17.07% of the adjectives), or any other kind of perception, to the brand, implicating warmth as the most top-of-mind inference following brand-to-brand praise. For additional insight, we asked participants to take the perspective of a manager and indicate why they would or would not post a message praising the competition. We find that most participants would be willing to praise a competitor (69.33%). Again, the majority of the responses (61.5%) indicated warmth to be the primary reason for the action (e.g., "I want to show others that I act in selfless ways"). Moreover, some responses also positively noted competence-related reasons (e.g., "have confidence in my own brand and its qualities"; "we appear to be positive and not insecure about our place in the market") either exclusively or in combination with warmth (25%), suggesting that while warmth-related reasons are more top-of-mind, perceptions of competence are unlikely to be harmed (for pretest details, see Web Appendix I).
In light of these initial qualitative insights, we empirically test warmth as the driver of the benefits of brand-to-brand praise and compare it with competence in Study 4. We predict that brand-to-brand praise enhances perceptions of brand warmth, which predominantly drives the boost in brand evaluations, rather than competence. Ultimately, we expect the heightened brand evaluations to drive consumer action, which is measured in this study as their willingness to sacrifice their own time (for free) on behalf of the brand.
Two hundred participants (57% female) from Prolific took part in this two-cell between-subjects study (control, praise).
Participants were introduced to a real tea brand, Treecup Tea, from Kickstarter. They were told that Treecup Tea focuses on making high-quality tea blends and competes with other tea companies on Kickstarter for funds. Participants read that Treecup Tea's competitors include Teafir, Shisso Tea, and Phat Tea. Next, participants were introduced to Treecup Tea's Twitter page. For the control condition, participants saw only the brand's Twitter header. We use this control condition, different from the self-promotion control in prior studies, to show again that the boost for the brand results from people feeling positive about the praise message instead of negative toward self-promotion. For the praise condition, participants saw a Twitter page that consisted of three praise messages toward the three competitors interspersed throughout other tweets on the page (e.g., "@ShissoTea Your tea is fresh and sustainable! That's amazing!"; Web Appendix J).
Participants then completed the same measure of brand attitude as in prior studies. Participants were also asked how much time (scale from 0 to 4 minutes, in 30-second increments) they were willing to volunteer to answer some market research questions for Treecup Tea after the main study, without additional pay. We chose this dependent variable as an action of consequence to the consumer, given that volunteering time is a costly consumer behavior. Finally, participants completed measures for warmth ("warm, friendly" on a seven-point scale; r = .81) and competence ("competent, capable" on a seven-point scale; r = .90; [ 1]). The order of the warmth and competence measures was randomized to ensure that one did not explain the effect more than the other simply due to order. We again measured believability of the stimuli as a covariate in our analyses.
Replicating prior results, a one-way ANCOVA revealed a significant effect of message (F( 1, 197) = 17.61, p < .001, = .08), whereby praise (M = 5.86) led to greater brand attitude compared with the control (M = 5.33).
The data were skewed because 34.1% of the sample indicated that they would not volunteer any time,[14] so we conducted a binary split on volunteer time (1 = those who would volunteer any amount of time, 0 = those who would not volunteer). A binary logistic regression with condition (1 = praise, 0 = self-promotion) and believability as the predictors on willingness to volunteer revealed that those in the praise condition were significantly more likely to volunteer time compared with those in the self-promotion condition (β = .59, SE = .30, Wald χ2( 1, N = 200) = 3.76, p = .05).[15]
A one-way ANCOVA revealed a significant effect of message on warmth (F( 1, 197) = 22.54, p < .001, = .10): praise (M = 5.92) led to greater perceptions of warmth compared with the control (M = 5.20). A one-way ANCOVA also revealed a smaller but significant effect of message on competence (F( 1, 197) = 5.59, p = .02, = .03): praise (M = 5.49) led to greater perceptions of competence compared with the control (M = 5.17).
Next, we tested the effect of praise on willingness to volunteer through warmth and brand attitude as serial mediators. We find that the praise message leads to increased warmth perceptions, which leads to improved brand attitude and, subsequently, greater willingness to volunteer (indirect = .08, 95% CI = [.008,.228]). However, a serial mediation model replacing warmth with competence is not significant (indirect = .05, 95% CI = [−.005,.169]).
In Study 4, we replicate prior findings and also demonstrate a novel downstream consequence of praise whereby observers are more likely to give up some of their time, without additional compensation, to help the praiser brand after viewing a praise message. We find warmth to be the most direct driver of the effects of brand-to-brand praise. However, given that competence perceptions also benefit from a praise message, it is worth considering how brand-to-brand praise may usher brands into the "golden quadrant" ([32]) of warmth and competence dimensions. Lastly, the stimuli used for the praise message in this study consisted of three compliments to three different competitors on a single Twitter page. The fact that we still see a boost in brand attitude suggests that giving praise repeatedly, at least to some extent, may not harm the brand in this context (an idea we revisit in the "General Discussion" section).
In Study 5, we again test for warmth as the mechanism underlying the effect of brand-to-brand praise on evaluations, and we do so with process by moderation, directly manipulating (and again also measuring) brand warmth. Compared with brands that are lower in warmth, we predict that brand-to-brand praise will not increase brand evaluations as much for brands that are higher in warmth. We theorize that this is because brands that are already high in perceived warmth will not have as much need or space to grow in that aspect, thus rendering praise less influential. In contrast, brands with lower perceived warmth at baseline will benefit more from the boost given by brand-to-brand praise. Based on prior research demonstrating that nonprofit organizations are high in perceived warmth ([ 2]), we compare the effects of praise on nonprofit and for-profit organizations. Beyond its relation to our theory, comparing the effect of praise in for-profit versus nonprofit organizations is of practical importance, as it will help managers from these clearly identifiable sectors better assess how effective brand-to-brand praise may be for their brands. Lastly, we measure perceptions of the brand's arrogance as an alternative explanation, given that self-promotional messages may be perceived as a form of bragging. We also measure perceptions of authenticity and novelty of the message as other potential explanations for the effect.
Six hundred one participants (62.2% female) from Prolific took part in this 2 (message: control, praise) × 2 (organization type: nonprofit, for-profit) between-subjects experimental study that was preregistered.[16]
Participants were introduced to a fictitious internet service organization, Tech Dev. In the nonprofit condition, participants were told that Tech Dev was a nonprofit organization that had a goal of helping the community. In the for-profit condition, participants read that Tech Dev was a for-profit organization with a goal of increasing profits. Participants were also told that Tech Dev competed with another organization, Networks.org or Networks.com, for either donations or sales (depending on its nonprofit or for-profit status, respectively). Next, participants were shown the Twitter page of Tech Dev in which they read either a control message ("Want to know more about the quality of our services? Click here: techdev[.com/.org]/internet") or a message praising Networks ("Networks[.com/.org], we are impressed by the quality of your services—competitor or not!"; Web Appendix K).
Participants then indicated their interest in Tech Dev using two measures ("How likely are you to seek out more information about Tech Dev?" and "How willing are you to talk to a customer representative to learn more about Tech Dev?"; r = .79).[17] We also measured warmth, competence, and believability of the stimuli. As in the prior study, the order of warmth and competence was randomized. In addition, we measured the extent to which participants' assigned condition was perceived as arrogant (braggy, conceited, arrogant; α = .94), authentic (authentic, self-aware; r = .65), and novel (single-item measure) as alternative explanations,.
Using a two-way ANCOVA, we find a main effect of message (F( 1, 596) = 24.51, p < .001, = .04), a main effect of organization type (F( 1, 596) = 35.82, p < .001, = .06), and a significant interaction (F( 1, 596) = 4.79, p = .03, = .008). As predicted, in the for-profit conditions, we replicate prior results in that praise (M = 3.74) led to greater brand attitude than the control did (M = 2.78; p < .001). In the nonprofit conditions, praise (M = 4.25) also led to greater brand evaluations than the control did (M = 3.86; p = .04), but to a diminished extent.
We find a similar pattern for warmth. A two-way ANCOVA revealed a main effect of message (F( 1, 596) = 98.85, p < .001, = .14), a main effect of organization type (F( 1, 596) = 124.43, p < .001, = .17), and a significant interaction (F( 1, 596) = 8.44, p = .004, = .01). In the for-profit conditions, we again replicate prior results where praise (M = 4.78) led to significantly greater warmth compared with the control (M = 3.37; p < .001). In the nonprofit conditions, we find that praise (M = 5.67) led to weaker, though still significant, effects compared with the control (M = 4.88; p < .001).
We also find a similar pattern for competence. A two-way ANCOVA revealed a main effect of message (F( 1, 596) = 26.90, p < .001, = .04), a main effect of organization type (F( 1, 596) = 23.33, p < .001, = .04), and a significant interaction (F( 1, 596) = 4.49, p = .03, = .007). In the for-profit conditions, we find that praise (M = 5.22) led to greater competence compared with the control (M = 4.53; p < .001). In the nonprofit conditions, we find that praise (M = 5.48) led to weaker, though still significant, effects compared with the control (M = 5.18; p = .02).
We do not find the same pattern of results for bragging. A two-way ANCOVA revealed only a main effect of organization (F( 1, 596) = 55.77, p < .001, = .09) such that the nonprofit (M = 2.08) was perceived as less arrogant than the for-profit (M = 2.93).
For authenticity, we find a similar pattern to that of warmth in which there was a significant interaction (F( 1, 596) = 16.33, p < .001, = .03) where the boost from praise was stronger in the for-profit conditions. We find similar patterns for novelty (interaction F( 1, 596) = 6.58, p = .01, = .01).
We first tested the predicted model. Specifically, we tested for moderated mediation (PROCESS Model 7, 5000 bootstrap samples, [31]) for the effect of praise on brand attitude through warmth, moderated by organization type, controlling for believability. As expected, we find that organization type significantly moderated the mediation through warmth (index of moderated mediation = .36, 95% CI = [.119,.609]), and the mediation effect is stronger in the for-profit conditions (ab = .82, 95% CI = [.624, 1.039]) compared with the nonprofit conditions (ab = .46, 95% CI = [.299,.632]).
Then, to explore the role of alternative explanations, we entered all the potential mediators (warmth, competence, bragging, authenticity, and novelty) in parallel into the moderated mediation model. We find that warmth remains a significant mediator in the model (index of moderated mediation = .23, 95% CI = [.079,.408]), suggesting that none of these alternatives "swamp" warmth in explaining this effect. Next, we find that the indices of moderated mediation for competence (index of moderated mediation = .08, 95% CI = [.005,.180]), authenticity (index of moderated mediation = .13, 95% CI = [.022,.271]), and novelty (index of moderated mediation = .12, 95% CI = [.025,.231]) were also significant. However, the index for bragging was not significant (index of moderated mediation = −.02, 95% CI = [−.083,.038]). Although competence, authenticity, and novelty were also significant mediators in the model in addition to warmth, warmth remains significant with the largest index of moderated mediation.
Because warmth is most frequently compared with competence, we conducted a final analysis statistically comparing the sizes of their indirect effects in a parallel mediation model. We do this within the for-profit conditions in which the effects were more prominent (and because we are unable to statistically compare the full moderated mediation models). We find that the indirect effect for warmth as a mediator was significantly greater than that of competence (p < .001, 95% CI = [.205,.718]). Thus, though other factors may play a role, our results point to warmth as the primary driver of the effects.
Study 5 sheds further light on warmth as the primary mechanism underlying brand-to-brand praise. We demonstrate that brand-to-brand praise increases perceived warmth, which enhances brand interest, but this effect is attenuated when the brand is already high in perceived warmth (e.g., nonprofits), as there is less room and need for growth in warmth. Thus, the benefits of praise can be most clearly seen among brands that are perceived as less warm (e.g., for-profits) at baseline. In addition, while perceptions of competence, authenticity, and novelty may play a role in driving brand interest, we show warmth to be the more consistent, primary driver of the effect. Finally, we rule out bragging as an alternative explanation.
In Study 6, we look to another context in which the effects of brand-to-brand praise may be attenuated. Here, we examine skepticism toward advertising as an individual difference that may moderate the effects of praise. Advertising skepticism is an individual difference that has been defined as consumers' chronic doubt or mistrust in a marketer's message ([47]). Individuals who are more skeptical are generally less persuaded by marketing messages and are less trusting of brands. As brands are confronting an "age of cynicism" where skepticism is at an all-time high ([21]; [49]) and 71% of consumers report having little faith in brands ([30]), it is very important to understand its effects.
As seen in the previous study, the effects of brand-to-brand praise become more prominent when there is room for increasing perceptions of brand warmth. Consumers who are more skeptical of advertising distrust that brands have positive intentions, or warmth, and thus inherently provide brands with greater room to improve in warmth than their nonskeptical counterparts who are already trusting. Thus, brand-to-brand communication may be most effective among skeptics, as it can bypass their cynicism and boost perceptions of warmth. Moreover, such skeptics are generally more persuaded by nonadvertising sources of information and emotional appeals ([48]), which allow marketers to circumvent consumer resistance by decreasing the activation of persuasion knowledge ([27]). Because brand-to-brand praise does not explicitly aim to promote one's brand and instead relies on the more emotional signal that the brand is high in warmth, it is less likely to activate persuasion knowledge and more likely to be accepted. Thus, we predict that consumers high in skepticism will be the most affected by brand-to-brand praise. Study 6 tests this idea utilizing two well-known competitors, Lyft and Uber.
Six hundred participants were recruited from Prolific for this 2 (message: self-promotion, praise) × measured (advertising skepticism) preregistered study.[18]
Participants first completed the skepticism toward advertising scale ([47]; 1 = "strongly disagree," and 5 = "strongly agree") and then completed filler items. Next, participants saw Lyft's Twitter page, where they read either a self-promotion tweet ("Congratulations to us on all our achievements this past year!") or a praise tweet toward Uber ("@Uber Congratulations on all your achievements this past year!"; Web Appendix L). Then, participants completed the same measure of brand attitude, warmth, and competence as in previous studies. We also measured perceptions of bragging, authenticity, and novelty using the same measures as in Study 5 as potential alternative explanations, as well as believability of the stimuli as a covariate.
Replicating prior results, an ANCOVA controlling for believability revealed a main effect of message on brand attitude (F( 1, 597) = 22.12, p < .001, = .04), where praise (M = 5.22) outperformed self-promotion (M = 4.74). Next, we find a significant interaction between message and skepticism (skepticism scale reverse coded; β = .21, SE = .10, p = .03; Figure 2) on brand attitude, such that when skepticism toward advertising was higher, the praise message led to a greater increase in brand attitude (Johnson–Neyman point = 1.48, 75.67% of participants; β = .23, SE = .12, p = .05). The effect of message was attenuated at lower levels of skepticism below the Johnson–Neyman point.
Graph: Figure 2. Study 6: Moderation by skepticism.
We find that praise significantly boosted perceptions of warmth (Mpraise = 5.36 vs. Mself = 4.99; F( 1, 597) = 13.76, p < .001, = .02), competence (Mpraise = 5.52 vs. Mself = 5.32; F( 1, 597) = 4.88, p = .03, = .008), authenticity (Mpraise = 4.90 vs. Mself = 4.56; F( 1, 597) = 9.46, p = .002, = .02), and novelty (Mpraise = 4.58 vs. Mself = 4.28; F( 1, 597) = 6.59, p = .01, = .01) compared with self-promotion, controlling for believability. Self-promotion was seen as more arrogant than praise (Mpraise = 3.18 vs. Mself = 4.16; F( 1, 597) = 54.23, p < .001, = .08).
We then tested for the predicted moderated mediation (PROCESS Model 7, 5000 bootstrap samples, [31]) for the effect of praise on brand attitude through warmth, moderated by skepticism, controlling for believability. While the index of moderated mediation was not significant (index = .10, 95% CI = [−.046,.247]), we find that the indirect effect patterns were in line with our predictions, such that the indirect effect is stronger and significant for those higher in skepticism (+1 SD = 3.17, ab = .31, 95% CI = [.084,.550]) and nonsignificant for those lower in skepticism (−1 SD = 1.30, ab = .12, 95% CI = [−.026,.274]).
Next, we test for mediation with all of the potential alternative explanations in parallel (PROCESS Model 4, 5,000 bootstrap samples, [31]). Warmth (ab = .14, 95% CI = [.064,.230]), competence (ab = .04, 95% CI = [.004,.091]), bragging (ab = .09, 95% CI = [.049,.146]), authenticity (ab = .06, 95% CI = [.021,.116]), and novelty (ab = .03, 95% CI = [.004,.054]) mediate the effect of message on brand attitude. However, we again find that the indirect effect for warmth was the largest, pointing to its primary role in causing this effect.
Finally, we statistically compared the indirect effects of warmth and competence in the model and find the indirect effect of warmth to be significantly greater than that of competence (p = .001, 95% CI = [.085,.349]), again suggesting that warmth plays the primary role in driving the effect of brand-to-brand praise on brand attitudes.
In Study 6 (and in a behavioral replication in Web Appendix M),[19] we show that brand-to-brand praise can actually operate more effectively for people who are generally more skeptical of advertising, as brand-to-brand praise bypasses their suspicions and creates more favorable consequences for the brand. Further, while we find that competence, bragging, authenticity, and novelty play a role in driving the effects of brand-to-brand praise on brand evaluations, we still identify warmth as the primary, more consistent underlying driver.
We investigate how observing brand-to-brand praise affects consumers' brand evaluations and choices. Across a variety of different modes of communication (social media, print advertising, digital advertising), study methods (web scraping; field, lab, and online studies), contexts (consumer products and services), outcomes (brand attitudes, social media and advertising engagement, brand choice, purchase behavior), and praise content, we show that consumers who witness brand-to-brand praise between competitors form more favorable evaluations of the praiser brand than consumers who witness other forms of communication, including typical self-promotion messages, helpful messages, basic brand information, and even outside-industry praise (for a summary of contexts, conditions, and findings across studies, see Web Appendix N). In addition to showing robustness to a variety of praise messages in the presented studies, a preregistered supplemental study demonstrates that the effect is further robust to both general and specific praise (Web Appendix O). Furthermore, the varied study stimuli suggest that this effect is robust to brands in a wide range of industries—including car care, snacks, candy, eyewear, beverages, technology, and transportation—with competitive relationships varying in intensity. In other words, brand-to-brand praise seems to benefit the praiser in less competitively intense relationships (e.g., mom-and-pop popcorn brands) as well as more competitively intense relationships (e.g., PlayStation and Nintendo, Uber and Lyft) in a variety of industries. We trace this effect primarily to the notion that brand-to-brand praise signals a brand's warmth, which leads to improved brand evaluations and affects consumer choices.
In addition, we show that this effect only exists when the praise is deemed to be associated with significant cost or risk (Study 3). Importantly, we find that the effects of brand-to-brand praise are diminished in some situations or among some consumers, such as when the brand is already high in warmth (Study 5) or among consumers who are already trusting of brand intent (Study 6). These boundaries provide further evidence for the crucial role that perceptions of warmth play in driving the benefits of praise. While other mechanisms may also play a role, as consumer behaviors are generally multiply determined ([35]; [55]), we find warmth to be the most consistent, primary driver of the effects of brand-to-brand praise.
Our research makes several theoretical contributions to literature streams on brand perceptions and relationships, brand communication, and praise. First, we contribute to the brand perception literature by showing that consumers' perceptions of brands can be affected by viewing a brand's interactions with other brands. We demonstrate that observing brand-to-brand praise can positively affect perceptions of a brand's warmth and influence subsequent brand evaluations. Second, we add to the warmth and competence literature by introducing a novel context in which brand-to-brand praise increases warmth without harming perceptions of competence. Third, our research demonstrates that directing positive attention toward the competition instead of toward one's own brand brings about benefits for the praiser brand, unlike what brands would typically do in comparative advertising and two-sided messaging campaigns (e.g., [ 4]; [17]; [40]). Finally, we contribute to the literature on praise by identifying brand-to-brand communication between competitors as a feasible form of praise that is less likely to induce suspicion compared with praise in traditional person-to-person contexts.
As we have noted, our findings may be surprising to practitioners who have been regularly and reasonably advised to avoid bringing positive attention to their competitors. However, our studies show that in some circumstances, praise is a method of brand-to-brand interaction that can result in beneficial consequences for the praising brand. Managers might consider offering compliments to competitors to boost their own brand evaluations. In other words, brands can expand from solely focusing on brand-to-consumer relationships to also focusing on their brand-to-brand relationships. This is akin to the positive reactions that politicians sometimes receive when positively acknowledging their opponent ([12]). We suggest a new context in which positive acknowledgment can benefit brands. With the rise of the digital age, brands can easily "speak" with each other and be observed by consumers. While it is not uncommon for brands to speak via "feuds" on social media, such as when Wendy's teases McDonald's for using frozen beef ([19]), we show in a supplemental study (Web Appendix P) that positive communication provides unique advantages. Negative or snarky communication, directed at a competitor or even directed at the self (as in the case of two-sided messaging), does not provide the same increase in brand evaluations. Although there will likely be variation depending on the exact content and cleverness of snarky communication (a ripe area for future research), marketers would be wise to consider opportunities for brand-to-brand praise instead, perhaps utilizing social media as a platform, to foster a warmer image.
Brand-to-brand praise may also be a valuable way to respond to competitors' actions. While the norm for companies responding to a competitor's new product release is to avoid saying anything that would bring attention to that competitor, our studies suggest that responding positively can increase purchases for the praiser brand without boosting the competitor to the same degree. Importantly, while prior work has shown that a brand's positioning as an underdog or market leader has important implications ([50]), brand-to-brand praise may be appropriate regardless of the competitor's market status. In a preregistered supplemental study (Web Appendix Q), we find that the favorable effects of brand-to-brand praise on brand evaluations are robust to the market leadership status of the brand. Future research could further explore how characteristics about the firm such as market leadership or firm size may factor into the effect of brand-to-brand praise.
Similarly, while the variety of brands leveraged in our studies suggest that brand-to-brand praise is likely beneficial when directed toward both less intense (e.g., local popcorn brands) and more intense (e.g., PlayStation and Nintendo, Uber and Lyft) competitors, future research should more systematically explore the role of competitive intensity. Such research might explore brand-to-brand praise with indirect versus direct competitors, which would likely vary in levels of perceived costliness. Indirect competitors may include other brands that are in a similar industry but do not compete directly in the same product category, such as in the case of soda and water in the drinks industry. In Study 3, we find that praise toward a direct competitor is deemed to be costly and results in better brand evaluations compared with praise toward noncompetitors. However, would praise toward an indirect competitor be perceived to be costly enough to increase evaluations? Outside of direct competition, what else determines whether praise is deemed to be costly or not? How would consumers react if complementary brands, such as soda and popcorn brands, praised each other? While we suggest that praise should benefit the praiser as long as it is perceived as costly, understanding what kind of brand-to-brand praise is considered costly may be beneficial to marketers.
In addition, future research could also explore how brand-to-brand praise compares to other types of messages that convey warmth. For example, brands often communicate prosocial messages, such as a food brand showing active support for a local food bank. Brands might also post self-deprecating messages that recognize their own room for improvement. While these types of messages may signal warmth, they may not benefit the messenger brand in the same way as brand-to-brand praise does because of the unique aspect of costliness associated with praising a direct competitor. Future research could compare these various types of brand messaging to find ones that benefit brands the most.
Future work might also examine whether certain brand personalities benefit more from praise than others. Our for-profit versus nonprofit results in Study 5 hint that brands with personalities that are inherently associated with warmth (e.g., sincere brands) may benefit less. Relatedly, are there circumstances under which snarky interactions or backhanded compliments are better suited for some brands, such as Wendy's ([15])? We noted that negativity did not fare as well as positivity (Web Appendix P), but under what conditions might this be different? Cultural differences, such as collectivism and individualism, may also play an important role. Collectivists, known for valuing community and kinship, may be particularly likely to value the warmth associated with brand-to-brand praise ([ 3]). Furthermore, future work could also explore when the effects of brand-to-brand praise are affected by gender. While women may value warmth more in some circumstances ([66]), we do not find consistent moderation by gender in our studies (Web Appendix R), which is consistent with prior work showing that gender does not always moderate brand warmth perceptions ([ 5]). Further research is needed to better understand the possible role that factors such as brand personality, consumer culture, and gender play in affecting brand-to-brand praise responses.
Another direction for future research involves investigating the optimal frequency of brand-to-brand praise. Although we demonstrate that some repetition may be acceptable (Study 4), one limitation of this research is the focus on a single instance of praise in a short span of time. Praise when repeated over time may become less effective or even backfire. Excessive praise may trigger suspicion in observers (e.g., [23]), and previous research has shown that suspicious praise can lead to the praiser being perceived as less sincere ([11]; [44]). Thus, future research might explore the optimal levels or repetition of praise and the consequences of excessive praise to prevent such a backfiring effect.
To further prevent backfiring effects, future research might also explore what happens if the praised competitor leverages the praise "against" the praiser? For instance, what if a praised brand publicly suggests that they are so great that even their competition praises them? It is possible that, if used by the competitor in this manner, the praise could damage the praiser brand. However, it is also possible that this kind of act could be perceived as arrogant or manipulative, damaging the praised brand and eliciting sympathy for the praiser.
Overall, this work raises the question: Are brands missing an opportunity to build positive relationships with consumers by not (publicly) building positive relationships with competitors? We suggest that marketers would be wise to explore the intriguing benefits of brand-to-brand praise.
sj-pdf-1-jmx-10.1177_00222429211053002 - Supplemental material for Befriending the Enemy: The Effects of Observing Brand-to-Brand Praise on Consumer Evaluations and Choices
Supplemental material, sj-pdf-1-jmx-10.1177_00222429211053002 for Befriending the Enemy: The Effects of Observing Brand-to-Brand Praise on Consumer Evaluations and Choices by Lingrui Zhou, Katherine M. Du and Keisha M. Cutright in Journal of Marketing
Footnotes 1 Deborah J. MacInnis
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was funded by the Fuqua School of Business at Duke University and the Lubar School of Business at the University of Wisconsin–Milwaukee.
4 Lingrui Zhou https://orcid.org/0000-0003-2539-2234
5 Other labels for warmth include communality, morality, and social good/bad; other labels for competence include agency, instrumentality, and intellectual good/bad. For a discussion of the different labels used to reflect warmth and competence, see Cuddy, Glick, and Beninger (2011).
6 See https://aspredicted.org/k46e2.pdf.
7 In this study, we also took measures of relational-independence self-construal and brand trust as proxies for brand warmth. The items sequentially mediated the effect of praise on choice. Web Appendix D provides details.
8 As part of a separate study (Study 6), we found that this measure of believability indeed measures the degree to which the messages seem real. The data do not suggest that this covariate measure represents skepticism that the praise is genuine. For more details, see Web Appendix E.
9 As a summary, of the studies in this article that utilize the believability covariate, five of the six—Studies 2, 3, 4, 5, and 6—do not require believability to produce a significant (p < .05; Studies 2, 4, 5, and 6) or marginally significant (p < .10; Study 3) positive effect of brand-to-brand praise.
The focal brand (The Mad Popper) captured lower consumer preference than the competitor (Carolina Popcorn Shoppe) across both conditions in this study. This may be due to Carolina Popcorn Shoppe being shown first across conditions, or because Carolina Popcorn Shoppe had a name that resonated more with the local community. Importantly, however, the focal brand still captured greater preference in the praise condition than the self-promotion condition. This might lead one to question whether underdogs are more likely to benefit from offering competitive praise. We find that market leadership status does not affect reactions to brand praise differentially (see the supplemental study in Web Appendix Q).
See https://aspredicted.org/cb7nq.pdf.
In the current study and in all subsequent ANCOVAs, we control for the believability of the message. Again, details related to this measure can be found in Web Appendix E.
Adding novelty as a control variable does not alter the results of this study, such that costly praise continues to increase brand attitude compared with the other conditions (ps < .05). In addition, although novelty does mediate the effect of message (comparing the self-promotion and costly praise conditions; ab = .08, 95% confidence interval [CI] = [.024,.159]) on brand attitude in this study, we further test the mediating role of novelty compared with warmth in subsequent studies.
Of the 34.1% (71 participants) who indicated that they would not volunteer any time, 40.8% (29 participants) were from the praise condition, and 59.2% (42 participants) were from the control condition.
When keeping volunteer time as a continuous dependent variable, we find directional results consistent with the predicted pattern (Mpraise = 2.84 vs. Mcontrol = 2.23; F(1, 197) = 2.36, p = .13).
See https://aspredicted.org/44gm6.pdf.
We indicated that these two measures would be two separate dependent variables in the preregistration, but for brevity (given their high correlation), we report results combining the two measures. Results hold for the individual measures as dependent variables. For more details, see Web Appendix K.
See https://aspredicted.org/ud2xe.pdf.
We replicate this moderation by chronic skepticism using real behavior, assessing whether consumers went to pick up a New York Times newspaper from a university library after seeing it praise its competitor or not.
References Aaker Jennifer L. , Garbinsky Emily N. , Vohs Kathleen D.. (2012), " Cultivating Admiration in Brands: Warmth, Competence, and Landing in the 'Golden Quadrant' ," Journal of Consumer Psychology , 22 (2), 191 – 94.
Aaker Jennifer , Vohs Kathleen D. , Mogilner Cassie. (2010), " Nonprofits Are Seen as Warm and For-Profits as Competent: Firm Stereotypes Matter ," Journal of Consumer Research , 37 (2), 224 – 37.
Abele Andrea E. , Wojciszke Bogdan. (2007), " Agency and Communion from the Perspective of Self Versus Others ," Journal of Personality and Social Psychology , 93 (5), 751 – 63.
Barry Thomas E. (1993), " Comparative Advertising: What Have We Learned in Two Decades? " Journal of Advertising Research , 33 (2), 19 – 30.
Bennett Aronté Marie , Hill Ronald Paul. (2012), " The Universality of Warmth and Competence: A Response to Brands as Intentional Agents ," Journal of Consumer Psychology , 22 (2), 199 – 204.
Bennett Aronté Marie , Hill Ronald Paul , Oleksiuk Daniel. (2013), " The Impact of Disparate Levels of Marketplace Inclusion on Consumer–Brand Relationships ," Journal of Public Policy & Marketing , 32 (1), 16 – 31.
Bernritter Stefan F. , Verlegh Peeter W.J. , Smit Edith G.. (2016), " Why Nonprofits Are Easier to Endorse on Social Media: The Roles of Warmth and Brand Symbolism ," Journal of Interactive Marketing , 33 (1) , 27 – 42.
Bolton Lisa E. , Mattila Anna S.. (2015), " How Does Corporate Social Responsibility Affect Consumer Response to Service Failure in Buyer–Seller Relationships? " Journal of Retailing , 91 (1), 140 – 53.
Bradberry Travis. (2015), " 12 Things Truly Confident People Do Differently ," Forbes (April 1), https://www.forbes.com/sites/travisbradberry/2015/04/01/12-things-truly-confident-people-do-differently/.
Bratanova Boyka , Kervyn Nicolas , Klein Olivier. (2015), " Tasteful Brands: Products of Brands Perceived to Be Warm and Competent Taste Subjectively Better ," Psychologica Belgica , 55 (2), 57 – 70.
Campbell Margaret C. , Kirmani Anna. (2000), " Consumers' Use of Persuasion Knowledge: The Effects of Accessibility and Cognitive Capacity on Perceptions of an Influence Agent ," Journal of Consumer Research , 27 (1), 69 – 83.
Cavazza Nicoletta. (2016), " When Political Candidates 'Go Positive': The Effects of Flattering the Rival in Political Communication ," Social Influence , 11 (3), 166 – 76.
Chan Elaine , Sengupta Jaideep. (2013), " Observing Flattery: A Social Comparison Perspective ," Journal of Consumer Research , 40 (4), 740 – 58.
Chen Cathy Yi , Mathur Pragya , Maheswaran Durairaj. (2014), " The Effects of Country-Related Affect on Product Evaluations ," Journal of Consumer Research , 41 (4), 1033 – 46.
Cheng Andria. (2018), " How Wendy's Learned to Stop Worrying and Love Its Twitter Roasts of McDonald's ," Forbes (October 8), https://www.forbes.com/sites/andriacheng/2018/10/08/wendys-twitter-roasts-have-become-the-envy-of-marketers-heres-how-it-does-it/#204b01ccfea4.
Cislak Aleksandra , Wojciszke Bogdan. (2008), " Agency and Communion Are Inferred from Actions Serving Interests of Self or Others ," European Journal of Social Psychology , 38 (7), 1103 – 10.
Crowley Ayn E. , Hoyer Wayne D.. (1994), " An Integrative Framework for Understanding Two-Sided Persuasion ," Journal of Consumer Research , 20 (4), 561 – 74.
Cuddy Amy J.C. , Glick Peter , Beninger Anna. (2011), " The Dynamics of Warmth and Competence Judgments, and Their Outcomes in Organizations ," Research in Organizational Behavior , 31 , 73 – 98.
Dua Tanya. (2018), " Wendy's Took a Direct Shot at McDonald's Beef in a Savage Super Bowl Ad ," Business Insider (February 4), https://www.businessinsider.com/wendys-super-bowl-commercial-mcdonalds-beef-2018-2.
Dumenco Simon. (2018), " The New York Times Wants You to Consult Other Trusted News Sources Too (but Not, Ahem, Fox News) ," Ad Age (May 3), https://adage.com/creativity/work/world-press-freedom-day-campaign-read-more-listen-more/54448.
Edelman (2021), "Edelman Trust Barometer 2021," research report (accessed November 9, 2021), https://www.edelman.com/sites/g/files/aatuss191/files/2021-01/2021-edelman-trust-barometer.pdf.
Fein Steven , Spencer Steven J.. (1997), " Prejudice as Self-Image Maintenance: Affirming the Self Through Derogating Others ," Journal of Personality and Social Psychology , 73 (1), 31 – 44.
Ferraro Rosellina , Kirmani Amna , Matherly Ted. (2013), " Look at Me! Look at Me! Conspicuous Brand Usage, Self-Brand Connection, and Dilution ," Journal of Marketing Research , 50 (4), 477 – 88.
Fiske Susan T. (2018), " Stereotype Content: Warmth and Competence Endure ," Current Directions in Psychological Science , 27 (2), 67 – 73.
Fiske Susan T. , Cuddy Amy J.C. , Glick Peter. (2007), " Universal Dimensions of Social Cognition: Warmth and Competence ," Trends in Cognitive Sciences , 11 (2), 77 – 83.
Fournier Susan. (2009), " Lessons Learned About Consumers' Relationships with Their Brands ," in Handbook of Brand Relationships , Priester J. , MacInnis D. , Park C.W. , eds. New York: N.Y. Society for Consumer Psychology and M.E. Sharp , 5 – 23.
Friestad Marian , Wright Peter. (1994), " The Persuasion Knowledge Model: How People Cope with Persuasion Attempts ," Journal of Consumer Research , 21 (1), 1 – 31.
Gershon Rachel , Cryder Cynthia. (2018), " Goods Donations Increase Charitable Credit for Low-Warmth Donors ," Journal of Consumer Research , 45 (2), 451 – 69.
Gordon Randall A. (1996), " Impact of Ingratiation on Judgments and Evaluations: A Meta-Analytic Investigation ," Journal of Personality and Social Psychology , 71 (1), 54 – 70.
Havas Group (2021), " Meaningful Brands ," research report (accessed November 9, 2021), https://s3.amazonaws.com/media.mediapost.com/uploads/MeaningfulBrands2021.pdf.
Hayes Andrew F. (2013), Introduction to Mediation, Moderation, and Conditional Process Analysis. A Regression-Based Approach. New York : Guilford.
Judd Charles M. , James-Hawkins Laurie , Yzerbyt Vincent , Kashima Yoshihisa. (2005), " Fundamental Dimensions of Social Judgment: Understanding the Relations Between Judgments of Competence and Warmth ," Journal of Personality and Social Psychology , 89 (6), 899 – 913.
Kervyn Nicolas , Bergsieker Hilary B. , Fiske Susan T.. (2012), " The Innuendo Effect: Hearing the Positive but Inferring the Negative ," Journal of Experimental Social Psychology , 48 (1), 77 – 85.
Kervyn Nicolas , Chan Emily , Malone Chris , Korpusik Adam , Ybarra Oscar. (2014), " Not All Disasters Are Equal in the Public's Eye: The Negativity Effect on Warmth in Brand Perception ," Social Cognition , 32 (3), 256 – 75.
Kirmani Amna. (2015), " Neatly Tied with a Bow ," Journal of Consumer Psychology , 25 (2), 185 – 86.
Kirmani Amna , Hamilton Rebecca W. , Thompson Debora V. , Lantzy Shannon. (2017), " Doing Well Versus Doing Good: The Differential Effect of Underdog Positioning on Moral and Competent Service Providers ," Journal of Marketing , 81 (1), 103 – 17.
Kolbl Živa , Arslanagic-Kalajdzic Maja , Diamantopoulos Adamantios. (2019), " Stereotyping Global Brands: Is Warmth More Important Than Competence? " Journal of Business Research , 104 , 614 – 21.
Koschate-Fischer Nicole , Hoyer Wayne D. , Wolframm Christiane. (2019), " What if Something Unexpected Happens to My Brand? Spillover Effects from Positive and Negative Events in a Co-Branding Partnership ," Psychology & Marketing , 36 (8), 758 – 72.
Lemmink Jos , Mattsson Jan. (1998), " Warmth During Non-Productive Retail Encounters: The Hidden Side of Productivity ," International Journal of Research in Marketing , 15 (5), 505 – 17.
Levine Philip. (1976), " Commercials That Name Competing Brands ," Journal of Advertising Research , 16 (6), 7 – 14.
Luerssen Anna , Jhita Gugan Jote , Ayduk Ozlem. (2017), " Putting Yourself on the Line: Self-Esteem and Expressing Affection in Romantic Relationships ," Personality and Social Psychology Bulletin , 43 (7), 940 – 56.
MacInnis Deborah J. , Folkes Valerie S.. (2017), " Humanizing Brands: When Brands Seem to Be Like Me, Part of Me, and in a Relationship with Me ," Journal of Consumer Psychology , 27 (3), 355 – 74.
MacInnis Deborah J. , Park C. Whan , Priester Joseph R.. (2009), " Why Brand Relationships ," in Handbook of Brand Relationships, Deborah J. MacInnis, C. Whan Park, and Joseph R. Priester, eds. New York: Taylor & Francis, ix–xxi.
Main Kelley J. , Dahl Darren W. , Darke Peter R.. (2007), " Deliberative and Automatic Bases of Suspicion: Empirical Evidence of the Sinister Attribution Error ," Journal of Consumer Psychology , 17 (1), 59 – 69.
Meyvis Tom , Van Osselaer Stijn M.J.. (2017), " Increasing the Power of Your Study by Increasing the Effect Size ," Journal of Consumer Research , 44 (5), 1157 – 73.
Nan Xiaoli , Heo Kwangjun. (2007), " Consumer Responses to Corporate Social Responsibility (CSR) Initiatives: Examining the Role of Brand-Cause Fit in Cause-Related Marketing ," Journal of Advertising , 36 (2), 63 – 74.
Obermiller Carl , Spangenberg Eric R.. (1998), " Development of a Scale to Measure Consumer Skepticism Toward Advertising ," Journal of Consumer Psychology , 7 (2), 159 – 86.
Obermiller Carl , Spangenberg Eric , MacLachlan Douglas L.. (2005), " Ad Skepticism: The Consequences of Disbelief ," Journal of Advertising , 34 (3), 7 – 17.
O'Brien Kyle. (2021), "Brands Are Facing the 'Age of Cynicism,' from Skeptical Consumers," Adweek (May 25), https://www.adweek.com/agencies/brands-are-facing-the-age-of-cynicism-from-skeptical-consumers/.
Paharia Neeru , Avery Jill , Keinan Anat. (2014), " Positioning Brands Against Large Competitors to Increase Sales ," Journal of Marketing Research , 51 (6), 647 – 56.
Patel Niel. (2015), " Your Customers Already Know About Your Competitors, and Here's How You Should Handle It ," Huffington Post (December 6), https://www.huffingtonpost.com/neil-patel/your-customers-already-know-about-your-competitors-and-heres-how-you-should-handle-it%5fb%5f8476008.html.
Peeters Guido. (2002), " From Good and Bad to Can and Must: Subjective Necessity of Acts Associated with Positively and Negatively Valued Stimuli ," European Journal of Social Psychology , 32 (1), 125 – 36.
Pereira Chris. (2017), " PlayStation, Xbox, and More Congratulate Nintendo on Switch Launch ," Gamespot (March 3), https://www.gamespot.com/articles/playstation-xbox-and-more-congratulate-nintendo-on/1100-6448411/.
Peter Christina , Ponzi Milan. (2018), " The Risk of Omitting Warmth or Competence Information in Ads: Advertising Strategies for Hedonic and Utilitarian Brand Types ," Journal of Advertising Research , 58 (4), 423 – 32.
Pham Michel Tuan. (2013), " The Seven Sins of Consumer Psychology ," Journal of Consumer Psychology , 23 (4), 411 – 23.
Reeder Glenn D. , Kumar Shamala , Hesson-McInnis Matthew S. , Trafimow David. (2002), " Inferences About the Morality of an Aggressor: The Role of Perceived Motive ," Journal of Personality and Social Psychology , 83 (4), 789 – 803.
Rindfleisch Aric , Moorman Christine. (2003), " Interfirm Cooperation and Customer Orientation ," Journal of Marketing Research , 40 (4), 421 – 36.
Scott Maura L. , Mende Martin , Bolton Lisa E.. (2013), " Judging the Book by Its Cover? How Consumers Decode Conspicuous Consumption Cues in Buyer–Seller Relationships ," Journal of Marketing Research , 50 (3), 334 – 47.
Spence Michael. (1973), " Job Market Signaling ," Quarterly Journal of Economics , 87 (3), 355 – 74.
Vonk Roos. (2002), " Self-Serving Interpretations of Flattery: Why Ingratiation Works ," Journal of Personality and Social Psychology , 82 (4), 515 – 26.
Wang Ze , Mao Huifang , Li Yexin Jessica , Liu Fan. (2017), " Smile Big or Not? Effects of Smile Intensity on Perceptions of Warmth and Competence ," Journal of Consumer Research , 43 (5), 787 – 805.
Weissman Saya. (2013), " KitKat, Oreo Do Real-Time Twitter Banter ," Digiday (March 22), https://digiday.com/marketing/kitkat-oreo-do-real-time-twitter-banter/.
White Diana. (2014), " 7 Common Habits of Confident People ," Womanitely (August 21), https://womanitely.com/common-habits-confident-people/.
Wojciszke Bogdan , Bazinska Roza , Jaworski Marcin. (1998), " On the Dominance of Moral Categories in Impression Formation ," Personality and Social Psychology Bulletin , 24 (12), 1251 – 263.
Wu Jintao , Chen Junsong , Dou Wenyu. (2017), " The Internet of Things and Interaction Style: The Effect of Smart Interaction on Brand Attachment ," Journal of Marketing Management , 33 (1/2), 61 – 75.
Xue Jianping , Zhou Zhimin , Zhang Liangbo , Majeed Salman. (2020), " Do Brand Competence and Warmth Always Influence Purchase Intention? The Moderating Role of Gender ," Frontiers in Psychology , 11 , 248.
Zahavi Amotz , Zahavi Avishag. (1999), The Handicap Principle: A Missing Piece of Darwin's Puzzle. Oxford, UK: Oxford University Press.
~~~~~~~~
By Lingrui Zhou; Katherine M. Du and Keisha M. Cutright
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 12- Better Marketing for a Better World. By: Chandy, Rajesh K.; Johar, Gita Venkataramani; Moorman, Christine; Roberts, John H. Journal of Marketing. May2021, Vol. 85 Issue 3, p1-9. 9p. 1 Diagram. DOI: 10.1177/00222429211003690.
- Database:
- Business Source Complete
Better Marketing for a Better World
Consider the following quotes:
The marketing discipline today constitutes a great paradox. The nation stands deeply troubled. It seeks solutions to grave problems both within and without its own society. Marketing and marketers are an integral part of this picture, either as a dimension of the problems or as a source of their solutions. Yet the emphasis of marketing study is not directed toward resolving issues of its social relevance, and there is strong and vocal sentiment in the field against being pulled in this direction...Relevancy is to be judged in the context of the true life and death issues which currently exist, such as war, poverty, racism, contamination of the environment, loss of self-identity, and the alienation of youth. Certainly, it is an appropriate time for marketers to reflect upon the relevancy of the marketing discipline in such a context.
Profits will continue to be essential and basic to corporate survival, but the major challenge to business today may be to meet the societal needs of a changing environment.
These quotes reflect concerns shared by many marketing academics and practitioners today about marketing's role in creating a better world. Yet they also reflect opportunities lost. These quotes appeared almost exactly 50 years ago in the Journal of Marketing special issue on "Marketing's Changing Social/Environmental Role," published in July 1971 (first quote: [17], p. 68; second quote: [22], p. 1). Reading that issue today, one is struck by the sense of hope represented in those scholars' assessments of the gaps between the topics studied in contemporary research and the opportunities and obligations associated with contemporary society. An awareness of these gaps, the logic seemed to go, should yield interest and pressure on academics to fill them.
How has today's scholarly community fared in its pursuit of "better world" topics? How well have we lived up to the hopes of 50 years ago and the current imperatives? Some answers to these questions are evident in the responses we received from a survey conducted in February 2021 among JM's associate editors and advisory board members. We asked these scholars for their views about research on "Better Marketing for a Better World" (BMBW). By BMBW, we mean the use of marketing activities and ideas to impact outcomes beyond just what is good for the financial performance of firms: BMBW emphasizes marketing's role in enhancing the welfare of the world's other stakeholders and institutions. To our first question, "How important is the topic of BMBW to the field of marketing?" the mean response from these 44 scholars was 6.34 on a 7-point scale. More than 60% gave the highest score in response to this question. However, when asked "To what degree has the field addressed BMBW topics?" and "How effectively do you think the field has addressed BMBW topics?" over 80% rated the current status of the field as 4 or below. While this is a select sample, our discussions with many other scholars point to the same conclusion.
This BMBW special issue is motivated by the gap that remains between what is studied in our field and what is possible. We believe that we still know too little about marketing's role in improving—or harming—our world. Unless we broaden the set of outcomes we study and change how we interpret marketing's role, marketing scholars risk becoming detached from many of the most important challenges facing the world today—challenges to which marketing can contribute both positively and negatively ([25]), such as persistent poverty, inequity, illiteracy, insecurity, disease, climate change, pollution, and human trafficking, among many others. Even in wealthy nations such as the United States, large proportions of the population believe the world is getting worse and that the system is stacked against them ([35]). Those perceptions are not necessarily wrong. The "American Dream," which many marketers helped shape, is an illusion for many ([13]; [14]). Discrimination based on gender, race, ethnicity, religion, and sexual orientation continues to keep millions from achieving their hopes and dreams ([ 5]; [16]). "Deaths of despair" caused by suicide, drug overdoses, and alcohol-related liver disease among non-college-educated white men and women have become so high that life expectancy for the U.S. population had begun to decline even before the COVID-19 pandemic hit ([11]). Moreover, these issues are not confined to the United States. Social mobility has declined in many nations ([ 1]). Extreme weather caused by climate change is uprooting lives and threatening livelihoods, while markets for green solutions are still largely nascent or poorly organized.
Surveying the challenges facing the world, a recent report commissioned by the CEOs of companies such as Alibaba, Mars, Merck, and Unilever ([ 9], p. 19) concludes: "Despite the economic and social gains of the past 30 years, the world's current economic model is deeply flawed." Marketing and marketplace exchanges are not peripheral to the world's economic model; they are in fact central to it. The world has no shortage of consequential challenges that should interest marketing scholars.
Fifty years after JM authors and editors made the case for it, the need for the scholarly study of BMBW is even more intense. The research represented in this special issue demonstrates that our discipline has no shortage of talent or tools with which to address these challenges. The record-setting number of 239 submissions that we received for this special issue suggests an intellectual ferment in our discipline that foreshadows new developments. Against this backdrop, we believe the time is ripe for BMBW research to occupy a more central position in the mainstream of marketing scholarship.
During the past 50 years, the field has made noteworthy progress in its pursuit of BMBW research.[ 4] Many scholars—more than we can list—have drawn our field's attention to necessary and important research in this domain. Indeed, some have devoted their lives to this cause. Insights from research have helped the push for change on several important issues, including tobacco advertising, deceptive advertising, labeling, recycling, and the application of marketing tools to nonprofit and social marketing campaigns. Encouragingly, empirical work of this nature has surged in all the leading journals in the field. This work has covered topics as diverse as prosocial behavior, environmental sustainability, corporate political advocacy and fraud, consumer privacy, health, and education. Scholars have also begun to integrate empirical findings to develop conceptual frameworks and research agendas for specific BMBW topics (e.g., [42]).
Yet our survey results and discussions with members of the marketing scholarly community suggest that despite these important inroads, BMBW topics remain peripheral to most scholars' work. Even today, rarely do doctoral dissertations focus on BMBW topics, rarely do sessions at our largest conferences feature BMBW discussions, and rarely do promotion and tenure committees find themselves assessing records of extensive publications on BMBW topics in leading journals. We have not yet realized the full scope of better marketing ideas. Neither have we fully realized the impact that better marketing can have on a better world. This section describes three ways BMBW work can achieve a more central role in our field.
First, we believe that many topics considered mainstream in marketing can be fruitfully viewed from a better world perspective. The authors featured in this special issue, for example, cover a wide array of bread-and-butter marketing topics: sales force management, price promotion, pricing, labeling, product design, product management, social media, the use of influencers, marketing education, marketing consulting, advertising, and targeting. They apply (or study the application of) these familiar marketing tools to better world outcomes. Web Appendix 1 contains a table summarizing these and other key features of the papers in the special issue. Consider the topic of sales force management. [21] find that variable compensation incentive schemes have a negative effect on salespeople's mental and physical health, increasing sick days and stress, especially among those with fewer personal and social resources. These health outcomes detract from the sales gains achieved from this widely used sales force management tool. [45] examine the impact of social media and influencers on the adoption of an eco-friendly pesticide in rural China. [23] showcase how customer relationship management tools can improve fundraising approaches and outcomes for a nonprofit scientific research center. In a final example, [44] demonstrate how price promotions can increase prosocial giving.
Second, we believe marketing scholars should take far more inspiration to find a role for marketing amid better world challenges and opportunities. These topics might, at first glance, appear far from the domain of marketing. Take, for example, discrimination and inequity. Many in the field might view discrimination as the domain of sociologists and psychologists, not marketers. But consumer and consumption responses to stigmatization are surely squarely in the marketing domain (see [15]). Discrimination can be a by-product of mainstream marketing activities such as targeting and segmentation ([37]); it can be implicit in branding and marketing communications; it is often silently furthered by the algorithms used in marketing; and, importantly, it can be mitigated by marketing training and sales initiatives ([12]).
Or take poverty, which remains a persistent worldwide problem. It is often studied by economists, demographers, sociologists, and the occasional consumer researcher ([ 3]; [ 7]), but marketing can contribute to or help alleviate poverty. How do other marketing practices (e.g., franchise location decisions that limit access and opportunity and predatory financial services that prey on ill-informed consumers living paycheck-to-paycheck) contribute to the challenges poor consumers face? What drives these practices, and what can be done to mitigate and guard against them?
Third, we believe marketing scholars should consider how a better (or worse) world can and should influence marketing. As Jerry Zaltman—one of our interviewees for this editorial, who was also one of the authors published in the 1971 JM special issue—described it, "How can the internal world of a firm not be shaped by the external world?" Indeed, the environment around the firm—physical or otherwise—affects both what is done in marketing and how well it is done. How do (or should) positive or negative changes in the environment change the way marketers think and behave? For example, climate change could affect nearly all aspects of marketing, including new product design, channels of distribution, and brand positions.
We believe all three types of research questions belong in the mainstream of marketing scholarship. What gets in the way? Many things contribute, but we focus on a set of assumptions that drive thinking in the field as the most formidable barrier. We discuss these assumptions and issue a set of challenges to both scholars and gatekeepers.
At a superficial level, it may stand to reason that marketing scholars should focus solely on the activities of those with "marketing" in their job titles (e.g., marketing managers, CMOs). But, interestingly, many marketing activities are actually done by individuals who would not consider themselves marketers first and foremost—entrepreneurs, CEOs, general managers, data scientists, product developers, and pricing strategists, to name a few. Many others work in noncommercial organizations as government officials, regulators, and societal critics of marketing. Fixating on the objectives of a narrow set of actors can prevent us from understanding the full impact and potential of marketing activities. For example, many social entrepreneurs see activities that we would regard as marketing, such as generating customer insights, as critical to their work. Similarly, many of those making pricing decisions would not call themselves marketers, yet their decisions profoundly affect who accesses their firms' offerings and who does not. For example, accessible pricing for important services such as mobile telephone services has a profound social impact on everything from education to health to poverty alleviation. These topics should serve as legitimate and valuable bases for marketing scholarship.
We should engage with the entire phenomenon of marketing rather than solely on the activities undertaken by those who define themselves as marketers. In addition to marketing managers, the protagonists in this special issue include entrepreneurs, policy makers, social marketers, nonprofit and NGO leaders, and consumers. For example, [19] offer a marketing intervention that NGOs, policy makers, and financial institutions can use to increase consumer savings that works in both developed and developing markets. As another example, [41] offer guidelines for communications on healthy eating that can help policy makers fight the obesity epidemic.
Most marketing scholars hold their primary affiliation in business schools. Large corporations feature prominently in business school case study lists and recruiter rosters. For researchers, data on large companies are easier to come by. Nevertheless, there are good reasons why large corporations' problems should not disproportionately preoccupy marketing scholars.
Large corporations employ a small fraction of those engaged in marketing activities worldwide. They are also not the sole focus of business school graduates, who increasingly pursue careers in the social and public sectors. Further, even when they actively seek to do good, many (large) companies still define "better world" outcomes as peripheral to their strategic goals. Despite the efforts of some heroic CEOs and the pronouncements of groups such as the [ 8], the focus on BMBW in the context of large corporations too often seems to boil down to their corporate social responsibility ratings, and even those may be subject to spin and manipulation.
We should explore beyond the familiar large businesses most often studied in academic marketing research. There are two shining examples in the special issue. [ 2] focus on entrepreneurs and volunteer consultants and examine how marketing advice affects their growth and the survival of small firms in Uganda. [39] focus on villagers in India and Tanzania and show how marketing education in the form of marketplace literacy training promotes well-being and entrepreneurship outcomes.
Even in contexts beyond large businesses, we see a related recommendation: we should consider more than just the average effect of marketing among consumers, firms, and markets. Averages can conceal variance that is critical to understanding better world outcomes. In a world of few winners and many losers, of oligopolies and inequality, it is cold comfort to the many on the losing side if average outcomes improve. Analyses of heterogeneity in outcomes among consumers and firms offer the opportunity to explore asymmetries in gains and losses. For example, [40] show that cigarette excise taxes decrease smoking but result in stronger brands gaining share, while smoke-free restrictions result in stronger brands losing share. [30] similarly observe that social marketing nudges work better for low-knowledge consumers.
Of course, firms have good reasons to adopt a focus on business profits; many have a fiduciary obligation to do so. Profits offer a clear metric that imposes accountability on managers. Moreover, the financial logic of maximizing shareholder value can be consistent with win-win outcomes for customers, employees, suppliers, communities, and the world at large (Figure 1). However, even the finance field is recognizing the pitfalls of a single-minded and too-often myopic devotion to shareholder value (see [46]). As [32], p. 21) notes, "Ultimately, a corporation sinks or swims on whether it makes a desirable widget, but in order to do this sustainably, it has to weigh the interests of a broader set of stakeholders than just the shareholders."
Graph: Figure 1. The impact of marketing.Note: "Good" is defined as long-term positive outcomes.
Looking beyond profits is important in part because increasing evidence shows that too many contemporary markets are uncompetitive and that instead of earning profits by investing and innovating, powerful firms use political pressure to secure their advantages ([31]). In such contexts, markets fail to deliver. Rather than win-win, the outcome is win-lose—profits for firms and losses for the world at large (see Figure 1).
Marketing scholars have the opportunity to document evidence on the consequences of bad actors and explain why and how marketing contributes to bad societal outcomes. A potent example is the management consulting company McKinsey's set of marketing recommendations to Purdue Pharma, such as using distributor price rebates to promote opioids, which we now know have devastated individuals, families, and entire communities ([ 6]). What can marketing learn from its societal critics? How should our research amplify these voices in the spirit of shaping better marketing practices for the world? Considering lose-lose outcomes—bad for the world and bad for the firm in the long run—researchers could seek to understand why these actions persist, with the goal of making them less common.
We also see many opportunities for researchers to examine the lose-win cell in Figure 1. This scenario manifests in the many situations in which it is unprofitable for firms to engage in a marketing action that benefits the world at large. How can they nevertheless be incentivized to apply their resources and capabilities and to engage in activities that lead to a "better world"? Many social enterprises, for example, rely on a combination of profits and grants to try to do good. Others collaborate with governments, NGOs, and grassroots entities to do so. What business models are appropriate in these contexts? What collaborations are most effective? What marketing activities offer the most leverage toward better world outcomes?
We should examine all four cells described in Figure 1 to develop a full accounting of marketing's impact on the world and the conditions under which each applies. For example, [28] show how a lose-win situation of offering steep discounts for imperfect produce can be turned into a win-win situation by combining "ugly" labeling with moderate discounts to increase purchase of imperfect produce that would have otherwise been wasted.
The outcomes often studied in contemporary marketing research typically involve one of these two stakeholders. But what are the negative and positive spillovers of marketing activities beyond customers and firms? The impact of marketing travels much further than most of our research has acknowledged. As [43], p. 217) emphasize, "adopting the perspective of the aggregate marketing system helps a person 'see' the field of marketing in its true expanse and complexity. However, this perspective largely has disappeared from the marketing mainstream in recent years." Over 20 years later, not much has changed.
We should place a greater emphasis on the spillovers—both positive and negative—of marketing. For example, [36] observe a positive spillover of spending more to purchase luxury goods: that the products last longer and are more likely to end up in secondary markets rather than a landfill.
We believe that this assumption is not far removed from current reality. To illustrate this point, we conducted a text analysis[ 5] of the words used in the manuscripts submitted to the BMBW special issue: we compared these words with those used in a random sample of 184 reports that describe the United Nations Sustainable Development Goals.[ 6] As a further contrast, we compared the words used in these two sets of documents with those used in Marketing Management by [24]; hereafter K&K) using a natural language processing (word2vec) model ([ 4]). Results indicate very little overlap between the marketing and UN documents (see Web Appendix 2 for details).
In terms of the nature of this disconnect, we observe differences in the three sets of documents in ( 1) the stakeholders addressed, ( 2) the activities and decisions those stakeholders undertake or are subject to, and ( 3) outcomes (see Web Appendix 2). In terms of stakeholders, the BMBW submissions are similar to K&K: both documents focused on consumers, customers, and businesses. The UN documents, in contrast, address a larger set of stakeholders (e.g., government, women, environment). Comparing the types of activities and actions, we find that both the BMBW submissions and K&K describe "doing" activities (e.g., "branding," "designing," "pricing"), while the UN documents show a preponderance of advocacy and evaluative words (e.g., "ensuring," "assessing," "addressing"). Finally, in terms of outcomes, many of the BMBW and K&K words relate to final outcomes (e.g., "choice," "value," "profit"), while the UN outcomes are more of an intermediate nature, describing enabling factors such as "publication," "education," and "growth."
For our field to reach research, practitioner, and beneficiary communities beyond the traditional ones, we should develop more diverse points of connection through which we can share ideas and insights. Indeed, the differences in language and areas of focus in our text analysis suggest that we have a large gap to bridge between the world of our scholarly community and that of many practitioners who are actively involved in the pursuit of better world outcomes. Doing so might require us to break out of familiar bubbles and to immerse ourselves more fully in the contexts we seek to understand ([38]).
Another recommendation to bridge the gap is for authors to write "Marketing Implications" sections for their papers—a practice we encourage at JM. We emphasize that marketing implications are not restricted to managers in firms: all actors who could engage in or influence better marketing are relevant here, including policy makers, educators, and societal stakeholders who challenge marketing activities. We ignore this fact at our own peril ([29]). We are sure that journals have contributed to this narrowing of perspective, including JM—a choice that cuts us off from the full implications of our ideas for the world. For those scholars worrying we will leave managers behind, it is important to remember that managers can and should learn from policy and societal implications if they are to be effective.
Although all research involves choices that reflect our values, getting involved in BMBW may make some scholars uncomfortable. It is indeed important that we remain scientific and objective, and our call for papers made it clear that there was no place for advocacy in the special issue. Yet we should also recognize that most research is value laden, with some values being more accepted or more enduring than others. In fact, the very goal of maximizing business profits is value laden. We encourage scholars to remain truth tellers in their investigations and to not shy away from investigations that are social and political flashpoints.
We may sometimes have to acknowledge that controversy can be a by-product of engaging with a better world, where evidence is often scarce and opinions polarized. At the same time, we should strive for independence and objectivity. For example, [34] take on the challenge of organ donation. Strategies to increase donation rates are hotly debated in policy circles. These authors show that changing the design of the service encounter and the content of the appeal helps improve donation rates. Similarly, [20] take on the highly politicized banning of plastic bags in Chile to understand where the policy design fell short.
[18], p. 17) made this claim early: "The discussions of the 'social responsibilities of business' are notable for their analytical looseness and lack of rigor." A more contemporary quote from one of our survey respondents reiterates this view: "I think prior work in this area sometimes gets coded as not 'theoretical' or sophisticated enough." An assumption that also surfaces among academics discussing BMBW research is that it is difficult to undertake rigorously because of the lack of access to data sets and the difficulty of running field experiments to pin down the causal role of marketing actions in generating better world outcomes.
At the same time, if marketing academics believe that studying BMBW is critically important, as our survey results and numerous conversations indicate, then we need to determine how to bring rigor to this research. Big problems demand creative solutions. BMBW can actually provide an opportunity to apply new dimensions of rigor. Indeed, rigorous work is not unknown to the study of "better world" topics. Some scholars have even won Nobel Prizes for their work on topics such as poverty alleviation, externalities, and social justice. Crucially, articles in this special issue provide a clear refutation of the idea that the study of BMBW comes at the expense of rigor (see Web Appendix 1).
We should strive to maintain rigor in our BMBW investigations. And we should do so within and across methods. Given JM's broad mission, it is particularly gratifying to see that the articles in this special issue apply a diverse set of methods, including field experiments, quasi-experiments, lab and online studies, surveys, web-scraping, archival methods, netnography, and qualitative interviews. We venture to guess that this is perhaps the most method-diverse special issue ever published in the field of marketing. As highlighted in Web Appendix 1, many papers utilize multiple methods, further enriching this diverse portrait. For example, [33] use a series of lab, online, and field experiments to show that requests for small charitable donations can be broadly targeted, beyond prior donors and those who support the cause, simply by offering consumers an opportunity to express their identity.
Research that can create impact at scale is not always easy or cheap to pursue. The timelines involved can be long. For those working in international contexts, distances—both geographic and cultural—can be challenging. Objectives may not be widely agreed upon, partly for reasons noted previously. It follows that metrics for BMBW may also be contentious. Given all this, returns will appear uncertain relative to the risks involved. Well-meaning advisors might warn their students to "stay mainstream" and adopt dissertation topics that can be easily conquered in the time they have. The result is that many young scholars who care deeply about these issues do not pursue BMBW topics in their formative years, hoping instead to do so later in their careers. However, for many, this opportunity never materializes. We are reminded of Warren Buffet's quote ([10], p. D5): "The chains of habit are too light to be felt until they're too heavy to be broken."
We should be bold in following our passions and ideals. Many marketers entered the profession in the belief that their thinking and their actions could help contribute to a better world. Applicants to doctoral programs in marketing today frequently list this belief as a major rationale for their applications, and conversations with them suggest that many are sincere in their desire to contribute to a better world through marketing scholarship. Though challenges do exist, so does an openness to fresh new ideas. Moreover, these challenges are not qualitatively different from those in any new area at the cusp of going mainstream. Risky ideas can be combined with less risky ideas in a portfolio of research. Further, risks and the efforts required to mitigate them are potentially more feasible in the early stages of one's career, when teaching and service obligations may be fewer.
To our delight, this special issue involves papers in which doctoral students and recently hired junior faculty have played an important role, including Katherine Du (University of Wisconsin-Milwaukee), Claudia Gonzalez-Arcos (University of Queensland), Ashley Goreczny (Iowa State University), Alison Joubert (University of Queensland), Sungjin Kim (University of Hawai'i), Sid Mookerjee (University of British Columbia), Jacqueline Rifkin (University of Missouri-Kansas City), Zhengyu Shi (University of Hong Kong), Jennifer Sun (Columbia University), and Wanqing Zhang (City University of London), spanning four continents.
One of our survey respondents framed this assumption thus: "The major journals typically emphasize theoretical advances, and BMBW work, by its very nature, tends to be more applied." In the last year alone, many leading marketing journals have published special issues focused on better world outcomes, so this assumption is slowly being put to rest. Even so, this perception is widely prevalent, and the pace of change may need to pick up. Our survey respondents made such comments: "There are many barriers, including journals being less open to this type of research," "greater appreciation for the topic [is needed] by editors/journals," and "journals/reviewers are usually rather rigid in their thinking/reviewing style."
We urge authors to retain their ambition to speak to the mainstream of marketing when addressing BMBW topics. To do so, we cannot rely only on reviewers' open-mindedness. It is best to anticipate and preempt the question that confronts any scholar who seeks to introduce new topics to the field: "Why does this topic belong in marketing?" Sell your ideas as marketing relevant ([26]).
We also urge editors and reviewers of leading journals to adopt a forward-looking stance to determine what could (or indeed should) belong in our field. We should be prepared to champion and shepherd some papers through the process even if it means overruling reviewers. The JM editors are open to these papers, and we hope this special issue sends a signal about our commitment to this area.
The authors in our special issue made the leap across the chasm imposed by the aforementioned assumptions. This issue, to our delight, covers many important challenges facing the world, including sustainability and climate concerns, economic and social empowerment, health and well-being, and increasing prosocial giving as a way of mitigating some of these challenges. Web Appendix 1 lists the papers by each topic and also catalogues the set of geographies covered, including developed markets such as the United States, Canada, and Germany and developing and emerging markets such as China, Brazil, Chile, India, Uganda, and Tanzania.
We invite you to make this leap as well—to look at pressing social issues and to ask some simple questions: Does this topic belong in marketing? How could you frame this topic as a mainstream marketing question? From these questions would emanate other questions: Why is the outcome important to marketing? Does marketing exacerbate the problem? Does marketing have the potential to provide a solution to or an explanation for the problem?
We asked marketing and consumer research scholars from across the field to reflect on these questions in sessions that we hosted at AMA and ACR conferences as part of the call for papers for the special issue.[ 7] They generated many interesting perspective and angles to connect better marketing to a better world, and we urge you to take inspiration from their ideas as well as the topics suggested in our call.[ 8] Our hope is that both will serve as inspiration long after this special issue.
As a coda to this special issue, we are announcing a set of initiatives to help address the barriers and to support the marketing community in this area. Details and updates regarding these intiatives are available at the BMBW website at www.bmbw.org.
- BMBW Workshops, Conferences, and Competitions to build a community of interdisciplinary scholars: We plan to conduct a monthly online workshop series that will be initially funded by the Wheeler Institute for Business and Development and JM. This series will feature speakers addressing BMBW topics within the marketing discipline. This team of editors will do the initial outreach for the series. To help develop the field, preference will be given to research that is in development and could benefit from input from other scholars. We also see an opportunity for a field-wide annual conference on BMBW that could build more community and foster the interdisciplinary bonds that will likely unlock the best solutions to better world problems. We will host a competition that will encourage cross-disciplinary submissions of research proposals that address challenging BMBW topics. We envision the possibility of a pan-marketing award for the top BMBW paper.
- BMBW Training to impart knowledge and skills: Our doctoral training as well as socialization into research in marketing too often ignores BMBW and does not offer relevant knowledge and skills to new members of the profession or to those seeking to make the transition to work on BMBW topics. To address this barrier, we commit to launching a research proseminar covering BMBW topics in 2021. The proseminar will pool expertise from the global community of scholars and will be open to all scholars interested in learning about topics, tools, and methods that can help illuminate the more complex research problems posed by a BMBW focus.
- BMBW Data Initiative to provide a BMBW data repository: To address the limited availability of data, we will initiate a data collation exercise that draws researchers' attention to the possibilities for empirical research using new and existing data sets. We will work with the creators of these data sets to offer input and training on how to use them effectively to study BMBW topics. The BMBW website (bmbw.org) will include a repository of these data sets and links to existing datasets. The training in item 2 will also address how these data sets can be leveraged to study BMBW.
The winds of change in science, regulation, demographics, and the physical environment are creating new opportunities for marketing to make an impact on the world at large. New technologies are connecting ideas, resources, individuals, firms, societies, and markets in unprecedented ways. Those who harness these changes can shape aspirations, identities, and notions of right and wrong. Thanks in part to the activities of those who have already done so, in many ways the world has never been more prosperous, safe, educated, or equal (see [27]; [35]). Yet the world has no shortage of challenges to address. The marketing discipline has no shortage of talent with which to tackle these challenges and these opportunities.
If we cannot transcend our own scholarly tribes and explore beyond the familiar, then it will be a failure of ambition on our part. If we cannot demonstrate to the next generation of scholars that better world outcomes are central to our field, then it would be our failure to inspire. If we cannot marshal the power of ideas and facts to speak with the powerless and to speak truth to power, it would not just be a failure of imagination. It would be a tragedy and a dereliction of our duty as scholars.
We can do more. We can do better. Let's work together to develop better marketing for a better world.
Supplemental Material, sj-pdf-1-jmx-10.1177_00222429211003690 - Better Marketing for a Better World
Supplemental Material, sj-pdf-1-jmx-10.1177_00222429211003690 for Better Marketing for a Better World by Rajesh K. Chandy, Gita Venkataramani Johar, Christine Moorman and John H. Roberts in Journal of Marketing
Footnotes 1 Online supplement: https://doi.org/10.1177/00222429211003690
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 BMBW topics have appeared in specialized journals (e.g., those at the intersection of marketing and public policy, ethics, and macro issues) and in articles and occasional special issues such as the current and 1971 JM issues, JMR on education and marketing, Marketing Science on health, JCR on transformative consumer research, JCP on consumer psychology for the greater good, and JPP&M and JACR on COVID-19). Such topics are evident in discussions in special interest groups (e.g., the AMA's Marketing and Society special interest group) and in movements such as the Transformative Consumer Research initiative.
5 We are grateful to Sanjana Rosario for undertaking this analysis on our behalf.
6 These documents were drawn from the UN Department of Economic and Social Affairs (UN DESA) in the Sustainable Development Goals Division and can be found at https://journalofmk.tg/30HR209.
7 https://www.ama.org/rethinking-marketing-scholarship-from-a-better-marketing-for-a-better-world-perspective/.
8 https://www.ama.org/2018/09/12/better-marketing-for-a-better-world-special-issue-journal-of-marketing/.
References Alesina Alberto, Hohmann Sebastian, Michalopoulos Stelios, Papaioannou Elias. (2021), "Intergenerational Mobility in Africa,"Econometrica, 89 (1), 1–35.
Anderson Stephen, Chintagunta Pradeep, Germann Frank, Vilcassim Naufel. (2021), "Do Marketers Matter for Entrepreneurs? Evidence from a Field Experiment in Uganda,"Journal of Marketing, 85 (3), 78–96.
Andreasen Alan R. (1975), The Disadvantaged Consumer. New York: Free Press.
Berger Jonah, Humphreys Ashlee, Ludwig Stephan, Moe Wendy W., Netzer Oded, Schweidel David A. (2020), "Uniting the Tribes: Using Text for Marketing Insight," Journal of Marketing, 84 (1), 1–25.
Bertrand Marianne, Mullainathan Sendhil. (2004), "Are Emily and Greg More Employable than Lakisha and Jamal? A Field Experiment on Labor Market Discrimination,"American Economic Review, 94 (4), 991–1013.
Bogdanich Walt, Forsythe Michael. (2020), "McKinsey Issues a Rare Apology for Its Role in OxyContin Sales,"New York Times(December 8), https://www.nytimes.com/2020/12/08/business/mckinsey-opioids-oxycontin.html.
Bryant Andrew, Hill Ronald P. (2019), "Poverty, Consumption, and Counterintuitive Behavior,"Marketing Letters, 30 (3), 233–43.
Business Roundtable (2019), "Business Roundtable Redefines the Purpose of a Corporation to Promote 'An Economy That Serves All Americans,'"https://www.businessroundtable.org/business-roundtable-redefines-the-purpose-of-a-corporation-to-promote-an-economy-that-serves-all-americans
9 Business and Sustainable Development Commission (2017), Better Business, Better World. Davos: Business and Sustainable Development Commission.
Carricaburu Lisa. (1996), "Wizard of Wall Street Holds Audience Spellbound in Rare Appearance at WSU,"Salt Lake Tribune (September 25), D5.
Case Anne, Deaton Angus. (2020), Deaths of Despair and the Future of Capitalism. Princeton, NJ: Princeton University Press.
Chaney Kimberly E., Sanchez Diana T., Maimon Melanie R. (2019), "Stigmatized-Identity Cues in Consumer Spaces,"Journal of Consumer Psychology, 29 (1), 130–41.
Chetty Raj, Hendren Nathaniel, Kline Patrick, Saez Emmanuel, Turner Nicholas. (2014), "Is the United States Still a Land of Opportunity? Recent Trends in Intergenerational Mobility,"American Economic Review, 104 (5), 141–47.
Coskuner-Balli Gokcen. (2020), "Citizen-Consumers Wanted: Revitalizing the American Dream in the Face of Economic Recessions, 1981–2012,"Journal of Consumer Research, 47 (3), 327–49.
Crockett David. (2017), "Paths to Respectability: Consumption and Stigma Management in the Contemporary Black Middle Class,"Journal of Consumer Research, 44 (3), 554–81.
Crockett David, Grier Sonya A. (2021), "Race in the Marketplace and COVID-19,"Journal of Public Policy & Marketing, 40 (1), 89–91.
Dawson Leslie M. (1971), "Marketing Science in the Age of Aquarius,"Journal of Marketing, 35 (3), 66–72.
Friedman Milton. (1970), "A Friedman Doctrine—The Social Responsibility of Business Is to Increase Its Profits,"New York Times (September 13), 33.
Garbinsky Emily, Mead Nicole, Gregg Daniel. (2021), "Popping the Positive Illusion of Financial Responsibility Can Increase Personal Savings: Applications in Emerging and Western Markets,"Journal of Marketing, 85 (3), 97–112.
Gonzalez-Arcos Claudia, Joubert Alison, Scaraboto Daiane, Guesalaga Rodrigo, Sandberg Jörgen. (2021), "'How Do I Carry All This Now?' Understanding Consumer Resistance to Sustainability Interventions,"Journal of Marketing, 85 (3), 44–61.
Habel Johannes, Alavi Sascha, Linsenmayer Kim. (2021), "Variable Compensation and Salesperson Health,"Journal of Marketing, 85 (3), 130–49.
Kelley Eugene S. (1971), "Marketing's Changing Social/Environmental Role,"Journal of Marketing, 35 (3), 1–3.
Kim Sungjin, Gupta Sachin, Lee Clarence. (2021), "Managing Members, Donors, and Member-Donors for Effective Nonprofit Fundraising,"Journal of Marketing, 85 (3), 220–39.
Kotler Philip, Keller Kevin Lane. (2011), Marketing Management, 14th ed. Upper Saddle River, NJ: Prentice Hall.
Kotler Philip, Levy Sidney J. (1969), "Broadening the Concept of Marketing,"Journal of Marketing, 33 (1), 10–15.
MacInnis Deborah J., Morwitz Vicki G., Botti Simona, Hoffman Donna L., Kozinets Robert V., Lehmann Donald R., Lynch John G.Jr, Pechmann Cornelia. (2020), "Creating Boundary-Breaking, Marketing-Relevant Consumer Research,"Journal of Marketing, 84 (2), 1–23.
Maddison Angus. (2001), The World Economy: A Millennial Perspective. Paris: OECD Publishing.
Mookerjee Siddhanth, Cornil Yann, Hoegg JoAndrea. (2021), "From Waste to Taste: How 'Ugly' Labels Can Increase Purchase of Unattractive Produce,"Journal of Marketing, 85 (3), 62–77.
Moorman Christine, Heerde Harald J. van, Page Moreau C., Palmatier Robert W. (2019), "JM as a Marketplace of Ideas,"Journal of Marketing, 83 (1), 1–7.
Mrkva Kellen, Posner Nathaniel A., Reeck Crystal, Johnson Eric J. (2021), "Do Nudges Reduce Disparities? Choice Architecture Compensates for Low Consumer Knowledge,"Journal of Marketing, doi/pdf/10.1177/0022242921993186.
Philippon Thomas. (2019), The Great Reversal: How America Gave Up on Free Markets. Cambridge, MA: Harvard University Press.
Rajan Raghuram. (2020), "'50 Years Later, It's Time to Reassess': Raghuram Rajan on Milton Friedman and Maximizing Shareholder Value," in Milton Friedman 50 Years Later, Zingales Luigi, Kasperkevic Jana, Schechter Asher, eds. Chicago: Stigler Center at the University of Chicago, 17–21.
Rifkin Jacqueline, Du Katherine, Berger Jonah. (2021), "Penny for Your Preferences: Leveraging Self-Expression to Encourage Small Prosocial Gifts,"Journal of Marketing, 85 (3), 204–19.
Robitaille Nicole, Mazar Nina, Tsai Claire, Haviv Avery, Hardy Elizabeth. (2021), "Increasing Organ Donor Registrations with Behavioral Interventions: A Field Experiment,"Journal of Marketing, 85 (3), 168–83.
Rosling Hans, Rosling Ola, Rönnlund Anna Rosling. (2018), Factfulness: Ten Reasons We're Wrong About the World – and Why Things Are Better Than You Think. New York: Flatiron Books.
Sun Jennifer, Bellezza Silvia, Paharia Neeru. (2021), "Buy Less, Buy Luxury: Understanding and Overcoming Product Durability Neglect for Sustainable Consumption,"Journal of Marketing, 85 (3), 28–43.
Ukanwa Kalinda, Rust Roland T. (2020), "Discrimination in Service,"Marketing Science Institute Working Paper Series Report No. 18–121.
Van Heerde Harald J., Moorman Christine, Page Moreau C., Palmatier Robert W. (2021), "Reality Check: Infusing Ecological Value into Academic Marketing Research,"Journal of Marketing, 85 (2), 1–13.
Viswanathan Madhu, Umashankar Nita, Sreekumar Arun, Goreczny Ashley. (2021), "Marketplace Literacy as a Pathway to a Better World: Evidence from Field Experiments in Low-Access Subsistence Marketplaces,"Journal of Marketing, 85 (3), 113–29.
Wang Yanwen, Lewis Mike, Singh Vishal. (2021), "Investigating the Effects of Excise Taxes, Public Usage Restrictions, and Anti-Smoking Ads Across Cigarette Brands,"Journal of Marketing, 85 (3), 150–67.
Weihrauch Andrea, Huang Szu-Chi. (2021), "Portraying Humans as Machines to Promote Health: Unintended Risks, Mechanisms, and Solutions,"Journal of Marketing, 85 (3), 184–203.
White Katherine, Habib Rishad, Hardisty David J. (2019), "How to Shift Consumer Behaviors to Be More Sustainable: A Literature Review and Guiding Framework,"Journal of Marketing, 83 (3), 22–49.
Wilkie William L., Moore Elizabeth S. (1999), "Marketing's Contributions to Society,"Journal of Marketing, 63 (4), 198–218.
Zhang Kuangjie, Cai Fengyan, Shi Zhengyu. (2021), "Do Promotions Make Consumers More Generous? The Impact of Price Promotions on Consumers' Donation Behavior,"Journal of Marketing, 85 (3), 240–55.
Zhang Wanqing, Chintagunta Pradeep, Kalwani Manohar. (2021), "Social Media, Influencers, and Adoption of an Eco-Friendly Product: Field Experiment Evidence from Rural China,"Journal of Marketing, 85 (3), 10–27.
Zingales Luigi. (2020), "Friedman's Legacy: From Doctrine to Theorem," in Milton Friedman 50 Years Later, Zingales Luigi, Kasperkevic Jana, Schechter Asher, eds. Chicago: Stigler Center at the University of Chicago, 128–35.
~~~~~~~~
By Rajesh K. Chandy; Gita Venkataramani Johar; Christine Moorman and John H. Roberts
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 13- Blame the Bot: Anthropomorphism and Anger in Customer–Chatbot Interactions. By: Crolic, Cammy; Thomaz, Felipe; Hadi, Rhonda; Stephen, Andrew T. Journal of Marketing. Jan2022, Vol. 86 Issue 1, p132-148. 17p. 1 Diagram, 2 Charts, 5 Graphs. DOI: 10.1177/00222429211045687.
- Database:
- Business Source Complete
Blame the Bot: Anthropomorphism and Anger in Customer–Chatbot Interactions
Chatbots have become common in digital customer service contexts across many industries. While many companies choose to humanize their customer service chatbots (e.g., giving them names and avatars), little is known about how anthropomorphism influences customer responses to chatbots in service settings. Across five studies, including an analysis of a large real-world data set from an international telecommunications company and four experiments, the authors find that when customers enter a chatbot-led service interaction in an angry emotional state, chatbot anthropomorphism has a negative effect on customer satisfaction, overall firm evaluation, and subsequent purchase intentions. However, this is not the case for customers in nonangry emotional states. The authors uncover the underlying mechanism driving this negative effect (expectancy violations caused by inflated pre-encounter expectations of chatbot efficacy) and offer practical implications for managers. These findings suggest that it is important to both carefully design chatbots and consider the emotional context in which they are used, particularly in customer service interactions that involve resolving problems or handling complaints.
Keywords: customer service; artificial intelligence; conversational agents; chatbots; anthropomorphism; anger; expectancy violations
The use of artificial intelligence (AI) in marketing is on the rise, as managers experiment with the use of AI-driven tools to augment customer experiences. One relatively early use of AI in marketing has been the deployment of digital conversational agents, commonly called chatbots. Chatbots "converse" with customers, through either voice or text, to address a variety of customer needs. Chatbots are increasingly replacing human service agents on websites, social media, and messaging services. In fact, the market for chatbots and related technologies is forecasted to exceed $1.34 billion by 2024 ([71]).
While some industry commentators suggest that chatbots will improve customer service while simultaneously reducing costs ([16]), others believe they will undermine customer service and negatively impact firms ([34]). Thus, while customer service chatbots have the potential to deliver greater efficiency for firms, whether—and how—to best design and deploy chatbots remains an open question. The current research begins to address this issue by exploring conditions under which customer service chatbots negatively impact key marketing outcomes. While many factors may influence customers' interactions with chatbots, we focus on the interplay between two common features of the customer service chatbot experience.
The first feature relates to the design of the chatbot itself: chatbot anthropomorphism. This is the extent to which the chatbot is endowed with humanlike qualities such as a name or avatar. Currently, the prevailing logic in practice is to make chatbots appear more humanlike ([ 9]) and for them to mimic the nature of human-to-human conversations ([44]). However, anthropomorphic design in other contexts (e.g., branding, product design) does not always produce beneficial outcomes (e.g., [36]; [38]). Accordingly, we examine circumstances under which anthropomorphism of customer service chatbots may be harmful for firms.
The second dimension explored in this research is a commonly occurring feature in customer service interactions, irrespective of the modality: customer anger. Anger is one of the most prevalent specific emotions occurring in customer service contexts; estimates suggest that as many as 20% of call center interactions involve hostile, angry, complaining customers ([24]). Furthermore, the prevalence of anger increased during the COVID-19 pandemic ([57]; [60]), so a higher proportion of interactions are likely to be with angry customers. Thus, it is both practically relevant to consider how customer anger interacts with chatbot anthropomorphism and theoretically relevant due to the specific responses (e.g., aggression, holding others accountable) evoked by anger that impact the efficacy of more humanlike technology.
Across five studies including the analysis of a large real-world data set from an international telecommunications company and four experiments, we find that when customers in an angry emotional state encounter a chatbot-led service interaction, chatbot anthropomorphism has a negative effect on customers' satisfaction with the service encounter, their overall evaluation of the firm, and their subsequent purchase intentions. However, this is not the case for customers in nonangry emotional states. The negative effect is driven by an expectancy violation; specifically, anthropomorphism inflates preinteraction expectations of chatbot efficacy, and those expectations are disconfirmed. Our findings suggest that it is important to both carefully design chatbots and consider the emotional context in which they are used, particularly in common types of customer service interactions that involve handling problems or complaints. This research contributes to the nascent literature on chatbots in customer service and has managerial implications both for how chatbots should be designed and for context-related deployment considerations.
Deliberate marketing efforts have made anthropomorphism, or the attribution of humanlike properties, characteristics, or mental states to nonhuman agents and objects ([20]; [68]), especially pervasive in the modern marketplace. Product designers and brand managers often encourage customers to view their products and brands as humanlike, through a product's visual features (e.g., face-like car grilles; [40]) or brand mascots (e.g., the Pillsbury Doughboy; [66]). In digital settings, advances in machine learning and AI have ushered in a new wave of highly anthropomorphic devices, from humanlike self-driving cars ([70]) to voice-activated virtual assistants with human names and speech patterns (e.g., Amazon's Alexa; [32]).
Extant research generally suggests that inducing anthropomorphic thought is linked to improved outcomes. "Humanized" products and brands are more likely to achieve long-term business success because they encourage a more personal consumer–brand relationship ([ 1]; [66]). Anthropomorphic product features can make products more valuable ([28]) and can boost overall product evaluations in categories, including automobiles, mobile phones, and beverages ([ 1]; [39]; [40]). [67] found that anthropomorphized products increased consumers' preference and subsequent choice of those products.
Anthropomorphism of technology has also been shown to improve marketing outcomes. Humanlike interfaces can increase customer trust in technology by increasing perceived competence ([ 5]; [70]) and are more resistant to breakdowns in trust ([18]). Avatars (anthropomorphic virtual characters) can make online shopping experiences more enjoyable, and both avatars and anthropomorphic chatbots can increase purchase intentions ([27]; [31]; [72]). Anthropomorphic digital messengers can even be more persuasive than human spokespeople in some contexts ([63]) and can increase advertising effectiveness ([13]). Anthropomorphized digital devices can even become friends with their users ([56]), such that the consumer resists being disloyal by replacing the product ([12]), leading to greater customer brand loyalty.
Although most evidence points to beneficial effects of anthropomorphism, there are drawbacks. For example, anthropomorphic helpers in video games reduce enjoyment of the gaming experience by undermining a players' sense of autonomy ([36]). Other research shows that for agency-oriented customers, brand anthropomorphism exaggerates the perceived unfairness of price increases ([38]) and hurts brand performance amid negative publicity ([54]). Low-power customers perceive risk-bearing entities (e.g., slot machines) as riskier when the entities are anthropomorphized ([37]). Further, research suggests that when customers are in crowded environments and want to socially withdraw, brand anthropomorphism harms customer responses ([53]). Thus, it would be overly simplistic to assume that anthropomorphism positively impacts customers' encounters with brands, products, or companies. The consequences are more nuanced, with outcomes depending on both customer characteristics and the context ([64]).
While customers' downstream responses to anthropomorphism are mixed, one consistent consequence of anthropomorphism is that customers attribute more agency to anthropomorphic entities ([20]). "Agency" refers to the capacity to plan and act ([25]). Because anthropomorphism leads customers to perceive a mental state in another entity, it increases individuals' perception that the entity is capable of acting in a deliberate manner ([69]). This increases expectations that the agent has abilities such as emotion recognition, planning, and communication ([25]). These heightened expectations lead individuals to ascribe moral responsibilities to anthropomorphic entities ([69]), to believe that the entity should be held accountable for its actions ([18]), and to think the entity deserves punishment in the case of wrongdoing ([25]).
Of course, anthropomorphic entities do not always perform in a manner consistent with the high levels of agency customers expect. In fact, some researchers suggest that one reason behind the "uncanny valley" (i.e., the tendency for a robot to elicit negative emotional reactions when it closely resembles a human; [48]) is because robots do not perform in the agentic manner that their human resemblance would imply ([69]). In other words, the robots' behavior violates the expectations elicited by their highly anthropomorphic facade. These violations arguably apply to current chatbots, given that their performance is not expected to reach believable levels of human intelligence before 2029 ([58]). Thus, expectancy violations play an important role in chatbot-driven customer service settings.
Before using a product or service, customers form expectations regarding how they anticipate the target product, brand, or company will perform. Postusage, customers evaluate the target's performance and compare that to their preusage expectations ([10]). When performance fails to meet expectations, the negative disconfirmation is known as an expectancy violation ([62]), which arises because ( 1) preusage expectations are high or ( 2) postusage performance is poor ([10]). Expectancy violations not only harm customer satisfaction ([50]; [51]) but also negatively impact other consequential downstream outcomes, including attitude toward the company ([10]) and purchase intentions ([11]; [50]). Importantly, customer responses to expectancy violations are highly influenced by their emotional states, particularly anger ([ 4]).
Two theories help explain why anger increases customers' negative responses to expectancy violations. The functionalist theory of emotion suggests that anger is an activating, high-intensity emotion with an evolutionary purpose: it evokes quick decision making and heuristic use to react quickly to immediate threat ([ 8]). Anger is often used as a strategy to respond to obstacles ([43]; [46]) or retaliate against an offending party ([14]) because of its tendency to increase action and aggression, compared with other emotions that are deactivating (e.g., sadness; [15]; [41]) or nonaggressive ([43]).
This retaliation is also predicted by appraisal theorists, who suggest that even in situations of incidental anger, anger increases the tendency to hold others responsible for negative outcomes ([35]) and to respond punitively toward them ([23]; [42]; [43]). This is markedly distinct from emotions such as frustration or regret, which are more likely to manifest when people hold the situation or themselves responsible for negative outcomes, respectively ([21]; [55]). Thus, angry (vs. nonangry) customers are more likely to blame others and retaliate when another's performance falls short of expectations. This is particularly the case if their goals are obstructed ([46]), as angry customers especially feel the need to achieve a desirable outcome ([55]).
Drawing from the extant theories and research, we hypothesize that anthropomorphism heightens customers' preperformance expectations about a chatbot's level of agency and performance capabilities, resulting in expectancy violations. Further, angry customers are more likely to suffer from expectancy violations due to their need to overcome obstacles, to blame and hold others accountable, and to respond punitively to such expectancy violations due to their action orientation (i.e., giving lower satisfaction ratings, poor reviews, or withholding future business from the offending party). This logic would suggest that angry customers might be better served by nonanthropomorphic agents. Recent research supports this notion by demonstrating that in unpleasant service situations, reducing human contact (e.g., through technological barriers; [ 6]) can help attenuate customer dissatisfaction and limit negative service evaluations ([22]).
Building on these arguments, we predict that individuals who enter a chatbot service interaction in an angry emotional state will respond negatively to chatbot anthropomorphism, whereas individuals in nonangry emotional states will not. While the most immediate negative reaction is likely to manifest in reduced customer satisfaction ratings of the service encounter with the chatbot, this can also carry over to harm more general firm evaluations and result in lower future purchase intentions, which are known consequences of dissatisfaction ([ 3]). Formally, we hypothesize the following:
- H1: For angry customers, chatbot anthropomorphism has a negative effect on (a) customer satisfaction, (b) company evaluation, and (c) purchase intention. This negative effect does not manifest for customers in nonangry emotional states.
- H2: Chatbot anthropomorphism leads to inflated expectations of chatbot efficacy, which, for angry customers, results in the negative effect described in H1.
Our proposed conceptual framework is illustrated in Figure 1. Across five studies, using a combination of real-world and experimental data, we test the different parts of our theorizing to collectively support our proposed framework. In Study 1, we analyze a large data set from an international mobile telecommunications company that captures customers' interactions with a customer service chatbot. We use natural language processing (NLP) on chat transcripts and find that for customers exhibiting an angry emotional state during a chatbot-led service encounter, anthropomorphic treatment of the bot has a negative effect on their satisfaction with the service encounter (consistent with H1a). In Study 2, the first of four experiments, we manipulate chatbot anthropomorphism and customer anger and find that angry customers display lower customer satisfaction when the chatbot is anthropomorphic versus when it is not (consistent with Study 1 and H1a). Study 3 shows that the negative effect extends to company evaluations (H1b) but not when the chatbot effectively resolves the problem. Study 4 shows that the negative effect of chatbot anthropomorphism for angry customers extends further to reduce customers' purchase intentions (H1c) and provides evidence that this effect is driven by inflated preinteraction expectations of chatbot efficacy (H2). Finally, Study 5 manipulates preinteraction expectations and demonstrates that the negative effect dissipates when people have lower expectations of anthropomorphic chatbots (further supporting H2).
Graph: Figure 1. Illustration of proposed model.
Study 1 analyzes a real-world data set from an international mobile telecommunications company capturing customers' interactions with a customer service chatbot. The chatbot was available via the company's website and mobile app and was a text-only bot driven by machine learning, specifically, advanced NLP. The chatbot was highly anthropomorphic; the avatar was a cartoon illustration of a young female avatar with long hair, makeup, and modern casual clothing. Her name appeared in the chat, and customers could visit a profile webpage with her bio describing her personality and listing some of her likes and dislikes.
The main purpose of the study was to examine how treating a chatbot as more or less human (i.e., higher or lower anthropomorphic treatment) impacted customer satisfaction with the encounter and, critically, whether this effect was moderated by customer anger (i.e., H1a). Because the chatbot was anthropomorphic and this could not be varied experimentally, we focused on the anthropomorphic treatment of the chatbot. If a customer treats a chatbot in a more human-consistent way, then we assume that is a consequence of a customer having more anthropomorphic thoughts resulting from perceiving the chatbot as more anthropomorphic. Specifically, we operationalized anthropomorphic treatment as the extent to which customers used the chatbot's name in their text-based conversation. As a name makes an object more anthropomorphic ([68]), the use of the chatbot's name indicates treating it as more human and serves as a reasonable proxy for anthropomorphic treatment.
Data were provided by a major international mobile telecommunications company. The data set covers 1,645,098 lines of customer text entries from 461,689 unique customer chatbot sessions that took place between September 2016 and August 2017 in one European country served by this company. At the end of each session, customers were given the option to rate their satisfaction with the chatbot encounter from one to five stars. Approximately 7.5% of sessions were rated (34,639 out of 461,689). In addition, for each line of customer text entered, there were metadata from the underlying chatbot NLP system that indicated the system's confidence in it having correctly "understood" each line of customer input, which was expressed as a percentage and termed the "bot recognition rate." We used the 1–5 satisfaction rating as the dependent variable. The distribution of this variable is shown in Figure 2, and the mean (SD) satisfaction rating was 2.16 (.79). We controlled for quality of the chatbot experience using the bot recognition rate, drawing on the assumption that for a given chatbot session, a higher average and lower variance in recognition rate indicated that the chatbot consistently understood more of a customer's inputs, which likely meant that the customer had an overall better communication experience.
Graph: Figure 2. Distribution of user satisfaction ratings following interaction with anthropomorphic service chatbot.
We processed chat transcript data (i.e., unstructured text) using the dictionary-based Linguistic Inquiry and Word Count (LIWC) package[ 4] ([52]) to classify each consumer text entry with respect to anger and to build our measure of the extent to which each customer treated the chatbot anthropomorphically.
In line with our theorizing, anger was the key emotion in customers' inputs to the chatbot. Our measure of anger was the corresponding LIWC item ("anger") that indicates the proportion of words in the input that are classified as being associated with anger. To arrive at the session-level measure, we averaged the LIWC anger value of each customer input within a given session. Unfortunately, this implicitly assumes that anger is exogenous by ignoring the initial emotional state of the customer and the dynamics of the consumer–bot exchange. We subsequently examine the robustness of our results when relaxing this assumption.
The chatbot was anthropomorphic because it had been endowed with extensive humanlike features (e.g., name, avatar, likes/dislikes). As we had no control over this, we instead aimed to measure anthropomorphic treatment, or the extent to which customers treated the chatbot in a humanlike manner.
There are several possible measures to derive for anthropomorphic treatment. Our approach was to use a simple measure based on the frequency of use (or nonuse) of the chatbot's name, assuming that if a customer used the chatbot's name, then they were treating it in a more humanlike manner than if they did not use it. Thus, our measure of anthropomorphic treatment was the total number of times in a customer's chatbot session that they used the chatbot's name. Repetition of the chatbot's name may also be an implicit acknowledgement by a customer of the chatbot's agency, which is another key indicator of humanlike treatment. Examples of this are customer inputs such as, "Hello [Bot Name], my name is [Customer Name], can you please help me with my bill?" and finishing a conversation with "Thank you [Bot Name]." The mean (SD) of this count was.032 (.178), ranging from zero to six times the bot's name was used per user session.
As noted previously, the chatbot's NLP system produced a confidence value, expressed as a percentage, of the likelihood that it correctly understood customer text input. The average and standard deviation of these values within a user session provide control variables for the quality and consistency of that customer experience. The mean (SD) of this value was 73.09 (23.6).
As a control, we captured the number of times the customer interacted with the chatbot in each session (i.e., the number of customer inputs per session). The mean (SD) was 3.56 (4.21), with a range of 1 to 491.
Finally, we controlled for the type of interaction the customer had with the service chatbot. This categorization was used by the chatbot itself, retained as metadata, connecting the requests received with a broad categorization of different types of service encounters: General Dialog, Questions and Answers, Providing Links, Frequently Asked Questions, and Feedback. Examples of individual dialogs include "Invoices," "SIM Card Activation," and "PIN Recovery."
Our goal was to estimate the extent to which anthropomorphic treatment, anger, and their interaction affected customer satisfaction. Considering that satisfaction was measured on a 1–5 scale (with five being the highest level of satisfaction), but with a great deal of mass in the distribution at the scale midpoint and both endpoints, we treated this as an ordinal variable and analyzed it using an ordinal logistic regression. Thus, we accounted for the potential for heterogeneity in distances between scale points. Vitally, we econometrically handled the obvious potential for a selection bias because only 7.5% of all customer chat sessions in our data set included a satisfaction rating. Thus, our analysis is based on the 34,639 chat sessions for which we had a satisfaction measure, but we make use of all 461,689 chat sessions in our treatment of endogenous sample selection.
To account for the ordinal nature of our data and the sample selection, we used an extended ordinal probit model, which estimates the probit selection and the ordinal satisfaction ratings equations simultaneously and with correlated errors ([26]).[ 5] The first equation was a binary probit model for leaving a satisfaction rating ( 1) or not (0), as described in Equation 1 (with i denoting the chat session and error esi). The second equation was an ordered probit model, as shown in Equation 2 (with i denoting the chat session and error ei). The error terms in Equations 1 and 2 were correlated. Note that in Equation 1 we used bot interaction type as an exclusion restriction because it impacts the likelihood of leaving a rating, P(Feedback = 1)i, but not the satisfaction rating, Ratingi.
Graph
( 1)
Graph
( 2)
Table 1 reports descriptive statistics and correlations. Table 2 reports results from the model described previously. First, considering the selection model in Table 2, we see that the type of customer–bot interaction (which we use as an exclusion restriction) is a significant predictor of the likelihood of the customer providing feedback to the firm, especially when, compared with general conversation (the base case in the model), the bot is focused on eliciting feedback (α6-feedback = 2.973, p <.001). Anger had a negative and significant effect on the likelihood of providing feedback (α2 = −.008, p <.05), whereas the number of exchanges with the bot during the session had a positive and significant effect (α5 = .033, p <.001).
Graph
Table 1. Descriptive Statistics in Study 1.
| Mean | SD | 1 | 2 | 3 | 4 | 5 | 6 |
|---|
| 1. Rating | 2.165 | .799 | — | — | — | — | — | — |
| 2. Anthropomorphic treatment | .031 | .177 | .05 | — | — | — | — | — |
| 3. Anger | .064 | 1.706 | .06 | .05 | — | — | — | — |
| 4. Bot language recognition rate | 73.09 | 23.60 | .00 | .07 | .01 | — | — | — |
| 5. Bot language recognition variance | 21.16 | 19.39 | .04 | .09 | .02 | .01 | — | — |
| 6. Number of interactions with bot | 3.563 | 4.218 | .01 | −.03 | −.01 | .02 | −.02 | — |
1 Notes: Boldface indicates p < .05.
Graph
Table 2. Probit Selection Model—Likelihood of Customer Providing a Rating After Interaction and Ordinal Probit—Customer Rating (Study 1).
| Variable | Coefficient | z-Value |
|---|
| Intercepts | | |
| 1|2 | −.0215 | −.15 |
| 2|3 | 1.0421*** | 6.04 |
| 3|4 | 1.7599*** | 9.24 |
| 4|5 | 1.8442*** | 9.56 |
| Rating: Second-Stage Model | | |
| Anthropomorphic treatment | −.0554 | −1.00 |
| Anger | −.0015 | −.07 |
| Anthropomorphic treatment × Anger | −.1665* | −1.96 |
| Bot language recognition rate | .0126*** | 12.81 |
| Bot language recognition variance | .0132*** | 17.30 |
| Number of interactions with bot | .0064† | 1.71 |
| P(Feedback): First-Stage Model | | |
| Anthropomorphic treatment | .0152 | .65 |
| Anger | −.0080* | −2.16 |
| Bot language recognition rate | .0002 | .97 |
| Bot language recognition variance | −.0051*** | −17.05 |
| Number of interactions with bot | .0334*** | 18.27 |
| Interaction type | | |
| FAQ | .7680*** | 19.77 |
| Feedback | 2.9731*** | 6.84 |
| Link | −.2184 | −1.46 |
| Question and answer | .0881*** | 3.13 |
| Constant | −1.6001*** | −48.03 |
| ρ | −.4921*** | −10.86 |
| Number of observations | 461,689 |
| Akaike information criterion | 74,082 | |
- 2 †p < .10.
- 3 *p < .05.
- 4 **p < .01.
- 5 ***p < .001.
Next, considering the main model for satisfaction ratings in Table 2, after accounting for the likelihood of providing feedback, both main effects of anthropomorphic treatment and anger were nonsignificant (β1 = −.055, n.s.; β2 = −.002, n.s.). However, their interaction was significant and negative (β3 = −.167, p = .05). This was after controlling for the technical performance of the bot during that session (recognition rate mean and variance) and the number of customer interactions in a session. Probing this interaction revealed the hypothesized effects across the distribution of consumer anger scores. When anger is higher (1 SD above the mean), the marginal effect of anthropomorphic treatment on satisfaction rating is significant and negative (β1 = −.350, p = .02), consistent with H1a. Interestingly, we also found that when anger was lower, but still present (1 SD below the mean), the marginal effect of anthropomorphic treatment on satisfaction rating remained negative, albeit with a smaller effect size than in the higher anger case (β1 = −.329, p = .02). Thus, it appears that the mere presence of anger can result in a negative relationship between anthropomorphic treatment and satisfaction. Further probing this interaction, we found that when anger was zero (i.e., not at all present), the marginal effect of anthropomorphic treatment on satisfaction rating was nonsignificant (β1 = .011, p = .32).
We were restricted in our analysis of Study 1 given that the data were provided as an outcome of the firm's operations and the conditions around each customer could not be assigned or manipulated. We could not control whether a customer provided feedback, the level of anthropomorphic treatment, or the customers' level of anger upon entering the chatbot interaction. While the first two limitations were addressed with a selection model and by exploiting variance in exhibited behavior, the levels of anger were taken as a given.
As a robustness check, we instead considered anger as a binary treatment effect and estimated an additional augmented model that accounted for the ordinal nature of ratings, the selection bias in providing feedback, and the anger of customers as a treatment condition. The inherent weakness of this model comes from a loss of information by dichotomizing anger into a binary (angry/not angry) condition, thus losing the nuance of levels of anger interacting with the level of anthropomorphic treatment. However, as a check, it shows the robustness of results to the specification of anger as an endogenous component of the customer experience, and for the estimation of drivers of anger, again with correlated errors (for the outcome, selection, and anger).
The model and results are presented in Web Appendix B. The endogenous anger treatment was positively but weakly correlated with the decision to provide feedback (r = .047) and negatively correlated with satisfaction rating (r = −.449). In the endogenous anger condition model, anthropomorphic treatment was positive and significant (γ1 = .289, p < .001). Critically, in the model for satisfaction rating, anthropomorphic treatment was nonsignificant for customers in the nonangry treatment (β1a = −.059, n.s.) and negative and significant for customers in the angry treatment (β1b = −.573, p = .04). We confirmed the significant difference between those two groups using a Wald test (χ2( 2) = 6.62, p = .04), which highlights that even if we model anger as a binary outcome of an endogenous process, our conclusions remain essentially the same.
In addition, we test whether alternative negative emotions (anxiety and sadness) or a positive mood valence meaningfully explain our customer satisfaction ratings or interact with the degree of chatbot anthropomorphic treatment. A reestimated extended ordinal probit model containing additional emotions is presented in Web Appendix C. While sadness is a significant predictor in our first stage model of feedback selection, no other negative emotions were significant. However, positive valence interacts meaningfully with anthropomorphic treatment (β = .0508, p = .001) in explaining satisfaction, consistent with prior research showing positive consequence of anthropomorphism. But, importantly, the hypothesized effect between anger and anthropomorphic treatment remains unchanged (β = −.1617, p = .05).
By leveraging real-world data from customers actively engaging with a chatbot across numerous chat sessions, we find initial evidence in support of H1a. An increase in the average level of anger exhibited by the consumer during their session resulted in a lower level of satisfaction with the service encounter, but only when the chatbot was treated anthropomorphically. In situations where the bot was not treated anthropomorphically, higher levels of anger did not meaningfully affect consumer satisfaction. Of course, this study has limitations. First, all customers were presented with the same highly anthropomorphic chatbot, so we had to rely on the variance in customers' anthropomorphic treatment of the bot, as opposed to variation in chatbot anthropomorphism, per se. Second, we initially assumed that customers entered the chat angry, independent from their exchange with the chatbot; however, our robustness checks confirm that anger is not strictly exogenous but also arose out of characteristics of the exchange with the bot (e.g., the number of exchanges, variance in language recognition). Finally, both anthropomorphic treatment and anger were measured from customer behaviors, rather than manipulated. These limitations motivated the four follow-up experiments.
The purpose of Study 2 was to test our theory under a controlled experimental setting and further show that, for angry customers, chatbot anthropomorphism has a negative effect on customer satisfaction. Accordingly, this study manipulated both chatbot anthropomorphism (via the presence/absence of anthropomorphic traits in the chatbot) and customer anger, allowing us to infer a causal relationship on satisfaction. In addition, careful chatbot selection enabled us to rule out idiosyncratic features of the Study 1 chatbot. Specifically, two of its specific features pose potential confounds in trying to generalize the results. First, it was clearly female, and previous research suggests that female service employees are more often targets of expressed frustration and anger from customers than are male service employees ([59]). In addition, she had a smiling expression, which is incongruent with the emotional state of participants who were angry, and such affective incongruity may cause a negative reaction and lower satisfaction.
We pretested avatars to select one that was both gender and affectively neutral. Twenty-five participants from Amazon Mechanical Turk (MTurk) evaluated a series of avatars on bipolar scales assessing both gender ("definitely male–definitely female") and warmth ("extremely cold–extremely warm") and indicated their agreement with one seven-point Likert item: "This avatar has a neutral expression." Drawing on the results of this pretest, we used the avatar pictured in Web Appendix D for the anthropomorphic chatbot condition. Specifically, our analysis confirmed that this avatar was neutral in both gender and warmth, with scores that did not significantly differ from the corresponding scale midpoints (Mgender = 3.64, t(24) = −1.23, p = .23; Mwarmth = 3.80, t(24) = −1.16, p = .26) and agreement with the neutral expression item was significantly above the midpoint (M = 5.76, t(24) = 7.80, p < .001).
We also wanted to choose a gender-neutral name for the anthropomorphic chatbot. Twenty-seven participants from MTurk evaluated a series of names on a seven-point bipolar scale assessing gender ("definitely male–definitely female"). From the results, we chose the name "Jamie," which was not significantly different from the midpoint (M = 3.89, t(26) = −.68, p = .50).
We created two customer service scenarios (neutral vs. anger) to use in this study (for the full scenarios used in both conditions, see Web Appendix E). In the neutral condition, the scenario described how the participant had purchased a camera for an upcoming trip, but upon receipt, the camera was broken. After searching the website, they diagnosed the issue as a problem with the lens and read about how to exchange the camera. It must be mailed back to the company before receiving a new camera, which is expected to arrive after they depart for a trip, the reason they wanted it originally.
In the anger condition, the scenario contained additional details designed to evoke anger. The original camera shipping was delayed, diagnosing the issue was time consuming, and they already tried to contact customer service and were placed on hold and passed from one representative to another. To ensure this scenario was successful in invoking anger compared with the neutral condition but did not differ in realism, we conducted a pretest on MTurk. Fifty participants were randomly assigned to read one of the two scenarios and then indicated how angry the scenario would make them feel (two items: "This situation would leave me feeling angry [frustrated]"; r = .65) and how realistic they found the scenario (two items: "How realistic [true-to-life] is this scenario?"; r = .61) on seven-point Likert scales (1 = "not at all," 7 = "extremely"). Our analysis confirmed that those in the angry condition reported significantly greater feelings of anger than those in the neutral condition (Mneutral = 5.46 vs. Manger = 6.06; F( 1, 48) = 4.31, p = .04). However, there was no significant difference in scenario realism (Mneutral = 5.48 vs. Manger = 5.69; F( 1, 48) = 2.09, p = .16).
We created two versions of the customer service chatbot (control vs. anthropomorphic). In the control condition, participants were told they would interact with "the Automated Customer Service Center," and in the anthropomorphic condition, they were told they would interact with "Jamie, the Customer Service Assistant." Furthermore, in the anthropomorphic condition, the chatbot featured the avatar selected from the pretest, and the chat text consistently used a singular first-person pronoun (i.e., "I") and appeared in quotation marks.
To ensure that this manipulation was successful, we conducted a pretest on MTurk. One hundred one participants were randomly assigned to one of the two chatbots and then indicated how anthropomorphic the chatbot was on nine seven-point Likert scales (adapted from [19]] and [38]]: "Please rate the extent to which [the Automated Customer Service Center/Jamie]: came alive (like a person) in your mind; has some humanlike qualities; seems like a person; felt human; seemed to have a personality; seemed to have a mind of his/her own; seemed to have his/her own intentions; seemed to have free will; and seemed to have consciousness"; α = .98). Analysis confirmed that those in the anthropomorphic condition reported significantly greater anthropomorphic thought (Mcontrol = 3.37 vs. Manthro = 4.82; F( 1, 99) = 16.64, p < .001).
Two hundred one participants (48% female; Mage = 37.29 years) from MTurk participated in this study in exchange for monetary compensation. The study consisted of a 2 (chatbot: control vs. anthropomorphic) × 2 (scenario emotion: neutral vs. anger) between-subjects design. Participants were randomly assigned to read one of the aforementioned scenarios (neutral or anger). Then, participants entered a simulated chat with either "the Automated Customer Service Center" in the control chatbot condition or "Jamie, the Customer Service Assistant" in the anthropomorphic condition.
In the simulated chat, participants were first asked to open-endedly explain why they were contacting the company. In addition to serving as an initial chatbot interaction, this question also functioned as an attention check, allowing us to filter out any participants who entered nonsensical (e.g., "GOOD") or non-English answers ([17]). Subsequently, participants encountered a series of inquiries and corresponding drop-down menus regarding the specific product they were inquiring about (camera) and issue they were having (broken and/or damaged lens). They were then given return instructions and indicated they needed more help. Using free response, they described their second issue and answered follow-up questions from the chatbot about the specific delivery issue (delivery time is too long) and reason for needing faster delivery (product will not come in time for a special event). Finally, participants were told that a service representative would contact them to discuss the issue further. The interaction outcome was designed to be ambiguous (representing neither a successful nor failed service outcome; however, we manipulate this outcome in Study 3). The full chatbot scripts for both conditions and images of the chat interface are presented in Web Appendices F and G.
Upon completing the chatbot interaction, participants indicated their satisfaction with the chatbot by providing a star rating (a common method of assessing customer satisfaction; e.g., [61]) between one and five stars, on five dimensions (α = .95): overall satisfaction, customer service, problem resolution, speed of service, and helpfulness. Lastly, participants indicated their age and gender and were thanked for their participation.
Four participants failed the attention check (entering a nonsensical response for the open-ended question), leaving 197 observations for analysis. Analysis of variance (ANOVA) results revealed a significant main effect of scenario emotion on satisfaction (i.e., averaged star rating on the five dimensions), in that those in the anger scenario condition were less satisfied than those in the neutral scenario condition (F( 1, 193) = 33.45, p < .001). Importantly, we found a significant chatbot × scenario emotion interaction on customer satisfaction (F( 1, 193) = 5.26, p = .02). Consistent with Study 1, a simple effects test revealed that participants in the anger scenario condition were less satisfied when the chatbot was anthropomorphic (M = 2.09) versus when it was not (M = 2.58; F( 1, 193) = 4.13, p = .04). For those in the neutral scenario, chatbot anthropomorphism had no significant influence on satisfaction, but satisfaction was directionally higher in the anthropomorphic condition (Mcontrol = 3.16 vs. Manthro = 3.44; F( 1, 193) = 1.46, p = .23). Figure 3 presents an illustration of means.
Graph: Figure 3. The effect of chatbot anthropomorphism and anger on customer satisfaction (Study 2).
Whereas Study 1 provides initial support for the interactive effect of anthropomorphism and anger on customer satisfaction, Study 2 tests our theorizing in a controlled experimental design. This allowed us to more definitively conclude that when customers are angry, anthropomorphic traits in a chatbot lower customer satisfaction with the chatbot (consistent with H1a)[ 6] and rule out alternative explanations based on the chatbot's gender or expression. While not central to our main theorizing, we ran an identical study manipulating sadness instead of anger. Both anger and sadness are negative emotions, but anger represents an activating emotion, whereas sadness is a deactivating emotion ([15]; [41]). We predicted that only angry customers are activated to respond negatively to anthropomorphic chatbots due to their need to overcome obstacles, blame others, and respond punitively to expectancy violations ([23]; [43]; [42]). Interestingly, participants in the sad condition were more satisfied when the chatbot was anthropomorphic versus when it was not (Mcontrol = 1.90 vs. Manthro = 2.53; F( 1, 188) = 8.49, p < .01), which is consistent with prior literature ([27]; [72]) that demonstrates the positive effect of anthropomorphic chatbots in other situations. Full details are available in Web Appendix I.
There were three main goals of Study 3. First, while our previous study induced anthropomorphism via a simultaneous combination of visual and verbal cues (with an avatar and first-person language, respectively), the current study aimed to show that the effect diminishes with the reduction of anthropomorphic traits. Thus, we remove the visual trait of anthropomorphism (i.e., the avatar) and anticipate the negative effect of anger to attenuate, providing further support that the degree of humanlikeness is responsible for driving the effect. Second, we wanted to test H1b by exploring whether the negative effect of anthropomorphism for angry customers would extend to influence their evaluations of the company itself. Finally, we wanted to provide initial evidence that expectancy violations are responsible for these observed negative effects. To do so, we examined whether the outcome of the chat interaction—namely, if the chatbot was able to indubitably resolve the customer's concerns—could serve as a boundary condition. We predicted that if the chatbot could meet the high expectations of efficacy, the negative effect of anthropomorphism should dissipate.
We selected a new avatar in this study to increase the robustness of our examination and generalizability of our findings. Twenty-six participants from MTurk evaluated a series of avatars as in the Study 2 pretest. Our analysis confirmed that the avatar (pictured in Web Appendix D) was considered neutral in both gender and warmth (Mgender = 4.04, t(25) = .12, p = .90; Mwarmth = 4.27, t(25) = 1.32, p = .20) and had a neutral expression (M = 5.12, t(25) = 4.35, p < .001).
As in Study 2, we created two customer service scenarios (neutral vs. anger; for the full scenarios, see Web Appendix J). In the neutral condition, the scenario described a situation where the participant was interested in buying a camera from the company, "Optus Tech," with a specific feature (advanced video stabilization). After searching the website, it was difficult to tell whether Optus Tech's camera had this feature. In addition, the delivery window was wide, which meant that the expected delivery may or may not occur after they depart for a trip, the whole reason they wanted the camera.
In the anger emotion condition, there were additional details designed to evoke anger: researching the feature was time consuming, they already tried to contact customer service and were placed on hold and passed from one representative to another, the representative could not answer their question, and they had to call a second time to ask about shipping. To ensure that this scenario was successful in invoking anger compared with the neutral condition, we pretested 48 MTurk participants who were randomly assigned to read one of the two scenarios and then indicated both their anticipated feelings of anger and how realistic they found the scenario (as measured in the Study 2 pretest; r = .77 and r = .63, respectively). Indeed, those in the anger condition reported significantly greater anticipated feelings of anger than those in the neutral condition (Mneutral = 3.63 vs. Manger = 5.46, F( 1, 46) = 18.00, p < .001). However, there was no significant difference in how realistic participants found the two scenarios (Mneutral = 5.50 vs. Manger = 4.92, F( 1, 46) = 2.54, p = .12). We only used the anger scenario in this study, but an upcoming study used both scenarios.
We created three versions of the customer service chatbot: control, verbal anthropomorphic, and verbal + visual anthropomorphic. The first and last chatbots were similar to Study 2 except using the new avatar. The additional chatbot used the verbal anthropomorphic traits (i.e., the bot introduced itself as Jamie and used first-person language in quotations) but not the visual trait (i.e., the avatar). One hundred twenty-one MTurk participants were randomly assigned to one of the three chatbots: the Automated Customer Service Center (control chatbot condition), Jamie without an avatar (verbal anthropomorphic condition), or Jamie with an avatar (verbal + visual anthropomorphic condition) and then indicated how anthropomorphic the chatbot was on nine seven-point Likert scales (as in Study 2; α = .97). We coded the control, verbal anthropomorphic, and verbal + visual anthropomorphic conditions as 0, 1, and 2, respectively, to represent the strength of the anthropomorphic manipulation. Results demonstrated that the linear trend was significant (Mcontrol = 2.38 vs. Mverbal = 4.14 vs. Mverbal + visual = 4.39; F( 1, 118) = 39.16, p < .001).
Four hundred nineteen participants (61% female; Mage = 38.50 years) from MTurk participated in this study in exchange for monetary compensation. This study consisted of a 3 (chatbot: control vs. verbal anthropomorphic vs. verbal + visual anthropomorphic) × 2 (scenario outcome: ambiguous vs. resolved) between-subjects design. First, all participants read the anger scenario and then entered a simulated chat with the chatbot. In the ambiguous outcome condition, participants encountered a series of questions and corresponding drop-down menus regarding the specific product (camera) and feature (advanced video stabilization) they were inquiring about. They were then given basic product information about the feature that was purposefully ambiguous. Then, participants indicated they needed more help. Using free response, they described their second issue and answered follow-up questions from the chatbot about the specific delivery issue (delivery window/timing) and reason for needing faster delivery (product will not come in time for a special event). Participants were told that a service representative would contact them to discuss the issue further. In the resolved outcome condition, there were two critical differences: participants were directly given the product feature information that resolved their query and explicitly informed of the specific delivery time information, which confirmed they would receive their delivery in time for their special event. The entire chatbot scripts for both conditions and images of the interface are presented in Web Appendices K and L.
Upon completing the interaction, participants evaluated the company, Optus Tech, on four seven-point bipolar items (α = .95): "unfavorable–favorable," "negative–positive," "bad–good," and "unprofessional–professional." As a manipulation check for the scenario outcome, participants responded to three items: "My question was sufficiently answered," "My problem was appropriately resolved," and "I got the help I needed" (1 = "strongly disagree," and 7 = "strongly agree"; α = .97). To assess whether participants knew they were interacting with a chatbot (vs. a human), we asked participants to indicate the extent to which they felt they interacted with a human versus an automated chatbot (1 = "definitely a real live human," 7 = "definitely an automated chatbot".[ 7] Participants indicated demographics and were thanked for participating.
Fifty-two participants failed the attention check (entering a nonsensical response for the open-ended question), leaving 365 observations for analysis.
Participants in the resolved condition indicated that their problem was more appropriately resolved (M = 6.36) than participants in the ambiguous condition (M = 4.32; t(363) = 12.40, p < .001), indicating a successful manipulation.
Unsurprisingly, ANOVA results revealed a significant main effect of scenario outcome on company evaluation, such that participants reported lower evaluations of the company when the outcome was ambiguous (M = 4.68) versus when it was resolved (M = 5.36; F( 1, 359) = 18.44, p < .001). There was no main effect of chatbot anthropomorphism (F( 1, 359) = 1.38, p = .25). Importantly, there was a marginally significant chatbot anthropomorphism × anger scenario interaction on company evaluation (F( 2, 359) = 2.64, p = .07). A simple effects test revealed that there was no significant difference between the chatbot conditions when the outcome was resolved (F < 1). This provides some evidence that effectively meeting expectations eliminates the negative effect, which is conceptually consistent with H2, because if high preinteraction expectations of efficacy are met, there should be no resultant expectancy violations. However, there was a significant difference when the outcome was ambiguous (F( 2, 359) = 3.78, p = .02). Planned contrasts revealed when the outcome was ambiguous, participants reported lower company evaluations when the chatbot was verbally and visually anthropomorphic (M = 4.28) versus the control (M = 5.06; t(359) = 2.75, p < .01), providing evidence in support of H1b. However, the verbal anthropomorphic condition (without an avatar) did not significantly differ from the control (Mverbal = 4.69 vs. Mcontrol = 5.06; t(359) = 1.36, p = .17) or the verbally and visually anthropomorphic condition (vs. Mverbal + visual = 4.28; t(359) = 1.48, p = .14). Because the verbal anthropomorphic condition both theoretically and empirically fell between the two other conditions, we tested whether our anthropomorphism manipulation demonstrated a linear trend. We coded the control, verbal anthropomorphic, and verbal + visual anthropomorphic conditions as 0, 1, and 2, respectively, to represent the strength of the anthropomorphic manipulation. Results demonstrated that the linear trend was not significant when the outcome was resolved (F < 1) but was significant when the outcome was ambiguous (F( 1, 359) = 7.55, p < .01). These findings suggest that visual and verbal anthropomorphic traits likely produce an additive effect, where multiple traits lead to greater anthropomorphic thought, and accordingly results in lower company evaluations (at least in the case of angry consumers, which we exclusively examined in this study). Figure 4 presents an illustration of means.
Graph: Figure 4. The effect of chatbot anthropomorphism and anger on company evaluation (Study 3).
Study 4 serves two key purposes. First, we extend our investigation to an even further downstream negative outcome by examining purchase intentions (H1c). Second, we build on the findings of Study 3 and directly test our proposed underlying process: expectancy violations driven by preperformance expectations (H2). Specifically, we predict anthropomorphism increases preperformance expectations that a chatbot would display greater agency and performance. While people in a neutral state will perceive the expectancy violation, they are less motivated to retaliate or respond punitively. Angry people, in contrast, punish the company by lowering their purchase intentions.
One hundred ninety-two participants (55% female; Mage = 37.31 years) from MTurk participated in exchange for payment. This study consisted of a 2 (chatbot: control vs. anthropomorphic) × 2 (scenario emotion: neutral vs. anger) between-subjects design. Participants were randomly assigned to read one of the neutral or anger information search scenarios pretested in the prior study. Then participants were told they were about to enter a chat with either the Automated Customer Service Center (control condition) or Jamie (anthropomorphic condition). At this point, all participants saw the brand logo for Optus Tech, but those in the anthropomorphic chatbot condition also saw the avatar.
Next, participants indicated their preinteraction efficacy expectations regarding the chatbot's upcoming performance on four seven-point Likert items ("I expect the Automated Service Center/Jamie to: do something for me; take action; be proactive in resolving my issues; say things to calm me down"; α = .89). Participants completed the same interaction as in the ambiguous condition from Study 3 and indicated their purchase intentions for the camera on two seven-point Likert items: "I would buy the camera from Optus Tech," and "I would try to find a different company to buy the camera from" (the latter was reverse-coded; r = .65). Afterward, participants rated their postinteraction assessment of the chatbot's efficacy, on four seven-point Likert items that corresponded to the preinteraction items ("I felt the Automated Service Center/Jamie: did a lot for me; took action; was proactive in resolving my issues; said things to calm me down"; α = .92). Lastly, participants indicated their age and gender and were thanked for their participation.
Twenty-one participants failed the attention check used in prior studies, leaving 171 observations for analysis. ANOVA results revealed a significant main effect of anger on purchase intentions, where participants in the anger scenario condition reported lower purchase intentions than those in the neutral scenario condition (F( 1, 167) = 20.04, p < .001). Consistent with the pattern of results predicted in H1c, there was a significant chatbot anthropomorphism × anger scenario interaction on purchase intentions (F( 1, 167) = 4.29, p = .04). A simple effects test revealed that participants in the anger scenario condition reported lower purchase intentions when the chatbot was anthropomorphic (M = 2.73) versus when it was not (M = 3.57; F( 1, 167) = 5.79, p = .02). For those in the neutral scenario, the chatbot had no significant influence on purchase intentions. Figure 5 presents an illustration of means.
Graph: Figure 5. The effect of chatbot anthropomorphism and anger on purchase intentions (Study 4).
We predicted that encountering an anthropomorphic chatbot at the start of the service experience would increase participants' preinteraction expectations about the efficacy of the chatbot, relative to the control chatbot. However, the postinteraction efficacy assessments of the chatbots should not differ because they performed equally, resulting in greater expectancy violations for anthropomorphic chatbots (H2).
To assess this hypothesis, we ran a repeated-measures ANOVA with chatbot anthropomorphism as the between-subjects variable and time (preinteraction expectations at Time 1 and postinteraction assessments at Time 2) as the within-subjects factor. We did not find a significant overall main effect of chatbot anthropomorphism on efficacy (F( 1, 169) = .91, p = .34). Importantly, there was a significant interaction of chatbot anthropomorphism and time (F( 1, 169) = 7.31, p = .01). Probing this interaction, as we expected, preinteraction expectations of the chatbot's efficacy at Time 1 were significantly higher in the anthropomorphism condition than in the control condition (Mcontrol = 4.94 vs. Manthro = 5.50; F( 1, 169) = 6.91, p = .01), but there was no difference in the postinteraction assessments at Time 2 (Mcontrol = 4.09 vs. Manthro = 3.88; F < 1). These results are consistent with the logic that a greater expectancy violation is more likely in the anthropomorphism condition than in the control because of inflated preinteraction expectations of chatbot efficacy stemming from more humanized traits.
We also calculated an expectancy violation score for each participant by subtracting their postinteraction assessment score at Time 2 from their preinteraction expectation score at Time 1 ([45]). As we expected, an ANOVA with chatbot anthropomorphism and anger scenario as predictors and expectancy violation as the dependent variable produced only a significant main effect of chatbot anthropomorphism on expectancy violation (Mcontrol = .85 vs. Manthro = 1.62; F( 1, 169) = 7.31, p = .01).
Importantly, our theorizing suggests that anthropomorphism inflates preinteraction expectations of chatbot efficacy for all customers. Yet, angry customers are more motivated than nonangry customers to respond punitively by lowering their purchase intent. Accordingly, we performed a moderated mediation analysis based on 10,000 bootstrapped samples ([29], Model 15). While the index of moderated mediation did not reach significance (indirect effect = .0279; 95% confidence interval [CI]: [−.0778,.1562]), we examined the separate indirect effects at each emotion condition based on our a priori predictions ([ 2]; [30]). In other words, while we did not have predictions for what might drive purchase intention for those in the neutral condition, we did predict that for angry customers, lowered preinteraction expectations would explain the decreased purchase intention. As per our theorizing, results demonstrated that for individuals in the anger condition, preinteraction expectations mediated the effect of chatbot anthropomorphism on purchase intention (indirect effect = .0675; 95% CI: [.0012,.1707]). However, for participants in the neutral condition, the indirect effect was not significant (indirect effect = .0396; 95% CI: [−.0408,.1468]). These results suggest that, as we predicted, the inflated preinteraction expectation of efficacy caused by the anthropomorphic chatbot is the underlying mechanism lowering purchase intentions for angry participants.
Study 4 demonstrated that anthropomorphic chatbots result in lower purchase intentions when customers are angry by elevating preinteraction expectations of efficacy. Yet, it is theoretically and managerially important to understand how this effect can be remedied. Some companies attempt to explicitly temper customer expectations of their chatbots. For example, Slack's chatbot introduces itself by explaining that, "I try to be helpful (But I'm still just a bot. Sorry!)" ([65]). Study 5 explores whether explicitly lowering customer expectations of anthropomorphic chatbots prior to the interaction effectively reduces negative customer responses.
For the Study 5 pretest, 31 participants from MTurk evaluated a series of avatars as in the prior pretests. Our analysis confirmed that the avatar (see Web Appendix D) was considered neutral in both gender and warmth (Mgender = 5.55, t(30) = 1.52, p = .14; Mwarmth = 4.97, t(30) = −.19, p = .85) and had a neutral expression (M = 5.94, t(30) = 8.91, p < .001).
Three hundred two participants (52% female; Mage = 40.78 years) from MTurk participated in exchange for monetary compensation. The study consisted of a 2 (chatbot: control vs. anthropomorphic) × 2 (expectation: baseline vs. lowered) between-subjects design. All participants read the anger information search scenario. Afterward, participants saw they would chat with either "the Automated Customer Service Center" in the control or "Jamie, the Customer Service Assistant" in the anthropomorphic chatbot condition. In the lowered expectation condition, they also read, "The Automated Customer Service Center/Jamie, the Customer Service Assistant will do the best that it/I can to take action but sometimes the situation is too complex for it/me (it's/I'm just a bot) so please don't get your hopes too high."
Participants then indicated their preinteraction efficacy expectations (as in Study 4; α = .89), completed the product information interaction and evaluated the company (as in Study 3; α = .97), rated their postinteraction assessment of the chatbot's efficacy (as in Study 4; α = .92), answered demographic questions, and were thanked for their participation.
Consistent with the manipulation intention, there was a main effect of anthropomorphism on preinteraction expectations (F( 1, 298) = 4.36, p = .04), where participants had higher expectations when the chatbot was anthropomorphic (M = 4.53) compared with the control (M = 4.18). There was also a main effect of expectations on preinteraction expectations (F( 1, 298) = 45.61, p < .001), where, as we intended, lowering expectations resulted in lower preinteraction expectations (M = 3.78) than in the baseline expectation condition (M = 4.93). There was a significant chatbot × expectation interaction on preinteraction expectations (F( 1, 298) = 7.00, p < .01). Simple effects tests revealed that in the baseline expectation conditions, the people in the anthropomorphic condition had higher expectations of chatbot efficacy than in the control condition (Mcontrol = 4.53 vs. Manthro = 5.34; F( 1, 298) = 11.35, p = .001). Yet, in the low-expectation conditions, there was no difference between the preinteraction expectations of efficacy for the anthropomorphic and control chatbot (Mcontrol = 3.83 vs. Manthro = 3.73; F < 1). For postinteraction evaluations, consistent with predictions, there were no significant differences (i.e., no main effect of anthropomorphism, no main effect of expectation, and no interaction between anthropomorphism and expectation). This indicates preinteraction expectations are responsible for changes to expectancy violations.
Expectancy violations were calculated by subtracting postinteraction evaluations from preinteraction expectations, with higher numbers indicating greater violations. There is a main effect of anthropomorphism on expectancy violations (F( 1, 298) = 7.38, p < .01), where participants indicated greater expectancy violations when the chatbot was anthropomorphic (M = .65) compared with the control (M = .10). There was also a main effect of the expectation manipulation on expectancy violations (F( 1, 298) = 36.73, p < .001), where there were greater expectancy violations in the baseline expectation condition (M = .99) compared with the lowered-expectation condition (M = −.24). Importantly, there was also a significant chatbot × expectation interaction on expectancy violations (F( 1, 298) = 13.26, p < .001). As we expected, when participants had the baseline expectation (i.e., no information given), they experienced greater expectancy violations driven by preinteraction expectations when the chatbot was anthropomorphic compared with the control (Mcontrol = .35 vs. Manthro = 1.63; F( 1, 298) = 20.47, p < .001). Yet when participants were told to have lower expectations, there was no difference between the expectancy violations for the anthropomorphic chatbot and the control (Mcontrol = −.14 vs. Manthro = −.33; F < 1).
The ANOVA results revealed a marginal main effect of anthropomorphism on company evaluation; participants reported marginally lower evaluations of the anthropomorphic chatbot (M = 4.11) versus control (M = 4.44; F( 1, 298) = 2.86, p = .09). There was no main effect of expectation (F < 1). There was a significant chatbot × expectation interaction on company evaluation (F( 1, 298) = 4.35, p = .04). Consistent with prior studies, a simple effects test revealed that participants in the baseline expectation condition rated the company lower when the chatbot was anthropomorphic (M = 3.90) versus when it was not (M = 4.63; F( 1, 298) = 4.13, p = .04). For those in the lowered-expectation condition, chatbot anthropomorphism had no significant influence on company evaluations (Mcontrol = 4.25 vs. Manthro = 4.32; F < 1), indicating that lowering customer expectations of anthropomorphic chatbots effectively mitigated the negative effect of anger on company evaluations. Figure 6 presents an illustration of means, and Web Appendix M provides additional analysis.
Graph: Figure 6. The effect of chatbot anthropomorphism and expectations on company evaluation (Study 5).
The deployment of chatbots as digital customer service agents continues to accelerate as the underlying machine learning technologies improve and as the practice becomes more common across industries. Customers are increasingly interacting with firms through chatbots, and there has been a significant push for more humanlike versions of such bots. Prior research has begun to demonstrate some negative implications of anthropomorphism in specific situations, including video games ([36]), gambling ([37]), and overcrowded environments ([53]), as well as for some types of people (agency-oriented customers; [38]). Yet, our research is the first to demonstrate the negative effect of anthropomorphism in the wider context of customer service and connect the use of these humanlike chatbots to negative firm outcomes.
We find, using a large data set of real-world customer interactions and four experiments, that anthropomorphic chatbots can harm firms. An angry customer encountering an anthropomorphic (s. nonanthropomorphic) chatbot is more likely to report lower customer satisfaction, lower overall evaluation of the firm, and lower future purchase intentions. This negative effect is driven by expectancy violations due to inflated preinteraction expectations of efficacy caused by the anthropomorphic chatbot. Angry customers respond more punitively to these expectancy violations compared with nonangry customers.
The decision to anthropomorphize a chatbot is a deliberate and strategic choice made by the firm. The current research shows that this choice has a significant impact on key marketing outcomes for a substantial (and increasing due to the pandemic; [57]; [60]) group of customers: specifically, those who are angry during the service encounter. As such, firms should attempt to gauge whether a customer is angry either before or early in the conversation (e.g., using NLP) and deploy a chatbot with an appropriate level of anthropomorphism or lack thereof. A less precise solution would be to assign nonanthropomorphic chatbots to customer service roles that tend to involve angry customers (e.g., customer complaint centers) while continuing to employ anthropomorphic agents in more neutral or promotion-oriented settings (e.g., searches for product information) due to their previously documented beneficial effects ([27]; [72]) and the current empirical evidence (i.e., Study 1, Web Appendix I). This strategic deployment of chatbots should help firms deliver better chatbot-mediated service experiences. Moreover, appropriate chatbot deployment can improve immediate customer satisfaction, company evaluations, and future purchase intentions (e.g., customer retention).
Given our finding that the negative effect of anthropomorphism for angry customers is driven by an expectancy violation due to inflated preinteraction efficacy expectations, another practical implication is that marketers should consider how to frame their customer service chatbots to customers. As our final study shows, if an anthropomorphic chatbot is deployed to angry customers, it is best to downplay its capabilities. Some companies seem to have intuited this, as illustrated by the aforementioned Slack bot example. Similarly, the Poncho weather app told people, "I'm good at talking about the weather. Other stuff, not so good" ([65]). Explicitly informing customers that they are conversing with an imperfect chatbot lowers preinteraction efficacy expectations that were inflated by anthropomorphic traits. Yet, this is not obvious to all companies; there are plenty of examples of chatbots that inadvertently increase preinteraction efficacy expectations. For example, Madi, Madison Reed's chatbot, is labeled a "genius" ([49]) and Tinka, T-Mobile's chatbot, is given a 18,456 IQ ([47]). Of course, Study 3 shows that meeting the high expectations for service can also reduce the negative impact of anthropomorphism. Thus, by utilizing these strategies, all customers can be handled well via AI technology.
Alternatively, firms could transfer angry customers directly to a real live person to assist them, thus avoiding an anthropomorphic chatbot–based expectancy violation entirely. Yet this option incurs additional costs and assumes that the human agent has greater agency and efficacy. While it is plausible that human agents will deliver higher quality service, in actuality, human agents suffer from constraints that limit their effectiveness. Thus, future research might address how people respond to chatbots compared with humans. It would be interesting to explore whether higher expectations of quality and agency would be compensated for by social norms of polite interactions and compassion for others.
In addition, anger may not be the only relevant emotion to consider managerially or theoretically. While our data indicate that anger is of primary importance and is the most commonly identified emotion in service contexts, it is possible that with more sophisticated language processing tools, other emotions, different sources of those emotions, and social conventions could become relevant. For example, it could be that anger remains relevant in customer service contexts but that the source of the anger, such as whether it arises from a lack of procedural or interactional fairness, also impacts the success of anthropomorphic bots ([ 7]). Thus, it is important for future research to continue to investigate how the complexities of emotion, sources of emotion, and social norms interact to influence the effectiveness of anthropomorphic digital customer service agents.
It is worth noting that there might be a point in the future when the conversational performance of AI becomes sufficiently advanced and its implementation so commonplace that expectancy violations simply cease to be a concern. In this future, chatbots might be capable of greater freedom of action, in addition to performing intuitive and empathetic tasks ([33]). In approaching such a point, the difference between the reactions of angry and nonangry customers would likely diminish until the groups are nondistinct, and anthropomorphism might cease to conditionally influence customer outcomes. However, this future does not appear to be imminent ([58]).
In the short and medium term, therefore, as firms experiment with conversational agents in a variety of customer-facing roles, it remains important to consider the anthropomorphic traits of the chatbot, including simple features such as naming (i.e., "Alexa"), language style, and the degree of embodiment, along with the specific customer contexts in which the interactions are likely to occur. Specific contexts can vary from traditional corporations to government (e.g., the Australian government chatbot), law (e.g., the "DoNotPay" chatbot), and psychotherapy (e.g., the "Woebot" chatbot). It is worthwhile to decide, and important for future research to explore, which chatbot is most appropriate for any given interaction, according to the chatbot's characteristics and the specific context.
Altogether, chatbots deliver a multitude of benefits to the business (e.g., scalability, cost reductions, control over the quality of interactions, additional customer data). As such, they will continue to be a valuable tool for marketers as the technology matures. Here, we have shown that the unconditional deployment of humanized chatbots leads to negative marketing outcomes, from dissatisfaction to lowered purchase intentions. However, with careful and conscientious implementation, considering the customer's emotional state (e.g., anger), firms can reap the benefit of this burgeoning technology.
sj-pdf-1-jmx-10.1177_00222429211045687 - Supplemental material for Blame the Bot: Anthropomorphism and Anger in Customer–Chatbot Interactions
Supplemental material, sj-pdf-1-jmx-10.1177_00222429211045687 for Blame the Bot: Anthropomorphism and Anger in Customer–Chatbot Interactions by Cammy Crolic, Felipe Thomaz, Rhonda Hadi and Andrew T. Stephen in Journal of Marketing
Footnotes 1 Praveen Kopalle
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Future of Marketing Initiative (FOMI) at Saïd Business School, University of Oxford.
4 While LIWC provides a validated approach to text classification, there are several more modern techniques within NLP that may be capable of the same task. However, each technique carries its own requirements, advantages, and disadvantages. State-of-the-art NLP for emotion detection would involve a combination of a contextualized word representation model, (e.g., Bidirectional Encoder Representations from Transformers [BERT]), which excels in language comprehension, followed by a classifier neural network. However, this approach requires a preclassified data set to train the model. Unfortunately, a training data set with our target label does not exist, and its absence motivated the need for LIWC's classification in the first place. See the alternative NLP techniques section in Web Appendix A for a more detailed discussion as well as a documentation of our attempts to apply the BERT technique to anger classification.
5 Alternatively, a Heckman two-step sample selection correction approach could be used, where the inverse Mills ratio from the first-stage probit selection model is entered as a covariate in a second-stage response regression. However, [26], p. 234) suggest that this is only appropriate in the case of a linear response with normally distributed errors, which is not the case here. For robustness, however, we redid this analysis using a two-step approach and inverse Mills ratio in the response equation, and the results were consistent.
6 We ran a supplementary study that conceptually replicates Study 2 with company evaluation as the dependent variable. Details of this study are presented in https://journals.sagepub.com/doi/suppl/10.1177/00222429211045687.
7 Across the three chatbot conditions, there were no significant differences in the extent to which participants believed they were chatting with a real human (Mcontrol = 5.76 vs. Mverbal = 6.05 vs. Mverbal + visual = 6.13; F(2, 364) = 2.01, p = .14). Importantly, all three scores were significantly higher than the midpoint (all ps < .001), indicating that participants knew they were interacting with an automated bot.
References Aggarwal Pankaj , McGill Ann L.. (2007), " Is That Car Smiling at Me? Schema Congruity as a Basis for Evaluating Anthropomorphized Products ," Journal of Consumer Research , 34 (4), 468 – 79.
Aiken Leona S. , West Stephen G.. (1991), Multiple Regression: Testing and Interpreting Interactions. Newbury Park, CA : SAGE Publications.
Anderson Eugene W. , Sullivan Mary W.. (1993), " The Antecedents and Consequences of Customer Satisfaction for Firms ," Marketing Science , 12 (2), 125 – 43.
Ask Karl , Landström Sara. (2010), " Why Emotions Matter: Expectancy Violation and Affective Response Mediate the Emotional Victim Effect ," Law and Human Behavior , 34 (5), 392 – 401.
Bickmore Timothy W. , Picard Rosalind W.. (2005), " Establishing and Maintaining Long-Term Human-Computer Relationships ," ACM Transactions on Computer-Human Interaction , 12 (2), 293 – 327.
Bitner Mary Jo. (2001), " Service and Technology: Opportunities and Paradoxes ," Managing Service Quality , 11 (6), 375 – 79.
Blodgett Jeffrey G. , Hill Donna J. , Tax Stephen S.. (1997), " The Effects of Distributive, Procedural, and Interactional Justice on Postcomplaint Behavior ," Journal of Retailing , 73 (2), 185 – 210.
8 Bodenhausen Galen V. , Sheppard Lori A. , Kramer Geoffrey P.. (1994), " Negative Affect and Social Judgment: The Differential Impact of Anger and Sadness ," European Journal of Social Psychology , 24 (1), 45 – 62.
9 Brackeen Brian. (2017), " How to Humanize Artificial Intelligence with Emotion ," Medium (March 30) , https://medium.com/@BrianBrackeen/how-to-humanize-artificial-intelligence-with-emotion-19f981b1314a.
Cadotte Ernest R. , Woodruff Robert B. , Jenkins Roger L.. (1987), " Expectations and Norms in Models of Consumer Satisfaction ," Journal of Marketing Research , 24 (3), 305 – 14.
Cardello Armand V. , Sawyer Fergus M.. (1992), " Effects of Disconfirmed Consumers Expectancy on Food Acceptability ," Journal of Sensory Studies , 7 (4), 253 – 77.
Chandler Jesse , Schwarz Norbert. (2010), " Use Does Not Wear Ragged the Fabric of Friendship: Thinking of Objects as Alive Makes People Less Willing to Replace Them ," Journal of Consumer Psychology , 20 (2), 138 – 45.
Choi Yung Kyun , Miracle Gordon E. , Biocca Frank. (2001), " The Effects of Anthropomorphic Agents on Advertising Effectiveness and the Mediating Role of Presence ," Journal of Interactive Advertising , 2 (1), 19 – 32.
Cosmides Leda , Tooby John. (2000), " Evolutionary Psychology and the Emotions ," in Handbook of Emotions , 2nd ed. New York : Guilford.
Cunningham Michael R. (1988), " What Do You Do When You're Happy or Blue? Mood, Expectancies, and Behavioral Interest ," Motivation and Emotion , 12 , 309 – 31.
De Akansha. (2018), " A Look at the Future of Chatbots in Customer Service ," ReadWrite (December 4), https://readwrite.com/2018/12/04/a-look-at-the-future-of-chatbots-in-customer-service/.
Dennis Sean A. , Goodson Brian M. , Pearson Chris. (2018), " MTurk Workers' Use of Low-Cost 'Virtual Private Servers' to Circumvent Screening Methods: A Research Note ," working paper , SSRN, https://ssrn.com/abstract=3233954.
De Visser Ewart J. , Monfort Samual S. , McKendrick Ryan , Smith Melissa A.B. , McKnight Patrick E. , Krueger Frank , et al. (2016), " Almost Human: Anthropomorphism Increases Trust Resilience in Cognitive Agents ," Journal of Experimental Psychology: Applied , 22 (3), 331 – 49.
Epley Nicholas , Akalis Scott , Waytz Adam , Cacioppo John T.. (2008), " Creating Social Connection Through Inferential Reproduction: Loneliness and Perceived Agency in Gadgets, Gods, and Greyhounds ," Psychological Science , 19 (2), 114 – 20.
Epley Nicholas , Waytz Adam , Cacioppo John T.. (2007), " On Seeing Human: A Three-Factor Theory of Anthropomorphism ," Psychological Review , 114 (4), 864 – 86.
Gelbrich Katja. (2010), " Anger, Frustration, and Helplessness After Service Failure: Coping Strategies and Effective Informational Support ," Journal of the Academy Marketing Science , 38 (5), 567 – 85.
Giebelhausen Michael D. , Robinson Stacey G. , Sirianni Nancy J. , Brady Michael K.. (2014), " Touch Versus Tech: When Technology Functions as a Barrier or a Benefit to Service Encounters ," Journal of Marketing , 78 (4), 113 – 24.
Goldberg Julie H. , Lerner Jennifer S. , Tetlock Philip E.. (1999), " Rage and Reason: The Psychology of the Intuitive Prosecutor ," European Journal of Social Psychology , 29 (5/6), 781 – 95.
Grandey Alicia A. , Dickter David N. , Sin Hock-Peng. (2004), " The Customer Is Not Always Right: Customer Aggression and Emotional Regulation of Service Employees ," Journal of Organizational Behavior , 25 (3) , 397 – 418.
Gray Heather M. , Gray Kurt , Wegner Daniel M.. (2007), " Dimensions of Mind Perception ," Science , 315 (5812), 619.
Greene William H. , Hensher David A.. (2010), Modeling Ordered Choices: A Primer. Cambridge, UK : Cambridge University Press.
Han Min Chung. (2021), " The Impact of Anthropomorphism on Consumers' Purchase Decision in Chatbot Commerce ," Journal of Internet Commerce , 20 (1), 46 – 65.
Hart Phillip M. , Jones Shawn R. , Royne Marla B.. (2013) " The Human Lens: How Anthropomorphic Reasoning Varies by Product Complexity and Enhances Personal Value ," Journal of Marketing Management , 29 (1/2), 105 – 21.
Hayes Andrew F.. (2013), Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach. New York : Guilford Press.
Hayes Andrew F. (2015), " An Index and Test of Linear Moderated Mediation ," Multivariate Behavioral Research , 50 (1), 1 – 22.
Holzwarth Martin , Janiszewski Chris , Neumann Marcus M.. (2006), " The Influence of Avatars on Online Consumer Shopping Behavior ," Journal of Marketing , 70 (4), 19 – 36.
Hoy Matthew B. (2018), " Alexa, Siri, Cortana, and More: An Introduction to Voice Assistants ," Medical Reference Services Quarterly , 37 (1), 81 –8 8.
Huang Ming-Hui , Rust Roland T.. (2018), " Artificial Intelligence in Service ," Journal of Service Research , 21 (2), 155 – 72.
Kaneshige Tom , Hong Daniel. (2018), " Predictions 2019: This is the Year to Invest in Humans, as Backlash Against Chatbots and AI Begins ," Forrester (November 8) , https://go.forrester.com/blogs/predictions-2019-chatbots-and-ai-backlash/.
Keltner Dacher , Ellsworth Phoebe C. , Edwards Kari. (1993), " Beyond Simple Pessimism: Effects of Sadness and Anger on Social Perception ," Journal of Personality and Social Psychology , 64 (5), 740 – 52.
Kim Sara , Chen Rocky Peng , Zhang Ke. (2016), " Anthropomorphized Helpers Undermine Autonomy and Enjoyment in Computer Games ," Journal of Consumer Research , 43 (2), 282 – 302.
Kim Sara , McGill Ann L.. (2011), " Gaming with Mr. Slot or Gaming the Slot Machine? Power, Anthropomorphism, and Risk Perception ," Journal of Consumer Research , 38 (1), 94 – 107.
Kwak Hyokjin , Puzakova Marina , Rocereto Joseph F.. (2015), " Better Not Smile at the Price: The Differential Role of Brand Anthropomorphization on Perceived Price Fairness ," Journal of Marketing , 79 (4), 56 – 76.
Labroo Aparna A. , Dhar Ravi , Schwarz Norbert. (2008), " Of Frog Wines and Frowning Watches: Semantic Priming, Perceptual Fluency, and Brand Evaluation ," Journal of Consumer Research , 34 (6), 819 – 31.
Landwehr Jan R. , McGill Ann L. , Hermann Andreas. (2011), " It's Got the Look: The Effect of Friendly and Aggressive 'Facial' Expressions on Product Liking and Sales ," Journal of Marketing , 75 (3), 132 – 46.
Lench Heather C. , Tibbett Thomas P. , Bench Shane W.. (2016), " Exploring the Toolkit of Emotion: What Do Sadness and Anger Do for Us? " Social and Personality Psychology Compass , 10 (1), 11 – 25.
Lerner Jennifer S. , Goldberg Julie H. , Tetlock Philip E.. (1998), " Sober Second Thought: The Effects of Accountability, Anger, and Authoritarianism on Attributions of Responsibility ," Personality and Social Psychology Bulletin , 24 (6), 563 – 74.
Lerner Jennifer S. , Keltner Dacher. (2000), " Beyond Valence: Toward a Model of Emotion-Specific Influences on Judgment and Choice ," Cognition and Emotion , 14 (4), 473 – 93.
Luff Paul , Frohlich David , Gilbert Nigel G.. (2014), Computers and Conversation. Burlington, MA : Elsevier Science.
Madden Charles S. , Little Elson L. , Dolich Ira J.. (1979), " A Temporal Model of Consumer S/D Concepts as Net Expectations and Performance Evaluations ," in New Dimensions of Consumer Satisfaction and Complaining Behavior , Ralph L. Day and H. Keith Hunt, eds. Bloomington : Indiana University.
Martin René , Watson David , Wan Choi K.. (2000), " A Three-Factor Model of Trait Anger: Dimensions of Affect, Behavior, and Cognition ," Journal of Personality , 68 (5), 869 – 97.
Morgenthal Jan F.. (2017), " Tinka—The Clever Chatbot of T-Mobile Austria ," Deutsche Telekom B2B Europe Blog (October 25) , https://www.b2b-europe.telekom.com/blog/2017/10/25/tinka-the-clever-chatbot-of-t-mobile-austria.
Mori Masahiro. (1970), " Bukimi No Tani (The Uncanny Valley) ," Energy , 7 (4), 33 – 5.
Murdock Susannah. (2016), " Meet Madi ," (November 17), https://www.madison-reed.com/blog/meet-madi.
Oliver Richard L. (1980), " A Cognitive Model of the Antecedents and Consequences of Satisfaction Decisions ," Journal of Marketing Research , 17 (4), 460 –6 9.
Oliver Richard L. , Swan John E.. (1989), " Equity and Disconfirmation Perceptions as Influences on Merchant and Product Satisfaction ," Journal of Consumer Research , 16 (3), 372 – 83.
Pennebaker James W. , Francis Martha E.. (1996), " Cognitive, Emotional, and Language Processes in Disclosure ," Cognition and Emotion , 10 (6), 601 – 26.
Puzakova Marina , Kwak Hyokjin. (2017), " Should Anthropomorphized Brands Engage Customers? The Impact of Social Crowding on Brand Preferences ," Journal of Marketing , 81 (6), 99 – 115.
Puzakova Marina , Kwak Hyokjin , Rocereto Joseph F.. (2013), " When Humanizing Brands Goes Wrong: The Detrimental Effect of Brand Anthropomorphization Amid Product Wrongdoings ," Journal of Marketing , 77 (3), 81 – 100.
Roseman Ira J.. (1984), " Cognitive Determinants of Emotions: A Structural Theory ," in Review of Personality and Social Psychology , Vol. 5. Beverly Hills, CA : SAGE Publications.
Schweitzer Fiona , Belk Russell , Jordan Werner , Ortner Melanie. (2019), " Servant, Friend or Master? The Relationships Users Build with Voice-Controlled Smart Devices ," Journal of Marketing Management , 35 (7/8), 693 – 715.
Shanahan Lilly , Steinhoff Annekatrin , Bechtiger Laura , Murray Aja L. , Nivette Amy , Hepp Urs , et al. (2020), " Emotional Distress in Young Adults During the COVID-19 Pandemic: Evidence of Risk and Resilience from a Longitudinal Cohort Study ," Psychological Medicine , 1 – 10 , doi: 10.1017/S003329172000241X.
Shridhar Kumar. (2017), " How Close Are Chatbots to Passing the Turing Test? " Chatbots Magazine (May 2) , https://chatbotsmagazine.com/how-close-are-chatbots-to-pass-turing-test-33f27b18305e.
Sliter Michael , Jex Steve , Wolford Katherine , McInnerney Joanne. (2010), " How Rude! Emotional Labor as a Mediator Between Customer Incivility and Employee Outcomes ," Journal of Occupational Health Psychology , 15 (4), 468 – 81.
Smith Louise E. , Duffy Bobby , Moxham-Hall Vivienne. (2020), " Anger and Confrontation During the COVID-19 Pandemic: A National Cross-Sectional Survey in the UK ," Journal of the Royal Society of Medicine , 114 (2), 77 – 90.
Sun Li-Yun , Aryee Samuel , Law Kenneth S.. (2007), " High-Performance Human Resource Practices, Citizenship Behavior, and Organizational Performance: A Relational Perspective ," Academy of Management Journal , 50 (3), 558 – 77.
Sundar Aparna , Noseworthy Theodore J.. (2016), " Too Exciting to Fail, Too Sincere to Succeed: The Effects of Brand Personality on Sensory Disconfirmation ," Journal of Consumer Research , 43 (1), 44 – 67.
Touré-Tillery Maferima , McGill Ann. (2015), " Who or What to Believe: Trust and Differential Persuasiveness of Human and Anthropomorphized Messengers ," Journal of Marketing , 79 (4), 94 – 110.
Valenzuela Ana , Hadi Rhonda. (2017), " Implications of Product Anthropomorphism Through Design ," in The Routledge Companion to Consumer Behavior. Abingdon, UK : Routledge.
Waddell Kevah. (2017), " Chatbots Have Entered the Uncanny Valley ," The Atlantic (April 21) , https://www.theatlantic.com/technology/archive/2017/04/uncanny-valley-digital-assistants/523806/.
Wan Jing , Aggarwal Pankaj. (2015), Strong Brands, Strong Relationships. Abingdon, UK : Routledge.
Wan Echo W. , Chen Rocky Peng , Jin Liyin. (2017), " Judging a Book by Its Cover? The Effect of Anthropomorphism on Product Attribute Processing and Consumer Preference ," Journal of Consumer Research , 43 (6), 1008 – 30.
Waytz Adam , Epley Nicholas , Cacioppo John. (2010), " Social Cognition Unbound: Insights into Anthropomorphism and Dehumanization ," Current Directions in Psychological Science , 19 (1), 58 – 62.
Waytz Adam , Gray Kurt , Epley Nicholas , Wegner Daniel M.. (2010), " Causes and Consequences of Mind Perception ," Trends in Cognitive Sciences , 14 (8), 383 –8 8.
Waytz Adam , Heafner Joy , Epley Nicholas. (2014), " The Mind in the Machine: Anthropomorphism Increases Trust in an Autonomous Vehicle ," Journal of Experimental Social Psychology , 52 (May), 113 – 7.
Wiggers Kyle. (2018), " Google Acquires AI Customer Service Startup Onward ," VentureBeat (October 2) , https://venturebeat.com/2018/10/02/google-acquires-onward-an-ai-customer-service-startup/.
Yen Chiahui , Chiang Ming-Chang. (2021), " Trust Me, if You Can: A Study on the Factors That Influence Consumers' Purchase Intention Triggered by Chatbots Based on Brain Image Evidence and Self-Reported Assessments ," Behaviour & Information Technology , 40 (11), 1177–94.
~~~~~~~~
By Cammy Crolic; Felipe Thomaz; Rhonda Hadi and Andrew T. Stephen
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 14- Buy Less, Buy Luxury: Understanding and Overcoming Product Durability Neglect for Sustainable Consumption. By: Sun, Jennifer J.; Bellezza, Silvia; Paharia, Neeru. Journal of Marketing. May2021, Vol. 85 Issue 3, p28-43. 16p. 1 Graph. DOI: 10.1177/0022242921993172.
- Database:
- Business Source Complete
Buy Less, Buy Luxury: Understanding and Overcoming Product Durability Neglect for Sustainable Consumption
The authors propose that purchasing luxury can be a unique means to engage in sustainable consumption because high-end products are particularly durable. Six studies examine the sustainability of high-end products, investigate consumers' decision making when considering high-end versus ordinary goods, and identify effective marketing strategies to emphasize product durability, an important and valued dimension of sustainable consumption. Real-world data on new and secondhand accessories demonstrate that high-end goods can be more sustainable than mid-range products because they have a longer life cycle. Furthermore, consumers engage in more sustainable behaviors with high-end goods, owning them for longer and disposing of them in more environmentally friendly manners. Nevertheless, many consumers prefer to concentrate their budget on multiple ordinary goods in lieu of fewer high-end products partly because of product durability neglect, a failure to consider how long a product will last. Although consumers generally believe that high-end products last longer, they fail to take such a notion into account when making purchases. Finally, this research offers actionable strategies for marketers to help consumers overcome product durability neglect and nudge them toward concentrating their budget on fewer high-end, durable products.
Keywords: product durability neglect; sustainable consumption; sustainable luxury; sustainability
The proof that you did something good is the fact that you can use it again and again.
—Miuccia Prada, head designer of Prada ([57])
Luxury and sustainability are one and the same.
—François-Henri Pinault, chief executive officer of Kering ([61])
The rise of fast-fashion retailers such as H&M and Zara has enabled consumers to increasingly adopt a habit of buying disposable clothing and accessories. More than half of fast-fashion products are worn for less than a year, contributing to a 36% decrease in the average number of times an item is worn compared with 15 years ago ([20]). Although fast fashion offers consumers access to trendy, albeit short-lived, attire at affordable prices, it also exacts high environmental costs, not only in the production phase but also in the postproduction stages of use and disposal. Indeed, the fashion industry has become one of the largest polluters ([31]), contributing to 10% of global carbon emissions as well as 20% of global wastewater ([69]).
Faced with this reality, several trends have emerged over the past decade to counterbalance fast fashion. Notable examples include the rise of sustainable luxury consumption ([ 2]), the concepts of "buy less, buy better" ([13]) and "slow-fashion" ([60]), and the trend of celebrities wearing identical outfits at multiple ceremonies ([10]). Consumers advocating such lifestyles strive to purchase fewer, higher-end products that will last longer, rather than many inexpensive products that will be quickly thrown away. However, these trends and movements still represent niche segments, as products with expensive price tags do not fit the stereotype of sustainable consumption generally associated with restraint and moderation ([ 5]).
Focusing on the clothing and accessories industries, this research explores three aspects of sustainable luxury consumption: ( 1) whether high-end[ 6] products are more sustainable by virtue of their longer product life cycles, ( 2) how consumers process information regarding the durability of these high-end products, and ( 3) how marketers can help consumers overcome a failure to consider product durability and promote the purchase of fewer, higher-end products that will last longer.
Across six studies, including one in which we examine real-world data on new and secondhand shoes and bags, we demonstrate that high-end goods can be more sustainable than ordinary products because of their longer life span and environmentally friendly ways in which they are disposed of. Yet we find that many consumers prefer to allocate the same budget on multiple lower-end products instead of purchasing fewer, higher-end products. We show that these preferences are due to product durability neglect, a failure to consider how long a product will last. In addition to deepening the theoretical understanding of durability as an important dimension of sustainable consumption ([35]; [48]; [71]), the present research also provides actionable strategies for marketers of high-end brands to emphasize the durability of their products and, thus, nudge consumers toward a more sustainable world with fewer, higher-end products that last longer. Given that the clothing and accessories industries are among the top-polluting businesses ([31]), the present work focuses on apparel goods (e.g., shoes, bags, clothes); however, as we elaborate in the "General Discussion" section, the insights from this research can be applied to many other industries as well.
In general, sustainability in consumption refers to "the consumption of goods and services that meet basic needs and quality of life without jeopardizing the needs of future generations" ([52]). Building on prior work in operations and marketing that addresses sustainability from various stages of the product cycle ([15]; [64]), our conceptualization identifies three key dimensions of sustainability: ( 1) sourcing of materials in the supply chain; ( 2) production and manufacturing processes, including labor practices; and ( 3) durability and life span of products, including use and disposal.
We focus on the third dimension of sustainability: product durability and life span. This dimension has mostly been overlooked, with a vast amount of research on sustainability focused on the first two dimensions related to the sourcing of raw materials and the manufacturing processes (for a review, see [71]]). Consistent with extant literature that identifies both the functional and stylistic elements of durability ([14]; [46]), we define a product as durable if it provides extended functional benefits (e.g., it does not deteriorate after a few washes in the case of apparel goods), as well as stylistic benefits (e.g., it does not quickly go out of style, reflecting its timelessness).
Product durability not only contributes to less waste production, but also offers tangible benefits to both consumers and companies. First, given that consumers not only want to be sustainable but also be mindful of personal financial resources ([35]), they can achieve both by selectively purchasing fewer products. By extending the life span of their purchases (i.e., using selectively purchased products for longer duration, and reselling or donating them), consumers can make strategic use of their financial budget, while actively participating in the sustainability movement. The online retailer [21] underscores these benefits when promoting its "Second Life" consignment service, proposing that "by selling your pre-loved bag, you're extending its life and helping the environment."
Second, product durability can benefit companies as well: it is a timely attribute from a managerial standpoint that is highly consistent with the aforementioned trends of sustainable luxury and "slow-fashion" that have gained traction in recent years ([13]). In fact, many high-end entrepreneurial brands, such as Pivotte, Everlane, and Cuyana, as well as more established premium and luxury brands, such as Patagonia, Brunello Cucinelli, and Loro Piana, promote the use of high-quality, durable materials that reduce downstream environmental impact while online luxury retailers like Net-a-Porter allow shoppers to filter the products by their sustainability (for examples, see Web Appendix W1). Given that some consumers purchase more expensive green products to signal status ([34]), promoting the durability of the product can be an appealing strategy for high-end brands to promote not only the luxuriousness of their products, but also the sustainable nature of their goods. Thus, we propose that encouraging the purchase of fewer, high-end durable products can be a win for both consumers and companies.
Luxury products not only embody high prestige and rarity, but also entail longer life spans and durability ([39]; [72]). More specifically, we conceptualize luxury in line with [72], which proposes that luxury goods score high on the following four dimensions: financial dimension (e.g., price, resale price), functional dimension (e.g., durability, quality, reliability), individual dimension (e.g., hedonism, self-identity), and social dimension (e.g., conspicuousness, status signaling). Thus, durability—both its functional and stylistic elements—is central to the definition of luxury ([ 2]; [ 4]). Given that sustainable consumption and luxury overlap on the product durability dimension, we argue that the consumption of fewer, high-end goods can be an effective means to engage in sustainability.
Although both extant literature and industry reports reveal that luxury products and sustainability share some common traits, such as durability, many consumers disregard the sustainable nature of high-end products ([ 5]). In fact, we propose that consumers may outright neglect product durability when contemplating high-end purchases because durability is not a salient attribute when considering these products. Such overlooking is consistent with prior work demonstrating that consumers are prone to making decisions based on easily accessible cues and background context ([22]; [68]) and often fail to consider attributes that are not readily salient ([45]). For instance, when consumers choose between two stereo systems, they may focus on comparing readily available attributes, such as price and the technical specifications (e.g., watt per channel) while neglecting nonsalient, yet important, opportunity costs considerations ([25]). Our product durability neglect hypothesis is also related to prior work showing that consumers disregard the frequency of usage when contemplating purchases of various appliances (e.g., microwaves, monitors, phones) because such information is not readily available ([26]; [29]; [50]).
Although previous work has explored various neglect biases, none has directly considered product durability. We propose that when consumers think of high-end luxury apparel, product durability may not be readily salient because they imagine other, more exemplary instances of luxury consumption (e.g., wearing high-end clothing for status signaling, splurging on a particular item for indulgence). In other words, high-end products are particularly susceptible to product durability neglect because consumers spontaneously focus more on the individual (e.g., hedonism, self-identity) and social (e.g., conspicuousness, status signaling) aspects of luxury goods ([41]; [72]). Accordingly, when choosing between different options, thinking of such prototypical occurrences related to high-end goods may crowd out consumers' ability to consider the relatively longer-lasting nature of these products in the consideration set. This theorizing is also consistent with the accessibility-diagnosticity model ([49]) and the scope insensitivity bias ([12]), suggesting that the accessibility of a given input (e.g., the associations of high-end products with hedonism and status signaling) increases the likelihood that such input will be used to form judgments.
Therefore, we predict that, even when holding the total spending and the time horizon constant, consumers considering different product options will prefer to spend their budget on multiple ordinary items in lieu of fewer, high-end goods because, at least in part, they neglect product durability. More formally, we hypothesize:
- H1: Holding the total budget and time horizon of consumption constant, consumers prefer to purchase multiple mid-range products over fewer high-end products.
- H2: The effect specified in H1 is mediated by product durability neglect.
With growing concerns about environment preservation, many luxury brands are increasingly embracing sustainability. Executives at leading luxury brands and conglomerates, such as LVMH Louis Vuitton Moët Hennessy and Kering, have announced initiatives to make sustainability and the production of sustainable luxury products a top priority ([42]; [59]). We propose that focusing on the durability aspect of sustainability can be an effective marketing strategy for high-end brands to promote their products, while at the same time nudging consumers toward buying fewer, better goods. That is, emphasizing product durability may shape consumers' actual purchase behavior while promoting an attribute central to luxury brands.
Work by behavioral economists and marketing researchers on nudging and choice architecture has found that careful message framing and product positioning can be an effective intervention to prompt behavioral change ([44]; [65]). With specific regards to product choices, making an overlooked attribute more salient or emphasizing explicit cues can help individuals overcome their neglect of various product attributes or decision factors ([25]; [50]). For example, explicitly stating that buying a cheaper stereo system will leave more money available for other purchases helps consumers overcome opportunity cost neglect ([25]). Accordingly, we predict that making product durability salient when choosing among different options will nudge consumers toward selecting fewer high-end products over multiple ordinary ones. More formally, we hypothesize the following:
- H3: Increasing the salience of product durability encourages the choice of fewer high-end products over multiple mid-range products.
With real-world evidence grounded in actual consumption contexts and responses from real product owners, Studies 1 and 2 demonstrate that high-end products can be sustainable because they have longer life spans. In particular, Study 1 provides empirical evidence from the web, with data from over 4,600 new and secondhand shoes and handbags scraped from online stores, and demonstrates that high-end goods are more sustainable than mass-market goods because they are more likely to be sold again as secondhand products. Study 2 finds that consumers engage in more sustainable behaviors with high-end goods (vs. low-end goods), as they desire to keep these items for a longer duration and engage in sustainable behaviors after use (i.e., resell or donate the products) instead of disposing of them. Despite the sustainable nature of high-end goods, Studies 3 and 4 demonstrate that consumers prefer to buy multiple ordinary items over fewer high-end items because, at least in part, they fail to consider the durability of the high-end products. Complementing these findings, the last set of studies also explores the managerial implications of the present research for marketers. Specifically, Study 4 identifies an effective strategy for marketers of high-end products to make durability salient and encourage the sustainable consumption of durable products. Finally, Studies 5a and 5b examine consumers' revealed preferences in two choice-based conjoint surveys, one of which was conducted in collaboration with a clothing company (Pivotte). When consumers have to consider durability and cannot neglect it by design, our results show that they do value durability as an important product attribute relative to other attributes, such as price and style (Study 5a), and that durability can be marketed as a valuable dimension of sustainability (Study 5b).
The objective of the preregistered Study 1 is to provide evidence in favor of the premise that high-end goods can be more sustainable than ordinary goods because they are more durable. To this end, we collect data on more than 4,600 secondhand and new products sold online and examine the presence of luxury products in secondhand markets. In line with our proposition that high-end goods are more durable than ordinary products, we expect to observe a prevalence of high-end brands on websites for secondhand products.
The preregistration detailing the methods and the analysis is available at https://aspredicted.org/blind.php?x=uj7k8h. To acquire relevant data in an objective manner, we identified the most frequently searched online retailers of clothing accessible to U.S. consumers through organic results on Google Search. Next, we constructed a list of the top 20 online retail stores for secondhand products and new products (for a detailed description of the methods, see Web Appendix W2). The top retailers for secondhand products based on the total tallied count were eBay, Grailed, Poshmark, Swap, The RealReal, thredUP, Tradesy, Vestiaire Collective, and Vinted. The top retailers for new products were Anthropologie, Boohoo, Charlotte Russe, Macys, MissGuided, NastyGal, Nordstrom, Target, Walmart, Zappos, and Zaful. Given that some retailers of new clothes only listed a small number of items, we scraped for information from a slightly larger number of retailers selling new clothes (11) than secondhand retailers ( 9) to have a similar number of items collected for each type of apparel (i.e., at least 2,000 products for each category). Moreover, to provide a more conservative test of our hypotheses, we wanted to perform robustness analyses in the absence of products from Target and Walmart (two retailers known for their affordable products) and have the same number of retailers in each list.
After we compiled the list of retailers, automated web crawler scripts scraped information from the 20 websites on both shoes for men and women, and handbags for women. We selected these categories given our focus on apparel and accessories. For each product, we collected the following information (if available): current price, original price, brand name, and detailed product category (e.g., kitten heels). For each website, the crawler collected information on the first 100 available products listed in men's shoes, women's shoes, and women's handbags categories. If a particular retailer listed fewer than 100 products or did not have a specific category of goods (e.g., did not sell handbags), information on all available products was collected. Web Appendix W3 reports summary statistics on the total number of items scraped, organized by product category and type. We collapse the data for shoes and bags for ease of exposition and report the pooled results below; analyzing data by separate product categories does not change the results (all reported in Web Appendix W4).
We collected data for 4,694 secondhand and new shoes and bags from 812 brands. To test our prediction that high-end goods are more prevalent on secondhand retailers than in new product retailers, we asked 1,800 Amazon Mechanical Turk (MTurk) respondents from the United States (60% female; Mage = 37.4 years) to classify the brands of the scraped products as high-end, mid-end, or low-end (or unfamiliar, if they did not know the brand). Each participant rated a random set of 20 brands; we converted the ratings into a numerical brand status score by assigning high-end a value of 3, mid-end a value of 2, and low-end a value of 1. Of the 812 brands, we constructed status scores for 268 brands based on respondents' familiarity with the brands, leading to a total of 2,990 ratings.
To test the prevalence of high-end branded products on secondhand markets, we examined the average status scores of the brands in the new and secondhand product categories. As we predicted, the respondents perceived the average status of the brands listed on secondhand retailers as higher-end than those listed on new product retailers (M2ndhand = 2.47 vs. Mnew = 2.05; t( 2,988) = 28.90, p <.001, d = 1.06). The difference was also significant without Target and Walmart (M2ndhand = 2.47 vs. Mnew = 2.09; t( 2,658) = 24.07, p <.001, d =.94). As an additional test, we confirm that respondents perceived the brands listed on the secondhand websites as higher-end than the midpoint ( 2) of the high/low scale (M2ndhand = 2.47; t( 1,429) = 41.62, p <.001, d = 1.10).
To examine these results at a more granular level and test the robustness of our prediction, we also evaluated the average status scores by percentiles of price (Web Appendix W5). Specifically, we observed that the average status of secondhand branded products was higher than the average status of new products across different percentiles of price. Thus, the significant difference in the average status scores of the secondhand and new products was not simply driven by the large differences in the extreme ends of the data set (i.e., differences in a small number of the most and least expensive items for these secondhand and new products). Consistent with our prediction, the results indicate that secondhand products had higher status than new products across all price points.
The average price for new shoes and bags was $247.28 (SD = $506.71), and for secondhand shoes and bags was $92.64 (SD = $189.91). Because the price distribution was skewed to the right, we logged the price to deal with outliers: the average logged price for new products was 1.68 (SD =.44) and for secondhand products was 2.01 (SD =.59). As expected, the products from secondhand retailers were listed at higher prices than those from new product retailers (M2ndhand = 2.01 vs. Mnew = 1.68; t( 4,692) = 22.02, p <.001, d =.65). The difference was also significant without Target and Walmart (M2ndhand = 2.01 vs. Mnew = 1.76; t( 4,092) = 15.60, p <.001, d =.49; for additional robustness checks, see Web Appendix W6).
Ancillary analyses cast doubt on several alternative explanations. One might wonder whether these results could be driven by secondhand products being unique or having better aesthetics, leading to a higher average brand status and price relative to the new products. To rule out these possibilities, we scraped the photos of ten products from each category from each of the 20 websites, for a total of 500 product images. Then, we recruited 1,000 U.S. respondents (74% female; Mage = 34.5 years) on MTurk to rate these product images on uniqueness and liking. Specifically, each respondent looked at two randomly chosen product images and answered the following questions for each product on a seven-point Likert scale: ( 1) "How unique does the product look to you?" (1 = "Not unique at all" to 7 = "Very unique [one-of-a-kind]") and ( 2) "How much do you like the design of the product?" (1 = "Do not like at all" to 7 = "Like it very much"). The new and secondhand products were rated similarly in terms of uniqueness (Mnew = 4.75 vs. M2ndhand = 4.75; t(498) =.00, n.s.). The respondents liked the new products more than the secondhand products (Mnew = 4.37 vs. M2ndhand = 4.06; t(498) = 2.35, p =.019, d =.21), which was opposite of what the results would have been had the alternative account been at play. Importantly, controlling for these factors by conducting an analysis of variance with average brand status scores as the dependent variable, product type as the main factor, and uniqueness and liking ratings as two covariates revealed that product type (new vs. secondhand) was the only significant factor (F( 1, 319) = 95.78, p <.001, η2 =.23), whereas the two covariates had no significant effect (uniqueness: F( 1, 319) =.02, n.s.; liking: F( 1, 319) = 3.58, n.s.).[ 7]
By directly scraping field data from 20 retailers selling secondhand products, our preregistered Study 1 provides correlational support for the notion that high-end products have a longer life cycle because they are more prevalent on online secondhand retailers than ordinary goods. One may wonder whether the presence of high-end goods on secondhand markets is a mere by-product of a higher starting original price. That is, perhaps more high-end products are listed on secondhand retailers just because they are more expensive. While this is a possibility, if high-end apparels were merely expensive but not long-lasting, our thesis that these high-end products are more sustainable by virtue of their durability would not be supported. On the contrary, the evidence stemming from this data set suggests that, in addition to possibly being more costly, high-end goods also last for a long time and make it to additional life cycles in the market.
To find further support for the notion that high-end goods can be more sustainable than lower-end items because high-end products are used for more extended periods and are discarded in more environmentally friendly manners, we directly asked owners of high- and low-end accessories to provide information about some of their belongings. We predict that the more high-end an owned item is, the longer the intended duration of ownership, and the lower the intention to throw it away instead of engaging in sustainable disposal behaviors, such as reselling, donating, or giving away the product to someone else. In line with Study 1, we expect that high-end items will be more durable and discarded more sustainably than ordinary goods.
We recruited 340 wealthy women from the United States on Qualtrics (Mage = 30.4 years; Mincome ≥ $121,000[ 8]) for an online study. We purposely recruited female respondents with high annual income to control for gender and financial background and to increase the likelihood that they would own products from diverse price ranges. We randomly assigned respondents to one of two between-subjects conditions (high-end vs. low-end) and asked them to provide information about both a pair of shoes and a bag that they owned (order counterbalanced). In the case of shoes, for example, respondents were told: "Please think about a high-end[ 9] [low-end] pair of shoes that you own." If they did not own any products that fit the description, respondents in the high-end condition thought of the most expensive products they owned, whereas those in the low-end condition thought of the least expensive products they owned: "If you do not have any pair of high-end [low-end] shoes, please think about the most [least] expensive pair of shoes you own." We used identical phrases to collect information about the respondents' bags.
Then, respondents answered a series of questions about their owned products, including ( 1) purchase price ("How much did you pay for the pair of shoes/bag?"), ( 2) length of planned ownership ("How long do you plan on wearing your shoes/using your bag before you no longer want them [it]?" on a seven-point Likert scale: 1 = "0–6 months," 2 = "6 months–1 year," 3 = "1 year–1 year and 6 months," 4 = "1 year and 6 months–2 years," 5 = "2 years–2 years and 6 months," 6 = "2 years and 6 months–3 years," and 7 = "> 3 years–specify"), and ( 3) disposal ("What will you do with the pair of shoes/bag when you no longer want them/it?" with the options "sell it," "give it to someone else," "throw it away," "donate it," "keep it even though I will not wear it," and "other–specify"). We recoded the disposal responses as a binary dependent variable depending on whether the answer was a sustainable behavior (1 if the respondent indicated selling it, giving it to someone else, donating it, or keeping it) or an unsustainable behavior (0 if the respondent indicated throwing it away). No value was assigned for "other—specify" (1% of responses). We also collected a series of ancillary variables on these products (e.g., physical product condition, who bought them). Controlling for all these variables in the analyses does not change the results.
The average price of shoes across the two conditions (high-end and low-end) was $183.67 (SD = $535.54). We found a significant difference between high-end and low-end conditions in purchase price of the owned shoes (Mhigh = $242.90 vs. Mlow = $127.17; F( 1, 338) = 4.00, p =.046, η2 =.01). The average price of bags was $264.18 (SD = $624.62). Similar to shoes, we found a significant difference between the two conditions in purchase price (Mhigh = $385.19 vs. Mlow = $148.74; F( 1, 338) = 12.59, p <.001, η2 =.04).
The significant differences between the purchase prices of the high-end and low-end products confirm that respondents indeed thought of a high-end or a low-end pair of shoes and a bag depending on the condition to which they were randomly assigned (high-end vs. low-end). Note that the average prices for the low-end products were not trivial (e.g., $148.74 for "low-end" bags). This was likely a by-product of recruiting high-income respondents and provides a more stringent test of the durability of high-end products.
For ease of exposition, we collapse the data for shoes and bags. However, all results are also significant when analyzing the two product categories separately. Consistent with our prediction, we found that the expected duration of ownership was significantly longer for the high-end products than the low-end products (Mhigh = 5.05 vs. Mlow = 4.13; F( 1, 678) = 39.74, p <.001, η2 =.06).
As predicted, there was a significant difference in the overall responses by condition (χ2( 1) = 17.77, p <.001, ϕ =.16). Specifically, owners of the high-end products displayed a greater willingness to engage in sustainable disposal behaviors (%high = 91.10) compared with the owners of the low-end products (%low = 79.54); the owners of the low-end products were more likely to throw away the products than the owners of the high-end products (%low = 20.46 vs. %high = 8.90).
Study 2 provides further empirical support that high-end goods are more sustainable than low-end products because consumers who own high-end goods intend to own them for longer and dispose of them in more sustainable ways. One potential weakness of Study 2 could be that the owners of high-end products were motivated to justify their purchases and, thus, stated that they would use these products for longer. To address this possible issue of postpurchase justification, in the next studies, we ( 1) directly explore consumers' preferences between high-end and lower-end apparel before making a purchase and ( 2) test the premise that high-end goods last longer regardless of ownership status. The next two studies also directly test our proposed product durability neglect account.
In Study 3, we investigate whether consumers prefer multiple mid-range products over a high-end product (H1) because they neglect product durability (H2). The study aims to provide evidence on the process in two ways. First, building on established methods to detect neglect biases in research (e.g., [29]; [50]; [63]), we test whether product durability neglect underlies consumers' preferences toward relatively less sustainable product choices by examining respondents' thoughts as they decide between different options. Second, we assess whether consumers' differing intertemporal preferences make certain consumers more susceptible to product durability neglect than others. Given that the benefits of sustainable consumption are often realized over a long time horizon, those who are more patient and have a more future-oriented mindset tend to engage in more sustainable consumption behaviors compared with myopic consumers, who have a stronger present bias ([ 3]; [38]). In the case of product durability, consumers who have a more future-oriented mindset should recognize that durable products yield benefits in the future because these products have longer life spans. Thus, if product durability neglect is indeed at play, we expect consumers with relatively lower intertemporal discount rates ([24]) to favor fewer high-end products (vs. multiple mid-range products) compared with consumers with higher intertemporal discount rates.
We recruited 201 U.S. respondents for a paid online survey on MTurk (44% female; Mage = 34.7 years). To increase the generalizability of our findings and confirm that our results are not driven by the specifics of the product category, we tested two products, different price points, and different time horizons. To this end, all respondents were randomly assigned to one of two between-subject replicates (product type: shoes vs. winter coat) and asked to make a purchase decision about shoes or winter coat. For shoes, respondents read, "Imagine that you typically have a shoes budget of $400[10] per year. You have two options regarding how you want to spend the $400. Which would you prefer?" Then, respondents selected either "buy one high-end pair of shoes for $400" or "buy four mid-end pairs of shoes for $100 each" (the order of appearance of the two options was randomized). Similarly, for winter coats, respondents read, "Imagine that you have a winter coat budget of $2,000[11] for the next ten years. You have two options regarding how you want to spend the $2,000. Which would you prefer?" Next, respondents chose either "buy one high-end winter coat for $2,000 this year" or "buy one mid-end winter coat for $200 every year" as their response (order of appearance randomized).
Then, all respondents listed at least one and up to five thoughts about the decision that they just made about the shoes or the winter coats ("In the form below, please list at least one reason why you decided to choose that option"; open-ended). To assess the presence of durability-related content, we developed a corpus of words that contained the following durability-related roots: "last" and "dura" (allowing to detect relevant terms such as "long-lasting," "last," "durability," and "durable"). Then, we counted the number of times these key terms appeared in the comments using the function grepl() in R. For instance, if a particular respondent mentioned the word "durable" in a given comment, this was tallied once.
Finally, all respondents completed the Dynamic Experiments for Estimating Preferences ([66]), which involved 12 rounds of adaptive questions related to one's time preferences (i.e., a choice between a smaller, immediate gain and a larger, later gain). The data were analyzed using a hierarchical Bayesian approach to estimate individual-level parameters in the quasihyperbolic time discounting model, including the estimates of beta, delta, and the discount rate r ([51]; [66]).
Regarding shoes, 78.85% of respondents preferred to buy multiple mid-range products, whereas only 21.15% of respondents preferred to buy one high-end product. Similarly, regarding winter coats, 76.29% indicated that they would prefer multiple mid-range products, whereas only 23.71% indicated that they would like one high-end product. As in previous studies, we collapse the two products—and report the results in aggregate (separate analyses of each category led to similarly significant results). Across the two products, 77.61% of respondents preferred to buy multiple mid-range products, whereas only 22.39% of respondents indicated that they would like to buy one high-end product. Thus, the majority of respondents preferred to consume multiple mid-range products (χ2( 1) = 61.30, p <.001, h = 1.17).
Respondents generated a total of 647 comments, with an average of 3.22 thoughts per person. A two-sample t-test revealed no significant difference in the average number of thoughts generated between those who chose the high-end option and those who chose the mid-range option (Mhigh = 3.09 vs. Mmid = 3.26; t(199) =.65, n.s.). Only 6.96% of all comments containing durability-related content, regardless of their product choice. However, a two-proportions z-test revealed that a significantly higher proportion of respondents who chose the high-end option mentioned durability in their thoughts (%high = 14.39) compared with respondents who chose the mid-range option (%mid = 4.92, χ2( 1) = 13.69, p <.001, h =.33). In support of our predictions, these results suggest that those who chose to allocate their budget on multiple mid-range products neglected product durability to a greater extent. In contrast, durability considerations were relatively more accessible for those who opted to concentrate their budget on one high-end option.
To test our account through intertemporal preferences, we ran a logistic regression with choice as the dependent variable (coded as 1 for choice of one high-end product and as 0 for choice of multiple mid-range products), discount rate r as the predictor, and product type (shoes vs. winter coat) as a covariate. The discount rate r was a negative and significant predictor of choice (β = −100.01, χ2( 1) = 4.96, p =.026). As expected, respondents with a lower discount rate were more likely to choose the high-end option instead of the ordinary options. The product type did not predict choice (β = −.11, χ2( 1) =.11, n.s.). Given that the higher discounting rate r indicates a greater present bias, and less patience, these results are consistent with product durability neglect and demonstrate that having a present-bias may impede consumers in recognizing the value of durability.
To increase statistical conclusion validity ([49]), we replicated the main findings in another study involving 248 respondents (33% female; Mage = 19.5 years; see Web Appendix W7) recruited at the behavioral lab of a U.S. university.
Although the lack of durability-related content in respondents' open comments suggests that consumers neglect product durability, it is possible that instead of neglecting product durability, consumers simply do not believe that high-end products are more durable and, thus, are reluctant to choose them. To address this possibility, we recruited 200 respondents in the lab at a U.S. university (57% female; Mage = 19.5 years) and asked them to rate, between-subjects, the durability of a high-end or a mid-range pair of shoes. If the alternative account—that consumers are doubtful that high-end products can be more durable—were supported, we would find no significant differences in the life span estimates of the high- and mid-range products. Our results go against such an account: respondents indicated that the high-end item would last for a significantly longer time than the mid-range item, in support of the lay belief that high-end products are more durable (Mhigh = 4.84 vs. Mmid = 3.05; t(198) = 7.48, p <.001, d = 1.06; see Web Appendix W8).
Study 3 demonstrates that when presented with two options, most respondents preferred to spend the same amount of money on multiple ordinary goods instead of on one high-end good (H1) because, at least in part, they did not consider the durability of the high-end product (H2). Consistent with our account, product durability neglect was stronger for respondents who chose multiple mid-range products (vs. one high-end product). Moreover, those who had a higher discount rate r tended to prefer multiple mid-range products.
Although these results support our product durability neglect hypothesis, there remain other potential alternative accounts. For instance, it is possible that the respondents opting for multiple goods, in addition to neglecting durability, were also driven by variety-seeking motives or risk aversion. It is also conceivable that the respondents opting for high-end goods may have mentioned durability for self-presentation motives ([23]) or as a justification for choosing a more indulgent product ([43]). Because these motives may be concurrently at play, the next study shows more unequivocally that product durability neglect underlies part of the observed effects by experimentally manipulating the salience of durability in a marketing-relevant context.
The purpose of Study 4 is twofold. First, consistent with previous research on neglect biases ([25]), we manipulate the salience of durability to further establish product durability neglect as the process underlying the preference for multiple mid-range products (vs. fewer high-end products). In doing so, we also control for potential alternative explanations such as variety seeking. Second, we explore the effectiveness of a marketing-relevant intervention to nudge consumers toward more durable products using realistic stimuli embedded in online product pages.
The preregistration detailing the methods and the analysis is available at https://aspredicted.org/blind.php?x=yy6z3y. We recruited 421 U.S. respondents (51% female; Mage = 32.2 years) on Prolific Academic for a paid online survey. We randomly assigned respondents to one of two conditions between-subjects (control vs. durability). Respondents considered two product pages—one for a high-end item priced at $80 and another for a mid-range item priced at $20[12]—featuring a black sweater sold by two fictitious brands, "Luyana" and "Cooper."
We opted for fictitious brand names to control for preexisting brand associations with well-established brands ([ 8]). To rule out potentially confounding effects of different models, styles, and brand names used in the stimuli, we created two versions—A and B—of the ad for all the conditions described next. In one version, a particular model, style, and brand name, "Cooper," was used in the high-end condition. In another version, another model, style, and brand name, "Luyana," was used in the high-end condition. This design serves as a between-subjects replicate, and we expect to observe the predicted results for both versions of the stimuli. In addition, to account for variety seeking, we embedded the focal product in a product page featuring three different colors (i.e., black, pink, and camel) to prime the notion that one could opt for multiple items of various colors. We also priced the items so that one could opt for several ordinary products with the same budget of one high-end item. Finally, we matched respondents' gender to the gender of the model featured to increase relevance. For ease of exposition, we report stimuli and results consistent with version A, in which Luyana was the mid-range retailer and Cooper was the high-end retailer.
All respondents read the following information about the two retailers: "Luyana is a retailer that offers mid-range clothing. Luyana typically sells sweaters priced around $10–$20. Cooper is a retailer that offers high-end clothing. Cooper typically sells sweaters priced around $70–$80." Then, they saw two product pages, each with an ad copy promoting the products. In the control condition, the high-end option read, "A high-end sweater with long sleeves, and ribbing at neckline and hem." The mid-range option read, "A mid-range sweater with long sleeves, and ribbing at neckline and hem." In the durability condition, the high-end option read, "A high-end, durable sweater. You can think of this sweater as a one-time purchase in one product that will last for many years" (see Web Appendix W15 for a complete set of the stimuli).[13] The mid-range option read the same as in the control condition. Then, to check whether our manipulation increased the salience of durability and to ensure that respondents were actually paying attention, we asked, "In the box below, please type about 2–3 keywords from the webpage above."
On the next page, all respondents read, "Imagine that this year, you have a clothing budget of $80 to spend on sweaters. You have two options regarding how you want to spend the $80." Then, respondents saw the following two options, buying "one high-end sweater for $80 at Cooper" or buying "four mid-range sweaters for $20 each at Luyana," and were asked, "Which would you prefer?" As in Study 3, all respondents listed at least one and up to five thoughts about the choice that they just made and we counted the number of times durability-related terms appeared in the comments.
Confirming the success of the durability salience manipulation, an analysis of the keywords that the respondents wrote down as they were looking at the two images (i.e., an ad for Cooper and an ad for Luyana) revealed that those in the durability condition mentioned durability-related words (%durability = 42.28) more than those in the control condition (%control = 0, χ2( 1) = 223.19, p <.001, h = 1.42).
We ran a logistic regression with choice as the dependent variable (coded as 1 for choice of one high-end product and as 0 for choice of multiple mid-range products) and with condition (control vs. durability) and version (A vs. B) as the independent variables. As predicted, respondents chose the high-end option significantly more in the durability condition than in the control condition (%durability = 27.36 vs. %control = 15.79, β =.70, χ2( 1) = 8.14, p =.004). Importantly, we observed the predicted effect of the durability manipulation even when variety seeking is potentially at play (given the three colors and the possibility of buying up to four items with the same budget). Although not central to our hypothesis, there also was a significant effect of version such that respondents were more likely to choose the high-end option for the brand and style of Cooper (%A = 25.59 vs. %B = 17.62, β = −.48, χ2( 1) = 3.91, p =.048).[14]
Respondents generated a total of 1,209 thoughts, with an average of 2.87 thoughts generated per person. A two-sample t-test revealed no significant difference between the average number of thoughts generated by those who chose the high-end option and those who chose the mid-range options (Mhigh = 2.97 vs. Mmid = 2.85; t(419) =.84, n.s.). Replicating results from Study 3, the vast majority of respondents, regardless of their product choice, did not mention any durability-related content in their thoughts, with only 7.28% of all comments containing such content. At the same time, a two-proportions z-test revealed that the magnitude of neglect was higher for those opting for multiple mid-range products (3.41% of all comments related to durability) over those choosing the high-end product (20.74%, χ2( 1) = 90.80, p <.001, h =.57).
We performed a mediation analysis (PROCESS Model 4, [36]) with choice as the dependent variable, condition (control vs. durability) as the independent variable, and the number of durability-related thoughts generated as the mediator. As predicted, the extent to which a consumer chose the high-end option was mediated by the number of durability-related thoughts generated (indirect effect =.64; 95% confidence interval [CI95%] = [.41,.94]).
By manipulating the salience of product durability, preregistered Study 4 provides additional support for the underlying process of product durability neglect and offers an effective strategy in online communication to promote high-end products. The findings suggest that making product durability more salient by mentioning the word "durable" is an effective and actionable intervention to encourage the sustainable consumption of fewer, better goods.
Studies 3 and 4 demonstrate that consumers tend to neglect product durability unless this attribute is made salient. However, even when durability is brought to consumers' attention, some important questions remain for marketers: Do consumers neglect durability because it is not on their radar at the time of purchase or because it is actually irrelevant to their product choice? Study 4 provides some evidence in favor of the former, but how much do consumers value durability relative to other important product attributes, such as price or design? And with specific regard to sustainability, can durability be legitimately framed as an aspect of sustainability?
Conjoint analysis is particularly suitable for answering these questions. By including durability as one of the product attributes (Study 5a) or as one of the levels (Study 5b) in the design of the study, respondents cannot neglect durability and are forced to make trade-offs revealing their true preferences with regard to this particular product dimension. In other words, we explore how much consumers value durability relative to other product features when they are forced to consider it.
In addition, in these studies, we further investigate managerially relevant ways to emphasize durability. In Study 5a, we frame durability as a standalone product attribute, independent from sustainability, enabling us to understand how consumers value different levels of durability when they are made concrete (e.g., a product that lasts five years vs. ten years). Further, we are able to understand the value of durability, compared with other attributes such as price, style, and the dimensions of sustainability (i.e., sourcing and manufacturing). In Study 5b (in collaboration with Pivotte, a U.S.-based clothing company), we explicitly frame durability as a dimension of sustainability, enabling us to determine whether durability can effectively be positioned as an aspect of sustainability. Taken together, Study 5a sheds light on how durability framings can appeal to a broader segment of consumers, independent of sustainability messaging and Study 5b demonstrates how durability can be positioned as a dimension of sustainability and used to target a specific segment of green consumers.
We recruited 162 (41% female; Mage = 27.8 years) graduate students at a U.S. university who completed the survey for course credit. To evaluate consumers' revealed preferences regarding durability with explicit trade-offs relative to other important product attributes (e.g., price, style), we employed a choice-based conjoint (CBC) survey using Sawtooth Software. We chose Moncler coats as the stimuli for this study given that Moncler was a popular, desirable high-end brand among the sample population (35% of respondents reported that they owned at least one Moncler product or expressed a desire to buy one in the future; 61% had heard of the brand before).
We created a CBC survey with five attributes—price, style, color, durability, and sustainability—with three levels within each attribute. The durability attribute had the following three levels: low-level ("The textile used to make the coat will last about 5 years"), mid-level ("The textile used to make the coat will last about 10 years"), and high-level ("The textile used to make the coat will last about 15 years"). Importantly, with this configuration of attributes, we made the durability information explicitly concrete to emphasize the total life span (i.e., 5 years, 10 years, and 15 years). In addition, the sustainability attribute entailed the following three levels: the sourcing of materials ("Made with down feather meeting strict Down Integrity System and Traceability [D.I.S.T.] requirements for animal welfare"), the production process ("Manufactured at Fair Trade Certified™ facilities with fair wage and labor practices"), and use and disposal ("Certified to meet bluesign® criteria for advanced waste-reduction technologies to minimize carbon footprint after disposal"; for a full description of all the other attributes and levels, see Web Appendix W9).
Each respondent completed 12 choices in random order and chose the most preferred option out of three Moncler coats based on their price, style, color, durability, and sustainability. To generate the choice sets, we used a full profile, complete enumeration design, producing the most orthogonal design for each respondent with respect to the main effects. After the choice task, we collected measures regarding awareness ("Have you ever heard of the brand, Moncler, before?"; yes/no) and ownership ("Do you currently own any Moncler coat(s) or have you ever considered purchasing one?" with options "No, I don't own and I don't plan on owning any Moncler coats," "I currently don't own a Moncler coat, but I'm thinking of purchasing one," and "Yes, I do own Moncler coat(s). Please indicate how many."). Controlling for these factors does not impact the significance of the following results.
We used Sawtooth's HB-Reg Module, which estimates a hierarchical random coefficients model, to calculate part-worth utilities of different attributes, a widely used approach in marketing research ([11]). We followed the approach outlined by [54] to computed the degree of confidence with which an attribute level is preferred to another attribute level (for calculations, see Web Appendix W10).
Focusing on durability, we found significant differences among the part-worth utilities of each level from low-level (Mutility = −1.74), to mid-level (Mutility =.55) to high-level (Mutility = 1.19) durability. The mid- and high-levels of durability were preferred to the low-level with 100% confidence. The high level of durability was preferred to the mid-level with 99.84% confidence. Thus, respondents significantly preferred higher levels of durability compared with lower levels. For ease of interpretation, we also present the increase in part-worth utility from one level of durability to another in monetary ($) terms.[15] An increase from the low-level (5 years) to the mid-level (10 years) of durability equates to an increase of $296.35 in the value of a product. Similarly, an increase from the mid-level (10 years) to the high-level (15 years) equates to an increase of $76.97 in the value of a product (for calculations, see Web Appendix W11).
Looking at product profiles holistically, the relative importance weights indicated that style was the most important attribute (43.94%; CI95% = [40.65, 47.22]). As Figure 1 shows, price (21.59%; CI95% = [19.31, 23.87]) and durability (18.87%; CI95% = [16.96, 20.79]) were the second-most important attributes and did not significantly differ from each other. Finally, color (10.09%; CI95% = [8.27, 11.92]) and sustainability (5.51%; CI95% = [4.88, 6.13]) were the least important attributes. Overall, these results indicate that, when respondents were obliged to consider it, the durability of the textile was as important as price. Thus, durability emerged as a key factor in respondents' purchase decisions that was second only to style. In contrast, the sustainability of the product was not a particularly important attribute, and significantly less important than durability as a standalone attribute.
Graph: Figure 1. Study 5a: relative importance of attributes (%).Notes: The error bars denote 95% CIs.
We recruited 106 (89% female; Mage = 37.3 years) real consumers of Pivotte from the company's email listserv for a paid online survey. To evaluate their preferences, we employed a CBC survey with four attributes (i.e., price, style, color, and sustainability) with three levels within each attribute. Note that in this study, durability is not an attribute by itself but is framed as one of the levels within the sustainability attribute. Consistent with our conceptualization of the three dimensions of sustainability, as well as the company's existing strategy, the sustainability attribute, labeled as "textile" in the survey, consisted of three levels: the eco-friendly sourcing of materials ("Made with eco-friendly fabric with advanced waste-reduction technologies"), manufacturing process with fair labor practices ("Made in N.Y.C. by top manufacturers with impeccable labor practices"), and the durability of the clothing ("Made with durable, 4-way stretch, stain-resistant fabric that will last for years"; for a screenshot of what respondents saw, including all attributes and levels, see Web Appendix W13).
Similar to Study 5a, we used Sawtooth's HB-Reg Module to estimate the models. Confirming the relevance of durability, we found that the part-worth utility of the durability message was highest (Mutility =.23), followed by sourcing of materials (Mutility =.10) and manufacturing process (Mutility = −.33).[16] The respondents preferred the durability level of sustainability to the manufacturing level, with 99.30% confidence, and to the sourcing level, with 72.82% confidence. Thus, there was a significant difference between the part-worth utilities of durability and manufacturing levels, but not between durability and sourcing levels.
We also examined the relative importance weights across all attributes; the weights indicated that style was the most important attribute (44.63%; CI95% = [40.37, 48.88]), followed by sustainability (20.43%; CI95% = [16.92, 23.94]), color (17.98%; CI95% = [15.58, 20.39]), and price (16.96%; CI95% = [14.68, 19.23]). These results indicate that, in the case of Pivotte pants, style was significantly more important than the other three attributes. Notably, information about the sustainability of the product was as important as the product's price and color, suggesting that when durability was framed as a level of sustainability, sustainability emerged as an important and valued attribute for consumers.
In Study 5b, we purposely labeled the sustainability attribute as "textile" to diminish potential demand effects. To increase the label's face validity, we also replicated Study 5b explicitly naming the attribute as "sustainability" on Prolific Academic (n = 150; 100% female; Mage = 36.4 years; Mincome ≥ $100,000). These results enable us to confirm that durability is an important dimension of sustainability independent of the specific label (see Web Appendix W12).
Study 5a shows that when consumers have to trade off between durability and other product attributes, durability emerges as an important attribute that is second only to style and just as valued as price. Study 5b demonstrates that durability can be effectively positioned as a dimension of sustainability. In particular, when durability was compared with the other two dimensions of sustainability (i.e., sourcing and manufacturing), it was strictly preferred to fair manufacturing processes and comparable to eco-friendly sourcing of raw materials.
In conclusion, Studies 5a and 5b offer additional managerial insights regarding durability and how to promote it. Findings from Study 5a suggest that, whenever possible, marketers of high-end brands should provide concrete estimates of products' life spans (e.g., three vs. five years) and promote the durable nature of their goods. The results of Study 5b highlight that marketers can position durability as an appealing sustainability dimension that consumers genuinely value.
The present research finds that purchasing luxury can be a unique means to engage in sustainable consumption because high-end products are more durable. Yet consumers prefer to concentrate their budget on multiple ordinary goods over fewer high-end products. We demonstrate that this effect is, in part, driven by consumers' product durability neglect. Although consumers generally believe that more expensive products last longer, they fail to take such a notion into account when making purchases. Focusing on the domains of clothing and accessories, our studies explore durability as a central dimension of sustainability. Given that 10% of global carbon emissions arise from the fashion industry, nudging consumers toward fewer purchases of long-lasting, high-end apparel could lead to a reduction of emissions, thereby reducing a key factor driving global warming ([69]).
Our findings show that high-end products can be more sustainable than mid-range products by virtue of their longer life cycle (Studies 1 and 2), and as Studies 4, 5a, and 5b indicate, durability can be strategically used to make high-end products more appealing. As such, the present research offers actionable strategies for marketers of high-end brands and products.
One potential challenge for marketers of high-end brands is to understand how to best educate their potential consumers in discerning the intrinsic high quality and durability of their goods. When we entered the term "product durability," into the search engine AlsoAsked,[17] we found that two related queries included "Why is durability important in a product?" and "How do you check durability?" (see Web Appendix W14), suggesting that there is a demand to learn more about evaluating product durability. Marketers can take advantage of this opportunity to educate consumers through tutorials and advertisements or by making durability claims more concrete, as we did in Studies 4 and 5a. In fact, some luxury and premium brands have dedicated pages on their websites that specifically address this notion. For instance, Loro Piana underscores the exceptional durability of its Pecora Nera wool (https://ii.loropiana.com/en/our-world/pecora-nera) while Cuyana promises to deliver products that will "last for years to come" (https://www.cuyana.com/sustainability.html). Presumably, consumers who understand and can identify the characteristics that make products more durable should be more prone to choosing fewer high-end goods.
In addition, government agencies and policy makers can take an active role in educating consumers about product durability. Public campaigns might encourage consumers to think of product durability and recognize long-lasting materials when making purchases. For example, the French national anticounterfeiting committee CNAC, in collaboration with many high-end brands (e.g., Van Cleef & Arpels, Chanel), has conducted a campaign to educate consumers about the downsides of purchasing counterfeit luxury products, such as the inferior quality of these goods leading to shorter-term use ([18]). Luxury brands and government agencies can collaborate to educate consumers about purchasing fewer, better goods that benefit the consumers and the environment.
Product durability may be a vital element in the emerging sharing economy for luxury products. Companies such as Rent the Runway, DressYouCan, and Verstolo are revolutionizing how millennials consume high-end clothing and accessories. Rental models allow for maximum use of physical products, giving multiple consumers access to the same products over a prolonged period, while mitigating potential concerns such as dissatisfaction or satiation with the purchase. Durability becomes even more important in these contexts as the products must be able to sustain multiple uses.
Marketers and brands also have an active role in determining how quickly goods are consumed, as the speed with which brands launch new products influences how quickly the existing goods become old-fashioned and discarded ([ 6]). Indeed, many new lines and collections are designed to have quick turnovers as certain trends and aesthetics are meant to evolve from season to season ([17]; [40]). Some fast-fashion brands, such as Zara and H&M, launch new items at two-week cycles. Recently, however, some high-end brands have started to challenge this notion and advocate for slower fashion cycles. Louis Vuitton, Off-White, Gucci, and Dries Van Noten are actively trying to slow down their fashion cycles by creating collections with "less unnecessary products" and a focus on fewer, longer-lasting pieces that "can remove the idea that just because it's last season, it's devalued" ([27]; [37]). High-end brands slowing down the pace of the new collections may send a positive signal to consumers that they should buy less frequently and value the long-lastingness of the products.
Pertinent to our focal product category of luxury are the questionable and unethical practices often associated with the sourcing and production processes. For instance, certain luxury brands are known to use materials that may impede on consumers' desire to protect animal rights (e.g., inhumane sourcing of animal skin) or are produced by exploiting workers during the production process and devastate the local community (e.g., blood diamonds, products created by sweatshop laborers; [55]; [56]). Recognizing these darker sides of luxury, we acknowledge that product durability alone may not lead to comprehensively sustainable business practices.
Durability is ultimately a consumer-centric attribute that directly affects consumers' pocketbooks, as consumers decide for how long to keep their belongings and whether to resell them when no longer wanted. Further, owning and reselling durable products can positively influence consumers' happiness and feelings of empowerment ([19]; [67]). Although there may be a risk of dissatisfaction shortly after a purchase or satiation over time, these issues can be uniquely addressed through return policies, resale markets (as Study 1 demonstrates), product warranty and guarantees, or innovative business models such as rental subscriptions.
By establishing product durability as a critical dimension of both sustainability and luxury, we hope that this article is the first step toward a deeper understanding of durability in marketing research. Future research could address several theoretical aspects related to product durability.
As previously discussed, we conceptualize durability in terms of both functional and stylistic benefits ([46]). Indeed, some high-end brands prominently advertise the long-lastingness and sturdiness of their products, as seen in the "Buy Less, Demand More" campaign by Patagonia (see Web Appendix W1). At the same time, others focus more on promoting the stylistic durability of their offerings, such as Farfetch's "forever wardrobe" advertisement, which maintains that Farfetch's collection of products will not go out of trend and can be timeless, long-lasting staples (see Web Appendix W1). From a theoretical standpoint, is there a hierarchy between the functional and stylistic elements of durability, or do they contribute equally to the construct of product durability? Is one of the two benefits a sufficient condition for product durability, or are both necessary for an item to be perceived as truly durable?
As previously mentioned, some work suggests that consumers exhibit usage frequency neglect when choosing between different appliances, such as microwaves, ice cream makers, and monitors ([50]). In this case, the overlooked decision factor is the frequency of use (i.e., how often a consumer uses the product). When should we expect to see product durability neglect versus frequency neglect? Given that durability is directly related to both how physically sturdy a product is as well as how timeless its style is, one hypothesis is that product durability neglect may apply to categories in which both functional and stylistic benefits are particularly relevant, such as apparel consumption (our focus) and possibly more hedonic products in general. In contrast, it is plausible that frequency neglect may be more relevant in utilitarian product categories, such as kitchenware.
The present research has focused on the domains of clothing and accessories. Although we predict that our findings and insights will likely generalize to different industries and product categories, it may be a worthwhile pursuit to document consumers' choices and product durability neglect in other domains. For example, it is plausible that for product categories that are often bought in installments (e.g., dishwashers, refrigerators) or for which data on depreciation and maintenance is readily accessible (e.g., cars, phones), consumers may be more apt to open mental accounts and compute the costs per usage of these transactions ([32]; [62]) than for products that are typically paid in full at the time of purchase. Consistent with our results, if consumers can readily anticipate long-time use of a potential purchase, they may be less prone to product durability neglect and thus opt for the high-end option. Another potentially interesting industry to analyze is furniture. For instance, would IKEA be the equivalent of the fast-fashion brand, H&M? In line with the present research, it is plausible that product durability neglect also drives preferences for frequent purchases of inexpensive furniture in lieu of long-term investments in high-end furniture that will last many years.
Our research can be further applied to explore additional aspects of sustainability and luxury brands.
The present research examines high-end, luxury goods and lower-end, ordinary goods in the context of apparel consumption. To broaden the scope of our inquiry, we have not distinguished between high-end, premium brands (e.g., Patagonia, Woolrich) and top luxury brands (e.g., Hermès, Louis Vuitton). However, these brands vary significantly on the luxury spectrum ([39]). Thus, future research might adopt a more nuanced approach and explore the meaning of durability at a more granular level for different types of high-end brands. For instance, when the top luxury watchmaker Patek Philippe promotes product durability, it may have to make a strong claim to justify the purchase (i.e., an intergenerational claim implying that the watch will last across three generations, spanning over a century; see Web Appendix W1). However, it is possible that other watchmakers that are positioned as premium brands may be able to make effective product durability appeals with shorter life span claims.
Despite certain benefits associated with luxury consumption, such as attribution of status, preferential treatment, and affiliation with desirable social groups and mates ([ 7]; [33]; [70]), recent work documents many social costs associated with the consumption of high-end, expensive products. For example, consumers who own luxury goods are considered less warm and authentic and more driven by impression-management motives than consumers who do not own them ([ 9]; [23]; [28]; [30]). These negative perceptions may also be driven by a failure to consider the durability of high-end products at the observers' end. In fact, our preliminary data (available upon request), which explore how others judge luxury shoppers, demonstrate that high-end consumers who spend the same amount of money as consumers opting for more ordinary goods across the same time horizon are perceived as more wasteful and materialistic, even though they ironically purchase fewer products. Given this finding, future work could further explore the negative nuances associated with perceptions of high-end buyers and uncover how such perceptions may be ameliorated.
If some avoid purchasing high-end products because of the aforementioned wasteful and materialistic perceptions associated with such goods, would highlighting product durability possibly help consumers justify these purchases to themselves and others? If so, they may be able to use product durability as a functional alibi for purchasing high-end items and increase their willingness to buy these goods ([43]).
How consumers define and conceptualize the term "waste" is also a topic that may further enhance our understanding of sustainability. While some may define "waste" purely in financial terms of wasting money (i.e., buying one expensive sweater when cheaper ones are available), others define waste in physical terms of wasting material objects (i.e., buying many inexpensive sweaters). From a financial perspective, it may seem more wasteful to spend more on a single item. However, from a sustainability perspective, it may seem more wasteful to purchase an abundance of cheaper clothing that will deteriorate quickly and be thrown away. One hypothesis that warrants further investigation is whether having different conceptions of waste (i.e., overspending financially vs. overconsuming physically) lead to different consumption behaviors. For instance, some consumers may not consider spending money on high-end purchases negatively but, instead, penalize a "quantity over quality" mentality. Indeed, in a follow-up study (available upon request), we found that those who were more averse to wasting physical objects (vs. wasting money) judged high-end consumers (who own fewer items) less negatively than consumers of multiple mid-range items.
We propose that luxury goods possess a unique, sustainable trait as they can have a longer life span than lower-end products. Despite the long-lasting nature of high-end goods, sustainable luxury can be a paradoxical concept for consumers, as many of them neglect the durability inherent in luxury products. With growing concerns about sustainable consumption, many luxury brands are increasingly becoming more committed in their efforts to embrace sustainability. Focusing on and promoting product durability could be an effective strategy to align a sustainability dimension with a high-end positioning while encouraging consumers to engage in a more sustainable consumption lifestyle for a better world.
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921993172 - Buy Less, Buy Luxury: Understanding and Overcoming Product Durability Neglect for Sustainable Consumption
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921993172 for Buy Less, Buy Luxury: Understanding and Overcoming Product Durability Neglect for Sustainable Consumption by Jennifer J. Sun, Silvia Bellezza and Neeru Paharia in Journal of Marketing
Footnotes 1 Kristin Diehl
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Jennifer J. Sun https://orcid.org/0000-0003-1165-2728
5 Online supplement: https://doi.org/10.1177/0022242921993172
6 The article uses the terms "luxury" and "high-end" interchangeably ([58], p. 58) and examines both top luxury brands and high-quality premium brands.
7 An identical analysis of variance with log price as the dependent variable also revealed that product type was the only significant factor (F(1, 496) = 41.73, p <.001, η2 =.08), whereas the two covariates were not significant (uniqueness: F(1, 496) =.33, n.s.; liking: F(1, 496) = 1.88, n.s.).
8 We chose the $121,000 cutoff because previous research on status ([7]; e.g., [1]) has identified this level as the highest income bracket.
9 Across all studies, we always use the term "high-end" instead of "luxury" in the stimuli read by respondents to avoid potential negative stereotypes and associations linked to the term "luxury."
We calibrated the price of a high-end pair of shoes using the average prices of shoes from Tod's, Church's, and Stuart Weitzman; for mid-range shoes, we used average prices from Zara, J.Crew, H&M, and Banana Republic.
We calibrated the price of a high-end winter coat using the average prices of coats from Moncler, Fay, and Loro Piana; for mid-range winter coats, we used average prices from Zara, J.Crew, H&M, and Banana Republic.
We calibrated the price of a high-end sweater using the average prices from Everlane, Naadam, and Cuyana; for the mid-range sweater, we used average prices from Zara, Madewell, and H&M.
For the high-end product stimuli in the durability condition, we purposely removed the words, "with long sleeves, and ribbing at neckline and hem," so that the two products' stimuli had a comparable number of words in the text.
As a further check, we also ran the same regression including the interaction term between condition and version and confirm the significance of condition as a predictor of choice (β =.67, χ2(1) = 4.29, p =.038).
Note that these dollar-equivalent estimates across different levels of durability are for ease of interpretation only; we did not use a market simulation approach, and these values should not be interpreted as estimated market value of the willingness to pay ([53]; for detailed calculations, see Web Appendix W11).
Note that a negative value reflects that the manufacturing process is valued less relative to the two other dimensions, not that respondents value it negatively.
AlsoAsked is a website that uses data from "People Also Asked" section of Google Search results and generates a tree diagram of related queries.
References Adler Nancy E., Epel Elissa S., Castellazzo Grace, Ickovics Jeannette R. (2000), "Relationship of Subjective and Objective Social Status with Psychological and Physiological Functioning: Preliminary Data in Healthy White Women," Health Psychology, 19 (6), 586–92.
Amatulli Cesare, De Angelis Matteo, Costabile Michele, Guido Gianluigi. (2017), Sustainable Luxury Brands: Evidence from Research and Implications for Managers. London: Palgrave Macmillan.
Arnocky Steven, Milfont Taciano L., Nicol Jeffrey R. (2014), "Time Perspective and Sustainable Behaviour: Evidence for the Distinction Between Consideration of Immediate and Future Consequences," Environment and Behaviour, 46 (5), 556–82.
Athwal Navdeep, Wells Victoria K., Carrigan Marylyn, Henninger Claudia E. (2019), "Sustainable Luxury Marketing: A Synthesis and Research Agenda," International Journal of Management Reviews, 21 (4), 405–26.
Beckham Daisy, Voyer Benjamin G. (2014), "Can Sustainability Be Luxurious? A Mixed-Method Investigation of Implicit and Explicit Attitudes Towards Sustainable Luxury Consumption," in Advances in Consumer Research, Vol. 42, Cotte June, Wood Stacy, eds. Duluth, MN: Association for Consumer Research, 245–50.
Bellezza Silvia, Ackerman Joshua M., Gino Francesca. (2017), "'Be Careless with That!' Availability of Product Upgrades Increases Cavalier Behavior toward Possessions," Journal of Marketing Research, 54 (5), 768–84.
Bellezza Silvia, Berger Jonah. (2020), "Trickle-Round Signals: When Low Status Is Mixed with High," Journal of Consumer Research, 47 (1), 100–127.
Bousch David M., Loken Barbara. (1991), "A Process-Tracing Study of Brand Extension Evaluation," Journal of Marketing Research, 28 (1), 16–28.
Cannon Christopher, Rucker Derek D. (2019), "The Dark Side of Luxury: Social Costs of Luxury Consumption," Personality and Social Psychology Bulletin, 45 (5), 767–79.
Cantor Carla. (2020), "A Greener Red Carpet," Columbia News (February 7), https://blogs.ei.columbia.edu/2020/02/07/red-carpet-sustainable-fashion/.
Chakravarti Amitav, Grenville Andrew, Morwitz Vicki G., Tang Jane, Ulkumen Gulden. (2013), "Malleable Conjoint Partworths; How the Breadth of Response Scales Alters Price Sensitivity," Journal of Consumer Psychology, 23 (4), 515–25.
Chang Hannah H., Pham Michel Tuan. (2018), "Affective Boundaries of Scope Insensitivity," Journal of Consumer Research, 45 (2), 403–28.
Cline Elizabeth. (2016), "The Power of Buying Less by Buying Better," The Atlantic (February 16), https://www.theatlantic.com/business/archive/2016/02/buying-less-by-buying-better/462639/.
Cooper Tim. (2010), Longer Lasting Products: Alternatives to the Throwaway Society. Surrey, UK: Gower Publishing Limited.
Cronin J. JosephJr, Smith Jeffery S., Gleim Mark R., Ramirez Edward, Martinez Jennifer Dawn. (2011), "Green Marketing Strategies: An Examination of Stakeholders and the Opportunities They Present," Journal of the Academy of Marketing Science, 39 (1), 158–74.
Cuyana (2020), "Sustainability: Our Fewer, Better Promise," (February 10), https://www.cuyana.com/sustainability.html.
Desmichel Perrine, Ordabayeva Nailya, Kocher Bruno. (2020). "What If Diamonds Did Not Last Forever? Signaling Status Achievement Through Ephemeral Versus Iconic Luxury Goods," Organizational Behavior and Human Decision Processes, 158 (2020), 49–65.
Diderich Joelle. (2012), "French Unveils New Anticounterfeit Ads," WWD (May 30), https://wwd.com/business-news/marketing-promotion/france-unveils-new-anti-counterfeit-ads-5932220/.
Donnelly Grant E., Lamberton Cait, Reczek Rebecca Walker, Norton Michael I. (2017), "Social Recycling Transforms Unwanted Goods into Happiness," Journal of the Association for Consumer Research, 2 (1), 48–63.
Ellen MacArthur Foundation (2017), "A New Textiles Economy: Redesigning Fashion's Future," (November 28), https://www.ellenmacarthurfoundation.org/publications/a-new-textiles-economy-redesigning-fashions-future.
Farfetch (2021), "Farfetch Second Life," (January 7), https://www.farfetch.com/positively-farfetch/secondlife.
Feldman Jack M., Lynch John G. (1988), "Self-Generated Validity and Other Effects of Measurement on Belief, Attitude, Intention, and Behavior," Journal of Applied Psychology, 73 (3), 421–35.
Ferraro Rosellina, Kirmani Amna, Matherly Ted. (2013), "Look at Me! Look at Me! Conspicuous Brand Usage, Self-Brand Connection, and Dilution," Journal of Marketing Research, 50 (4), 477–88.
Frederick Shane, Loewenstein George, O'Donoghue Ted. (2002), "Time Discounting and Time Preference: A Critical Review," Journal of Economic Literature, 40 (2), 351–401.
Frederick Shane, Novemsky Nathan, Wang Jing, Dhar Ravi, Nowlis Stephen. (2009), "Opportunity Cost Neglect," Journal of Consumer Research, 36 (4), 553–61.
Friedman Elizabeth M.S., Dhar Ravi. (2021), "Neglect of Ownership Duration in Consumer Choice," working paper, Columbia Business School, Columbia University.
Friedman Vanessa. (2020), "On the Eve of New York Fashion Week, What's Next?" The New York Times (September 14), https://www.nytimes.com/2020/09/14/style/Fashion-Week-2020.html.
Garcia Stephen M., Weaver Kimberlee, Chen Patricia. (2018), "The Status Signals Paradox," Social Psychological and Personality Science, 10 (5), 690–96.
Goodman Joseph K., Irmak Caglar. (2013), "Having Versus Consuming: Failure to Estimate Usage Frequency Makes Consumers Prefer Multifeature Products," Journal of Marketing Research, 50 (1), 44–54.
Goor Dafna, Ordabayeva Nailya, Keinan Anat, Crener Sandrine. (2020), "The Imposter Syndrome from Luxury Consumption," Journal of Consumer Research, 46 (6), 1031–51.
Gordon Jennifer Farley, Hill Colleen. (2015), Sustainable Fashion: Past, Present and Future. New York: Bloomsbury Academic.
Gourville John T., Soman Dilip. (1998), "Payment Depreciation: The Behavioral Effects of Temporally Separating Payments from Consumption," Journal of Consumer Research, 25 (2), 160–74.
Griskevicius Vladas, Tybur Joshua M., Sundie Jill M., Cialdini Robert B., Miller Geoffrey F., Kenrick Douglas. (2007), "Blatant Benevolence and Conspicuous Consumption: When Romantic Motives Elicit Strategic Costly Signals," Journal of Personality and Social Psychology, 93 (1), 85–102.
Griskevicius Vladas, Tybur Joshua M., Van den Bergh Bram. (2010), "Going Green to Be Seen: Status, Reputation, and Conspicuous Conservation," Journal of Personality and Social Psychology, 98 (3), 392–404.
Haws Kelly L., Winterich Karen Page, Naylor Rebecca Walker. (2014), "Seeing the World Through Green-Tinted Glasses: How Green Consumers' Use Motivated Reasoning to Prefer Environmentally Friendly Products," Journal of Consumer Psychology, 24 (3), 336–54.
Hayes Andrew F. (2013), Methodology in the Social Sciences. Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach. New York: Guilford Press.
Indvik Lauren. (2020), "The Fashion Industry Wants to Slow Down. Can It?" Financial Times (May 30), https://www.ft.com/content/8bd9fe5e-a02e-11ea-b65d-489c67b0d85d.
Joireman Jeffrey A., van Lange Paul A.M., van Vugt Mark. (2004), "Who Cares About the Environmental Impact of Cars? Those with an Eye Toward the Future," Environment and Behavior, 36 (2), 187–206.
Kapferer Jean-Noel. (2010), "All That Glitters Is Not Green: The Challenge of Sustainable Luxury," European Business Review (November/December), 40–45.
Kapferer Jean-Noel, Bastien Vincent. (2012), The Luxury Strategy: Break the Rules of Marketing to Build Luxury Brands. London: Kogan Page.
Kapferer Jean-Noel, Klippert Cindy, Leproux Lara. (2014), "Does Luxury Have a Minimum Price? An Exploratory Study into Consumers' Psychology of Luxury Prices," Journal of Revenue and Pricing Management, 13 (1), 2–11.
Keinan Anat, Crener Sandrine, Goor Dafna. (2020), "Luxury and Environmental Responsibility," in Research Handbook on Luxury Branding, Morhart Felicitas, Wilcox Keith, Czellar Sandor, eds. Cheltenham, UK: Edward Elgar Publishing, 300–323.
Keinan Anat, Kivetz Ran, Netzer Oded. (2016), "The Functional Alibi," Journal of the Association for Consumer Research, 1 (4), 479–96.
Klotz Leidy, Weber Elke, Johnson Eric, Shealy Tripp, Hernandez Morela, Gordon Bethany. (2018), "Beyond Rationality in Engineering Design for Sustainability," Nature Sustainability, 1 (5), 225–33.
Legrenzi P. Vaolo, Girotto Vittorio, Johnson-Laird Philip N. (1993), "Focusing in Reasoning and Decision Making," Cognition, 49 (1/2), 37–66.
Levinthal Daniel A., Purohit Devavrat. (1989), "Durable Goods and Product Obsolescence," Marketing Science, 8 (1), 35–56.
Loro Piana. (2020), "Pecora Nera," (February 10), https://ii.loropiana.com/en/our-world/pecora-nera.
Luchs Michael G., Naylor Rebecca Walker, Irwin Julie R., Raghunathan Rajagopal. (2010), "The Sustainability Liability: Potential Negative Effects of Ethicality on Product Preference," Journal of Marketing, 74 (5), 18–31.
Lynch John G., Bradlow Eric, Huber Joel, Lehmann Donald. (2015), "Reflections on the Replication Corner: In Praise of Conceptual Replications," International Journal of Research in Marketing, 32 (4), 333–42.
Mittelman Mauricio, Gonçalves Dilney, Andrade Eduardo B. (2019), "Out of Sight, Out of Mind: Usage Frequency Considerations in Purchase Decisions," Journal of Consumer Psychology, 30 (4), 652–59.
O'Donoghue Ted, Rabin Matthew. (2001), "Risky Behavior Among Youths: Some Issues from Behavioral Economics," in Risky Behavior Among Youths: An Economic Analysis, Gruber Jonathan, ed. Chicago: University of Chicago Press, 29–68.
Organisation for Economic Co-operation and Development (2002), Towards Sustainable Household Consumption? Trends and Policies in OECD Countries. Paris: OECD Publishing.
Orme Bryan K. (2001), "Assessing the Monetary Value of Attribute Levels with Conjoint Analysis: Warnings and Suggestions," Sawtooth Software Research Paper Series.
Orme Bryan K., Chrzan Keith. (2017), Becoming an Expert in Conjoint Analysis: Choice Modeling for Pros. Provo, UT: Sawtooth Software.
Paharia Neeru. (2020), "Who Receives Credit or Blame? The Effects of Made-to-Order Production on Responses to Unethical and Ethical Company Production Practices," Journal of Marketing, 84 (1), 88–104.
Paharia Neeru, Vohs Kathleen, Deshpandé Rohit. (2013), "Sweatshop Labor Is Wrong Unless the Shoes Are Cute: Cognition Can Help and Hurt Moral Motivated Reasoning," Organizational Behavior and Human Decision Processes, 121 (1), 81–8.
Palmer Alexandra. (2005), "Vintage Whores and Vintage Virgins: Second Hand Fashion in the Twenty-first Century," in Old Clothes, New Looks: Second Hand Fashion, Palmer Alexandra, ed. Oxford, UK: Berg Publishers, 197–214.
Pandelaere Mario, Shrum L.J. (2020), "Fulfilling Identity Motives Through Luxury Consumption," in Research Handbook on Luxury Branding, Morhart Felicitas, Wilcox Keith, Czellar Sandor, eds. Cheltenham, UK: Edward Elgar Publishing, 57–74.
Paton Elizabeth. (2017), "François-Henri Pinault, Kering Chief, on Why Green Is the New Black," The New York Times (January 25), https://www.nytimes.com/2017/01/25/fashion/franois-henri-pinault-kering-sustainability.html.
Pierre-Louis Kendra. (2019), "How to Buy Clothes That Are Built to Last," The New York Times (September 25), https://www.nytimes.com/interactive/2019/climate/sustainable-clothing.html.
Pinault Francois-Henri. (2019), "Sustainability: Our Approach," Kering (accessed February 3, 2021), https://www.kering.com/en/sustainability/our-approach/.
Prelec Drazen, Loewenstein George. (1998), "The Red and the Black: Mental Accounting of Savings and Debt," Marketing Science, 17 (1), 4–28.
Sela Aner, LeBoeuf Robyn A. (2017), "Comparison Neglect in Upgrade Decisions," Journal of Marketing Research, 54 (4), 556–71.
Seuring Stefan, Muller Martin. (2008), "From a Literature Review to a Conceptual Framework for Sustainable Supply Chain Management," Journal of Cleaner Production, 16 (15), 1699–710.
Thaler Richard, Sunstein Cass. (2008), Nudge: Improving Decisions About Health, Wealth, and Happiness. New York: Penguin Putnam Inc.
Toubia Olivier, Johnson Eric, Evgeniou Theodoros, Delquié Philippe, (2013), "Dynamic Experiments for Estimating Preferences: An Adaptive Method of Eliciting Time and Risk Preferences," Management Science, 59 (3), 613–40.
Turunen Linda Lisa Maria, Cervellon Marie-Cecile, Carey Lindsey Drylie. (2019), "Selling Secondhand Luxury: Empowerment and Enactment of Social Roles," Journal of Business Research, 116 (August), 474–81.
Tversky Amos, Kahneman Daniel. (1974), "Judgment Under Uncertainty: Heuristics and Biases," Science, 185 (4157), 1124–31.
United Nations Economic Commission for Europe (2018), "UN Alliance Aims to Put Fashion on Path to Sustainability," press release (July 13), https://unece.org/forestry/press/un-alliance-aims-put-fashion-path-sustainability.
Veblen Thorstein. (1899/2007), The Theory of the Leisure Class. New York: Oxford University Press.
White Katherine, Habib Rishad, Hardisty David J. (2019), "How to SHIFT Consumer Behaviors to be More Sustainable: A Literature Review and Guiding Framework," Journal of Marketing, 83 (3), 22–49.
Wiedmann Klaus-Peter, Hennigs Nadine, Siebels Astrid. (2007), "Measuring Consumers' Luxury Value Perception: A Cross-Cultural Framework," Academy of Marketing Science Review, 7 (7), 1–21.
~~~~~~~~
By Jennifer J. Sun; Silvia Bellezza and Neeru Paharia
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 15- Caffeine’s Effects on Consumer Spending. By: Biswas, Dipayan; Hartmann, Patrick; Eisend, Martin; Szocs, Courtney; Jochims, Bruna; Apaolaza, Vanessa; Hermann, Erik; López, Cristina M.; Borges, Adilson. Journal of Marketing. Sep2022, p1. DOI: 10.1177/00222429221109247.
Ahead of Print- Database:
- Business Source Complete
Record: 16- Can Encroachment Benefit Hotel Franchisees? By: Kim, TI Tongil; Jap, Sandy D. Journal of Marketing. Mar2022, Vol. 86 Issue 2, p147-165. 19p. 1 Diagram, 8 Charts, 2 Graphs. DOI: 10.1177/00222429211008136.
- Database:
- Business Source Complete
Can Encroachment Benefit Hotel Franchisees?
Franchise encroachment, or the addition of an outlet in the vicinity of existing franchisees, is largely viewed as resulting in revenue cannibalization of incumbent locations. Against this backdrop, the authors consider the possibility that the addition of same brand outlets can, in fact, also create positive effects via customer utility and ultimately benefit franchisees, due to a range of mechanisms such as quality signaling, learning, or brand awareness, resulting in a positive pathway on franchisee performance. The authors unpack this possibility using an experiment and detailed proprietary and publicly available data sets from the hotel industry over a five-year period. Their results show evidence of positive effects on customer utility for same-brand outlets and stronger effects for newer brands, cross brands, and online travel agency channel bookings. Counterfactual simulations indicate that although encroachment hurts franchisees on average, it can modestly benefit same-brand franchisees in low-brand-density markets. Together, the findings illustrate the potential "sunny side" of encroachment, underscoring the need to update our view of encroachment as context-dependent. The novel emphasis on customers versus the dominant firm view suggests customer and incumbent responses to encroachment should be accounted for in the development of franchise strategy and public policy decisions.
Keywords: brand management; demand estimation; franchise encroachment; franchise sales; franchise systems; structural model
Franchise encroachment ("encroachment" hereinafter) is a key phenomenon in franchise management. Encroachment occurs when a franchisor places a new outlet in proximity to an existing franchisee(s). Incumbent franchisees view these actions negatively, as the new outlet increases competition and cannibalizes their revenues. This view is reflected in substantial literature in marketing, which primarily focuses on documenting this negative impact ([21]; [33]) and managing the resulting conflict ([23]). The literature also examines ways to safeguard franchisee interests through litigation ([ 4]), contracting ([22]), and a range of governance mechanisms ([ 3]). In broad strokes, this literature has painted a picture of encroachment as being predominantly negative for incumbent franchisees and has focused on firm roles.
We explore the possibility that encroachment might in fact be positive for incumbent franchisees by incorporating both a customer view and brand differences into this picture. Specifically, we introduce the possibility that customers may gain utility from more outlets of the same brand, subsequently increasing demand and, ultimately, franchisee performance. This utility might come about via a range of mechanisms such as quality signaling, learning, or brand awareness. While researchers have acknowledged conceptually that encroachment may improve franchisee performance ([ 8]; [21]), little empirical evidence exists to date. The closest finding is limited: [13] show that multiple exposures of a trademark hotel brand name on a third-party intermediary's website generated an overall net increase in consumer clicks. However, no studies thus far have provided evidence of encroachment activities directly improving performance via customer utility and demand.
While research has examined encroachment effects on customer demand in a static setting with endogenous franchisee pricing ([35]; [39]), these studies do not make the distinction between same versus different brand effects on customer utility, and they focus on a single franchise brand. In contrast, we consider a range of brand characteristics within a franchise chain to inform a broader view of encroachment, performance outcomes, and a more explicit incorporation of the customer's viewpoint. For example, an entry of a Residence Inn hotel will have same brand effects on other Residence Inn hotels and different brand effects on Westin hotels, Element hotels, and so on in the market. We examine these differing effects of brand encroachment on consumer choices.
Our context is a single franchisor, incumbent franchisee(s) of multiple brands, and a range of encroachment types across various markets. We model the net impact using publicly available data, a proprietary data set from one of the largest hotel groups in the United States over a five-year period, and an experiment. We find that encroachment effects are not only statistically significant but also economically relevant. Although the effects are predominantly negative, we nevertheless find positive effects on customer utility that are stronger for same brands, newer brands, cross brand locations (i.e., brands with overlapping monikers such as Hyatt Place and Hyatt House), and bookings made through online travel agencies such as Expedia and Kayak.
Counterfactual simulations reveal both positive and negative economic outcomes. Overall, encroachment reduces incumbent revenues and profits by 5.3% ($43,810 in revenue and $26,090 in profit per month), consistent with the historical bent of the literature (i.e., increased competition). However, this impact is not statistically significant for same-brand franchisees, suggesting the negative impact is reduced. Further analysis reveals significant positive effects in low-brand-density markets: same-brand franchisees improve performance by approximately 2% per month (i.e., $4,300 in revenue and $2,900 in profit), with 70% (30%) of these outlets realizing higher (lower) revenues as a result.
To the best of our knowledge, ours is the first study to directly model encroachment effects on customer preference and identify the conditions under which it might benefit franchisees. We find that while encroachment hurts franchisees on average, the prevailing assumption that it always results in revenue losses for all incumbent franchisees is inaccurate; we find a distinct difference in its impact on same versus different brand outlets.
The stakes for better understanding franchising phenomena are substantial. Franchising is one of the most pervasive organizational forms in the U.S. economy, accounting for as much as $890 billion, or 50% of all retail sales across 75 industries. This amounts to approximately 3% of the 2016 U.S. gross domestic product in nominal dollars.[ 6] Our research context—the hotel industry—accounted for approximately $660 billion in 2019.[ 7]
In the following sections, we provide a motivating example, overview the relevant literature, and develop a customer utility model that estimates the net effect of encroachment, franchisee competition, endogenous pricing, and heterogeneous customer preferences. We next describe data, and the estimation section then applies a logit and a full demand model. We explore heterogeneous brand characteristics to gain further insights and present counterfactual simulations to consider market composition differences. A general discussion and implications for management conclude.
In this section, we review the related literature on franchise encroachment. Table 1 summarizes the range of terms that occur throughout the article and their definitions.
Graph
Table 1. Terms and Definitions.
| Term | Definition | Present study |
|---|
| Franchisor | The organization that owns the franchise brand(s) and oversees and coordinates the franchise system | Examines data from one hotel franchisor and its franchise system. |
| Chain | The totality of retail outlets in the franchise system | Examines a chain that comprises multiple distinct hotel brands, most of which share no common brand-related terms, although two brands have a key brand-related term in common (e.g., brands "AB" and "ABC"). |
| Franchisee | Entity that owns and operates outlet(s) in the franchise chain | Examines franchisees that have one hotel with one of the chain's brands. |
| Encroachment | The addition of a new outlet of the same franchise chain in proximity to an existing franchisee. | Examines three distinct types of encroachment: Same brand (the new hotel franchisee has the same brand as an incumbent) Different brand (the new hotel has a different brand within the hotel chain) Cross brand (the new hotel franchisee shares a component of the incumbent's brand name; e.g., Hyatt and Park Hyatt)
|
Encroachment arises because franchise business models are marked by a fundamental incentive misalignment: franchisors earn a royalty rate on franchisee sales and an upfront fee ([27]), but franchisees are profit maximizing in their pursuits ([23]). The fixed fee averages 8% of all payments from franchisees ([28]), leaving 90% of the franchisor's income from royalties. Given that franchisors cannot extract all rents via a fixed fee (see [24]), they have an incentive to add more outlets as long as revenues are positive, but this creates a downward pressure on franchisee profits. This behavior is widely viewed as anticompetitive ([ 8]), and the incentive misalignment has led to an active regulatory context and lobbying efforts aimed to minimize the resulting conflict.
Several states have implemented measures to safeguard franchisees' interests through disclosure requirements and relationship laws, and research has shown that such laws do in fact lower litigation incidences ([ 4]). [22] show how the franchisor's use of ex ante contracts and extra-contractual incentives influence subsequent monitoring and enforcement, while [ 3] examine the relationship between governance mechanisms and franchisees' motivation and bankruptcy.
In a related vein, other research considers the factors that drive a franchisor's decision to add a new location to a market. The academic literature mainly focuses on the trade-offs between increased franchisee competition and firm-side benefits such as preemption and firm size spillovers related to entry of hamburger chains ([10]; [17]) or new entrant sales at the expense of incumbents ([20], [21]; [30]). In contrast, practitioners predominantly talk about "demand generators," or businesses that bring in out-of-town visitors, which seem to be less studied. Our analyses corroborate this and show the number of businesses as a strong predictor of hotel entry (Web Appendix W1). However, we find that local population characteristics are less relevant in predicting hotel entry after controlling for the number of businesses.
Encroachment can have positive and negative effects on incumbents. Herein, we consider potential theoretical mechanisms that drive positive and negative pathways and their performance impact, although we are not able to empirically tease them apart due to data or modeling limitations. Figure 1 displays an overview of the pathways.
Graph: Figure 1. Positive and negative pathways of encroachment on franchisees' financial performance.
A long-standing prediction of industrial organization is that an incumbent's prices and revenues will decrease if new firms enter the market through business stealing ([ 2]; [31]; [36]; [40]). This prediction is easily extended to a franchising context: the entry of additional franchisees will lead to greater revenue losses for incumbent franchisees.
Ample evidence supports the resulting negative pathway from competition. For example, [21] finds that encroachment leads to significant decreases in hotel franchisee revenues per room ($51–$66.81 per quarter). [18] show that a hotel in Manhattan was less likely to survive as the number of chain units in the area increased. [33] shows the presence of competing convenience stores in the same market, regardless of their chain affiliation, reduces store-level revenues significantly. In fast food franchise settings, evidence for negative pathways approximates on average an 18 percentage point decrease in revenues ([45]). [35] show that incumbents' revenues are decreased by 1%–2%, but distance can help mitigate this negative pathway; for every one mile increase between stores, revenue loss is stemmed by as much as 28.1%. Collectively, these results show that encroachment generally hurts incumbent performance via negative pathways by increasing competition.
An alternative point of view is that a cluster of same-brand locations can be useful for customers, while simultaneously benefiting incumbents ([ 8]; [21]). This possibility has been understudied. We next consider three possible drivers of positive pathways that can arise from customers.
More same-brand outlets can increase awareness of the brand or simply make options more noticeable. Franchisee locations can act as a "living billboard to build awareness and positive brand association" ([ 5], p. 97). Although most hotel customers are not local, this effect might be operative in online choice settings in which a consumer sees multiple same-brand locations available in a specific area (cf. [13]). One extension of brand awareness impacting positive pathways is via cross-brand effects. In our context, an example of cross-brands is hotel locations with overlapping monikers, such as Hyatt Place and Hyatt House. Given the prominence of the Hyatt chain name, its use in multiple locations might boost demand for these outlets. This may increase brand awareness via multiple exposures and market prevalence.
Multiple locations can also act as a quality signal, much as advertising communicates quality ([ 1]; [25]). When multiple products share a common brand name, this acts as a nontrivial "performance bond," a credible indicator assuring consumers that products are of high quality, thereby lifting sales of all products under certain conditions ([43]). Costly brand and infrastructure investments communicate a pledge to a prespecified quality level for both existing and new hotels.
This logic might be moderated by brand type and booking channel. For example, a quality signal might be more diagnostic for newer than for older, more established brands because consumers may perceive greater purchase uncertainty associated with new brands, and the quality signal acts as an assurance in the consumer's choice process. Similarly, the signal might be more pronounced in channels in which customers do not have strong brand loyalty or product experience. As an example, when a consumer sees multiple Marriott locations in a purchase channel that overviews multiple brands, the cluster of outlets may be a more diagnostic quality signal (i.e., the brand's commitment to a market area) than for a customer using Marriott's booking channel. A prominent avenue for quality signaling is online travel agency (OTA) channels: platforms such as Expedia and Kayak rely heavily on sponsor advertising, and these efforts are another means by which firms can signal brand quality to customers.
Consumer learning across similar brands occurs ( 1) at a point in time, such as when a new brand enters a market, and ( 2) over time via consumption and exposure to firm communications ([19]). Thus, the first scenario can be purely cognitive: consumers form a quality perception, and existing brand quality inferences spill over to the new, similar brand entrant. In this context, online reviews from other guests may enable a customer to pool their experiences to infer hotel quality, potentially improving utility for specific brands. In contrast, the second scenario may involve a gradual process over time owing to consumers' experiences. Learning can be particularly valuable with experience goods such as a hotel stay.
The net effect of positive and negative pathways on incumbent franchisee performance is not obvious. The new outlet may increase customer preference for franchisees and benefit their performance but, at the same time, heighten competition, which can put pressure on prices and decrease revenues. Thus, a model is needed to quantify the net effect on incumbent performance. While we attempt to rule out as many alternative explanations as possible, completely distinguishing among the positive pathway mechanisms—quality signaling, consumer learning, and brand awareness—remains a fertile ground for future research.
We employ a structural model to examine customer response to encroachment in the presence of franchisees' endogenous prices as well as subsequent outcomes, such as whether their sales increase or are cannibalized. [ 7] provide a useful starting point in their model of an individual customer choosing a product with the greatest indirect utility out of a discrete set of differentiated goods that compete on price; this is a natural framework of static competition. We build on their model and account for a positive pathway effect on customer utility, customer heterogeneity, and price competition as a modification to the standard model.
Past research models the addition of a new franchise outlet via reduced form analyses using revenue data and estimates costs related to the firm's decisions (e.g., [10]; [16]; [17]; [38]). In contrast, we micro-model customers' heterogeneous demand response to encroachment while abstracting away from modeling the firm's dynamic entry. Further, we control for the firms' entry decision using fixed effects and uncertainty in the exact timing of hotel entries. A more comprehensive approach would be to micro-model both heterogeneous demand response and a firm's dynamic entry decision with endogenous prices accounting for the demand response to the number of same brand outlets, but we leave this to future research.
Consider a geographic market m in time t, in which H (or to be precise) hotels compete. In each market, customer i either decides to stay at one of the H hotels or chooses the outside option. Our data are from one of the largest hotel chains in the world with a wide range of brands; H hotels are the focal chain's franchisees, and the outside options are to stay at hotels that are not part of this chain or not stay at any hotel. Customer i gets the following utility if she stays at hotel h in time t. Otherwise, she gets the normalized mean zero utility (i.e., the outside option):
Graph
( 1)
includes price, and seasonality dummies for summer (June–August) and winter (November–January). We operationalize a positive pathway effect by the number of same-brand hotels ( ) in as hotel h in market m at time t, representing consumer response to same brand encroachment. For example, with three Hilton hotels in a market, there would be two same brand hotels for each Hilton hotel. Trials of other functional forms of the number of same brand hotels, such as a nonlinear function, produce similar results (for robustness checks, see Web Appendix W2.1).
captures the product characteristics observed by customers but not the researcher. It is possible that both price and are correlated with , which raises endogeneity concerns that can bias parameter estimates. In the estimation section, we address this issue using instrumental variables, generous fixed effects, and institutional knowledge in the timing of hotel entries. is an idiosyncratic random term, assumed to be i.i.d. Type I extreme value.
We further model as follows:
Graph
( 2)
captures individual hotel characteristics such as amenities, size, capacity, and location characteristics. is a set of market × six-month dummies to control for time-changing market-specific unobserved heterogeneity. We include these fixed effects to address potential endogeneity issues with identifying the parameter for . is then considered a residual common demand shock.
For the kth observed product characteristic , we specify the following random coefficient:
Graph
( 3)
is the mean sensitivity parameter that is common across customers for product characteristic k. denotes a vector of customer 's observed attributes, and represents a taste coefficient for product characteristic k that varies with observed attributes. is a coefficient for the standard deviation of unobserved customer attributes on , assumed to be independently distributed standard normal. This parametrization allows covariance between product characteristics and observed customer attributes; if customers with higher have a higher preference for X, the model estimates a positive .
By grouping demand parameters into , where are macro parameters that are common across customers, and are micro parameters that are individual taste dependent (i.e., ), the customer utility is rewritten as follows:
Graph
( 4)
is the mean utility that is common across individuals, and accounts for customer heterogeneity in preferences for different product characteristics, where denotes the interaction between and customer heterogeneity and . The predicted market share is then calculated by integrating the probability of each customer staying at hotel h in market m at time t over all customers:
Graph
( 5)
The franchisee sets its own hotel price (per night) without restrictions from the franchisor. Thus, we assume that hotel h in market m at time t decides on the room price to maximize the following profit function:
Graph
, , , and are price, marginal cost, predicted market share, and the total market potential, respectively. The optimal price is determined via the first-order conditions specified next. Note that hotel 's market share depends on other hotels' optimal prices , which is part of X. In a given market m at time t, the J first-order conditions with respect to price competition are then given by
Graph
( 7)
Marginal cost is not observed but estimated by inverting the first-order conditions to the following equation:
Graph
( 7)
where , p, and denote the vectors of marginal cost, price, and predicted market shares, respectively. is own-price share derivative . We use [ 7] supply-side model by adding observable margin cost components at hotel :
Graph
( 8)
where is the vector of marginal cost parameters, and is the unobserved component. includes the number of same-brand hotels and various fixed effects to study the effect of potential economies of scale (e.g., cost savings via volume discounts on supplies such as shampoo, soap, towels, etc.). Franchisees pay a portion of revenue to the franchisor as a royalty fee. Royalty fees are hotel specific and are time consistent in a typical 15- to 20-year franchise contract; therefore, we include individual hotel dummies to absorb the fees as part of marginal costs, along with other hotel-specific characteristics. Note that hotel dummies are more granular than market dummies; therefore, any market-specific cost variables are absorbed into hotel dummies. We also include market × six-month fixed effects for unobserved factors that affect marginal cost differently across markets over time.
We obtained a unique data set from one of the largest hotel franchisors in the world with multiple distinct brands, which provides substantial variation for investigating encroachment. More than 98% of the hotels are run by independent franchisees that set their own prices and pay a royalty fee to the chain. The royalty rate is assumed to be 10.61% of the gross franchisee revenue, which is the average for the focal chain in HVS Global Hospitality Services studies without much variation across brands.[ 8] Due to confidentiality concerns, we cannot disclose the number of brands in the focal chain. Suffice it to say that the individual brands are distinctive with unique logos, brand names, and service quality. The data include each hotel's brand, price, monthly demand data, and so on.
We have monthly revenue and room demand of the franchisor's hotels operated between September 2007 and March 2012 (55 months), from which we derive price (revenue per room night), consistent with previous work such as [14], [21], and [38]. The data include information on the percentage of business stays at each hotel in a given month, providing some insight into customer heterogeneity. Publicly available data on industry cost shifters, such as the median hourly wages of hotel employees from the U.S. Bureau of Labor Statistics, serve as instrumental variables for price.
Market share is defined as the demand (nights × rooms sold) divided by the monthly market potential. Market potential is defined as the maximum of the sum of a franchisee's demand and the demand of its competitors (including other hotels in the chain and hotels that are not part of the focal chain) in a particular market over the data period; this value is constant across time periods. The competitor demand information was collected by the chain to ensure that a relevant, competitive set determines each franchisee's market; this value is verified via mutual discussion with the franchisee. For example, suppose two franchisees are in a given market: franchisee 1 and franchisee 2. Also suppose that the demand for month t is 10 and 20 for the two hotels, respectively. Additionally, assume that the surveyed competitors' demand for the period is 30 and 35, respectively. Market potential in month t is then max(franchisee 1 market potential, franchisee 2 market potential) = max(10 + 30 = 40, 20 + 35 = 55) = 55 hotel nights. We then again take the maximum of such market potential across time periods for the market. This information is likely more accurate than an industry association such as Smith Travel Research might provide.
The outside market share is defined as staying at hotels that are not part of the focal chain or not staying at any hotel. Web Appendix W3 provides further justification of the market potential being constant over time and discusses the outside option related to time trends.
To properly assess encroachment and its impacts, our definition of a market needs to correctly capture the true local competition faced by each franchisee. One constraint is that the data do not include the address or location of each hotel. The location information provided is a geographic tract, which is defined by the combination of a metropolitan statistical area designation and location type (e.g., near an airport, highway or resort; suburban). For example, in the San Francisco-Oakland-Fremont metropolitan statistical area, the Oakland airport and San Francisco downtown area are provided as two separate tracts.
One problem is that some tracts may encompass multiple local markets. For example, three major highways crossing the state lie within the "Wyoming Interstate" tract. It would be erroneous to assume that all the hotels along these highways compete against one another (for a map of this area, see Web Appendix W4.1). Their inclusion could result in unreasonable parameter estimates, and more importantly, they do not reflect local franchisee competition, which is crucial to correctly specify Equation 5. We exclude rural metro/towns and suburban areas citing a similar argument. Therefore, we exclude such tracts (i.e., suburban, rural metro/town, and interstate locations) from estimation and retain tracts that are more likely to reflect a single market: airports, resorts, and urban areas. Absent more granularity and location information, we are unable to accord all the excluded tract data to the necessary unit of analysis. We also exclude markets with a single hotel of the focal chain during the entire period, as they may not be representative markets where our counterfactual simulations are relevant.
Web Appendix W4 details the process of tract exclusion and considers various empirical tests to justify their exclusion. Logit model results suggest that our main model results are robust to the inclusion of single hotel tracts[ 9] and tracts with missing variable information such as wages. Importantly, including suitably matched excluded suburban and rural metro locations that are likely small, well-defined markets yields similar results (see Web Appendix W4.4), which suggests that our findings are not limited to urban, airport, and resort locations but can generalize to suburban and rural metro/town locations.
The final estimation sample consists of 21,644 hotel-market-month observations of 447 unique hotels in 128 geographic markets across 34 states. Although the data set spans 55 months, we do not necessarily observe 55 months for every tract, depending on the chain's presence in a market. Out of 447 franchisees, 127 entered the market during the data period. Forty-three percent of the entries (54 hotels) involved one or more instances of same brand encroachment, whereas the remainder did not. The data include 6,924 market-month observations.
The focal chain consists of a mix of multiple economy and luxury hotel brands; we are prohibited from revealing each brand's market share and price point for reasons of confidentiality. Instead, we provide summary statistics of monthly demand, market share, mean hotel size and more in Table 2. The total market share of the chain is approximately 28%, which is generally consistent with its known overall market share in the industry. The average room price is $107.45 (SD = $39.05) per night. The large standard deviation is driven by the fact that the chain has multiple brands: the highest-end brand has a 9.19% market share with the average price of $192.79, whereas the lowest-end brand has a 6.84% market share with an average price of $73.68 per room per night.
Graph
Table 2. Hotel Summary Statistics.
| Statistic | Mean | SD | p25 | p50 |
|---|
| Average monthly demand (room nights) | 4,028.59 | 3,061.87 | 2,001.00 | 3,021.00 |
| Total monthly market potential (room nights) | 61,214.93 | 42,647.24 | 30,628.00 | 47,111.00 |
| Average market share at a focal hotel | .077 | .046 | .043 | .069 |
| Average room price per night ($) | 107.45 | 39.05 | 83.97 | 99.55 |
| Average hotel size (room nights) | 200.49 | 132.66 | 104.00 | 150.00 |
1 Notes: Summary statistics are calculated across 21,644 hotel-month observations.
Based on the 6,924 market-month observations, the average number of hotels is 3.13 per market in a month (SD = 2.13). All brands have at least one market with same brand encroachment (.99 hotels on average per market). Figure 2 overviews the distribution of this chain's hotels across market-month observations. The top graph shows that approximately 90% of market-month observations have multiple hotels, and the bottom graph indicates that approximately 38% of the market-month observations with multiple hotels have at least one brand with same brand encroachment. Some market-month observations even have two or more such brands.
Graph: Figure 2. Distribution of market-month observations.
Business stays account for 52% of hotel visits, and we use this proportion to estimate different demand response types. We employ multiple moderators to understand brand heterogeneity (e.g., brand type and age) and the number of cross brands (two brands in the chain that share a common moniker in the market). Overall, 21.8% (SD = 41.2%) of the 21,644 hotel-month observations represent luxury brand hotel observations, and 16.2% (SD = 36.9%) represent newer brands (introduced after 1997). The average number of cross brands is.76 hotel (SD = 1.14) across hotel-month observations. We use wage data from the U.S. Bureau of Labor Statistics: the mean of median wages of desk clerks (maids) was $10.20/hour ($9.66/hour) with a standard deviation of $1.38/hour ($1.52/hour) during the data period.
The first component of the estimation matches the predicted market shares to the observed market shares in the data :
Graph
( 9)
where is the set of parameters to be estimated. [ 6] shows that a unique value of matches these two market shares through a contraction mapping (i.e., inner loop), as in [ 7].
We then construct moment conditions by assuming that demand shock is uncorrelated with Z, a set of instrumental variables including exogenous variables as well as a constant:
Graph
(10)
Note that price can be potentially correlated with the demand shock , which may create an endogeneity problem that could potentially bias the price coefficient estimate ([41]). To deal with this issue, we use [ 7] instruments, which are the averages of other hotel characteristics. Specifically, the first instrument is the average hotel size (i.e., number of hotel rooms) of all other hotels in the chain, and the second instrument is all hotels of differing brands of the focal chain in the market. The number of rooms in each hotel is determined by the combination of the availability of land for commercial use and the local government's zoning restrictions on the land use (e.g., the total number of floors due to height restrictions or the shape of the building) that are deemed to be exogenous with respect to the demand shock .[10] Web Appendix W5 provides the test results of various instrumental variables. In the estimation, we also include functions of these instrumental variables and cost shifters such as the median wages of desk clerks and room service personnel.
Another endogeneity concern involves the positive pathway variable as defined by the number of same-brand hotels in a market. Part of may contain unobserved heterogeneity that affects the number of hotels of the encroached brand in a given market. One way to deal with this concern is to include market dummies. However, market dummies may not be sufficient to control for unobserved heterogeneity. For instance, a hotel within walking distance to a popular tourist attraction (e.g., Disney World) would draw higher customer preference over a hotel that is far from it, despite competing in the same geographic market. We therefore include individual hotel dummies ( ) as part of product characteristics.
While individual hotel dummies control for endogeneity from time-consistent demand heterogeneity, concerns remain about market-specific time-varying unobservables. Because the positive pathway is proxied by the number of same-brand hotels and its change in a market, if such unobserved heterogeneity is correlated with different hotel entry patterns, it can ultimately bias the estimate ([29]; [35]). We address this issue in several ways. First, we include fixed effects specified at the market × six-month level to flexibly control for the unobservables that may influence hotel entry and, thus, the change in the number of same-brand hotels. The fixed effects also account for time-changing outside options to some degree by allowing them to have different intercepts.
Second, we exploit the institutional knowledge of the industry regarding the timing of hotel entry. This entry decision may be decided at large time intervals (e.g., hotel entry planned in the second half of 2015); interviews with chain executives reveal that the exact timing of entry is difficult to predict due to the nature of the commercial real estate development process. This timing varies within a six-month window, depending on many unpredictable external forces, such as financing commercial real estate mortgages, passing inspections, and obtaining permits from local government. Therefore, we assume that the variation in the number of same hotel brands is exogenous within a six-month window.
This approach is consistent with the precedent of [12], who similarly assumes exogenous entry timing of generic drugs due to unpredictable FDA approval processes. We first checked the validity of this assumption by showing that the month in the six-month window cannot predict entry (see Web Appendix W2.2). We checked robustness of the results with additional control variables (see Web Appendix W2.3). Additionally, we examine incumbents' demand and price surrounding the entries of same and different brands and find that both increase only for same brands post entry (Web Appendices W6 and W7, respectively). These results corroborate our modeling approach that customer utility or willingness to pay increases only in the case of same brand encroachment.
Since we only observe hotels from one chain, it is difficult to completely rule out the possibility that the entry timing in this chain may coincide with entry/exit of unobserved nonchain hotels. Two factors mitigate this concern. First, the uncertainty in the commercial real estate development process equally applies to unobserved hotels. Thus, it is unlikely that their entry/exit patterns will systematically coincide with our chain's entry timing within the six-month window and result in an overestimate of our results.[11] Second, we do not observe the full range of competitive options, which can lead to an attenuation bias via measurement error (i.e., our estimates may be conservative; for a similar setting in the airline industry, see [34]). Finally, we consider other endogeneity concerns (e.g., the granularity of our fixed effects, independent hotels rebranding as new hotels in our data) in Web Appendix W8.1.
The estimation strategy is based on the generalized method of moments estimation combined with micro moments (e.g., Berry, Levinsohn, and Pakes 1995 2004) to estimate customers' heterogeneous demand response in terms of the percentage of business stays. Supply moments are estimated to evaluate the degree to which an economies of scale explanation might be operative after controlling for hotel and time-varying market factors. Technical details of the estimation are included in Web Appendix W8.
We estimate a homogeneous logit demand model ( in Equation 3) to assess data variations and address endogeneity issues using instrumental variables and fixed effects before estimating the full random coefficient demand model. The logit model regresses on price and the number of same brand hotels in the market (SB), as well as hotel seasons (June–August for summer and November–January for winter). is the market share of the outside option. This is Model 1, and the results are displayed in Table 3. For Model 2, we add hotel fixed effects to control for unobserved hotel-specific characteristics (e.g., amenities, size, capacity, location). Model 3 adds market × six-month fixed effects to control for time-varying market-specific unobservables that may result in hotel entry, which in turn can bias the SB estimate. Finally, we run an instrumental variable regression to address the endogeneity of price (Model 4).
Graph
Table 3. Logit Model Results.
| Model | (1) | (2) | (3) | (4) |
|---|
| Regression | OLS1 | OLS2 | OLS3 | IV |
|---|
| Hotel FEs | N | Y | Y | Y |
| Market × six-month FEs | N | N | Y | Y |
| Price ($00) | .121*** | .552*** | .549*** | −1.297*** |
| (.012) | (.013) | (.015) | (.367) |
| # Same Brand Hotels (SB) | −.105*** | .199*** | .090*** | .059*** |
| (.004) | (.008) | (.008) | (.015) |
| Constant | −2.363*** | — | — | — |
| (.014) | | | |
| Observations | 21,644 | 21,644 | 21,644 | 21,644 |
| R2 | .079 | .841 | .875 | - |
- 2 *** p < .01.
- 3 Notes: The dependent variable represents a transformation of customer utility in a logit model, expressed as a logarithmic function of market share minus a logarithmic function of the outside option market share. Hotel season dummies for summer and winter are included for all specifications. Robust standard errors are reported in parentheses. FE = fixed effects. OLS = ordinary least squares regression estimation. IV = instrumental variables regression estimation.
The results of the logit regression using price and SB in Model 1 imply that customers prefer higher prices and dislike having more of the same-brand hotels in a market, as SB is negative and significant. When we include hotel fixed effects in Model 2, a positive pathway emerges, and the number of same-brand hotels increases customer utility, suggesting that demand is positively associated with same brand encroachment. The price coefficient remains positive and statistically significant, an unconventional finding. Including additional market × six-month fixed effects in Model 3 substantially reduces the SB estimate (from.199 to.09, although still significant at 1%), underscoring the value of controlling for time-varying market-specific unobservables.
However, after instrumenting price with the average size of the focal chain's other hotels and that of different brand hotels in the market, the price coefficient in Model 4 is now negative (−1.297, p < .01); this suggests that an instrumental variable approach results in more reasonable results (i.e., hotel demand decreases as price increases). The first-stage regression results show that the instruments have enough explanatory power with the first-stage F-statistics of 112.9. Collectively, the logit results suggest consumers respond positively to same brand encroachment—a positive pathway that can create revenue benefits among same brand franchisees if the negative pathway or market competition is not severe.
The results are robust to different functional forms of SB, such as a log function of the same-brand hotels (see Web Appendix W2.1). In comparison to the log specification, the linear function of same brand hotels results in a better fit ( =.714 versus.694 under the log function) and is therefore used throughout.
We now estimate the full model with heterogeneous customer preferences ( in Equation 3). Table 4 provides macro parameter estimates of common customer preference on all observed hotel and market characteristics explaining consumer utility ( in Equation 3).
Graph
Table 4. Demand Macro Parameter Estimates in Equation 3.
| Coef. | SE |
|---|
| Price ($00) | −1.544*** | .302 |
| Number of Same Brand Hotels (SB) | .097*** | .037 |
| Summer (June–August) | .092*** | .008 |
| Winter (November–January) | –.426*** | .023 |
| Luxury | .759*** | .013 |
- 4 *** p < .01.
- 5 Notes: A full set of individual hotel dummies and market × six-month dummies are included in the estimation. We obtained estimates for the luxury brand dummy using a minimum-distance procedure.
The mean price parameter estimate is negative and significant (−1.544, p < .01), implying that customers have a preference for lower prices. Combined with the price heterogeneity estimates in Table 5, this results in an average own-price elasticity of −1.33. Seasonality has a significantly negative (positive) effect for winter (summer), which suggests that demand peaks in summer. We draw additional insights on customer taste by running a generalized least squares regression ([11]; [32]) in which we regress hotel dummy estimates on a luxury brand dummy (=0 if economy brand). As expected, customers prefer luxury over economy hotels (.759, p < .01). Parameter estimates for all dummy variables ( and in Equation 2) are summarized in Web Appendix W9.
Graph
Table 5. Customer Heterogeneity Parameter Estimates in Equation 3.
| Coefficient | SE |
|---|
| Price − Business Stays | .009*** | .003 |
| Price − Unobserved Heterogeneity | .516*** | .010 |
| # Same Brand Hotels (SB) − Business Stays | −.037*** | .012 |
| # Same Brand Hotels(SB) − Unobserved Heterogeneity | .042 | .279 |
6 Notes: *** p < .01.
Customer heterogeneity coefficients ( for observed characteristics and for unobserved heterogeneity in Equation 3) are presented in Table 5. A positive price estimate for business stays implies that business stays exhibit lower price sensitivity, in line with the conventional wisdom in the industry. The estimate of unobserved heterogeneity (or the standard deviation of the normally distributed random coefficient) is statistically significant for price, capturing a large variance in customer taste. The SB estimates suggest that business stays are less sensitive to the number of same brand hotels than nonbusiness stays (.097 –.037 = .060 for business stays versus.097 for nonbusiness stays). The estimate of unobserved heterogeneity for SB is nonsignificant.
The estimation of the supply side (Equation 8) shows that the effect of SB on marginal costs after controlling for hotel and market-time fixed effects is slightly negative but not significant (–.008, p = .233), suggesting a lack of economies of scale (for detailed estimation results, see Web Appendix W9). An alternative covariate is the number of same-brand hotels in the nation in a given time period if the level of economies of scale is at the national as opposed to local markets. Our analysis shows that the economic significance of such an effect is very close to zero. The estimated marginal cost is $27.11 per room night, which approximates the marginal cost found in the hotel industry (e.g., [16]).
We can estimate the economic value of this effect by comparing the observed data with a scenario in which the SB parameter is set to zero; conceptually, this approximates a scenario in which consumers are not responsive to an increase in the number of same brand hotel outlets (i.e., no positive pathway effect). Note that in this scenario, franchisees may set different optimal prices. The first-order conditions defined in Equation 6 allows us to recalculate the new optimal price. We then compare the scenario's outcome against our data in Table 6. The results suggest that a positive pathway contributes to a slight increase in average price and accounts for approximately 9.4% (8.3%) of the average hotel franchisee monthly revenue (profit) in the focal market. Also, 4% of the $707.5 million gross royalty income is estimated to result from the positive pathway effect. Therefore, the positive pathway is not only statistically significant, but also economically relevant regarding the franchisee's economic welfare.
Graph
Table 6. Positive Pathway Economic Value.
| Variable | Model with SB = 0 (Simulation) | Model with SB (Data) | % Change |
|---|
| Avg. Price ($) | 116.11 | 117.09 | .85 |
| Avg. Market Share (%) | 6.47 | 6.63 | 2.58 |
| Avg. Franchisee Revenue ($000) | 577.39 | 631.53 | 9.38 |
| Avg. Franchisee Profit ($000) | 358.73 | 388.58 | 8.32 |
| Total Royalties ($M) | 707.50 | 734.56 | 3.82 |
7 Notes: Franchisee revenue and profit are monthly figures, and royalties are summed up across the data period.
In this section, we investigate whether the positive pathway varies across a range of brand dimensions such as quality (luxury vs. economy brands), brand age (more vs. less established brands), and cross-brands (brands with an overlapping moniker).
In Model 1 of Table 7, we reestimate the logit model (Model 4 in Table 3) with an interaction term between and (an indicator function equals 1 if hotel h is a luxury brand and 0 otherwise). The parameter estimate of the interaction term is not statistically significant, implying that the positive pathway for luxury brands is similar to economy brands.
Graph
Table 7. Heterogeneous Effects of the Number of Same Brand Hotels.
| Model | (1) | (2) | (3) | (4) |
|---|
| Heterogeneity | Brand Quality | Brand Age | Cross-Brand | Combined |
|---|
| Price ($00) | −1.326*** | −1.264*** | −.936*** | −.888** |
| (.370) | (.362) | (.358) | (.351) |
| Number of Same Brand Hotels (SB) | .056*** | .057*** | .056*** | .053*** |
| (.015) | (.014) | (.013) | (.013) |
| .033 | | | .010 |
| (.049) | | | (.043) |
| | .237*** | | .246*** |
| | (.063) | | (.065) |
| | | .048*** | .051*** |
| | | (.013) | (.013) |
| Observations | 21,644 | 21,644 | 21,644 | 21,644 |
| R2 | .709 | .720 | .771 | .778 |
- 8 ** p < .05.
- 9 *** p < .01.
- 10 Notes: The dependent variable represents a transformation of customer utility in a logit model, expressed as a logarithmic function of market share minus a logarithmic function of the outside option market share. Hotel season dummies for summer and winter, hotel fixed effects, and market × six-month fixed effects are included in all specifications. Robust standard errors are reported in parentheses. SB = same brand.
Prior to 1992, the chain was composed of 57% of the chain's brands relative to the data period. Three or more brands were added to the chain's portfolio after 1997, and then the full set of brands developed together. We define a dummy variable equal to 1 for these newer, less established brands and 0 for older brands. In Model 2, the interaction of with is positive and significant (.237, p < .01), implying that newer brands are associated with even higher positive pathways relative to older brands, which we surmise could be because well-established, older brands face ceiling effects that limit the positive pathway. This result suggests that encroachment may help newer brands enter the market, leading to more hotel choices that can potentially increase consumer welfare.
We consider the possibility of additional positive pathways through cross-branding, in which two outlets share a common aspect of the brand name. For example, a "Hyatt Place" hotel may benefit whenever another hotel outlet incorporating "Hyatt" (e.g., "Hyatt House") is added to the market. Among the chain's brands, two have similar names (e.g., brands A and AB): brand A's moniker is one word, and brand AB has two words starting with brand A. The second word (B) does not directly signal superior overall quality but refers instead to the speed of service. If positive pathways are generated via the number of outlets with same brand monikers, we may observe a positive pathway not only between two outlets of the A brand, but also between A and AB brands. We assess this possibility by adding , a count of the number of cross-brands in the market. For instance, if there are two A hotels and one AB hotel, for A (AB) hotels would be 1 ( 2). For brands other than A and AB hotels, equals 0. Model 3 results indicate that, in addition to the same-brand encroachment effect, there exists a smaller positive cross-brand effect on customer utility between A and AB hotels (.048, p < .01), corroborating the existence of a positive pathway. Model 4 reflects the combined model with robust results. Collectively, these analyses unpack how a positive pathway effect might be heterogeneous in brand characteristics. While we do not observe significant additional positive pathways with regard to brand quality, we do observe additional positive pathway effects for newer and cross-brands.
To further assess robustness, we conduct a placebo test to see if a cross-brand effect is no longer observed for the count of the number of hotel brands with dissimilar names. For instance, in place of brand A's , we count the number of brand E outlets in the market, where brands A and E are distinct names without moniker overlap. As expected, the coefficient for the placebo brand term is not statistically significant. We also conduct additional placebo tests using the number of future same brand entries and do not find significance. See Web Appendix W10 for further details.
Booking channels offer varying forms of product comparison and purchase convenience in the customer's purchase process. We consider positive pathway effects across booking channels that support varying degrees of purchase intent and brand loyalty. One advantage of OTA channels such as Expedia or Orbitz is that they allow customers to compare across brands and options in terms of price, dates, and locations, which may be particularly valuable to customers who are seeking a specific type of deal and are open to a range of brands. OTA customers may be more likely to make occasional purchases and may not be brand loyal. In contrast, customers who book through a chain's official website are likely to have a preference for the brand and may not need an array of brand options for their purchase; they may be brand loyal or repeat customers. A positive pathway for demand may be less likely for this group due to saturation or ceiling effects, relative to OTA channel customers. Put differently, OTA consumers are more likely to benefit from a positive pathway, and this impact should be observable via heightened demand as encroachment increases.
We investigate this possibility by estimating a series of ordinary least squares (OLS) regressions for different online channel demand over brief windows (±3 months) around hotel entries for incumbent hotels, excluding the entered hotels. Details and results of the estimation are included in Web Appendix W11. We find that the incumbent's demand (9.946, p < .10) and the proportion of demand (.013, p < .05) through OTAs increase relative to other brands when a same-brand hotel enters the market. However, there is no significant corresponding effect through the chain's website channel. These significant interaction terms support our contention that same brand encroachment spurs a positive pathway effect and increases customer preference in OTA channels. Thus, incumbents benefit more from purchases in the OTA channel than the chain's channels. We evaluated the price response to determine whether a demand increase in the OTA channel decreases the overall price (hotel revenue per room). We find an increase in price for same brand incumbents post-entry, suggesting higher willingness to pay in customer utility (see Web Appendix W7 for further details).
We conduct similar analyses with offline booking channels associated with the chain, namely, traditional travel agents and call centers, but did not find any significant results. We conjecture that because travel agents are experienced in hotel bookings, they are less influenced by the addition of same brands. In addition, customers who book through the chain's call centers, like those who book through the chain's website, may have brand loyalty or preference that would make them less sensitive (i.e., a ceiling effect) to same brand encroachment.
We next consider various explanations for the observed positive pathways. Although we are not able to unequivocally isolate their effects, a careful consideration of the empirical evidence can suggest which explanations might be operative.
Our results in prior sections suggest that brand quality signaling and awareness mechanisms are likely operative. Evidence of an additional positive pathway on customer utility for cross-brands and a moderating effect of newer brand outlets further corroborates this possibility. Additionally, the demand patterns show that consumers who purchase through the chain's booking channels (as opposed to OTA channels) are less likely to reflect a positive pathway effect, possibly due to brand loyalty and familiarity. Online advertising in OTA channels may also result in greater exposure of same brand hotels, which would increase brand awareness and quality signaling. This is consistent with our finding in Table W22.
We have suggested that a possible positive pathway might arise if consumers learn from the presence of multiple brand outlets. Although our hotel-level data prohibit us from isolating this effect at the customer level, we explore this potential mechanism via an online experiment. If learning improves as the number of hotels in a market (i.e., brand exposure) increases, consumers may learn about hotel quality more accurately (i.e., correct quality association of the brand with more outlets), absent more information such as budget constraints, pricing, location, and so on. Better learning would thus be evidenced by greater accuracy in markets with more hotels regardless of the hotel quality level.
We test this possibility using a design in which we manipulate hotel quality (high vs. low) and number of hotels in a market (two vs. five) in a 2 × 2 between-subjects design. The dependent variable is the recall accuracy of the focal hotel brand quality. If the consumer learning mechanism holds, one would expect that recall inaccuracy would be higher in the two-location condition than the five-location condition. Web Appendix W12 contains more details regarding data collection, stimuli, and analyses.
The results are supportive: inaccuracy is significantly higher in the two-location than the five-location conditions across both high- and low-quality hotels. When there are fewer hotel outlets, the inaccuracy is mainly driven by the participants' reported ratings gravitating toward the condition mean, consistent with an explanation of not learning well. As a result, for higher-than-average hotel quality (i.e., as in our data), more same-brand hotels are associated with higher perceived quality. Collectively, these results provide evidence for the possibility that more outlets can increase preference via consumer learning, supporting a positive pathway effect.
More broadly, our results suggest that customer utility and market demand can be positively impacted by the addition of a new outlet when the addition is of the same brand, a newer brand, a cross brand, or purchased in an OTA channel.
If consumers prefer scarce boutique hotels (e.g., Hotel Indigo) over more standard hotel offerings with wide market coverage (e.g., Hilton Garden Inn), then the existence of multiple establishments of the same brand would decrease utility for the encroached brands. Research on luxury goods consumption is consistent in this regard, showing that scarcity and exclusivity creates value for customers ([37]; [44]). However, our results do not support differing pathway effects for luxury hotel brands compared with economy brands, suggesting that scarcity is an unlikely explanation.
Television advertising is difficult to target or tie to a specific outlet opening, given the difficulty of predicting a hotel opening even within a six-month window. Unlike fast food customers, hotel customers are not necessarily local, which can complicate advertising targeting. Therefore, we consider television advertising unlikely in our context.
Although our focus has been on demand-side mechanisms for positive pathways, several supply-side mechanisms have often been discussed in the franchise literature as well, such as co-ownership, cost agglomeration, economies of scale, and capacity constraints. While these mechanisms are useful, they cannot adequately explain the positive pathway effects on customer utility or are irrelevant in our research context, as explained in Web Appendix W13.
A key advantage of a structural modeling approach, relative to regression analyses, is the ability to conduct counterfactual simulations. In this section, we explore two counterfactual simulations. The first simulates the impact of encroachment reduction, which has implications for shaping policy and represents a possible action for conflict reduction. It illustrates how removal of an encroaching outlet impacts the remaining hotels' optimal pricing and customers' subsequent hotel choice. The second counterfactual explores a key boundary condition of intrabrand competition: markets with varying same brand density. These counterfactuals have both practical and theoretical implications, and they allow us to explicitly illustrate the role of positive and negative pathways on incumbent performance.
We simulate encroachment reduction by removing one same brand franchisee and allowing the remaining franchisees to reoptimize their prices through Equations 6 and 8. We compare this simulated market outcome to the current model estimates and quantify the net aggregate effect on the remaining franchisees' performance. Because we do not have data on hotels outside the chain, our results assume that such hotels would not respond to the simulated policy changes.
We consider all combinations of the removal of a single same brand hotel using the following process: first, in each market, one same brand hotel is removed. If there are three brand A hotels, then one is randomly selected (A1) for removal. Second, the remaining hotels (A2 and A3) reoptimize their prices, according to Equations 6 and 8. Third, market outcomes—including franchisee price, demand, revenues, and profits, as well as franchisor revenues and profits—are then recalculated. Finally, the foregoing three steps are repeated for each remaining hotel (A2 and A3) and across all markets.
We further decompose the impact of encroachment reduction by considering its effects on same and different brand outlets. Table 8, Panel A, reflects the performance impact on revenues and profits with one fewer same brand outlet. The first column reflects the simulation results (i.e., less encroachment), the second column reflects our model estimates (i.e., more encroachment), the third column reflects the mean percentage change between the foregoing, and the fourth column reports 95% confidence intervals. The overall impact of more encroachment results in an average revenue and profit loss of 5.3% (i.e., −$43,810 in revenue and −$26,090 in profit per month) for all incumbents. This result is consistent with the negative pathway findings in the empirical literature. This value is meaningful, as the "impact threshold," or the minimum drop in revenue that must be observed for a franchisee to bring suit for encroachment in Iowa, is 5% (Iowa Code 523 H, 1995).
Decomposing this further for same- and different-brand incumbents, we find that the effect is even greater on different-brand franchisees in the franchise system, with revenues and profits declining significantly by 7%–8% (i.e., −$59,834 in revenue and −$35,586 in profit per month). In contrast, more encroachment has little to no impact on same brand franchisees: although revenues decrease significantly (i.e., −$5,914 per month or −.58%, 95% CI [−.68%, −.48%]), the impact on profits is not statistically significant (−$3,659 per month or −.07%, 95% CI [−.18%,.03%]). Given that franchisors earn a royalty rate on revenues and franchisees are profit maximizing, these effects are economically meaningful. The results suggest that while a negative pathway dominates for different brands, a positive pathway seems to mostly offset a negative pathway for same brands (i.e., mitigation). This represents a critical contribution to our understanding of encroachment.
Another dimension to consider is the number of existing brands, or a brand's "density," in a market. The fewer the number of same brand outlets, the lower the brand's density, and potentially, the weaker the negative pathway from adding yet another outlet. Positive pathways might then dominate negative pathways in the case of same brands, potentially improving franchisee performance. In contrast, in markets in which brand density is high, firms have more competition for customers; thus, we are likely to observe that negative pathway effects prevail.
Suppose there are two markets that vary in terms of the number of A hotels:
- Low-density market ("Market L") consists of two A hotels, one B hotel, one C hotel, and one D hotel. In this market, we can calculate a brand density ratio for A hotels to be 40%.
- High-density market ("Market H") contains three A hotels, and one B hotel, and one C hotel. In this market, the same-brand density ratio for A hotels would be 60%.
If one A hotel is removed, we would expect its impact on the remaining A hotels to be smaller in Market H because the marginal positive pathway is likely lower. We divide the data set into low (≤50%) and high (>50%) same brand density ratio markets.[12] Note that the brand density ratio reflects the number of same-brand hotels over the franchisor's total number of hotels in the market. We provide further evidence that the brand density ratio is a good proxy for the same brand concentration with respect to all hotels, including hotels that are not part of the chain, in Web Appendix W14.1.
Table 8, Panel B, displays the performance impact. Consistent with a negative pathway mechanism, encroachment generally hurts franchisees, reducing average revenue and profits by 4%–6% regardless of the market's brand density. This performance decrement increases for the remaining different brand franchisees, resulting in a reduction in revenue and profits of 7%–8% and replicating results from the previous counterfactual. In contrast, the performance decrement is minimized and can even be positive for same brand franchisees. In high-density markets, they experience only a 3% reduction in performance (−$18,954 in revenue and −$12,011 in profit per franchisee per month). In low-density markets, incumbent performance improves by approximately 2% ($4,349 in revenue and $2,914 in profit per franchisee per month).
Graph
Table 8. Counterfactual Simulations of More Encroachment on Franchisee Monthly Performance.
| A: More Encroachment Impact Across Franchisee Brand Types |
|---|
| Impact on remaining franchisees | Outcome | Less Encroachment ($000) | More Encroachment ($000) | Mean % Change, [95% CI] |
|---|
| Same-brand franchisees only | Revenue | 567.41 | 561.50 | −.58, [−.68, −.48] |
| Profit | 342.00 | 338.34 | –.07, [−.18,.03] |
| Different-brand franchisees only | Revenue | 87.82 | 81.99 | −7.23, [−7.28, −7.18] |
| Profit | 475.30 | 439.72 | −7.52, [−7.57, −7.47] |
| All franchisees | Revenue | 78.63 | 736.82 | −5.25, [−5.30, −5.20] |
| Profit | 435.67 | 409.58 | −5.31, [−5.36, −5.25] |
| B: More Encroachment Impact Across High and Low Brand Density Markets |
| Mean % change, [95% CI] | Outcome | Low-brand-density markets () | High-brand-density markets () |
| Same-brand franchisees only | Revenue | 1.65%, [1.55%, 1.75%] | −3.41%, [−3.57%, −3.26%] |
| Profit | 2.27%, [2.17%, 2.37%] | −3.05%, [−3.22%, −2.89%] |
| Different-brand franchisees only | Revenue | −7.19%, [−7.24%, −7.14%] | −7.90%, [−8.09%, −7.71%] |
| Profit | −7.48%, [−7.53%, −7.42%] | −8.34%, [−8.54%, −8.14%] |
| All franchisees | Revenue | −5.43%, [−5.49%, −5.37%] | −4.35%, [−4.48%, −4.22%] |
| Profit | −5.53%, [−5.60%, −5.47%] | −4.15%, [−4.30%, −4.01%] |
11 Notes: In the simulation, we iteratively remove same brand encroaching hotels in the market one at a time and average the outcomes. Same brand franchisees have the same brand as the removed hotel in each iteration. Less encroachment means one less encroaching chain outlet in simulation. More encroachment reflects the main model estimates.
Figure 3 sheds further light on these point estimates by illustrating the distributions of franchisees' monthly revenue change: the midpoint of the graphs reflect no change, while areas to the right reflect the net outcome of a positive pathway dominating a negative pathway, reflected in higher franchisee revenues. Panel A of Figure 3 reflects the relative distributions for the first two cells of the last rows in Panel B. It is worth noting that the brand-density markets' plot is bimodal. We explore this further in Figure 3, Panels B and C.
Graph: Figure 3. Distribution of incumbent revenue change across high- and low-brand-density markets.
Figure 3, Panel B, shows revenue change from the perspective of same brand outlets (i.e., the first two rows in Table 8, Panel B). In low-brand-density markets, a clear positive pathway impact is evident for same-brand franchisees; this plot is similar to the positive pathway impact for low-brand-density markets in Figure 3, Panel A. This distribution shift suggests that on average, 70% of franchisees benefit from having a same brand hotel added in their vicinity. With high brand density markets, revenues are generally lower, consistent with a negative pathway dominating; however, even under these conditions there is a sizable portion of revenue for which positive pathways dominate.
Figure 3, Panel C, illustrates franchisee revenues decrease from the perspective of different brand outlets (i.e., the middle rows in Table 8, Panel B). This plot suggests that the addition of a different brand hotel hurts franchisee revenues regardless of market brand density. It also suggests that there are no positive pathway gains for outlets whose brand is different from the new outlet—only a negative pathway.
Collectively, the counterfactual analyses and associated plots provide a more nuanced understanding of the impact and market conditions under which positive and negative pathways to franchisee performance occur. Web Appendix W14.2 presents histograms of selected market revenues pre- and post-simulation.
A chief advantage of franchising organizations is rapid growth in the number of outlets and service availability. This growth may inevitably result in more same brand franchisees being located close to one another. Prior to our research, encroachment was only considered in terms of negative pathways. However, our work provides important insights into the circumstances under which this might not be the case. Adding a same brand outlet can in fact lead to positive pathways via customer utility enhancement, which can subsequently mitigate negative pathways and even improve franchisee revenue and profits for same brand outlets in low-brand-density markets.
The reduced negative performance impact of encroachment on same brand franchisees suggests that encroachment may not be as universally harmful as once thought. Moreover, the prevailing positive pathway increasing both revenues and profits in low-brand-density areas suggests that firms should seek to encroach more in those areas (i.e., strength in numbers). To this end, we find that newer brands benefit more, holding promise for market entry strategies, as do cross brand locations and OTA bookings. Our work calls for managers and researchers to rework their view of encroachment to recognize that it may not be unilaterally negative.
Beyond the hotel industry, if encroachment involves mostly negative pathways dominating positive pathways, one may not observe any increase in performance related to same brands. Alternatively, if products are well differentiated such that negative pathways are relatively small, newer brand franchisees may achieve improved performance with same brand encroachment. On the other hand, different brand encroachment does not involve positive pathways, so we expect it to be seen as an adverse event to incumbents in other industries as well.
Encroachment has been a contentious issue in the marketplace, but our research is the first to identify potential benefits and positive pathways from the addition of a same brand via a customer-centric viewpoint. Our ability to trace these pathways to improved customer utility underscores the power of the customer in a research area that has mostly focused on the franchisor's view of the market. Encroachment can be positive through customer utility, and we demonstrate where and when these benefits occur.
Being the first to distinguish the performance impact for same- versus different-brand outlets in a franchise system adds nuance to our understanding of how franchise systems should go to market and how brand characteristics play into it. Specifically, franchisors might prioritize markets with low same brand density for more locations. In this regard, we observe that a positive pathway effect can increase franchisee revenues by 1.7%, reflecting a 2.3% improvement in profits. This illustrates the potential "sunny side" of this controversial phenomenon. Importantly, franchisors can be strategic about when and where to reduce or expand locations, particularly as it pertains to franchisee conflict or relationship improvement.
Regarding channel conflict, franchisors face a fundamental trade-off in achieving their expansion goals; prevailing in a particular conflict, such as encroachment, can come at the cost of less location expansion. Our work identifies an alternate path, suggesting that additional same brand locations can result in a win-win outcome in low-brand-density markets and with various brand characteristics.
It is also worth noting that franchise contract specifics may make it difficult to generalize our results. For instance, encroachment is irrelevant in franchise chains that guarantee exclusive territories. Also, the type of franchising may affect generalizability. While many new franchises are business format franchising (i.e., franchisees pay running royalty payments to the franchisor as a percentage of sales), the traditional franchising model is still prevalent in sectors such as automobile dealerships and gasoline stations ([ 9]). In traditional formats, the franchisor sells the product to franchisees with the right to serve a certain geographic area instead of collecting ongoing royalties. Regardless of the format differences or other unique aspects of the hotel industry such as capacity constraints, we may yet observe positive pathways if brand awareness and quality signaling are relevant. We leave these issues for future research to verify.
From a consumer welfare perspective, our research sheds a new light on how to consider hotel choice offerings across a range of markets. Our findings of additional positive pathways (e.g., cross brands, newer brands, purchase channel types) to hotel performance suggest that such offerings might enhance consumer welfare in new markets and in the creation of valued product offerings. These insights are important for policies such as those described in The Network Expansion Handbook, which is developed by the International Franchise Association's Franchise Relations Committee to assist in establishing codes of conduct and setting expectations and compliance for encroachment practices.
Our counterfactual simulations help inform the fractious efforts around franchise regulation. Although franchisors may allow exclusive territories to individual franchisees, established franchisors often maintain legal rights in the final encroachment decision (for more detailed discussion, see [21]). However, the threat of litigation motivates franchisors to maintain goodwill. Our findings of prevailing positive pathways in low-brand-density markets and the substantially reduced negative pathway impact on same-brand incumbents indicate customer utility gains from having more same brand hotels. Future research might consider how a ban on same-brand encroachment, such as the Iowa Franchise Act of 1992, would affect customers.
Our finding that encroachment can be economically beneficial for same brand franchisees contributes to the literature, which has historically shown the opposite. Incorporating a consumer perspective represents a substantial departure from the dominant approach, which has historically focused on the franchisor's revenue and market share. Moreover, the customer's perspective enables us to illuminate the value of encroachment in terms of increased customer utility and demand for same brand franchisees. Although we find customer heterogeneity differences for hotel industry-specific factors such as business stays versus leisure stays, generalizability to other industries remains an area for future investigation. As an example, considering that hotel franchisees rely on out-of-town customers more than a fast food franchisee might, pathways for these two industries may differ.
We expand the scope of research in the marketing literature on franchise management, which has mostly focused on issues of governance and enforcement ([ 3]; [22]), or organizational forms such as the proportion of a chain that is franchisor versus franchisee owned ([ 4]). Our research highlights the phenomenon of encroachment as a substantive direction for future research in its own right.
Our work contributes to the growing body of work on firm entry. Similar to [15] and [42], we allow both negative and positive outcomes from entry, albeit in a franchising context. Their results support the positive effects that arise from economy-of-scale gains on the supply side or a concentration of product category brand locations. We, in contrast, focus on the effect of multiple same brand outlets on customer demand as opposed to the use of multiple-store formats or category effects.
Generalizing our findings beyond the hotel industry depends on industry-specific factors that may affect pathways.[13] For example, the magnitude of negative pathways may be different in other industries. One measure of this negative pathway is own-price elasticity. For example, [35] find an elasticity of −2.09 among fast food franchisees, and [26] finds −2.82 among auto radiator distribution franchisees. Our price elasticity of −1.33 suggests weaker negative pathways relative to these industries.
The findings also suggest that one would expect stronger positive pathways with newer brands. If an industry mostly consists of long-established brands with a high degree of familiarity, positive pathways may be small or difficult to measure.
Future inquiries could consider these possibilities as well as identify additional moderating factors in which positive pathways might be operative. Historical or other secondary data analyses would assist in documenting the real financial and relational impact of encroachment for same brand outlets in low-brand-density markets. More broadly, such results could be investigated across other franchised service industries.
We acknowledge several considerations in the summation of our results. First, the data limitation on location information of hotels led us to mainly focus on well-defined markets such as downtown areas, airports, and resort towns to isolate encroachment effects. Can we observe the effect of encroachment in other markets? Including suburban and rural metro locations that are likely small, well-defined market yields similar results (Web Appendix W4.5), suggesting that our findings are not limited to downtown, airport, and resort locations. Further information on hotel location data in the excluded markets would allow us to break them into local markets for inclusion in our analyses.
Second, while consumer learning remains a possible explanation for our findings, our static model is unable to fully capture potential dynamics. A dynamic model based on individual choice data and experience would be required to determine that learning has occurred. We leave this as an avenue for future research.
Finally, from a modeling perspective, while our results are robust with more granular fixed effects, our market × time fixed effects may not fully capture all unobserved demand shocks, which can bias results. Individual-level customer panel data would have been useful for determining which demand side explanations—consumer learning, brand awareness, or quality signaling—dominate. Although our study focuses on the presence of same brand hotels, the role of online advertising and consumer search behavior would further shed light on this topic. These topics remain fruitful directions for the future.
Footnotes 1 Hari Sridhar
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 TI Tongil Kim https://orcid.org/0000-0003-3277-4337 Sandy D. Jap https://orcid.org/0000-0003-2661-8549
5 Online supplement:https://doi.org/10.1177/00222429211008136
6 https://legacy.trade.gov/topmarkets/pdf/Franchising_Executive_Summary.pdf.
7 https://www.ahla.com/press-release/hotel-industry-supports-more-1-25-american-jobs#::text=Click%20here%20to%20download%20the,billion%20annually%20to%20U.S.%20GDP
8 The best information we can find is HVS Global Hospitality Services, which biannually publishes data on the average royalty rate information of different hotel brands. We take information from 2007 to 2013.
9 Web Appendix W15 shows that when we include single hotel tracts in the main model, the pattern of results remains consistent. We thank an anonymous reviewer for this suggestion.
It is known that the hotel industry constantly deals with fickle zoning restrictions in New York City (http://observer.com/2013/02/rise-of-the-sliver-hotel-why-blah-buildings-are-blighting-midtown/).
We acknowledge the possibility that concurrent entries may occur in a spurious manner despite these uncertainties, which would then affect our estimates.
Suppose there are three A hotels and two B hotels of the focal chain in a market. When we remove an A hotel, this market is considered a high-brand-density market, as A hotels account for the majority of the hotels in the market. When we remove a B hotel, in contrast, this market is considered a low-brand-density market, as B hotels consisted of less than 50% of the hotels.
We thank the associate editor and an anonymous reviewer for these insights.
We thank the associate editor for this suggestion.
We are grateful to an anonymous reviewer for the suggestions that motivate this section.
References Ackerberg Daniel A. (2001), " Empirically Distinguishing Informative and Prestige Effects of Advertising ," The RAND Journal of Economics , 320 (2), 316 – 33.
Anderson Simon P. , Renault Regis. (1999), " Pricing, Product Diversity, and Search Costs: A Bertrand-Chamberlin-Diamond Model ," The RAND Journal of Economics , 300 (4), 719 – 35.
Antia Kersi D. , Mani Sudha , Wathne Kenneth H.. (2017), " Franchisor–Franchisee Bankruptcy and the Efficacy of Franchisee Governance ," Journal of Marketing Research , 540 (6), 952 – 67.
Antia Kersi D. , Zheng Xu , Frazier Gary L.. (2013), " Conflict Management and Outcomes in Franchise Relationships: The Role of Regulation ," Journal of Marketing Research , 500 (5), 577 – 89.
Avery Jill , Steenburgh Thomas J. , Deighton John , Caravella Mary. (2012), " Adding Bricks to Clicks: Predicting the Patterns of Cross-Channel Elasticities Over Time ," Journal of Marketing , 760 (3), 96 – 111.
Berry Steven T. (1994), " Estimating Discrete-Choice Models of Product Differentiation ," The RAND Journal of Economics , 250 (2), 242 – 62.
Berry Steven , Levinsohn James , Pakes Ariel. (1995), " Automobile Prices in Market Equilibrium ," Econometrica: Journal of the Econometric Society , 63 (4), 841 – 90.
Blair Roger D. , Lafontaine Francine. (2002), Legislating Exclusive Territories: Franchising Encroachment and Legislative Proposals. Ann Arbor, MI : Mimeo, University of Michigan.
Blair Roger D. , Lafontaine Francine. (2005), The Economics of Franchising. Cambridge, UK : Cambridge University Press.
Blevins Jason R. , Khwaja Ahmed , Yang Nathan. (2018), " Firm Expansion, Size Spillovers, and Market Dominance in Retail Chain Dynamics ," Management Science , 640 (9), 4070 – 93.
Chamberlain Gary. (1982), " Multivariate Regression Models for Panel Data ," Journal of Econometrics , 180 (1), 5 – 46.
Ching Andrew T. (2010), " Consumer Learning and Heterogeneity: Dynamics of Demand for Prescription Drugs After Patent Expiration ," International Journal of Industrial Organization , 280 (6), 619 – 38.
Chiou Lesley , Tucker Catherine. (2012), " How Does the Use of Trademarks by Third-Party Sellers Affect Online Search? " Marketing Science , 310 (5), 819 – 37.
Chung Wilbur , Kalnins Arturs. (2001), " Agglomeration Effects and Performance: A Test of the Texas Lodging Industry ," Strategic Management Journal , 220 (10), 969 – 88.
Datta Sumon , Sudhir K.. (2011), "The Agglomeration-Differentiation Tradeoff in Spatial Location Choice," working paper, Yale School of Management , Yale University.
Hollenbeck Brett. (2017), " The Economic Advantages of Chain Organization ," The RAND Journal of Economics , 480 (4), 1103 – 35.
Igami Mitsuru , Yang Nathan. (2016), " Unobserved Heterogeneity in Dynamic Games: Cannibalization and Preemptive Entry of Hamburger Chains in Canada ," Quantitative Economics , 70 (2), 483 – 521.
Ingram Paul , Baum Joel A. C.. (1997), " Chain Affiliation and the Failure of Manhattan Hotels, 1898–1980 ," Administrative Science Quarterly , 420 (1), 68 – 102.
Janakiraman Ramkumar , Sismeiro Catarina , Dutta Shantanu. (2009), " Perception Spillovers Across Competing Brands: A Disaggregate Model of How and When ," Journal of Marketing Research , 460 (4), 467 – 81.
Kalnins Arturs. (2003), " Hamburger Prices and Spatial Econometrics ," Journal of Economics & Management Strategy , 120 (4), 591 – 616.
Kalnins Arturs. (2004), " An Empirical Analysis of Territorial Encroachment Within Franchised and Company-Owned Branded Chains ," Marketing Science , 230 (4), 476 – 89.
Kashyap Vishal , Antia Kersi D. , Frazier Gary L.. (2012), " Contracts, Extracontractual Incentives, and Ex-Post Behavior in Franchise Channel Relationships ," Journal of Marketing Research , 490 (2), 260 – 76.
Kaufmann Patrick J. , Kasturi Rangan V.. (1990), " A Model for Managing System Conflict During Franchise Expansion ," Journal of Retailing , 660 (2), 155.
Kaufmann Patrick J. , Lafontaine Francine. (1994), " Costs of Control: The Source of Economic Rents for McDonald's Franchisees ," The Journal of Law and Economics , 370 (2), 417 – 53.
Kihlstrom Richard E. , Riordan Michael H.. (1984), " Advertising as a Signal ," Journal of Political Economy , 920 (3), 427 – 50.
Kim Tongil T. (2021), " When Franchisee Service Affects Demand: An Application to the Car Radiator Market and Resale Price Maintenance ," Marketing Science , 400 (1), 101 – 21.
Lafontaine Francine. (1992), " Agency Theory and Franchising: Some Empirical Results ," The RAND Journal of Economics , 230 (2), 263 – 83.
Lafontaine Francine , Shaw Kathryn L.. (1999), " The Dynamics of Franchise Contracting: Evidence from Panel Data ," Journal of Political Economy , 1070 (5), 1041 – 80.
Manuszak Mark D. , Moul Charles C.. (2008), " Prices and Endogenous Market Structure in Office Supply Superstores ," The Journal of Industrial Economics , 560 (1), 94 – 112.
Mazzeo Michael J. (2002), " Product Choice and Oligopoly Market Structure ," The RAND Journal of Economics , 330 (2), 221 – 42.
Narasimhan Chakravarthi , John Zhang Z.. (2000), " Market Entry Strategy Under Firm Heterogeneity and Asymmetric Payoffs ", Marketing Science , 190 (4), 313 – 27.
Nevo Aviv. (2000), " A Practitioner's Guide to Estimation of Random-Coefficients Logit Models of Demand ," Journal of Economics & Management Strategy , 90 (4), 513 – 48.
Nishida Mitsukuni. (2014), " Estimating a Model of Strategic Network Choice: The Convenience-Store Industry in Okinawa ," Marketing Science , 340 (1), 20 – 38.
Orhun A. Yesim , Guo Tong. (2018), " Reaching for Gold: Frequent-Flyer Status Incentives and Moral Hazard ," available at SSRN 3289321.
Pancras Joseph , Sriram Srinivasaraghavan , Kumar Vineet. (2012), " Empirical Investigation of Retail Expansion and Cannibalization in a Dynamic Environment ," Management Science , 580 (11), 2001 – 18.
Perloff Jeffrey M. , Salop Steven C.. (1985), " Equilibrium with Product Differentiation ," The Review of Economic Studies , 520 (1), 107 – 20.
Stegemann Nicole. (2006), " Unique Brand Extension Challenges for Luxury Brands ," Journal of Business & Economics Research , 40 (10), https://doi.org/10.19030/jber.v4i10.2704.
Suzuki Junichi. (2013), " Land Use Regulation as a Barrier to Entry: Evidence from the Texas Lodging Industry ," International Economic Review , 540 (2), 495 – 523.
Thomadsen Raphael. (2005), " The Effect of Ownership Structure on Prices in Geographically Differentiated Industries ," The RAND Journal of Economics , 36 (4), 908 – 29.
Tirole Jean. (1988), The Theory of Industrial Organization. Cambridge, MA : MIT Press.
Villas-Boas J. Miguel , Winer Russell S.. (1999), " Endogeneity in Brand Choice Models ," Management Science , 450 (10), 1324 – 38.
Vitorino Maria Ana. (2012), " Empirical Entry Games with Complementarities: An Application to the Shopping Center Industry ," Journal of Marketing Research , 490 (2), 175 – 91.
Wernerfelt Birger. (1988), " Umbrella Branding as a Signal of New Product Quality: An Example of Signalling by Posting a Bond ," The RAND Journal of Economics , 190 (3), 458 – 66.
Wiedmann Klaus-Peter , Hennigs Nadine , Siebels Astrid. (2009), " Value-Based Segmentation of Luxury Consumption Behavior ," Psychology & Marketing , 260 (7), 625 – 51.
Yang Nathan. (2012), " Burger King and McDonald's: Where's the Spillover? " International Journal of the Economics of Business , 190 (2), 255 – 81.
~~~~~~~~
By TI Tongil Kim and Sandy D. Jap
Reported by Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 17- Capturing Marketing Information to Fuel Growth. By: Du, Rex Yuxing; Netzer, Oded; Schweidel, David A.; Mitra, Debanjan. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p163-183. 21p. 1 Chart. DOI: 10.1177/0022242920969198.
- Database:
- Business Source Complete
Capturing Marketing Information to Fuel Growth
Marketing is the functional area primarily responsible for driving the organic growth of a firm. In the age of digital marketing and big data, marketers are inundated with increasingly rich data from an ever-expanding array of sources. Such data may help marketers generate insights about customers and competitors. One fundamental question remains: How can marketers wrestle massive flows of existing and nascent data resources into coherent, effective growth strategies? Against such a backdrop, the Marketing Science Institute has made "capturing information to fuel growth" a top research priority. The authors begin by discussing the streetlight effect—an overreliance on readily available data due to ease of measurement and application—as contributing to the disconnect between marketing data growth and firm growth. They then use the customer equity framework to structure the discussion of six areas where they see substantial undertapped opportunities: incorporating social network and biometric data in customer acquisition, trend and competitive interaction data in customer development, and unstructured and causal data in customer retention. The authors highlight challenges that obstruct firms from realizing such data-driven growth opportunities and how future research may help overcome those challenges.
Keywords: analytics; biometrics; competitive intelligence; field experiments; growth; social network; text analysis; trendspotting
Marketing is the functional area primarily responsible for driving a firm's organic growth. With increasingly rich data from an ever-expanding array of sources, marketers can now capture abundant information to derive more actionable insights on customers and competitors. These insights can fuel firm growth, yet there is also the potential for clutter, confusion, and misuse. Indeed, [99] reports a low level of contribution of marketing analytics to firm performance and no improvement in that contribution over the last eight years. This raises the question of how marketers can wrestle massive data flows into information relevant for effective growth strategies, turning data into a driver of long-term growth. The Marketing Science Institute (MSI) has made tackling this challenge a top priority.
To address the disconnect between marketing data growth and firm growth, we recognize that the use of new data is not, in and of itself, a growth strategy. Rather, a firm's marketing data and application mix must align with its growth strategy and, in doing so, provide the firm with a powerful strategic capability ([22]). However, evidence suggests that firms' data and application mix may not always align with resource allocations across growth strategies. The "streetlight effect" is one threat to this alignment, which stems from managers' overreliance on readily available data due to ease of measurement and application, irrespective of their growth objective. The streetlight effect manifests in managers leveraging "big data" for small problems while making do with small data for big problems or, even worse, neglecting data altogether. For example, the streetlight effect caused by the abundance of advertising data leads managers to focus more on managing advertising than on distribution or product line, despite advertising elasticities being much lower than the elasticities of the latter two ([43]). Likewise, the streetlight effect of abundant near-market knowledge leads firms to choose proximity over favorable growth opportunities in distant international markets ([75]).
To guard against the streetlight effect, marketers must explicate how a firm's growth strategy can be supported by its portfolio of marketing data and applications. The objective is to prioritize data and applications to maximize the value of the decisions they support.
Relying on the premise that the "valid definition of business purpose (is) to create a customer" ([27], p. 37), we suggest that the pertinent value must relate to customers. [ 9] describe growth in terms of maximizing the value of the customer base, which arises from three drivers of customer equity: ( 1) acquiring new customers, ( 2) developing existing customer relationships, and ( 3) retaining customers ([40]). Accordingly, we frame the quest for marketing data to fuel firm growth along these three customer equity dimensions.
The remainder of this article proceeds as follows. We first look at historical examples of the streetlight effect as it pertains to marketing data. We then structure our discussion of opportunities for marketing data–driven growth around the three customer equity components (customer acquisition, customer development and customer retention), highlighting how undertapped data sources offer growth opportunities. For each component, we discuss the challenges that can obstruct such opportunities and identify directions for future research. The opportunities that we discuss—the use of biometric and social network data in customer acquisition, trend and competitive interaction data in customer development, and unstructured and causal data in customer retention—are by no means comprehensive, but we hope they can spark future work that leverages data with growth objectives in mind.
Data innovations are often thrust on marketers, who then scramble to leverage these data for different purposes using a variety of applications that unleash a firehose of information but may not always contribute to the firm's growth in the long run. This is one important contributor of the streetlight effect. In this subsection, we review several prominent examples of past marketing data innovations that demonstrate the streetlight effect and the resultant blind spots.
The advent of retail scanner and scanner panel data was a major breakthrough in the 1980s and 1990s (e.g., [39]). However, as [65] caution, the increasing prevalence of real-time transaction data has made firms more myopic in their search for growth, overinvesting in promotions because the short-term effects are readily quantifiable and underinvesting in brand equity, new products, and distribution because their long-term effects are difficult to assess.
With the spread of loyalty programs and advances in database technology, the advent of all-encompassing customer relationship management (CRM) data has enabled marketers to track the entire history of interactions with their own customers. However, most CRM data contains little information on the interactions that customers have with competitors or how customers' needs and wants evolve over time ([31]). Such an imbalance between data on internal versus external relationships can lead to growth strategies that are overly inward- and backward-looking.
Clickstream data that track customers across digital channels have afforded marketers a near 360-degree view of a customer's journey online. This has led to tools that support growth by improving marketing return on investment (ROI) through better target selection and media planning, including multitouch attribution models. However, touchpoints are not equally trackable across channels. Discrepancies in measurability between digital and mass media (e.g., TV, radio, print, outdoor) may have contributed to shifts of advertising budgets toward digital due to ( 1) larger measurement errors in mass media and ( 2) the difficulty in quantifying the generative influences of mass media and cross-media synergies.
Recognizing the challenges inherent in drawing causal inferences from observational data, marketers increasingly turn to field experiments. Online testing has been the dominant driver behind the rise of field experimentation. Yet experimentation remains relatively rare when it comes to offline behaviors whose short-term responses are difficult to measure (e.g., [32]) or marketing levers that are expected to have long-term effects (e.g., proactive retention efforts). Field experiments may also not capture the full range of marketing decisions, whether it be promotion amounts or ad copies, and their results may not be reliable over time. A relatively small set of domains where field experiments are prevalent, combined with increasing faith in them, could lead marketers to focus disproportionately on activities that are easy to manipulate and are more effective at driving measurable short-term responses, potentially at the expense of long-term growth.
Social media data and other forms of UGC have revolutionized the way marketers listen to their customers, dwarfing the data available through traditional tools such as surveys and focus groups. Popular UGC platforms use such data to target advertising. However, the size, timeliness, and richness of UGC does not guarantee that these data are representative. Relying on social listening could bias perceptions of the marketplace due to differences in users across platforms ([94]; [92]).
Many firms have jumped on the big data–machine learning bandwagon as open-source algorithmic methods become readily implementable, without fully grasping the relative advantages of traditional methods taken by marketing science or recognizing the potential for "algorithmic biases" and other unintended consequences (e.g., [55]). In pursuing predictive ability, big data applications risk sacrificing the interpretability of the results. Consequently, marketers could inadvertently create social ills in their pursuit of growth that can harm society and, eventually, the firm.
A common thread across these examples is that while new data sources can yield insights that lead to growth, they come with trade-offs. As marketers focus on particular data sources, our field of vision narrows, resulting in potential blind spots that can weaken a firm's growth trajectory, including the following.
First, marketing data may result in prioritizing short-term growth ahead of long-term growth. While some marketing efforts yield short-term responses, others must be assessed over a longer time frame. Because data for measuring short-term effects (e.g., click-through rates) are easier to obtain, data-driven growth may favor investments in lower-funnel actions (e.g., price promotions) at the expense of longer-term, upper-funnel actions (e.g., brand building).
Second, marketers may overly rely on historical, internal data at the expense of forward-looking, external growth opportunities. Firms have rich data about interactions with their customers. Growth strategies built on such data may favor customers with limited expansion potential while those whose relationships could be substantially enhanced are neglected.
Third, marketing data may create a preference for more easily measured digital touchpoints at the expense of offline channels. Despite the continued growth of e-commerce—especially during the COVID-19 pandemic—it accounts for only 14.5% of retail sales ([33]). Given the prevalence of offline activities, efforts to leverage marketing data must balance online and offline sources.
Finally, marketers may rely on available data in lieu of representative or predictive data. This is exemplified by UGC, where research has cautioned about a vocal minority that may not represent the "silent majority" (e.g., [76]). Another example of this blind spot arises from assuming, often incorrectly, that each data point is equally informative. [96], [97]) identify features in commonly encountered contexts where specific data points are more informative than others. These features result in the superiority of measures harnessed from selected data points (e.g., maximum, top rank displacement) as predictors of growth versus those that seek to capture information from all data points (e.g., average, variance).
In the next three sections, we outline specific undertapped data and research opportunities for each of the three drivers of customer equity. Tackling customer acquisition, development, and retention in turn, we discuss how the streetlight effect may have led to these opportunities being overlooked and how marketers may leverage them to fuel growth.
Customer acquisition is critical to firm growth, as its success is necessary before a firm can focus on developing and lengthening the relationship. One way to achieve growth via acquisition is to use marketing data to increase the efficiency of acquisition efforts, such as by identifying prospects who are more likely to convert. While internal data on current customers may be rich, such data may contain limited information about prospects. Though customer profile data are readily available, there are questions as to their accuracy ([81]) and ability to capture prospects' state of mind at a given moment. Focusing on such available and limited data may lead marketers to overlook richer data sources. Next, we discuss two such sources that can provide timelier and often external insights into prospects that can improve acquisition efforts: biometric and social network data.
In recent years, we have seen an explosion in biometric data being collected. Biometric data are physiological measures and calculations collected from individuals. Genetic testing kits from providers such as 23andMe offer insights into people's health, including screening for the BRCA1/BRCA2 mutations that indicate an increased risk of breast and ovarian cancer ([102]). Beyond providing health insights, medical research is one use of genetic data ([83]). For business purposes, biometric data is being used to evaluate marketing creatives (e.g., [103]), enabling marketing research firms to collect data on how individuals respond to advertising and identify creatives that are most likely to resonate with the target audience.
Marketers have examined an array of biometric data sources, including eye tracking (e.g., [89]), electroencephalogram (e.g., [90]), functional magnetic resonance imaging (e.g., [107]), and emotion detection (e.g., [64]). A compelling aspect of biometric data is its real-time nature. Smartwatches and activity trackers monitor heart rate and blood pressure at a given moment. Such wearable devices also offer a means by which individuals can be motivated. For example, by incentivizing drivers to be monitored, insurance providers may find that delivering biometric information through wearables encourages the adoption of healthy habits. Digital music providers and advertisers can coordinate the audio with a user's level of activity at the time.
While physiological responses to stimuli can offer insights into consumers in a controlled environment, there have been limited efforts to apply these insights for marketing in the field. The availability of consumer-facing products that collect biometric data and tools that can collect such data from large groups can enable marketers to incorporate such information to engage consumers in real time. Biometric data also offer opportunities for healthcare and technology providers to develop new services and acquire new customers ([19]). Collecting biometric responses to marketing content can enhance content personalization and product recommendation. Biometric data can also aid in contextual targeting to inform when messages should be delivered to support prospect acquisition.
Biometric-based marketing research is being delivered by a range of providers, including Nielsen, Ipsos, and Kantar. In studies conducted by such providers, participants are often exposed to stimuli and their physiological responses are monitored. Though such studies are powerful additions to the marketing researcher's toolkit, the technology lacks portability. While functional magnetic resonance imaging and electroencephalogram studies can monitor how neurological activity is affected by external stimuli, such studies are often expensive and must occur in a laboratory environment ([103]).
One of the key challenges in the use of biometric data to support customer acquisition is the ability to collect such data both at scale and in the field. Connected devices such as smartwatches and activity trackers may facilitate field data collection, enabling the capture of heart rate, blood pressure, and the rate of oxygen consumption. Some techniques, such as eye tracking and emotion detection, can already be deployed in the field. Once the data collection challenge has been addressed, marketers can investigate the relationship between biometric measures and acquisition to better understand the stimuli and context in which different biometric measures arise. Doing so can then support prospect identification and content development to increase the efficiency of acquisition efforts.
Marketers must be cautious when harvesting personal information from individuals. Biometric data may be used for unintended purposes, such as facial recognition using data scraped from social media posts or retailers using video surveillance data to identify engaged prospects without obtaining their explicit consent ([91]). Not only are consumers likely to vary in their privacy preferences, they may also react differently to data protection measures such as anonymization or aggregation.
The opportunities for biometric data to support growth require that metrics be collectable in the field. Wearables provide a convenient means by which such data can be collected. Smartwatches can collect measures such as heart rate and oxygen consumption levels. The use of such data by marketers depends on its being made available by manufacturers. In discussing growth opportunities using biometric data, we focus on eye tracking (e.g., [89]) and emotion detection (e.g., [64]), which can be implemented at scale. Methods are being developed and improved to detect emotions from individuals in a crowd (e.g., [35]). However, we have yet to see these tools being applied to large crowds. Challenges remain in using eye tracking in the wild due to each individual's "idiosyncratic field of view at each point in time during the recording" ([14], p. 38).
Retailers, hospitality businesses, and entertainment venues could make use of biometric data to target and optimize marketing messages. Exterior signage equipped with video cameras would enable marketers to identify the characteristics of messages that are resonating with passersby based on the emotions they express and if the content is grabbing their attention. Identifying the content that both attracts attention and elicits a positive emotional response may better engage prospects and increase conversion. Another direction for research is the optimization of marketing content delivery timing. Using eye tracking and facial detection systems, future research could examine how to optimize the schedule with which content is shown to large groups to increase acquisition. While this would ideally involve linking an individual who is exposed to multiple marketing messages to their subsequent behavior, which could be achieved through facial recognition, researchers should work toward privacy-friendly methods to connect media exposures and subsequent activities.
Because one of the advantages of biometric data is its real-time availability, research could use such data to examine how consumers physiologically react to marketing throughout the customer journey. At different stages, certain biometric measures may be more informative of the likelihood of progressing to the next stage. In particular, how do consumers react to marketing in the steps leading up to the acquisition decision? In the prepurchase stage that precedes acquisition, consumers engage in need recognition, consideration, and search ([59]). Eye tracking may inform consumers' consideration sets and search processes based on the products viewed and the features that attract attention. The emotions expressed when different products are viewed can inform subsequent product recommendations. Research should also consider the intensity of emotions that consumers express. Real-time data collection can also enable analysts to examine how variations in biometric measures (e.g., difference, velocity, acceleration) relate to acquisition behavior, which can enable marketers to target consumers at the times when they would be most receptive to acquisition efforts. Emotional trajectories likely vary across categories, perhaps with consumers exhibiting more emotional variation in response to experiential and hedonic products while they may not express as much emotional variation in their search for and consideration of utilitarian products. This may suggest that eye tracking measures and emotional measures have varying degrees of informativeness on the likelihood of acquisition in different categories.
Biometric data can also create new opportunities to understand the drivers of acquisition by capturing physiological measures during media consumption activities (or interactions with salespeople in a business-to-business [B2B] context). An individual's biometric responses can be gathered while they are exposed to videos or music on mobile devices, whether it is through a smartwatch or earbuds that collect measures such as heart rate. Such biometric measures will fluctuate based on the media being consumed, including the focal content and embedded marketing. [100] report that media consumers who generate content related to the consumption experience report increased immersion and engagement. The extent of consumers' immersion may be measurable in real time using biometric data. Moreover, researchers can use biometric data to identify when consumers will be most open to marketing efforts and have an increased tendency for conversion, such as high arousal and positive emotion. These measures may be informative of purchase intent, offering marketers a time window during which to deliver incentives.
In addition to the effects of embedded marketing messages, researchers using biometric data may consider the sequence of emotional reactions that are produced along the path to purchase, and whether these reactions are affected by marketing messages or the content in which they are embedded. If consumers are immersed in the content (e.g., [47]), outside factors such as marketing messages may disrupt the consumption experience and elicit negative reactions that adversely affect the brand. By collecting biometric response throughout media consumption, such misfires can potentially be avoided.
While methods for customer ([61]; [93]) and corporate (e.g., [40]; [71]) valuation that account for customer acquisition have been developed, they are not without shortcomings. Such models often are constructed on the assumption that customers act independently. While this lowers data requirements and offers computational benefits, this assumption may also limit the accuracy of the valuation model.
Word-of-mouth (WOM) activity plays a critical role in customer acquisition. [101] report that WOM activity is often more impactful than traditional marketing actions in acquiring users. [104] find that customers acquired through WOM deliver more long-term value than those acquired through traditional marketing. Others have demonstrated similar findings regarding offline WOM activities (e.g., [ 6]). The source of WOM has also been identified as an important consideration (e.g., [48]).
One way in which firms can observe social ties is through customer referrals (e.g., [54]). However, referrals reveal only a portion of the social ties. Secondary data sources can provide a more complete picture. While social interactions are ubiquitous, social ties and the information flow across these ties are difficult to observe, which may have limited their use (in comparison to referrals) for acquisition. Social ties data are often collected by third-party providers, further contributing to the streetlight effect that leads marketers to overlook such data. Another contributing factor to marketers ignoring this potentially fruitful data source is that, as we discuss subsequently, it can be difficult to directly interpret, leading to another manifestation of the streetlight effect.
Despite the challenges associated with using social network data, it offers a significant opportunity to drive growth via acquisition. In prioritizing prospects for acquisition efforts, focusing on a prospect's value based solely on their own expenditures neglects the value they provide through their impact on other prospects (e.g., [46]; [53]). Incorporating such data into acquisition efforts can help align the firm's data efforts with its growth objectives by leveraging interactions among members of the customer base.
To incorporate social network structure into customer analytics and inform subsequent acquisition decisions, there are some fundamental challenges that must be addressed. We discuss two specific challenges as they pertain to customer acquisition.
Platforms for UGC have proven to be a rich data source, but limited research has documented differences across platforms ([92]; [94]). As users' motives for engaging with each other likely differ across platforms, inferring social ties from one venue will not capture all meaningful connections between users. While LinkedIn connections may arise for professional reasons, those on Instagram may reflect personal or familial ties. The relevance of social ties must also be identified. A specialized forum for physicians or health care professionals could reveal interactions that are most important and the most influential nodes with regard to procedure adoption, but social ties on consumer-oriented venues such as Instagram or Pinterest may be useful for identifying prospects with similar socioeconomic status.
The streetlight effect has led to a preponderant focus on online (rather than offline) social ties and WOM. However, with most consumers still making purchases and interactions offline, offline social ties may be particularly relevant in certain industries. Though offline social networks can be inferred through sources such as mobile location data (e.g., [109]), an overreliance on online social interaction data that are more readily available may bias ROI estimates of social network–based acquisition efforts ([17]).
Another challenge is the degree to which network ties and their strength are stable. The longevity of a connection and the frequency of interactions may inform the strength of the connection and the extent to which two individuals' decisions are correlated. Researchers often fall victim to the streetlight effect when they merely look at the existence of a social tie as opposed to the strength of the tie. It is not enough to observe the presence of ties, as data capturing the degree and direction of information flow along these ties is needed to infer the relevance of the relationship.
A related challenge to using social network data in customer acquisition is the shelf life of the data. While consumers add online connections, they rarely take steps to prune them. If individuals do not actively sever social ties, we observe a situation akin to latent attrition, with no activity occurring along a social tie for an extended period of time. Acquisition efforts require up-to-date information on the social ties and their strength that may influence prospects' choices, requiring that marketers distinguish between active and dormant social ties.
There also remain technical challenges associated with capturing and analyzing dynamic social network data. While dyad-level data can characterize the relationship between consumers, the data grow quadratically with the size of the network ([12]). Adding a temporal dimension to the interactions results in rapid growth of the data, which may require marketers to predetermine the prospects that warrant their attention in their network analysis.
Having identified the challenges pertaining to missing data in the construction of the social network and the dynamic nature of social ties, we now discuss specific opportunities that we see to leverage social network structure to drive growth via better targeted acquisition efforts.
[46] discuss the influencer value of a customer, recognizing that a customer's social influence may induce the acquisition of others. [41] recognize the impact of direct network effects in customer value, as the presence of nonpaying customers may help attract paying customers. In the context of a two-sided platform, [106] estimate the forgone value associated with an individual. Future research can reevaluate the way in which we consider the value that a customer delivers to an organization. Recognizing the knowledge value that individuals provide (e.g., [53]), research can view interactions between prospects as opportunities for knowledge to be shared, which may either be retransmitted to other prospects or directly affect a prospect's decision. The volume and content of such interactions allow the experiences of a single prospect to affect others, thereby affecting perceptions and future expectations of customer acquisition.
In the case of acquisition, one may consider the forgone value of an individual prospect. Not only are the expenditures by the prospect lost, but so is the value from others that the focal prospect may have affected. The value of some individuals, depending on their position in the network, may arise primarily through their impact on others. Though social effects have been probed in the context of customer retention (e.g., [ 3]), to the best of our knowledge, there have been no efforts to consider it with regard to acquisition.
A related opportunity is to derive the forgone value associated with combinations of (possibly socially connected) customers. Deriving the value of customer combinations upends traditional prospect scoring. Among the research opportunities that this creates is the development of techniques to evaluate the impact of different resource allocations across prospects. We expect that individuals with strong social ties have the ability to influence others' acquisition decisions. We also expect a stronger impact on customer acquisition in contexts where network density is high, as messages may propagate further. Developing a joint model for information flow through a prospect network and acquisition decisions could help evaluate counterfactuals. For instance, should the firm reduce its efforts on tightly connected prospects? This may be feasible if the firm can leverage social interdependence to achieve the same rate of acquisition using fewer resources.
Given the number of combinations of prospects, it may not be feasible to derive the forgone value of a unique prospect. One alternative may be to characterize prospects using internal classifications aligned with operations (e.g., marketing territories) to estimate the forgone value associated with one (or more) prospects of a certain type. Identifying actionable proxies for the social network structure and incorporating them into resource allocation models is another area in which future research is warranted.
Another opportunity lies in comparing the value of various external social interaction data sources for identifying high potential prospects. [81] report considerable variation in the veracity of third-party profiles. Marketers' reliance on such data sources to identify prospects could be improved if they can be merged with data on the social interactions of their current customers. Prospects' positions in the social network relative to current customers could inform which prospects warrant attention. By using social network data to screen prospects, marketers may increase conversion, boosting the ROI of their efforts and driving growth. One of the reasons that social network data can supplement third-party profiles is the ability to focus on activity within a particular time frame. As social interaction data include a timestamp, marketers may focus on recent interactions. If consumers typically conduct product research for a few weeks prior to making a purchase in a category, marketers may focus on the social connections with which an individual has interacted during this timeframe.
The benefits of using social network data must be weighed against privacy concerns ([20]). While consumers may accept the use of third-party profiles in gauging their purchase intent, they may react adversely to firms compiling their social ties. To avoid consumers' backlash, marketers must exercise caution in sourcing social interaction data, as well as in the extent to which marketing communications reveal the rationale by which prospects have been selected for targeting (e.g., [45]). Network typology as reflected in node-level summary statistics may be an alternative to the use of detailed social ties that can mitigate consumers' privacy concerns. Other metrics that may be informative of a prospect's likelihood of becoming a customer include the fraction of social ties who are current customers or the average expenditures of an individual's social connections. While the former captures the extent to which the social ties have a relationship with the firm, the latter may better reflect the monetary value of these social ties. Research can identify privacy-friendly measures that retain the predictive value of the social network with respect to acquisition tendencies.
Another opportunity to use social network data for customer acquisition is social selling, a lead generation method whereby members of a sales force reach out to prospects on social media platforms. Social selling can be particularly effective in the context of B2B marketing ([73]), enabling marketers to expand their audience. Accounting for the network structure among the individuals with which a firm engages is essential as it can affect the frequency with which prospects are exposed to content from the firm. Ignoring social connections may result in marketers saturating prospects with messages due to repeated exposure from social ties. In addition, in contrast to many consumer decisions, B2B purchases involve many individuals making a joint decision. The strength and density of the social connections among the individuals who comprise the decision-making team may inform the relationship between decision makers and the likely interactions among them.
Accounting for the network structure also has implications for sales force compensation. Salespeople should optimize their efforts across prospects, accounting for both expected sales to a prospect and their potential impact on others. The design of compensation plans should also minimize the potential for free riding off of the social effect generated by other salespeople's efforts. Developing compensation schemes that accommodate social interdependence in the presence of multiple salespeople is a promising direction for future research, which can better align the sales force and the company incentives, minimize undesirable behavior and serve as an opportunity to drive growth via acquisition in B2B settings.
Firms can grow by expanding the relationship with their existing customers. Data-driven approaches to customer development often rely on internal data on customer–firm interactions to identify targeting and personalization opportunities. Such data are readily available and have been widely used to build applications such as recommender systems for cross- and upselling. However, an overreliance on internal customer–firm interaction data for customer development can make a firm vulnerable to the streetlight effect, risking backward- and inward-looking myopia. Backward-looking myopia arises as customer needs and wants change due to shifting trends, which make past purchases less predictive of future preferences. Inward-looking myopia stems from customers interacting with multiple firms, with internal data providing only a partial view of a customer's category demand. In this section, we focus on two undertapped data sources—trend and competitive interaction—that can help mitigate these risks, leading to new growth opportunities through more forward- and outward-looking customer development efforts.
Changing customer needs and wants can manifest in shifting preferences for existing product features ([28]) or emerging demands for new features that can reshape well-established product category boundaries and cost–benefit trade-offs in customers' purchase decisions ([84]). Spotting trends in the evolution of customer mindsets and behaviors before the competition and adjusting the marketing strategies of existing offerings or developing new offerings are key opportunities for growth through customer development. The data that inform trends are often external, mitigating the risk of inward-looking myopia. In addition, trend data (by definition) focus on longer-term growth as opposed to short-term gains. Despite the importance of trendspotting to inform growth opportunities, marketing scholars have provided little guidance on how to systematically gather and analyze data to generate such foresights ([30]). Several challenges remain.
The biggest challenge in trendspotting is that, without the benefit of hindsight, it is difficult to distinguish short-lived fads from emergent trends that offer meaningful growth opportunities. Consequently, it is risky to implement a proactive growth strategy contingent on identifying emerging trends in customer needs.
Take gluten-free and aspartame-free as two contrasting examples. Gluten-free foods have grown from a niche category into a multibillion dollar business where avoiding gluten is now a lifestyle choice ([74]), despite the lack of scientific evidence for its benefits ([60]). Gluten-free products have become a source of sustained growth for many consumer packaged goods manufacturers ([85]). By contrast, PepsiCo's recent decision to remove aspartame (an artificial sweetener) from its flagship Diet Pepsi turned out to be a debacle akin to the 1985 "New Coke" fiasco, having to be pulled and replaced with the original after three years of declining sales ([52]). The root cause of PepsiCo's blunder is that it mistook a fad (health concerns about aspartame) for a trend, despite the due diligence and market research accompanying such a high-stake decision ([34]).
Another challenge with identifying relevant trends is that they often transcend industries, technologies, and products, representing major shifts in consumer mindsets and behaviors that arise from social, economic, environmental, political, and technological changes ([84]). Identifying such transcendental trends requires surveying and integrating a multitude of sources across domains (e.g., expert interviews, trade publications, industry reports), which can often diverge from or even contradict with one another. Such cross-domain, cross-source synthesis presents a major methodological hurdle.
A practical trendspotting challenge is the collection of archival data to quantify the magnitude and momentum of trends and fads at any given moment in time, including the pretakeoff period. Because historical sales data could be difficult to obtain for a large number of trends and fads, an alternative data source could be Google Trends (trends.google.com/trends/), which can be used to gauge Google users' interest in virtually any topic ([30]; [28]). Keyword search volume by market can be obtained dating back to 2004, which has been shown to be correlated with sales ([30]).[ 4] Researchers can also gather historical data from social listening platforms (e.g., Brandwatch, Meltwater) to quantify shifts in consumer interests based on the volume and content of online posts, which have been shown to be informative of consumers' mindset and behavior ([94]; [108]).
Beyond natively digital sources such as search and social media, other sources of text can prove useful for trendspotting. Researchers can take advantage of digitized texts that have appeared in books, periodicals, patents, and other publications (e.g., [105]). For example, researchers can now easily access a corpus of digitized texts containing 4% of all books ever printed (https://books.google.com/ngrams). The data set spans over 200 years (1800 through 2019) and comprises yearly data on occurrence frequency of two billion one- through five-n-grams.[ 5][72] illustrate how such data enable quantitative investigations of cultural trends, generating insights about fields as diverse as technology adoption, lexicography, collective memory, the pursuit of fame, censorship, and epidemiology.
Beyond compiling the necessary data sources for trendspotting, new econometric or machine learning methods for conducting large-scale trend analysis are needed to better separate emergent trends from fads as early as possible and to predict long-term trend trajectories. To that end, methodological advances can be made in several fronts.
For each historical trend or fad, there can be many relevant keywords and topics manifesting across many sources (e.g., online searches, posts in different social media platforms, mentions in different news media). One challenge to integrating different data sources to derive a meaningful index at any period in time lies in their varying periodicities. While some measures can be collected on a daily or weekly basis, others may only be available at a monthly (e.g., trade publications), quarterly (e.g., financial reports), or yearly (e.g., letters to shareholders) level. Research is needed to synthesize trend data with disparate reporting frequencies to isolate the common underlying signal (e.g., [ 4]) while accounting for other intrinsic differences that may exist across sources (e.g., search volumes may be a stock variable of consumer interest while social media mentions may be a flow variable, UGC may capture shifts in demands while company press releases may capture shifts in the way that managers view the marketplace).
Another methodological opportunity for researchers is the development of methods to integrate both structured and unstructured data in deriving trend indexes. Structured data such as the volume of keyword searches or topic mentions in textual social media posts can be combined with unstructured data such as nontextual social media posts to allow researchers to go beyond textual data and tap into visual, audio, and video data to spot trends that can drive growth through customer development ([ 7]).
Projecting the trajectory of a trend or fad is a fundamental challenge in time series modeling, especially during the pretakeoff period when data are limited and accurate long-range forecasts are highly valuable ([16]). One way to solve this "cold start" problem is to build a large training sample of historical trends and fads that cover a wide range of industries, pairing it with pattern recognition methods to identify the most similar historical counterparts to help make forecasts for the emergent trend or fad ([44]). Unfortunately, no such a training sample exists, the construction of which will require painstaking efforts. To do so, one may follow the principles that have been successfully applied to discover empirical regularities in the diffusion of technological innovations and new products (e.g., [37]).
Another direction that researchers may consider exploring is the identification of generalizable factors that can predict the trajectory of a trend or fad. In their application to customer base analysis, [25] demonstrate how dynamic behavior can be decomposed into underlying components including calendar effects and individual-specific factors that affect transaction behavior. Researchers could consider a similar decomposition to the aforementioned data sources for better distinguishing trends from fads. In addition to the time at which the data are generated, researchers could also consider geographic variation based on the origin of the data. Deep learning methods such as long short-term memory networks may offer a useful tool, provided that an adequate training set is developed.
The timing with which a trend or fad unfolds may vary across markets or segments, some of which could be "harbingers" that send early warning signals about what is to come ([ 1]). Identifying these trendsetting markets or segments could help spotting the rise or fall of a trend or fad. For example, in which product categories do coastal or urban markets tend to lead inland or suburban markets, or vice versa? One promising data source for addressing such an empirical question about shopping trends would be Google Shopping Insights (https://shopping.thinkwithgoogle.com/), which provides daily shopping search data for 55,000+ products, 45,000+ brands, and nearly 5,000 categories, across all 210 designated market areas in the United States.
A related direction for future research is to investigate the nature of contagion among markets or segments. For instance, trends may spread via both online and offline WOM, with the latter relying more on geographic proximity ([ 6]). A geotemporal diffusion model that distinguishes between online (global) and offline (local) contagion may help identify how quickly a trend spreads both across and within markets, providing guidance to marketers on where and when their resources may best be deployed to increase customer expenditures.
Finally, thinking about the social ties discussed in the acquisition section, understanding the social ties in which a trend has emerged may also help assess the likelihood of its success ([26]). One way in which researchers may explore this opportunity would be to examine whether an increased focus on marketing to trendsetters among existing customers will increase expenditures by other customers with whom they have strong social ties.
Firms face a fundamental information asymmetry at the customer level. While they observe their interactions with customers, they know little to nothing about a customer's interactions with competitors, resulting in an overreliance on readily available internal data. Obtaining customer-level competitive intelligence can enrich a firm's knowledge of both what a customer purchases from competitors and the offers the customer receives from competitors, which is essential to increasing the share of their category expenditures with the firm.
Using internal data alone, a firm may misjudge a customer's growth potential because it cannot distinguish a customer with a small wallet of which it gets a large share from one with a large wallet of which it gets a small share. The latter may have more growth potential if the customer were correctly identified and targeted. Researchers have proposed methods for estimating size and share of customer wallet that comprise two main approaches: ( 1) augmenting internal transaction and customer characteristics data with external transaction data for a sample of customers (e.g., [31]; [50]) using customer surveys or purchase panels run by syndicated data providers or third-party data aggregators, or ( 2) estimating size and share of customer wallet using only internal data (e.g., [18]). While the first approach may incur significant data acquisition costs, the latter approach requires imposing strong assumptions about the data generating processes and can only be validated indirectly due to the lack of external data.
Firms also struggle to assess the effectiveness of relationship expansion efforts because they cannot observe the offers or counteroffers that customers receive from competitors. In the absence of such customer-level competitive intelligence, it is difficult to tailor offers that account for the competitive context for each customer. Few statistical and econometric methods are available for addressing the challenge of missing data on customer-level competitive activities (for an exception, see [77]]).
Conceptually, the pitfalls of ignoring competitive intelligence in managing customer relationships has long been recognized (e.g., [11]). However, overcoming inward-looking myopia remains difficult for one primary hurdle: data on customers' activities with competitors have been difficult to obtain. The schism in availability between internal and external data, we contend, may have grown over time as firms increasingly track interactions with their own customers, increasing the detail and volume of data available to them, while customer-level competitive intelligence remains elusive due to increasingly personalized marketing efforts, media and channel fragmentation, and privacy concerns.
The challenges described previously present growth opportunities for firms that can tackle the internal versus external information asymmetry to yield a competitive advantage in customer development, and promising directions for researchers to develop methodologies for the collection and analysis of customer-level competitive intelligence data.
While surveys are a common source for size and share-of-wallet data, self-reports can be time-consuming, costly, and error-prone. [51] find that the rank ordering in perceptions and attitudes among brands used by a consumer tend to be highly predictive of a brand's share of wallet. Researchers may identify similar survey measures (e.g., consideration set, purchase intent, willingness to recommend) that can strike a balance between burden on the respondent and predictive ability. Once such measures prove cost-effective and reliable, they could be integrated into brand trackers. Beyond survey data, for a sample of customers firms may acquire size and share of wallet data from vendors that run purchase panels (e.g., comScore Networks, GfK, IRI, Kantar, Nielsen, NPD) or third-party aggregators that track customer transactions across competing vendors in a category (e.g., Acxiom, credit bureaus, IMS, IXI, Rakuten Intelligence).
Researchers may also try to integrate newer data sources that can be used to infer size and share of wallet. For example, just as social media mentions have been used to map brands' competitive positions (e.g., [79]), researchers may explore the use of such data to infer brands' share of wallet at the customer level. Paired with customer transaction data, research may assess if the frequency and concentration of competitor mentions is informative of customers' size and share of wallet. Researchers may also explore the potential for clickstream data (via cookies installed by the firm's own website or through third-party aggregators) and mobile location data (e.g., Mogean, PlaceIQ). Specifically, to what extent do share of website and offline store visits relate to share of wallet? Collaborations with credit card panels such as Second Measure could provide a means of assessing such a relationship empirically. Wearables may also serve as tools to assess a customer's overall usage in some categories, or even the use of a specific competitor's products. For example, distance traveled recorded by a fitness tracker could be used to gauge demand for running shoes. Should a relationship between these newer data sources and share/size of wallet be borne out, such data could provide means for estimating the extent to which customers engage with competitors, mitigating the asymmetry caused by differences in the availability of internal and external transaction data.
Customer valuation and cross- and upselling models rely almost exclusively on internal data ([80]; [93]), confounding the processes governing the evolution of wallet size and wallet share. When the former remains stable, existing models would be well suited as changes in internal transactions are driven mainly by changes in share of wallet. However, in situations where consumers' needs and preferences evolve over time (e.g., [29]), ignoring the distinct dynamics between wallet size and share could lead to erroneous assessments of growth potential.
Developing models that disentangle the evolution of wallet size and share remains an opportunity for future research. Changes to the wallet size are driven mainly by customers' purchasing power and needs, whereas shifts in the share of wallet are indicative of relationship strength. Own and competitive marketing efforts can affect both the size and share of wallet. For example, product development that incorporates new features by a single brand may spill over to other brands in a category by increasing the perceived category benefit. Exogenous events such as a global pandemic or brand crisis may affect both category demand and wallet share, depending on the strength of an individual's relationship with brands in the category. This could result in a form of double jeopardy, with shocks reducing both total category demand and the share of wallet of weaker brands. Modeling the joint evolution of these processes enables forecasting future expenditures for a given firm and for the entire category. Such efforts are nontrivial, as they must account for strategic behavior of the firm and its competitors.
Comparing predictive performance across models that ( 1) use only internal data and ignore the distinct dynamics between wallet size and wallet share; ( 2) use only internal data but account for the distinct dynamics between wallet size and wallet share; and ( 3) augment internal data with external data and account for the distinct dynamics between wallet size and wallet share could reveal the incremental value of each component, which could vary depending on the empirical context. For example, in categories where customer wallet sizes tend to vary substantially but predictably over time, it might be most important to model the dynamics of wallet size and wallet share separately, with or without external data. Purchase panel data that covers an extended time window (say, between five and ten years) and breaks out transactions by competing vendors would offer a promising testing ground (e.g., IXI, Kantar's Worldpanel).
Marketers have long recognized the importance of cross-effects in aggregate market response models. However, the effects of competitors' marketing efforts have been generally ignored in customer analytic models. While survey data could be collected, it may be difficult for customers to recall specific competitive offers and touchpoints. It might be feasible to ask customers for perceived relative levels of marketing efforts across competitors. Though imprecise, such information can be used to impute the general level of competitive efforts targeted toward a customer, which could highlight those customers who are potentially being targeted by competitors and whose share of wallet should be monitored for signs of a weakening relationship. In addition, it is possible for firms to monitor some of the online and direct marketing efforts by their competitors (e.g., Competiscan, Comperemedia) and incorporate these into their customer analytic models.
Competitive marketing efforts (e.g., which customer receives how much, and from whom?) may reflect what competitors know about customers that the firm does not. If this is the case, capturing competitive marketing efforts and incorporating them into a firm's customer analytic models may identify customers with more "winnable" growth potential. However, given the complexity of competitive efforts (e.g., high-dimensional, customer-specific, time-sensitive), care must be taken to account for the strategic nature of the marketing efforts by the firm and its competitors, and the possibility of signaling and counter response by competitors ([95]). When customer-level competitive intelligence is unavailable, data on aggregated competitive activities, such as those used for market response modeling, could offer a means to augment data for customer response modeling (e.g., modeling competitive efforts received by a customer as a function of aggregated competitive activities that vary across time and markets).
In the presence of competition, information asymmetries across firms with regard to customers' interactions will persist. Leveraging various data sources and statistical tools will not solve this problem completely. Compared with data on internal interactions, measures of external transactions and competitive marketing offers at the customer level are bound to be less accurate, comprehensive, or timely. This raises questions about the implications of firms' attempts to mitigate this information asymmetry. Chief among these questions is whether firms that acquire data on customers' external activities will have a competitive advantage over their counterparts and, if so, how large such an advantage would be and for how long it would persist.
In markets with multiple firms with significant market shares, the use of customer-level competitive intelligence may reveal a path to growth, be it through promoting products that customers have only purchased from competitors or refining the timing and level at which promotions are deployed in an effort to co-opt a customer journey that would have likely ended with a transaction with a competitor.
It is critical for research to consider the impact of acquiring customer-level competitive intelligence on the competitive equilibrium. Factors including the concentration of the market, heterogeneity in user preferences, and the extent to which customers prefer their data not to be shared by firms may determine whether customer-level competitive intelligence can support growth for those who acquire the information. If a firm can acquire and integrate customer-level competitive intelligence more efficiently than its competitors, such as by enticing its customers to actively share competitive efforts in exchange for benefits, it may have an opportunity to reap a competitive advantage over others. The firm's ability to sustain this advantage, though, depends on how quickly it can erect a moat around its newfound gains and stave off competitors' counter responses. The desire for customer-level competitive intelligence may also create an incentive for firms to pool information, whether collaborating directly or through a third-party entity, depending on the extent to which participants are collectively better or worse off.
Another potential outcome is that all firms acquire customer-level competitive intelligence, incurring the costs associated with acquiring and integrating the data but not having an opportunity to benefit from them, resulting in a Bertrand supertrap ([15]). Theoretical models can evaluate the new competitive equilibrium, examining the effects on firms and customers, as well as identifying the factors that may moderate this equilibrium (e.g., [78]; [95]). Empirical research can also help address these questions through the use of single-source data in which all purchases made by a customer panel and their exposures to marketing are observed. Such data would enable researchers to build and evaluate customer response models under various data availability scenarios. Such analyses may offer firms guidance as to the potential growth they could gain from acquiring and incorporating customer-level competitive intelligence.
Arguably the most financially relevant aspect of the customer equity framework is retention ([40]), putting churn management at the heart of CRM. Firms have traditionally relied on purchase and usage data to predict customer churn (e.g., [ 4]; [57]). These internal data sources have been complemented with clickstream data, adding information on the customer's search process ([ 5]). While research on customer retention has focused on predicting churn, little attention has been given to mitigating churn, or such efforts have yielded little value ([ 5]).
This focus on churn prediction is yet another example of the streetlight effect. Once it has occurred, churn is observed, and it is relatively straightforward to leverage existing and emerging data sources to predict this outcome. Yet the goal for marketers should be to prevent churn that would have occurred otherwise. That is, mitigating churn is a counterfactual scenario that requires understanding when and why a customer may churn (e.g., [13]) so that it can be prevented. We propose the use of two underutilized data sources in customer retention to focus on churn prevention: unstructured data and causal data that focus on proactive churn management.
If one wishes to move toward mitigating churn, marketers must explore why customers are unsatisfied and may be at risk of churn. One of the ways in which research can support firm growth through improved retention is by leveraging unstructured customer–firm interaction data to better inform who is likely to churn when and why. There are several challenges to doing so.
One of the key challenges associated with analyzing customer–firm communications is that the data from these interactions are unstructured, including textual data for chats, audio data for call center conversations, and video data for service encounters. Until recently, methods for automated unstructured data analysis have been limited. While we have seen increased interest in automated analyses of textual data ([ 7]), the development of methods for visual (e.g., [63]), audio, and video data (e.g., [64]) lags. Moreover, the analysis of unstructured data must be conducted over time and linked to customer behaviors to assess the relationship between customer–firm interactions and retention.
A contributing factor to the limited use of customer–firm interaction data is that such communications may include personal data such as credit card information or customers' names that are difficult to anonymize. As such, firms are often reluctant to share these data with service providers or researchers. Scrubbing such personal information requires analyzing the data with the same advanced unstructured analytic tools that the firm is trying to acquire to gain insights from customer–firm interactions.
Unstructured data can inform what customers do, why these customers may be unsatisfied, and how they can be retained. Firms spend billions of dollars on customer communications via advertising ([38]), attending to nuances such as the font or background. Yet when the customer wants to communicate with the firm, despite being told that "this call may be recorded," the content of the interaction is often ignored. Apart from quality assurance, firms rarely investigate the conversation content systematically. While customers are notified of the recording for legal purposes, this seemingly benign statement may raise service expectations.
One of the biggest opportunities in managing churn is the development of automated methods to understand customer–firm interactions, be it through online chats, telephone calls, or direct encounters with service providers.[ 6]
From a methodological perspective, studying unstructured customer–firm interactions is novel for several reasons. In contrast to most common use of automated text analysis in marketing (e.g., analyzing UGC), which examines consumers' mass communication ([ 7]), understanding how customer–firm interactions affect retention requires analyzing the creator–receiver dyad. Considering a customer and the firm representative, analyses must identify the content creator and take conversational dynamics into account. Who says what and the preceding remarks from both sides of the dyad must be taken into account to recognize the context in which a comment is made. For instance, a comment about the price of service following a complaint about service quality may be handled differently than a comment about price following mentions of competitors.
In addition, based on the way in which customers interact with the firm, be it text, audio, or video, there is an opportunity to explore the stylistic similarity of the two agents (customer and firm representative). Previous research has demonstrated that a linguistic match between people can affect the interaction (e.g., [56]; [66]) due to aspects such as mimicry and homophily. The (mis)matching between the language, voice, or gestures of the firm representative and the customer may affect customer retention efforts. Future research could investigate how the stylistic similarity in text, audio, or imagery affects customer satisfaction and retention, paving the way for firms to train their representatives and best match representatives with customers based on their dispositions as revealed from prior interactions.
In the analysis of textual or voice conversations, the researcher is often interested in either exploring how the text or voice affects the receiver and/or what the text or voice reflects about the originator ([ 7]). Firms can improve their retention efforts by listening to their customers to identify what the conversation reflects about the customers' intentions. The weights that customers place on different aspects of service (e.g., price, quality, convenience) may be learned from the conversations that agents have over the course of the relationship. Such insights can be provided to agents prior to the start of a customer interaction so that they may be prepared with the language and potential offers that may be most effective at reducing the risk of customer churn. Alternatively, looking at the service representative, firms can investigate how the language that the service representative uses affects the customer's subsequent retention decisions. In addition to the impact of marketing offers that are provided based on what a customer has said, research could also investigate the effects of service representative empathy and listening. If such factors are found to affect customer retention decisions, this may call into question performance metrics related to the volume of calls that service representatives can field and the speed with which customer calls are dispatched.
Customer–firm interactions may be multimodal, and the choice of communication mode may provide insight into the customer's mindset. Customers may choose to communicate through chatbots for the efficiency with which service can be delivered, and in the process of doing so only provide textual responses. Those deciding to contact the firm via phone for customer service provide both textual and audio information. The free-form nature of a dialog may be preferable for those who want to elaborate on their remarks or are seeking to express frustration. Other customers may prefer a person-to-person encounter, allowing them to express themselves through text, audio, and physical movements. These distinct modes of engagement pose a modeling challenge and an opportunity of how to best combine them to identify not only the topics of the conversation, but also the customer's state of mind ([69]) and the urgency with which the firm must respond to avert churn. Audio characteristics such as volume and tempo, as well as physical movements captured on video, may enable researchers to identify the interaction characteristics that permit the earliest indication that a customer is at risk and for what reason. Such data may be more informative than the text transcript of the interaction in predicting the possibility of churn and understanding its possible causes.
Audio characteristics such as tone, pitch, and amplitude can be recorded and may reflect the customer's (and agent's) current state of mind during a customer–firm interaction. Such tools are sometimes used by call center management companies to identify the customer's mood from the pitch of their voice and, if necessary, escalate calls to more experienced agents. It has also been used to evaluate the impact of affective states during conference calls on firm performance ([70]). Investigating the content of past customer–firm interactions rather than simply focusing on the current interaction could capture changes in the relationship and help identify the customers who are most likely to churn. Deep learning methods have also been developed to analyze emotions based on audio content (e.g., [87]), which can be used to assess a customer's mindset across a sequence of interactions so as to spot changes in how customers communicate with the firm that may signal changes in the underlying relationship.
The content of customer–firm interactions can reveal not only whether a customer is likely to churn (e.g., mentioning the name of a competitor during a call center conversation), but also the underlying cause of churn. Interpreting statements such as "The competitor offered me a lower price" or "I don't get good reception in my basement" identifies the cause of churn, enabling firms to move from predicting to managing churn. In some cases, churn is outside the control of the firm (e.g., geographic relocation, death). When churn is preventable, understanding the cause can guide the firm's response ([13]). A customer complaining about the price may be "saved" by a discount, whereas churning due to product features may be mitigated by a product upgrade.
Beyond the presence of specific keywords, other aspects of a dialog may be informative about the tendency to churn and its reasons. The use of pronouns by customer service agents, for example, have been found to affect customer satisfaction, with first-person pronouns being preferred to third-person pronouns ([86]). An analysis of customer churn based on the agents with whom customers have interacted may support such a pattern, indicating that training should be sufficiently detailed to coach agents on their linguistic choices. The pronoun choices by customers can also be examined to assess whether their usage patterns are indicative of churn in the near future. In addition to pronouns, other aspects of a customer's linguistic style can provide meaningful information about the individual, such as their traits or their emotional state during the interaction ([88]). Identifying the customer's emotional state can be useful in not only predicting churn, but also in establishing the timing and possible delivery of messages to prevent it.
Just as biometric data can inform acquisition efforts, it can also be gathered during customer–firm interactions and linked to the risk of churn. Research could focus on developing techniques to analyze real-time biometric data to inform how likely a customer is to churn, as well as to provide feedback to a service associate to mitigate churn. As discussed with regard to customer acquisition, research that applies these tactics to evaluate offline service encounters would be invaluable in shedding light on the drivers of churn.
Just as the audio of customer–firm interactions reflects a customer's emotional state, biometric measures can be extracted from videos of service encounters to analyze what contributed to them being a success or failure. Facial expressions and eye tracking may reveal a level of satisfaction with a given interaction that relates to a customer's retention decision (e.g., [10]). Future research could embark on the identification of signals from video footage that are indicative of future churn. Once such patterns have been identified, which requires establishing a link between what is being conveyed by an associate and the ensuing physiological response of the customer, tools can be developed to train service associates by identifying key moments in an encounter at which the customer responded (un)favorably. Research could also use biometric data to provide associates with real-time guidance on the information that should be shared with customers or when they would be better served by moving on to another customer.
Much of firms' retention efforts are reactive as opposed to proactive ([ 5]), with firms waiting for the customer to exhibit indicators of impending churn and then offering incentives to stay, or even investing in win-back efforts after the customer churn. Such efforts often come up short because the customer has already made up their mind. If firms could identify customers who are on a path that is likely to result in churn, they may be able to deploy less costly efforts to change the customer's relationship trajectory.
Obtaining insights into customers' impending churn and the effectiveness of the firm's retention efforts can help firms move from reactive to proactive retention management ([23]). Academic research on the effectiveness of such proactive campaigns is quite sparse ([ 2]). Two key elements are needed to establish such proactive campaigns: ( 1) Which customers are at risk of churning? and ( 2) What are the effects of different proactive campaigns or messages on the risk of churning of different customers? While the first question has been heavily researched, the second question remains largely underexplored.
Evaluating retention campaigns inherently requires a counterfactual analysis to assess their potential impact, resulting in several challenges that marketers must confront.
As with other forms of field data, and particularly with respect to firms' targeted actions, customers' exposures and responses to retention efforts do not occur at random. Firms may target specific customers or groups of customers with retention efforts based on their past behaviors. Moreover, the content of these targeted campaigns is often nonrandom. Thus, efforts to isolate the effects of retention campaigns from field data require the ability to identify the target selection process or comparable groups of customers who were not targeted to serve as "controls."
Just as firms may be strategic in their deployment of retention efforts, so too are customers in their responses. Taking advantage of the fact that firms often provide incentives to dissatisfied customers to stay, strategic customers may complain to elicit such offers despite not having a grievance with the firm. Moreover, firm responses to complaints can increase customer expectations, resulting in more complaints ([67]). As such strategic customer behavior is likely to be more pronounced in response to reactive rather than proactive campaigns, focusing on proactive retention may help mitigate such behavior.
In contrast to responses to digital marketing efforts that can often be observed from click-through rates, evaluating retention campaigns requires a longer time horizon. One of the challenges this raises is which marketing efforts affect retention decisions throughout a customer's relationship with the firm. While the most recent customer touchpoint with the firm will affect retention decisions, so too may prior touchpoints, akin to the impact of earlier marketing activities in attribution models (e.g., [62]). A customer's touchpoint history with the firm must be considered to evaluate the impact of marketing interventions on retention. An additional challenge to the extended time frame over which marketing interventions must be evaluated to assess their impact on retention is that other factors may affect customers' retention decisions during that period. Among these is the potential for spillover via WOM, which can result in customers who were not actually exposed to retention efforts becoming aware of them.
A final consideration related to the long-time horizon over which retention decisions occur is estimating the value associated with a retained customer. That is, suppose that we had a means of quantifying the incremental effect of a proactive retention campaign on a given customer's annual renewal decision. It would be shortsighted to only credit an amount equal to the expected increase in revenue attributable to the campaign for a single year. While the increase in residual lifetime value might be a theoretically sound measure, this would attribute credit for revenue that has not yet been realized.
One way in which counterfactual scenarios can be evaluated is by gathering data that arise with naturally occurring variation. Such variation may create conditions akin to experiments that allow marketers to rule out other factors that might affect retention decisions.
Though identifying exogenous shocks may be difficult, we provide some illustrations of where they may occur. For service providers, exogenous shocks such as unanticipated service interruptions may affect certain customers. For example, poor reception due to weather conditions may affect only an isolated geographic area. An unexpected software glitch may cause a system outage that only affects those customers who tried to access the system during a narrow period of time. Introducing the service in one market may overwhelm the service center and have unintended consequences on the service in another unrelated geographical area. Researchers may examine the behavior of individuals before and after the service interruption, identifying those who were recipients of proactive retention efforts. As the service interruption may have affected the strength of customer relationships with the firm, we would anticipate that those who received the proactive marketing intervention would be less likely to churn following the service interruption, compared with those who did not receive such marketing efforts. The difference between these two groups offers a means of quantifying the extent to which proactive retention campaigns can mitigate the effects of service failures. Such quasi-experiments, or different matching algorithms through which customers who were "treated" with proactive retention efforts are matched with those who were untreated but have similar characteristics, may offer a means of controlling for nonrandom behavior by customers and firms.
An alternative to searching for naturally occurring variation is to undertake field experiments that have been designed to evaluate the long-term effects of retention campaigns. Field experiments have been used extensively in the domains of customer acquisition and response to marketing actions such as display advertising, website design, and catalog mailings ([36]).
Despite their success in assessing the effectiveness of other aspects of marketing, the use of field experiments for the purpose of retention has been quite limited (for exceptions, see [ 2]] and [58]]). Field experiments have been demonstrated to be useful in focusing the attention of management on customers who are likely to respond positively to retention efforts as opposed to those who are merely at risk of churning ([ 2]). There are considerable opportunities to use field experiments to inform which retention efforts are likely to be useful for which customers and at what times throughout the relationship. Future research that combines field experiments with analysis of the reasons to churn can then provide firms with a means of matching the marketing effort that is likely to be most effective with each customer. For example, some services like online games or apps have fairly high churn rates within the first few days of usage. Tracking user behavior in these first few experiences and understanding which aspects of the product grab customers' attention may help designing targeted proactive retention campaigns that better meet customer needs. Field experiments can also get at the importance of the timing with which marketing efforts are deployed.
In designing and deploying retention efforts, incorporating individual-level measures such as customer satisfaction is a promising direction to pursue. Mobile, digital and in-store electronic devices enable firms to track satisfaction in real time, not just with respect to the product but also with respect to other aspects of the experience. Such data allow firms to experiment with retention efforts that match aspects of the experience with time periods of low satisfaction (e.g., [82]). We encourage future research to combine the rich work on customer satisfaction with the causal effects of proactive retention efforts.
Just as social ties may affect customer acquisition, they may be instrumental in retention decisions, as a focal customer with social ties who have a high churn rate may also have a high risk of churning ([23]). Because it is difficult to separate social effects from homophily in historical data ([68]), data derived from social field experiments can isolate the causal social effects of churn (e.g., [ 3]). One potential avenue to explore is the derivation of social scores for customers with regard to their influence on the retention decisions of others. Such measures may be derived based on the focal customer's ties to other customers and the focal customer's role in being a bridge between different communities of customers. We expect such social effects to be particularly important for services that benefit from network externality such as online games, cloud collaboration platforms and telecommunication services. By seeding firm-generated proactive retention messages with different customers based on their location in the customer network, researchers can estimate the extent to which social influence may be a tool they can leverage to amplify the impact of retention efforts.
Overall, we believe that causal data, whether through naturally occurring variation or field experiments, can help focus research and CRM practice on proactive churn management, moving beyond mere prediction. In doing so, the field of CRM can move beyond descriptive and predictive research, and toward more theory-driven causal inference and prescriptive research.
In response to a top MSI research priority—capturing information to fuel growth—we focused on growth via the customer equity framework and discussed six data areas where we see substantial opportunities. Table 1 summarizes the promising research directions we identified.
Graph
Table 1. Summary of Future Research Directions.
| Focus Area | Opportunity | Future Research Directions |
|---|
| Customer acquisition | Incorporating biometric data | Identifying biometric responses throughout the customer journey Using biometrics to measure the effectiveness of acquisition efforts Optimizing marketing messages in real time by using biometric data
|
| Incorporatingsocial network | Redefining the value of prospects for acquisition by incorporating their impact on others in their social network Supplementing existing third-party profile data with social network data to improve targeting Incorporating social influence into effort allocation and compensation models in B2B settings
|
| Customer development | Leveraging trend data | Identifying trends by synthesizing across data sources and domains Conducting early long-range forecasting to separate meaningful trends from fads Identifying trendsetting markets or segments for leading indicators of emergence, propagation and decline of trends
|
| Incorporating competitive intelligence | Imputing size and share of wallet using multiple external data sources such as surveys, purchase panels, third party aggregators, clickstreams, social media, wearables, and mobile locations Projecting the evolution of both size and share of wallet Imputing customer-level competitive marketing activities Quantifying the value of customer-level competitive intelligence
|
| Customer retention | Usingunstructured data | Analyzing the dyadic nature of customer–firm interactions between the customer and the service provider. What does the conversation reflect about the customer and how can the service provider affect the customer? Exploring customers' state of mind and response to service interactions from audio and video data Using textual data to identify reasons to churn and their implications for proactive churn management Using biometric data to improve service encounters
|
| Harnessing causal data for proactive retention | Leveraging naturally occurring variation such as exogenous shocks or service failures to examine the causal impact of retention efforts Leveraging field experiments and measures such as real-time satisfaction to match retention efforts with customers Exploring the social effects of proactive retention efforts
|
| Managerial challenges | | Quantifying the incremental value of marketing data and applications Recognizing the potential ethical and legal costs of marketing data and their impact on customers and firms Prioritizing the data and application mix for firms along the analytics maturity curve Understanding the sustainability of data-driven growth
|
Given the breadth of the topic, it would be impossible to enumerate all the paths through which data can be turned into growth opportunities. At a fundamental level, the entirety of marketing analytics is focused on deriving information from data. Our focus is on illustrating the connection between marketing data and the three drivers of customer equity as organic growth avenues—acquiring new customers, developing existing customers, and retaining them. In doing so, we highlight six areas of data and applications that may have been overlooked due to the streetlight effect but can be potent for firms pursuing organic growth.
We hope our (admittedly selective) discussions illustrate the opportunity costs that a data availability bias creates when it comes to searching for data-driven growth. While marketing technology and analytics have advanced, they have also exacerbated the risk of falling victim to the streetlight effect as newer, bigger, and richer data are constantly thrust on marketers, often without a clear roadmap to drive growth. Despite the siren song of novel data sources, marketers must stay apprised of the ever-expanding data landscape and become familiar with the potential use cases of a wide range of data sources. Indeed, companies have emerged to create more transparency and trust in the data marketplace (e.g., Datarade, AlternativeData.org, AWS Data Exchange, ProgrammableWeb), offering to connect curated data providers with data buyers.
Marketers must stop thinking about the use of data solely in terms of how the available information can be analyzed. Rather, they need to consider the use of data as a component of a strategy problem. That is, how can existing and nascent data resources be brought into alignment with the firm's growth strategy? We believe this seemingly subtle difference in managerial orientation can make a substantial difference in turning data opportunities into growth opportunities, reducing preoccupation with short-term operational goals and tactical problems.
While we have focused on opportunities associated with leveraging select data sources, resolving the disconnect between data and firm growth requires addressing not only data and analytics challenges, but also managerial challenges. [24] discuss drivers of data and analysis adoption by management, finding that less centralized and less formalized organizations are more likely to make use of data analysis. The misalignment between data mix and growth strategy may stem from the analytics teams and marketing decision makers being at arm's length in a centralized organization. They also find that the involvement of management in data collection and research is crucial for adoption of data-driven decision making. Surprising results of data analysis were found to limit the use of data analysis in decision making, risking a confirmation bias of leveraging data only when it leads to the preconceived directional results. Of course, researchers need to further examine how robust these organizational phenomena are and identify the underlying processes. To conclude, we discuss a few broader areas that can affect the adoption of data-driven growth by firms and in which our specific observations can be couched.
Marketers must evaluate the incremental benefit of each data resource and make cost–benefit trade-offs in determining whether to generate it internally, acquire it, or forgo it, as well as which application are best fitted for each data source. The challenge in determining this trade-off is exacerbated when firms are faced with an explosion of alternatives. The literature offers little guidance on how marketers can build a data and application portfolio suitable for their growth strategies and budgets. For targeting and personalization, how can one quantify the marginal value versus cost of each additional piece of individual data? For monitoring brand equity, what is the optimal mix of data collected through tracking surveys and social listening? For media measurement and planning, what is the optimal mix of individual data for multitouch attribution modeling and aggregate data for marketing-mix modeling? For idea generation, what is the optimal mix of data collected through qualitative methods (e.g., focus groups, depth interviews, ethnographic observations) versus data gathered through machine learning based on UGC?
We have little evidence that quantifies the ROI firms can generate from investing in different data sources and applications ([ 8]). Such research would offer tremendous value to marketers as they aim to optimize their data and application mix. One approach that firms may consider is conducting field experiments (akin to the one discussed in the "Customer Retention" section) in which customers are targeted with different treatments based on different sources of data, or with different levels of data usage (e.g., with or without augmenting internal data with customer profile data from third-party providers). The firm would then be able to evaluate how its marketing decisions and customer responses would differ if a particular data source were incorporated into its decision making. The incremental performance can then be weighed against the incremental costs of data acquisition and analysis. Similarly, data suppliers can experiment with the data they provide to different organizations or use quasi–field experiments to build case studies that demonstrate the ROI of using their data.
Marketers should also consider the potential ethical and legal costs of the data they employ. For example, investing in cross-device identity resolution can help connect customer behavior from diverse devices to create a more holistic profile that informs targeting decisions and personalization. However, such rich individually identifiable data may have hidden costs that are difficult to quantify, such as changes in customer behavior due to privacy concerns, heightened risks of data breaches, and increasing costs of regulatory compliance with General Data Protection Regulation, California Consumer Privacy Act, and other emerging legislations and regulations ([20]).
With the trend toward giving customers more control over how and by whom their data are collected, there can be costly implications for data-driven growth. What if a portion of customers decide they do not want to share their data with a firm? This could lead to imbalanced data across customers, inadvertently creating data-driven discrimination that can have unintended consequences. Future research should aim to develop methods by which available data can be used to derive inferences about those who elect to not share their data with firms. Using data from third-party aggregators (e.g., Acxiom) and public social media posts, for instance, could allow some of the missing individual data to be imputed. In doing so, firms can make use of the data that consumers have purposefully shared with the public while respecting their privacy.
We are also seeing more regulations of specific types of data (e.g., the ban on sales of location data from cell phone companies in New York City) and calls to not use information for specific purposes (e.g., banning facial recognition use by police). How do marketers engage customers and regulators to convey the value of marketing data? Firms must be transparent with customers and disclose how customer-provided data will be used, and what value customers will receive in exchange for such data. Delivering on this promise will entail firms going on a data "diet," only requesting essential data from customers. Such transparency and a reduction in the stockpiling of customer data may assuage privacy concerns. It may also be prudent for marketers and future researchers to explore privacy-friendly approaches to data-driven growth. For example, the adoption and reporting of K-anonymity ([98]), in which an individual in a data set cannot be distinguished from at least K-1 other individuals who are also in the data set, could allay concerns about the invasiveness of marketing data collection and analytics.
While the academic literature favors research using data from novel and emerging sources, firms differ in terms of their state of data and analytics maturity ([21]). Many firms are still in the early stages of developing a marketing data infrastructure. Research should identify the optimal path forward for firms at different stages of data-driven decision-making readiness. Should firms relatively new to data-driven decisions first focus on building a robust CRM system, followed by a system to track marketing expenses and then collecting the "digital exhaust" (e.g., search behavior, online clickstream, social media posts)? When is the right time to think about brand tracking or competitive intelligence?
Firms also need guidance on the appropriate balance between investments in data generation/acquisition and data applications. Firms in the early stages of the analytic maturity curve may focus on building application capabilities, which can help harvest the "low-hanging fruits" from existing data. As firms mature analytically, they may rebalance their investments to data generation/acquisition efforts that can yield a competitive edge. Similarly, should young firms primarily focus on data related to customer acquisition, and invest in data related to customer development and retention only in later stages? Without evidence-based guidance, firms that are less mature analytically will face difficulty in leveraging data for growth.
A key managerial challenge about data-driven growth lies in the development of a sustainable competitive advantage when the same data sources and applications are available to all. For example, American Airlines pioneered yield management to reap large but short-lived benefits, as yield management is now the norm among modern airlines. Marketers need to outthink and out-execute their competition in ensuring the uniqueness of their data mix and uses, lest the growth opportunities they expected be short-lived due to the commoditization of data.
This challenge raises many important research questions. For example, what types of information are harder for competitors to match (e.g., combining proprietary internal data with external data to derive insights)? What types of analytical tools are harder for competitors to duplicate, thus allowing firms to see what everybody else sees, but think what nobody else does (e.g., proprietary algorithms for extracting insights from unstructured data)? To what extent can those unique data and analytical tools sustain a competitive advantage? Having answers to these questions can illuminate our thinking about the strategic challenge of turning marketing data and analytics into a sustainable driver of superior growth.
Building on the work of [42], we see several key aspects that can help determine when marketing data create sustainable competitive advantages. First, is the data proprietary? This condition certainly favors internal over external data. We believe it also favors unstructured over structured data because the former could potentially allow far more degrees of freedom in how it is used for data-enabled learning.
Second, how long do data remain relevant? Short-shelf-life data make it difficult to build sustainable data-driven competitive advantages. Marketers need to be keenly aware of the shelf life of their data, models, and algorithms. More research is needed in devising scientific methods for determining how often data and applications need to be refreshed. Using forward-looking data and analytics such as the trendspotting analysis we discussed previously can be useful to mitigate that risk.
Third, how much does the data application benefit the customer? Applications such as data-enabled personalization can create more defensible moats than, say, targeting, because the former offers more value to the customer. However, while data-enabled personalization increases the switching cost for that one customer, it does not provide an advantage in competing for new customers. More sustainable data-enabled learning would allow information "spillovers" from existing customers to new customers.
Fourth, how fast can the insights from data be incorporated into offerings? The faster the firm can bring data-enabled learnings to the market while managing data privacy concerns ([49]), the harder it is for competitors to catch up. Research should guide marketers to balance insights generated from analytics projects and the amount of time and resources it takes to run them. The "test and learn" framework is useful in expediting the implementation of data-driven insights. Furthermore, how difficult is it to imitate product improvements that are based on customer data? The key factor affecting firms' ability to overcome this challenge is whether the data-enabled improvements are hidden or deeply embedded in a complex production process, making them hard to replicate.
Whereas the data available to marketers are vast and proliferating, too often academics and practitioners view the acquisition of data and subsequent analyses as the end goal. Data that are most accessible or lend themselves to easily interpretable analyses may be preferred. Such a streetlight effect in the utilization of marketing data and applications can result in missed growth opportunities if they require data that are less accessible or harder to interpret.
Rooted in the customer equity framework, we encourage viewing marketing data and the applications they enable as a component that supports firm growth. Beyond the identification of relevant data sources and the ability to implement the appropriate analyses, this perspective requires that firms be mindful of the managerial challenges they may face in leveraging data for firm growth. We hope our discussions illustrate several data-driven growth opportunities that may have gone overlooked and serve as a call to action for researchers and practitioners to better align marketing data and analytics with firm growth.
Footnotes 1 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
2 Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
3 ORCID iDs Oded Netzer https://orcid.org/0000-0002-0099-8128 David A. Schweidel https://orcid.org/0000-0003-2665-3272
4 For example, Google Trends indexes for searches by U.S. consumers including the keyword "gluten" versus "aspartame" (https://trends.google.com/trends/explore?date=all&geo=US&q=gluten, https://trends.google.com/trends/explore?date=all&geo=US&q=aspartame).
5 An n-gram is a contiguous sequence of n words from a given sample of text (see http://www.culturomics.org).
6 While we emphasize in this section the use of unstructured data for customer retention, the use of such data in all aspects of CRM, including customer acquisition and development, has been scant. Such data can also be beneficial in identifying leads for customer acquisition and opportunities for cross-sell, upsell, or share-of-wallet measurement.
References Anderson Eric, Lin Song, Simester Duncan, Tucker Catherine. (2015), "Harbingers of Failure," Journal of Marketing Research, 52 (5), 580–92.
Ascarza Eva. (2018), "Retention Futility: Targeting High-Risk Customers Might Be ineffective," Journal of Marketing Research, 55 (1), 80–98.
Ascarza Eva, Ebbes Peter, Netzer Oded, Danielson Matthew. (2017), "Beyond the Target Customer: Social Effects of Customer Relationship Management Campaigns," Journal of Marketing Research, 54 (3), 347–63.
Ascarza Eva, Hardie Bruce G.S. (2013), "A Joint Model of Usage and Churn in Contractual Settings," Marketing Science, 32 (4), 570–90.
Ascarza Eva, Neslin Scott A., Netzer Oded, Anderson Zachery, Fader Peter S., Gupta Sunil, et al. (2018), "In Pursuit of Enhanced Customer Retention Management: Review, Key Issues, and Future Directions," Customer Needs and Solutions, 5 (1/2), 65–81.
Bell David R., Song Sangyoung. (2007), "Neighborhood Effects and Trial on the Internet: Evidence from Online Grocery Retailing," Quantitative Marketing and Economics, 5 (4), 361–400.
7 Berger Jonah, Humphreys Ashlee, Ludwig Stephan, Moe Wendy W., Netzer Oded, Schweidel David A. (2020), "Uniting the Tribes: Using Text for Marketing Insight," Journal of Marketing, 84 (1), 1–25.
8 Berman Ron, Israeli Ayelet. (2021), "The Added Value of Data-Analytics: Evidence from Online Retailers," working paper.
9 Blattberg Robert C., Deighton John. (1996), "Manage Marketing by the Customer Equity Test," Harvard Business Review, 74 (4), 136–44.
Bolton Ruth N. (1998), "A Dynamic Model of the Duration of the Customer's Relationship with a Continuous Service Provider: The Role of Satisfaction," Marketing Science, 17 (1), 45–65.
Boulding William, Staelin Richard, Ehret Michael, Johnston Wesley J. (2005), "A Customer Relationship Management Roadmap: What Is Known, Potential Pitfalls, and Where to Go," Journal of Marketing, 69 (4), 155–66.
Braun Michael, Bonfrer André. (2011), "Scalable Inference of Customer Similarities from Interactions Data Using Dirichlet Processes," Marketing Science, 30 (3), 513–31.
Braun Michael, Schweidel David A. (2011), "Modeling Customer Lifetimes with Multiple Causes of Churn," Marketing Science, 30 (5), 881–902.
Bulling Andreas, Wedel Michel. (2019), " Pervasive Eye-Tracking for Real-World Consumer Behavior Analysis," in Handbook of Process Tracing Methods, Schulte-Mecklenbeck Michael, Kuehberger Anton, Johnson Joseph G., eds. Abingdon, UK : Routledge, 27–44.
Cabral Luís, Villas-Boas J. Miguel. (2005), "Bertrand Supertraps," Management Science, 51 (4), 599–613.
Chandrasekaran Deepa, Tellis Gerard J. (2007), "A Critical Review of Marketing Research on Diffusion of New Products," Review of Marketing Research, 3 (1), 39–80.
Chen Xinlei, Chen Yuxin, Xiao Ping. (2013), "The Impact of Sampling and Network Topology on the Estimation of Social Intercorrelations," Journal of Marketing Research, 50 (1), 95–110.
Chen Yuxin, Steckel Joel H. (2012), "Modeling Credit Card Share of Wallet: Solving the Incomplete Information Problem," Journal of Marketing Research, 49 (5), 655–69.
Cheng Andria. (2019), "Healthcare May Eventually Become a Bigger Business for Best Buy Than Selling Electronics," Forbes (September 24), https://www.Forbes.Com/Sites/Andriacheng/2019/09/24/This-Business-May-Eventually-Be-Bigger-For-Best-Buy-Than-Selling-Electronics/.
Cui Toni, Ghose Anindya, Halaburda Hanna, Iyengar Raghuram, Pauwels Koen, Sriram S., et al. (2021), "Informational Challenges in Omnichannel Marketing: Remedies and Future Research," Journal of Marketing, 85 (1), 103–20.
Dallemule L., Davenport T.H. (2017), "What's Your Data Strategy?" Harvard Business Review, 95 (3), 112–21.
Davenport Thomas, Harris Jeanne. (2017), Competing on Analytics: Updated, with a New Introduction: The New Science of Winning. Boston : Harvard Business Press.
De Matos, Miguel Godinho, Pedro Ferreira, Rodrigo Belo. (2018), "Target the Ego or Target the Group: Evidence from a Randomized Experiment in Proactive Churn Management," Marketing Science, 37 (5), 793–811.
Deshpandé Rohit, Zaltman Gerald. (1982), "Factors Affecting the Use of Market Research Information: A Path Analysis," Journal of Marketing Research, 19 (1), 14–31.
Dew Ryan, Ansari Asim. (2018), "Bayesian Nonparametric Customer Base Analysis with Model-Based Visualizations," Marketing Science, 37 (2), 216–35.
Dover Yaniv, Goldenberg Jacob, Shapira Daniel. (2012), "Network Traces on Penetration: Uncovering Degree Distribution from Adoption Data," Marketing Science, 31 (4), 689–712.
Drucker Peter F. (1954), The Practice of Management. New York : Harper & Row Publishers, Inc.
Du Rex Y., Hu Ye, Damangir Sina. (2015), "Leveraging Trends in Online Searches for Product Features in Market Response Modeling," Journal of Marketing, 79 (1), 29–43.
Du Rex Y., Kamakura Wagner A. (2006), "Household Life Cycles and Lifestyles in the United States," Journal of Marketing Research, 43 (1), 121–32.
Du Rex Y., Kamakura Wagner A. (2012), "Quantitative Trendspotting," Journal of Marketing Research, 49 (4), 514–36.
Du Rex Y., Kamakura Wagner A., Mela Carl F. (2007), "Size and Share of Customer Wallet," Journal of Marketing, 71 (2), 94–113.
Eastlack, Joseph O. Jr., Rao Ambar G. (1989), "Advertising Experiments at the Campbell Soup Company," Marketing Science, 8 (1), 57–71.
eMarketer (2020), "US Retail Sales to Drop More Than 10% in 2020," (June 7), https://www.emarketer.com/content/us-retail-sales-drop-more-than-10-2020.
Ester Mike, Mickle Tripp. (2015), "PepsiCo to Drop Aspartame from Diet Pepsi: Consumer Backlash, Slumping Sales Prompt Beverage Giant to Switch Artificial Sweeteners," The Wall Street Journal, (April 24), https://www.wsj.com/articles/pepsico-to-replace-aspartame-with-sucralose-in-diet-pepsi-in-u-s-1429885941#:∼:text=PepsiCo%20Inc.,controversial%20but%20still%20artificial%20sweetener.
Favaretto Rodolfo Migon, Knob Paulo, Musse Soraia Raupp, Vilanova Felipe, Costa Ângelo Brandelli. (2019), "Detecting Personality and Emotion Traits in Crowds from Video Sequences," Machine Vision and Applications, 30 (5), 999–1012.
Feit Elea Mcdonnell, Berman Ron. (2019), "Test & Roll: Profit-Maximizing A/B Tests," Marketing Science, 38 (6), 1038–58.
Golder Peter N., Shacham Rachel, Mitra Debanjan. (2009), "Findings—Innovations' Origins: When, by Whom, and How Are Radical Innovations Developed?" Marketing Science, 28 (1), 166–79.
Gordon Brett R., Jerath Kinshuk, Katona Zsolt, Narayanan Sridhar, Shin Jiwoong, Wilbur Kenneth C. (2021), "Inefficiencies in Digital Advertising Markets," Journal of Marketing, 85 (1), 7–25.
Guadagni Peter M., Little John D.C. (1983), "A Logit Model of Brand Choice Calibrated on Scanner Data," Marketing Science, 2 (3), 203–38.
Gupta Sunil, Lehmann Donald R., Stuart Jennifer Ames. (2004), "Valuing Customers," Journal of Marketing Research, 41 (1), 7–18.
Gupta Sunil, Mela Carl F. (2008), "What Is a Free Customer Worth? Armchair Calculations of Nonpaying Customers' Value Can Lead to Flawed Strategies," Harvard Business Review, 86 (11), 102–09.
Hagiu A., Wright Julian. (2020), "When Data Creates Competitive Advantage," Harvard Business Review, 98 (1), 94–101.
Hanssens Dominique M., Pauwels Koen H. (2016), "Demonstrating the Value of Marketing," Journal of Marketing, 80 (6), 173–90.
Heist Greg, Tarraf Sarah. (2016), "Trend Analytics: A Data-Driven Path to Foresight," Marketing Insights, Spring, 18–19.
Hersh Eitan D., Schaffner Brian F. (2013), "Targeted Campaign Appeals and the Value of Ambiguity," Journal of Politics, 75 (2), 520–34.
Ho Teck-Hua, Li Shan, Park So-Eun, Shen Zuo-Jun Max. (2012), "Customer Influence Value and Purchase Acceleration in New Product Diffusion," Marketing Science, 31 (2), 236–56.
Hoffman Donna L., Novak Thomas P. (1996), "Marketing in Hypermedia Computer-Mediated Environments: Conceptual Foundations," Journal of Marketing, 60 (3), 50–68.
Iyengar Raghuram, Van Den Bulte Christophe, Valente Thomas W. (2011), "Opinion Leadership and Social Contagion in New Product Diffusion," Marketing Science, 30 (2), 195–212.
Kalaignanam Kartik, Tuli Kapil R., Kushwaha Tarun, Lee Leonard, Gal David. (2021), "Marketing Agility: The Concept, Antecedents, and a Research Agenda," Journal of Marketing, 85 (1), 35–58.
Keiningham Timothy L., Aksoy Lerzan, Buoye Alexander, Cooil Bruce. (2011), "Customer Loyalty Isn't Enough. Grow Your Share of Wallet," Harvard Business Review, 89 (10), 29–31.
Keiningham Timothy L., Buoye Alexander, Ball Joan. (2015), "Competitive Context Is Everything: Moving from Absolute to Relative Metrics," Global Economics and Management Review, 20 (2), 18–25.
Kelso Alicia. (2018), "Diet Pepsi Is Bringing Aspartame Back, Again," Food Dive (February 22), https://www.fooddive.com/news/diet-pepsi-is-bringing-aspartame-back-again/517618/#:∼:text=Dive%20Brief%3A,but%20only%20via%20e%2Dcommerce.
Kumar V. (2018), "A Theory of Customer Valuation: Concepts, Metrics, Strategy, and Implementation," Journal of Marketing, 82 (1), 1–19.
Kumar V., Andrew Petersen J., Leone Robert P. (2010), "Driving Profitability by Encouraging Customer Referrals: Who, When, and How," Journal of Marketing, 74 (5), 1–17.
Lambrecht Anja, Tucker Catherine. (2019), "Algorithmic Bias? An Empirical Study of Apparent Gender-Based Discrimination in the Display of STEM Career Ads," Management Science, 65 (7), 2966–81.
Lemaire Alain, Netzer Oded. (2020), "Linguistic-Based Recommendation: The Role of Linguistic Match Between Users and Products," working paper, Columbia University.
Lemmens Aurélie, Croux Christophe. (2006), "Bagging and Boosting Classification Trees to Predict Churn," Journal of Marketing Research, 43 (2), 276–86.
Lemmens Aurélie, Gupta Sunil. (2020), "Managing Churn to Maximize Profits," Marketing Science, 39 (5), 956–73.
Lemon Katherine N., Verhoef Peter C. (2016), "Understanding Customer Experience Throughout the Customer Journey," Journal of Marketing, 80 (6), 69–96.
Levinovitz Alan. (2015), The Gluten Lie: And Other Myths About What You Eat. New York : Simon and Schuster.
Lewis Michael. (2006), "Customer Acquisition Promotions and Customer Asset Value," Journal of Marketing Research, 43 (2), 195–203.
Li Hongshuang (Alice), Kannan P.K. (2014), "Attributing Conversions in a Multichannel Online Marketing Environment: An Empirical Model and a Field Experiment," Journal of Marketing Research, 51 (1), 40–56.
Liu Liu, Dzyabura Daria, Mizik Natalie. (2020), "Visual Listening in: Extracting Brand Image Portrayed on Social Media," Marketing Science, 39 (4), 669–86.
Liu Xuan, Shi Savannah Wei, Teixeira Thales, Wedel Michel. (2018), "Video Content Marketing: The Making of Clips," Journal of Marketing, 82 (4), 86–101.
Lodish Leonard M., Mela Carl F. (2007), "If Brands Are Built over Years, Why Are They Managed over Quarters?" Harvard Business Review, 85 (7/8), 104–12.
Ludwig Stephan, De Ruyter Ko, Friedman Mike, Brüggen Elisabeth C., Wetzels Martin, Pfann Gerard. (2013), "More Than Words: The Influence of Affective Content and Linguistic Style Matches in Online Reviews on Conversion Rates," Journal of Marketing, 77 (1), 87–103.
Ma Liye, Sun Baohong, Kekre Sunder. (2015), "The Squeaky Wheel Gets the Grease—An Empirical Analysis of Customer Voice and Firm Intervention on Twitter," Marketing Science, 34 (5), 627–45.
Manski Charles F. (1993), "Identification of Endogenous Social Effects: The Reflection Problem," Review of Economic Studies, 60 (3), 531–42.
Matz Sandra C., Netzer Oded. (2017), "Using Big Data as a Window into Consumers' Psychology," Current Opinion in Behavioral Sciences, 18, 7–12.
Mayew William J., Venkatachalam Mohan. (2012), "The Power of Voice: Managerial Affective States and Future Firm Performance," Journal of Finance, 67 (1), 1–43.
McCarthy Daniel M., Fader Peter S., Hardie Bruce G.S. (2017), "Valuing Subscription-Based Businesses Using Publicly Disclosed Customer Data," Journal of Marketing, 81 (1), 17–35.
Michel Jean-Baptiste, Shen Yuan Kui, Aiden Aviva Presser, Veres Adrian, Gray Matthew K., Pickett Joseph P., et al. (2011), "Quantitative Analysis of Culture Using Millions of Digitized Books," Science, 331 (6014), 176–82.
Minsky Laurence, Quesenberry Keith A. (2016), "How B2B Sales Can Benefit from Social Selling," Harvard Business Review (November 8), https://hbr.org/2016/11/84-of-b2b-sales-start-with-a-referral-not-a-salesperson.
Mintel (2018), "US Gluten-Free Foods Market Report," https://store.mintel.com/us-gluten-free-foods-market-report.
Mitra Debanjan, Golder Peter N. (2002), "Whose Culture Matters? Near-Market Knowledge and Its Impact on Foreign Market Entry Timing," Journal of Marketing Research, 39 (3), 350–65.
Moe Wendy W., Schweidel David A. (2012), "Online Product Opinions: Incidence, Evaluation, and Evolution," Marketing Science, 31 (3), 372–86.
Moon Sangkil, Kamakura Wagner A., Ledolter Johannes. (2007), "Estimating Promotion Response When Competitive Promotions Are Unobservable," Journal of Marketing Research, 44 (3), 503–15.
Musalem Andrés, Joshi Yogesh V. (2009), "Research Note—How Much Should You Invest in Each Customer Relationship? A Competitive Strategic Approach," Marketing Science, 28 (3), 555–65.
Netzer Oded, Feldman Ronen, Goldenberg Jacob, Fresko Moshe. (2012), "Mine Your Own Business: Market-Structure Surveillance Through Text Mining," Marketing Science, 31 (3), 521–43.
Netzer Oded, Lattin James M., Srinivasan Vikram. (2008). "A Hidden Markov Model of Customer Relationship Dynamics," Marketing Science, 27 (2), 185–204.
Neumann Nico, Tucker Catherine E., Whitfield Timothy. (2019), "Frontiers: How Effective Is Third-Party Consumer Profiling? Evidence from Field Studies," Marketing Science, 38 (6), 918–26.
Nevskaya Yulia, Albuquerque Paulo. (2019), "How Should Firms Manage Excessive Product Use? A Continuous-Time Demand Model to Test Reward Schedules, Notifications, and Time Limits," Journal of Marketing Research, 56 (3), 379–400.
Nocera Chris. (2020), "People with This Blood Type Were at Less Risk of Experiencing Severe COVID-19 Symptoms, Study Showed," ABC (July 11), https://abc13.com/Covid-19-Blood-Type-O-23andme/6310236/.
Ofek Elie, Wathieu Luc. (2010), "Are You Ignoring Trends That Could Shake Up Your Business?" Harvard Business Review, 88 (7/8), 124–31.
Packaged Facts (2016), Gluten-Free Foods in the U.S., 6th ed, https://www.packagedfacts.com/Gluten-Free-Foods-10378213/.
Packard Grant, Moore Sarah G., Mcferran Brent. (2018), "(I'm) Happy to Help (You): The Impact of Personal Pronoun Use in Customer–Firm interactions," Journal of Marketing Research, 55 (4), 541–55.
Papakostas Michalis, Spyrou Evaggelos, Giannakopoulos Theodoros, Siantikos Giorgos, Sgouropoulos Dimitrios, Mylonas Phivos, et al. (2017), "Deep Visual Attributes vs. Hand-Crafted Audio Features on Multidomain Speech Emotion Recognition," Computation, 5 (2), 26.
Pennebaker James W. (2011), "The Secret Life of Pronouns," New Scientist, 211 (2828), 42–45.
Pieters Rik, Wedel Michel. (2004), "Attention Capture and Transfer in Advertising: Brand, Pictorial, and Text-Size Effects," Journal of Marketing, 68 (2), 36–50.
Pozharliev Rumen, Verbeke Willem J.M.I., Van Strien Jan W., Bagozzi Richard P. (2015), "Merely Being with You Increases My Attention to Luxury Products: Using EEG To Understand Consumers' Emotional Experience with Luxury Branded Products," Journal of Marketing Research, 52 (4), 546–58.
Puntoni Stefano, Reczek Rebecca Walker, Giesler Markus, Botti Simona. (2021), "Consumers and Artificial Intelligence: An Experiential Perspective," Journal of Marketing, 85 (1), 131–51.
Schoenmueller Verena, Netzer Oded, Stahl Florian. (2020), "The Polarity of Online Reviews: Prevalence, Drivers and Implications," Journal of Marketing Research, 57 (5), 853–77.
Schweidel David A., Bradlow Eric T., Fader Peter S. (2011), "Portfolio Dynamics for Customers of a Multiservice Provider," Management Science, 57 (3), 471–86.
Schweidel David A., Moe Wendy W. (2014), "Listening in on Social Media: A Joint Model of Sentiment and Venue Format Choice," Journal of Marketing Research, 51 (4), 387–402.
Shin Jiwoong, Sudhir K. (2010), "A Customer Management Dilemma: When Is It Profitable to Reward One's Own Customers?" Marketing Science, 29 (4), 671–89.
Shugan Steven M., Mitra Debanjan. (2009), "Metrics—When and Why Non-Averaging Statistics Work," Management Science, 55 (1), 4–15.
Shugan Steven M., Mitra Debanjan. (2014), "A Theory for Market Growth or Decline," Marketing Science, 33 (1), 47–65.
Sweeney Latanya. (2002), "K-Anonymity: A Model for Protecting Privacy," International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems, 10 (5), 557–70.
The CMO Survey (2020), "Highlights and Insights Report," (February), https://CMOsurvey.org/Results/February-2020/.
Tonietto Gabriela N., Barasch Alixandra. (2020), "Generating Content Increases Enjoyment by Immersing Consumers and Accelerating Perceived Time," Journal of Marketing (published online September 10), DOI:10.1177/0022242920944388.
Trusov Michael, Bucklin Randolph E., Pauwels Koen. (2009), "Effects of Word-of-Mouth Versus Traditional Marketing: Findings from an Internet Social Networking Site," Journal of Marketing, 73 (5), 90–102.
U.S. Food and Drug Administration (2018), "FDA Authorizes, with Special Controls, Direct-to-Consumer Tests that Reports Three Mutations in the BRCA Breast Cancer Genes," press release (March 6), https://www.fda.gov/news-events/press-announcements/fda-authorizes-special-controls-direct-consumer-test-reports-three-mutations-brca-breast-cancer.
Venkatraman Vinod, Dimoka Angelika, Pavlou Paul A., Vo Khoi, Hampton William, Bollinger Bryan, et al. (2015), "Predicting Advertising Success Beyond Traditional Measures: New Insights from Neurophysiological Methods and Market Response Modeling," Journal of Marketing Research, 52 (4), 436–52.
Villanueva Julian, Yoo Shijin, Hanssens Dominique M. (2008), "The Impact of Marketing-Induced Versus Word-of-Mouth Customer Acquisition on Customer Equity Growth," Journal of Marketing Research, 45 (1), 48–59.
Watts Jameson. (2018), "Trend Spotting: Using Text Analysis to Model Market Dynamics," International Journal of Market Research, 60 (4), 408–18.
Yao Song, Mela Carl F. (2008), "Online Auction Demand," Marketing Science, 27 (5), 861–85.
Yoon Carolyn, Gonzalez Richard, Bettman James R. (2009), "Using fMRI to Inform Marketing Research: Challenges and Opportunities," Journal of Marketing Research, 46 (1), 17–19.
Zhong Ning, Schweidel David A. (2020), "Capturing Changes in Social Media Content: A Multiple Latent Changepoint Topic Model," Marketing Science, 39 (4), 827–46.
Zubcsek Peter Pal, Katona Zsolt, Sarvary Miklos. (2017), "Predicting Mobile Advertising Response Using Consumer Colocation Networks," Journal of Marketing, 81 (4), 109–26.
~~~~~~~~
By Rex Yuxing Du; Oded Netzer; David A. Schweidel and Debanjan Mitra
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 18- Carbon Footprinting and Pricing Under Climate Concerns. By: Bertini, Marco; Buehler, Stefan; Halbheer, Daniel; Lehmann, Donald R. Journal of Marketing. Mar2022, Vol. 86 Issue 2, p186-201. 16p. 2 Diagrams, 2 Charts. DOI: 10.1177/0022242920932930.
- Database:
- Business Source Complete
Carbon Footprinting and Pricing Under Climate Concerns
This article studies how organizations should design a product by choosing the carbon footprint and price in a market with climate concerns. The authors develop a model and first show how the cost and demand effects of reducing the product carbon footprint determine the profit-maximizing product design. They find that stronger climate concerns reduce the product carbon footprint, demand, the overall corporate carbon footprint and profit, but have an ambiguous impact on price. Next, the authors establish that offsetting carbon emissions can create a win-win outcome for the firm and the climate if the cost of compensation is sufficiently low. Going net zero leads to a win for society if the cost of offsetting is sufficiently low compared to the social cost of pollution created by the corporate carbon footprint. Third, the authors show how regulation in the form of a cap-and-trade scheme or a carbon tax affects product design, firm profitability, and green technology adoption. Finally, the authors extend the analysis to a competitive scenario and show that going net zero creates a win-win-win outcome for the firm, the climate, and society if the offset technology is sufficiently effective.
Keywords: carbon footprint; carbon offsetting; climate impact; net-zero emissions; pricing
The consequences of climate change have become apparent and touch every corner of our society. Public opinion has reached a point where "business as usual" is hard to justify, and many organizations are pressed to find solutions. For instance, "flight-shaming" and the European Green Deal ([22]) pose a threat to the business model of airlines ([ 7]). Amid much fanfare, most major carriers are studying or already adopting approaches that are broadly in line with the three-step process "measure, reduce, compensate" outlined in the United Nations Climate Neutral Now initiative ([57]), essentially pledging to operate "net-zero" flights in the short run. Similarly, the automotive industry must urgently find ways to replace combustion engines to meet the increasing demand for low-emission vehicles and more stringent emission targets ([27]). Car makers around the world are rushing to bring electric vehicles to the market at reasonable prices.
These recent developments, which generalize to organizations in logistics, fashion, retailing, and other sectors in the economy, underscore the importance of understanding climate concerns and making the necessary adjustments to one's offerings and prices. Marketing professionals play a critical role here because they are often tasked with sensing changes in consumer preferences and channeling them within an organization. According to a recent article, "chief marketing officers should be involved in the development of the sustainability strategy based on what they can bring to the table: customer data, market analysis and audience insights" ([ 8]). At the same time, however, marketing officials may lack the confidence to contribute to the debate and provide meaningful guidance to internal stakeholders ([43]).
This article develops a model that helps marketers address climate concerns by optimizing carbon footprinting and pricing. It is well documented that consumers have climate concerns ([63]; [64]) and that media coverage of climate change motivates consumers to make more sustainable consumption decisions ([11]; [30]). The starting point of our analysis is a monopoly setting in which the firm designs a product by choosing its product carbon footprint and price, hereinafter referred to simply as product design. Calculating product carbon footprints—the climate impact per unit of product in carbon dioxide equivalent (CO2 eq) emissions—is now common practice ([42]; [60]), and these footprints are routinely certified based on international accounting standards ([26]; [34]). Using terminology from the [26], we define the product carbon footprint as "cradle-to-gate emissions," which include production emissions (Scope 1) and emissions from purchased energy (Scope 2).[ 5] Importantly, reducing the product carbon footprint has both a cost effect due to the change in the unit cost to produce a greener product and a demand effect due to consumers' climate concerns.
The key difference from a standard model where the firm chooses price and (environmental) quality is that the total number of purchases made by consumers determines not only the firm's profit but also its corporate carbon footprint ([28])—the aggregate climate impact of the firm across all units sold. The corporate carbon footprint causes a market externality that depends on the strength of the climate concerns and that a regulator may want to control. Figure 1 illustrates how the interplay between firm and consumers drives the market outcome (comprising product design, firm performance, and climate impact) and climate externality, and the role played by the regulator intervening to limit corporate carbon footprints.
Graph: Figure 1. The interplay between firm, consumers, and regulator and the resulting market outcome under climate concerns.
We derive three key results from this initial framework. First, we show how the profit-maximizing product carbon footprint depends on the relative size of the cost and demand effects of reducing the product carbon footprint. This insight reflects the familiar return-on-quality logic in the marketing literature ([50]; [51]; [52]) but accounts for consumers' climate concerns.
Next, we show the impact of stronger climate concerns on the profit-maximizing product design. We demonstrate that it is optimal for a firm to decrease the product carbon footprint, and analyze the impact on price. In addition, we show that stronger climate concerns reduce firm profit.
Finally, we show that stronger climate concerns reduce the overall corporate carbon footprint. This result occurs because stronger climate concerns reduce both the product carbon footprint and demand, which leads to a reduction in overall emissions. In other words, the greener product design further contributes to the reduction in overall emissions that results from the lower sales volume. Listening to the voice of consumers who demand greener products therefore helps organizations to reduce their corporate carbon footprint.
We then extend the analysis in several directions. First, we consider the profitability of carbon offsetting, which compensates for a carbon footprint by reducing, avoiding, or sequestering carbon emissions elsewhere on the planet ([25]). Carbon offsetting is feasible because carbon emissions are a global (rather than local) environmental problem. Projects that result in carbon offsets tend to focus on renewable energy (such as building wind farms that replace coal-fired power plants) or carbon sequestration in soils or forests (such as agroforestry and tree-planting activities). Specifically, we allow the firm to purchase carbon offsets to attain a net-zero corporate carbon footprint. As a result, a firm may be able to offer a climate-neutral product even if its carbon footprint prior to offsetting is positive. We show that it is optimal for firms to go net zero if the compensation cost is sufficiently low relative to the demand-enhancing effect of reducing the product carbon footprint to net zero. In this case, going net zero is a win-win strategy for the firm and the climate.
Second, we examine the profit-maximizing product design from a welfare perspective, effectively complementing the profit motive of the firm with respect for the environment and social justice ([ 6]; [32]; [35])—often referred to as the triple bottom line of profit, planet, and people ([21]). We show that, in the absence of carbon offsetting, the profit-maximizing corporate carbon footprint generally deviates from the socially optimal level. A net-zero corporate carbon footprint, in turn, is economically efficient if the cost of offsetting is sufficiently low compared with the social cost of the corporate carbon footprint.
Third, we analyze how carbon regulation affects product design and the corporate carbon footprint. We study three common market interventions ([66]): carbon caps, cap-and-trade systems, and a carbon tax. We find that these interventions typically reduce firm profit. In addition, we show that these instruments are generally effective in curbing both the product carbon footprint and the corporate carbon footprint when taking the profit-maximizing price response into account. We also show how carbon regulation can accelerate green technology adoption.
Finally, we extend our analysis to competition and illustrate how competitive carbon offsetting emerges in equilibrium if the offset technology is sufficiently effective. From a policy perspective, this suggests that providing efficient carbon removal technologies can accelerate the transition to a low-carbon economy. Table 1 provides an overview of the key findings and highlights the insights for marketers.
Graph
Table 1. Key Results and Insights for Marketers.
| Topic | Insight |
|---|
| Product design (Propositions 1 and 2) | Climate concerns affect the product carbon footprint and price, and thereby profit. Stronger climate concerns reduce the product carbon footprint and profit, but have an ambiguous impact on price. |
| Climate impact (Proposition 3) | Stronger climate concerns reduce demand and the corporate carbon footprint. |
| Carbon offsetting (Proposition 4) | Offsetting carbon emissions can create a win-win outcome for the firm and climate if the demand effect of reducing the corporate carbon footprint to net zero is sufficiently large compared with the cost of carbon removal. |
| Corporate social responsibility (Proposition 5) | Offsetting carbon emissions can create a win-win-win outcome for the firm, climate, and society if the cost of carbon removal is sufficiently low compared with the social cost created by the corporate carbon footprint. |
| Regulation (Propositions 6–9) | Carbon regulation in the form of binding carbon caps, cap-and-trade systems, and carbon taxation reduces firm profitability, stimulates green technology adoption, and typically leads to the design of greener products. |
| Competitive strategy (Proposition 10) | Stronger climate concerns reduce the product carbon footprint and the corporate carbon footprint of each firm. If the offset technology is sufficiently effective, going net zero creates a win-win-win outcome for the firm, the climate, and society. |
Taken as a whole, our results contribute to research on green product development ([10]) by showing how carbon footprinting and pricing are determined by the interplay of consumers' climate concerns ([37]), firm technology, and market regulation ([49]). By endogenizing product design, this article also adds to the return-on-quality literature ([50]; [50]; [52]). Importantly, we provide a welfare analysis to understand the implications of product-design decisions for corporate social responsibility and thereby add to the sustainability literature in marketing ([ 9]; [14]; [32]; [39]; [47]). Finally, we extend [10] and related literature in supply chain management and engineering ([12]; [17]; [29]; [67]) by accounting for the climate externality and providing the first analysis of carbon offsetting.
Our results also contribute to the literature on regulation in economics ([ 4]) by showing how carbon caps and carbon taxes ([13]) affect product design. In addition, we show that climate regulation can trigger investments in green technologies, thereby adding to the insights of [49] on the dynamic impact of regulation and the economics of climate science more broadly ([31]; [46]; [54]).
Consider a firm that designs a product (or service) by choosing the price and product carbon footprint . The set indicates the technologically feasible product carbon footprints, where the firm offers a green product with zero emissions if and a maximally polluting brown product if . The technology of the firm results in the unit cost function defined on , where is the change in unit cost in response to a change in the product carbon footprint κ.[ 6] If , reducing the product carbon footprint increases unit cost. The opposite is true if .
We consider a market with consumers who have climate concerns and evaluate the product based on not only its intrinsic features and price p but also its carbon footprint κ. Without loss of generality, the mass of consumers is normalized to unity. A buyer derives utility
Graph
1
where is the valuation of the intrinsic features; measures the disutility from purchasing a product with carbon footprint κ, with capturing the strength of climate concerns; and is the disutility from the climate externality caused by other buyers. Because a single buyer has no impact on the climate externality, E is the same irrespective of whether or not the consumer purchases the product. By normalizing the (intrinsic) utility of the outside option to zero, a consumer purchases the product if v exceeds the perceived price .
The unobserved valuation v is distributed independently across consumers according to the cumulative distribution function . The disutility is assumed to increase at an increasing rate in the product carbon footprint κ, reflecting the increasing guilt or "cold prickle" ([ 3]) of consumers from purchasing a product that affects the climate. Formally, letting subscripts denote first and second partial derivatives, the convexity assumption can be restated as and . We set the disutility to zero if consumers do not have climate concerns or if the product is green, that is, .[ 7] The other boundary case occurs if consumers have strong climate concerns, in which case we assume that . We further assume that stronger climate concerns increase the disutility from a given carbon footprint, that is, . Finally, we assume that stronger climate concerns increase the marginal disutility of increasing κ, that is, zκλ(κ; λ) > 0.
Consumers purchase if the utility from the product exceeds the utility from the outside option. Therefore, the demand for the product is derived as
Graph
2
Demand is decreasing in the product carbon footprint and price. Interpreting the product carbon footprint as an inverse measure of product quality, a lower κ means higher quality and therefore higher demand. Lowering the product carbon footprint implies demand neutrality when consumers do not care about the climate impact of the product and demand expansion when consumers have climate concerns . The novel aspect of our modeling approach is that "product quality" affects not only demand but also the corporate carbon footprint (i.e., the overall climate impact of the firm).
The corporate carbon footprint results from multiplying the product carbon footprint by demand and is therefore given by . Note that if buyers do not fully account for their carbon emissions, they create a climate externality—"the biggest market failure the world has seen" ([54], p. 1). The climate externality results from adding up the noninternalized carbon emissions across buyers:
Graph
3
This climate externality is reduced to zero when consumers have strong climate concerns ( ) and equals the corporate carbon footprint if consumers do not care about purchasing a product that affects the climate ( ). Therefore, the corporate carbon footprint has an impact on all consumers if buyers do not fully account for the product carbon footprint when making their purchase decision.
This section first derives the profit-maximizing product carbon footprint and price of a product. We then study the impact of stronger climate concerns on these variables. Finally, we consider the impact of product design on the corporate carbon footprint. We assume throughout that the profit function is strictly concave in κ and p and thus has a unique constrained global maximum.
The firm chooses the product carbon footprint κ and the price p of the product to maximize profit. More formally, the firm solves
Graph
4
Graph
The profit function shows that the product carbon footprint and price have a dual impact on markup and demand. Proposition 1 characterizes the profit-maximizing product design with product carbon footprint and price . To facilitate exposition, all proofs are relegated to the Appendix.
- Proposition 1: If reducing the product carbon footprint lowers unit cost, the firm should offer a green product with at price , irrespective of the demand effect. If reducing the product carbon footprint increases unit cost but not demand, then it is optimal to offer a brown product with at price . Finally, if the demand effect is sufficiently strong compared with the cost effect, then it is optimal to offer a product with at price .
Proposition 1 mirrors the familiar return-on-quality logic in the marketing literature ([50]; [50]; [52]) and has two important implications. First, if lowering the product carbon footprint reduces unit cost, then it is optimal to increase efficiency and thereby increase "process quality" ([15]; [16]). Green cost cutting is more attractive when lowering the product carbon footprint not only reduces cost but also increases demand ([48]). This result helps explain why many sustainability efforts increase firm profit ([65]).
Second, if lowering the product carbon footprint increases unit cost, there may be a trade-off between the cost effect and the demand effect. Absent a demand effect, reducing the product carbon footprint below that of the brown product only results in higher unit cost and is therefore suboptimal under profit maximization. However, when the increase in demand outweighs the impact of higher unit cost, firms should reduce the product carbon footprint relative to the brown product. In contrast to cost-cutting sustainability, cost-increasing sustainability reflects the idea that "major pressure for changing marketing practices may come from consumers themselves" ([37], p. 133) and can be viewed as one of the "sustainability programs worthy of the name" ([19]). Figure 2 summarizes the product design strategies derived in Proposition 1.
Graph: Figure 2. Profit-maximizing product design as a function of the cost effect (due to the change in unit cost) and the demand effect (due to consumers' climate concerns).
Stronger climate concerns affect demand and thereby product design and firm profitability. The next result summarizes the implications.
- Proposition 2: If reducing the product carbon footprint lowers unit cost, stronger climate concerns do not affect the profit-maximizing product carbon footprint and price, and leave profit unchanged. Instead, if reducing the product carbon footprint increases unit cost, stronger climate concerns decrease the product carbon footprint, have an ambiguous impact on price, and reduce profit.
Proposition 2 has two important implications. First, it shows how climate concerns affect the profit-maximizing product design. Stronger climate concerns increase the consumers' marginal disutility of raising κ, which motivates the firm to reduce the product carbon footprint to make the product more attractive to consumers. The impact on the price is ambiguous because stronger climate concerns not only increase the unit cost due to the lower product carbon footprint, but also compress the price-cost margin.[ 8] Note that if we interpret the product carbon footprint as an inverse measure of product quality, then Proposition 2 implies an ambiguous relationship between product quality and price, which contributes to the literature on price-quality relationships ([24]; [48]).
Second, Proposition 2 implies that a monopoly firm has a motive to downplay climate concerns due to their negative impact on profit. This suggests an intuitive explanation for "dither and denial" ([62]) by polluting firms in the face of climate change ([38]; [40]). This result also points to a potential tension between product managers who tend to focus on profit and managers who are in charge of corporate social responsibility. As we will show next, one way firms can resolve this tension is by broadening the scope of performance measurement beyond profit to include climate and societal impact.
The first two propositions extend the logic of profit-maximizing product design to a setting where consumers have climate concerns. The goal of this subsection is to provide new insights on how changes in climate concerns affect the climate impact of the firm.
- Proposition 3: Stronger climate concerns reduce demand and the corporate carbon footprint .
Proposition 3 shows that stronger climate concerns necessarily reduce the corporate carbon footprint. The reason is that stronger climate concerns reduce both the product carbon footprint and demand, which leads to a reduction in overall emissions. This is a strong result, because the positive demand effect of offering a greener product could be expected to compensate for the lower product carbon footprint—similar to the rebound effect from technological progress (Alcott 2005), which suggests that higher efficiency leads to an initial reduction in demand for a resource that is outweighed by a corresponding increase in demand due to relatively lower resource cost ("Jevons paradox"). In our setting, the rebound effect cannot occur because stronger climate concerns increase the disutility from a given carbon footprint (zλ (κ; λ) > 0), which translates into an overall reduction in demand. In contrast, in a setting where zλ (κ; λ) < 0, stronger climate concerns may lead to a rebound effect because of the resulting increase in demand, an outcome that is conceivable if consumers use the brown product (rather than the green product) as a reference point.[ 9] In sum, whenever listening to the voice of consumers leads to greener product design and lower demand, the consumer pressure for greener products motivates profit-maximizing organizations to become greener, even though the impact on profit is negative. Proposition 3 thus suggests that consumers play an important role in making both products and organizations greener.
While producing a green product is perhaps the most obvious means for a firm to achieve climate neutrality, an increasingly popular alternative is to adopt an offset strategy whereby the corporate carbon footprint is fully compensated for (by funding projects that achieve an equivalent level of carbon dioxide saving), thereby creating a net-zero corporate carbon footprint. While carbon offsetting is arguably not the solution to climate change, it allows firms to achieve climate neutrality even if the available production technology does not yet allow it. In principle, any company can go net zero by buying offset services (that promote the planting of trees, renewable energy, etc.) from providers such as Carbon Footprint Ltd or Gold Standard.
Accordingly, the purpose of this section is to study under what conditions firms can benefit from adopting an offset strategy. Suppose that an offset provider charges a fixed price per unit of carbon offset. The firm then chooses the product carbon footprint κ and the price p to
Graph
5
Graph
where is the total offsetting cost of reaching a net-zero corporate carbon footprint if the product carbon footprint prior to offsetting is κ. That is, with carbon offsetting purchase decisions and demand depend on the net-zero product carbon footprint rather than the product carbon footprint prior to offsetting. The next result points to the possibility of a win-win outcome for the firm and the climate, where the benchmark is provided by the no-offset strategy.
- Proposition 4: Adopting an offset strategy is optimal for a firm if the compensation cost is sufficiently low compared with the additional profit from the demand-enhancing effect of reducing the product carbon footprint to net zero. Stronger climate concerns make the adoption of an offset strategy more attractive. The downside of an offset strategy is that it motivates a firm to increase the product carbon footprint before offsetting if the price per unit of carbon offset is sufficiently low.
Proposition 4 shows that offsetting carbon emissions can boost profit and fight climate change. The key driver of this result is that relieving consumers from disutility resulting from consuming a product with a positive carbon footprint has a demand-enhancing effect that directly translates into higher profit. A firm is more likely to adopt an offset strategy if the price per unit of carbon offset is low. This suggests that providing low-cost carbon offset options to firms might curb their corporate carbon footprints even when the standard tools of carbon regulation have no bite.
Sustainability is an umbrella term generally viewed as comprising economic profitability, respect for the environment, and social justice ([ 6]; [32]; [35]). To integrate these three ingredients into the analysis, we say that a firm behaves in a manner that is consistent with corporate social responsibility if it maximizes welfare. To do so, it must consider the triple bottom line of profit (firm and offset provider), planet (climate impact), and people (consumer surplus). Our next result shows that the adoption of an offset strategy can create a win-win-win outcome.
- Proposition 5: Without offsetting, the corporate carbon footprint is generally nonzero and different from the socially optimal level. Adopting an offset strategy that leads to net-zero carbon emissions improves welfare if the cost of carbon offsetting is sufficiently low compared with the social cost created by the corporate carbon footprint.
Proposition 5 confirms the notion that focusing exclusively on profit leads firms to make decisions that are generally inconsistent with corporate social responsibility. Intuitively, a firm has an incentive to strategically distort the product carbon footprint to exploit pricing power, which leads to an economically inefficient product carbon footprint ([53]). Interestingly, under an offset strategy, profit-maximization may result in a net-zero corporate carbon footprint even if it is socially undesirable to fully compensate for the emissions because the firm does not factor in the social cost of carbon removal. However, if the carbon removal technology is sufficiently cost effective, the win-win outcome for the firm and the climate under an offset strategy translates into a win-win-win outcome and therefore produces benefits for society at large.
In addition, Proposition 5 sheds light on the controversial debate about carbon offsets that "have been used by polluters as a free pass for inaction" ([58]). The cost efficiency of carbon offsetting stems from the fact that emissions are compensated for in places where the cost of offsetting is low, typically in developing countries. While this makes sense from an economic perspective, managers have to bear in mind "whose mess this is" and that "some of these places would welcome investment in reforestation and afforestation, but they would also need to be able to integrate such endeavours into development plans which reflect their people's needs" ([20]).
Regulators increasingly try to limit carbon emissions of firms to meet climate targets and address climate change. The most recent examples include the Green New Deal in the United States and the European Green Deal, which address climate change by introducing various regulatory interventions. We show how a firm should respond to carbon caps, cap-and-trade systems, and carbon taxes, which are by far the most common regulatory market interventions today ([66]), and study their impact on expected firm profitability. While the institutional details of these interventions vary across industries and legislations, we focus on their key characteristics and show that the risk of regulation accelerates investments in green technology.
The most direct approach to limit the corporate carbon footprint is to impose a binding carbon cap . An example is the European Union's fleet-wide binding emissions target for new cars imposed on manufacturers ([23]). In the face of such regulation, the firm solves the following problem:
Graph
6
Graph
where is the corporate carbon footprint. The next result summarizes the impact of a binding carbon cap.
- Proposition 6: A binding carbon cap reduces the corporate carbon footprint and profit, and translates into a lower product carbon footprint if the sales expansion that results from offering a greener product is sufficiently small.
A binding carbon cap has the obvious effect of reducing the corporate carbon footprint and profit. More interestingly, restricting overall emissions forces the firm to adjust the profit-maximizing product design by lowering the product carbon footprint because this helps to relax the carbon constraint if the sales expansion from offering a greener product is sufficiently small compared with the direct reduction in overall emissions. In real-world markets, however, carbon caps are often coupled with a carbon market, where firms can sell or purchase carbon allowances, which gives rise to cap-and-trade systems.
The leading examples of cap-and-trade systems are California's Cap-and-Trade Program, the Chinese National Carbon Trading Scheme, and the European Union Emissions Trading System. Cap-and-trade systems have an important advantage over carbon caps: firms with low compliance costs can sell carbon allowances in the emissions market and turn them into a source of revenue. For example, Tesla generates significant revenues by selling zero-emission vehicle credits in the United States ([41]).
The society's need to tackle climate change creates considerable uncertainty for businesses regarding their regulatory environment. To address how a firm can proactively deal with the possible introduction of regulation, we assume that a regulator is expected to implement a cap-and-trade system with probability , with a given carbon cap . Under regulation, the firm can choose between two options: ( 1) adjust the product design to meet the potential regulatory constraint at the firm level (profit ) or ( 2) stick to the current product design and purchase carbon allowances at a market price . The following result summarizes the impact of a binding carbon cap coupled with the possibility to buy carbon allowances in the emissions market.
- Proposition 7: The expected cost of a cap-and-trade system to the firm is given by , where is the expected reduction in profit if the firm complies with the carbon cap by adjusting product design, and is the expected reduction in profit if the firm purchases carbon allowances to offset the emissions. The expected cost increases when the implementation probability ρ is higher, when the carbon cap R is more severe, and when the carbon price ϖ is higher.
Proposition 7 confirms the intuition that uncertain cap-and-trade regulation reduces the expected profit of the firm. Furthermore, the cost of regulation to the firm is increasing in the probability of regulation and the market price for emissions. This is important because companies should anticipate changes in the regulatory environment and thus want to invest in the adoption of a greener technology to comply with expected regulation.
In December 2019, the International Monetary Fund issued a report suggesting that a global average carbon price of $70 a ton would be sufficient for many (but not all) countries to meet their Paris accord mitigation targets ([33]). While a carbon cap directly limits the climate impact of the firm, such a price alters the cost structure of the firm with the goal of reducing the corporate carbon footprint to a socially desirable level. To reflect this, assume that is the fixed (Pigouvian-style) tax rate on carbon emissions. Under such a proportional carbon tax, the firm solves
Graph
7
Graph
The next result summarizes the impact on the product design and firm profit, where the optimized profit under a carbon tax is denoted by .
- Proposition 8: A proportional carbon tax reduces not only the profit-maximizing product carbon footprint but also the corporate carbon footprint. The expected cost of taxation is given by , which increases in the probability ρ that a tax will be implemented and in the tax rate t.
Proposition 8 shows that a higher carbon tax effectively reduces the corporate carbon footprint via its impact on the profit-maximizing product carbon footprint and price in response to higher cost, which leads to an overall reduction in sales. The result also shows that the uncertain introduction of a carbon tax reduces expected profit: The increase in the carbon tax increases unit cost, but only part of this can be passed on to consumers in the form of higher prices—a result akin to the imperfect pass-through of trade deals documented in the channels literature ([44]; [45]). The adverse impact on profit helps explain why firms lobby against regulation ([61]).
That said, the result also shows why an offsetting strategy is an interesting option: a net-zero corporate carbon footprint makes regulation unnecessary and has an immediate positive impact on the climate. A carbon tax, in turn, affects the product carbon footprint and raises revenue for the government without offsetting the emissions. However, carbon offsets do not provide an incentive for firms to invest in green technologies and are therefore often considered an interim measure until new technologies become available.
The need to comply with carbon regulation may trigger investments in green technologies. To demonstrate this, we consider the case of a carbon cap and assume that an existing brown technology can be replaced with a green technology at a fixed cost . This green technology enables the firm to reach any product carbon footprint at a lower unit cost, that is, for all κ. Letting ρ denote the probability that carbon regulation becomes effective, we derive the following result.
- Proposition 9: The threat of carbon regulation stimulates green technology adoption if the anticipated carbon regulation reinforces the profit advantage of adopting the green technology.
Proposition 9 shows that regulatory risk provides an incentive for the firm to adopt the green technology. In other words, the mere threat of carbon regulation can prompt a reduction of the corporate carbon footprint and lead to process innovation ([49]). More broadly, from a policy perspective, the threat of effective regulation allows the government to put some of the burden of technology adoption on the shoulders of the firm.
In this section, we extend the baseline case to include competition. We first describe the interaction between two firms and consumers and then study conditions under which adopting an offset strategy is consistent with pursuing a triple bottom line.
We consider a market with two single-product firms that simultaneously choose the product carbon footprint and price . The technology of firm i is represented by the unit cost function , where is a firm-specific cost parameter. Carbon offsetting to achieve a net-zero product carbon footprint is provided by an independent provider at cost per unit of carbon emissions. The carbon removal technology of the provider is represented by the unit cost and fixed cost . Each firm can choose between two strategies: a no-offset strategy where the product is marketed with carbon footprint , or an offset strategy where the product is marketed with a net-zero carbon footprint.
The products are differentiated horizontally and vertically. Horizontal differentiation is à la Hotelling and reflects consumer heterogeneity with respect to intrinsic product features. We assume that the firms are located at the extremes of the characteristics space , that is, and . Vertical differentiation on the carbon footprint reflects the notion that a lower product carbon footprint enhances the worth of the product in the minds of consumers. Category demand is fixed, and the market consists of a unit mass of consumers. We assume that individual preferences are described by the conditional indirect utility function
Graph
8
where v is the valuation of the intrinsic product features, is the disutility from purchasing a product with carbon footprint , and is the disutility from the climate externality caused by other buyers in the market. Following convention, we let denote the consumer's preferred product characteristic and denote the horizontal distance to the product of firm i ([ 2]). The preferred product characteristics are drawn independently across consumers from a uniform distribution over the interval . Demand for the product of firm i as a function of the product carbon footprints and prices can be derived as
Graph
9
Each firm can therefore obtain a competitive advantage over its rival by offering a product with a lower carbon footprint, by charging a lower price, or both.
In a setting with two firms and two strategic options per firm, there are a total of four possible outcomes: both firms adopt a no-offset strategy, both firms adopt an offset strategy, or one firm adopts an offset strategy while the other firm adopts a no-offset strategy. To illustrate the emergence of competitive carbon offsetting, we consider symmetric firms with and let λ = 1 represent strong climate concerns. The following result holds.
- Proposition 10: Suppose that firms are symmetric and consumers have strong climate concerns. Then, if the offset technology is sufficiently effective, each firm benefits from adopting an offset strategy irrespective of the rival's choice of strategy. This leads to a net-zero industry carbon footprint and improves welfare.
To understand the intuition for this result, consider the profits that each firm can earn in the four possible outcomes represented in Figure 3. From firm 1's point of view, it is always more profitable to adopt an offset strategy than a no-offset strategy, no matter whether firm 2 chooses a no-offset strategy (which yields profit ) or an offset strategy (which yields profit ). Because the firms are symmetric, the same logic applies to firm 2, no matter whether firm 1 chooses a no-offset strategy or an offset strategy. In other words, each firm can create a win-win for itself and the climate by adopting the offset strategy—irrespective of the rival's choice of strategy. Therefore, the offset strategy is strictly dominant for each firm, and competitive carbon offsetting emerges in equilibrium.
Graph: Figure 3. Possible outcomes and corresponding profits for c10=c20=1 and λ=1, where the top left number is the profit of Firm 1 and the bottom right number the profit of Firm 2.
Interestingly, Proposition 10 further shows that if the offset technology is sufficiently cost effective, competitive forces can create a win-win-win outcome for each firm, the climate, and society. Therefore, choosing an offset strategy is consistent with pursuing corporate social responsibility. This has an important implication for policy makers: Providing efficient carbon removal technologies can accelerate the transition to a zero-carbon economy by providing incentives for firms to offer products and services with a net-zero product carbon footprint.
This article explored how organizations should design a product by choosing the carbon footprint and price in a market with climate concerns. We also analyzed how changes in product design affect profitability and the organization's overall climate impact—the corporate carbon footprint. Furthermore, we analyzed how offset strategies and carbon regulation can be used to limit the corporate carbon footprint, and how they affect green technology adoption. Finally, we examined the role of competition for product-design decisions and carbon offsetting.
Throughout, the underlying objective was to help marketing professionals understand how climate concerns translate into optimal carbon footprinting and pricing decisions. With this in mind, the current section elaborates on the implications of our results for organizations, policy makers, and consumers. We end by discussing some of the limitations of our work and avenues for future research.
When confronted with climate concerns, the first response of an organization should be to assess the carbon footprint of its product and understand the impact on cost. If it is possible to reduce the product carbon footprint and reduce cost, eliminating waste (e.g., improving energy management) and adjusting price is the obvious consequence. However, if reducing the product carbon footprint is costly, it is imperative for marketers to understand the trade-off with demand (via, e.g., market research). They can then advise their organizations on how to adjust the product design in response to stronger climate concerns.
Another key consideration is how changes in product design affect the organization's overall climate impact. Because marketers lower the product carbon footprint and adjust price in response to stronger climate concerns, the changes in product design lead to a lower corporate carbon footprint as both the product carbon footprint and demand are reduced. Therefore, greener products lead to greener organizations. At the same time, the consumer pressure for greener products reduces profit. To save the cost of making the product greener, organizations may decide to go net zero by offsetting the carbon emissions. For example, UPS offers a carbon-neutral shipping service with net-zero carbon emissions ([59]), while Kering "will become carbon neutral within its own operations and across the entire supply chain" ([36]). EasyJet announced its decision to go net zero and claims to be "the first major airline to offset the carbon emissions from the fuel used for every single flight" ([18]). In theory, net-zero carbon footprints are consistent with corporate social responsibility if the social cost of carbon compensation is sufficiently low. However, in practice, carbon compensation remains an imperfect solution that falls short of green product development.
A third consideration is whether competition forces otherwise brown organizations to offset their carbon emissions. We showed that competition has the potential to prevent an industry from a race to the bottom where firms offer brown products. For instance, several European airlines offset their carbon emissions on domestic flights to compete for climate-concerned consumers.
While carbon regulation effectively limits the organization's overall climate impact, it imposes a cost of regulation on organizations. This is arguably the reason why policy makers hesitate to implement effective carbon regulation. Our analysis shows that the mere threat of regulation negatively affects firm profitability. On the positive side, a well-designed market intervention benefits consumers and society at large, and stimulates green technology adoption.
More generally, our work suggests that society should put a price on carbon emissions. Carbon offsets and carbon taxes achieve this goal. However, carbon offsets do not provide an incentive for firms to invest in greener technologies and, therefore, should be considered an interim measure until new technologies become available. The recent call by the United Nations Global Compact to set an internal price at a minimum of $100 per metric ton by 2020 is an attempt to price carbon emissions and put climate change at the heart of corporate strategy ([56]).
Our research suggests that organizations should act on consumers' climate concerns even if it reduces their profit, because not doing so would result in even lower profit. Voicing stronger climate concerns in our model necessarily reduces the corporate carbon footprint. However, in pressuring organizations into offering greener products, consumers may end up paying higher prices.
In addition, our research shows that consumers with stronger climate concerns cause a smaller climate externality and thereby reduce the burden they impose on society. Stronger climate concerns also increase the profitability of carbon offsetting, which may stimulate the transition to a net-zero carbon economy. More broadly, our analysis suggests that "green consumerism" has a real impact on market outcomes.
Future research could study how climate concerns are shaped and how they affect the consumers' utility function. Exploring preferences is key to understanding the impact of stronger climate concerns on product design and the overall corporate carbon footprint. One approach is to assume that climate concerns are influenced by opinion leaders. Another option is to assume that the organization can influence climate concerns via persuasive advertising. Yet another issue for future research is whether changes in climate concerns monotonically affect purchase decisions.
Second, future research could consider emissions that occur during the consumption stage (Scope 3). This would allow marketers to understand what drives the life-cycle carbon footprint of a product (a cradle-to-grave approach). The interesting aspect of such an extension is that the emissions in the consumption phase are driven by consumer behavior that cannot be easily influenced by the firm.
Third, it would be interesting to study the role of competition in a more nuanced way. A limitation of our approach is that it ignores the possibility of market expansion. Researchers could also study the impact of carbon regulation and taxation on industry dynamics and their potential to accelerate the transition to a zero carbon economy.
Overall, this article highlights some of the complexities and consequences of climate concerns on product design and corporate carbon emissions. Hopefully, this will spur further research into understanding the impact of climate-dependent preferences and exploring the system-wide effects of government actions, including the determination of offset prices. We also hope to see multiple approaches brought to bear in the area, including agent-based simulations, data-based empirical analyses, and natural experiments. Ideally, this will lead to creative regulations and behaviors that result in win-win-win outcomes for consumers, organizations, and the environment or at least, absent that, better understanding of the trade-offs being made among the three parties.
Proof of Proposition 1. Assuming that the profit function is strictly concave in , the profit-maximizing product carbon footprint and price must satisfy the following necessary and sufficient Kuhn–Tucker conditions (the multipliers and are associated with the inequality constraints):
Graph
A1
Graph
A2
Graph
Depending on the slope of the unit cost function, we distinguish two cases. First, we consider the case where . Suppose that and . Then, Equation A1 leads to a contradiction as , so that . This result holds a fortiori if . Second, we assume that . If , then a solution that involves leads to a contradiction in Equation A1, so that . Next, if , then the choice of the product carbon footprint is governed by the relative strength of the cost effect and the demand effect of increasing κ: If , then , whereas if , then ; otherwise, there is an interior solution with . □
Proof of Proposition 2. First, from Proposition 1, and if . Because for , stronger climate concerns leave product design and profit unchanged.
Second, if , there are two subcases: the emergence and the reinforcement of climate concerns. In the absence of climate concerns ( ), profit at is given by . Instead, when consumers have climate concerns ( , profit at is given by . Because the emergence of climate concerns reduces demand and (weakly) increases unit cost, this implies that
Graph
A3
where and , which means that the emergence of climate concerns reduces profit. Instead, when climate concerns are reinforced, applying the envelope theorem yields
Graph
A4
where from Equation 2 and by assumption, which means that the reinforcement of climate concerns reduces profit.
To understand the impact of reinforced climate concerns on product design, suppose that , so that the multipliers and are zero in Equations A1 and A2. Substituting Equation A2 into Equation A1 yields , which can be expressed in model primitives as
Graph
A5
by using that and . Applying the implicit function theorem to Equation A5 yields
Graph
A6
because the denominator is positive by the concavity assumption and by assumption. To see the impact of stronger climate concerns on price, note that Equation A2 can be expressed in model primitives as
Graph
A7
From the implicit function theorem,
Graph
A8
The concavity assumption and imply that the first term on the right-hand side of Equation A8 is negative, whereas the second term is positive by Equation A6 and the assumption that . Therefore, the impact of stronger climate concerns on the profit-maximizing price is ambiguous.
Proof of Proposition 3. The corporate carbon footprint results from multiplying the product carbon footprint by demand and can therefore be written as
Graph
A9
Differentiating Equation A9 with respect to λ yields
Graph
A10
where the second equality follows from substituting the expressions for and in Equations A6 and A8, respectively, and using that from Equation A5. Because and by the concavity assumption, stronger climate concerns reduce demand and therefore the corporate carbon footprint because by Proposition 2.
Proof of Proposition 4. Suppose that . Under carbon offsetting, the product has a net-zero carbon footprint, which implies that z(0,λ) = 0 and thus that demand D(0,p) is independent of λ. An offset strategy yields the profit
Graph
where is the product carbon footprint prior to offsetting and is the corresponding price. Noting that , it follows from Equation A3 that . Applying the envelope theorem to the optimal profit under offsetting yields . This implies there exists such that for , which means that the firm can benefit from adopting a climate neutral strategy when ω is sufficiently low. Next, stronger climate concerns reduce profit in the benchmark case absent carbon offsets by Equation A4 ( ), whereas they leave profit unaffected under an offset strategy ( ) because demand is independent of λ. Consequently, stronger climate concerns make the adoption of an offset strategy more attractive to the firm.
Absent carbon offsets, the optimal product carbon footprint is determined by the condition from Equation A5. Under an offset strategy, the firm chooses to maximize the markup because demand does not depend on the choice of κ. Therefore, at an interior solution, is determined by the condition . Clearly, if . Note that this result holds a fortiori at a corner solution where . Thus, an offset strategy motivates a firm to increase the product carbon footprint before offsetting the emissions if ω is sufficiently low.
Proof of Proposition 5. Following convention, we define welfare as the sum of consumer surplus and profit. Consumer surplus is obtained by adding up the utilities from buyers and nonbuyers:
Graph
Graph
Graph
A11
where the third equality uses the definition of demand in Equation 2, the definition of the market externality in Equation 3, and where denotes the corporate carbon footprint. Summing consumer surplus in Equation A11 and profit in Equation 4 yields welfare:
Graph
A12
Drawing on [53], let and . The ratio of maximized profit to maximized welfare is defined as Taking logs and differentiating, it follows that Now let denote the profit-maximizing product carbon footprint. Because by definition, it follows that Thus, the product carbon footprint exceeds the socially optimal level if and conversely, which implies that the firm's choice of the product carbon footprint is not necessarily consistent with corporate social responsibility.
Adopting an offset strategy is consistent with corporate social responsibility if it increases welfare compared with the no-offset strategy. To this end, consider an offset market in which an offset provider compensates emissions at variable cost , where is an efficiency parameter, and fixed cost . In this scenario, welfare is obtained by adding up consumer surplus and the profits from the firm and the offset provider:
Graph
A13
Because the offset cost is a transfer from the firm to the offset provider, it cancels out in the welfare calculation. Using Equations A12 and A13, adopting a climate neutral strategy is economically efficient if . This condition is satisfied if the social cost of carbon offsetting is sufficiently low compared with the climate damage that results from the corporate carbon footprint under a no-offset strategy, given by .
Proof of Proposition 6. The profit-maximizing product carbon footprint and price satisfy the following Kuhn–Tucker conditions:
Graph
A14
Graph
A15
Graph
We denote the unique constrained profit-maximizing product design by and assume that the carbon constraint is binding so that , which implies that . Substituting Equation A15 into Equation A14 and rearranging yields
Graph
A16
The third term on the left-hand side of Equation A16 is negative if , as and . Consequently, if , that is, if the sales expansion from offering a greener product is sufficiently small to not compensate for the direct reduction in overall emissions.
Proof of Proposition 7. Suppose that a firm with product design ( ) and profit faces regulation that is implemented with probability ρ. If the firm decides to meet the carbon cap by adjusting the product design, the expected reduction in profit is given by , where is the constrained optimal profit. Instead, if the firm decides to leave the product design unchanged and to purchase carbon allowances, the expected reduction in profit is given by . Clearly, the firm chooses the option that minimizes the negative profit impact. Therefore, the expected cost of a cap-and-trade regulation to the firm is given by .
Proof of Proposition 8. To analyze the impact of a carbon tax, we use the same approach as in Proposition 2. Letting denote the effective unit cost, the problem in Equation 7 is structurally equivalent to the problem in Equation 4. Therefore, using Equation A5, the profit-maximizing carbon footprint at an interior solution satisfies , which can be rewritten as
Graph
A17
Applying the implicit function theorem to Equation A17 yields
Graph
A18
where the inequality follows from the second-order condition. From Equation A7, the first-order condition for the profit-maximizing price reads . Implicit differentiation yields
Graph
A19
The corporate carbon footprint can be written as
Graph
Differentiating with respect to the tax rate t and substituting for and given in Equations A18 and A19, respectively, yields
Graph
A20
Therefore, a proportional carbon reduces not only the profit-maximizing carbon footprint but also the corporate carbon footprint.
The expected cost of carbon taxation is the difference between the actual profit and the expected profit under uncertain taxation, which can be expressed as , where is the profit under a zero tax rate. Because from the envelope theorem and using the definition of the corporate carbon footprint, we have that
Graph
which implies that uncertain taxation reduces profit.
Proof of Proposition 9. In the absence of carbon regulation, the firm adopts the green technology if . With regulation, the firm adopts the green technology if . Therefore, if , the threat of carbon regulation captured by relaxes the standard adoption constraint.
Proof of Proposition 10. Demand for each firm i, , can be derived from the location of the consumer who is indifferent between buying from firm 1 and firm 2, denoted . From the indirect utility function in Equation 8, this location solves the indifference condition . With linear mismatch, the consumer located at segments the market, that is, consumers located to the left of purchase from firm , while consumers located to the right of purchase from firm . Demand of firm i can therefore be derived as
Graph
A21
To illustrate the emergence of competitive carbon offsetting, we focus on the case where and . Firm i then solves
Graph
A22
The (necessary and sufficient) first-order conditions are given by
Graph
A23
Graph
A24
Simultaneously solving the first-order conditions of firm i by substituting Equation A24 into Equation A23 and using the properties of demand in Equation A21 (with ) yields . The optimal prices are then determined by solving for equilibrium in a standard Hotelling model with given unit costs and demand. The solution is indeed the profit maximum because is concave in both and and the determinant of the Hessian matrix evaluated at ( ) is strictly positive (specifically, ). By substitution, , (the upper-left cell in Figure 3), and . Consumer surplus for buyers of firm 1 is obtained as
Graph
A25
Because consumers fully internalize their climate externality ( ), it follows that . By substitution, Equation A25 reduces to , and symmetry implies that . Welfare is obtained by aggregating consumer surplus and profit net of the climate impact across firms:
Graph
A26
Second, we analyze the setting in which firm 1 uses an offset strategy and firm 2 uses a no-offset strategy. Firm 1 therefore solves
Graph
A27
where ω denotes the offset cost per unit of carbon emissions. Instead, firm 2 solves
Graph
A28
Using a similar approach as in the case where both firms adopt a no-offset strategy, simultaneously solving the first-order conditions and substituting the unique solutions back into the profit functions yields the optimal profits and (the lower-left cell in Figure 3). Note that these profits are reversed in a setting in which firm 1 uses a no-offset strategy and firm 2 uses an offset strategy (the upper-right cell in Figure 3).
Third, we analyze the setting in which both firms adopt an offset strategy. Therefore, firm i solves
Graph
A29
Using a similar approach as in the previous cases, simultaneously solving the first-order conditions and substituting the unique solutions back into the profit functions yields the optimal profits (the lower-right cell in Figure 3).
Inspection of Figure 3 shows that adopting an offset strategy is a strictly dominant strategy for each firm. The reason is that the offset strategy is more profitable than the no-offset strategy, no matter what the competitor may choose because and for all .
These equilibrium strategy choices are consistent with corporate social responsibility if welfare is improved over the benchmark case where both firm use a no offset strategy. Welfare under offset strategies can be derived as
Graph
A30
where is the profit of the offset provider.
Carbon offsets improve welfare over the case absent offsets if . Clearly, this holds if the marginal cost ϕ and the fixed cost F are sufficiently small, that is, as long as the offset technology is sufficiently cost effective.
Footnotes 1 Brian Ratchford
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Daniel Halbheer gratefully acknowledges the financial support of the HEC Foundation and Labex Ecodec (ANR-11-LABEX-0047).
4 Marco Bertini https://orcid.org/0000-0003-3010-855X
5 The model abstracts from consumption emissions (Scope 3) because they are difficult to measure in practice ([42]).
6 To focus on the interesting cases where changes in κ affect unit cost, we abstract from .
7 Alternatively, one can interpret the disutility as the extent to which the product deviates from the "should expectation" ([5]; [55]) of a green product.
8 To see the ambiguous price effect, consider the example where v is uniformly distributed on [0,1], , and , with and . The impact of stronger climate concerns on the profit-maximizing price is , which is positive if and negative if .
9 To illustrate the possibility of a rebound effect outside of our model, consider the example where v is uniformly distributed on [0,1], , and , with and . The impact of stronger climate concerns on the corporate carbon footprint is , which is strictly positive.
References Alcott Blake. (2005), "Jevons' Paradox ," Ecological Economics , 54 (1), 9 – 21.
Anderson Simon P. , de Palma André , Thisse Jacques-François. (1992), Discrete Choice Theory of Product Differentiation. Cambridge, MA : MIT Press.
Andreoni James. (1995), " Warm-Glow Versus Cold-Prickle: The Effects of Positive and Negative Framing on Cooperation in Experiments ," Quarterly Journal of Economics , 110 (1), 1 – 21.
Armstrong Mark , Sappington David E.M.. (2007), " Recent Developments in the Theory of Regulation ," in Handbook of Industrial Organization , Vol. 3 , Armstrong Mark , Porter Robert , eds. Amsterdam : North-Holland , 1557 – 1700.
Boulding William , Kalra Ajay , Staelin Richard , Zeithaml Valarie A.. (1993), " Dynamic Process Model of Service Quality: From Expectations to Behavioral Intentions ," Journal of Marketing Research , 30 (1), 7 – 27.
Boyd Chris. (2001), " Sustainability Is Good Business ," OECD Observer , 228 (September).
Bryant Chris. (2019), " Business Class Flying Is Under Attack ," Bloomberg , (accessed January 6, 2020) , https://www.washingtonpost.com/business/energy/business-class-flying-is-under-attack/2019/12/13/c563ae08-1d80-11ea-977a-15a6710ed6da%5fstory.html.
Cavanaugh Chris. (2018), " The CMO's Role in Driving Sustainability ," Forbes (April 20) , https://bit.ly/2wyzjfW.
Chandy Rajesh K. , Johar Gita Venkataramani , Moorman Christine , Roberts John H.. (2021), " Better Marketing for a Better World, " Journal of Marketing , 85 (3), 1 – 9.
Chen Chialin. (2001), " Design for the Environment: A Quality-Based Model for Green Product Development ," Management Science , 47 (2), 250 – 63.
Chen Yubo , Ghosh Mrinal , Liu Yong , Zhao Liang. (2019), " Media Coverage of Climate Change and Sustainable Product Consumption: Evidence from the Hybrid Vehicle Market ," Journal of Marketing Research , 56 (6), 995 – 1011.
Cheng Yonghong , Zhang Pan. (2017), " Optimal Pricing and Product Carbon Footprint Strategies with Different Carbon Policies and Its Implications ," International Conference on Service Systems and Service Management. New York, IEEE , 1 – 6.
Cremer Helmuth , Thisse Jacques-François. (1999), " On the Taxation of Polluting Products in a Differentiated Industry ," European Economic Review , 43 (3), 575 – 94.
Cronin J. Joseph Jr. , Smith Jeffery S. , Gleim Mark R. , Ramirez Edward , Martinez Jennifer Dawn. (2011), " Green Marketing Strategies: An Examination of Stakeholders and the Opportunities They Present ," Journal of the Academy of Marketing Science , 39 (1), 158 – 74.
Crosby Philip B.. (1979), Quality Is Free: The Art of Making Quality Certain. New York : McGraw-Hill.
Deming W. Edwards. (1986), Out of the Crisis. Boston : MIT Center for Advanced Engineering Study.
Diabat Ali , Simchi-Levi David. (2010), " A Carbon-Capped Supply Chain Network Problem ," in IEEE International Conference on Industrial Engineering and Engineering Management. Piscataway, NJ : IEEE , 523 – 27.
EasyJet (2019), " Climate Change, Carbon Emissions and Carbon Offsetting " (accessed May 15, 2020) , https://bit.ly/2A8hk1v.
The Economist (2014), " Schumpeter—The New Green Wave ," (August 30) , econ.st/2vqKeEM.
The Economist (2019), " Climate Change—The Necessity of Pulling Carbon Dioxide out of the Air ," (December 7) , econ.st/36nQopi.
Elkington John. (1999), " Triple Bottom-Line Reporting: Looking For Balance ," Australian CPA , 69 (2), 18 – 21.
European Commission (2019), " A European Green Deal ," (accessed May 8, 2020) , https://bit.ly/2AdSpKd.
European Commission (2020), " Reducing CO2 Emissions from Passenger Cars—Before 2020 ," (accessed May 15, 2020) , https://bit.ly/37hLavC.
Gerstner Eitan. (1985), " Do Higher Prices Signal Higher Quality? " Journal of Marketing Research , 22 (2), 209 – 15.
Goodward Jenna , Kelly Alexia. (2010), Bottom Line on Offsets. Washington, DC : World Resources Institute.
Greenhouse Gas Protocol (2011), " Corporate Value Chain (Scope 3) Accounting and Reporting Standard ," (accessed November 29, 2018) , bit.ly/2SeYVDY.
Hannappel Ralf. (2017), " The Impact of Global Warming on the Automotive Industry ," AIP Conference Proceedings , 1871 , 060001.
Harangozo Gabor , Szigeti Cecilia. (2017), " Corporate Carbon Footprint Analysis in Practice—With a Special Focus on Validity and Reliability Issues ," Journal of Cleaner Production , 167 , 1177 – 83.
He Bin , Liu Yongjia , Zingbin Lingbin , Wang Shuai , Zhang Dong , Yu Qianyi. (2019), " Product Carbon Footprint Across Sustainable Supply Chain ," Journal of Cleaner Production , 241 , 118320.
Holt Diane , Barkemeyer Ralf. (2012), " Media Coverage of Sustainable Development Issues—Attention Cycles or Punctuated Equilibrium? " Sustainable Development , 20 (1), 1 – 17.
Hsiang Solomon , Kopp Robert E.. (2018), " An Economist's Guide to Climate Change Science ," Journal of Economic Perspectives , 32 (4), 3 – 32.
Huang Ming-Hui , Rust Roland T.. (2011), " Sustainability and Consumption ," Journal of the Academy of Marketing Science , 39 (1), 40 – 54.
International Monetary Fund (2019), " Putting a Price on Pollution ," Finance & Development , 56 (4), 16 –19.
International Organization for Standardization (2006), "ISO 14064-1:2018 Greenhouse Gases – Part 1: Specification with Guidance at the Organization Level for Quantification and Reporting of Greenhouse Gas Emissions and Removals ," report , https://www.iso.org/standard/66453.html.
Johnson Robert L.. (2009), " Organizational Motivations for Going Green or Profitability Versus Sustainability ," Business Review , 13 (1), 22 – 28.
Kering (2020), " Soaring Toward a Carbon-Neutral Future " (accessed May 15, 2020) , https://bit.ly/2LpPS1u.
Kotler Philip. (2011), " Reinventing Marketing to Manage the Environmental Imperative ," Journal of Marketing , 75 (4), 132 – 35.
Krugman Paul. (2018 , November 26), " The Depravity of Climate-Change Denial: Risking Civilization for Profit, Ideology and Ego ," The New York Times (November 26) , nyti.ms/2P2OdBX.
Luo Xueming , Bhattacharya C.B.. (2006), " Corporate Social Responsibility, Customer Satisfaction, and Market Value ," Journal of Marketing , 70 (4), 1 – 18.
Mann Michael E. , Toles Tom. (2016), The Madhouse Effect: How Climate Change Denial Is Threatening Our Planet, Destroying Our Politics, and Driving Us Crazy. New York : Columbia University Press.
McGee Patrick , Campbell Peter. (2019), " Fiat Chrysler Pools Fleet with Tesla to Avoid EU Emissions Fines ," Financial Times , (April 6) , on.ft.com/2G4Lp2r.
Meinrenken Christoph J. , Kaufman Scott M. , Ramesh Siddharth , Lackner Klaus S.. (2012), " Fast Carbon Footprinting for Large Product Portfolios ," Journal of Industrial Ecology , 16 (5), 669 – 79.
Moorman Christine , Kirby Lauren. (2019), " Top Marketing Trends of the Decade ," The CMO Survey (accessed March 6, 2020) , https://bit.ly/3axxFtx.
Moorthy Sridhar. (2005), " A General Theory of Pass-Through in Channels with Category Management and Retail Competition ," Marketing Science , 24 (1), 110 – 22.
Nijs Vincent , Misra Kanishka , Anderson Eric T. , Hansen Karsten , Krishnamurthi Lakshman. (2010), " Channel Pass-Through of Trade Promotions ," Marketing Science , 29 (2), 250 – 67.
Nordhaus William. (2019), " Climate Change: The Ultimate Challenge for Economics ," American Economic Review , 109 (6), 1991 – 2014.
Papadas Karolos-Konstantinos , Avlonitis George J. , Carrigan Marylyn. (2017), " Green Marketing Orientation: Conceptualization, Scale Development and Validation ," Journal of Business Research , 80 , 236 – 46.
Parasuraman A. , Zeithaml Valarie A. , Berry Leonard L.. (1985), " A Conceptual Model of Service Quality and Its Implications for Future Research ," Journal of Marketing , 49 (4), 41 – 50.
Porter Michael E. , Linde Class van der. (1995), " Toward a New Conception of the Environment-Competitiveness Relationship ," Journal of Economic Perspectives , 9 (4), 97 – 118.
Rust Roland T. , Moorman Christine , Dickson Peter R.. (2002), " Getting Return on Quality: Revenue Expansion, Cost Reduction, or Both? " Journal of Marketing , 66 (4), 7 – 24.
Rust Roland T. , Zahorik Anthony J.. (1993), " Customer Satisfaction, Customer Retention, and Market Share ," Journal of Retailing , 69 (2), 193 – 215.
Rust Roland T. , Zahorik Anthony J. , Keiningham Timothy L.. (1995), " Return on Quality (ROQ): Making Service Quality Financially Accountable ," Journal of Marketing , 59 (2), 58 – 70.
Spence A. Michael. (1975), " Monopoly, Quality, and Regulation ," Bell Journal of Economics , 6 (2), 417 – 29.
Stern Nicholas. (2008), " The Economics of Climate Change ," American Economic Review: Papers and Proceedings , 98 (2), 1 – 37.
Tse David K. , Wilton Peter C.. (1988), " Models of Consumer Satisfaction Formation: An Extension ," Journal of Marketing Research , 25 (2), 204 – 12.
United Nations (2019), " Put a Price on Carbon ," (accessed May 15, 2020) , https://bit.ly/3cBLtom.
United Nations (2020), " Climate Neutral Now ," (accessed May 8, 2020) , https://bit.ly/2SPcooV.
United Nations Environment Programme (2019), " Carbon Offsets Are Not Our Get-Out-of-Jail Free Card ," (June 10) , bit.ly/349DZ6M.
UPS (2019), " Receiving a UPS Carbon Neutral Shipment ," (accessed May 15, 2020) , https://bit.ly/2LyZ9EV.
Vandenbergh Michael P. , Dietz Thomas , Stern Paul C.. (2011), " Time to Try Carbon Labelling ," Nature Climate Change , 1 (1), 4.
Viscusi W. Kip , Harrington Joseph E. Jr. , Sappington David E.M.. (2018), Economics of Regulation and Antitrust , 5th ed. Cambridge, MA : MIT Press.
Watts Jonathan , Blight Garry , McMullan Lydia , Gutiérrez Pablo. (2019), " Half a Century of Dither and Denial—a Climate Crisis Timeline ," The Guardian (October 9) , http://bit.ly/2EVPVim.
Whitmarsh Lorraine , Capstick Stuart. (2018), " Perceptions of Climate Change ," in Psychology and Climate Change , Clayton S. , Manning C. , eds. Cambridge, MA : Academic Press , 13 – 33.
Wicker Pamela , Becken Susanne. (2013), " Conscientious vs. Ambivalent Consumers: Do Concerns About Energy Availability and Climate Change Influence Consumer Behaviour? " Ecological Economics , 88 , 41 – 48.
Winston Andrew , Favaloro George , Healy Tim. (2017), " Energy Strategy for the C-Suite ," Harvard Business Review , 95 (1), 139 – 46.
The World Bank (2015), " State and Trends of Carbon Pricing 2015 ," Washington, DC.
Yalabik Baris , Fairchild Richard J.. (2011), " Customer, Regulatory, and Competitive Pressure as Drivers of Environmental Innovation ," International Journal of Production Economics , 131 (2), 519 – 27.
~~~~~~~~
By Marco Bertini; Stefan Buehler; Daniel Halbheer and Donald R. Lehmann
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 19- Caring for the Commons: Using Psychological Ownership to Enhance Stewardship Behavior for Public Goods. By: Peck, Joann; Kirk, Colleen P.; Luangrath, Andrea W.; Shu, Suzanne B. Journal of Marketing. Mar2021, Vol. 85 Issue 2, p33-49. 17p. 1 Color Photograph, 2 Charts, 1 Graph. DOI: 10.1177/0022242920952084.
- Database:
- Business Source Complete
Record: 20- Commentary: A Strategic Perspective on Capturing Marketing Information to Fuel Growth: Challenges and Future Research. By: Morgan, Neil A.; Lurie, Robert S. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p184-189. 6p. 1 Diagram, 1 Chart. DOI: 10.1177/0022242920973036.
- Database:
- Business Source Complete
Commentary: A Strategic Perspective on Capturing Marketing Information to Fuel Growth: Challenges and Future Research
Firms are increasingly swimming in a growing sea of data ([ 5]). Yet, as evidenced by Marketing Science Institute's research priorities, most are having difficulties translating the opportunity potential afforded by these data into significant growth outcomes. [ 2], hereinafter Du et al.) highlight a number of areas in which researchers can help harness and leverage newly available data sources and tools in an effort to enable managers to uncover business opportunities. However, we believe that the root of the growth problem facing managers is both bigger and broader than data capture, requiring a significant refocusing of the academic research agenda. In this commentary, we identify two key problems limiting current knowledge and offer two new lenses to address them. We use these lenses to shed light on key challenges firms face in connecting data potential to realized growth outcomes in an effort to refocus the research agenda in a meaningful way.
First, the primary focus of most of the data and tools produced by recent developments in technology is on better understanding and improving the quality or productivity of the firm's execution of individual marketing activities such as pricing or message customization. Yet in most firms and markets, significant growth is usually a consequence of managers making new and different insight-driven strategic decisions ([ 6]). While the term "insights" is widely used and with varying meaning, we think it is useful to view market insights as a new understanding of both what is happening (or likely to happen) in a marketplace and why, which has action implications with nontrivial performance potential for the firm. For example, Starbucks's pivot from roaster to coffee shop was built on the insight that for Italians, coffee drinking was a social activity built into their daily routines—leading to the testing of Howard Schultz's "third place"[ 3] supposition. Such insight-driven strategic decisions, the decision makers involved, and the data that enable them have not been the focus of—or even a major consideration in—most of the data-capture technology developments and allied research to date in marketing. This leaves large and important gaps in current knowledge related to insights. Most importantly, how can firms build and enhance their insights capability?
To expand the content focus of future research to address such questions, we offer a new lens outlined in Figure 1 that extends beyond data and their capture. In this view, while data are a necessary "fuel," the "engine" that creates firm growth is the firm's decisions and their marketplace executions, and the "driver" of that engine is the decision maker involved. Given this lens, the key elements that researchers need to study are therefore the decisions being made, the decision makers who make them, the data they use in doing so, and the organization and marketplace contexts within which they operate.
Graph: Figure 1. Key elements in understanding firm growth challenges.
Second, as revealed by the research discussed in Du et al., most technology developments and allied research to date have focused on enabling and exploring the capture of previously unobservable marketplace phenomena at scale and speed. However, in practice, the majority of the challenges firms face in generating growth arise after data capture. Unfortunately, extant research has largely ignored these post-data-capture stages and implicitly assumed that either the rest of the process runs smoothly or that, if not, improving capture will enhance the remaining steps required "to fuel growth." We question both assumptions and believe that research focused on exploring them could be highly productive.
To expand the focus of future research in this direction, Figure 2 offers a new process lens outlining a simplified model of the data-to-growth pathway, which identifies the major stages involved in connecting marketplace data to firm growth outcomes. The model reveals a long and complex series of interdependent steps required to connect available data to firm growth. At each step in the process, critical activities must be undertaken and decisions made. Decision makers also face important challenges that weaken these steps in many organizations that are flush with data. These steps and their associated challenges offer a rich agenda for important new research.
Graph: Figure 2. The data-to-growth pathway.
The next section combines these lenses to identify and illuminate key practical problems that academic research could help solve. Many of these challenges are inevitably "messy" in a real-world sense, which means they involve multiple elements of Figure 1 and often more than one stage of Figure 2. Given this, we point to where each problem "fits" with our two lenses when it is informative to do so.
We organize the key practice-based problem areas we identify around the decision maker, data, decision, and organizational context domains in Figure 1 and conclude with the more general—but vitally important—data-to-growth process challenge. This leaves many additional process questions raised in Figure 2. It also leaves unaddressed the marketplace context challenges for firms that do not (and may never be able to) sell directly to end-user customers and capture data when they do as highlighted in Figure 1. This is not to suggest that these are unimportant questions and contexts for future research, merely a prioritization choice given space constraints for exposition.
Barring their sale to others, data are only valuable to the firm if they are processed and used by decision makers to make consequential decisions—hence our focus on insight as emerging from the intersection of these three factors in Figure 1. Yet, most extant research is focused on data and is oriented in what we call a "marketplace forward" direction, that is, focused on observing more of what is happening in the marketplace and bringing this inside the firm—largely ignoring the decision maker as well as the decisions being made. Adopting an alternative "decision maker back" perspective would both better align tools and data with "user" requirements and waste fewer resources in capturing and sharing data of less value (see also [ 1]). It may also help reduce frictions in moving from both Stages 2 to 3 and 3 to 4 in the data-to-growth pathway (Figure 2) as the data fit the decisions and the decision maker.
Researchers have much work to do to uncover which data characteristics allow marketers to make better decisions. What are the dimensions of better information for different types of decisions? Are there ways to better elicit precisely what information is most valuable to a decision maker (as researchers have already done for customers)? In addition, how data are presented makes an enormous difference in how they are comprehended and used. Understanding when and how different approaches to data presentation and visualization affect decision maker comprehension and use in making different kinds of decisions would allow data to be presented in ways that maximize its value to decision makers.
As noted, the decision maker is a central actor in the insights-to-growth process. Much of the existing data and tool-focused research to date adopts a rational, objective, fact-based perspective, which implicitly assumes rational decision makers. However, it has long been determined that decision makers are inefficient and systematically biased seekers and users of information. We believe that in practice, insights generation and use is a domain in which such biases form particularly large and challenging barriers with very significant economic consequences. For example, local search biases may be a cause of "stickiness" in firms' identification of new marketplace data sources (from Stage 1 to 2 in Figure 2) and tools and decision makers' subsequent use of them (from Stage 2 to 3). Similarly, even if new data are captured and used to generate insights, biases may limit changes in decision makers' marketplace mental schema and decision options generated and selected. For example, loss aversion and the associated sunk-cost fallacy may lead a decision maker to continue with an existing course of action, and groupthink and other group-level biases must be overcome to persuade a group of decision makers. Such bias-related problems likely permeate the entire data-to-growth process. However, current approaches to dealing with this issue that focus on simply raising awareness and understanding of such biases seem to be insufficient. This raises interesting new questions, not least of which is this: Can new tools and approaches be developed and used to reduce or even counter such human decision-making biases in insight generation and activation?
In practice, decision makers need to evaluate both cost and revenue implications of their growth-related options and actions, including the data involved. As highlighted in Figure 1, this creates the challenging problem of estimating the likely benefits of data in the face of nontrivial investments required for data capture, and the contractual relationships often involved in purchasing data. Calibrating likely data-use benefits is important not only for making such decisions in moving from Stage 1 to 2 in Figure 2, but also for subsequent accountability evaluations of the data investments made at Stage 5, when outcomes are observed. Yet, little research is available to help managers estimate and track the value of insights investments. What approaches to valuing data potential can be usefully employed? For example, can approaches used to evaluate likely customer utilization of new data sources and tools be adapted for use inside the organization? Are there valuation approaches available in data science that may be adapted? Equally, while noted by Du et al., researchers tend to ignore data cost considerations, which include not only the data themselves but also analysis costs and the time and effort involved in using it. Opportunity costs also need to be considered as data budgets and managers' time are finite resources. Research focused on understanding how managers do—and should—calibrate these costs would be helpful. Identifying common errors and offering new frameworks such as options theory would be useful in creating new process tools to guide marketers' evaluations.
The marketplace phenomena that new technologies have made observable at scale and speed have focused mainly on transactions and short-term responses to actions such as advertising, price, assortments, and so on. These data can be used to make tactical decisions that may not even require human intervention (contributing to Du et al.'s streetlight problem). For example, marketing automation makes rapid and frequent market response data-driven adjustments to firm tactics such as pricing and advertising placements. Such local optimization enhances the efficiency of the firm's marketing execution. However, significant and sustainable growth generally comes from identifying, quantifying, and characterizing "new-ish" strategic opportunities—new segments in existing markets, new platforms or product ideas, and so on—which require larger investments, integrated program changes, and longer-term paybacks. These opportunities are what we characterize as having a high degree of strategic "relevance" in Figure 1 (at the intersection of data and decisions). For example, [ 4] recount how uncovering a new segment led a business-to-business firm to make significant changes to its marketing program design and sales force organization that led to substantial growth. We need research that helps identify how insights used for longer-term strategic decisions differ from regular "research" findings used in more tactical and short-term decisions. How are such growth-relevant insights for strategic decisions generated? What inhibits their generation? What kinds of data and tools are most useful in doing so? Can marketing technology systems be designed to look for deeper patterns and opportunities that have more strategic potential for companies?
Identifying growth opportunities and generating strategic decision options that may best exploit them generally requires data from a large number and variety of sources and human syntheses that allow "why" insights in addition to the kind of "what" data required for local optimization.[ 4] In practice, synthesizing different analyses and data sources in ways that allow potentially growth-relevant relationships to be uncovered is a particularly challenging area in insight generation (from Stage 2 to 3 in Figure 2). It usually involves a combination of data (e.g., different sources, tools, and techniques), decision maker (e.g., ability and willingness to synthesize, recognize patterns, and hunt for clues and supporting evidence as to why), and organizational (e.g., time and resource availability, analysts to support decision makers during the process) components. Why is this typically so hard to routinely do in firms? Which of the various components are the most problematic in generating the kinds of syntheses required for growth-relevant insight opportunity discovery? Are there "best practices" that may aid all firms in enhancing their ability to engage in such syntheses, or is this a contingency-based firm-specific process?
While technology developments that allow for the creation of "data lakes"[ 5] have helped firms combine new and existing data in ways that can be leveraged into "single sources of truth" (which lay at the intersection of decision makers and data in Figure 1), this does not automatically translate into shared understanding among decision makers (the intersection of decision makers and decisions in Figure 1). Since consequential firm decisions are usually a shared responsibility, this shared understanding of both the meaning of available data and analyses and its implications is vital (see the discussion of Salesforce's Customer Transformation Disciplines in [ 7]). Without this, decision makers spend most of their time arguing about the "facts" and engaging in political maneuvering to "sell" their individual viewpoints rather than jointly exploring how best to use their shared understanding of the meaning and implications of data and analysis to generate and select among relevant decision options. [ 3] detail just how difficult it is to adjust decision makers' marketplace mental schema. And yet questions remain. What are the biggest barriers? Is the problem mainly a lack of shared understanding with respect to data and analysis meaning, or is it disagreements regarding its decision and action implications—or both? What are the roles of using common definitions of constructs (e.g., customer engagement) and variables (e.g., loyalty) in creating shared interpretations of data and analyses? Does the use of common analysis and decision-making frameworks enable a shared understanding with respect to the action implications of insights? Can shared understanding be accomplished without sacrificing diversity of thinking in generating insight-driven growth options?
As highlighted in Figure 1, the organizational context is a critical factor that influences all aspects of the data-to-growth pathway. Aligned with the Marketing Science Institute framing of the problem, Du et al. focus on technology developments and how they can be used to generate and capture marketplace data. Much less attention has been paid to the fundamental organizational questions involved in this process. For example, how can firms organize to better develop and benefit from marketplace insights? Should the "insights" function be centralized, dispersed, or some kind of hybrid? Staffed by specialists or generalists? Report to the chief marketing officer or someone else? Managing relationships with the insights organization is also an issue. Are insights personnel internal service providers, internal consultants, or partners in decision making? If partnering, how is responsibility and accountability for decisions managed? If service providers, are they organizers of external vendors or do they engage directly in data capture? Who should make decisions with respect to what to in-source versus outsource, and how should such decisions be made? In addition, how do you organize to make both scanning (marketplace forward) and research (decision maker back) work, and what should be the balance between the two?
Another key organizational context issue concerns the "how-to" capability issues underpinning insight generation and activation. These organizational capabilities have been sidestepped by researchers interested in data and methods thus far, but therein lies the key to their effective and efficient use over time. Most importantly, we encourage marketing leaders to focus on how they can build and enhance their firms' ability to manage the stages outlined in Figure 2, which we think can form the basis of an effective insights capability. The focus needs to be on designing common processes for accomplishing these tasks using shared frameworks and tools across the firm. Researchers can help by addressing the questions outlined in Figure 2. In particular, it would be helpful to determine the following: What are the key tasks involved in generating and using insights? What processes, frameworks, and tools may be the most useful in accomplishing these tasks under different circumstances? Building capabilities also requires personnel with the knowledge and skills needed to use the tools and frameworks and who are held accountable for doing so. What are the commonly required knowledge and skill sets? How can these be developed or acquired? What are the benefits and costs of different approaches to holding people accountable for using common insights approaches across the organization?
As revealed in Figure 2, there are many interdependent steps required and challenges associated in moving from one stage to another in the data-to-growth process. For prioritization purposes, research calibrating the extent of leakage—loss of value—between Figure 2's steps within stages, as well as between stages, would be a useful guide for managers and researchers. Research identifying the key sticking points and relative importance of various aspects of the problems uncovered would also be valuable in designing and selecting solutions to help reduce frictions in accomplishing required steps and in smoothing transitions between stages. For example, we believe that a great deal (possibly the majority) of leakage may be in moving from Stage 2 to 3—using data captured by the firm to generate insights. Is this supposition correct? If so, what steps are the most problematic tasks to accomplish and why? What distinguishes firms with lower leakage at this stage from others? We also believe that the fastest-growing area of leakage is in moving from Stage 1 to 2. This raises many questions that researchers could usefully explore. For example, how much do firms miss with respect to either not knowing or being able to evaluate what data they could possibly capture? What are the scale and nature of barriers to identifying new data and tools? How can they be overcome? What are the costs and benefits of doing so under various conditions?
As revealed in Figure 2, there are many things that must go right—and many opportunities for them not to—between new types of marketplace data generation and capture becoming available and managers subsequently obtaining and using new marketing data to change firm and customer behavior in ways that deliver firm growth. We believe that adopting a strategic perspective based on this understanding as shown in Figure 1—focusing on more than just the tools and technologies designed to observe and capture more of what is currently unobservable and marketplace contexts in which generating such data is easiest—offers researchers a much more valuable future opportunity. Our view is that effective insights lie at the intersection of data, decisions, and decision makers and leverage the organizational context as a basis for a profitable and sustainable platform for moving from data to growth. Marketing studies to date have tended to ignore these intersections, and we hope our research agenda has illuminated their importance. Further, we offer a set of process insights that need to be better understood and managed to move the firm closer to the goal of generating and activating valuable growth-relevant insights. In each case, we point to a range of challenges and research questions our field needs to address if we are to reach the goal of "insight about insight" that is required to support practitioners in this important area.
Footnotes 1 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
2 Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
3 https://www.businessinsider.com/starbucks-reimagine-third-place-2019-3.
4 While artificial intelligence tools can help establish the existence of patterns in a data set, to date they have been much less useful in answering "why" questions, which is often where most of the insight lies.
5 Centralized systems or repositories of data stored in their natural (or raw) format.
References Andreasen Alan R. (1985), "Backward Marketing Research," Harvard Business Review, 63 (May), 176–80.
Du Rex Yuxing, Netzer Oded, Schweidel David, Mitra Debanjan. (2021), "Capturing Marketing Information to Fuel Growth," Journal of Marketing, 85 (1), 163–83.
Gebhardt Gary F., Farrelly Francis J., Conduit Jodie. (2019), "Market Intelligence Dissemination Practices," Journal of Marketing, 83 (3), 72–90.
Jaworski Bernard J., Lurie Robert S. (2020), The Organic Growth Playbook: Activate High-Yield Behaviors to Achieve Extraordinary Results – Every Time. Chicago and Bingley, UK : American Marketing Association and Emerald Publishing.
Kalaignanam Kartik, Tuli Kapil R., Kushwaha Tarun, Lee Leonard, Gal David. (2021), "Marketing Agility: The Concept, Antecedents, and a Research Agenda," Journal of Marketing, 85 (1), 35–58.
6 Rodriguez-Vila Omar, Bharadwaj Sundar, Morgan Neil, Mitra Shubu. (2020), "Do You Have the Right Marketing Organization? A Framework for Aligning Growth Strategies and Capabilities," Harvard Business Review, 98 (November/December), 104–13.
7 Wild Jason. (2021), "Commentary: Beyond Data: The Mindsets and Disciplines Needed to Fuel Growth," Journal of Marketing, 85 (1), 190–95.
~~~~~~~~
By Neil A. Morgan and Robert S. Lurie
Reported by Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 21- Commentary: Artificial Intelligence: The Marketer's Dilemma. By: Kozinets, Robert V.; Gretzel, Ulrike. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p156-159. 4p. DOI: 10.1177/0022242920972933.
- Database:
- Business Source Complete
Commentary: Artificial Intelligence: The Marketer's Dilemma
[ 9]; hereinafter Puntoni et al.) have written a useful, interesting, and comprehensive summation of many of the promises and potential pitfalls of artificial intelligence (AI) for consumers. In this short commentary, we extend their contributions by focusing on the hidden challenges, and thus the dilemma, that marketers face when they utilize AI. It seemed to us that Puntoni et al. suggest in several places that companies and organizations have a degree of control over AI or its algorithms which might be unrealistic. For example, the authors write that companies could "strive to develop AI that is less, rather than more, humanlike" (p. 175) and that managers should "design both debiased and antibias AI experiences" (p. 170). We found a similar perspective in [ 1]. For example, their idea that "marketing managers" need to "specify valid objective functions" (pp. 91–92) may overstate the amount of control that most marketers have over AI. It is fair to say that a firm can customize an AI offering or its outputs without altering its core algorithms, in the same way that the passenger on an Uber ride can adjust their seat, listen to their own music, and open the window. However, most marketers in the world today are consumers, not creators, of AI technology. Creating and maintaining algorithms and AI is a complicated, expensive process that requires skilled personnel and continual monitoring and adjustment. Although AI is extremely useful to marketers, most do not produce or control it.
Although it offers a sense of the degrees of freedom marketers may have to address consumer-facing challenges, Puntoni et al.'s experience design conception and advice may unhelpfully blur the line between what most marketers are and are not able to control about AI. Their article also obscures the power shifts and practical alterations accompanying the implementation of AI in organizations today. In the larger trend, which we might term "the technologizing of marketing," marketers have largely become the users of technology rather than its masters, increasingly dependent on technology and its keepers—both those inside and outside their organizations—to do their job.
This status is not new. Since computerized sales reports, customer relationship management (CRM), e-commerce, and web pages became important parts of the marketer's toolkit, influence over the implementation of marketing has been shifting to organizational information technology departments and the types who populate them. Platforms and AI further accelerate this technologizing shift of influence and power over the customer interface. Just as with prior changes, marketers may set the goals, but they are increasingly subordinate to technologists for development, implementation, and interpretation. Even the goals of marketing may ultimately be informed by AI.
We build on Puntoni et al. by describing three important challenges that marketers face when they apply AI. After naming and explaining each of these challenges, we offer some recommendations on how marketers can face the important dilemma these challenges present.
Artificial intelligence gives marketers unprecedented knowledge about large-scale consumer patterns, but it also obscures fundamental insights about consumer behavior.
The kind of understanding that AI offers is qualitatively different from the kind of understanding that traditional marketing is based on. Artificial intelligence brilliantly and rapidly recognizes patterns in huge amounts of customer data, allowing sophisticated market prediction of user attention, traffic, sales, or engagement. As Puntoni et al. explain, it even facilitates many types of consumer nudging (or manipulation, depending on your view). But although AI recognizes patterns, it does not and cannot interpret them to give them meaning. Lindebaum, [ 5], p. 248) refer to AI-based algorithms as "supercarriers of formal rationality" in contrast to the human capacity for contextualized, value-related rational reflection and action or "substantive rationality." Applied to understand a variety of marketing problems, AI replaces substantive with "formal rationality." It recognizes patterns, but it does not understand people, their motivations, and their intentions.
Artificial intelligence provides marketers with powerful tactical tools that are bound by particular contexts, but, without adequate critical attention and caution, marketers risk offloading their own market "understanding" to the algorithms, keeping marketers out of the loop, and losing the ability to manage the marketing process. Because of the "black box" model of AI, even its creators lose control of it ([ 3]). Recently, one of the heads of digital marketing from Netflix was speaking to a small group of professors and students at the University of Southern California, which included the first author. He described how Netflix had enjoyed unprecedented success promoting one of its television shows using Facebook's machine learning combinations of rapid-fire A/B testing and programmatic advertising buys. When speaking with his advertising contacts at Facebook, however, he found that they could not explain how the AI had achieved these results. Moreover, the same type of AI-based advertising buys for other television shows yielded uneven results, none of which approached the first one. This lack of success by the AI was impossible to explain.
The Netflix example is valuable because, although the AI system "learned" and was able to achieve impressive results for a single campaign, it did not contribute to any organizational or managerial marketing knowledge. No one gained any insight into why a particular desired marketing action was or was not achieved. No one could describe the receptive target markets in a comprehensible or relatable manner. None of the people involved could explain why the system worked well at one point in time, for one show, in one context, but not in others.
Using AI for marketing campaigns may result in gains in revenue and sales. A sophisticated AI working across a large platform may be a "better" marketer than any human being ever could be. However, the employment of that AI may result in a loss of learning as well as a lack of transferability across marketing domains. The Netflix example shows how AI led to a lost opportunity to gain a general understanding about customers. The subtle differences between local markets are likewise obscured. Like a muscle that atrophies from lack of use, over time, AI-dependent marketers might lose the ability to build generalized and local understanding of customers, and to use that knowledge to market to them.
Marketers may be failing to perceive or question this atrophy because of a type of learned helplessness that [ 5], p. 257) say accompanies AI use due to the high status and legitimacy associated with the presumed rationality and objectivity of algorithmic procedures. But without recognizing that there are different—and equally valid—ways to understand consumers than those offered by AI, marketers risk losing their valuably nuanced and institutionally informed perspective. Absent an appreciation for different types of customer understanding, the incomprehensible black boxes of AI marketing—as useful and profitable as they may be—may impede the ability of organizations to build a marketing orientation ([ 4]) and could adversely affect the learning and human capital development of effective marketing organizations ([ 7]).
Artificial intelligence multiplies marketing efficiency, automation, and digital touchpoints, but it also removes valuable opportunities for marketer–customer contact and relationship-building.
Artificial intelligence fundamentally alters the opportunities for relationship building and customer contact. [ 1], p. 100) invoke the "paradox of automation" to suggest that automating the mundane and repetitive tasks of marketing will deny "customer service agents, sales representatives, content marketing editors, CRM specialists, or target marketing experts" of valuable chances to initiate, develop, and hone their marketing skills. But what might be lost pertains to more than knowledge or skills—it is also about the mutual care, trust, relationships, and loyalty that come with meaningful and continued connection between marketers and customers.
Artificial intelligence imposes a layer between marketers and customers that may lead to disconnection as well as distraction. Systems such as Google's PageRank, which is vital for online marketing, are not single algorithms or formulas. Instead, they involve many thousands of calculations, measurements, and optimizations happening with extreme rapidity and working in coordination with many inputs. Because of this complexity and indeterminacy, many contemporary marketers engage in sophisticated guesswork about the various elements of the platform algorithms that are going to affect their search rankings on Google's PageRank or, as another example, how they can tilt Facebook's programmatic advertising auctions to their advantage. [ 2] note that online advertising similarly involves a game of cat and mouse between AI and the ad fraud perpetrators who try to guess its algorithms.
Marketers are consequently spending increasing amounts of their time trying to decipher these complex puzzles so that they can "beat the system" of AI, albeit usually temporarily. Their efforts, though necessary in the short term, may lead them to miss the point of marketing. Marketers (who are consumers of AI) and their customers (also AI consumers) may each be building stronger relationships with technology platforms than they traditionally did with each other. This is not the "disintermediation" promised by [ 8] but, rather, a platform-led AI intermediation.
Without a doubt, the automation of business contact through AI and the aggregation of people's attention onto online platforms allow for vast increases in marketing efficiency. Marketing AI and the platforms it operates on provide multiple new touchpoints with customers and provide them at scale. However, those touchpoints connect to the platforms. The opportunities for brand and organizational contact on which sales, service, communication, and other marketing functions have been traditionally based may become diminished and overshadowed. Brand loyalty could also decrease as, over time, marketing grows less personal and personable. When an AI such as Alexa recommends a brand to a consumer and takes the order for the brand, it stands between the brand and the consumer. Who, in this case, does the consumer trust and build the relationship with? This divided trust is not a new problem for brands, and it is not unique to AI, but it may be amplified by the embedded, everyday presence of an AI application such as Alexa. Without additional types and forms of contact to supplement fading direct connections, marketers risk losing not only skills but also the most valuable thing they possess: meaningful customer relationships.
Marketers' increasing use of AI drives power imbalances and makes them more vulnerable to changes in algorithms.
Marketing organizations are increasingly drawn to large technology companies' platforms and AI, but, by relying on them, they may be trading off short-term gains for long-term control over their fate. Artificial intelligence benefits from the same network effects that encourage the growth of massive communication and marketing platforms such as Facebook and Amazon. It works well for marketers on these large platforms precisely because algorithms do not work as well at a smaller scale. Algorithms and AI are hungry for data, and, because they function best with massive amounts of data, they push marketers toward engaging with the large platforms. The more data points, the clearer the patterns, and the better the prediction models. For instance, a small or medium-sized enterprise will likely find only limited and diminishing success mining its own customer data from its CRM or corporate website, but the potential gains from pattern matching with Facebook or Amazon's massive data sets are far greater. Yet this network effect creates powerful inequities that advantage large platforms. Money flows toward these massive platforms and their AI, and they invest enormous amounts in platform expansion and building more algorithms. Without antitrust intervention or practical regulation of any kind, companies such as Facebook, Google, and Amazon have become the new railroad barons of online commerce and advertising.
As platforms and AI extend their technological capacities, they offer more marketing services to businesses. As these investments yield results for marketers both large and small—which are the platforms' business-to-business customers—across industries and around the world, their use grows. Core elements of the marketing profession have already become outsourced to large technology platforms and their AI. Businesses are increasingly relying on platforms to make sales, report results, gather customer information, provide service, and even communicate with customers. Marketing is becoming more automated by these features—and often, it must be said, more efficient and effective. But therein lies the problem, as these benefits may not last.
The more a business relies on large technology companies' platforms, AI, and algorithms, the more vulnerable it becomes to changes in them that could affect it adversely. For example, in 2018, the women's lifestyle publisher Little Things was hit by a change to Facebook's algorithm that was intended to prioritize content from friends and family in users' news feeds. The changes cut Little Things' influencer and organic traffic by 75%. Blaming the change on the new Facebook algorithm, the company shut down. Consider next a hypothetical example of a sportswear marketer that becomes highly dependent on Amazon as its retail channel. The more that the marketer relies on Amazon for its sales, the more likely it is that changes in Amazon's search algorithms (e.g., to favor a new Amazon sportswear brand), will affect the marketer. In these examples, the marketers who rely more on AI are also made more vulnerable. He who lives by the algorithm, dies by the algorithm. Algorithms can change at any time, they remain hidden from marketers' view, and marketers have no ability to control them. Ultimately, technology companies are building and tweaking those algorithms to look out for their own need to make profits and reward their shareholders. Often their purposes are the same as their marketer clients, but not always. Yet those goals will always supersede the needs of the increasingly vulnerable marketers who depend on AI.
[ 6] identified a paradox of enslavement and empowerment inherent in all technology consumption. Because most marketers are consumers and not producers of AI, they seem to be subject to this selfsame paradox. In this short commentary on Puntoni et al., we discuss AI's usefulness and value to marketers. But we add to this praise the consideration that it can lead to interrelated challenges of incomprehensibility, disconnection, and vulnerability that present contemporary marketers with a genuine dilemma. This dilemma, centered on what we called the technologizing of marketing, asks marketers to consider what marketing becomes when it is increasingly performed by technologies and their keepers. If we conceive of marketing as an act of understanding and fulfilling customer needs, what happens when that understanding pertains to increasingly formal and automated, rather than substantive, matters? Rather than serving as an augmentation to marketing, will AI preside over the eventual decline of marketing into an automated computational exercise?
We don't think so. Many of these challenges can be met, at least partially, through the deliberate pursuit and use of supplemental alternatives to (proprietary) AI—and with better oversight. First, even as marketers use AI to understand large patterns in customer behavior, they should continue to use and develop techniques that emphasize empathic understanding, human insight, and substantive rationality. These skills may be assisted by machine learning but never replaced by it. Second, marketers need additional conduits for customer relationship-building that take them beyond the big technology platforms and their AI. These channels might include owned media, expanded customer service contacts, deliberate outreach (e.g., mailing lists, email) with a more personal touch, and even platforms based on open AI standards. Diversifying marketing channels will also ensure that businesses are less vulnerable to unwanted changes in algorithms. Third, the already-established battle between marketers and technologists ([10]) must proceed, inside and outside organizations, and it has higher stakes today than ever before. Marketers demonstrate their value to an organization when they contribute discernment and understanding that technologists never could, such as about cultural contexts, customer journeys, motivations, fantasies, and a myriad of other meaningfully multidimensional concepts that are difficult to determine through decontextualized data mapping.
By diversifying their perspectives on understanding, their links to customers, and their use of platforms—and by advocating for smart oversight of technology companies—marketers can avoid AI's technology paradox pitfall of enslavement and still enjoy many of the remarkable gifts of its empowerment.
Footnotes 1 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
2 Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
References De Bruyn Arnaud, Viswanathan Vijay, Beh Yean Shan, Brock Jürgen Kai-Uwe, Wangenheim Florian von. (2020), "Artificial Intelligence and Marketing: Pitfalls and Opportunities," Journal of Interactive Marketing, 51, 91–105.
Gordon Brett R., Jerath Kinshuk, Katona Zsolt, Narayanan Sridhar, Shin Jiwoong, Wilbur Kenneth C. (2021), "Inefficiencies in Digital Advertising Markets," Journal of Marketing, 85 (1), 7–25.
3 Knight Will. (2017), "The Dark Secret at the Heart of AI," MIT Technology Review (April 11), https://www.technologyreview.com/2017/04/11/5113/the-dark-secret-at-the-heart-of-ai/.
4 Kohli Ajay K., Jaworski Bernard J. (1990), "Market Orientation: The Construct, Research Propositions, and Managerial Implications," Journal of Marketing, 54 (2), 1–18.
5 Lindebaum Dirk, Vesa Mikko, Hond Frank den. (2020), "Insights from 'The Machine Stops' to Better Understand Rational Assumptions in Algorithmic Decision Making and Its Implications for Organizations," Academy of Management Review, 45 (1), 247–63.
6 Mick David Glen, Fournier Susan. (1998), "Paradoxes of Technology: Consumer Cognizance, Emotions, and Coping Strategies," Journal of Consumer Research, 25 (2), 123–43.
7 Moorman Christine, Day George S. (2016), "Organizing for Marketing Excellence," Journal of Marketing, 80 (6), 6–35.
8 Negroponte Nicholas. (1996), Being Digital. New York : Vintage.
9 Puntoni Stefano, Reczek Rebecca Walker, Giesler Markus, Botti Simona. (2021), "Consumers and Artificial Intelligence: An Experiential Perspective," Journal of Marketing, 85 (1), 131–51.
Workman John P. (1993), "Marketing's Limited Role in New Product Development in One Computer Systems Firm," Journal of Marketing Research, 30 (4), 405–21.
~~~~~~~~
By Robert V. Kozinets and Ulrike Gretzel
Reported by Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 22- Commentary: Beyond Data: The Mindsets and Disciplines Needed to Fuel Growth. By: Wild, Jason. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p190-195. 6p. 1 Color Photograph, 1 Chart, 1 Graph. DOI: 10.1177/0022242920972398.
- Database:
- Business Source Complete
Commentary: Beyond Data: The Mindsets and Disciplines Needed to Fuel Growth
Common wisdom holds that big data should be well suited to drive growth through marketing motions such as those described by [ 2]; hereinafter Du et al.). However, our work with chief executive officers and their teams over the years has pointed to a more urgent theme—the pressing need to accelerate business transformation that extends beyond data-driven marketing techniques.
For most organizations, data and technology alone are not enough to drive new growth. Rather, leaders must determine how to put customers at the center of everything they do. As Salesforce's President and Chief Product Officer, Bret Taylor, says, "Transformation happens when leaders stop focusing internally on technology, products, departments, or systems—and re-center around their customers" ([ 5]). While this is not a new idea, as the extensive literature on market and customer orientation and customer-centricity attest, our experience has yielded new insights, which I share in this commentary.
As we work with CEOs and their teams to accelerate business transformation, we have seen that technology and data can help inform that effort. More often than not, however, it requires an organizational shift—one we call "Customer Transformation." This shift reimagines the experiences customers have with the organization's products and services. To put it simply, every person across the organization needs to understand the customer's context and then focus on serving that context every day. Companies that fail to make this shift may already be in the midst of disruption. Such companies are like whales: when a whale dies, it takes years for it to sink to the bottom of the sea. Until it does, other sea creatures feed off its remains. Companies that fail to adopt a customer-centric mindset may not even realize they are already dead and being consumed.
Our engagements with leaders across industries have made it clear: to create change, you need to first shift the organization's mindset and then develop the critical disciplines needed to ensure success.
A key challenge is that most leadership teams have a variety of conflicting beliefs about how to transform. Many of the executives with whom we work adhere to one of two mindsets—the common assumptions, orthodoxies, and actions that both hold companies back and propel them forward.
With regard to the first mindset, we meet executives in many industry-leading organizations who see the current wave of new technologies as just the latest in a regular flow of incremental innovations. They interpret new technologies as an extension of previous digitization efforts and believe that optimizing for productivity and efficiency will lead to reliable outcomes. We call this the Renovate mindset. This tactical approach can generate isolated successes, but it typically fails to scale benefits across the organization. And, unless cost savings are reinvested in a more ambitious transformation, it remains a purely defensive strategy that is vulnerable to disruption. It also carries the added risk of exposing the organization to "data blind spots," as discussed by Du et al., where misplaced emphasis on readily available data may result in prioritizing short-term growth ahead of long-term growth.
In contrast, we see a different mindset in almost every disruptor entrepreneur, which we call the Transcend mindset. These leaders and their organizations imagine new ways to create customer value and reject traditional definitions of competitors and markets. They reject legacy processes in favor of a new and relentless focus to the customer. They leverage new technologies across a broad ecosystem to innovate.
Of course, existing companies cannot simply flip a switch and build a new technology-ready organization from the ground up, like most digital disruptors have. These organizations need to find ways to construct new business models, develop new offerings, and deliver compelling customer experiences within the realities of their existing middle and back offices.
Practically, this means transforming their existing culture and organizational capabilities—the business processes that have been honed into muscle memory over decades—and the underlying technology stack that supports them. It is a difficult challenge that requires a mindset that serves as a bridge between the two, as shown in Figure 1.
Graph: Figure 1. The customer transformation mindset shift.
We see successful leaders decisively commit to move out of the Renovate mindset to a new Evolve mindset, with a deliberate plan to shift from product-centricity to customer-centricity. This bridge to a future Transcend view recognizes the reality of where an organization is today. It does not mean there will not be Transcend-type initiatives—these are imperative. Yet, everything cannot change at once. Some aspects of the business—such as accounting procedures, legal functions, inventory management, and so forth—may continue to operate in a Renovate fashion for some time, driving efficiency and creating savings that can be put toward true transformation.
Given this requirement, the central question is this: how does an organization move from Renovate to Evolve? We see a clear, actionable answer to that question: a company that is committed to Customer Transformation must develop four disciplines. Think of these four disciplines as four muscles or capabilities a company needs to build to make Customer Transformation a reality: ( 1) customer-centric business processes, ( 2) one team aligned around the customer, ( 3) leanest possible technology stack, and ( 4) sense and respond.
Based on our experience with industry leaders, we have synthesized what we have learned, including step-by-step strategies, into a leadership playbook called the Customer Transformation Framework.[ 3] I next dig deeper into what each discipline entails—and the steps required to make it a reality. Figure 2 summarizes this framework.
Graph: Figure 2. Disciplines required for customer transformation.
Many businesses grew up in an era when success meant creating new product brands, advertising through broadcast media, and then distributing those products to customers. The only way to achieve scale was through standardization, and given the limits of what was possible with data, the focus was purely on transactions and transaction volumes.
Those approaches no longer work in the Fourth Industrial Revolution—a term coined by Klaus Schwab of the World Economic Forum to describe the blurring of boundaries between the physical, digital, and biological worlds ([ 6]). In a connected, digital world, companies must pivot from product-centric business processes that drive internal productivity to customer-centric experiences that focus on the customer at every touchpoint and business processes that align the organization around the customer. It is no longer a matter of B2C (business to consumer), B2B (business to business), or even B2B2C (business to business to consumer); rather, organizations must now focus on B2 H: business to humans.
This new framing helps companies begin by seeking to understand the very human needs of their customers first. This practice of human-centered rather than product-centered process design is a critical first step on the path to customer-centricity. A company focused only on selling its products or services risks misunderstanding its customers' desired end goal—their "jobs to be done." Disruptors capitalize on the gaps created by legacy companies who fail to fully understand the human needs that ultimately drive the buying process. To stay competitive, organizations must focus on clearly defining and benchmarking their customers' ideal buying journey and user experience. The customers' needs, once brought into focus, should then drive planning, resourcing, and budgeting (rather than an operations or supply-chain view). Success metrics should be recalibrated to track lifetime customer value over individual transactions (see also Du et al.). Opportunities to leverage artificial intelligence–optimized tools to automate repetitive tasks must be identified and implemented in order to free employees to better serve customer needs. And finally, organizations must seek opportunities to connect with customers on a human level by proactively providing an emotional connection to the brand.
The hospitality industry in particular has experienced the impact of this revolution firsthand. Disruptors like Airbnb, Hotels.com, Google, and Amazon now give travelers more options and control. With over 7,000 properties and 30 brands, Marriott turned to Salesforce to help them leverage technology to respond to this new competitive pressure. Salesforce helped them shift from "room-centric" property management systems to processes and systems organized around each Marriott guest.[ 4] Universal guest profiles recognize customers across properties and brands—streamlining reservation, check-in, room personalization, and concierge services—and allowing Marriott to use a mobile app to serve guests directly and in a customized manner.
To deliver personalized and connected experiences at scale, firms must move beyond mere transactions and develop a holistic, contextual, and actionable view of the customer. Over time, they must understand who their most valuable customers are and find new ways to identify customer needs and execute against those needs in near real time.
When an organization functions as a team in service of the customer, internal information silos typical of complex businesses are more accurately viewed as obstacles. Breaking down those silos enables teams across the company to deliver the full power of the organization in service of the customer. It is about moving from a partitioned view of responsibilities in which specializations matter most to a holistic view in which the whole company comes together with shared metrics and cross-functional collaboration. This requires abandoning the organizational structures we inherited from the early twentieth century—with rigid divisions between sales, marketing, service, and production—and instead adopting more flexible and flatter team structures.
To achieve this goal, organizations must commit to optimizing their organizational structure, in addition to their business processes, in service of customer needs. This structural shift requires a customer-centric culture throughout the organization. A truly customer-centric culture is driven by values that the entire organization understands, internalizes, and works to achieve every day. Board and executive leaders must serve as visible, active champions of customer-centricity. Teams must be structured around customer journeys and optimized for jobs to be done, rather than tied to product lines or sales channels. Success metrics must be based on customer outcomes, rather than units sold. None of this is possible without cross-functional visibility of customer interactions at every stage, cross-functional collaboration, and a shared understanding of success with clear accountability.
To help Humana's care teams meet their patients' changing needs during the COVID crisis, Salesforce helped them develop the ability to easily collaborate by offering a single, central view of each member's complete clinical history and insight into social, environmental, and lifestyle factors impacting that member's health. This integrated health experience, designed through the lens of Humana's members, has reshaped the firm's approach to patient care. This single view extends to providers, as noted by William Fleming, President of Clinical and Pharmacy Solutions at Humana: "Humana's journey toward integrated care is so important to delivering the best possible health outcomes for our members. As part of this important work, we're advancing interoperability so providers and participants in a member's care team have connected, simplified healthcare experiences."[ 5]
The discipline of maintaining the leanest possible technology stack focuses first on reducing the complexity of an organization's information technology (IT) footprint so spend can shift from an ongoing need to "keep the lights on" to a greater focus on innovation on behalf of the customer and the ecosystem supporting such efforts. Most large enterprise companies will spend 70%–80% of their annual IT budget on maintaining existing technologies (or just "keeping the lights on"), leaving just a small residual budget for technology-driven innovation, experimentation, and governance. Leading companies have consciously sought to streamline their technology footprint in order to effectively invert their IT spend to be 70%–80% focused on customer innovation, experimentation, and governance. This shift ensures IT is well positioned to develop and support tools required to meet emerging customer needs and streamline internal jobs so employee time can be reallocated to more creative tasks.
A second shift involves reducing the expertise required to integrate different parts of the stack, using modern approaches like APIs and low-code development tools instead of legacy bespoke integration and developer-led approaches. This reduction also enables firms to quickly reconfigure their stack as customer needs and data sources evolve. This, combined with an emphasis on enterprise-wide transparency and coordination, provides a jolt of energy to IT. Organizations that commit to reducing the overall complexity of their suite of technical tools must focus on delivering up-to-date, accessible, and accurate customer information as it evolves.
In keeping with the idea of moving from product supply and transactions, the relative differentiation power of enterprise resource planning declines and customer relationship management begins to play a more central role in the stack given its ability to provide a single view of the customer (see Du et al.). A customer profile is never truly complete; rather, it captures and synthesizes the ongoing relationship between the company and the customer over time. Employees throughout the company must all be able to access this "single source of truth" in real time and keep it updated as they interact with customer themselves. For example, Unilever used Salesforce to consolidate 50 different systems serving more than 90,000 employees globally. This enabled them to deliver personalized, relevant experiences to every customer across their portfolio of brands, according to Unilever Chief Digital and Growth Officer Peter ter Kulve ([ 1]).
Putting the previous three disciplines into place provides an incredible foundation for customer-centricity, but not quite enough. The problem is that customers are changing—their needs, their expectations, and their behavior—faster than ever before ([ 4]). And unless an organization builds the capacity to sense and respond to those changes in near real time, it risks falling behind more agile competitors.
For many years, the biggest companies dominated. Then the fastest companies took the lead, whether in terms of speed to market or the ability to launch new products quickly to meet rapid changes in demand. Now, companies that are most able to adapt are emerging as leaders. The ability to listen to and internalize customer feedback, in as close to real time as possible, helps the organization evolve as fast as the market moves. Today, a real-time or "zero latency" organization is the aspiration.
Many companies struggle to respond quickly and effectively to customer-generated data. Indeed, as indicated by Du et al., organizations often fall victim to the "streetlight effect," meaning they are either overly reliant on readily available data or neglect data altogether when making business decisions. This risk can be addressed by prioritizing the Sense and Respond discipline, which emphasizes rigor in the collection and consumption of customer-generated data in close to real time. Companies that aspire to leverage real-time customer insights to shape future customer experiences must embrace experimentation. Given the freedom to experiment, teams can use iterative processes to rapidly solve problems in response to shifting customer needs (such as the trend information identified in Du et al.). Employees with ready access to on-the-go customer information, and supported by AI-driven responses, can efficiently receive, process, and respond to free-form customer feedback, which then becomes a key input to sharpen and validate the organization's core purpose.
Adidas worked with Salesforce to use the data collected during online interactions with an individual consumer over time to tailor a subsequent engagement with that specific consumer. The Vice President of Digital Experience Design for Adidas, Jacqueline Smith-Dubendorfer, confirms: "those data points then enable us to adapt what we present, when we present it, and how we present it to ensure that we deliver as close to what that customer is looking for as we can."[ 6] Beyond simply presenting tailored content to a shopper, Adidas also uses these data to create better products and offer custom-made products that can be manufactured on the fly and delivered to consumers remarkably fast.
When everyone from the CEO to a customer-facing service agent has access to transparent, up-to-date, and actionable insights, the impact can be enormous. But simply rolling out new technology is not enough. It has to be accompanied by a practice of continuous iteration, underpinned by an ability to make small, meaningful adjustments as new information enters the system.
Crucially, at the level of organizational culture, this is about moving from information as power—to be hoarded and leveraged for personal or departmental goals—to an expectation of transparency and access for all.
Over the past 20+ years, I've had the pleasure (and pain) of working with leaders who are grappling with large-scale change. I have led customer engagements and projects in 38 countries—with clients ranging from Disney to the North Atlantic Treaty Organization. Of course, leaders can and should do many things. However, their most fundamental role is to successfully manage change. But in the past 30 years, only a scant handful of Fortune 500 companies have successfully pivoted from one core business model to another. It is a rare thing to truly transform.
Trillions of dollars have been spent over the past decades on "digital transformation," with very few tangible results to show for it. Du et al. propose that big data is the answer to this problem, and many experts still make the claim that "data is the new oil." I believe this is a shortsighted approach that mischaracterizes a critical point. Data are not something to be exploited. Data must be earned to truly have value. To create truly enduring relationships with customers, organizations must earn the right to ask for and receive data from their customers and ecosystems—a point overlooked by Du et al. How is this done? By building relationships based on shared purpose and outcomes. In fact, the exchange of high-quality data for high-quality customer experiences creates a flywheel effect, as shown in Figure 3.
Graph: Figure 3. The data flywheel.
When a company delivers on its promise of customer-centric business practices, it generates customer trust. In turn, that trust earns it access to high-quality customer data, and high-quality customer data informs more efficient delivery of excellent customer experiences. Humans create insights. Insights create better data. And the virtuous cycle—built on mutual trust—continues.
I saw this effect in action when we worked with Bajaj, a financial services company based in India. A deeper look in to their "voice of the customer" data revealed that during Diwali, a holiday focused on gift giving, financial services products were in high demand precisely at the moment that most banks were closed ([ 3]). By understanding what their customers wanted to do and how they wanted to interact, Bajaj made the decision to be there to engage. They were able to originate hundreds of thousands of dollars in loans on a day that traditionally had generated zero dollars in revenue.
Leaders and CEOs cannot delegate the future. They must actively create a culture that surfaces raw information about what customers are saying, and they must build processes that use data to drive deeper relationships with their customers. The great leaders that inspire me have a set of common traits: they model the mindsets and behavior of the future, they involve and engage many in cocreating the future of the organization, they value clarity of vision and an action bias, and they create mutual trust with key stakeholders through a shared purpose driven by clear values.
Becoming a customer-centric company is about more than just collecting the right data to deliver the right marketing messages to a target audience. Rather, putting the customer at the center of every organizational function is more likely to lead to Customer Transformation—including redesigned business structures and processes, streamlined technology implementations, and investments that enable the organization to rapidly sense and respond to customer needs as they change. At its core, the mindsets and disciplines outlined in the Customer Transformation Framework shape a company's ability to transform.
Simply pouring new technology over old thinking will not work. As the CEO of a large Israeli bank once said to me, "It's like putting a digital skin on an analog body." Most business models were designed for a world that no longer exists. We were convinced this was true before the COVID-19 pandemic, and we are even more convinced now. COVID has disrupted global markets and has affected countless lives. But it has also been an accelerator of long-overdue changes in corporate mindsets and has helped refocus many organizations on clarity of purpose.
From the Fourth Industrial Revolution to the economic crises driven by a global pandemic, the context for business has materially changed.
The urgency is real—and so is the potential.
Footnotes 1 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
2 Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
3 Salesforce Customer 360 Playbook: https://www.salesforce.com/form/pdf/customer-360-playbook/?d=cta-body-promo-63&nc=7010M000002E9ShQAK.
4 https://www.salesforce.com/customer-success-stories/marriott/.
5 https://investor.salesforce.com/press-releases/press-release-details/2020/Humana-Selects-Salesforce-to-Deliver-Connected-Personalized-Healthcare/default.aspx.
6 https://www.salesforce.com/customer-success-stories/adidas/.
References Dreamforce Keynote (2018), https://www.salesforce.com/video/3402956/
Du Rex Yuxing, Netzer Oded, Schweidel David A., Mitra Debanjan. (2021), "Capturing Marketing Information to Fuel Growth," Journal of Marketing, 85 (1), 163–83.
Horn Robert. (2019), "Does Design Pay Off? Yes. (And There's Data to Prove It.)" Fortune (March 5), https://fortune.com/2019/03/05/design-results-roi-data/
Kalaignanam Kartik, Tuli Kapil, Kushwaha Tarun, Lee Leonard, Gal David. (2021), "Marketing Agility: The Concept, Antecedents, and a Research Agenda," Journal of Marketing, 85 (1), 35–58.
Mulcahy Simon. (2020), "Customer 360 Transformation: The Importance of Changing Mindsets in Your Organization," https://www.salesforce.com/blog/2020/01/360-perspectives-change-mindsets.html.
Schwab Klaus. (2016), "The Fourth Industrial Revolution: What It Means, How to Respond," https://www.weforum.org/agenda/2016/01/the-fourth-industrial-revolution-what-it-means-and-how-to-respond/.
~~~~~~~~
By Jason Wild
Reported by Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 23- Commentary: Governing Technology-Enabled Omnichannel Transactions. By: John, George; Scheer, Lisa K. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p126-130. 5p. DOI: 10.1177/0022242920972071.
- Database:
- Business Source Complete
Commentary: Governing Technology-Enabled Omnichannel Transactions
In "Informational Challenges in Omnichannel Marketing: Remedies and Future Research," [ 3]; hereinafter Cui et al.) discuss advances in information and communications technology (ICT) that promise better support for fractured customer journeys. Today's customers often engage a seller's offering via one touchpoint within a specific channel and then subsequently interface with a seller, the seller's agents, or resellers at different touchpoints within other channels. We may begin searching a category on Amazon, then investigate a brand at the manufacturer's website, visit a discount retailer to evaluate the product in person, obtain more information via online chat using the manufacturer's mobile app, and ultimately buy the product at a local store that offers free training. The central role of ICT in this process is self-evident in the explosion of websites, mobile apps, opportunities for chat, and online retailing.
Behind each touchpoint is an array of upstream firms that create and support the services a customer receives through that interface. Before and while a customer engages in the journey, a plethora of transactions take place between diverse sets of firms that produce, promote, and provide the goods and services accessed during the journey. These interorganizational networks are undergoing massive changes and facing new challenges spurred by omnichannel marketing and enabled by ICT advances.
The payday loan industry illustrates these disruptions. Historically, walk-in prospects at a storefront were converted into completed loan applications, which were then sent to precontracted lenders who processed them to make lending decisions. As online lending has grown, aggregators have emerged that pay other websites (aka "affiliate publishers") on a per-click basis to direct traffic to the aggregator's website. The aggregator converts the visitor into an applicant and then presents the accepted application for sale to potential lenders in an automated sequence, where each lender gets only seconds to accept or pass on the application. The lender who buys that application may or may not ultimately decide to offer a loan. Although the vast majority of applications are still rejected by lenders, industry volumes have expanded greatly; however, so have private lawsuits and regulatory investigations arising from controversial issues centering on automated, online transactions.[ 3]
Transaction cost analysis (TCA) offers a unique perspective from which to examine these changes and challenges. Drawing on the extensive research within this paradigm, we offer propositions regarding the efficient design and structuring of the upstream relationships within ICT-enabled omnichannels.
Transactions occur when "a good or service is transferred across a technologically separable interface" ([10], p. 1). Some otherwise profitable transactions are not undertaken because of the hazards the potential trading partner would incur, specifically ( 1) lock-in due to required investments in transaction-specific assets, ( 2) the need for revisions to initial agreements over time, and ( 3) undersupply of imprecisely measured output.
A firm's transaction-specific assets cannot be fully redeployed if the trading partner were to be replaced, creating lock-in to that partner. For example, a dealer is unlikely to implement a manufacturer's program to develop brand loyalty in downstream customers, because if the dealer were successful, the manufacturer could sell directly to those cultivated customers, in effect appropriating the dealer's investments. Dealers will not fully implement such a program absent some governance mechanism(s) to mitigate this hazard (e.g., contractually assigned dealer territories, up-front subsidy payments to cover their costs).
Agreements must evolve to be relevant for fast-changing environments, but postagreement adaptations can be gamed to disadvantage the partner. To illustrate, a manufacturer may try to change dealers' territory boundaries to adapt to growth or mergers among downstream customers, but dealers may suspect that the manufacturer might use the revision to capture a greater share of channel profit. Potentially profitable revisions will be forgone absent some governance mechanism(s) to mitigate this hazard (e.g., franchise buy-back provisions, territory override commissions that provide compensation for sales in dealers' territories regardless of attribution).
Without verifiable measures of trading partners' contributions, it is hazardous to implement performance-based compensation. For example, to encourage prospecting, a manufacturer may offer dealers greater compensation for sales attributed to dealer-identified leads. When multiple parties other than the dealer are involved in converting leads into sales, however, attributing credit is complicated. Dealers may fear that the manufacturer will self-servingly attribute some dealer-generated leads to manufacturer sources, thereby not fully compensating dealers for their efforts. Dealers will not fully support the lead development effort absent some governance mechanism(s) to mitigate this hazard (e.g., uniform commissions on all sales regardless of attribution of leads).
Marketing researchers have examined a plethora of governance mechanisms that vary in degree of hierarchical control between unfettered market exchange and the pure hierarchy of vertical integration, such as commission payments, royalties, component branding, dual distribution, exclusive territories, franchising, and alliances. According to TCA, parties craft governance mechanisms commensurate with the trading hazards present in a specific setting. They aim to minimize opportunity losses of forgone profitable transactions and the out-of-pocket costs of reaching and enforcing agreements. Broadly speaking, when hazards are minimal, market governance, which relies primarily on performance-based compensation and self-enforcement, suffices. As potential trading hazards increase, market discipline and self-interest enforcement are increasingly insufficient to mitigate the hazards, so more hierarchical governance is added, introducing administrative fiat, supervision, and/or flat payments replacing performance-based compensation.
Omnichannel journeys require sellers to make products available in multiple touchpoints through a wide variety of emerging channels. As Cui et al. document, ICT advances now enable new, separable interfaces (e.g., permissioned block-chains, federated and smart contracts), thereby providing new opportunities for firms to specialize in specific tasks and expanding the universe of potential transactions. Obtaining and maintaining a presence in the plethora of potential consumer touchpoints requires a provider to transact with new channel partners in novel ways. How might new omnichannel transactions be organized?
We examine these technologies from a TCA perspective, discussing transaction-relevant dimensions identified previously ([ 2]; [ 4]). We apply a TCA lens to link specific dimensions of these novel ICT applications to the magnitude of potential trading hazards, seeking insight on how governance mechanisms may mitigate those hazards in omnichannels. Our goal is to derive propositions regarding the efficient governance of transactions so that these promising ICT omnichannel solutions can be profitably implemented.
Modularity is the degree to which a technology enables an economic exchange to be subdivided into smaller modules. [ 4] discuss the relentless increase in modularity, from interchangeable machine-produced firearms components introduced during the American Civil War to today's internet technologies, enabling ever more specialization. Contemporary trends in ICT such as "virtualized" and "containerized" design loosen ties to specific hardware, enhancing modularity and specialization. To illustrate, early ICT solutions such as electronic data interchange required significant partner-specific investments in hardware and software. Over time, electronic data interchange applications became more generic and modular, with many of today's supply chain applications using browser-accessible client–server architectures that can be repurposed across trading partners.
Although modularity enables more trading partners and smoother, less costly transitions in switching partners, there is an inherent trade-off: the sacrifice of greater security of transactions and the customized applications that enhance unique performance aspects that can provide a distinct competitive advantage. For example, in the securities industry, high-frequency securities transactions prioritize execution speed over everything else, motivating firms to develop advanced ICT applications that rely on partner-specific infrastructures, comprising specialized hardware and customized software tied to that hardware. Similarly, in the digital advertising business, OpenX has colocated its ad exchange infrastructure adjacent to Google's data centers to reduce latency.[ 4]
Consider seamless entry and exit, a key feature of market-based governance, inherent in the blockchain technologies discussed in Cui et al. Bitcoin—the prototypical permissionless blockchain—offers a completely self-governed system that seemingly enables anonymous traders to forgo costly verification of trading partners prior to entering agreements by relying solely on unilateral exit to enforce agreements. Cui et al. posit that permissioned blockchains are superior because they benefit from the same verifiable ledger but also verify the participants' identities through permissioning authorities, such as IBM and Maersk, in the TradeLens global supply chains in which they are a member. They opine that TradeLens participants will voluntarily cede control to IBM and Maersk as the "trusted parties."
We note that permissioning significantly elevates participants' potential lock-in, as unilateral entry and/or exit is not available. Although modularity in itself reduces some vulnerability to opportunism, as it decreases the need to make significant transaction-specific investments ([ 2]), authority-based regulation of which participants may/may not enter enables the permissioning authority to use its power opportunistically to disadvantage potential competitors and thereby garner additional margins. For example, Maersk may use its access to data on a fast-growing participant to advantage itself or even to expel that participant. Such conflicts are reminiscent of Amazon's ongoing controversies regarding its use of data on third-party sellers to assist its own retail operations. Per TCA logic, TradeLens participants would seek mitigation of this potential trading hazard, but the smaller, weaker participants are in no position to bargain for contractual safeguards from powerful permissioning authorities. What might motivate a prospective TradeLens participant to have confidence that an authority such as IBM or Maersk is unlikely to exploit its position?
Empirical TCA-based research suggests that weaker parties may cede control to powerful trading partners, provided the exchange is supported by shared norms of trust. Benevolence-based trust may not be necessary if sufficient confidence can be achieved through the alignment of permissioning authorities' self-interests with the self-interests of blockchain participants ([ 7]). For instance, large, well-known firms as permissioning authorities may engender sufficient confidence because their brand reputations serve as "hostages"—in other words, they would incur losses in brand equity if it were publicly revealed that they abused their power. Our logic also implies that a new, unknown entrant would be hard-pressed to successfully introduce a permissioned blockchain system. Thus,
- Proposition 1 : Permissioned blockchains in omnichannels present greater lock-in hazard and require costlier, more hierarchical governance than permissionless blockchains that are governed via market-based self-enforcement (i.e., unilateral entry and/or exit).
Omnichannel transactions that are ICT-enabled have become nearly instantaneous, such as internet search engines that provide oceans of organized information at a click. Transactions among the firms producing, promoting, and providing goods and services are likewise accelerated. For example, producers can now access near real-time shelf prices for each stockkeeping unit at every store in a retail chain. Real-time information shrinks decision horizons and increases the frequency of adjustments in response to shifting customer preferences, market trends, competitor initiatives, and unexpected events such as Oreo's creation and placement of an ad during the 38-minute-long 2013 Super Bowl power failure.
Transaction cost analysis has long recognized the need to support postagreement adjustments. Where more frequent and significant ex post adaptations are anticipated, administrative fiat exercised within more hierarchical governance mechanisms lowers transaction costs by avoiding repeated bargaining over contract revisions that would arise in market governance. For example, digitally enabled programmatic ad buying empowered quicker, more frequent ex post adaptations, leading advertisers to bring these activities in-house. Thus,
- Proposition 2 : More frequent ex post adjustments arising from faster ICT-enabled omnichannel transactions align with the administrative adjustment processes of more hierarchical governance.
Firms prefer performance-based compensation of a trading partner wherever feasible because it provides better motivational benefits as well as lower administrative costs of market governance. Research on TCA has long documented that credible, precise measurement is the sine qua non to employing performance-based compensation. When performance-based compensation is based on less precise measures,[ 5] however, it invites gamesmanship and/or fraud. Pervasive fraud in pay-per-click online advertising settings illustrates the problem of compensation based on imprecise measures.[ 6]
Blockchains increase the granularity, timeliness, and precision of transaction-relevant data, thus appearing to improve the potential for performance-based compensation. Other prominent ICT advances that offer promise include real-time inventory reporting systems, supply chains offering "identity-preservation" of high-value products such as non-GMO grains, and the Internet of Things (IoT), which combines ICT with sensor-equipped machinery to enable real-time monitoring of equipment at client sites. One might expect that ICT advances such as these would enable more performance-based compensation.
However, TCA offers a more cautious appraisal of these opportunities. [ 5] show that not all types of more precise measures invite more performance-based compensation. Specifically, more precise output measures (e.g., sales) support greater use of performance-based compensation, whereas more precise measures of inputs or behavior (e.g., inventory levels) actually invite more flat wages, greater supervision, and monitoring of effort.
Consistent with this perspective, [ 8] found that when IoT-based performance compensation was included in business-to-business sales and maintenance contracts, disputes became more frequent and intense; the IoT infrastructure afforded more precise measurement of the machines' behavior, but not of the machines' output. Effective performance-based compensation thus depends on the type of precision afforded by ICT advances such as blockchains. Thus,
- Proposition 3 : In omnichannel transactions, more precise ICT-enabled output measures in omnichannel transactions align with greater use of performance-based compensation within market governance, while more precise ICT-enabled measures of inputs and behaviors align with flat payments under hierarchical governance.
[ 9] spawned a large stream of research on cocreation. Distributing a complex task across a multitude of participants via numerous transactions is enabled by ICT-supported integration, coordination, and interconnections among the participants and focal firm. In digitally enabled cocreation networks, a large number of voluntary participants each contribute a tiny part of the aggregate output.[ 7] In omnichannel settings, the most visible cocreation activities are the ubiquitous user-generated ratings and reviews. Any individual's evaluations constitute a single data point among thousands and provide only a miniscule contribution to the aggregate output. [ 2] proposes that neither hierarchies nor markets provide effective governance in such systems. The plethora of participants and permeable boundaries between voluntary participation and exit render hierarchical governance infeasible, but concurrently, the small size and scope of each participant's task also undermines the use of market-based financial incentives to motivate participation. Primarily, in this context, governance harnesses social rewards to motivate the vast majority of participants' contributions.
Nevertheless, some participants aim to monetize their efforts as expert reviewers, brand ambassadors, and so on. A growing number of firms proactively encourage these users to become influencers by offering financial incentives such as cash payouts or complementary products for review. Creating a governance structure to encourage both the socially motivated participants and the financially compensated mavens requires careful attention; providing financial compensation to any participant can undermine the social-motivation norm that spurs contributions from other participants. This is particularly problematic with very large user populations where the median participant's contributions are so small that only a tiny fraction can expect meaningful remuneration. Thus,
- Proposition 4 : More widely distributed cocreation in omnichannel transactions align with market governance, but with greater reliance on social rewards than financial rewards for participants.
In TCA, efficient governance choices depend on the exogenous legal, political, and social environment in which transactions occur. When the external environment does not adequately support contract enforcement in a predictable, low-cost manner, customers are wary of engaging in impersonal, digitally enabled transactions with geographically distant sellers. Stronger exogenous support invites more impersonal transactions and market governance, while weaker exogenous systems invite additional, more costly governance.
This substitutability principle is illustrated by contrasting Amazon in the United States and Alibaba's Taobao and TMall platforms in China, the largest online trading platforms in their respective countries. [ 6] describe how the weak Chinese legal system supporting commercial transactions motivated Taobao and TMall to develop their costly, internal online User Dispute Resolution Center. This quasijudicial, jury-like system reportedly had adjudicated over 2 million disputes as of 2018, using voting panels of 13 public assessors selected from millions of volunteers. Amazon offers no such adjudication despite access to the same ICT as the Chinese platforms. The U.S. legal system offers external support to enforce transaction agreements, whereas the Chinese legal system was insufficient to do so, consequently requiring the Chinese platforms to substitute their own costly enforcement system.
Some contend that ICT-enabled smart contracts can substitute for a weak external legal system, because they are enacted via immutable blockchain ledgers that verify participants' past actions and include auto-enforcement code (Cui et al.). We assert that the smart contracts discussed in Cui et al. exhibit a complementary relationship with the external environment. Smart contracts are self-enforcing, similar to a rental car key fob that is programmed to turn off automatically at the end of the rental period. They were first introduced to govern self-contained cryptocurrencies where the content of the transaction ledger is itself the object of the transaction (e.g., Bitcoin, Ethereum). However, as [ 1] documents, the decentralized autonomous organization episode shows that even these self-contained systems cannot replace external legal support. The decentralized autonomous organization was launched on Ethereum as a venture capital fund with thousands of investors contributing Ether cryptocurrency to secure voting rights on projects. In June 2016, a hacker siphoned off nearly one-fourth of the $250 million raised. Efforts to rewrite the Ether smart contract to invalidate the hacker's account illustrated the impossibility of simultaneously maintaining the intended immutability of the smart contract and the need to revise to unforeseen circumstances. Ultimately, after considerable acrimony and controversy, Ethereum Classic branched off from Ether because the original Ether contract could not be modified.
The challenges facing smart contract enforcement in omnichannel contexts are much greater. In contrast to self-contained cryptocurrency trading, omnichannel transactions exchange goods and services for remuneration that exists outside the blockchain ledger (Cui et el. 2021). Changes in the production of, provision of, and demand for goods and services require frequent revisions to the smart contract ([ 1]). For instance, unexpectedly low demand may necessitate an unforeseen increase in marketing program funds designated to dealers or influencers. Such revisions to initial contract provisions are virtually impossible to accomplish with auto-enforcement codes in smart contracts. Therefore, although smart contracts are promising devices for enabling novel omnichannel transactions, changing circumstances in the real economy where goods and services are exchanged invite ex post revisions; the necessary flexibility to make postagreement adjustments conflicts with smart contracts' auto-enforcement quality and intended immutability. Indeed, stronger legal system enforcement enables all but trivially simple smart contracts. Complex smart contracts pose greater ex post revision needs, indicating the necessity to rely on a strong legal system. Thus,
- Proposition 5 : Stronger external legal systems complement and thus enable greater utilization of smart contracts in omnichannel transactions.
Digital enablement of fractured customer journeys in omnichannel markets is fast becoming commonplace. [ 3] call our attention to important advances in ICT that further promote these trends. In our commentary, we expand on their response to these trends from a governance viewpoint that goes beyond resource allocation and competitive strategies and formulate propositions that suggest directions for further research.
Footnotes 1 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
2 Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
3 For a class-action lawsuit, see Rilley v. Money Mutual, United States District Court, District of Minnesota, Case No.: 16-cv-04001-DWF/LIB. The FTC reports and actions are available at https://www.ftc.gov/terms/payday-lending.
4 OpenX: Powering the Future of Advertising with Google Cloud (https://cloud.google.com/customers/openx/).
5 Many factors, including team production, crossover journeys, and multi-item purchases, can contribute to lower precision of measures.
6 Among the impediments to credible measurement of digital ads is the unwillingness of the major ad tech firms to allow organizations such as the Audit Bureau of Circulation to certify the measures.
7 [2] refers to this as granularity.
References Arruñada Benito. (2018), "Blockchain's Struggle to Deliver Impersonal Exchange," Minnesota Journal of Law, Science and Technology, 19 (1), 55–104.
Benkler Yochai. (2006), The Wealth of Networks: How Social Production Transforms Markets and Freedom. New Haven, CT : Yale University Press.
Cui Tony Haitao, Ghose Anindya, Halaburda Hanna, Iyengar Raghuram, Pauwels Koen, Sriram S., Tucker Catherine, Venkataraman Sriraman. (2021), "Informational Challenges in Omnichannel Marketing: Remedies and Future Research," Journal of Marketing, 85 (1), 103–20.
John George, Weiss Allen M., Dutta Shantanu. (1990), "Marketing in Technology-Intensive Markets: Toward a Conceptual Framework," Journal of Marketing, 63 (4), 78–91.
Lafontaine Francine, Slade Margaret. (2007), "Vertical Integration and Firm Boundaries: The Evidence," Journal of Economic Literature, 45 (3) 629–85.
Liu Lizhi, Weingast Barry R. (2018), "Taobao, Federalism, and the Emergence of Law, Chinese Style," Minnesota Law Review, 102 (4), 1563–90.
Scheer Lisa K. (2012), " Trust, Distrust and Confidence in B2B Relationships," in Handbook of Business-to-Business Marketing, Lilien Gary L., Grewal Rajdeep, eds. Cheltenham, UK : Edward Elgar Publishing, 332–47.
8 Shekari Saeed, Ray Sourav. (2018), " Monitoring Technologies and Performance Contracts for Multi-Component Systems," working paper, McMaster University.
9 Vargo Stephen L., Lusch Robert F. (2004), "Evolving to a New Dominant Logic for Marketing," Journal of Marketing, 68 (1), 1–17.
Williamson Oliver E. (1985), The Economic Institutions of Capitalism. New York : The Free Press.
~~~~~~~~
By George John and Lisa K. Scheer
Reported by Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 24- Commentary: "Half My Digital Advertising Is Wasted...". By: Pritchard, Marc. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p26-29. 4p. DOI: 10.1177/0022242920971195.
- Database:
- Business Source Complete
Commentary: "Half My Digital Advertising Is Wasted..."
In 1994, the first digital banner ad was launched, and the digital media revolution was born. What followed has been an extraordinary transformation of ways to connect with, entertain, inform, and engage consumers, with digital media disrupting every aspect of the advertising, media, and marketing ecosystem. Advertising formats have exploded. Worldwide digital media spending of $300 billion has surpassed television media spending ([ 2]). Over-the-top streaming services and ecommerce are growing exponentially, fueled by data and digital media.
There have been many positive benefits from this revolution. Creativity has expanded beyond the constraints of the 30-second TV ad, with new ways of expressing brand benefits and points of view. Anything discoverable can now be found instantly through search engines. Shopping for new products has never been easier or more frictionless through ecommerce. And we can connect with people and engage the consumers we serve in novel and entertaining ways we never thought possible.
But there has been a dark side to this revolution—most notably, massive inefficiencies resulting in media waste, outright fraud, and issues of brand safety as documented in [ 3], hereinafter Gordon et al.). As digital media grew, the marketing industry faced the inconvenient truth that it was operating in a nontransparent, murky, and sometimes even fraudulent digital media supply chain. Marketers and industry associations have taken steps to clean up this supply chain through multiple initiatives, but much more needs to be done to ensure it is truly efficient and effective.
More than a century ago, marketing pioneer John Wanamaker said, "Half my advertising is wasted; the trouble is, I don't know which half." At the time, newspaper advertising was the dominant form of media for marketers—over time, radio and TV broadcasting became dominant. When the digital media revolution began more than 25 years ago, marketers had high hopes that digital media would solve the persistent problem of advertising waste that has plagued the industry since the advent of mass marketing. With digital now the dominant form of media, substantial waste still exists, but there are approaches for addressing inefficiencies. This commentary provides an overview of the key problems facing marketers today and recommends possible solutions.
Viewability means the opportunity to see an ad. Did the ad make it onto the screen where human eyes can see it? Having a common standard for assessing digital ad viewability is important for conducting business transparently and comparatively across platforms. However, for many years, each platform created its own viewability metric, which was used to set payment terms for advertising. For example, Facebook considered an ad viewable and therefore "billable" for payment as soon as one pixel entered the screen, while YouTube considered an ad billable only after the entire ad was shown. Not having a standard meant Procter & Gamble (P&G) and other marketers wasted time and money trying to understand, analyze, and explain the differences between various viewability metrics claiming to be the right metric for each platform.
In response, the Media Ratings Council (MRC) proposed a common viewability standard for digital ads in 2014 ([ 5]). Display or banner ads would be considered viewable when 50% of pixels were on the user's screen for.5 seconds. Videos would be considered viewable when 50% of the video was on the user's screen for 2 seconds.
Although this standard had been developed and validated through thousands of studies and millions of dollars spent in research, digital platforms and publishers were slow to adopt, fearing that it would change how billable payment terms were determined and therefore potentially reduce revenues. Advertisers knew it was a minimum standard and, despite flaws, were willing to accept the standard in order to conduct business on a level-playing field across platforms and publishers. P&G set the expectation that all agencies, media suppliers, and platforms needed to adopt the standard and demanded that viewability be measured by a third party and accredited by the MRC. The Association of National Advertisers (ANA), which includes more than 2,000 marketing companies, also demanded compliance. Eventually, digital platforms and publishers developed systems to implement third-party viewability measurement and agreed to get MRC accreditation.
With this common standard and transparent third-party measurement implemented, substantial waste was exposed. P&G found that the average view time for digital ads was approximately 1.7 seconds, little more than a glance. We realized that we were spending far too much money on static digital display ads, which consumers were skipping past with little to no engagement. Consequently, we shifted substantial spending away from static digital display ads and into more effective media, including television. Although today we do not waste time debating viewability standards, we do know that one of the key sources of waste in digital media continues to be lack of viewability.
Every marketer should insist that every digital platform and publisher provide third-party, MRC-accredited viewability measurement. From that measurement, marketers can analyze the view time for digital ads to ensure they are being seen and are effective in driving awareness, engagement, and sales. If not effective, media money can then be shifted into more effective vehicles.
For media to be efficient, ads need to reach the intended audience at a minimum frequency (Gordon et al.). P&G proprietary research on daily essential hygiene, health, and cleaning products demonstrates that for a brand ad to effectively communicate its intended message, a consumer needs to view that ad at least one time a week to register awareness of the brand's message. At most, an ad generally needs to be viewed three times a week to lead to purchase. Ad frequency of more than three times a week is generally considered wasteful.[ 3]
In the digital media world, measurement of media reach and frequency was not commonplace for many years; if it was done at all, it was dependent on self-reporting from digital platforms and publishers. This was a major concern for advertisers, as it was the equivalent of "the fox guarding the henhouse." In 2017 at the Interactive Advertising Bureau Annual Conference, P&G called for third-party, MRC-accredited measurement verification of audience reach and frequency for all digital players. The ANA joined the call for compliance. After initial resistance, the digital platforms developed third-party measurement systems, and they are now increasingly committing to submit to MRC audits and accreditation to verify that the systems for measuring reach and frequency are reliable. This commitment is evident even in the largest platforms, termed "walled gardens" because they control audience data about billions of people, which, for consumer privacy reasons, they understandably do not share.
As these measurement systems have become established, substantial waste has been discovered. P&G found that brands were serving the same ad to the same person multiple times. This duplication wastes money and annoys consumers. In fact, a recent study by HubSpot indicated that 7 of 10 consumers say digital ads are annoying ([ 1]). That leads to digital ad blocking: up to 21.1% of the U.S. population and 25.8% of U.S. internet users now claim to use ad blocking tools to avoid viewing advertising completely ([ 4]). To mitigate these problems, P&G brands attempt to place limits on ad frequency within every digital publisher and platform. Unfortunately, many digital players, especially the "walled gardens," provide no access to consumer data for privacy reasons. For the most part, reach and frequency can be measured and optimized only within each individual walled garden, whether by MRC-accredited third parties or having their own first-party reporting audited by the MRC. However, any excess frequency can only be discovered after the waste has already occurred through third-party reporting. And until audits are completed to verify the accuracy of the measurement, we cannot be certain of the reliability of the reported results.
Marketers should establish clear reach and frequency targets for digital media to avoid annoying excess frequency and wasted spending. Marketers should always insist on third-party, MRC-accredited measurement of audience reach and frequency from digital media providers. The analysis of that measurement can reduce waste by reducing excess ad frequency.
Gordon et al. note that ad fraud continues to be a major source of digital media waste. Ad fraud occurs when ads are not served to consumers, but instead are served to "invalid traffic," often called "bots" that mimic human online activity on fraudulent sites where payment for ads goes to criminals. Digital ad fraud is estimated to be as high as 20% of all digital media spending based on multiple studies. It is rampant because criminals prey on big money—especially in inefficient, nontransparent systems with multiple touch points from many suppliers.
Digital media providers are responsible for detecting invalid traffic to either eliminate it or reimburse advertisers for fraudulent ad activity. In 2017, P&G, with support from the ANA, called for standard, third-party audits of digital media providers to ensure that fraud is being properly handled. Most publishers are conducting these third-party audits. However, the most dominant digital players do not, citing privacy reasons. Instead, they provide first-party audits, reporting that invalid traffic is almost nonexistent. Although this may be true, we worry that media money is still wasted on ad fraud because these big players are essentially "grading their own homework."
Marketers should continue to call for third-party, MRC-accredited validation of anti-fraud on all platforms and publishers, including the big digital platforms. Until that happens, we cannot be certain that marketers are not wasting money on fraudulent ads. At the same time, publishers can adopt best practices from the Trustworthy Accountability Group, the leading global certification program fighting criminal activity and increasing trust in the digital advertising ecosystem.
Persistent lack of transparency throughout the media supply chain is one of the main causes of digital waste and inefficiency. Ideally, advertisers would have sufficient information to buy media in real time and reach people on a more personalized basis, without annoying excess frequency and at a cost that creates value for all. Unfortunately, advertisers do not have sufficient information and are generally subject to what a media agency leader once bragged about when pitching the agency's capabilities to P&G: this person said they had achieved "information asymmetry"—whoever has the most information has an advantage. And year in, year out, media providers have the most information when it comes to buying and placing media—leading to waste through excessive ad frequency as well as persistent media price inflation.
In digital media, marketers face a great deal of information asymmetry, as the "walled gardens" control audience data about billions of people, which they do not share for consumer privacy reasons. Media pricing rates are generally not negotiated up front, because media is bought through an auction within each individual major digital platform. Through the ANA, marketers have asked for some form of data in order to transparently compare media choices across platforms. Marketers have asked for a data signal when an individual has seen an ad, in order to avoid serving too many ads to the same person. All requests have been denied due to privacy reasons. Consequently, marketers cannot be certain that we are getting the best value across alternatives, and we cannot be certain that we are not serving too many ads to the same person.
Marketers want a level playing field, where all players—digital and TV alike—participate in a cross-platform data measurement system to enable reaching consumers without excess ad frequency at a price that creates value. Although work has been done at the industry level to create a framework for such a system, it has been a slow process as there is little incentive for any of the media providers to implement this system because it reduces their "information asymmetry" advantage. Marketers should demand that validated pilots are implemented and ready to scale across the industry within the next year.
However, even if cross-platform measurement is implemented, it is unlikely to truly level the playing field because digital platforms have amassed such vast amounts of consumer data. That is why P&G is taking more control by buying more digital media through programmatic media buying across multiple ad exchanges available outside the major digital platforms. These ad exchanges allow automated, real-time buying across a marketplace of thousands of sites and publishers. In China, for example, nearly 90% of P&G media spending is digital, and more than 80% of that is programmatic. In the United States, programmatic spending is growing in double digits and is nearly P&G's largest digital media investment. We do not want to be dependent on a few and prefer to work with the many. So, P&G's preferred media providers are those with programmatic capability. It levels the playing field among thousands of media companies, enabling more to participate and to compete on effectiveness and value, not information asymmetry.
Harmful online content is another prevalent form of digital media waste. Harmful content includes posts, videos, and commentary that is not appropriate for brand advertising, such as violence, terrorism, nudity or pornography, hate speech, and denigrating or discriminatory content. Unacceptable content continues to be available online and is still being viewed alongside brand ads. It is bad for the consumer experience and it wastes billions of dollars in advertising because people pay little or no attention to ads that are next to shocking or harmful content. Those people who do notice the ads often lose trust in the brand because they associate that brand with the harmful content.
Progress has been made with the digital platforms, with agreement on common definitions for reporting through the Global Alliance for Responsible Media (GARM), but enforcement is challenging. Marketers need common metrics, such as the number and percentage of hateful content that exists on each platform and how much is eliminated before it is seen. Marketers need third-party auditing to objectively verify results. And marketers need ways to buy media so they can be certain that brands advertise in safe places, never on or near bad content—such as how YouTube worked with P&G to establish "safelisted" channels proven by a third party to be virtually free of bad content. As always, we fully support freedom of expression across a wide spectrum and recognize there is gray area on what constitutes hateful content, but we expect digital platforms to eliminate content that is plainly offensive to most.
In addition to GARM, marketers from the ANA and World Federation of Advertisers are meeting regularly with digital platforms to review commitments, plans, and timetables for eliminating harmful online content. But while reviews make a difference, marketers have been astonished to find that this problem still exists. Eliminating harmful content should be "table stakes" for any media company—whether a broadcaster or digital platform—and needs strict enforcement. Can we self-regulate? Yes, but that is not where we want to spend time. All marketers need to insist that digital platforms urgently apply content standards properly so we can spend time together on creating value. History has shown that when industries cannot sufficiently self-regulate, governments could step in. In fact, we are already seeing governments around the world doing just that. So, it is time for the platforms to get moving to eliminate harmful content.
John Wanamaker's famous quote about advertising waste is still valid today, more than a century later. The difference is, today we have a greater understanding of that waste along with potential solutions for eliminating it. Advancements in viewability standards, third-party measurement of audience reach and frequency, anti-fraud actions and audits, cross-platform transparency, and eliminating harmful content are helping marketers identify sources of waste, eliminate wasteful spending, and improve effectiveness. More progress is dependent on working together across the entire media supply chain—marketers, agencies, media platforms, and providers—to address persistent inefficiencies. With all of the brainpower, creativity, and technology available in the industry, marketers hope to find "which half of digital media is wasted" and make the digital media industry efficient and effective at driving growth and value creation.
Footnotes 1 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
2 Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
3 https://www.marketingscience.info/wp-content/uploads/2016/01/Institute-and-CNCB-Synergy-White%5fPaper.pdf
References An Mimi. (2020), "Why People Block Ads (And What It Means for Marketers and Advertisers)," https://blog.hubspot.com/marketing/why-people-block-ads-and-what-it-means-for-marketers-and-advertisers.
eMarketer (2020), "Global Digital Ad Spending 2019," https://www.emarketer.com/content/global-digital-ad-spending-2019.
Gordon Brett R., Jerath Kinshuk, Katona Zsolt, Narayanan Sridhar, Shin Jiwoong, Wilbur Kenneth C. (2021), "Inefficiencies in Digital Advertising Markets," Journal of Marketing, 85 (1), 7–25.
4 He Amy. (2019), "Ad Blocking Growth Is Slowing Down, but Not Going Away," eMarketer, July 26, https://www.emarketer.com/content/ad-blocking-growth-is-slowing-down-but-not-going-away
5 Media Rating Council (2014), MRC Viewable Ad Impression Measurement Guidelines, Media Rating Council, Inc. June.
~~~~~~~~
By Marc Pritchard
Reported by Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 25- Commentary: How AI Shapes Consumer Experiences and Expectations. By: Cukier, Kenneth. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p152-155. 4p. DOI: 10.1177/0022242920972932.
- Database:
- Business Source Complete
Commentary: How AI Shapes Consumer Experiences and Expectations
There is genuinely something known as artificial intelligence (AI). There is certainly something called consumer experience. How the two interact is of critical importance to marketers. Yet practitioners and scholars interested in where these two trends in business come together will miss essential issues from reading [ 3], hereinafter Puntoni et al.), including what is distinctive about AI and how it will be deployed in the world. Moreover, by relying on extreme exceptions rather than mainstream AI uses, the authors' view of customer experience overlooks more important everyday activities that are informative. It blames AI for problems that more reasonably attach to digital technology or business and society generally. And it overlooks the most fundamental aspect of AI: the technology will enable profound product and service improvements, thereby enhancing the consumer experience. As a result of the article's weaknesses, real and important issues are missed.
There are merits to their work as well. The article does a useful job of conceptualizing the consumer experience of AI into four areas: data collection, algorithmic output, delegating human tasks to AI, and social interactions with it. Several of the points raised are essential and deserve deeper consideration, which I return to subsequently.
This response first explains where the analysis by Puntoni et al. should be challenged or could be usefully extended. Then, it offers a different framing, underscoring what is beneficial and distinctive about AI, and how it may be deployed with respect to customer experience. It concludes by positing several novel issues related to AI and customer experience and highlighting several of Puntoni et al.'s points that merit further study.
The first dimension of the consumer AI experience is what is called "data collection" in international privacy law and most scholarship, which the authors call "data capture." They write: "Despite AI's ability to predict and satisfy preferences, consumers can feel exploited in data-capture experiences" (Puntoni et al., p. 163). The reason is that the collection is "increasingly intrusive" and consumers "are not aware of how this information is aggregated" (p. 164). Of course, the problem of customer awareness is not a shortcoming of AI per se, but in how the company communicates what data it collects and how these data power AI features; AI is only incidental.
However, the result, the authors argue, is "actual and perceived loss of personal control" that "can turn into demotivation and helplessness" (p. 165). This is extreme. We have now strayed far from AI and marketing. It is difficult to see how speech-recognition software or product recommendation systems have made us helpless. But the example the authors use clouds things even further: "the case of Leila, a sex worker, who shielded her identify on Facebook" but whose regular clients popped up on the "People You May Know" function.
This is an odd example, considering the harm is unrelated to data collection, and one can easily imagine a situation whereby the user was happy to connect with the community of clients. It is not clear why AI was at fault; most people find this feature remarkable and highly useful. The deeper problem, the authors note, is that it could be a privacy invasion for political activists and others. But it is difficult to make this case for a service that people use on their own volition, whose terms of service require people to use their real names, and that is specifically designed to connect to other people.
The second dimension of the AI customer experience is the result of the algorithmic processing, which the authors call "classification" (a classifier is one of several types of machine-learning systems). The fear here is that customers may be "classified as a certain type of person" (p. 168), which will hurt their self-image. Once again, AI is portrayed in a negative light. It need not be. The system could misclassify the user as rich and glamourous, thereby improving their self-image. It is not even clear that this is a real harm: people may be classified as they truly are.
Moreover, a misclassification is not so much a problem with AI but marketing, which has always played with people's sense of self, from the need for Palmolive to get rid of "dishpan hands" (in the 1960s through the 1980s) to poor "Marvin," whose handsome looks, wealth, and intelligence were ignored by ladies because he did not use Listerine for his bad breath (in 1929). Before there was AI to bash, people fretted that supermodels made ordinary women feel inferior. The point is that the harm is from the classification, not AI. And classification communications can boost self-esteem. For example, Google famously ran abstruse recruitment ads in the early 2000s that had hidden math problems for the cognoscenti to catch and apply for jobs.
The chief harm from AI processing that the authors identify is bias and discrimination (specifically, "feelings" and a "perception" of discrimination, as this discussion is in the context of customer experience). This is a real worry: AI, even when it is accurate—perhaps especially when it is accurate—may have the appearance of discriminatory behavior. Marketers will need to be transparent about the systems to alleviate concerns.
The third area is AI delegation: having the technology handle a task that humans once performed. This should be an area where most people will cheer. The point of technology through the centuries is to hand to the machine that which can be automated so that humans can do things that are not "repeatable," in the phraseology of the scholar Roberto Unger of Harvard Law School. But to the authors, such delegation is "replacing" humans with machines and threatens to "deprive humans of the sense of accomplishment" (p. 172) The examples Puntoni et al. use are fishing with GPS maps and Google's sentence auto-complete feature: the authors found one fisherman who takes umbrage at the assistance and an old-school journalist who is angst ridden about computer-generated suggestions. Never mind that fishing services have stocked ponds for years, and thesauruses line writers' bookshelves: AI is too seductive a target to avoid blame.
The bigger point is that the analysis grossly misunderstands customer expectations and how this affects usage. There is typically a divergence between what customers say they want and the "revealed preference" of how they act. People may grumble about a product, but if they continually use it, we can assume they in fact like it. So just because consumers express unease with AI does not mean they will not find it beneficial and stick with it. The authors cite the journalist John Seabrook of The New Yorker fretting about Google Smart Compose that completed his sentences as he wrote. But such AI delegation may be a boon for those who are not professional writers and precious about their words.
What is important is that the experiences of the anti-AI fisherman and anti-AI writer are outliers. Rather than generalize from their experience, it would be more useful to understand how the majority of people who use AI technology regard delegation. They are probably grateful for the productivity gain, reduction in toil, and increase in economic growth. Puntoni et al. earnestly suggest that "outsourcing labor to machines prevents consumers from practicing and improving their skills, which can negatively influence self-worth" (p. 172). Yet this is difficult to substantiate. After all, the authors probably do not feel diminished because they used a computer instead of a quill, or electric lighting instead of candles—unless they regret not "practicing and improving their skills" of wax dipping.
There is a broader problem with the criticism of AI delegation: it anchors the baseline on human capability. Yet the essence of AI is to exceed human capabilities. That is crux of the AI revolution, exemplified by achievements in image recognition, the ancient Chinese board game Go, and six-player No-Limit Texas Hold'em poker—all areas where AI has beat humans. Hence, arguing that AI should let people do the work to keep their "self-worth" misunderstands the technology and stunts it. It also ignores the question of agency: customers choose to use the technology in the first place.
A central issue of delegation (which might presumptively be regarded as a good, not bad, thing) is the matter of control and the perception of control. The two are different. We will want AI at times to take control because it will do things better. (For example, supermarkets could have human door-openers, but the automated versions work just as well.) Yet in some customer interactions, we will want to give people the perception of control, so they feel a sense of autonomy.
The example of how Betty Crocker cake mix was deliberately designed so people did a basic level of preparation to give the illusion of classical baking is extremely apt. And the idea the authors present (from André et al.) that self-driving cars should allow for "customization of peripheral features" is shrewd. In fact, one can build on this, by looking across the consumer AI landscape and asking: Where do companies already take active steps to provide the perception of control—or where could they?
For instance, when online advertising networks provide a link and the message "Why did I see this ad?" or "Report this ad," it gives consumers a sense of agency. Even if they do not exercise it, they still may appreciate it, enabling the relationship to change from unwitting recipient to active participant in AI. As the consumer AI industry evolves, this may become one of the most essential areas of study and a practical way that companies can win consumer acceptance.
The fourth and final dimension of the AI customer experience is the social one: when customers know they are interacting with an AI (such as with Amazon Alexa) and when they do not initially know (e.g., a company chatbot or an automated scheduler like "Evie" that is not obvious or forthcoming). The problem, the authors note, is that anthropomorphic products and services based on AI may exhibit traditional biases and prejudices, notably sexism. Indeed, the preponderance of AI digital assistants with female names suggests something truly awful about the technology industry, as well as society generally. This alienates customers, the authors note. And it rightly should.
But the problem is not with AI but how companies implemented it. Tech firms scrambled to add masculine names and voices to their services to avoid just that charge of social stereotyping. However, the author's argument is broader, that people feel "discomfort" when reacting with "social robots."
This does not hold up, considering that people seem happy to use the Google search engine—a service that used to require a secretary or travel agent or librarian (i.e., a human social interaction). Instead, the authors justify the claim with the example of the discomfort experienced by a person with learning disabilities and dyslexia. It raises the question whether AI is really to blame, or if the person might face similar unease in an analog setting. Once again, an exceptional case is used to make a broader point, which portrays AI in a negative light at the expense of a more balanced view of AI's good and bad aspects relative to customer experience.
There is a more useful way to understand the issue of AI and consumer experience, but it requires reframing the issue. First, we need to understand what AI is to tease out the unique attributes that affect the customer experience. Second, we need to see the relationship between the technology and customer as a trade-off and value exchange.
First, the technology. At its origin in the 1950s, AI referred to using computers to do tasks that were once only done by people. By the 1990s that meant everything from statistical machine translation and chess-playing computers to so-called expert systems that could make deductions based on inputs, such as diagnosing illnesses or managing supply chains. However, progress was limited. A radical change in approach took place over the past two decades that affects what is meant today by AI.
The classical method of hand-coded, logic-based AI systems did not scale, and performance improvements were meager. But the technique of machine learning (relying on statistics rather than rules) produced impressive results because more data were used to train the algorithms. It infers the right answer rather than having it explicitly programmed, but that means the outputs are probabilistic and nondeterministic. Moreover, the systems are not instructed with what variables are relevant prior to the analysis; they are able to tease out the relevant inputs, a process called "feature extraction." For this reason, the systems work best when they have "all" the "raw" data (i.e., high-dimensional, unstructured data, in AI lingo), rather than a curated set of processed data—in other words, imputing as little human judgment as possible tends to yield the best results.
This is what is meant by AI that companies are using as the foundation of their products and services, which consumers access today. It will produce novel and beneficial customer experiences and pose unique challenges to marketers. A germane feature is that AI is not something one buys directly; it is an "intermediate good," in the parlance of economics. Like electricity or microprocessors, it powers the thing but is not the thing itself. Customers experience it, but only as it is embedded in products to improve their performance. AI will perform tasks better than a human in terms of speed, scale, and often accuracy—and thus, costs.
This framing, depicting what AI actually is, challenges the four basic tenets of Puntoni et al. The "data capture" enables the systems to work, not act as a form of exploitation. The "classification" improves product quality, rather than diminishes people. The "delegation" is precisely why the customer chose to use the product or service. The social interaction with AI is simply a matter of establishing the right interface, which is a common problem with all technologies.
The consumer experience with AI will entail data collection, algorithmic analysis, delegation, and possibly social interaction. But underlying this is a value exchange, particularly for free web services like Google or Facebook. It is a trade-off. Data power the service, and customers are not passive subjects but agents—after all, they agreed to use the service or product. They are rational beings capable of evaluating benefits and choosing freely. This idea of a value exchange and agency in the AI customer experience is wholly missing from the article.
In fact, AI may at times remedy the harms that the authors put forward. For example, they worry that "data capture" and "classification" may exploit, alienate, and be biased; but there are cases when it overcomes discrimination. When Amazon Go stores opened in 2018 with face-recognition payment systems, the experience of many customers was not one of menace but convenience. A black woman explained that she felt liberated: throughout her life she was always tense in stores, told by her parents to never pick up an item she did not intend to buy, make sure the store clerk always sees you, get a receipt, etc. The Go store did not have potentially biased human clerks. AI was a step to overcome racism, not perpetuate it.
What makes AI unique in regard to marketing is not articulated in Puntoni et al. There are, however, attributes that are indeed novel and require consideration. They can be broken down into four categories: data, value, explainability, and servitization.
First, data. Unlike traditional data analysis based on classical statistics, AI needs all the raw data to work very well. The systems perform best when the engineers use the fewest assumptions about what the most effective variables will be. Often, it is several seemingly unrelated variables in unison that will improve the system's performance. But this is different than how classical privacy law works, in which processors (i.e., a company) need to request the consent of data subjects (i.e., the consumers) and the data must be both specified and limited in scope.
How can we have a truly high-performing AI economy under these constraints? The answer is we cannot. So, a new arrangement needs to be established that wins customer acceptance. Businesses, academia, media, and even regulators need to educate consumers why collecting the data is necessary.
Second, value. As noted previously, a more useful way to frame the interaction between company and consumer is a value exchange. But markets need information and transparency in order to work well, and the nature of the value exchange is cloudy. Machine-learning algorithms require troves of data, and consumers need a way to see the value they receive. Some information has such limited value that customers have no problem giving it away for free (e.g., clicks on Google search results feeding back into the ranking algorithm, Gmail seeing what spelling variants are corrected versus kept to improve its spellchecker).
But other interactions entail high-value data, such as shopping information. How the industry provides customers with transparency on the value exchange still needs to be worked out. But just framing it as a value exchange, not a taking, is part of the solution. One reform proposed by technology analyst [ 1] is that major web platforms like Google and Facebook be required to offer a paid product devoid of advertising, as a way of giving consumers choice about how their data are used. It would make explicit the value they get and let them choose.
Third is explainability. That is the term used in AI to describe the fact that the most sophisticated systems are based on such intricate networks of correlations that it is unknowable how they reached an answer even if we can validate the degree to which the answers are correct. This prevents advanced AI from being used in certain industries like health care, insurance, and finance, where full transparency of algorithmic decisions is required. But it creates havoc for other sectors as well, if the output decisions cannot be clearly explained.
A trade-off will be required in society between our interest in understanding and the performance improvement from AI. It will be the job of marketers to communicate the benefits of forgoing explanation for performance. Most people will probably come to accept it, in the same way they do not understand how their computers or phones work. In this case, though, the reason for the AI's output is actually unknowable. But in most use cases—such as why a given exercise routine is best via an AI fitness tracker, or why a language-translation earpiece chose a particular word—convenience will win out.
Last is servitization (with apologies for the gangly neologism). It describes the effect of AI on many everyday consumer items, insofar as it transforms them from stand-alone products into services, by dint of giving them "senses" and "intelligence" to render judgments. An example is a map. It used to be a paper product. But because of digitization and AI, maps have become services, showing users the most effective route and updating in real time based on traffic flow. In some cases, the business model changes from a single sale to recurring subscription income (e.g., a digital watch becoming a fitness tracker or health monitor).
Marketing professionals will need to reorient their work to support this commercial transition and help customers understand the new features and value. As people rely on their products to "do" things not just "be" things, product use entails a degree of intimacy with the item and trust with the company. Explaining its features and giving consumers more control will be essential. Google does this with its "incognito mode" and "clear search history." Likewise, consumer AI companies will need make an explicit virtue of promoting these sorts of tools.
The implications of these trends are profound. Soon, AI will force companies to answer the classic question posed by [ 2] of Harvard Business School: "What business are you really in?" It will force companies to make some counterintuitive replies, and marketing will need to support the strategy. Firms will need to redefine their markets. For example, to improve product performance, innovate on features, and develop new product categories, companies will need to have a larger data footprint of customer activities. This will encourage the formation of large "AI conglomerates" and "AI keiretsu" (tight networks of independent firms) that span multiple sectors. It will also create incentives for tighter integration among products and services, at the expense of third-party innovation by outside firms.
A taste of this is already playing out with the tech giants. But the same logic will apply to firms in nondigital industries. AI-centric firms will understand customer needs better, develop products more efficiently, and innovate faster. Marketers may move higher into the C-suite in the mid-twenty-first century, just as finance directors did at the end of the twentieth century.
We are in the "egg timer" phase of AI, in which for all the fancy potential uses of the technology, we still mostly ask it to replicate what we did in previous settings with less performant technologies, like having Alexa play music or time a pot of boiling water. But this will change. When it does, AI will become as commonplace as electricity, the microchip, or plastics. Getting the marketing of consumer experience right will be crucial for success.
Footnotes 1 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
2 Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
References Keen Andrew. (2018), How to Fix the Future. New York : Atlantic Monthly Press.
Levitt Theodore. (1960), "Marketing Myopia," Harvard Business Review, 38 (July-August), 45–56.
3 Puntoni Stefano, Reczek Rebecca, Giesler Markus, Botti Simona. (2021), "Consumers and Artificial Intelligence: An Experiential Perspective," Journal of Marketing, 85 (1), 131–51.
~~~~~~~~
By Kenneth Cukier
Reported by Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 26- Commentary: Inefficiencies in Digital Advertising Markets: Evidence from the Field. By: Porter, Jonathan. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p30-34. 5p. 1 Chart. DOI: 10.1177/0022242920970133.
- Database:
- Business Source Complete
Commentary: Inefficiencies in Digital Advertising Markets: Evidence from the Field
The Competition and Markets Authority (CMA) is the United Kingdom's primary competition and consumer authority. It is an independent, nonministerial government department responsible for carrying out investigations into mergers, markets, and regulated industries as well as enforcing competition and consumer law. The CMA's statutory duty is to promote competition, both within and outside the United Kingdom, for the benefit of consumers.
As described in [ 6], hereinafter Gordon et al.), the CMA carried out an in-depth review of online platforms and digital advertising in the United Kingdom between July 2019 and June 2020. In addition to considering the advances and benefits for consumers from online platforms and digital advertising, the review explored whether practices in those markets could have an adverse effect on consumers. It also considered what steps could be taken to remedy, mitigate, or prevent any adverse effects. The resulting report not only presented the CMA's findings but also made a series of recommendations to government to inform its deliberations on future regulatory structures for online markets in the United Kingdom ([ 3]). The CMA's work on online platforms and digital advertising has sparked interest from a range of commentators and has already been used as the basis of articles making the case for competition enforcement against both Google and Facebook in the United States (e.g., [12]).
Gordon et al. lists four areas of market inefficiency in digital ad markets—measurement of ad effectiveness, internal frictions, ad blocking, and ad fraud—and suggests that these areas have received less attention from various regulatory and competition reviews. In fact, the CMA did explicitly consider the theoretical and practical challenges of measuring ad effectiveness in the United Kingdom (see [ 3]; Appendix O) as part of a broader analysis of the challenges of assessing and evaluating the quality of digital advertising. The CMA approached the assessment of the quality of digital advertising in terms of a process involving several discrete stages: ( 1) verification: checking the viewability of the advertising, the context in which it was displayed, and identifying the potential for ad fraud; ( 2) attribution: tracking what actions the consumer took after being exposed to the ad; and ( 3) measuring effectiveness: did the advertising meet the campaign objectives the advertiser had set? For instance, did it produce an incremental uplift in sales?
On the key theoretical and methodological issues relating to measuring effectiveness, the CMA reached findings similar to Gordon et al. We also identified issues related to verification and attribution, which were relevant to the operation of digital advertising markets in the United Kingdom.
In this commentary, I begin by considering why measurement matters for effective competition in digital advertising markets. I then pick up from Gordon et al. and consider the extent to which the experimental techniques for the robust measurement of ad effectiveness proposed in that article have actually been adopted by U.K. advertisers. I present data collected by the CMA showing that the level of experimental testing being carried out in the United Kingdom is currently very low. In exploring why this is the case, I find that this is due to internal frictions (e.g., considerations of time and cost) rather than lack of access to data or deliberate actions on the part of the digital advertising platforms to impede measurement of ad effectiveness. I then discuss challenges related to the verification and attribution of digital advertising. Here, the CMA did find that lack of access to data and actions on the part of Facebook and Google were having an adverse impact on U.K. advertisers' ability to properly and independently evaluate the quality of the digital advertising inventory they were buying. I briefly summarize the CMA's recommendations to address these findings and to improve the functioning of digital advertising markets in the United Kingdom. Finally, I conclude with some suggestions as to where Gordon et al.'s research agenda could usefully be supplemented.
Competition analysis often focuses on the supply side of a market—that is, the extent to which firms compete for customers. However, the CMA examined the issue of measuring the effectiveness of digital advertising from the perspective of whether advertisers were able to evaluate the quality of the digital advertising inventory they were buying. If advertisers are not able to measure ad effectiveness—as a dimension of quality—then they will not be able to make well-informed purchasing decisions, and this will undermine effective competition in digital advertising markets.
The CMA report starts from the position that advertising campaigns have the characteristics of experience goods ([ 8]). That is, it is possible to specify a clear set of objectives (e.g., in terms of reach, conversions) for a campaign up-front, but it is only possible to evaluate effectiveness after a campaign has been purchased and run. In theory, the availability of significant volumes of user-level data in relation to digital advertising and the ability to track subsequent purchasing behavior should mean that it is now possible to measure the effectiveness of digital advertising much more precisely compared with other advertising media. However, it could be argued that advertising campaigns are, in fact, credence goods in that even after the campaign has been run, in practice it remains difficult and costly to measure quality—and even then, the results may be imperfect.
I offer three main observations regarding measurement challenges in digital advertising markets. The first is that the CMA's discussions with advertisers and media agencies reveal that they recognize the importance of focusing on the incremental impact of advertising. Furthermore, they are also aware of the risks associated with standard observational methods outlined by Gordon et al., and they recognize the superiority of experimental approaches and randomized control trials—often viewed as the "gold standard" of measuring the advertising effectiveness ([11]).
The second observation is how closely the leading online platforms (i.e., Google and Facebook) have been involved in the academic research and the development of new experimental approaches and how quickly these platforms have made tools for carrying out experiments available to advertisers. For instance, both Google and Facebook offer "Conversion Lift" tools to measure the incremental impact of advertising using experimental methods. Google's Conversion Lift tool is based directly on the "Ghost Ads" methodology described in [ 7], and Facebook's equivalent tool is an intent-to-treat experimental approach. This speed of adoption is illustrated in that the first reference to the Ghost Ads methodology appears in Google's marketing literature in 2015 ([ 4]), indicating that a tool for advertisers was being developed in parallel with the developments in the academic research. The speed of adoption suggests that platforms saw a distinct competitive advantage in the ability to demonstrate the efficacy of advertising on their platforms to advertisers.
The third observation is that although advertisers report that they are aware of the measurement challenges and have the tools available to carry out robust measurements, the amount of advertising expenditure actually being exposed to experimental testing in the United Kingdom is very modest. Table 1 presents data on the number of U.K. advertisers using Google and Facebook's Conversion Lift tools over the period 2017–2019, together with the amount of advertising expenditure that has been subject to testing using these tools.
Graph
Table 1. Use of Conversion Lift Testing in the United Kingdom: Google and Facebook.
| Google Conversion Lift Tool | Facebook Conversion Lift Tool |
|---|
| Year | Number of U.K. Advertisers | U.K. Ad Spend Tested | Number of U.K. Advertisers | U.K. Advertiser Spend Tested |
|---|
| 2017 | 0–10 | £100,000–£200,000 | 100–150 | £10–£20 million |
| 2018 | 30–40 | £600,000–£700,000 | 900–1,000 | £20–£30 million |
| 2019 | 10–20 | £400,000–£500,000 | 700–800 | £40–£50 million |
1 Notes: Exact numbers have been redacted for reasons of confidentiality. Source: Tables O.1 and O.3 of Appendix O ([ 3]).
The modest levels of testing should be set against a backdrop of increasing digital advertising expenditures—the CMA estimated that Facebook accounted for over half of the £5.5 billion spent on online display advertising in the United Kingdom in 2019—and technological advances on user tools that make it easier to automate experiments. Clearly not all advertisers would make use of experimental approaches, as such approaches are only likely to be relevant to larger advertisers and, even then, principally for their larger, conversion-focused campaigns (as opposed to longer-term brand awareness campaigns). Furthermore, Google does not support the Conversion Lift tool on Google Search—its main product. Even so, the amounts being subject to testing are extremely modest.
Gordon et al. suggest that advertisers could make more use of experimentation across geographic areas. In fact, the CMA found that Google already appears to put an emphasis on geo-experiments—experimentation across geographic areas—as opposed to user-based experiments. For instance, Google is reported to have expertise in using geographical experiments to measure the causal impact of increased advertising ([ 9]), and geo-experiments are now its standard tool for the causal measurement of online advertising ([ 2]).
As part of its review, the CMA asked Google for information on the extent to which U.K. advertisers were using geo-experiments instead of Conversion Lift testing. Although U.K. advertisers are using geo-experiments more than Conversion Lift tests, the scale of use still appears to be very modest. Google has countered this suggestion by arguing that advertisers could be carrying out geo-experiments independently of Google's geo-experiment tool.
Following [13], [ 6] suggest that internal frictions between, for example, the marketing and finance departments could be another source of inefficiency. Given that advertisers seem to be aware of the measurement challenges and that they have the tools available to them, the CMA was interested in understanding the range of factors—both external and internal—that could explain the modest amount of online advertising that was subject to experimental testing.
In terms of external factors, the data indicate that advertisers are running more user-level randomized control trials (RCTs) on Facebook than Google. This appears to be due to a clear understanding of the relative advantages that Facebook has over Google in this area. [ 5] described Facebook's ability to track users via its "single-user login" across devices and sessions as representing a significant measurement advantage over more common "cookie-based" approaches. The single-user login means that Facebook can not only associate all exposures and conversions across devices and sessions with a particular user but also maintain the integrity of the random assignment process that is crucial to an experimental approach. The ability to identify and track a user across different devices is critical because users might be exposed to advertising on one device (e.g., a mobile phone) and subsequently convert on a different one (e.g., a desktop computer); thus, Facebook's single-user login appears to give it an advantage in terms of measuring ad effectiveness. Indeed, one media agency described the Facebook platform as being the most advanced for RCT-based testing.
In terms of internal factors that could affect advertisers' propensity to make use of experiments, discussions with media agencies suggest that reasons why RCTs are not frequently deployed include ( 1) costs (i.e., advertisers want to maximize the amount of money going into the advertising itself); ( 2) lack of time—advertisers need time to analyze and apply the insights from an RCT to other campaigns before carrying out retesting; ( 3) suitability—not all campaigns are suitable for RCTs (particularly branding campaigns); and ( 4) difficulties in carrying out RCTs outside of walled garden platforms.
In the course of its review, the CMA found that advertisers and agencies were aware of [ 1] study, which raised the possibility that, under certain circumstances, digital advertising might actually be ineffective. As a result, one might have expected advertisers and agencies to be more interested in measuring ad effectiveness to ensure that their advertising budgets were being spent efficiently. However, discussions with agencies suggested that Blake, Nosko, and Tadelis's study is typically viewed as an outlier; instead, agencies emphasize that subsequent research had been able to demonstrate that digital advertising does have a positive impact on driving conversions—albeit at a more modest level than was perhaps originally claimed.
The evidence available to the CMA suggests that advertisers and agencies are aware that RCTs represent the "gold standard" in terms of measuring advertising effectiveness, and they have the tools and access to the data to carry them out. However, in practice, it seems that many regard them as just one of several "tools in the bag" for evaluating the effectiveness of an advertising campaign. The CMA's overall impression was that there is a wariness to relying heavily on any one single metric—whatever its academic credentials—and instead, advertisers and agencies prefer to triangulate across different metrics. As a result, they might make use of experimental approaches to test the effectiveness of a limited number of key campaigns, but they continue to use observational methods—albeit with caveats—as well as other proxies to assess the delivery of a campaign (e.g., click-through rate, cost per click, cost per action). Each approach (experimental and observational) is perceived as having pros and cons, and multiple stakeholders voiced the opinion to the CMA that there was "no single point of truth" when it came to measuring advertising effectiveness.
The CMA regarded the assessment and evaluation of the quality of digital advertising inventory as a process involving not only measurement of the advertising's effectiveness in terms of outcomes but also the verification and attribution of the delivery of the digital advertising. In its review, the CMA examined concerns that large platforms had the ability to obstruct or place restrictions on advertisers accessing and assessing the data they needed to make a proper, independent evaluation of verification and attribution of the advertising inventory they were purchasing. The CMA noted that there had historically been concerns about the misreporting of data, including data on viewability, by Facebook ([10]), together with concerns about brand safety (i.e., companies do not want their advertising to appear on websites with content at odds with their brands) in relation to Google.
In terms of attribution, the CMA noted concerns that actions by Google had already made attribution by third parties more difficult: for instance, no longer sharing DoubleClick user IDs. Google's proposals to block third-party cookies in its Chrome browser in the future could also make third-party attribution still more difficult because it will constrain the ability of independent analytics providers to track consumer behavior across different websites.
The CMA was concerned that if advertisers are unable to independently assess the relative merits of advertising across different platforms and are forced to rely on the measurement tools of platforms that had significant market power, then they could be overpaying for the advertising inventory supplied by those platforms and misallocating their advertising expenditure relative to other sources of supply. Furthermore, if advertisers are forced to rely on data and metrics provided by those platforms, it could hinder other platforms from demonstrating that they can offer a competitive alternative, thereby running the risk of undermining effective competition.
In the case of verification for viewability and brand safety purposes, the CMA found that large, walled garden platforms such as Google and Facebook did have the ability to obstruct or place unnecessary restrictions on the advertiser's ability to access the data needed to carry out a proper, independent evaluation of the advertising inventory owned by those platforms. It was not clear to the CMA whether the data involved in verification of viewability and brand safety had to include personal data. For instance, the CMA understood that for viewability, verification involved determining whether the ad was served; whether the ad appeared on the screen; how much of the ad appeared on the screen; for how long the ad appeared on the screen; and, if the ad was a video, whether and how long the ad played and whether the sound was on.
The CMA found that, by restricting full independent verification of their own inventory, Facebook and Google had introduced a degree of opacity into the buying and selling of advertising of their own inventory. It is clear that the buying and selling of display advertising is already a complex process and platforms introducing additional restrictions on access to data will add to the "friction" in terms of evaluating market outcomes.
In the case of attribution, there were two related concerns about the exploitation of market power. First, platforms were removing or preventing access to the underlying user data necessary for attribution, which made it more difficult for third parties to implement their own attribution solutions. Second, and in parallel, platforms were increasing customers' reliance on the analytical products and services the platforms themselves offered. Without the ability to carry out independent attribution, there is a risk that advertisers are tied to using Google and Facebook's own evaluation tools. As a result, they could be paying higher prices for advertising purchased from those platforms and so misallocating ad spend relative to other sources of supply.
The CMA noted that Facebook and Google (and others) have made available services such as data "clean rooms" to advertisers to allow them to analyze campaign performance in a controlled environment. However, feedback from media buying agencies indicates that this form of data analysis is still a nascent area. At present, the use such services by advertisers appears to be very limited. and there are concerns that it is not possible to carry out analysis at the level of individual users. The ability to extract data from the "clean room" environments is also considered a constraint on the usage of this tool.
The CMA's analysis of digital advertising markets in the United Kingdom highlighted several areas where the operation of competition did not appear to be as effective as it could be owing to actions by the major platforms. Together with a range of other concerns about the entrenched market position of major platforms funded by digital advertising (e.g., Google, Facebook), this led the CMA to recommend a new, procompetition regulatory regime to govern the behavior of such platforms. The intention was to address the concerns about the current position of digital advertising platforms while promoting more competitive outcomes.
The CMA has proposed that a new regulatory body in the form of the Digital Markets Unit (DMU) should have the ability to ensure that platforms with market power, such as Google and Facebook, do not engage in exploitative or exclusionary practices or use practices likely to reduce trust and transparency; in addition, the DMU should have the authority to impose financial penalties if necessary. This would involve a mandated Code of Conduct involving three high-level objectives that the platforms would have to meet: ( 1) the Fair Trading objective addresses concerns about exploitative behavior on the part of the regulated platforms, ( 2) the Open Choice objective addresses the potential for exclusionary behavior, and ( 3) the Trust and Transparency objective ensures that regulated platforms provide sufficient information to users so they are able to make informed choices. Under the Trust and Transparency objective, the CMA specifically recommended that platforms that were subject to regulation should allow advertisers access to the data and tools necessary to allow for independent third-party verification of viewability and brand safety on the platforms' owned and operated inventory. The CMA also suggested that the DMU should have the power to make specific interventions related to the sharing of data to address competition concerns and to improve the efficiency of digital advertising.
The CMA's review has highlighted the importance of advertisers' access to data to independently evaluate the effectiveness of their digital advertising on an ongoing basis to support the operation of effective competition in digital advertising markets. Building on the CMA's findings, and to better understand issue on the demand side of digital advertising markets, the extensive research agenda in [ 6] could usefully be supplemented with further research to explore whether there are significant differences in outcomes across countries. For instance, are there significant differences in the adoption of experimental testing between the United Kingdom and the United States? Or is the experience in the United Kingdom common to other countries? If there are differences, what are the factors—either internal or external to the firm—that drive those outcomes? Do U.S. advertisers face similar issues in relation to independent verification for viewability and brand safety purposes as U.K. advertisers? Are there examples of countries where advertisers believe they have access to sufficient data to make robust, informed assessments of the effectiveness of digital advertising? If so, is that experience easily transferable to other countries? These and similar questions could shed further light on this important topic for marketers, for the competitiveness of digital advertising markets and for the benefit of consumer welfare.
Footnotes 1 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
2 Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
References Blake Tom, Nosko Chris, Tadelis Steven. (2015), "Consumer Heterogeneity and Paid Search Effectiveness: A Large-Scale Field Experiment," Econometrica, 83 (1), 155–74.
Chen Aiyou, Au Timothy C. (2019), "Robust Causal Inference for Incremental Returns on Ad Spend with Randomized Paired Geo Experiments," (November 26), https://arxiv.org/abs/1908.02922.
3 CMA (2020), "Online Platforms and Digital Advertising: Market Study Final Report," (July 1), https://assets.publishing.service.gov.uk/media/5efc57ed3a6f4023d242ed56/Final_report_1_July_2020_.pdf.
4 Google (2015), "A Revolution in Measuring Ad Effectiveness," Think with Google (May), https://www.thinkwithgoogle.com/intl/en-gb/marketing-resources/data-measurement/a-revolution-in-measuring-ad-effectiveness/.
5 Gordon Brett R., Zettelmeyer Florian, Bhargava Neha, Chapsky Dan. (2019), "A Comparison of Approaches to Advertising Measurement: Evidence from Big Field Experiments at Facebook," Marketing Science, 38 (2), 193–225.
6 Gordon Brett R., Jerath Kinshuk, Katona Zsolt, Narayanan Sridhar, Shin Jiwoong, Wilbur Kenneth C. (2021), "Inefficiencies in Digital Advertising Markets," Journal of Marketing, 85 (1), 7–25.
7 Johnson Garrett, Lewis Randall A., Nubbemeyer Elmar I. (2017), "Ghost Ads: Improving the Economics of Measuring Online Ad Effectiveness," Journal of Marketing Research, 54 (6), 867–84.
8 Nelson Philip. (1970), "Information and Consumer Behavior," Journal of Political Economy, 78 (2), 311–29.
9 Owen Art B., Launay Tristan. (2016), "Multibrand Geographic Experiments," (October), https://arxiv.org/pdf/1612.00503.pdf.
Peterson Tim. (2017), "FAQ: Everything Facebook Has Admitted About Its Measurement Errors," Marketing Land (May 17), https://marketingland.com/heres-itemized-list-facebooks-measurement-errors-date-200663.
Poynter Cassidy, and Duckworth (2014), "The Expert Guide to Measuring Not Counting. How to Evaluate Social Media for Marketing Communications," #IPASocialWorks (accessed October 6, 2020), https://ipa.co.uk/media/7347/ipa_socialworks_expert_guide.pdf.
Scott-Morton Fiona M., Dinielli David C. (2020), "Roadmap for a Digital Advertising Monopolization Case Against Google," Omidyar Network (May), https://www.omidyar.com/sites/default/files/Roadmap%20for%20a%20Case%20Against%20Google.pdf.
Simonov Andrey, Rao Justin M. (2019), "Firms' Reactions to Public Information on Business Practices: Case of Search Advertising," Quantitative Marketing and Economics, 17 (2), 105–34.
~~~~~~~~
By Jonathan Porter
Reported by Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 27- Commentary: Managing Human Experience as a Core Marketing Capability. By: Lieberman, Scott. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p219-222. 4p. DOI: 10.1177/0022242920973035.
- Database:
- Business Source Complete
Commentary: Managing Human Experience as a Core Marketing Capability
Today's marketers face innumerable challenges to meaningfully engage consumers and entice them to start and stay with their brand. [ 2]; hereinafter Morewedge et al.) have added a formidable challenge to the mix as they lay out an argument for how digitization has transformed consumer perceptions of ownership and the implications for marketers. While psychological ownership is important, other more central objectives such as improving customer engagement, deepening loyalty, and sustaining competitive differentiation is where marketers are focused in the face of digitization.
The authors imply that the shift from legal ownership to legal access and the shift from material goods to liquid experiences are equally important. Based on my practical experience, having worked directly with more than ten chief marketing officers and their teams, I propose that the shift from "solid" material goods to "liquid" experiential goods is a broader concept and deserves more weight in considering how to deepen perceived ownership in the eye of the consumer. The shift to legal access is more a result of the shift to experiences than it is a driver of change.
Additionally, Morewedge et al. argue that experiences are replacing material goods (e.g., shifting car ownership to car sharing). I propose that experiences are not only replacing material goods but also being used to enhance material goods. The authors somewhat acknowledge this implication, as they say firms are making "investments in servitization" that involve embedding products into experiences (p. 000), but it is a minor part of their framework. In my experience with established businesses navigating the change, it occupies much of their attention.
Moreover, while the authors identify tactical marketing demand creation actions, for firms to fully realize market opportunities resulting from the shift to experience, marketers should also consider the importance of building internal capabilities to manage this shift. From my experience, marketing leaders continually trade off capability investments in skills, organization, and process with demand generation activities. My commentary lays out a framework to help marketers build these internal capabilities as delivering human experiences becomes increasingly important in the face of digitization.
This moment in time—marked by a global pandemic that is abruptly shifting consumer behavior—has created an unprecedented opportunity to accelerate digitization. A recent study found "seventy-five percent of CEOs say the pandemic has accelerated the creation of a seamless digital customer experience."[ 4] Simultaneously, the pandemic has heightened the need for companies to focus on the human aspects of the experience by demonstrating transparency, trust, and empathy in times of uncertainty.
The shift to human experiences that merge digital with physical, personalization with unlimited options, and ease of access with privacy and security gives marketers a wide array of options to craft experiences that create compelling value for consumers, while offering effective and efficient engagement models for brands. To guide the development of internal capabilities to deliver human experiences, I propose marketers adopt four design principles: ( 1) adopt a Comprehensive approach to experience design, ( 2) create Open experiences, ( 3) foster experience Resiliency, and ( 4) apply data-driven insights to enable Evidence-Based experiences. These principles, which I label the CORE Human Experience framework, provide a model to guide marketers' thinking on this topic.
The experience begins before the consumer engages with a particular product, during the discovery and search step, and develops as the brand relationship evolves through sales, service, and beyond. Comprehensive experiences designed to create value for the consumer and the company should contemplate all aspects of the discovery, purchase, and engagement cycle. End-to-end design also entails seeking, capturing, and acting on customer feedback throughout the life cycle to enable brands to act with empathy and build trust.
One of the more visible examples of how experience is moving from fit-for-purpose interactions to comprehensive engagement is the rise of super apps ([ 1]). Made popular by apps from the East like WeChat and more recently evidenced by the efforts of major retailers like Target and Walmart to consolidate their mobile experiences, these applications give consumers and brands an end-to-end experience from discovery, to purchase/payment, to service and loyalty ([ 3]).
They not only make the experience more efficient for consumers, but also raise the expectations for marketers to design experiences that keep and grow customer relationships. One of these expectations is that marketers will act on feedback across the life cycle, which requires a strong discovery and coordinated response capability. Consider a fictitious high-end retail furniture company that delivers furniture to five customers over the course of two months. In four of the five deliveries, the customer reports the piece arrived warped. The fulfillment team immediately orders replacements, which are delivered without question; however, the incidents have affected customer satisfaction and eroded trust, and the fulfillment team does not provide the feedback to the marketing and merchandising teams. If the marketing and merchandising team had an experience feedback and discovery process, the team would have learned that upon further investigation, the way the supplier was recommending pieces be stored in the retailer's warehouse was causing the furniture to warp.
Effectively delivering comprehensive experiences requires marketers to extend their influence beyond the funnel to the postsales experience. This shift reflects both a mindset and role shift for leading marketers to engage throughout the customer life cycle and develop strong relationships with colleagues in sales and service. One way this can be accomplished is by embedding marketing expertise into sales and service functions through coordinated recruiting, skill development, or rotational programs. Another option is to design sales and service planning processes to have regular input from marketers.
An open experience is one that is easily shaped by partners and internal stakeholders. The digitization of experiences provides the opportunity to apply a similar approach to experience design as has been applied in software development—from closed proprietary systems to open-sourced common standards. Designing experiences to common standards can improve marketers' ability to create more connected experiences that quickly integrate with partners.
Creating open experiences requires capabilities to be engineered to integrate with brands in the ecosystem. When brands first took on this mandate, innovation involved the use of simple APIs to support integration between specific experiences. Consider major airlines' early partnerships with ride-sharing companies to allow passengers to book a ride using the airlines' mobile applications. This capability offered convenience but did not meaningfully differentiate the experience. Today, entire platforms have emerged that are designed to offer brands the ability to personalize interactions with the aim of creating greater differentiation, while allowing marketers to easily link experiences. These types of open platforms will migrate further into the consumer experience, streamlining the development of human experiences.
Designing insurance experiences that can integrate with ecosystems offers marketers the opportunity to create more value for consumers. The market for homeowners' insurance can serve as an example in which marketers can apply an open approach. The rapid rise of IoT (internet of things) technology adapted into the home to support home automation and home security gives insurance companies a wide variety of ways to differentiate products and personalize solutions. For example, being able to offer homeowners premium discounts if they install water and temperature sensors throughout their home to provide early warning of floods or frozen pipes requires that insurance companies develop experiences that easily connect to home automation technology partners.
Developing open human experiences implies marketers need to be partnership savvy, not only with traditional digital advertising partners, but also with emerging digital engagement platforms. Sourcing, developing, and managing external partnerships with ecosystem platforms will become a core competency of leading marketing organizations. Some ways this could be accomplished are to establish a focused partnership team within marketing or to expand traditional co-marketing partnership teams to include digital platforms and experiences. Another option to quickly develop strong relationships with digital engagement platforms is by establishing cross-functional teams spanning information technology, product development, and marketing to focus on alignment with external partners.
Building resilience into experiences means that critical elements of the consumer engagement cycle can be changed, updated, and moved to more effective delivery channels, as needed. Resiliency has historically been applied to operations—providing organizations the ability to continue to operate in the event of disruptive events. A similar philosophy can be applied to designing the experience to create resilience even in times of regular operation. For example, retailers could implement buy online/pick up in-store promotions for a specific set of stores—versus systemwide—in response to promotional item demand in a local geography.
Service experiences are ripe to be redesigned with resiliency in mind. For example, retail banks have considered ways to more personally service mass affluent customers to expand relationships, but the cost to serve that market at the same level as private banking clients is often prohibitive. Given the new acceptance of virtual work and consumers who are more willing to engage via video conference in their homes, banks could service clients through a team of service professionals who receive correspondence, engage digitally, and truly get to know the customer wherever and however the customer chooses to interact. Creating such service "pods" reflects a new level of resilience that simultaneously lowers cost to serve and increases customer engagement.
Another way to build resilience is to leverage immersive technology. A shift from predominately static (e.g., words, pictures, video) content to immersive content applying extended reality technologies (the combination of augmented and virtual reality) has emerged as a popular way marketers can create more human-like digital experiences. Early adopters were beauty and fashion brands that have rolled out virtual try-ons, thus alleviating the need to make a trip to the store to check out a specific look. Warby Parker was one of the first major brands to introduce an extended reality virtual try-on experience for its made-to-order prescription glasses. Sephora introduced similar technology to allow consumers to use a photo of their face to create a skin tone match and find the perfect makeup color. These immersive experiences are designed to increase the likelihood the consumer will complete the purchase through a fully digital experience with lower odds of returns. They also provide greater resilience for brands to continuously learn from what the customer is or is not doing as circumstances change.
Creating experience resiliency has direct implications for how design is considered in the marketing planning and execution cycle. As experiences become more flexible, user experience design will need to shift from a downstream marketing execution activity to become more central to the strategy and planning function. Marketers can use the experience itself to create differentiated marketing actions, as in the Sephora example, where the intent of the virtual try-on is to increase customer conversion. One way marketing organizations may change to accommodate this shift is to build technology-savvy design teams or form partnerships with the design and technology teams within their organization.
Marketers can shift from simple data-driven campaigns, for which there is a correlation between a consumer action and response, to responding to actual behavior, so that marketing meets consumers where they are in their lives. To that end, experiences can be designed based on evidence consumers provide through their observed behavior. The ability for a brand to "notice" a pattern, behavior, or a change is an important aspect to deliver human experiences.
If every time I take a trip from New York to Chicago during the week, I book the trip on my corporate credit card, stay at the same downtown hotel, and eat out at similarly priced restaurants, this pattern could be noticed and my travel experience anticipated and enhanced by any number of brands that participate in this routine experience. Rather than asking consumers to provide specific preference information, the ability for marketers to use evidence—while honoring privacy and security—of specific patterns will create opportunities for innovation.
Home entertainment offers an example of how evidence-based approach can be used to increase customer engagement and offer marketers new opportunities. Consider the latest streaming devices from Amazon, Apple, and Google that serve to integrate entertainment options from multiple streaming services. Personalized recommendations are presented in an integrated interface as these services learn which type of entertainment options appeal to customers. Each member of the household can even have his or her own customized interface and options based on both stated preference and evidence-based usage patterns. Going forward, savvy marketers could apply an evidence-based design approach that integrates customized content—perhaps a personalized highlight reel from last night's games of their favorite teams or a personalized preview of first-run movies—based on observed behavior.
Another example of evidence-based experiences is the use of dynamic data-driven interactions enabled by AI to provide marketers with the ability to create "always on" campaigns that dynamically respond to specific consumer situations, rather than predefined campaigns that marketers initiate on a specific schedule (see [ 4]). For example, consider the rapid growth of voice assistants in the home. In many ways, these artificial intelligence (AI)–based devices are becoming consumers' personal agents, constantly on the lookout for consumers' interests and preferences. Amazon's September 2020 announcement that it would enable its home assistant Alexa to participate in a conversation brings AI fully into the daily human experience. Alexa uses multisensory AI, including natural language programming, acoustic, linguistic, and visual cues in a conversation involving two individuals by identifying what questions are directed toward her and responds by presenting offers based on the information gathered during the interaction.
Marketers can adjust their approach to brand building by shifting from specific point-in-time programs to an evidence-based method that allows the customers' actions to illuminate the path toward greater engagement and loyalty. This approach aligns marketing objectives with consumers' implied needs. Savvy marketers can take personalization to the next level by applying a data-driven design model early in the strategy and planning process. The implication is that traditional marketing analytics functions may need to shift to not only include traditional funnel metrics and customer segmentation insights, but also maintain sophisticated analytical skills that can monitor and discern changes across the entire experience life cycle and provide quick access to strategy, design, and execution changes.
Technology has provided marketers with increasingly powerful ways to reach consumers. Digitization of products and services amplifies the need for marketers to become more technologically savvy. Marketers need to be able to market products and services that are increasingly digital, while simultaneously acquiring the technical skills to manage the marketing function, all while delivering experiences that create value. This is a tall order indeed.
As a result, new capabilities are required to build modern marketing organizations. These capabilities cut across skills, organizations, and processes. The CORE human experience framework gives marketers one tool to consider as they contemplate these new capabilities.
Footnotes 1 Author's Note This article represents the views of the author only, and does not necessarily represent the views or professional advice of KPMG LLP.
2 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 KPMG 2020 CEO Outlook: COVID-19 Special Edition (https://home.kpmg/xx/en/home/insights/2020/09/kpmg-2020-ceo-outlook-covid-19-special-edition.html, emphasis added).
References Huang Andrew, Siegal Mitch. (2019), "Super App or Super Disruption?" KPMG, (June), https://home.kpmg/xx/en/home/insights/2019/06/super-app-or-super-disruption.html.
Morewedge Carey, Monga Ashwani, Palmatier Robert, Shu Suzanne, Small Deborah. (2021), "Evolution of Consumption: A Psychological Ownership Framework," Journal of Marketing, 85 (1), 196–218.
Perez Sarah. (2020), "Walmart Grocery Is Merging with Walmart's Main App and Website," TechCrunch, (March 5).
Puntoni Stefano, Reczek Rebecca, Giesler Markus, Botti Simona. (2021), "Consumers and Artificial Intelligence: An Experiential Perspective," Journal of Marketing, 85 (1), 131–51.
~~~~~~~~
By Scott Lieberman
Reported by Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 28- Commentary: Music's Digital Dance: Singing and Swinging from Product to Service. By: Griffin, Jim. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p223-226. 4p. DOI: 10.1177/0022242920972704.
- Database:
- Business Source Complete
Commentary: Music's Digital Dance: Singing and Swinging from Product to Service
One good thing about music, when it hits you feel no pain.
So hit me with music, hit me with music now.
—[ 1], "Trenchtown Rock"
When Bob Marley performed "Trenchtown Rock" in 1971, he likely did not think of the song as a material object. His music partners and investors almost certainly did. To them, he created property that fans would want to purchase and make their own, or make "MINE," as [ 3]; hereinafter Morewedge et al.) put it.
I particularly enjoyed reading "Evolution of Consumption: A Psychological Ownership Framework" and commend the authors for a thought-provoking, fresh, and bold scholarly examination of technology-driven changes to consumption. Few industries experienced change in the relationship between consumers and their goods more than music, and because music is my field, I will focus my commentary there. The authors' main claim is that "preserving psychological ownership...should be a priority for marketers and firm strategy." It is a very interesting proposition, and I explore what it means for music and venture into some of the business-side complexities involved.
The paper proposes two ways consumption is changing. First, legal possession of private goods—for music, the most recent example is consumers' owning compact discs—is succumbing to on-demand access to the goods and services owned by others—for music, the digits that constitute a streamed phonorecord. Second, experiential goods are displacing material goods. Is that so for music?
Historically, music began as a performed service. It became a physical product with the invention of crude wire/string recordings and progressed through the first digital medium (piano rolls), to Edison's wax cylinders, to RCA/Victor phonograph records, and then to cassettes and compact discs. Now, it mostly appears in subscriptions to digital carrier services. Truly, music has traveled from service to product and back to service, adapting to new technologies as they appeared.
Music performances have changed, too. Acoustic became electric in the 1920s; before electric, artists controlled their music with their feet (if they were not in the room, you could not see them or hear them). Electricity brought loudspeakers and crowds bigger than could be counted from the stage. Radio and television followed with audiences that could only be sampled by media marketers; today, we can and do count and account for each stream delivered.
In 1971, sound recording products were a trickle by comparison to the flood that began in 1972 when they attained U.S. copyright status. Five decades later, the product would return to a service, with physical sound carriers an afterthought—a relic bin of compact discs, vinyl, 8-track tapes, cassettes, and Edison cylinders. In this latest twist, music fans buy access to music through subscriptions or ad-supported "feels free" access, totaling more than 110 million paid subscribers in the United States alone as of this writing. The just-in-time arrival of streaming digits on demand—where and when you want them—conveys the ownership feel of an otherwise unobtainable collection of music, replete with lyrics, photos, and liner notes.
As to the murkiness of rights, for centuries the authors of songs gleaned their experiential and printed value. In 1847, composer Ernest Bourget dined at the Paris Concert Café Ambassadeurs and refused to pay the check, reasoning that the Ambassadeurs played his music without paying him. A decade later, a French appeals court gave composers the right to collect payment for music use, which led to the formation of Agence Centrale for the joint administration of performance music rights, the predecessors of ASCAP, BMI, and SESAC in the United States today.
This is a useful backdrop for considering Morewedge et al.'s point that consumers' psychological ownership of a product or service is a valuable asset to consumers and to firms. The music industry has done a good job maintaining that feeling and responding to the associated threats, transfers, and opportunities noted in the article, perhaps because music consumption is inherently experiential. There are ways, perhaps unintentional, that music providers already support psychological ownership; for example, through playlists and reproducing in digital form some aspects of the physical form (cover art, lyrics). Morewedge et al.'s article is full of other ideas (see Table 3 in particular) that the industry could adopt to thrive in this new business environment.
What does it mean to own music in a digital age, and what rights come with paid or sponsored legal access? Lines blur as creativity moves from the center of the network to its edge. Where creative endeavors were once the province of centralized broadcast television and radio networks and the music companies that sold discs, today's creative production is often sourced from backyards and video corners.
Spotify, for one, just moved to permit the use of full-length tracks in podcasts uploaded to its network. This offers subscribers more MINE, a powerful advantage in both making and using content—and getting paid along the way.
Morewedge et al. propose that music is evolving from legal ownership to legal access, but to apply the proposition to music calls for nuance. To the fan, artist, production firm, or investor, to say the evolution in music is riding a trend toward "legal" access does not help. The problem is that there is no consensus yet on what makes access legal for many of the uses to which music is put.
Furthermore, fragmented ownership abounds, starting with the bifurcated structure of music ownership, with the author (songwriter/publisher) controlling the song and the musicians/performers/record company owning the sound recording. Today's songs can have 30 or more writers, each with an interest in declaring ownership.
Disputes over ownership and uses for music products are as old as music itself, as demonstrated in the Bourget example. Today, discs are used for performances but are not licensed for them, nor is their use permitted outside ambiguous terms around a vague concept of fair use. Likewise, your right to share them—whether within your household (allowed) or over a global network (not allowed)—is murky.
Many mistakenly think they can show Netflix on the restaurant television or play Spotify instead of a bar's jukebox. This confusion now transfers to Twitch streamers, many of whom think their legal access grants then ownership rights and go on to share the content, discovering too late that there are legal consequences—much as many a newly married couple finds out too late that their photographer owns the photo negatives (and the right to use them, having fixed them into the medium). Does sharing a song on a social network cede the rights? The Magic 8-Ball says "cloudy, but maybe."
Five years after "Trenchtown Rock," the U.S. Copyright Act of 1976 formalized the idea of MINE, declaring that as soon as Marley "fixed the words into a medium sufficiently permanent or stable to permit it to be perceived, reproduced, or otherwise communicated for a period of more than transitory duration," it could be owned, and investors could buy it from him. The song became material. The industry could distribute and sell it on sound carriers like vinyl, cassette tape, or compact disc. Broadcasters could transmit it to fans on airwaves, more service than product, to generate advertising revenue.
Now that more than 35 years have passed, the Copyright Act allows Marley (or his heirs) to say the song is MINE again. It reverts to him and once again becomes his property. And each time other artists "cover" the song by making a sound recording of their own performance of it (which they can do without permission under Section 115 of the Copyright Act), more property is made, which the cover artists own and which they in turn can sell.
There are many other areas of music consumption in which we do not know how rights in the music service economy will play out. To date, there is little consensus on the rights that attach to music used in a live-streamed video on a social network, artists performing their own music live, using music in background, or disc jockeys mixing music and streaming it. When podcasts include music, they operate in a domain that, although over 15 years old, still has no systematic licensing. Do users infringe when they copy media? Record from streaming services? And what of buffering and caching music for use in a time and place other than when it is offered? The right to buffer is not settled law and may not ever be, and these rights may differ among 150+ sovereign countries. This limbo presents firms and consumers with something I have called "Tarzan Economics" and Will Page wrote about in his upcoming book ([ 4]): we must let go of one jungle vine before we can see or feel the weight or arc of the new vine.
Make no mistake about it, fans use music for many purposes beyond their private consumption, especially when they use it to make more media, much of it collage from existing media available—but not legally cleared—for noninfringing use. "Ripping" remains an issue: consumers clearly do think it is MINE.
The Morewedge et al. article observes that firms are making (implicitly should be making) "significant investments in servitization and experiential offerings." Here too, the experience of music adds nuance to the advice. A question at the heart of music's experience with the shifts from legal ownership to legal access to experience is how to make money.
Ultimately, product pricing's buck-a-song led to the death of music sales, a price point that was untenable for filling portable music playing devices that can hold over 100,000 songs. Debundling the complete album into individual songs cost music dearly. Flat-fee digital buffets rebundle price while filling an important need to empower new, old, and unusual forms of music.
Copyright's bucket has always proved leaky, especially for music. The current state of music's digital availability makes it somewhat voluntary to pay: not legally, not contractually, not morally, but nonetheless a choice that renders product payment a sort of tip jar.
This is worsened by the fact that those who pay for music services are more interested in the service than the music, the ultimate form of medium over message, form over function. Fancy that song? YouTube likely has it, and numerous live versions and covers, all recordable. If not, another site will fill the need.
Civilized society cannot long tolerate purely voluntary payment for art, knowledge, or culture. Nor does it find that it has to. As new technologies drive music from its old product forms to new experiential forms, music is finding new ways to capture value. For example, the game platform Fortnite— which itself "feels free" because purchases are not required and convey no in-game advantage—thrives on selling music, clothing, and other merchandise. In April, rapper Travis Scott delivered five prerecorded concerts within Fortnite that drew an audience of 27.7 million and premiered a virtual concert track, "The Scotts," which was launched on streaming services a week later. There is much to learn from cross-pollinating business models and media. This exploration is in its infancy.
Music's creators are fighting music services for access to the data generated by the use of music, vital information not always made available to artists; it is vital because marketing efforts depend on it as the key strategy to pour gas on sparks. At the United Nations/World Intellectual Property Organization conference on Global Digital Content Markets in September 2020 in Geneva, Switzerland, many artists were motivated to express their desire for legally mandated access to the music data from services and from the intermediaries that stand between the creators and the markets for music.
Fans have different concerns about music and media data. Essentially, where once artists sold music, now they sell fans by serving as a gatekeeper to them. Observation and tracking can have a chilling effect on trying new media, especially music, so there must be limits on reporting and retaining information akin to those that surround library records. Fortunately, music payment's actuarial nature can function without granular analysis, just as it does in broadcast media environments.
Copyright is among those rights, like patents and trademarks, created by the hand of government. It is a new kind of property, as foreseen by [ 5], p. 733) in his seminal work:
The institution called property guards the troubled boundary between individual man and the state. It is not the only guardian; many other institutions, law, and practices serve as well. But in a society that chiefly values material well-being, the power to control a particular portion of that well-being is the very foundation of individuality.
Just as copyright law turned a creator's work into property, so, conceivably, may laws granting us rights over our personal data turn these data into property. Will private data become subject to a new property right? The answer is yes if lawmakers yield to demands that you own the data you generate, or have a right to access it or control its sharing. The hand of government will judge, but customers may do so first and impose their fickle values as they will by avoiding opening their wallets or devoting time off their clock to protect their data from potential privacy incursions. We may find ourselves as businesspeople wishing for a government prescription or global treaty that makes this thorny headache go away. I doubt this will happen and believe it will remain a contested balance.
In the case of music, new models need new shoes, a rhetorical call for distinctive terms that contrast the changes in operation, especially so when tangible objects manufactured and inventoried are obviated by friction-free, just-in-time, and customized digital delivery.
Music once consummated relationships with consumers ad seriatim without so much as finding out their names, rarely identifying and developing a relationship with a fan. Reliable surveys regularly found that 70% of fans did not know when their favorite artist released a new album of music, so disconnected was the product relationship. Today's music customer could not be more different than those from 20 years ago, and we cannot yet know the changes that will result.
Canadian media futurist Marshall McLuhan warned in the 1960s against drawing conclusions about the media of the times we live in, preferring instead retrospective views of media:
We look at the present through a rear-view mirror. We march backwards into the future....Because of the inevitability of an environment during the period of its invention, man is only consciously aware of the environment that has preceded it. In other words, an environment becomes fully visible only when it has been superseded by a new environment. Thus, we are always one step behind in our view of the world. ([ 2], pp. 74–75)
For the designers of strategy, this is not reassuring.
The authors of this paper have made good guesses at predicting effects of new models, but the verdict is out on what new vine, in the language of Tarzan economics, music marketers should grasp. It is too soon to tell, and much too soon to gauge the effects on creators, fans, and music's economy. Strategic counsel at such a time must contain a good measure of optionality.
As consultants and advisors, we are called upon for solutions, for paths and outcomes that absorb our clients' uncertainty. For music, and perhaps many other industries, five rules apply, in my opinion:
- Walk the tightrope. The case of music shows that there is a tightrope to be walked: cultivate enough psychological ownership that fans will pay for access, but not so much that they come to believe they own what they access and can freely share it with others.
- Power comes from aggregation. The power of music service networks grows exponentially with the number of users, as Metcalfe's Law observes, and so the strategic principle here is to pursue user scale. In the trade-off between the breadth of a fan base and the depth of its sense of ownership, first favor breadth.
- Laws change, and you can change them. Intellectual property rights are, like statues, creatures of popular government, not natural rights. The people's will manifests itself in ways that change over time. Where government creates property, it can take it away. Where the world once feared photocopy machines or audio/video recorders, it now sees fewer absolutes and more opportunities. Mark Cuban led Congress to a webcast license for music—access and ownership may be negotiable.
- The challenge of friction-free delivery. The Tarzan swing from product to service is transformative in basic ways that challenge traditional management, and even the law (digits know no borders). Where physical distribution was the result of matching supply to demand at extraordinary opportunity costs, this burden is lifted with just-in-time global digital delivery--the death of distance and inventory management.
The center of the network is giving way to its edge—the users. Basic communications that served "Channel We" to the masses now enable "Channel Me" from the opposite direction. User-generated content coming from the edge now competes for the spotlight.
A heady mix of change lies ahead. If you were dreaming of upsetting the apple cart, now is the time. Disruption is afoot and Morewedge et al.'s fine article is very well timed.
Footnotes 1 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
2 Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
References Bob Marley and the Wailers (1971), "Trenchtown Rock," Single Record, Trojan Records.
McLuhan Marshall. (1967), The Medium Is the Massage. New York : Bantam Books.
3 Morewedge Carey, Monga Ashwani, Palmatier Robert W., Shu Suzanne B., Small Deborah. (2021), "Evolution of Consumption: A Psychological Ownership Framework," Journal of Marketing, 85 (1), 196–218.
4 Page Will. (2021), Tarzan Economics: Eight Principles for Pivoting Through Disruption, London : Simon & Schuster.
5 Reich Charles A. (1964), "The New Property," Yale Law Journal, 73 (5), 733–87.
~~~~~~~~
By Jim Griffin
Reported by Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 29- Commentary: Omnichannel from a Manufacturer's Perspective. By: Ailawadi, Kusum L. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p121-125. 5p. DOI: 10.1177/0022242920972639.
- Database:
- Business Source Complete
Commentary: Omnichannel from a Manufacturer's Perspective
[ 3], hereinafter Cui et al.) offer a comprehensive understanding of why the information that marketers need to become omnichannel is so difficult to come by and how new technologies and new developments in data analytics can help. They highlight the fact that the challenges are even more daunting when information is needed across channels that are "external to the firm" than when the firm owns its channels. Yet, other than the advertising context, much of the discussion is drawn from the latter case, that of retailers or vertical brands that own their distribution channels. This is to be expected because omnichannel is primarily a retail concept (e.g., [ 1]; [10]).
But how is omnichannel relevant outside of retail? I aim to complement Cui et al. by taking the perspective of manufacturers who market products to end consumers through independent channel partners some or all of whom may want to be omnichannel, and possibly also through their own direct-to-consumer (DTC) channel. I make three points related to omnichannel from the manufacturer's perspective and offer some important managerial issues and research implications to go with each.
In omnichannel work, it is customary to include communication (not just distribution) channels among the touchpoints across which the customer experience is managed, so Cui et al. are not alone in covering both under omnichannel marketing. Indeed, [10], p. 3) state that it is important to broaden the scope of channels to include customer touchpoints that occur in one-way and two-way communication channels even if those channels are informational and not transactional. This is very much in line with [ 1], who do not prioritize one over the other but simply observe that the term "omnichannel" often includes channels of distribution and channels of communication.
For a manufacturer however, it is important to distinguish between communication and distribution channels, even if the line between them is increasingly blurry in the digital world. Although all distribution channels also play a communication role, the strategic decision to use or not use certain communication channels is very different from the one to sell products through certain distribution channels and not others. So are the relevant metrics to assess how well communication versus distribution channels are working.
In both types of channels, the fit of the product with the channel and its audience/customer base is important, and so is consistency in positioning. But beyond that, when making decisions about communication channels, the manufacturer is in the mode of advertising budgeting, media planning, return-on-investment assessment, and attribution analysis. In contrast, when making decisions about distribution channels, issues include the functions performed by each channel; the need to balance distribution coverage against conflict with channel partners; which parts of the product line to sell in which channels, at what price, with what types of incentives and contracts; and how to influence the pricing and efforts of the channel partners ([ 2]] discuss these and other issues in detail).
Suppliers know well the conflict that occurs with existing channel partners when they introduce new channels. Adding or dropping specific communication channels is far from trivial, but, unless we are talking about dropping advertising behemoths such as Google and Facebook, it is not nearly as consequential as adding or dropping distribution channels or even specific distribution channel partners.[ 3]
Of course, the demarcation between communication and distribution channels is far from ironclad. Traditionally, there has been no ambiguity that advertising is a communication channel, but certain kinds of digital advertising today can substitute for distribution. This is definitely the case for shoppable ads of the type that Instagram and Google introduced in 2019, in which a consumer can hover over a product in the image to not only see seller and price information but also buy the item with a click. Brands, especially small or niche ones aimed at younger market segments that can be reached effectively on social media, may find that shoppable ads are a way to make their products available for purchase without relying on too many independent channel partners. While they may be labeled digital advertising, the unique transactional role they play requires that their effects be analyzed separately from other types of digital advertising for manufacturers.
As an example, [ 4] found, using aggregate data from a retailer, that total elasticities and return on investment for display and search advertising were substantially higher than for traditional advertising, and this was largely attributable to the cross-channel effects of display and search. But it is not clear whether the result would generalize to a manufacturer's shoppable ads. If the consumer can simply click on the ad to make a purchase, that should reduce cross-channel effects, but many consumers viewing the ad may not be far enough along their path to purchase yet. In addition, previous research on feedback effects has found that traditional advertising can beget wider distribution, and distribution can make advertising more effective, but the longer-term and feedback effects of these types of ads may differ to the extent that they substitute for, rather than drive, distribution. An analysis of these effects would benefit from the integration of marketing-mix models with individual-level attribution analysis (e.g., [ 6] study of the effects of different types of display advertising at different stages along the consumer's funnel path) that Cui et al. (Table 2) call for.
Cui et al. view omnichannel as the ideal, saying that "in the ideal scenario, customers interact seamlessly with the firm across channels both internal and external to the firm, and the firm has full information on all customer touchpoints to provide a single unified experience across channels" (p. 104). Indeed, omnichannel is widely viewed as the holy grail for retailers.[ 4] For manufacturers, however, it is by no means clear that omnichannel should be the holy grail.
Would manufacturers like to have a 360-degree view of the end customer across all channels? And do they want the customer experience to be the best it can be in every channel? Certainly! But a seamless customer experience across multiple channels that they do not own is not feasible and, indeed, may not be desirable. For example, it may make sense for manufacturers with a tightly targeted and relatively small product line to try to be omnichannel by selling the same products in all channels and harmonizing prices with minimum advertised price or even minimum resale price policies. However, many manufacturers sell different brands, different products, or simply different versions of the same products, through different channels. There are many reasons for this, targeting different customer segments and reducing head-to-head price competition between channels being two of them. Such decisions open up an array of research questions.
Consumers search differently and likely buy different products in physical and digital channels. Recent research has focused on optimizing offline and online product assortment from the perspective of an omnichannel retailer (e.g., [ 5]; [ 9]). Manufacturers are directly affected by how omnichannel retailers make product assortment decisions across their offline and online channels, but manufacturers must deal with additional issues in designing their own assortment for different channel members.
For example, there are demands for temporary or long-term exclusivity and/or stockkeeping units that fit the channel's needs in terms of features, size, price point, or packaging; there are the production and logistics costs involved in meeting those demands; there is the challenge of determining the gains and losses of offering exclusive products to a given channel member (e.g., [ 7]) or to offline-only or online-only channels. In addition, manufacturers may have different goals for supplying a limited portion of their product line to a particular channel. For instance, Burberry negotiated a deal with Amazon to sell a limited number of items in exchange for assistance in controlling unauthorized sellers of Burberry on Amazon. Research is needed on the manufacturer's assortment decisions across channels and also to better understand the effectiveness of exclusivity and of Burberry-like deals.
We also need to understand how manufacturers' expansion into new, often online, channels affects the pricing, presentation, and performance of their brands, given the inevitable exacerbation of conflict with existing channel partners. In particular, while there is plenty of research on the impact of opening or closing an online or offline channel for a retailer, we know very little about the impact of a manufacturer opening a DTC channel in competition with its independent channel partners. Channels that are DTC provide visibility and control over the customer experience, but those who can afford to rely only on a DTC channel remain the exception to the rule.[ 5] When manufacturers open their own physical stores, the reactions of channel partners and the effect on the brand's overall performance will vary depending on whether there are a small number of flagship stores or factory outlet stores, or several regular physical stores intended to compete directly with independent channel partners. Similarly, with a DTC website, both effects will vary depending on whether the website is only informational or also transactional—and if transactional, whether it offers products not available to independent channel partners; whether it routes orders to independent channel partners or fulfills them in competition with channel partners; and whether it does so at the same, lower, or higher prices than those of its channel partners.
Cui et al. emphasize how the execution of omnichannel strategies depends on availability of reliable data across channels and across the stages of the customer's journey. As they explain, technologies such as blockchain have greatly improved the ability to share data across multiple sources within and beyond the boundaries of a firm while assuring users of the integrity of the data. However, despite their benefits, these technologies do not increase the willingness of independent channel members to share their customer data when the supplier does business with their competitors and increasingly also competes with them through a DTC channel.
Supply-chain-focused Walmart has historically had great success with vendor data-sharing systems (e.g., their Retail Link and Vendor Managed Inventory systems), and not surprisingly, the company is at the forefront in blockchain use as well. It is a canonical example in many discussions of blockchain. But many companies lack the channel power of Walmart, and sharing supply chain data with an upstream or downstream partner is quite different from sharing customer data with competitors, especially when insights from these data can be an important competitive advantage and a source of leverage in channel negotiations.
Cui et al. mention the need to explore how firms can be incentivized to share data with their channel partners as well as with their competitors, but this is easier said than done and warrants elaboration. I discuss some of the incentives that manufacturers can use as well as their likely effectiveness. One approach is for the manufacturer to gain access to customer data by taking responsibility for important functions that are onerous for the channel members to perform. For example, Natura provides payment, delivery, and data analytics support for its digital sales consultants when customers are registered through the consultants' personalized web pages on the company's Rede Natura website, and Progressive Insurance set up its "For Agents Only" website to provide similar services to its independent agents ([ 2], p. 300). In such instances, the channel members are individuals who often lack the resources to perform the functions themselves or the skills to do much with the customer data.
In another approach, when there are very close strategic relationships between manufacturers and a small number of channel partners with whom they do business, information sharing may be the norm. This is most likely in business-to-business contexts, such as in the electronic component industry, where distributors routinely register their customers and even prospects with suppliers. In most other cases, however, channel partners that operate at arm's length from manufacturers are reasonably well-resourced organizations that zealously guard their customer data and, increasingly, can put those data to good use to improve their own performance.
Manufacturers can incentivize such channel partners to share data by demonstrating to them that they are channel agnostic and care about performance across channels rather than about capturing sales in their DTC channel. For example, manufacturers can route orders received on their own website to the stores of their independent retailers. Or they may employ initiatives such as "buy online and pickup in store" or "buy online and return in store" in their DTC online channel to increase foot traffic to those stores. When a retailer uses these initiatives, they have been shown to benefit the retailer's offline sales. When a manufacturer employs them, the benefit to the retailer may come in the form of sales of other manufacturers' products, so the manufacturer may not see a direct effect on its sales. Still, the manufacturer can demonstrate to the retailers its commitment to their success, and perhaps even reduce its logistics costs in the bargain. Manufacturers can also optimize their advertising to drive online and offline sales of their channel partners, not just DTC sales, using targeting and location analytics services such as Placed.com, Facebook's Offline Sales Measurement, and Google's Store Visits, all of which connect ad exposures to offline store visits.[ 6] In turn, maybe, just maybe, some channel partners would be willing to share some of their own data.
In contrast to such channel-agnostic approaches, other suppliers may focus on driving more consumers to the DTC channel to own their customer experience, build up their 360-degree view, and gain influence over independent channels. For example, Nike is being hailed as an omnichannel trailblazer. It is emphasizing DTC with its SNKRS app, the Nike Plus Loyalty Program, and other initiatives. But it remains to be seen whether and how it can spread its omnichannel initiatives across independent retail partners, which, as of 2020, account for over 65% of the company's U.S. revenue. For example, should Nike allow consumers to earn and redeem points on Nike's own loyalty program when the transaction is with a retailer? If retailers are charged for this, how would they react? If they are not charged, would the additional data be worth the cost of the program? Assessing the effectiveness of these various approaches is a fruitful avenue for research.
With or without incentives, and with or without blockchain and other enabling technologies, conversations with executives from retailer as well as manufacturer organizations suggest little hope that many channel partners will share their granular data on identifiable customers anytime soon. But effectiveness of the aforementioned approaches is not only about customer data—and definitely not only about granular data. Researchers should examine the extent to which these approaches coordinate the channel by motivating channel partners to increase their selling efforts and improve their customers' experience with the manufacturer's products. After all, access to data is only a means to that end goal for the manufacturer.
Cui et al. make an important point in their concluding section about which data are really necessary (and for how long) for a firm to accomplish specific omnichannel objectives while not impinging on consumer privacy. That point is even more important for manufacturers. What type of data are really needed to achieve the manufacturer's goal of ensuring that customers get the best experience possible in whichever distribution channel and with whichever channel member they interact? As one example, much of the conflict that can make it harder for manufacturers to coordinate across multiple independent channels may come down to disagreements over the division of work and pay, especially when one type of channel is more likely to be "showroomed" while another is more likely to be able to free ride on the showroomed channel's investments and sell at lower prices. Precise attribution of a purchase is not needed to address such conflict, unlike in attribution models whose goal is to optimize spending in different communication and transaction channels. Identifying the showroomed channels and measuring the channel members' support of consumers along their purchase path may be a matter of tracking some of the metrics described by [ 1] through traditional means like consumer surveys and store audits.
Of course, manufacturers continue to use traditional means of piecing together their customer databases from a variety of sources such as DTC channel interactions, warranty and registration information, and information collected from redemption of rebates and other consumer promotional offers. The ability to merge these databases with the data available from third-party providers such as those described by Cui et al. gives them access to much more information about their own customers than they can collect on their own, as well as large panels of other consumers from which they can analyze lookalikes.
Cui et al. provide a rich and state-of-the-art discussion of the information needs for effective omnichannel marketing by retailers and vertical brands, the hurdles that exist, and how they can be overcome. Practicing managers will find great value in their discussion of new technologies, privacy regulations, new sources of data, and methods to make the best use of available data. Data scientists and marketing researchers will, in addition, find their discussion of marketing attribution a great way to get up to speed on the topic, along with plenty of ideas for new research. I offer a complementary view, looking at omnichannel from the perspective of manufacturers that work with independent channel partners. I have highlighted where the goals and challenges facing these managers differ from those facing retailers and offered avenues for related research.
Footnotes 1 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
2 Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
3 This is true even for a retailer, as exemplified by the large body of research on what happens when a retailer opens or closes a transaction channel. In contrast, there is no research (at least to my knowledge) on when a retailer starts or stops using a specific communication channel.
4 Even so, [8] conceptualizes a continuum from managing in silos to integrating the experience across different stages of a customer's journey and/or across a retailer's offline and online channels. He also raises the question of how far along the continuum it is advisable for a retailer to go.
5 Becoming a seller on a third-party marketplace such as Amazon provides more control than selling to the retailer and gives access to many more end customers than a DTC channel would afford, but no more visibility, because Amazon does not share individual customer data (see [2], p. 209).
6 Cui et al. provide a good summary of the types of services that firms can use to integrate their own first-party data with third-party data from multiple sources.
References Ailawadi Kusum L., Farris Paul W. (2017), "Managing Multi- and Omnichannel Distribution: Metrics and Research Directions," Journal of Retailing, 93 (1), 120–35.
Ailawadi Kusum L., Farris Paul W. (2020), Getting Multi-Channel Distribution Right. Hoboken, NJ : John Wiley & Sons.
Cui Tony Haitao, Ghose Anindya, Halaburda Hanna, Iyengar Raghuram, Pauwels Koen, Sriram S., Tucker Catherine, Venkataraman Sriraman. (2021), "Informational Challenges in Omnichannel Marketing: Remedies and Future Research," Journal of Marketing, 85 (1), 103–20.
Dinner Isaac M., Heerde Harald J. van, Neslin Scott A. (2014), "Driving Online and Offline Sales: The Cross-Channel Effects of Traditional, Online Display, and Paid Search Advertising," Journal of Marketing Research, 51 (5), 527–45.
Dzyabura Daria, Jagabathula Srikanth. (2018), "Offline Assortment Optimization in the Presence of an Online Channel," Management Science, 64 (6), 2767–86.
Ghose Anindya, Todri Vilma. (2016), "Towards a Digital Attribution Model: Measuring the Impact of Display Advertising on Online Consumer Behavior," MIS Quarterly, 40 (4), 889–910.
7 Gielens Katrijn, Gijsbrechts Els, Dekimpe Marnik G. (2014), "Gains and Losses of Exclusivity in Grocery Retailing," International Journal of Research in Marketing, 31 (3), 239–52.
8 Neslin Scott A. (2020), " The Omnichannel Continuum: How Far Is Far Enough? " working paper, Tuck School of Business, Dartmouth College.
9 Rooderkerk Robert P., Kök Gürhan. (2019), " Omnichannel Assortment Planning," in Operations in an Omnichannel World, Vol. 8, Gallino Santiago, Moreno Antonio, eds. New York : Springer, 51–86.
Verhoef Peter C., Kannan P.K., Jeffrey Inman J. (2015), "From Multi-Channel Retailing to Omnichannel Retailing: Introduction to the Special Issue on Multi-Channel Retailing," Journal of Retailing, 91 (2), 174–81.
~~~~~~~~
By Kusum L. Ailawadi
Reported by Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 30- Commentary: The Case for a Healthier Social Customer Journey. By: Forbus, Pamela. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p93-97. 5p. DOI: 10.1177/0022242920974680.
- Database:
- Business Source Complete
Commentary: The Case for a Healthier Social Customer Journey
Social media platforms such as Facebook, where Pernod Ricard USA spent more than half of its ad dollars last year, are indispensable business tools. They allow affordable, targeted advertising for even the smallest brands and businesses, and enable easy two-way communication with customers and brand fans. In addition, they help businesses facilitate search, increase website traffic, and demonstrate brand values and cultural relevance in real time. Social media platforms allow brands to interact directly with individuals in the midst of the customer journey, influencing customer decision making and purchase. As social media platforms continue to transform and evolve how brands and consumers engage, it is imperative for marketing researchers and practitioners to reevaluate current thinking on the customer journey framework. [ 3], hereinafter Hamilton et al.) offer a framework for understanding the current role of various social influences on that journey—what they call the "social customer journey."
Their work explores the potentially powerful impact of distal social others—"larger groups or the whole of society, whose members may not be individuated, present, temporally proximal, or even known to the customer" (Hamilton et al.)—at each step in the journey and the role these distal influences, or "traveling companions," have on consumers' motivations, search, evaluation, decisions, and postdecision sharing activities.
Marketers are currently grappling with one aspect of their model: the role of shared values in the social customer journey and the traveling companions, including brands, that leverage those shared values to influence decision making at each step of the journey. Hamilton et al. suggest the following questions for further research: How do consumers navigate conflicts between brand preferences and alignment with their broader social networks' values (e.g., political ideologies, social causes)? How do consumers manage situations in which proximal motivational inputs are in opposition to distal social norms?
In simpler terms, how do consumers make choices when the act of choosing is an "identity signal" (Hamilton et al.) that either aligns them with or puts them in opposition to the norms and values of their personal social networks? These questions are critical for practitioners, who are increasingly communicating their values via social channels, leading to both positive and negative outcomes including boycotts, buycotts, increases or decreases in reputation, changes to loyalty and consideration, and becoming a target of society's ever-growing "cancel culture."
This past summer's #StopHateForProfit boycott points to a critical question to explore about the role of brands in the social customer journey: What is the role of the business community in stopping "traveling companions" from toxifying the social media environment with hate speech and negativity? While Hamilton et al. acknowledge the influence these distal social others have on a specific customer journey, they do not address the influence they have on the overall environment in which individuals are making decisions. "Traveling companions" who spread hate in social media environments may have an outsized influence on the social customer journey, driving consumers to curtail or end engagement on social media, thereby diminishing trust in the social media platforms themselves and the viability of social media platforms as powerful business tools all brands can rely on.
This commentary extends Hamilton et al.'s view of the social customer journey by exploring the business imperative of addressing the toxic social environments created by "traveling companions" and a solution being developed by Pernod Ricard USA, in partnership with industry associations, to empower consumers, advertisers, and social media platforms to stop the spread of hate speech online.
In July 2020, more than 1,200 companies—including Pernod Ricard USA—joined the #StopHateForProfit boycott. Boycotters agreed to pause social media advertising on Facebook for the month to pressure the company's leadership to move with greater speed, effectiveness, and transparency in addressing the spread of hate, extremism, and misinformation on the platform. Many companies opted to pause advertising on all social channels for the month of July—and even for the remainder of 2020—to demonstrate the seriousness of the issue and to acknowledge that these problems exist on all social platforms. The boycott was intended to leverage the influence of the advertisers whose choices are directly responsible for Facebook's earnings, which reached $70 billion in 2019,[ 3] and to hold the platform accountable for the toxic environment created by individuals who leverage it to spread hate.
The evidence of hate on social media is abundant. Research from the Anti-Defamation League (ADL; [ 1]) shows that 35% of Americans report experiencing harassment online due to racial, religious, or sexual identity, with LGBTQ+ individuals, Muslims, Hispanics or Latinos, and African Americans facing "especially high rates of identity-based discrimination." And these levels are climbing, with research showing religion-based harassment nearly doubling year-over-year from 11% to 21%, and race- and ethnicity-based harassment jumping from 15% to 25% ([ 1]).
The negative impacts of hate are just as evident. Of Americans who reported being harassed online, more than 45% reported physical, mental, or emotional impacts—trouble sleeping, incidents of depression and anxiety, and increased sense of fear in the physical world, according to the ADL. Researchers in the United Kingdom found that as the number of "hate tweets"—those deemed antagonistic in terms of race, ethnicity, or religion—made from one location increased, so did the number of racially and religiously aggravated crimes, which include violence, harassment, and criminal damage ([ 2]). And with teen suicide rates rising by nearly 60% between 2007 and 2018, cyberbullying and online harassment are often cited as factors in those deaths ([ 5]).
For marketers, one particular finding in recent research highlights the tangible business imperative—more than 35% of Americans who experienced harassment online have changed, reduced, or stopped their online activity ([ 1]). Consumers are leaving platforms because of the toxicity in the environment, causing these channels to be less impactful tools for all businesses.
While the boycott led to a deeper understanding of the problem, it also reinforced a key point about Facebook's business—the 1,200 organizations who joined the boycott are just a fraction of the more than eight million groups that advertise on the platform. Furthermore, the top 100 advertisers reportedly generate less than 20% of Facebook's revenue ([ 4]), while "the bulk of the company's sales come from millions of smaller businesses that rely heavily on the platform" ([ 6]). Without the small and medium business (SMB) community's participation, boycotts are not an effective solution for companies looking to detoxify social media environments. But without action from businesses, consumers will continue to leave the social platforms in search of safer environments.
Pernod Ricard USA's core value of conviviality compels us to take action on the issue of social media safety. In a pandemic where physically distancing ourselves from one another is critical to physical health, the need to be socially connected—to create conviviality—becomes even more important. Social media was made for this moment. But if action is not taken quickly to stop the hate and harassment that make social media environments toxic for many users, the health and well-being of people on and off the platforms are at risk.
While there is no simple, singular solution to this problem, inspiration can be found in other contexts where corporations, brands, products, and consumers intersect to create positive change. A parallel can be drawn between environmental sustainability and the actions corporations have taken to reduce their negative impact on our physical environment. Hate speech and online harassment are the social media analog of greenhouse gases or chemical waste, suggesting that sustainability thinking extends to social media, with a focus on the actions corporations can take to reduce their negative impact on the online environment. The environmental sustainability movement also demonstrates that real change only occurs when individuals, communities, corporations, brands, and products come together and work toward a shared vision for the future.
Drawing on this analogy, the #EngageResponsibly initiative (engageresponsibly.org) is an effort to spark collective action among all key stakeholders—consumers, brands, and social media platforms—to go beyond the boycott and create change that will directly result in a safer social media environment for all. Specifically, it seeks to achieve four goals: ( 1) give consumers a voice; ( 2) help brands leverage their influence to be true drivers of change; ( 3) engage SMBs in thoughtful ways; and ( 4) unify vision and voices across platforms to drive greater impact, accountability, and transparency. Next, I discuss each of these goals in turn and provide details of the #EngageResponsibly program.
#EngageResponsibly gives consumers a voice in three ways: awareness, action, and advocacy. Consumers will learn more about the problem of online hate speech and the actions they can take to stop it—including awareness of how to use existing reporting tools on each social media platform—through a brand-led marketing campaign. The campaign will be amplified by influencers committed to promoting a safer social media environment and news stories about the initiative.
Importantly, individuals will have new tools at their disposal to take action, including a newly developed hate reporting tool that will allow social media users to report hate speech they encounter online using the direct message feature on Twitter, Instagram, and Facebook. Research indicates that people are looking for more transparent and credible ways to have their voices heard. The [ 1], p. 21) finds that 77% of online users want hate content reporting to be easier and 62% seek independent reporting of online hate. By building this functionality into the platforms, we make it as easy as possible for individuals to report hate.
Once individuals use the hate reporting tool, #EngageResponsibly will introduce them to the "Initiative Hub," which will house real-time social listening data on hate speech, trends in hate reporting based on used patterns from the hate reporting tool, details about coalition members, and more.
#EngageResponsibly will inspire brands and organizations to both empower consumers and hold themselves accountable to the highest standards of responsible marketing. To do this, platforms and brands will opt in to become "Anti-Hate Certified." This certification will come with a seal of approval that can be used internally and externally and that serves as an easy visual marker to all audiences that the brand is committed to stopping the spread of hate on social media platforms.
Certification begins with #EngageResponsibly, in partnership with a credible third-party partner, calculating a "Platform Safety Score" for each social media platform. Scores will be based on a predetermined set of criteria that measure the volume of hate and the progress being made by the social platform to address it. The scores will be updated on a quarterly basis to understand what drives fluctuations, along with the scope and scale of hate at key points in time.
Next, brands will calculate their "Hate Footprint." Using a formula based on the quarterly "Platform Safety Score" and the brand's advertising spend on said platform, brands will learn their social media "Hate Footprint," which will allow them to then invest in designated, pre-vetted nongovernmental and advocacy organizations supporting communities most affected by hate speech to offset their footprint. The "Hate Footprint" and size of the required offset will be developed in partnership with a credible third-party partner.
As discussed previously, the bulk of Facebook's revenue comes from SMBs and performance-driven advertising. Yet SMBs simply do not have the ability to participate in boycotts like #StopHateForProfit without putting themselves in financial jeopardy—especially given the fiscal impact of our current global pandemic. However, according to research conducted by WPP in September for an internal Pernod Ricard report, 68% of SMBs say that they would join a coalition to fight hate speech, and 68% also say they would fight much harder if they were equipped with more tools and resources. #EngageResponsibly will create a program and toolkit to help all SMBs—many of whom are partners and customers of the corporations who participated in the initial boycott—to encourage responsible marketing for businesses of all sizes.
A critical element of the initiative is to unify the vision and voices of all key stakeholders—including the social media platforms—to drive greater impact, accountability, and transparency by all, for all. Active engagement and partnership with the social media platforms will be critical to the success of #EngageResponsibly. The initiative will achieve this through three key actions: ( 1) generating new data that measure the problem, ( 2) providing a consistent standard of analysis, and ( 3) creating a program built around shared responsibility.
The actions taken by consumers and brands through #EngageResponsibly will provide significant new data and insights to help inform further changes to policy and functionality social media platforms can deploy to help stop the spread of hate speech. The hate reporting tool and social monitoring program led by the initiative will generate significant data on the scale of the problem, how many consumers are willing to use an independent reporting tool, what consumers need to more easily identify and report hate speech, what actions drive spikes or dips in hate speech, and more.
The Anti-Hate Certification element of the initiative, including the "Platform Safety Score" and "Hate Footprint" calculations, will use consistent, platform-agnostic formulas to assess the spread of hate speech on social media in a uniform way. This will allow consumers, brands, and the platforms themselves to more coherently assess the scope and scale of the problem on each platform.
The opt-in nature of the initiative, which requires brands to invest in the same level of transparency and accountability that consumers, communities, and businesses are seeking from the social media platforms, establishes a high level of trust and shared responsibility with the social media platforms from the outset.
Businesses, and marketers in particular, can no longer focus only on return on investment. They need to be equally focused on return on responsibility. Marketers must act responsibly if we are to be viewed credibly. #EngageResponsibly is the marketer's opportunity to lead on return on responsibility. Clear, common-sense actions by corporations that reduce the spread of hate speech online will have positive benefits for consumers, brands and, ultimately, the social media platforms themselves.
Through #EngageResponsibly, consumers will see brands empowering them to act; tangible, measurable actions they can take to stop hate speech; a visual marker that lets them know which brands share their anti-hate values; and transparent data about the action (or inaction) of social media platforms to stop hate speech.
Through #EngageResponsibly, brands will be able to engage their consumers and fans to drive real action in the fight against hate speech, transparently take accountability for the role they play in the problem, use the visual marker of certification to let the public know they walk the walk when it comes to being anti-hate, and provide tools to SMB customers and partners who want to demonstrate their anti-hate commitments. As a result, brands will experience greater trust, favorability and consideration from consumers, brand fans, and SMB customers and partners, as well credibility and legitimacy in demonstrating their commitment to the well-being of their consumers and communities.
Through #EngageResponsibly, social media platforms will have access to new data-driven insights on incidents of hate speech on their platforms and an informed understanding of how well they are doing at combating the problem relative to their peer set. Importantly, the consistent approach to "scoring" and reporting hate on all social media platforms will provide each with a better understanding of the scale of the problem, along with a better understanding of how users sparking hate travel from platform to platform, and what triggers those migrations. This will help platforms anticipate potential spikes in hate speech and act quickly to limit their spread, helping maintain the user activity that is critical to their business health.
As Hamilton et al. illuminate, the social customer journey provides a new understanding of the complex proximal and distal influences on customers at each point in their online decision-making journey. But a missing element in the current analysis is how the social environment in which the decision is being made—especially when that environment contains toxins generated by hate speech—affects the journey or a customer's willingness to remain on the social platform altogether.
The #EngageResponsibly initiative could help social media to become a safer, more responsible space for consumers and brands to interact, minimizing the negative impact of "traveling companions" who distract with hate. Yet, while #EngageResponsibly has strong potential to create positive impact, it also has limitations. As an opt-in approach, it relies on actors willing to dedicate resources to test, learn, and iterate, rather than a mandate or policy required of all actors. As a coalition effort, the need to continually balance needs and goals of all parties will present challenges along the way. Without the engagement of government, the relative ability of regulation to provide tangible solutions to the problem of online hate speech remain unknown. We welcome collaboration with researchers interested in evaluating #EngageResponsibly.
In addition, marketing researchers can play a critical role in advancing this effort by analyzing other impacts of toxicity in social media environments for both individuals and brands. For example, researchers can explore the following:
- The economic impact of hate speech on social media. If consumers flee the platforms, what impact does that have on advertisers' return on investment and the bottom line for companies engaging customers on social media? Is there a tipping point at which the platforms become less effective channels for advertising due to hate? What economic impact do individuals who abstain from social media participation face? What economic impact will social media platforms face if more consumers flee?
- The cost of values-based marketing. As companies choose not to advertise via mediums and outlets (print, broadcast, and social media) that do not adequately align with their corporate values, are there enough alternative channels to maintain or even increase impact? Where do economic impact and reputational impact intersect for businesses based on how they advertise? Are there greater or lesser economic impacts by channel (e.g., is there a greater or lesser economic impact for a company that chooses not to advertise on specific broadcast networks vs. specific social media channels)?
- Defining toxicity and toxicity tolerance for individuals. While platforms, organizations and brands work to define and diminish toxic social media environments, how do individuals define them? Do definitions of toxicity or tolerance for toxicity vary from individual to individual, or can correlations be made between groups with shared characteristics? How closely do individual's definitions and tolerance levels align with proximal social others and distal social others?
In the social media environment, brands are friends, connections, and "traveling companions" of consumers. When those relationships are built on shared values, brands have an incredible ability to influence the choices consumers make on the social customer journey. However, when distal others are actively working to toxify the social media environment with hate speech, brands have both a need and responsibility to stamp out the hate. Brands that condone hate speech, through their own inaction, where social connections are built on shared values, will lose credibility and connections over time.
Hate is not good for humanity. It is not good for business. And it is not good for the long-term viability of social media platforms. People need social media more than ever to maintain and strengthen their social connections in a world where physical distancing is critical to physical health. Now is the moment for brands, consumers, and social media platforms to work together to stop the spread of hate speech online and ensure the social media environment is safe, healthy, and viable for all.
Footnotes 1 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
2 Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
3 https://investor.fb.com/investor-news/press-release-details/2020/Facebook-Reports-Fourth-Quarter-and-Full-Year-2019-Results/default.aspx
References ADL (2020), "Online Hate and Harassment Report: The American Experience 2020," research report, https://www.adl.org/online-hate-2020.
Cardiff University (2019), "Increase in Online Hate Speech Leads to More Crimes Against Minorities," PhysOrg (October 15), https://phys.org/news/2019-10-online-speech-crimes-minorities.html.
Hamilton Ryan, Ferraro Rosellina, Haws Kelly L., Mukhopadhyay Anirban. (2021), "Traveling with Companions: The Social Customer Journey," Journal of Marketing, 85 (1), 68–92.
4 Iyengar Rishi. (2020), "Here's How Big Facebook's Ad Business Really Is," CNN Business (July 1), https://www.cnn.com/2020/06/30/tech/facebook-ad-business-boycott/index.html.
5 Reinberg Steven. (2020), "Suicide Rate Keeps Rising Among Young Americans," WebMD (September 11), https://www.webmd.com/depression/news/20200911/suicide-rate-keeps-rising-among-young-americans#1.
6 The New York Times (2020), "All the Companies Quitting Facebook," (June 29), https://www.nytimes.com/2020/06/29/business/dealbook/facebook-boycott-ads.html.
~~~~~~~~
By Pamela Forbus
Reported by Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 31- Commentary: The Ethical Use of Powerful Words and Persuasive Machines. By: Donath, Judith. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p160-162. 3p. DOI: 10.1177/0022242920973975.
- Database:
- Business Source Complete
Commentary: The Ethical Use of Powerful Words and Persuasive Machines
[ 6], hereinafter Puntoni et al.) have written a thought-provoking article, one that returns repeatedly to the social and ethical questions brought about by the various technologies that fall under the umbrella of artificial intelligence (AI) and that concludes with a call for professional guidelines that "acknowledge the new ethical challenges raised for marketers by the growth of AI" (p. 178). This is a timely and important goal, and herein I would like to contribute to it by highlighting two areas where the intersection of marketing and artificial intelligence raises important ethical questions.
The first area is vocabulary. That words are powerful is not news to anyone in the field of marketing. The right words create an aura of desirability around a product, imbuing it with mystery or coolness or bacteria-vanquishing cleaning power. Used unethically, they can mislead. Words shape how we think. And, while marketers are not the builders of computational systems, they are often the architects of how they are described and depicted.
Let's begin by examining the power of the phrase "artificial intelligence" and the use of terms that ascribe intention and volition to computational processes. "Artificial intelligence" can refer to everything from sophisticated interactive robots to auto-complete algorithms; it is credited with finding Martian craters, writing sports news, powering toothbrushes, and bringing us washing machines that "leverage advanced AI to deliver peace of mind and enhanced customer satisfaction, along with improved product performance and longevity."[ 3] It is a term that is both vivid and vague, imbuing products and processes with the aura of having insights and capabilities beyond our comprehension, but in fact adding nothing to our understanding of their actual capabilities or limits.
The term "artificial intelligence" was coined in 1956, in a proposal for a summer research workshop at Dartmouth, the goal of which was "to make machines use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves...[to] behave in ways that would be called intelligent if a human were so behaving" ([ 4], p. 2). The phrase was meant to be ambitious, flashy, and attention getting, to encourage big ideas rather than narrow technical studies ([ 5]). It was, one might say, a marketing phrase for the new field.
It quickly became clear that the researchers had vastly underestimated the complexity of creating a human-like intelligence. Subfields emerged—machine vision, natural language processing, knowledge representation—but the big problem of creating general artificial intelligence—human-like machines that could reason and adapt, even experience thinking and a sense of self—receded from mainstream research. The recent resurgence of interest in AI is due to the development of "deep learning" systems[ 4] that have made rapid advances in application in its own specific domain; they are achievements of new algorithms and massive data collections, not of an emerging machine mind. We remain far from achieving anything resembling general AI.
While the exigencies of research funding and career paths have steered scientific research toward practical applied problems, more vivid depictions of intelligent machines have proliferated in popular culture. These imagined artificial beings are sometimes cute (e.g., Astro Boy) or cooperative (e.g., Asimov's I Robot series); often they are destructive, their superhuman intelligence an existential threat to humanity (e.g., The Matrix, Hal in 2001: A Space Odyssey, Ellison's short story "I Have No Mouth, and I Must Scream"). And while these fictional representations of AI as entities with emotions, desires, and intentions bear little relationship to the statistical data analysis and other algorithmic programs that constitute real-world AI systems, it is the imaginary narratives that shape what the words "artificial intelligence" evoke, and therein lies the root of my first concern.
Powerful, superhumanly brilliant science fiction AIs imbue the phrase "artificial intelligence" with credibility and authority. In real life, too, breathless reports of how one human achievement after another has just been bested by an AI (particularly in the world of games, where humans have been defeated by computers playing chess, Go, and now even six-player no-limit Texas Hold'em) reinforce the impression that trying to match wits with an AI is as humbling and futile as trying to outrun a train.
While such an aura can be advantageous when applied to a product or service, it is also problematic, because it discourages critique and questioning. As computer scientist/critic Jaron Lanier notes, the term AI "adds a layer of religious thinking to what otherwise should be a technical field."[ 5] The danger here is that one does not challenge seers and oracles. If an "AI-powered" system says that this is the right policy or marketing strategy to enact, or that this is the best route to take, who are you to question it?
Encouraging critique is especially important when you consider that not all products and applications are beneficial. The vague authority of "artificial intelligence" can provide a smokescreen behind which lurk procedures that are not in the user's interest, from privacy-shredding data collection and biased decision-making algorithms to emotionally manipulative interaction techniques, as Puntoni et al. warn.
Phrases that refer to AI as an individual rather than as a technology or technique are especially problematic. The phrase "an AI" (as in "an AI beat the top chess champion" or "your application/your paper/your plea for asylum was rejected by the AI") ascribes behavior and decision making to an entity in the machine, one with a will, agency, and intentions. Yet there is no such being. The gameplay, the drug discovery, the application analysis are performed by programs—big, complex, perhaps incomprehensible, but earthly programs nonetheless, no more conscious than a spreadsheet. "An AI" conjures up a superhuman mind, preternatural in both its oracular abilities and its disembodied presence, an unquestionable authority.
Other common usages, such as using verbs such as "thinks," "wants," and "decides," more subtly invoke intentions and agency. For example, Puntoni et al. note, "Data collection devices listen, in the broad sense of gathering information from different sources" (p. 163). To listen implies a mind, the ability to pay attention and anticipate. Think of a simple organism: one would say that it may sense vibrations, but not that it is listening to anything—for that, one needs a brain. When we use words such as "listen" to describe a computational process, rather than, say, "data collecting," we imbue the machine with mental abilities it does not have.
The conflation of conscious beings and machines raises a number of ethical issues, central to which is the concept of moral rights. Peter Singer, among other ethicists, has argued that the basis of having moral rights—the right to life, the right to not be subjected to unnecessary pain—is being sentient, that is, having the capacity to feel sensation ([ 7]). If we convince ourselves (consciously or not) that various devices and programs are sentient entities, do we change our relationship with and responsibilities toward them? Do we need to take their experience into account and avoid, if possible, causing them harm? Joseph Weizenbaum, who in 1976 created the first chatbot and was subsequently deeply dismayed at people's willingness to confide in it and to see it as an empathic being, warned that the willingness to accept machines as fellow beings risked devaluing the significance and sanctity we accord to being human.
Words matter. The technologies we call "AI" have other, more accurate names. Phrases such as "big data analysis," "algorithmic curating," and "adaptive learning" may not sound as exciting and futuristic, but they are informative, descriptive, and empowering; they are more open to inquiry, to looking under the hood. I encourage using these terms where appropriate, and limiting "AI" and "artificial intelligence" to technologies that are meant to be perceived as entities or that have adaptive and generalized intelligence.
The second big concern at the intersection of AI and marketing encompasses the realm of persuasion ([ 1], [ 2]). A core goal of marketing is persuasion: persuading people to purchase a product, to make a donation, to recycle, to believe an idea, or to vote for a candidate. Machine learning and other technologies may soon be able to make preternaturally persuasive messages—but should they? As Puntoni et al. describe, a tremendous amount of data is being collected and analyzed about individuals as they go about their daily lives, whether online, where we spend more and more of our time working, socializing, and shopping, or via the increasingly ubiquitous mobile and embedded devices that track us through the physical world. The collected data report not only what we are doing, but also how we feel about it (at least as much as affect can be deduced from our facial and verbal expression or the output of our Fitbit devices). Unbeknownst to ourselves, as we peruse the news or our social media feeds, we are often also participating in massive A/B tests assessing which headline got more people to read this story or which color of background or obscure meme induced more people to click on the ad teaser. These data are then used to create increasingly targeted and presumably more persuasive presentations. These systems are still early in their development, still yielding laughably mistargeted ads at times, but it is likely that coming years will see rapid improvement of their algorithms—potentially incurring a significant effect on their target audience's autonomy. We human beings are instruments of great emotional depth and complexity—and computers are getting better and better at playing us.
The biggest contribution of AI to persuasion might be in its capacity to appear to be an intelligent, interactive entity. People, we have seen, develop emotional relationships with interactive devices ([ 8]). The simple keychain pet, the Tamagotchi, which made no pretense of intelligence, was able to persuade its millions of users to schedule their day—often at considerable inconvenience and to the irritation of parents, friends and teachers—around caring for it, by demanding attention at arbitrary moments, and "dying" if it was ignored. Owners of Aibo, the robot dog, professed to have the same affection and sense of responsibility for their mechanical companion that owners of flesh and blood canines do for theirs. People count digital assistants such as Siri and Alexa as friends—and their voice and conversation algorithms, their use of "I" and of feeling and thinking verbs, have been carefully designed to elicit that response and to gain the user's trust. It is easy imagine how such persuasive relationships can be. Now imagine how powerful several of them working together could be, chatting among themselves and suggesting that you need a weekend getaway, a pair of pink sandals, or to rethink your position on an upcoming referendum: our desire to be an accepted member of the group is powerful and exploitable.
The potential for technologies such as machine learning and artificial entities to be extraordinarily persuasive is immense—but not yet realized. Now is a propitious and critical time for those who care about the ethics of AI in marketing to think deeply about how such persuasive abilities, should they come to fruition, ought to be used. It goes without saying that using them to promote exploitive causes is unethical—but what about ethical or simply neutral ones? Is there a point beyond the limits of natural persuasiveness that is wrong? We are a social and persuadable species, and people have long been persuaded by charismatic others to act in ways contrary to their own ethics or self-interest. Are such technologies never acceptable—or is there a moral imperative to use them for the greater good?
My response to these questions is inevitably personal. I believe that the development of highly persuasive techniques is not stoppable, but it can and should be limited, and that the central ethical issue is what is being promoted, more than how is it being promoted. Specifically, I suggest the following.
- First, limit the collection of personal data. This is far from a novel suggestion, but an important one because the development of many excessively persuasive techniques requires detailed information about people's reading and viewing habits, eating patterns, physical movements, and so on.
- Second, when AI entities mix factual and sponsored material, the sponsored material should be delivered in a distinctly different voice. While this breaks the illusion of a smooth, "natural" conversation—well, that is the point.
- Third, encourage the creation of public interest marketing in which firms pledge to use their AI-persuasive clout to promote goods and ideas they believe to be socially beneficial. In general, promote serious consideration of social values within the field, including in education (ethical decision-making courses in the marketing curriculum), industry publications, and so on. As marketers acquire increasingly powerful tools of persuasion, the imperative to use them beneficially grows.
Footnotes 1 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
2 Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
3 http://www.lgnewsroom.com/2020/01/lg-introduces-next-generation-of-laundry-with-new-ai-powered-washer/.
4 "Deep learning" uses neural networks—adaptive statistical models inspired by biological neural networks—to find patterns in big data sets ([3]). "Deep" refers to the use of many network layers, not to any profundity in the system (a common misperception).
5 https://edge.org/conversation/jaron_lanier-the-myth-of-ai.
References Donath Judith. (2018), " The Robot Dog Fetches for Whom? " in A Networked Self and Human Augmentics, Artificial Intelligence, Sentience, Papacharissi Zizi, ed. London : Routledge, 26–40.
Donath Judith. (2020), " Ethical Issues in Our Relationship with Artificial Entities," in The Oxford Handbook of Ethics of AI, Dubber Frank Pasquale Markus D., Das Sunit, eds. Oxford, UK : Oxford University Press, 53–73.
LeCun Yann, Bengio Yoshua, Hinton Geoffrey. (2015), "Deep Learning," Nature, 521 (7553), 436–44.
McCarthy John, Minsky Marvin L., Rochester Nathaniel, Shannon Claude E. (1955), " A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence," Grant Proposal, Rockefeller Foundation.
McCorduck Pamela. (2004), Machines Who Think. Boca Raton, FL : Taylor and Francis.
6 Puntoni Stefano, Reczek Rebecca, Giesler Markus, Botti Simona. (2021), "Consumers and Artificial Intelligence: An Experiential Perspective," Journal of Marketing, 85 (1), 131–51.
7 Singer Peter. (2011), Practical Ethics. Cambridge, UK : Cambridge University Press.
8 Turkle Sherry. (2007), "Authenticity in the Age of Digital Companions," Interaction Studies, 8 (3), 501–17.
9 Weizenbaum Joseph. (1976), Computer Power and Human Reason. San Francisco : W.H. Freeman.
~~~~~~~~
By Judith Donath
Reported by Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 32- Commentary: The Future of Marketing Is Agile. By: Lewnes, Ann. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p64-67. 4p. DOI: 10.1177/0022242920972022.
- Database:
- Business Source Complete
Commentary: The Future of Marketing Is Agile
Agility has become a central marketing principle. A profound global event like the COVID-19 pandemic has accelerated the need for teams to move quickly, assess, and adapt. As economic conditions dramatically change, people work remotely, and many businesses sell exclusively online, marketers need to step up.
Adobe was very early to digital, but we, too, have had to stretch the bounds of what it means to be agile during COVID-19. Marketing at Adobe has been on the frontline throughout this crisis, and it has forever changed what we do and are accountable for. Now more than ever we have found that the ability to monitor and quickly identify shifts in the marketplace and customer base, rapidly respond and shift direction, reskill and bring in new talent, and consistently measure impact in real-time have become requirements for modern marketers. As noted in [ 7], hereinafter Kalaignanam et al.; p. 35), "The digital transformation of enterprises, emergence of new channels (e.g., social media, mobile devices), and a deluge of customer data is altering the practice of marketing." In short, marketing agility has become a mandate, and leaders need to orient their people, process, and technology around the customer and drive innovation to do that effectively.
Agility isn't just the domain of marketers—it's key to business transformation. Adobe knows a bit about that. We took a very profitable packaged software business and moved it to a subscription business in the Cloud. We also built a completely new category around digital marketing and customer experience management (CXM). As our business transformed and we redefined our product and go-to-market, our marketing department—on the frontline with our customers—led the charge in many ways.
We quickly realized that succeeding in a digital-first business required new skills. So we reskilled where we could and brought in new skills where we needed to—our market researchers reinvented themselves and became data analysts; our media team learned to build new ad formats and programmatic; our designers became content machines to keep pace with digital's demands, and we brought in data scientists to do marketing-mix modeling and media attribution analysis.
But above all, we looked for people with a new mindset that is an amalgamation of creativity and data. There's a myth that these two are orthogonal, but we have found them to be amazingly complementary. I have never met a creative who does not want to know how their campaign is doing or an analyst that does not want creative to work harder. At Adobe, we run marketing by the numbers while still holding our belief that creativity is our core. Data scientists, web product managers, analysts, filmmakers, and copywriters...this is the twenty-first century's marketing organization.
From initial brainstorm to execution, companies need a range of voices at the table—from different backgrounds, ages, ethnicity, experience, and education—to ensure that they're fostering a rich stream of ideas while also identifying blind spots. A recent study in Harvard Business Review found that companies with above-average total diversity (measured as the average of six diversity dimensions) had both 19% higher innovation revenues and 9% higher earnings before interest and taxes (operating) margins ([ 8]). Likewise, a Deloitte report noted that organizations with inclusive cultures were six times more likely to be innovative and agile ([ 3]).
This view is consistent with Kalaignanam et al.'s (p. 49) view that "the exchange and cross-fertilization of diverse knowledge and perspectives can spark creative ideas and processes, allowing teams to uncover and test novel marketing ideas." Adobe has always believed in the importance of hiring diverse talent and bringing different perspectives to the table. We know that when people feel respected and included, they can be more creative and innovative.
In June 2020, as the United States witnessed acts of racial injustice, we came together as a marketing team to determine how we can do better as a company and organization. The diversity of our team was vital in helping us better understand how to move forward.
[ 7] (p. 49) suggest that "increased autonomy for teams acts to reduce bureaucratic constraints, enabling team members to more effectively identify and respond to new situations," and that "for teams to adopt marketing agility, they need distributed empowerment." This has been my experience at Adobe. In an agile marketing organization, teams must make decisions quickly and have an appetite for risk.
When COVID-19 hit, we were weeks away from holding Adobe Summit, our annual digital marketing conference, in Las Vegas. Expecting over 20,000 people, we swiftly made the difficult decision to turn it into an online event. In mere weeks, we turned our keynotes, product demos, news announcements, and more than 100 breakout sessions into an online, on-demand format. And our teams did all of it from home. Now, Adobe Summit has become a digital destination, where more than 500,000 people have already viewed content.
Throughout this experience, we understood that leaders need to trust and empower their teams. This meant giving them platforms to provide feedback, opportunities to voice their ideas about our vision, and space and encouragement to execute them. It is essential to trust your teams if agility is going to become a marketing reality.
Great marketing requires cross-functional cooperation. But the truth is, most departments across an enterprise collect data, run analyses, and report findings very differently (see [ 5]). Different teams have different customer views, requirements, and practices, and these differences create friction. At Adobe, we not only use data and analytics to inform decisions but also surface and update the most important metrics for our teams, creating a single source of truth. At one point, we were using over 100 nonstandardized key performance indicators to measure the health of the business. Evaluating the impact of marketing campaigns would take weeks. From the product organization to finance, sales, and marketing, everyone struggled to agree on a single interpretation of the data.
In 2016, our information technology and marketing teams worked together to develop a new data-driven operating model (DDOM) for our Creative Cloud business. The DDOM is Adobe's playbook for deriving and acting on data-driven insights ([ 4]). This model fundamentally shifted how we operate by creating a common language around data. It helped break down silos and ensured that our teams were all marching toward a common goal.
To swiftly and effectively act on data, organizations need to develop new processes and roles that establish a cadence of reporting and facilitate a shared understanding. At Adobe, we now hold weekly cross-functional DDOM meetings, where owners of each stage of the customer journey—from awareness to purchase to renewal—are empowered to make decisions and drive action. In addition, the leadership team has assigned internal champions across key organizations—information technology, marketing, finance, product engineering, support, and more—enabling new data insights to be communicated quickly and brought into the DDOM process to be acted on throughout the organization.
The economy has fundamentally changed; we've moved from a "world with digital" to a digital-only world. The ability to deliver great customer experiences is no longer just a nice-to-have—it is a competitive requirement. It is also a massive undertaking that won't get far without a solid digital foundation.
"Speed, scope, and impact are all operating at a different velocity than they did even 10 years ago," said Craig Gorsline, president of ThoughtWorks. "And enterprises, more than ever, are really having to face the stark reality that they need not just to augment legacy technology, but they need to really rewire the entire enterprise" ([ 1]).
This "rewiring" requires that companies have the "technology chops" necessary to understand the current customer experience they are providing and to determine where improvement is needed. Strong analytics capabilities are needed to guide real-time decisions. Beyond analytics, you need a culture of testing. Our website (Adobe.com) has billions of unique visits per year. Prospects and customers come to be inspired, learn about our products, download trials, purchase software, and get help. This is where a large percentage of our revenue is derived. Optimizing our website for different audiences is Job 1 and requires a strong digital platform and constant testing to ensure content and experiences are personalized for each visitor. This same rigor is applied to all paid and earned media. Test, learn, and optimize is our mantra.
Having a robust digital foundation gave Adobe the agility we needed to create, personalize, and deliver new messaging quickly when COVID-19 first hit ([ 2]). Within a couple of days, we had updated the Adobe website with the latest resource pages for our customers and communities. We immediately launched new programs, including provisioning free at-home access to Creative Cloud for millions of students across the United States. We paused marketing campaigns that weren't directly in service of what our customers were facing. Internally, we redesigned hundreds of pages on Inside Adobe, our company-wide intranet, to prioritize COVID-19 updates, launched dedicated Slack channels, and held regular employee meetings and town halls to keep employees connected and informed.
At Adobe, we have rallied around the concept of CXM. It is phase two of the customer-centric business transformation.
The challenge today is that consumer expectations are higher than ever, and they demand that we meet those expectations across a growing set of personalized channels. Consumers do not want or need to follow a predefined path. They do what they want, on the channels they choose—across web, mobile, social, in-store—and they expect brands to interact with them with one voice.
Enterprises are faced with the need to close the gap between the customer, the channels they use, and most importantly, the experience they expect. At Adobe, we close that gap with CXM, defined as orchestrating and personalizing the entire end-to-end customer experience, moment to moment, at scale, on any channel, in real-time.
How can your organization make this happen, effectively and efficiently? It begins with five foundational elements: an open and real-time customer profile, creative agility, cross-channel, ecosystem, and intelligence.
An open and real-time customer profile is achieved via a unified customer data platform in place. The real-time customer profile stitches together data from all over the organization—behavioral, transactional, financial, operational, and more—to get a true end-to-end view of customers for immediate actionability. The more detailed and up-to-the-minute an organization's customer profile is, the more marketers can understand customer behavior and customize journeys that are more relevant and valuable.
Creative agility is the ability to create the right content quickly for every step of the customer journey. It is foundational for delivering on the promise of CXM and is core to marketing agility; yet it is overlooked by [ 7]. Creativity agility lets companies continuously test and optimize experiences in real time, which fosters innovation. We chose to in-source a lot of creative development to ensure that we can launch, test, and modify campaigns quickly if necessary.
Your CXM strategy requires the right platform for cross-channel orchestration. The customer experience must be correctly sequenced and personalized. This means you must design, connect, deliver, and manage experiences across diverse channels and devices to maintain a singular voice throughout the context of the customer journey. That could be sending a promotional email, presenting a new mobile offer, or pushing an app upgrade to a consumer's smartphone. Organizations that personalize reduce acquisitions cost by 50%, increase global revenue by up to 15%, and improve marketing spend efficiency by 30% ([ 6]).
Importantly, the CXM technology platform must support a single data model, customer experience apps, and an open ecosystem that helps accelerate innovation. The technology provider behind your platform should have established partnerships with other service companies. For example, while we believe that Adobe Experience Cloud is the best-of-breed CXM solution, our technology, software, and data partners are a key component to our success.
Finally, advanced targeting and personalization at scale aren't possible without artificial intelligence (AI). You need AI to make real-time decisioning a reality. Moreover, AI should be the backbone of your cross-channel orchestration, uncovering hidden opportunities, making your processes faster, and helping you provide contextually relevant experiences to every customer, every time, based on interest; behavior; and transactional, financial, and operational patterns.
We believe that these five foundational CXM elements will not only help organizations deliver real-time, personalized experiences to customers but also separate the leaders from the laggards in 2021—and beyond.
The ability to innovate in marketing today relies on the marketing organization's ability to move fast enough to keep up with always-on, tech-empowered consumers' ever-changing expectations. COVID-19 has only accelerated this need to move fast—and there's no going back. But, agility isn't enough. Today's most successful digital-first companies are authentic, transparent, and intent on doing good for their customers and communities. Focused on innovation with their people, processes and technology, they never lose sight of their mission and purpose. It's these companies, driven by modern marketers, that will thrive most in the future.
Footnotes 1 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
2 Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
References Abramovich Giselle. (2017), "The 3 Fundamentals of CX Innovation," CMO by Adobe (accessed October 16, 2020), https://cmo.adobe.com/articles/2017/2/the-3-fundamentals-of-cx-innovation-tlp.html#gs.gbxn2e.
Adobe (2020), "Agility Is Key When Pivoting Your Communications in a Pandemic," (accessed October 16, 2020), https://www.adobe.com/experience-cloud/insights/business-continuity/communication-agility.html.
3 Bourke Juliet, Dillon Bernadette. (2018), "The Diversity and Inclusion Revolution," Deloitte Review, 22, 82–95.
4 Cox Eric. (2019), "How Adobe Drives Its Own Transformation," Adobe Blog (March 27), https://blog.adobe.com/en/publish/2019/03/27/how-adobe-drives-its-own-transformation.html#gs.gc08g1.
5 Cui Tony Haitao, Ghose Anindya, Halaburda Hanna, Iyengar Raghuram, Pauwels Koen, Sriram S., Tucker Catherine, Venkataraman Sriraman. (2021), "Informational Challenges in Omnichannel Marketing: Remedies and Future Research," Journal of Marketing, 85 (1), 103–20.
6 Gregg Brian, Heller Jason, Perry Jesko, Tsai Jenny. (2018), "Unlocking the Next Wave of Growth by Unifying Creativity and Analytics," McKinsey & Co. (June 18), https://www.mckinsey.com/business-functions/marketing-and-sales/our-insights/the-most-perfect-union.
7 Kalaignanam Kartik, Tuli Kapil, Kushwaha Tarun, Lee Leonard, Gal David. (2021), "Marketing Agility: The Concept, Antecedents, and a Research Agenda," Journal of Marketing, 85 (1), 35–58.
8 Lorenzo Rocio, Reeves Martin. (2018), "How and Where Diversity Drives Financial Performance," Harvard Business Review (January 30), https://hbr.org/2018/01/how-and-where-diversity-drives-financial-performance.
~~~~~~~~
By Ann Lewnes
Reported by Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 33- Commentary: Toward Formalizing Social Influence Structures in Business-to-Business Customer Journeys. By: Grewal, Rajdeep; Sridhar, Shrihari. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p98-102. 5p. 1 Graph. DOI: 10.1177/0022242920974669.
- Database:
- Business Source Complete
Commentary: Toward Formalizing Social Influence Structures in Business-to-Business Customer Journeys
[ 5]; hereinafter HFHM) build on extant research on customer journeys to conceptualize the pivotal role of social influence. HFHM (p. 69) recognize "the changing and pervasive role of social influence throughout the consumer decision making process" and that the extant literature focuses on individual customer journeys such that the focus is on "isolated consumers as decision making units." With this observation, HFHM's goal is "to create a more 'accurate' journey map" to conceptualize and elucidate the "social customer journey." The authors achieve this goal by defining social others and theorizing the changing role of social others across the customer journey stages while laying out the case of joint journeys "wherein two or more customers journey occur together"—that is, "traveling companions."
HFHM focus on the business-to-consumer (B2C) customer journey, but they recognize the importance of social others in the business-to-business (B2B) customer journey. Specifically, HFHM (2021, p. 73) rely on [ 6] to suggest that in B2B "decision making can be viewed through the lens of joint customer journey as there are often several individuals playing a role in the decision-making process." Indeed, B2B marketing is social by definition as it "encompasses the activity of building mutually value-generating relationships...between organizations..., [and] the many individuals within them" ([ 3], p. 3). However, the B2B literature has evolved beyond the joint decision-making model to examine buyer–seller relationships (e.g., [ 2]) and buyer/seller networks (e.g., [ 4]). Despite this, B2B research typically receives far less attention from marketing scholars (e.g., [ 3]), and the literature on customer journeys is not an exception.
We build on HFHM to highlight unique challenges in studying social customer journeys for B2B firms. We start by describing the various stages of a B2B customer journey and underscore the multiplicity of social influence at various stages of the journey. We then develop three novel elements (n-ad, m-hierarchy, and p-s social influence structures), which serve as concepts to study the nature and impact of social influence on the buyer side, seller side, and buyer–seller interface. In the process, we raise specific questions for future research. Before concluding, we also stress the need for research in B2B social customer journey related to structural dynamics and key performance indicators.
We begin by adapting these B2C customer journey stages in HFHM for the B2B context. As an illustrative example, an energy company pursuing oil and gas exploration and production operations across many sites maintains an on-site production crew on a 24/7 basis. If the energy company decides to outsource on-site hospitality and catering functions, what steps should it go through to search for, finalize, and work with a facilities management supplier? A typical B2B customer journey such as this is comprised of nine steps: need activation and consideration, information search and shopping process, purchase/buy, billing/payment, delivery/install/setup, usage/consumption, maintenance/repair/resolution, disposal, and repurchase/rebuy/new buy ([ 1]).
This energy and the facility management seller will have multiple influencers on the buyer and seller sides at each stage of this journey. For example, finance, procurement, and site managers serve as influencers on the buyer side during the information search and shopping process stage, while project management and sales and bidding serve as influencers on the seller side during the same stage. It is important to understand the role of the influencers, the multiplicity of influences they exert within their organization and on the other party, and how this role and these influences change across the customer journey. The multiplicity of social influences in B2B buying is pervasive, as in B2C, but the social dynamics may be more complex given each influencer has a different vantage point of the organization's objective function, which they jointly seek to achieve. In the next section, we take steps toward formalizing the social influences in B2B customer journeys to stimulate future research in this important area.
We demarcate social influences in B2B markets into those within buyer-firm stakeholders and those within seller-firm stakeholders as well as those reflecting relationships between buyer stakeholders and seller stakeholders. B2B buying tasks involve many internal buyer-firm stakeholders with diverse functional backgrounds, such as finance, procurement, research and development (R&D), and production. Similarly, B2B selling involves many internal selling-firm stakeholders with diverse skill sets such as sales and bidding, project management, and safety ([ 1]). The nature, roles, skill sets, and influence of these stakeholders change through the stages of customer journey for both the buyer and seller firms. For example, an oil and gas company creates a buying committee with members from finance and procurement, as well as a site manager. Procurement vets legal hurdles and contract terms to produce the request for proposal. Finance approves the budget, and supplier search begins with ongoing feedback from site employees.
We introduce three concepts to capture the social influence structure within buyer firm and within seller firm: n-ad social structure, m-hierarchy social structure, and primary-support (p-s) social influences. We introduce each with an eye toward exposing how each may impact the journey and future research opportunities.
At each stage of a B2B journey, stakeholders arrive at important decisions on behalf of the buying/selling organization. To understand such decision making, we need to conceptualize the extent of social influence among these decision makers. We capture links among equals with the concept of n-ad social structures, such as a dyad between two individuals or a triad among three individuals. For example, in Panel A of Figure 1, we depict a simple 3-ad structure among finance, procurement, and site manager, who might come together on the buyer side during the information search stage. In Panel A, three bidirectional arrows capture three different bidirectional social influences among the individuals.
Graph: Figure 1. Illustrative example of n-ad and m-hierarchy structures.Notes: Panel A represents a simple 3-ad or triad structure to depict relationship among three functions (finance, procurement, and site manager) such that the bidirectional arrow suggests similar level of primary social influence roles at this stage of customer journey. Panel B replicates this three-ad structure and adds two 2-hierarchy structures for finance and procurement, indicated by unidirectional arrows and lighter shaded diamond and circle. Panel C replicates Panel B and adds a support 2-ad influence between lower hierarchy finance and procurement employees, indicated by dotted bidirectional arrows. Thus, the n-ad influence can be n-ad-p (as in the upper triad, solid bidirectional arrows) or n-ad-s (as in the lower dyad, lower bidirectional arrow).
What does n-ad tell us about social influence in the buying/selling firm? As n increases, more stakeholders have a similar level of influence on the decision at a customer journey stage. The benefit of increasing n for the buying firm is that the decision is evaluated from multiple vantage points that might reduce decision risk, while the downside is that decision-making costs increase. An important question to ask is what is the right level of n that balances risks and costs at each stage of customer journey? Does the buying situation (rebuy/new-buy) moderate this effect? For the seller, knowing the n-ad buyer structure provides valuable information about buyer's revealed trade-off between risks and costs. Specifically, if the buyer chooses a large n, the seller might infer that the buyer is risk averse and that, therefore, risk-mitigating seller strategies (e.g., customized communication for each buyer stakeholder) might be fruitful. Further, the buyer's n-ad structure should inform the communication strategy of the seller. Knowing the elements in the buyer's n-ad, the seller should infer the actors/experts it needs to persuade the buyer. For example, if the seller knows that the site manager is part of the n-ad, the seller would need to provide detailed information on execution of facility management services and might want to include a service employee in its own n-ad to interface with the buyer. Thus, the n-ad structure of the buyer informs the n-ad structure and the associated communication strategy of the seller.
Currently, we know little about the n-ad structures of buyers/sellers across journey stages. Unanswered questions include the following:
- Why and how do buyers choose n-ad structures at each stage of the customer journey? Antecedents could include extent and nature of risk (e.g., internal/external, supply/demand), cost drivers, and buyer characteristics (e.g., size).
- How and why do n-ad structures change across stages of customer journey in B2B firms, and how do these structures influence buyer efficacy (e.g., coordination costs, bargaining power, profitability) across customer journey stages as needs change?
- Given a particular n-ad buyer structure, how and why should a selling firm develop its n-ad structure across customer journey stages to satisfy customer needs and optimize account profitability? Answering this question could help sellers plan the appropriate n-ad structures ex ante to enhance performance.
Most organizational structures have embedded hierarchies, which rank individuals according to status or authority. Typically, the lower-ranked employee supports and advises the higher-ranked employee, either within or across functions. To enrich and capture hierarchical social influences within buying and selling firms, we supplement the n-ad social structure with the notion of m-hierarchy social structures. In Panel B, we augment the triad n-ad social structure in Panel A by introducing two 2-hierarchy social structures for finance and procurement (indicated by lighter-shaded diamond and circle, respectively). The unidirectional arrow from the lighter-shaded circle to the darker-shaded circle indicates that a lower-ranked procurement employee supports and advises the higher-ranked procurement employee for this customer journey stage. The addition of m-hierarchy to n-ad increases the fidelity with which we can assess information flow and social influence within buying and selling firms.
So how does the inclusion of m-hierarchy enrich understanding of social influence in the buying firm? When m increases, the levels of hierarchy increase (e.g., if m goes to three for procurement, then we have three levels of procurement employees). Further, when the number of m-hierarchy structures increase, a larger number of higher-ranked employees are supported by lower-ranked employees (e.g., in Panel B we have two 2-hierarchy social structures, one for finance and one for procurement).
As m increases, the benefit for the buying/selling firm is that the information-processing burden on top-ranked employee in the m-hierarchy decreases, and the decision quality should improve. Essentially, if a top procurement employee has support from a lower-ranked procurement employee, the top employee can focus on strategic aspects (such as their negotiations with their n-ad connections in finance and site manager) rather than administrative activities. However, an increase in m not only increases the costs (of an additional employee), it might also result in information loss due to miscommunication through the ranks.
An important question, therefore, is this: What is the appropriate balance of n and m at each stage of customer journey for buying and selling firms? As n increases, coordination and information sharing costs across the n-ad structure increases, and the complexity faced by top-ranked employees increases; thus, increasing complexity necessities the need for m-hierarchy support. However, as m increases, costs and potential for information loss increases as well. Further, an increase in n increases the need for hierarchal support across several top-ranked employees in the n-ad; as a result, the number of m-hierarchy structures increase. Therefore, the interplay between n-ad and m-hierarchy across the customer journey stage and for buyer/seller outcomes at various journey stages remains an open research area. Understanding drivers (e.g., contract size), moderators (e.g., buyer competitive market position), meditators (e.g., buyer risk tolerance), and outcomes (e.g., costs) of this interplay for customer journey stages would be insightful.
Further, the buyer's interplay of n-ad and m-hierarchy social structure should inform the n-ad–m-hierarchy structure choice and strategy formulation for the seller. To what extent and how should the seller's communication strategy focus on n-ad top-ranked buyer employees as opposed to the lower-ranked m-hierarchy employees? In certain situations (e.g., product-service provision situations such as brand guides and social media audits), targeting the lower-ranked m-hierarchy employees (e.g., millennial employees) might be beneficial, while in other situations, (e.g., getting qualified for RFP), targeting top-ranked n-ad employees is more crucial for the seller. Other research questions include the following[ 3]:
- What is the appropriate buyer n-ad–m-hierarchy social structure for each customer journey stage? Some journey stages (e.g., need activation) involve lower decision risk but higher need for due diligence than other stages. Understanding the interplay between risk (due diligence) and the associated n-ad–m-hierarchy social structure would benefit theory development and provide managerial insights.
- Given a particular n-ad–m-hierarchy buyer structure, how and why should a selling firm develop its n-ad–m-hierarchy structure across customer stages journey to satisfy customer needs and optimize account profitability?
In many buying/selling firms, lower-ranked employees interact to form their own n-ad social structures (e.g., an assistant procurement manager might interact with an assistant finance manager). Such interactions among lower n-ad social structures can influence higher n-ad interactions and (indirectly) the eventual outcomes across customer journey stages. To conceptualize these n-ad social structures across levels of buyer/seller hierarchies, we introduce the concept of p-s social influences. We define p or primary social influences among top-ranked n-ad and s or support social influences among lower ranked n-ad social structure. We provide an illustrative example in Panel C of Figure 1, which replicates Panel B and adds a support 2-ad influence between lower hierarchy finance and procurement employees, indicated by dotted bidirectional arrows. Thus, we differentiate between primary n-ad among top-ranked employees (i.e., n-ad-p, as in the upper triad, solid bidirectional arrows) and support n-ad among lower-ranked employees (i.e., n-ad-s, lower bidirectional arrow for the lower dyad).
How does the inclusion of n-ad-s social structure enrich our understanding of social influence in the buying/selling firm?[ 4] As n increases for n-ad-s social structure, the resolution of many administrative and tactical issues occurs at the support level as opposed to the primary level. Such resolution reduces cognitive load in the n-ad-p structure and enables efficacious decision making for top-ranked employees. However, as n increases for the n-ad-s social structure, the potential for information loss across the n-m-p-s structure increases, which might increase the risk of suboptimal decisions. Any miscommunication among n-ad-s employees might go unnoticed until a subsequent customer journey stage. Theoretically, as n increases for n-ad-s social structures, the potential for free riding and blame gaming could increase given the difficulty in tracking down information flow and social influence in a complex network.
An important question is this: What is the appropriate balance of n-ad-p, m-hierarchy, and n-ad-s social structures at each stage of customer journey for buying/selling firm? For example, should the n-ad-s social structure exactly mirror the n-ad-p structure? For the 3-ad-p triad structure in Panel C, using a 3-ad-s structure would imply that finance, procurement, and the site manager at the primary level influence one another, as do their counterparts at the support level. Such a structure would reduce information loss because three-way communications occurs at primary and support levels. However, such a structure does introduce redundancies due to duplication of n-ads and thus alter organizational cost structure for the journey stage. How these trade-offs play out remains unexplored.
Extant B2B research only studies an aggregate conceptualization of these social influence structures and therefore does not distinguish between n-m-p-s structures or consider the impact of customer journey stage. For example, [ 4] study how within-seller network density (i.e., the ratio of actual connections within selling organization to maximum number of connection possible within selling organization) influences seller account profitability. However, Gupta et al. do not differentiate between n-m-p-s structures and thus cannot shed light on when and how n-ad-p, m-hierarchy, or n-ad-s dominates.
Although we focus on n-m-p-s social structures in B2B firms, some concepts might also be relevant for B2C journey research. For example, when a consumer looks to purchase a high-ticket item, they may consider inputs from other household members, thereby creating n-ad social influence structures. To the extent that consumers use shopping assistants (e.g., human, AI) to help them reduce information burden, they create m-hierarchy social structures. However, n-ad-p and n-ad-s structures are not as relevant to B2C markets, as consumers' assistants seldom interact with one another. B2C sellers already account for the notion that consumers engage other household members and shopping assistants while making purchase decisions. If sellers understand customers' n-ad–m-hierarchy structures (e.g., decision maker, spouse, and AI), and how they change during journey stages (e.g., only decision maker makes the final purchase decision, but the other n-ad actors engage in information search), they can tailor their communications. Thus, we hope that our conceptualization of n-m-p-s social structures in B2B firms also spurs future B2C research.
The richness of B2B customer buying and the three elements of social structure (i.e., n-ad, m-hierarchy, and p-s) raise at least two additional research topics worthy of future research: structural dynamics and key performance indicators (KPIs).
The social structure elements we propose (i.e., n-m-p-s) evolve across- and within-customer journey stages, presenting many opportunities to study structural dynamics (i.e., coevolution of social structure elements and buyer/seller decisions and outcomes). For example, a buyer may want some common actors across journey stages. Which actor, n-ad-p and/or n-ad-s, should continue across particular journey stages? There is an inherent trade-off as greater overlap would reduce information loss but might also perpetuate groupthink. Similarly, institutionalized knowledge about structural dynamics and associated outcomes can help buyer/seller use the knowledge base to navigate future social structure decisions in customer journeys. Knowing the relationship between buying/selling configuration and associated outcomes at each stage should build organizational memory for managing future buyer/seller relationships. Structural dynamics for customer social journey in B2B marketing remain unresearched due to aggregate conceptualization of social networks.
Researchers can also offer insight into KPIs that measure the benefits and costs associated with n-ad, m-hierarchy, and p-s social influences at different journey stages. These metrics can provide buyers and sellers with potential governance mechanisms. For example, as n-ad increases for primary influence, the potential for free riding should also increase. The principal could use process and outcome KPIs to monitor the contributions of the n-ad agents. Similarly, as m-hierarchy becomes complex, the power dependence becomes asymmetric, and the principal might need to rely on contractible metrics to mitigate the deleterious effects of the asymmetry. The whole domain of governance within buyer, within seller, and for buyer–seller relationships across the customer journey stages remains an open area for research.
The B2B customer journey is social by definition, with the actors, their roles, and their relationships changing across the journey as it is cocreated. Thus, we augment the conceptualization of HFHM to highlight the unique social aspects of the B2B customer journey. In doing so, we formalize the social influence structure in the B2B customer journey by introducing elements of n-ad social structures, m-hierarchy social structures, and p-s social influences. This formalization enriches the current sparse research on B2B buying that adopts a social network perspective. We suggest illustrative research questions that we hope will spawn future research on the B2B social customer journey.
Footnotes 1 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
2 Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
3 Many research questions parallel those posed by HFHM, such as the following: (1) How are conflicts within the selling firm at different stages of the customer journey reconciled? and (2) What are the sources of power and social influence within the selling firm at each stage of the customer journey?
4 We hope it is evident that our earlier discussions on n-ad apply for the n-ad-p social structure.
References Best Roger, Mittal Vikas, Sridhar Shrihari. (2021), Market-Based Management, 7th ed., forthcoming (Independent).
Dwyer F. Robert, Schurr Paul H., Oh Sejo. (1987), "Developing Buyer–Seller Relationships," Journal of Marketing, 51 (2), 11–27.
Grewal Rajdeep, Lilien Gary L. (2012), " Business-to-Business Marketing: Looking Back, Looking Forward," in Handbook of Business-to-Business Marketing, Lilien Gary L., Grewal Rajdeep, eds. Northampton, MA : Edward Elgar, 3–12.
Gupta Aditya, Kumar Alok, Grewal Rajdeep, Lilien Gary L. (2019), "Within Seller and Buyer Seller Network Structures and Key Account Profitability," Journal of Marketing, 83 (1), 108–32.
5 Hamilton Ryan, Ferraro Rosellina, Haws Kelly L., Mukhopadhyay Anirban. (2021), "Traveling with Companions: The Social Customer Journey," Journal of Marketing, 85 (1), 68–92.
6 Sheth Jagdish N. (1973), "A Model of Industrial Buyer Behavior," Journal of Marketing, 37 (4), 50–56.
~~~~~~~~
By Rajdeep Grewal and Shrihari Sridhar
Reported by Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 34- Commentary: Trajectories and Twists: Perspectives on Marketing Agility from Emerging Markets. By: Hughes, Nick; Chandy, Rajesh. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p59-63. 5p. DOI: 10.1177/0022242920973037.
- Database:
- Business Source Complete
Commentary: Trajectories and Twists: Perspectives on Marketing Agility from Emerging Markets
Age-old ways of making marketing decisions—caricatured by annual plans, quarterly reports, and inflation-adjusted budgets—are being questioned today, and new, agile, and presumably more effective approaches are being pitched and embraced (see [ 1]; [ 9]). So it feels timely and right to focus on marketing agility as a research subject, especially as markets are changing dramatically and technology allows us to radically change the way we develop and market products and services. Given this reality, [ 6], hereinafter Kalaignanam et al.) provide a valuable contribution to the field by offering a precise definition of a concept that is—like many concepts that attain buzzword status—too often fuzzily defined and loosely framed. By differentiating marketing agility from other related concepts and highlighting some its drivers and boundary conditions, they have also provided a means for practitioners to think clearly and comprehensively about the concept and to benefit from the insights of others who have implemented the idea.
In this commentary, we offer a perspective on Kalaignanam et al. by focusing on digital products in emerging markets. Why? Because just as automakers developing new vehicles assess their robustness and durability in extreme conditions of heat and cold, we aim to assess the robustness and durability of the Kalaignanam et al. perspective on marketing agility in extreme conditions of uncertainty and change. Digital products and emerging markets, for reasons we argue subsequently, represent such extreme conditions in which to assess the idea of marketing agility. Those making decisions in extreme conditions can serve as—to use the memorable phrase of [ 7]—the "advance guard" for others seeking to make agile decisions in periods of great change. Having witnessed and (in the case of Nick Hughes) participated in the dramatic growth of digital products and services in Africa and Asia, we hope to contribute a perspective on marketing agility that assesses and complements Kalaignanam et al.
If there is any context that demands marketing agility—the rapid iteration between making sense of markets and executing marketing decisions to adapt to changing market assessments—then digital products in emerging markets represent such a context. As Kalaignanam et al. rightly note, marketing agility is especially appropriate in contexts where the environment is characterized by rapid change and market response is unpredictable. Many emerging markets are currently undergoing a process of change that can seem dizzying by Western standards. Consider the market environment facing the typical resident of East Africa compared with the typical American: electrical appliances, telephones, banking services, smartphones, and digital commerce all arrived more or less concurrently in East Africa, and brand-new highways are now connecting the rapidly urbanizing workforce to their rural families. Climate change and global warming are a living reality and are forcing large-scale changes to already fragile livelihoods and lives.
The changes that much of the contemporary West has experienced (or is yet to experience) was scattered over decades, even centuries. Emerging markets, in contrast, are undergoing change that is compressed in time ([ 2]). In some ways, William Gibson's memorable quote "The future is already here—it's just not evenly distributed" (Economist 2001) could be restated as "The past, present, and future are already here—they're in emerging markets." Like rivers joining together to create unpredictable eddies and uncertain navigation, the confluence of technological, social, economic, and environmental forces amplifies unpredictability and creates uncertainty for digital businesses seeking to succeed in emerging markets (see [ 3]). If agility should matter anywhere, it is here.
Our experience (as a practitioner and an academic, respectively) with digital products in emerging markets has been that agility in thinking and execution is crucial for successful marketing decisions in these rapidly changing markets. This experience also suggests areas for further conceptual and empirical development around the concept of agility.
Kalaignanam et al. propose four dimensions to marketing agility: sensemaking, iteration, speed, and marketing decisions. Of these dimensions, we focus on sensemaking and iteration in this commentary because (in our opinion) these dimensions are especially ripe for new insights.
We propose that sensemaking for agility involves not only responding to "unexpected or ambiguous" developments (Kalaignanam et al., p. 000); it also involves drawing trajectories—simplified representations of patterns that underlie changing market phenomena. Trajectories can be contextual (involving patterns related to customers, competitors, and the macro-environment) and/or strategic (involving patterns related to a focal firm). Contextual trajectories could include trends in technology, demographics, regulations, or environmental factors. Strategic trajectories could include trends in thinking or actions on the part of decision makers in the focal firm or stakeholders such as investors or suppliers. Why is an understanding of trajectories important to marketing agility?
First, trajectories can help decision makers anticipate outcomes in fast-changing environments ("skate where the puck is going to be," as Wayne Gretzky famously said) and help them quickly prioritize decisions and actions. Trajectories offer direction to decision makers and help them aggregate across the many underlying forces that might cause a particular pattern of change. Moreover, trajectories help decision makers contextualize change and more easily understand which of the forces of change operating around the world are most relevant to their own context. By drawing trajectories, decision makers can not only make sense of unexpected or ambiguous new contextual developments, but also position these developments in the context of an evolving understanding of their own options regarding the future. Second, by integrating contextual and strategic trajectories, firms can make sense of entirely new developments, building confidence before investing substantial resources rapidly in new courses of action. Furthermore, by integrating trajectories, managers can narrate a story that helps external and internal stakeholders make sense of the need for change—and thus justify agile behavior in anticipation of or in response to change.
We propose that iteration involves not only "repeatedly refining marketing decisions" (Kalaignanam et al.); it also involves engaging in strategic twists: swift and substantial changes in the direction of activities. Unlike the generic notion of iteration, which involves the "repetition of a sequence of operations to yield results successively closer to a desired result" ([ 8]), engaging in twists is a process that can require doing entirely different tasks from what is currently being done (and therefore not repeating a particular sequence of tasks). It involves discovery of new insights, possibilities, and goals relative to what is currently being understood or pursued. Moreover, rather than refining existing marketing decisions, it can involve actively seeking to reframe the marketing decisions that are warranted given changes in marketplace phenomena and communicating the ensuing changes in internal and external narratives.[ 3]
In the sections that follow, we illustrate these concepts and arguments by outlining two case studies of agility in marketing decisions. Both case studies are of digital products in Africa: the M-PESA mobile money service, and the M-KOPA solar energy service. One of the authors of this commentary— Nick Hughes—helped lead the development of each of these services; so we offer first hand insights that embody agility within a framework of trajectories and twists. We note that both cases followed somewhat analogous processes of sensemaking by anticipating and integrating trajectories. Moreover, both cases involved discovering unexpected market insights and reframing ideas to implement twists.
The M-PESA mobile money service is responsible for financial transactions equivalent to almost half of Kenya's gross domestic product. This financial service contributed over $787 million in 2019–2020 revenues to Safaricom, the Kenyan mobile operator (or carrier) and affiliate of UK-based Vodafone; M-PESA now contributes more to Safaricom's revenues than traditional mobile services such as texts and mobile data, and close to its revenues from voice calls ([10]). But just 15 years ago, M-PESA was a prototype, and the team working on it was pursuing an application that was—to their surprise—not what very many customers wanted. We illustrate the trajectories and twists involved in marketing agility by first describing the M-PESA case.
M-PESA was created in part because Nick Hughes and his team were able to anticipate and integrate multiple trajectories.
Despite the obvious need for financial services, the availability of formal financial services had been limited for most of those living in emerging markets. Almost three-fourths of Kenyans had no access to formal financial services in 2006 ([ 4]). At the same time, micro-credit was growing in prominence thanks in part to Muhammad Yunus's Nobel Prize–winning work in Bangladesh. Further, the trajectory of mobile phone penetration in many emerging markets including Kenya was steep and positive. The original logic for M-PESA was therefore based on the anticipation that the micro-finance and mobile penetration trajectories could be applied to address the need for financial services in emerging markets.
To integrate these distinct trajectories and conceive of profitable products from them, Nick and his team used the analogy of the last mile problem in transportation: suburban commuters often need more time getting from their nearest train station to their homes a mile away than for getting from the station to a city center much farther away. Similarly, there were many hassles (and high costs) in moving cash between customers and micro-loan branches. "What if customers and their micro-loan providers could simply text their payments to each other?" was the original idea. Just as commuter parking lots could solve the last mile problem in suburban transportation, could mobile payments solve the last mile problem in microfinance?
To take forward the idea of using mobile telephony for financial services, Nick had to integrate the new idea with the strategic trajectories of the company as well as external allies. Within Vodafone, he pitched the idea as a growth platform that matched its emerging market aspirations. To external funding agencies, he noted its potential to improve the trajectory of financial services for the unbanked. His first major break came when he won a grant from the UK government to conduct a pilot of a mobile-based micro-credit platform. Thanks in part to this funding, Safaricom, then a young company within the Vodafone Group —and one that was fast establishing a strong, trusted brand in Kenya and a dominant (>60%) market share—agreed to implement the idea in a Kenyan test market. Nick's team built and deployed a prototype mobile loan disbursement and repayment scheme, and started receiving a stream of transaction data related to customer accounts, balances, and transactions.
The pilot revealed surprising lessons: customers were behaving very differently from what the team had expected. By sketching out patterns of customer behavior, they made an unexpected discovery that led to a conceptual twist that reframed their product and market strategy.
Nick and his team discovered from the pilot data that customers were doing unexpected things. For example, some loan group members started sending other group members money that did not relate to a loan repayment; many customers were loading more funds into their accounts than was required to repay their loans and leaving these funds in their digital wallets for extended periods. These unexpected applications of the product were clearly valued by customers—storing funds safely in electronic form and having a convenient, low-cost way to send electronic money to friends, family, and business contacts. Importantly, the M-PESA team discovered that when customers had the opportunity to borrow easily from family, friends, and acquaintances, they were less keen to borrow from a micro-finance company. This behavior on the part of customers also alerted the team to a trajectory that they had not emphasized before: the rapid increase in urbanization in Kenya, which involved migration to cities by those who would then contribute through remittances to those "back home" in villages.
The M-PESA team quickly engaged in a strategic twist. The grant funding from the UK government meant that Nick's team had the autonomy to change strategy quickly in response the market insights. They abandoned the complexity of a micro-credit platform and launched M-PESA under the marketing banner "send money home." Safaricom's leadership accepted the new framing of the M-PESA proposition, and contributed their in-market resources—their brand, market presence, and distribution—to the new proposition. Their chief executive officer and chief financial officer provided the strong operational drive once the reframed proposition was clear and compelling. The product tapped into a very rich vein of demand; customers signed up by the thousands per day, and usage took off at a rate that was both unexpected and unprecedented. M-PESA is now used in 96% of Kenyan households ([11]).
M-KOPA's foundations lie in the pay-as-you-go solar energy space, pioneering what is now an established sector for financed energy access across Africa. It counts over one million customers for its digital products and services across East Africa. In our view, this case provides another illustration of the trajectories and twists involved in marketing agility.
M-KOPA was also created in part because the company could anticipate and integrate multiple trajectories.
Over 600 million people—or almost two-thirds of the inhabitants of sub-Saharan Africa—lack access to reliable grid electricity. They use dirty, expensive, and potentially dangerous fuels like kerosene for lighting. Despite the intense need for energy, this trajectory has been stubbornly flat. Other trajectories offered cause for optimism and opportunities for innovation. The price of solar energy per kilowatt hour has fallen by a factor of three since 2010. The rapid rise of electric vehicles and portable consumer electronics meant that batteries have become better and cheaper. Mobile signal coverage is increasingly ubiquitous and can be used to create an "internet of things." And, of course, the emergence of digital payment services such as M-PESA has meant that consumers can engage in financial transactions easily and cheaply.
M-KOPA's initial concept was to sell connected solar home power systems that customers could pay for using digital money over the course of 12 months. By remotely controlling the solar home system and only allowing it to only work on receipt of a digital payment, M-KOPA was effectively operating a "coin-in-the-meter" model to provide access to clean energy. The company could risk taking the hardware onto its balance sheet and "micro financing" access to this asset for customers with sporadic household income. To the customer, the proposition is simple: "spend 50 Kenya Shillings (roughly US$.50) a day with M-KOPA instead of spending that amount on kerosene and you'll get clean, safe power—plus you are buying your household an asset that once paid for, will bring clean energy for nothing." M-KOPA's analysis suggests that the typical household saves around $650 over the lifetime of the solar home system.
The promise of profitably fulfilling an intense customer need by applying the powerful trajectories outlined in this section meant that the founders could credibly integrate the M-KOPA concept with the strategic objectives of a variety of impact-oriented investors. These included investors with a particular interest in addressing the trajectory of environmental degradation and global warming, foundations seeking to improve the trajectory of social and economic impact and development finance institutions such as the Foreign, Commonwealth and Development Office, part of the UK government.
Very quickly after launch, a competitive trajectory became evident as well. Other players, who were able to manufacture lower-cost hardware, began to match M-KOPA on cost and on the product bundle (e.g., number of lights, accessories). M-KOPA had to find a way to differentiate. A twist involving an unexpected discovery and a reframing of the product idea and market approach emerged.
The M-KOPA team discovered that customers who had built a digital payment relationship with M-KOPA were demonstrating unique behavioral trajectories. Because M-KOPA can observe the payment history of each customer, the team was able to determine that many customers were paying off the M-KOPA system more quickly than expected. Because each solar home system is always connected to a mobile network, the team is able to assess the load on each battery, which indicates which customers use their systems most heavily. The payment and usage history of each customer provided M-KOPA with data to segment customers into those who might be profitably served (as described subsequently) by loans that used the solar home system as collateral.
This discovery has meant that M-KOPA has gone from offering just a solar home system to offering a financial service. While the initial offer still involves a hardware product(s) in the form of the solar home system (comprising, e.g., a photo voltaic panel, battery, LED lights, DC TV), M-KOPA has engaged in a strategic twist of its model toward a financial service that uses the payment and usage profile of each paying customer to offer customized loans. Today, more than 50% of M-KOPA's customers have taken a second or third digitally enabled loan from M-KOPA, collateralized against their solar power system. Loan types range from "cash back" to school fee loans, credit for water tanks, and even fertilizer for small farms. This twist is predicated on maintaining a payment relationship and offering relevant, timely credit products. It would not have been possible if M-KOPA did not have information on customer trajectories for segmentation and offer (as well as finance) an ever-changing set of products and services for the "connected African home." The company is now building AI tools to help it profile customers as early as possible in their relationship and thus move quickly to grow lifetime customer revenue opportunities where margins are higher.
Kalaignanam et al. have provided a valuable contribution to research and practice. As with many thought-provoking articles—and especially so for articles that promote a new research area—their framework raises even more questions than it answers directly. The qualitative insights in their article are largely based on input from large firms in the U.S. market. Our assessment, which is based on the somewhat "extreme" context of digital products in emerging markets, suggests that the definition and framework of agility that they provide could be widely applicable across contexts.
We seek to complement the Kalaignanam et al. perspective by introducing the concepts of trajectories and twists as a way of framing how firms might apply marketing agility in fast-moving environments. We present M-PESA and M-KOPA as two cases of new digital services that embody elements of marketing agility, the need for which is made acute by being set in the context of emerging markets. In both cases, decision makers anticipated multiple market trajectories, integrated market trajectories with the strategic trajectories of the firm and its partners, and engaged in strategic twists that involved discovery of new market insights and reframing of problems or opportunities based on these insights.
Given the dramatic changes that are evident throughout the world today, we are hopeful that the early signals of interest in marketing agility will yield a steep trajectory of new research and informed practice.
Footnotes 1 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
2 Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
3 Others have used phrases such as "pivots" ([9]) and "dynamic course corrections" ([5]) to refer to ideas similar to what we mean by twists. We prefer the latter term because it (1) allows for changes in mental models and narratives (e.g., "twists in the tale"), not just actions, and (2) it implicitly acknowledges that there may be a cost to engaging in twists (i.e., twists can be injurious as well).
References Brown Shona L., Eisenhardt Kathleen. (1998), Competing on the Edge: Strategy as Structured Chaos. Boston : Harvard Business Press.
Chandy Rajesh, Om Narasimhan. (2015), "Millions of Opportunities: An Agenda for Research in Emerging Markets," Customer Needs and Solutions, 2 (4), 251–63.
Davies Richard. (2020), Extreme Economies: What Life at the World's Margins Can Teach Us About Our Own Future. London : Farrar, Straus and Giroux.
4 FinAccess (2019), "The 2019 FinAccess Household Survey," (accessed 14 October 2020), https://fsdkenya.org/publication/finaccess2019/.
5 Giesen Edward, Riddleberger Eric, Christner Richard, Bell Ragna. (2010), "When and How to Innovate Your Business Model," Strategy & Leadership, 38 (4), 17–26.
6 Kalaignanam Kartik, Tuli Kapil R., Kushwaha Tarun, Lee Leonard, Gal David. (2021), "Marketing Agility: The Concept, Antecedents, and a Research Agenda," Journal of Marketing, 85 (1), 35–58.
7 Keynes John M. (1932), " Economic Possibilities for Our Grandchildren (1930)," in Essays in Persuasion. London : Palgrave Macmillan, 321–32.
8 Merriam-Webster's Collegiate Dictionary (1999), "Iteration." Springfield, MA : Merriam-Webster Incorporated.
9 Reis Eric. (2011), The Lean Startup. New York : Crown Business.
Russel A. (2020), "M-PESA Now Makes Up a Third of All Safaricom Revenue," Tech in Africa, (April 29), https://www.techinafrica.com/16629-2/.
Suri Tavneet, Jack William. (2016), "The Long-Run Poverty and Gender Impacts of Mobile Money," Science, 354 (6317), 1288–92.
~~~~~~~~
By Nick Hughes and Rajesh Chandy
Reported by Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 35- Communication in the Gig Economy: Buying and Selling in Online Freelance Marketplaces. By: Ludwig, Stephan; Herhausen, Dennis; Grewal, Dhruv; Bove, Liliana; Benoit, Sabine; de Ruyter, Ko; Urwin, Peter. Journal of Marketing. Jul2022, Vol. 86 Issue 4, p141-161. 21p. 5 Charts, 2 Graphs. DOI: 10.1177/00222429211030841.
- Database:
- Business Source Complete
Communication in the Gig Economy: Buying and Selling in Online Freelance Marketplaces
The proliferating gig economy relies on online freelance marketplaces, which support relatively anonymous interactions through text-based messages. Informational asymmetries thus arise that can lead to exchange uncertainties between buyers and freelancers. Conventional marketing thought recommends reducing such uncertainty. However, uncertainty reduction and uncertainty management theories indicate that buyers and freelancers might benefit more from balancing—rather than reducing—uncertainty, such as by strategically adhering to or deviating from common communication principles. With dyadic analyses of calls for bids and bids from a leading online freelance marketplace, this study reveals that buyers attract more bids from freelancers when they provide moderate degrees of task information and concreteness, avoid sharing personal information, and limit the affective intensity of their communication. Freelancers' bid success and price premiums increase when they mimic the degree of task information and affective intensity exhibited by buyers. However, mimicking a lack of personal information and concreteness reduces freelancers' success, so freelancers should always be more concrete and offer more personal information than buyers. These contingent perspectives offer insights into buyer–seller communication in two-sided online marketplaces. They clarify that despite, or sometimes due to, communication uncertainty, both sides can achieve success in the online gig economy.
Keywords: business-to-business exchange; gig economy; multisided platforms; online freelance marketplaces; text analysis; uncertainty management
Online freelance marketplaces, such as Upwork, Fiverr, and PeoplePerHour, have prompted massive transformations in business-to-business (B2B) markets ([13]; [82]). In particular, they allow buyers to post gigs, or short-term service projects, which initiate reverse auctions whereby interested freelance workers submit bids to offer their services ([39]). In these digital environments, buyers and freelancers often devote rather limited time and attention to detailed assessments and instead make choices on the basis of rational value expectations or prices ([ 2]). In addition, online freelance marketplaces suffer from information asymmetries because they rely on text-based messages, which can create uncertainty and hinder the exchange ([34]; [77]). Imagine a buyer wants to hire a freelancer to optimize their pet website's search rankings, so they post a call for bids, requesting "someone for an SEO job." In response, Freelancer A might vaguely promise, "I have plenty of experience writing content that users find interesting to improve the quality and quantity of your traffic," whereas Freelancer B more concretely states, "I have four years of experience writing articles and blogs that engage users and are SEO-friendly. For example, I could focus on interest pieces like the everyday lives of pets." The communication of both the buyer and freelancer create different degrees of uncertainty, likely impacting who applies and who gets hired.
Uncertainty regarding communication can lead to various negative outcomes on both sides, including high rates (more than 50%) of service gigs that go unfulfilled ([36]), diminished bid success, and less-than-optimal pricing for freelancers ([ 2]). However, parties in B2B exchanges can also strategically leverage uncertainty in their communication to achieve more effective outcomes, such as when negotiators conceal information ([69]) or when ambiguous contracts help reduce litigation concerns and increase cooperation ([81]). Buyers and freelancers on online freelance marketplaces engage in a form of B2B exchange, so we propose that they similarly might balance their communication efforts by strategically reducing and increasing uncertainty to maximize their exchange success. In our previous example, by staying vague and without any specific direction from the buyer, Freelancer A might be trying to keep multiple options open and avoid overpromising outcomes.
In addition to fundamental questions regarding how to manage uncertainty in B2B exchanges ([50]; [63]), we seek to address the role of communication in such exchanges ([ 7]; Rajdeep et al. 2015; see also Web Appendices A and B). We integrate uncertainty reduction theory ([ 6]) and uncertainty management theory ([ 9]) to predict that, in online freelance marketplaces, various strategies for reducing and increasing the ability of message recipients to anticipate message senders' meaning and actions can benefit the exchange ([ 8]). Using Grice's ([27]) communication principles, we argue that greater provision of task and personal information might reduce uncertainty in service exchanges ([54]) but could also lead to information overload or disagreements ([18]; [40]). Presenting information in a more concrete (cf. abstract) manner or with greater affective intensity also can reduce uncertainty ([27]; [28]; [61]). But again, too much concreteness or affective intensity might lead to restrictive communication that hinders exchanges ([18]; [37]).
We apply this theoretical reasoning to exchanges in online freelance marketplaces, in which buyers post calls for bids to attract as many freelancers as possible to apply ([36]). These buyers face a trade-off between reducing uncertainty for freelancers (e.g., providing more information, using less ambiguity) and still efficiently granting them sufficient interpretative freedom. Theorists concur that principles for using relevant information or less ambiguity often get deliberately flouted in conversation, such as when an individual is attempting to save face ([23]) or please a counterpart ([44]). If different communication strategies might entice more freelancers to bid, buyers could establish optimal designs for calls for bids.
In response to those calls for bids, freelancers write and submit their bids. In doing so, these freelancers also must manage uncertainty. Thus, they might benefit from matching or mimicking the communication approach adopted by the prospective buyer that issued the call ([78]). Communicative mimicry can evoke similarity perceptions, which tend to increase receivers' sense of rapport and reduce their uncertainty ([75]). Research on adaptive selling recommends matching the buyer's communication (e.g., [57]; [72]). However, in some situations, deviations also may be beneficial ([ 1]), so we consider a more nuanced distinction related to the level at which the similarity occurs. Furthermore, if freelancers compete on price, they may become enmeshed in a self-defeating value trap ([34]; [76]) in which they win more bids but earn less revenue. Strategically mimicking or deviating from a buyer's communication might provide a viable means to winning more gigs without being trapped. We accordingly suggest how freelancers should calibrate their bid formulations to improve their bid success and achieve a price premium.
Using a unique, large-scale data set of calls for bids and bids, obtained from a leading online freelance marketplace, along with a series of multilevel models that account for endogeneity, we establish three main contributions. First, we determine the effects of buyers' strategic communications in two-sided online marketplaces ([ 7]). Rather than uncritically recommending that communication should always be informative and unambiguous, we specify the diminishing, even adverse consequences that can result if buyers relay too much task or personal information in a very concrete, intense manner. Second, in an extension of research into adaptive selling ([57]; [78]), we reveal how freelancers' dyadic communicative mimicry affects bid success. Mimicry effects are contingent on the communicative aspect and the buyer's relative uses of each aspect. As we show, mimicking buyers in terms of the provision of task information and use of affective intensity increases bid success. In contrast, we find that freelancers should always offer more personal information and be more concrete in their bid formulations than buyers' calls for bids. Third, we offer insights into how freelancers can avoid predatory pricing ([13]) and escape a value trap ([76]). By strategically managing uncertainty according to the information communicated, and by managing the manner in which they do so, freelancers can earn price premiums.
Online freelance marketplaces that feature reverse auctions rely on a three-stage process ([34]; [39]). First, in seeking a suitable freelancer, a buyer describes a gig or short-term service project in a call for bids. Second, multiple freelancers apply by formulating and submitting bids that describe themselves, the service offering, and the price requested. Third, the buyer compares the bids and selects a freelancer to complete the project. The outcome of each stage defines exchange success. That is, buyers' success results from a large pool of viable freelance offers. A higher number of bids increases the chances of finding a suitable freelancer for the gig ([35], [36]). Freelancers' success depends on whether their bids are chosen, preferably at a price premium ([13]; [32]). In this context, a price premium is the monetary amount in excess of the buyer's original payment offer (i.e., expected price; [26]; [73]). Buyers might pay a premium beyond their original payment offer for various reasons, including their willingness or "need to compensate the seller for reducing transaction risks" ([ 2], p. 248). In competitive online marketplaces, freelancers also might encounter value traps in which they wind up selling more of their services at a lower price ([ 2]; [76]). In this sense, freelancers' success depends on winning the bid but also earning price premiums (or avoiding discounts). Unlike traditional B2B exchanges, buyers' and freelancers' success hinges on textual communication ([13]; [36]). Comparing theories on uncertainty and the role of communication in producing or reducing it, we delineate how both buyers and freelancers may best strike a balance between providing more information and reducing ambiguity versus preserving some uncertainty and maintaining interpretative flexibility.
Uncertainty reduction theory ([ 6]) and uncertainty management theory ([ 9]) draw on a central tenet of information theory ([71])—namely, that communication, information, and uncertainty are inextricably linked. Thus, uncertainty is inherent to any interaction. [22] suggests uncertainty depends on the ability to draw inferences from provided information content and the manner in which it is provided. Whereas uncertainty reduction theory predicts how communication can reduce uncertainty, uncertainty management theory examines how people cope with uncertainty, which may include efforts to increase uncertainty to attain beneficial outcomes ([ 8]). Our conceptual development relies on these fundamental principles.
In online freelance marketplaces, buyers and freelancers depend on one another; all else being equal, they want their mutual exchange to succeed. In such interactions, [27] suggests that four generalized cooperative communication principles (or maxims) apply. Three principles refer to what should be said: the quantity of information ("give as much information as is required and no more than is required"), its quality ("do not say what is false or that for which you lack adequate evidence"), and its relevance. The fourth principle, manner (be clear and avoid ambiguity), pertains to "how what is said is to be said" ([27], p. 46). In our study context, neither a buyer nor a freelancer can know upfront whether the other party might be lying, so truthfulness would have to be assumed prior to the exchange. We also highlight that information does not have to be "correct" to influence uncertainty perceptions ([ 9]). Therefore, among the four maxims, we focus on the quantity of relevant information that buyers and freelancers offer and the manner in which they present it.
Communication outcomes are fundamentally uncertain ([ 6]). When people vary their use of communication principles ([27]), they create conversational implications such that message recipients must infer what speakers are trying to imply with their wording. Accordingly, the (un)certainty that buyers and freelancers encounter while making inferences should depend on the degree to which calls for bids and bids provide relevant information in an unambiguous manner, though the meaning of relevant information varies by context. In line with prior research (e.g., [ 7]), we define this degree as the proportion of specific lexical terms used relative to the total number of words in a message.
More information reduces uncertainty ([ 6]) and increases receivers' perceptions of the information's value ([79]). In service exchanges, the parties seek information about the task and the person who will complete it ([54]). A greater degree of task information should reduce uncertainty about functional service aspects ([54]). By self-disclosing greater degrees of personal information, a sender also provides a receiver with more information about the self ([15]). In line with the quantity principle ([27]), sparse provision of relevant task and personal information would make it difficult for the receiver to anticipate outcomes or distinguish among options, thus creating uncertainty ([19]).
Regarding the principle of manner ([27]), greater degrees of concreteness and affective intensity should reduce ambiguity and enhance clarity. Concrete terms describe something in a perceptible, precise, specific, or clear manner ([11]; [49]; [61]). A greater degree of concreteness reduces ambiguity because it makes it easier for receivers to perceive or recognize what the message sender is implying ([11]; [28]; [61]). Affective intensity reflects the proportion of affective terms included in a message. More affective terms as a proportion of the total word count produce a greater degree of affective intensity, which increases receivers' ability to make evaluative judgments ([28]; [37]).[ 5] We provide examples of these principles in Table 1.
Graph
Table 1. Communication Elements, Links to Uncertainty, and Examples.
| Communication Element | Definition | Link to Uncertainty | Example |
|---|
| Task information | A content element of communication. In service exchanges, it is conveyed through functional, duty terms (Ma and Dubé 2011). The proportion of task terms to the total number of words in a message defines the degree of task information. | Greater (lesser) degrees of task information decrease (increase) uncertainty. | Sparse degree of task information: "I saw your project description and I would like to work for you. I have plenty experience in different settings where I have written content which users find interesting."Dense degree of task information: "I saw your project description and would like to write the content for your website. I have experience in writing articles, blogs & E-books which is user engaging and SEO friendly as well." |
| Personal information | A content element of communication that is conveyed through self-disclosing terms (Derlega, Harris, and Chaikin 1973). The proportion of self-disclosing terms to the total number of words in a message defines the degree of personal information. | Greater (lesser) degrees of personal information decrease (increase) uncertainty. | Sparse degree of personal information: "Saw your project description and would like to write the content for your site. I have experience in writing articles, blogs & E-books which is user engaging and SEO friendly as well."Dense degree of personal information: "I saw your project description and I would like to write the content for your site. I have 12 years of work experience in copy writing for articles, blogs & E-books. I have a Master's in Journalism and have worked fulltime for companies like Adobe." |
| Concreteness | A manner element of communication conveyed by terms that are perceptible, precise, or specific (Brysbaert, Warriner, and Kuperman 2014; Packard and Berger 2020). The proportion of concrete terms to the total number of words in a message defines the degree of concreteness. | Greater (lesser) degrees of concreteness decrease (increase) uncertainty. | Sparse degree of concreteness: "I noticed your project description and I would like to do work on it. I have plenty of experience in scripting text, which is engaging, compelling, and SEO friendly."Dense degree of concreteness: "I saw your posted project description on Upwork, and I would like to write the contents for your website. I have a lot of experience in article and weblog writing in an SEO friendly fashion." |
| Affective intensity | A manner element of communication that is conveyed through affective terms (Hamilton and Hunter 1998). The proportion of affective terms to the total number of words in a message defines the degree of affective intensity. | Greater (lesser) degrees of intensity decrease (increase) uncertainty. | Sparse degree of affective intensity: "I saw your project description and I can write the required content for your site. I have plenty of experience in writing articles, blogs & E-books which is user engaging and SEO friendly as well."Dense degree of affective intensity: "I liked your project description and would be happy to write the content for your site. I have great experience in writing articles, blogs & E-books which is user engaging and SEO friendly as well." |
Cross-disciplinary research provides ample evidence that conversational partners generally prefer to reduce uncertainty ([ 5]). In B2B relationships, reducing uncertainty increases exchange effectiveness ([29]; [50]; [63]). In Web Appendix A, we offer an overview of some key empirical marketing studies on B2B communication aspects. Specifically in online freelance marketplaces, which are relatively anonymous, the required coordination and dependence between rational buyers and freelancers may increase their need for information and clarity ([13]; [34]). Thus, for example, reputation cues commonly appear in online freelance marketplaces as a way to reduce uncertainty and facilitate exchanges ([34]). More broadly, reducing uncertainty by adhering to [27] principles in dyadic buyer–freelancer communications may boost exchange success.
However, people experience uncertainty differently and do not always prefer to reduce it ([ 8]). Instead, according to uncertainty management theory ([ 9]), strategic communication choices that might not minimize uncertainty, and even cultivate it, can be effective and lead to better outcomes for consumers ([38]), organizations ([18]; [31]), and interorganizational governance ([81]). For example, [38] find that a lack of concreteness aids consumers' initial online searches because such vague queries return a greater variety of search results. In collective bargaining settings, seasoned negotiators use concealment and ambiguity to enhance the likelihood of agreement ([69]). In B2B exchanges, parties can use less information and more ambiguity strategically to accomplish specific goals ([ 3]; [81]). Even if such efforts are not universally favored, uncertainty-cultivating communication provides benefits by allowing different receivers to perceive multiple different meanings simultaneously ([18]). Moreover, communication theorists concur that people sometimes deliberately flout or violate [27] conversation principles, such as when they intentionally maintain uncertainty to save face ([23]) or please a counterpart ([44]). Subverting the principles is not necessarily less cooperative, and furthermore, the purpose of communication is not always to be as informative and clear as possible. Arguably, cooperative principles encourage reasonable adherence, not compulsion. Thus, strategically allowing recipients to develop a broader range of possible interpretations by maintaining some level of uncertainty might facilitate buyer–freelancer exchanges.
Freelancers choose whether to offer their services in response to a buyer's call for bids. The number of freelancers who choose to do so is consequential for the buyer, as more bids implies a greater likelihood of finding a suitable service provider ([36]). Managing freelancers' uncertainty through relevant information provision and the manner of communication in the calls for bids should influence freelancers' decisions to apply.
In calls for bids, buyers can vary the degree of task and personal information included in the description of the gig. If freelancers evaluate this information favorably, they develop more positive dispositions and are more likely to apply ([72]). As prior research establishes, more information enhances communication outcomes in business settings by reducing uncertainty. For example, studying web forums, [79] indicate that the breadth of information provided by a sender affects receivers' objective judgments of the value of that information. [49] find that greater degrees of monetary information increase peer-to-peer lending, and [41] shows that more task information increases the time and commitment sellers allocate to a buyer. Greater degrees of personal information also reduce uncertainty, increase trust ([55]), and enhance performance on crowdsourcing platforms ([68]). Such self-disclosure can strengthen ongoing buyer–seller relations as well ([14]). In contrast, a greater proportion of nonrelevant information (i.e., a lesser degree of relevant information) increases uncertainty ([ 9]). Because greater degrees of task and personal information in calls for bids help reduce freelancers' uncertainty, freelancers who believe they qualify for the gig should be more willing to submit bids.
However, excessive relevant information may be ineffective, even if it reduces freelancers' uncertainty. That is, if buyers provide excessive details about the task, the gig may appear too restrictive or prescriptive ([18]), which might not appeal to freelancers. For example, leaving detailed information out of contracts ([21]) or negotiations ([69]) represents a tactic for improving exchange performance. In a downsizing context, a greater degree of information provision can increase uncertainty and negative reactions ([31]). For freelancers, excessive information can feel overwhelming and can limit their motivation, opportunity, or ability to process the information and submit bids ([40]). A buyer that self-discloses a high amount of personal information might also appear less attractive as a prospective business contact ([12]). Because extensive self-disclosures are unusual in initial B2B online exchanges ([46]), such disclosures might be perceived as inappropriate ([59]).
In summary, we argue that moderate degrees of task and personal information in calls for bids relate to more freelancer bids. Buyers who provide greater degrees of task and personal information should attract more bids, but beyond a moderate degree (i.e., a very dense provision of relevant information), providing still greater degrees of task and personal information may decrease the number of bids. We thus propose a curvilinear relationship:
- P1: Extremely sparse and extremely dense degrees of (a) task and (b) personal information in calls for bids yield fewer freelance bids than moderate degrees.
In calls for bids, buyers can vary the concreteness and affective intensity with which they describe the gig. Researchers disagree about whether more or less ambiguous communication leads to more efficacious speech ([ 8]; [18]; [37]), but in an online freelance marketplace, we posit that buyers must reduce ambiguity to at least some extent by being more concrete and intense. Greater concreteness and affective intensity can be more efficient because recipients can process the information with less time and effort ([37]; [61]). These approaches also tend to result in communication that is more persuasive, memorable, and accessible than communication that uses predominantly abstract or unemotional wording ([28]; [37]). In other settings, greater concreteness increases consumer satisfaction with employee interactions and purchase likelihood ([61]). Greater degrees of intensity achieved through proportionally more affective words provide accessible, diagnostic signals to customers ([52]). They can also sway business partners' decisions when used as inspirational appeals ([72]). Finally, greater concreteness and affective intensity provide heuristic cues that allow freelancers to take mental shortcuts, which makes them more likely to bid ([37]).
However, if the calls for bids appear too concrete or too intense, the task might appear narrow, which reduces the appeal of performing the gig ([37]). [80] finds that greater vagueness (i.e., less concreteness) can enhance judgments of a speaker's character, message acceptance, and recall. Moreover, some research asserts that reducing uncertainty with more concrete formulations is ineffective ([ 9]; [18]), so managers instead should embrace strategic ambiguity to allow for interpretative freedom ([43]). In contracts, unexpected specificity even increases ex ante costs ([58]). Contrastingly, greater task ambiguity can lower costs as well as reduce the risk of litigation and enhance cooperation in B2B exchanges ([81]). Greater degrees of concrete terms in communications with investors also can have adverse effects ([64]), and excessive degrees of positive affective words diminish the impact of customer reviews ([52]). Thus, we predict a stylistic trade-off: Overly ambiguous calls for bids, lacking any concreteness or affective intensity, may undercut buyers' success in attracting freelancers, but some degree of ambiguity (i.e., avoiding overly concrete, affectively intense communication) can allow for divergent interpretations to coexist. Thus, moderate degrees of concreteness and affective intensity may be most effective in encouraging freelancers to bid.
- P2: Extremely sparse and extremely dense degrees of (a) concreteness and (b) affective intensity in calls for bids yield fewer freelance bids than moderate degrees.
Buyers also face uncertainty when deciding whom to hire and how much to pay ([ 2]; [13]). By managing these uncertainties through their bids, freelancers can affect their chances of winning bids and their price premiums. To establish relevant predictions, we integrate [27] communication principles with uncertainty research such that we anticipate a greater provision of relevant information communicated with greater concreteness and affective intensity allows buyers to draw inferences from freelancers' bids with more certainty. Beyond these communication principles, [ 6] suggest that perceived similarity to a message sender reduces receivers' uncertainty. Thus, both purchase likelihood and buyers' willingness to pay a price premium might be influenced by freelancers' adherence to certain communication principles, as well as by their communicative similarity to the buyer.
In other exchange contexts, research has established that when service employees relay greater degrees of service or personal information ([51] 2015; [62]), it improves customers' intentions to purchase. Willingness to purchase also increases if employees use greater concreteness in online service chats ([61]) or greater degrees of affective words in their emails ([72]).
However, the dense provision of relevant information in a bid risks information overload ([40]), and a freelancer being overly concrete or intense might signal a restrictive, narrow approach to the gig ([37]). Our reasoning here parallels that for buyers' formulations of calls for bids. We thus similarly predict that moderate degrees of task and personal information provided in a moderately unambiguous manner (i.e., moderate degrees of concreteness and affective intensity) enhance freelancers' chances of winning the gig.
Yet preferences for uncertainty also might be situational and dispositional ([ 9]), as reflected in buyers' own communicative choices ([30]). Specifically, calls for bids can reveal buyers' expectations and preferences for communication behaviors. For example, buyers might like to get to know freelancers, or they may prefer to keep their business relationships impersonal. The extent to which they disclose their own personal information in calls for bids should signal these preferences. An ambiguous bid offered in response to an ambiguous call for bids might lead the buyer to conclude that the freelancer is tactful, sensitive, and noncoercive ([10]). Adaptive communications also raise perceptions of credibility, common social identity, approval, and trust ([52]; [75]), as well as similarity perceptions, all of which in turn reduce uncertainty ([ 6]). Crafting responses that mimic the buyer's communication is a common personal selling recommendation ([78]). As [72] show, when sellers mimic buyers' communicative manner, it increases buyers' attention. Accordingly, freelancers who mimic a buyer's communication content and manner might improve their exchange success.
In some situations, though, deviating from buyers' communications may be more beneficial ([ 1]). Even in studies that note the performance benefits of adaption, researchers highlight the importance of the degree of adaptivity (e.g., the degree to which salesperson behaviors adjust for each customer during interactions; [78]). Similarly, studies of communication accommodation investigate the degree of accommodation used ([75]). Extending these insights, the outcomes of adaptation likely depend on communication levels (e.g., very informative vs. not informative). In keeping with uncertainty reduction theory, we expect that buyers are less likely to hire freelancers whose bids offer sparse information and are very ambiguous, even if the call for bids has these characteristics.
- P3: When the degrees of (a) task and (b) personal information, (c) concreteness, and (d) affective intensity provided by the buyer are at least moderate (sparse), freelancers can increase (decrease) their chances of bid success by mimicking buyers' communications.
Buyers' uncertainty about a freelancer should influence their willingness to pay a price premium ([ 2]). Although there are many reasons for price variations ([26]) in online freelance marketplaces, buyers compensate (penalize) freelancers for reducing (increasing) their transaction uncertainty by deciding to accept a price above (below) their original payment offer ([ 2]). In line with [70], freelancers' greater provision of relevant task and personal information in a more concrete and intense manner in bids likely reduces buyers' information asymmetry and exchange-specific risks. Therefore, buyers who want to transact with high certainty may render a price premium for such bids ([51] 2015).
The degree to which freelancers mimic buyers' communication also may influence the price premium. For example, [60] find that adaptive approaches for different customers help salespeople increase those customers' willingness to pay a price premium. However, in line with our arguments regarding bid success, we expect that the positive influence of a freelancer's communicative mimicry depends on the specific degree to which the buyer uses a specific communicative element. This reasoning aligns conceptually with the communication principles ([27]), the recommendation that uncertainty should be carefully managed ([ 8]), and the benefits of mimicry identified in studies of communication accommodation ([75]) and adaptive selling ([78]). However, we know of no studies that consider price premium implications of communicative trade-offs between reducing buyers' uncertainty and adapting to buyers' communication. In addition, we are not aware of any research that considers the possible negative effects when sellers mimic buyers who provide less task and personal information, are less concrete, or sparsely use affective intensity.
Buyers who want to transact with high certainty might render a price premium to freelancers who reduce uncertainty by providing greater degrees of relevant information in a more concrete and intense manner. But if buyers perceive that the provision of relevant information, degree of concreteness, and level of intensity surpasses their own reasonable level, they might feel overloaded or restricted and thus unwilling to pay a premium. We therefore predict that buyers offer a price premium to freelancers who provide degrees of relevant information, concreteness, and affective intensity at a level similar (but never too sparse) to their own communication, as only these bids help reduce buyers' exchange risks.
- P4: When the degrees of (a) task and (b) personal information, (c) concreteness, and (d) affective intensity provided by the buyer are at least moderate (sparse), freelancers can increase (decrease) their chances of earning a price premium by mimicking the communication of the buyer.
Graph: Figure 1. Effect of buyers' communication on call for bids success.
We conducted a large-scale field study with a proprietary data set of calls for bids and corresponding bids posted on a leading, global online freelance marketplace. The marketplace hosts seven freelance service submarkets: ( 1) design; ( 2) writing and translation; ( 3) video, photo, and audio; ( 4) business support; ( 5) social media, sales, and marketing; ( 6) software and mobile development; and ( 7) web development. The bidding process follows a sequential, sealed-bid reverse auction format, and it concludes when the buyer chooses one winning bid ([34]; [39]). As with recent marketing research that investigates large scales of communication (see Web Appendix B for an illustrative overview), this process depends on and is captured in text data. We used ( 1) text data from 343,796 calls for bids issued by 49,081 buyers (restricted to those who posted at least two gigs) to predict buyers' call for bids success, ( 2) 2,327,216 bids submitted by 34,851 freelancers (restricted to those who submitted at least two bids) to predict freelancers' bid success, and ( 3) 148,158 bids submitted by 30,851 freelancers (restricted to those who won and for which the payment was disclosed) to predict freelancers' price premium. Our multilevel approach required more than one observation (call for bid or bid) in each Level 2 unit (buyer or freelancer); otherwise, Level 2 and Level 1 variance might have been confounded ([74]). Web Appendix C summarizes the definitions and operationalizations, and Web Appendix J provides the descriptive statistics and correlations.
The number of freelancers who submit bids to offer their services provided the measure of success of buyers' call for bids. More submitted bids increases the probability that buyers can find an appropriate freelancer, whereas failing to find a suitable match is time consuming and costly because it requires further searches and delays the project ([35], [36]). We measured freelancers' bid success as a binary indicator of whether ( 1) or not (0) the freelancer was chosen by the buyer and won the bid ([32]). For freelancers' price premium, we gauged the percentage by which the accepted bid price for the project exceeded (or fell short of) the buyer's original payment offered (i.e., benchmark price; [20]). This operationalization accounted for the difference between the final price a buyer paid and the original price they offered (i.e., what the buyer expected to pay) ([73]).
To capture the independent communication variables, we mined the text of each call for bids and each bid. For the preprocessing and extraction steps, we used the R package Quanteda ([ 4]), as well as a combination of newly developed and prevalidated text mining dictionaries. For the degree of task information in each text, we inductively sourced a list of context-specific task descriptor words. To start, we acquired all 34,851 freelancers' service skill tags ([ 7]; for an illustration, see Web Appendix D), which freelancers list in their profiles to describe the service tasks they offer (e.g., "developer," "illustrator"). After removing stop words and duplicates, two coders reviewed the remaining word list, deleted any misspelled words, and removed terms that did not describe a service (e.g., "great," "reliable"). Using Quanteda ([ 4]), we stemmed the remaining words, leaving 1,912 unique word stems that describe service tasks. We mined each call for bids and bid, then summed word occurrences reflecting the new task dictionary. By dividing this sum by total words, we obtained a measure of the degree (ratio) of task information in each text. When people self-disclose personal information, they use singular, first-person pronouns. In line with previous research (e.g., [67]), we measured the degree of personal information as the ratio of first-person singular pronoun words (e.g., "I," "me") to the total words in each text. To determine the degree of communication concreteness, we mined each text for [11] list of generally known English lemmas that indicate whether a concept denoted by a term refers to a perceptible entity. Following their operationalization, we included all terms that received a rating of 3 or greater on their bipolar, five-point abstract-to-concrete rating scale.[ 6] That is, terms that score 3 or higher refer to relatively more specific objects, materials, people, processes, or relationships. We again divided the sum of the concrete terms by the total words in each text. Finally, the ratio of emotion-laden words (e.g., "problematic," "easy"; [28]; [37]) determined affective intensity. Using the Linguistic Inquiry and Word Count (LIWC) affect dictionary, we obtained a list of affect words, which we then summed for each text ([66]) and divided each by the corresponding total word count to obtain the degree of affective intensity.
To ensure the validity of our text-mined communication measures, we asked two coders to classify the texts of a random subsample of 100 calls for bids (Mlength = 129 words) and 100 bids (Mlength = 102 words). The coders indicated whether considerable task information, personal information, concreteness, and affective intensity were present in each text ( 1) or not (0). Comparing the coders' classifications with our text-mined classification revealed substantial agreement for both calls for bids (.73 to.94) and bids (.66 to.88) ([47]). The average F1 measure was sufficiently high for both bids (.79 to.95) and calls for bids (.80 to.95), as we detail in Web Appendix F.
To establish the internal validity of the chosen communication aspects on receivers' uncertainty perceptions, we conducted a series of experimental pilot studies. We used single-factor, within-subject designs for ( 1) task information, ( 2) personal information, ( 3) concreteness, and ( 4) affective intensity. For each pilot study, we recruited between 50 and 53 U.S. consumers with a mean age of 37.6 years (50% women) from Amazon Mechanical Turk (for details, see Web Appendix G). In line with previous research (e.g., [28]; [49]; [55]; [61]), we find that greater use of all four communication aspects in bids significantly reduces buyers' uncertainty perceptions and affects their hiring intentions.
In Web Appendix H, we summarize the model-free findings. The mean-level comparison indicates that calls for bids with significantly greater degrees of task information and concreteness, as well as significantly lower degrees of personal information and affective intensity, receive more freelance bids than an average call for bids (M = 5).
The success of calls for bids reflects a count variable. Noting the overdispersion in the data (p < .001), we used a negative binomial model instead of a Poisson model. Furthermore, calls for bids are nested within buyers, and thus, the call for bids and number of freelancers who offer their service might be interdependent. The significant between-group variance (p < .001) and ICC( 1) of.27 suggests a multilevel structure. We therefore specified a multilevel model with a random intercept to control for time-invariant unobserved differences between buyers (e.g., education, country, gender) that could relate to differences in their success, using the following base equation:
Graph
( 1)
where is the success of a call for bids i (i = 1, ..., 343,796) issued by buyer j (j = 1, ..., 49,081), is buyer task information, indicates buyer personal information, is buyer concreteness, and refers to buyer affective intensity in the call for bids. In turn, is buyer task information squared, is buyer personal information squared, is buyer concreteness squared, and is buyer affective intensity squared. Finally, is the random intercept and is the error term.
Some empirical challenges inhibited a robust model identification, which we addressed in several ways. To account for observed heterogeneity, we incorporated covariates that might influence how many freelancers respond to a particular call for bids. First, in line with extant text mining studies ([ 7]), we controlled for the word count in each call for bids. Second, as a reputation cue, we measured buyer experience as the number of projects a buyer had commissioned previously on the platform prior to posting the focal call for bids ([32]). Third, a higher payment offer may attract more freelancers ([36]), so we determined the payment offered by the buyer in U.S. dollars, multiplied by an undisclosed index for anonymity. We used a dummy for nondisclosed payments, but we replaced missing values with a grand mean to retain the observations. Fourth, we measured project duration, as longer projects attract more freelancers ([36]). A dummy variable indicated whether the project was slated to last more ( 1) or less than a month (0). Fifth, more buyers demanding freelance services at the same time creates a relative shortage of freelancers ([36]). To account for an excess supply of freelancers, we calculated the sum of all active freelancers in the specific submarket of the call for bids and divided by the sum of all calls for bids posted around the same time (±31 days) in the same submarket. Sixth, the marketplace grew over time, so we included fixed effects for the year of the call for bids. Seventh, we included fixed effects for the seven submarkets, since submarkets that feature more complex projects have fewer qualified freelancers.
Beyond these observed covariates, buyers' bid formulations might have varied by project characteristics unobservable to us. To the extent that these unobserved project characteristics influenced both the buyers' communication strategies and buyer outcomes, the estimated parameters might be biased. Therefore, we concatenated all service skill tags from the service profile of each freelancer who submitted a bid in response to a specific call. Then, to uncover the latent mixture of project types, we applied a latent Dirichlet allocation model to the project-specific skill tags (e.g., [ 7]; see Web Appendix I). We included the resulting 12 latent project characteristics as fixed effects to account for unobserved heterogeneity.
Buyers also strategically make their communication decisions in learned anticipation of a larger number of bids or other factors, which were potentially unobservable to us. This strategic behavior could make communication approaches endogenous ([42]). Because our data did not contain valid, strong instruments for buyers' communications, we adopted [65] approach and used Gaussian copulas to model the correlation between each buyer communication and the error term. We added regressors to Equation 1, such that
Graph
( 2)
where is the inverse of the normal cumulative distribution function and represents the empirical distribution functions of the four buyer communication approaches. The endogenous regressors must be nonnormally distributed for identification ([65]), and we confirmed this was true using Shapiro–Wilks tests (all p < .001). The updated equation to predict buyers' call for bids success, after correcting for endogeneity, was thus
Graph
( 3)
where is the vector of control variables, are year effects, are submarket effects, are latent project clusters, and are Gaussian copulas. We used a robust estimator to account for correlated and clustered standard errors.
The maximum variance inflation factor is 2.11, indicating no potential threat of multicollinearity. Table 2 contains the results of a main effects model and the full model, and Figure 1, Panels A–D, display the curvilinear effects from the full model. We have proposed that extremely sparse and extremely dense degrees of relevant information, concreteness, and affective intensity in calls for bids yield fewer freelance bids than moderate degrees of these communication elements. In line with our expectations, we find a positive linear effect (.152, p < .01) and negative squared effect for task information (−.026, p < .01), as displayed in Figure 1, Panel A. Moderate levels of the use of task information (50%:.222, p < .01) yield better results than sparse (10%: −.426, p < .01) and dense (90%: −.495, p < .01) uses. Furthermore, we find a positive linear effect (.052, p < .01) and negative squared effect for concreteness (−.080, p < .01) (Figure 1, Panel C). Moderate use (50%:.078, p < .01) yields better results than sparse use (10%: −.092, p < .01) or dense use (90%: −.251, p < .01) of concreteness. Contrary to our expectations, we find a negative linear effect (−.190, p < .01) and a positive squared effect (.032, p < .01) of personal information (Figure 1, Panel B). We also find a negative linear effect (−.084, p < .01) and a nonsignificant squared effect (.001, ns) of affective intensity (Figure 1, Panel D). Thus, it appears that any provision of personal information or greater use of affective intensity by the buyer is always ineffective. As a possible explanation, we note that in B2B online conversations, self-disclosure and emotions may be valued only after business relations have been established, not at the moment they form ([46]). Most of the exchanges in our data were between strangers, rather than being repeat exchanges, so it may be more appropriate for buyers to avoid personal details and appear rational rather than emotive.
Graph: Figure 2. Response surfaces for bid success and price premium.
Graph
Table 2. Predicting the Success of Buyers' Calls for Bids.
| Model 1:Main Effects | Model 2:Full Model |
|---|
| β | SE | 95% CI | β | SE | 95% CI |
|---|
| Buyer Communication | | | | | | |
| Task information | .123** | .003 | .117,.128 | .152** | .003 | .146,.159 |
| Personal information | −.149** | .004 | −.157, −.141 | −.190** | .004 | −.199, −.181 |
| Concreteness | .040** | .003 | .035,.045 | .052** | .003 | .046,.057 |
| Affective intensity | −.098** | .007 | −.112, −.085 | −.084** | .008 | −.100, −.068 |
| Buyer Communication Squared | | | | | | |
| Task information squared | | | | −.026** | .001 | −.028, −.024 |
| Personal information squared | | | | .032** | .002 | .029,.035 |
| Concreteness squared | | | | −.008** | .001 | −.011, −.007 |
| Affective intensity squared | | | | .001 | .001 | −.001,.004 |
| Controls | | | | | | |
| Word count | −.025** | .003 | −.031, −.019 | −.027** | .003 | −.033, −.021 |
| Buyer experience | −.033** | .007 | −.046, −.020 | −.033** | .007 | −.046, −.020 |
| Project payment | .170** | .005 | .159,.181 | .168** | .005 | .158,.179 |
| Payment not disclosed | .073** | .005 | .063,.083 | .071** | .005 | .061,.081 |
| Project duration | .116** | .003 | .110,.122 | .117** | .003 | .111,.123 |
| Excess supply of freelancers | .614** | .004 | .606,.622 | .606** | .004 | .598,.613 |
| Fixed Effects | | | | | | |
| Years | included | included |
| Submarkets | included | included |
| Unobserved Heterogeneity | | | | | | |
| Project characteristics | included | included |
| Endogeneity Corrections | | | | | | |
| Gaussian copulas | included | included |
| Buyers | 49,081 |
| Call for bids | 343,796 |
1 Notes: Standardized results. Significance is based on two-tailed tests. The dependent variable is the count of all bids received. The sample included all projects listed by buyers with at least two projects to which at least one freelancer submitted a bid. Effects for years, submarkets, project characteristics, and Gaussian copulas are detailed in Web Appendix Q.
To entice more freelancers to bid, buyers should keep their calls for bids brief (−.027, p < .01 for word count), which emphasizes the need for careful formulations. Higher payment offers (.168, p < .01), longer project durations (.117, p < .01), and an excess supply of freelancers (.606, p < .01) all increase the number of bids. Notably, the number of projects a buyer has previously commissioned relates negatively to the number of freelancers who bid (−.033, p < .01). These experienced buyers might have established relationships with specific freelancers, which reduces other freelancers' chances and causes them to refrain from bidding ([48]).
Bids that offer less personal information and greater task information, concreteness, and affective intensity are more successful in winning projects. Among bids that won, the mean-level comparisons indicate nonlinear effects of mimicry. That is, successful freelancers mimic buyers' use of task information, personal information, and concreteness closely. If a buyer uses very sparse or very dense degrees of these communication aspects, the winning freelancers deviate more, indicating a nonlinear impact of mimicry. We do not find evidence of this mimicry relationship for affective intensity (see Web Appendix H).
Previous studies often operationalize communication similarity as the absolute difference between two measures (e.g., [52]; [75]), but this approach suffers some implicit constraints ([17]). In particular, difference scores suggest that one party's communication increases at the same magnitude as the other's decreases. They also ignore the degree at which the relative mimicry occurs. As a preferable alternative, we use polynomial regression, which allows for simultaneous testing of similarity and dissimilarity effects on bid success, at different levels of freelancers' and buyers' uses of the four communication aspects. In their study of positive and negative emotional tone convergence, [24] also use polynomial regression to explore the nuanced effects of convergence in leader–follower relationships on leader–member exchange quality. A simple regression model that captures absolute deviation cannot simultaneously assess the degree of task information by the buyer and the potential nonlinear effects of task information mimicry by the freelancer. So, we performed a polynomial regression with response surface analyses for each communication aspect to capture the extent to which freelancers mimicked a prospective buyer's provision of relevant information and communication manner. We detail this polynomial modeling approach that led to Equation 4 and the calculation of all polynomial terms, using task information as an example, in Web Appendix E.
We tested freelancers' trade-off between adding more uncertainty-reducing communication versus mimicking the buyer's communication in a polynomial regression model that included linear terms, quadratic terms, and interactions. In the multilevel base equation to predict freelancers' bid success (ICC( 1) = .09, p < .001),
Graph
( 4)
is the success of bid k (k = 1, ..., 2,327,216) by freelancer l (l = 1, ..., 34,851), are the four freelancer communication aspects, indicate the four buyer communication aspects, are freelancer communication aspects squared, are interactions of freelancer and buyer communication aspects, are buyer communication squared, is the random intercept, and is the error term.
We incorporated several covariates that might influence freelancers' bid success. As in the buyer model, we controlled for word count, project payment, project duration, and excess supply of active freelancers. We also included fixed effects for years, submarkets, and latent project characteristics. We accounted for the number of projects the freelancer completed prior to submitting the focal bid as a reputation cue that might determine bid success ([32]). Freelancer rating is an average five-point satisfaction rating that a freelancer has received for all completed projects. To retain observations of unrated freelancers, we included a dummy for observations without star ratings and replaced the missing values with a grand mean rating.
Several additional controls relate to whether a bid is successful. First, following prior research, we assessed linguistic style matching, or the similarity between each bid and the respective call for bids, across nine function word categories ([52]). Second, we accounted for any previous relationship in which the freelancer had completed at least one project for the same buyer prior to the specific call for bids ([32]). Third, freelancers submit a bid price that may differ from the payment offered by the buyer, and a higher bid price may reduce the likelihood of bid success ([32]). In light of this, we measured each bid price as a ratio between the asking price and the average indexed bid price requested by all competing freelancers for the same call for bids. Fourth, the longer it takes freelancers to submit a bid, the lower their chances of success ([34]). So, we measured time-to-bid as the number of days between the posting of the call for bids and the bid submission. A dummy variable also indicates whether the bid was submitted late ( 1) or on time (0). Fifth, competition for a specific call for bid should impact each bid's success chances, so we controlled for the number of bids for the same call ([34]).
Similar to buyers, freelancers make communication decisions strategically in anticipation of higher bid success or other, unobservable factors. Thus, freelancer communication is potentially endogenous, so we again used Gaussian copulas (Shapiro–Wilk tests: all p < .001). The updated equation to predict freelancers' bid success is as follows:
Graph
( 5)
where is the vector of control variables, are year effects, are submarket effects, are latent project clusters, are Gaussian copulas for bid text, and are Gaussian copulas for calls for bids text.
The maximum variance inflation factor is 3.86, indicating no threat of multicollinearity. Table 3 contains the results of the freelancer bid success models, Web Appendix K summarizes the response surface coefficients, and Figure 2 displays these coefficients on three-dimensional surfaces, reflecting relationships among freelancer communication, buyer communication, and bid success. We also highlight the misfit line used to explore the trade-off between exceeding and falling short of buyers' communication levels.
Graph
Table 3. Predicting Freelancers' Bid Success.
| Model 3:Freelancer Communication | Model 4:Full Model |
|---|
| β | SE | 95% CI | β | SE | 95% CI |
|---|
| Freelancer Communication | | | | | | |
| y01: Task information | .014** | .001 | .013,.015 | .015** | .001 | .014,.016 |
| y02: Personal information | .018** | .001 | .016,.019 | .017** | .001 | .016,.017 |
| y03: Concreteness | .030** | .001 | .029,.031 | .031** | .001 | .030,.032 |
| y04: Affective intensity | .001 | .001 | −.001,.003 | .000 | .001 | −.002,.001 |
| Buyer Communication | | | | | | |
| y05: Task information | | | | −.009** | .000 | −.009, −.008 |
| y06: Personal information | | | | −.017** | .000 | −.018, −.017 |
| y07: Concreteness | | | | −.008** | .000 | −.009, −.008 |
| y08: Affective intensity | | | | .001** | .000 | .001,.002 |
| Freelancer Communication Squared | | | | | | |
| y09: Task information squared | −.006** | .000 | −.006, −.005 | −.006** | .000 | −.007, −.006 |
| y10: Personal information squared | −.005** | .000 | −.005, −.004 | −.005** | .000 | −.005, −.004 |
| y11: Concreteness squared | −.006** | .000 | −.007, −.006 | −.007** | .000 | −.007, −.007 |
| y12: Affective intensity squared | .000** | .000 | .000,.001 | .000** | .000 | .000,.001 |
| Freelancer–Buyer Interactions | | | | | | |
| y13: Task information interaction | | | | .015** | .000 | .015,.016 |
| y14: Personal information interaction | | | | −.002** | .000 | −.002, −.001 |
| y15: Concreteness interaction | | | | .005** | .000 | .004,.005 |
| y16: Affective intensity interaction | | | | .020** | .000 | .020,.021 |
| Buyer Communication Squared | | | | | | |
| y17: Task information squared | | | | .001** | .000 | .001,.002 |
| y18: Personal information squared | | | | −.004** | .000 | −.005, −.004 |
| y19: Concreteness squared | | | | .001** | .000 | .001,.001 |
| y20: Affective intensity squared | | | | .000* | .000 | .000,.000 |
| Controls | | | | | | |
| Word count | −.022** | .001 | −.024, −.021 | −.021** | .001 | −.022, −.020 |
| Linguistic style matching | .051** | .001 | .048,.053 | .051** | .001 | .048,.053 |
| Freelancer experience | .002** | .001 | .001,.003 | .002** | .001 | .001,.003 |
| Freelancer rating | .010** | .001 | .009,.010 | .010** | .001 | .009,.010 |
| Project payment | −.001** | .000 | −.001, −.001 | −.001** | .000 | −.001, −.001 |
| Payment not disclosed | −.028** | .000 | −.029, −.028 | −.029** | .000 | −.030, −.028 |
| Previous relationship | .078** | .001 | .076,.081 | .078** | .001 | .075,.080 |
| Bid price | −.006** | .000 | −.006, −.006 | −.006** | .000 | −.007, −.006 |
| Time-to-bid | .001 | .000 | .000,.001 | .001 | .000 | .000,.001 |
| Late submission | −.005** | .000 | −.005, −.004 | −.004** | .000 | −.005, −.004 |
| Competition | −.251** | .007 | −.265, −.238 | −.251** | .007 | −.264, −.238 |
| Excess supply of freelancers | −.044** | .000 | −.045, −.043 | −.042** | .000 | −.043, −.042 |
| Fixed Effects | | | | | | |
| Years | included | included |
| Submarkets | included | included |
| Unobserved Heterogeneity | | | | | | |
| Project characteristics | included | included |
| Endogeneity Corrections | | | | | | |
| Gaussian copulas | included | included |
| Freelancers | 34,851 |
| Bids | 2,327,216 |
- 4 Notes: Standardized results. Significance is based on two-tailed tests. The dependent variable is whether the freelancer was chosen and won the bidding process. The sample included all bids by freelancers with at least one winning and at least one losing bid. Effects for years, submarkets, project characteristics, and Gaussian copulas are detailed in Web Appendix Q.
- 5 *p < .05.
- 6 **p < .01.
We have proposed that when the degree of relevant information, concreteness, and affective intensity provided by the buyer is at least moderately dense (sparse), freelancers can increase (decrease) their chances of bid success by mimicking the buyer's communication. The surface-level tests along the plotted misfit line (Web Appendix K) display negative curvatures for task information (−.020, p < .01), personal information (−.007, p < .01), concreteness (−.011, p < .01), and affective intensity (−.020, p < .01). These results indicate that mimicking the buyer's communication increases bid success (see Web Appendix L for further clarification).
In line with our proposition, we qualify this effect for sparse degrees of task and personal information, concreteness, and affective intensity provided by the buyer in Web Appendix M. If we were to find positive slope coefficients at lower levels, it would suggest that freelancers can increase their chances of bid success by exceeding, rather than mimicking, the buyer's communication. This prediction holds for personal information (.020, p < .01) and concreteness (.024, p < .01), according to the slopes at low levels of buyer communication. However, contrary to our expectations, we find negative effects for the slopes of task information (−.008, p < .01) and affective intensity (−.030, p < .01) at low levels of buyer communication. Therefore, freelancers should always mimic the degree of task information and affective intensity provided by the buyer. For these two communication aspects, the tenets of communication accommodation theory ([75]) and adaptive selling ([78]) hold: mimicking the buyer is always better. To increase their chances of bid success further, freelancers also must keep their bids concise (−.021, p < .01 for word count). Reputation cues (experience:.002, p < .01; rating:.010, p<.01) increase freelancers' chances of bid success, as do linguistic style matching (.051, p < .01), previous business relations with the buyer (.078, p < .01), lower bid prices (−.006, p < .01), timely (cf. late) bid submissions (−.004, p<.01), lack of competition (−.251, p < .01), and reduced supply of freelancers (−.042, p < .01).
Bids with significantly less affective intensity and significantly more task information, personal information, and concreteness achieve greater price premiums than an average bid (M = 14% discount). Moreover, 96% of freelancers completed projects without any price premium, indicating the prevalence of value traps. The bids that achieved price premiums mimicked those buyers that made moderate use of task information, concreteness, and affective intensity closely, yet they deviated from buyers that made very sparse or very dense use of them. For personal information, we find a distinctive, positive, linear relationship for mimicry. Successful freelancers mimicked buyers that supplied a lot of personal details but deviated if buyers supplied very little or moderate degrees of personal information (Web Appendix H).
The price premium analysis is restricted to bids that win and buyers that disclose their payment offer upfront. Thus, our estimates may be biased by buyers' self-selection, in terms of which bid they chose and whether they disclosed payments. Therefore, we employed a two-stage selection model. In the first stage, we estimated a choice model, with the availability of the necessary data as a binary dependent variable (i.e., bid was won and payment was disclosed). Using this model, we computed the inverse Mills ratio to account for the potential selection bias (probit model in Web Appendix N) and included this correction term in the final model estimation. To identify second-stage parameters, there needed to be one term in the first-stage equation that was unrelated to the error term in the freelance price premium equation. We thus included the dummy that indicates if the bid was submitted late only in the first-stage equation because this term explained buyers' choice of the bid, but we did not expect it to be conceptually related with the eventual price premium. Thus, this term satisfied both relevance and exogeneity requirements. The updated equation of our multilevel model (ICC( 1) = .13, p < .001) is as follows:
Graph
( 6)
where is the price premium of bid k (k = 1, ..., 148,158) offered by freelancer l (l = 1, ..., 30,851), and is the correction term.
The maximum variance inflation factor is 2.74, indicating no threat of multicollinearity. Table 4 contains the results of the freelancer price premium models, Web Appendix K details the response surface coefficients, and Figure 2 displays the surfaces.
Graph
Table 4. Predicting Freelancers' Price Premium.
| Model 5:Freelancer Communication | Model 6:Full Model |
|---|
| β | SE | 95% CI | β | SE | 95% CI |
|---|
| Freelancer Communication | | | | | | |
| y01: Task information | .023** | .002 | .020,.026 | .022** | .002 | .019,.025 |
| y02: Personal information | .021** | .002 | .017,.025 | .021** | .002 | .017,.026 |
| y03: Concreteness | .006** | .001 | .004,.007 | .005** | .001 | .003,.007 |
| y04: Affective intensity | .004 | .003 | −.002,.010 | .003 | .003 | −.003,.009 |
| Buyer Communication | | | | | | |
| y05: Task information | | | | .003* | .001 | .001,.005 |
| y06: Personal information | | | | −.016** | .002 | −.019, −.012 |
| y07: Concreteness | | | | −.004** | .001 | −.006, −.002 |
| y08: Affective intensity | | | | −.001 | .002 | −.005,.003 |
| Freelancer Communication Squared | | | | | | |
| y09: Task information squared | −.001 | .001 | −.002,.000 | −.001 | .001 | −.002,.001 |
| y10: Personal information squared | −.004** | .001 | −.006, −.002 | −.004** | .001 | −.006, −.002 |
| y11: Concreteness squared | −.003** | .001 | −.004, −.002 | −.003** | .001 | −.005, −.002 |
| y12: Affective intensity squared | .000 | .000 | −.001,.000 | .000 | .000 | −.001,.000 |
| Freelancer–Buyer Interactions | | | | | | |
| y13: Task information interaction | | | | .025** | .003 | .024,.027 |
| y14: Personal information interaction | | | | −.004** | .001 | −.005, −.002 |
| y15: Concreteness interaction | | | | .002** | .000 | .001,.002 |
| y16: Affective intensity interaction | | | | .010** | .003 | .008,.012 |
| Buyer Communication Squared | | | | | | , |
| y17: Task information squared | | | | .003** | .001 | .001,.004 |
| y18: Personal information squared | | | | .003** | .001 | .002,.005 |
| y19: Concreteness squared | | | | −.002** | .001 | −.004, −.001 |
| y20: Affective intensity squared | | | | .002** | .001 | .000,.003 |
| Controls | | | | | | |
| Word count | −.014** | .002 | −.018, −.011 | −.014** | .002 | −.018, −.011 |
| Linguistic style matching | .022** | .004 | .013,.030 | .023** | .004 | .015,.032 |
| Freelancer experience | −.001 | .001 | −.004,.001 | −.001 | .001 | −.004,.001 |
| Freelancer rating | −.001 | .001 | −.003,.001 | −.001 | .001 | −.003,.001 |
| Project payment | −.029** | .010 | −.049, −.009 | −.029** | .010 | −.049, −.009 |
| Previous relationship | .057** | .001 | .054,.059 | .056** | .001 | .054,.059 |
| Time-to-bid | .009** | .001 | .007,.012 | .009** | .001 | .006,.011 |
| Competition | −.015** | .002 | −.019, −.011 | −.015** | .002 | −.019, −.011 |
| Excess supply of freelancers | −.001 | .001 | −.002,.001 | −.001 | .001 | −.002,.001 |
| Fixed Effects |
| Years | included | included |
| Submarkets | included | included |
| Unobserved Heterogeneity | | | | | | |
| Project characteristics | included | included |
| Endogeneity Corrections |
| Gaussian copulas | included | included |
| Sample-Selection Correction | | |
| Inverse Mills ratio | −.016** | .002 | −.020, −.012 | −.016** | .002 | −.020, −.012 |
| Freelancers | 30,851 |
| Bids | 148,158 |
- 7 Notes: Standardized results. Significance is based on two-tailed tests. The dependent variable is price premium for the chosen bid. The sample includes all winning bids for which the payment was disclosed. Effects for years, submarkets, project characteristics, and Gaussian copulas are detailed in Web Appendix Q.
- 8 *p < .05.
- 9 **p < .01.
We proposed that when the degree of relevant information and communication manner provided by the buyer is at least moderately high (low), freelancers increase (decrease) their chances of earning a price premium by mimicking this communication. Web Appendix O displays the misfit lines on two-dimensional planes. In line with our expectations, the surface-level tests along the plotted misfit line show a negative curvature for task information (−.023, p < .01), concreteness (−.007, p < .01), and affective intensity (−.008, p < .01), such that mimicking the buyer's communication increases bid success. However, for personal information, we find a positive curvature (.003, p < .05), which implies freelancers should always offer more personal information than the buyer. For these B2B services, the provider and the service are inseparable, which may lead buyers to place more value on personal information about freelancers, even if their own provision of personal details in the calls for bids is sparse.
If a buyer provides little relevant information or is less concrete (Web Appendix P), a positive slope would suggest that freelancers can increase their chances of earning a price premium by exceeding rather than mimicking the buyer. We find support for this prediction in the slope of personal information (.027, p < .01) at low levels of buyer personal information. However, negative effects emerge from the slopes of task information (−.016, p < .01) and affective intensity (−.012, p < .01), and we find a nonsignificant effect for concreteness (.002, n.s.). Mimicking the buyer's task information and affective intensity is always better, which is in line with accommodation theory and adaptive selling ([75]; [78]).
Freelancers also increase their price premiums by avoiding lengthy bids (−.014, p < .01). Although platform reputation cues (experience and rating) can boost freelancers' chances of bid success, they do not determine the final price buyers pay. The skew in the ratings toward very high scores may limit their ability to help prospective buyers determine an appropriate price ([45]. Linguistic style matching (.023, p < .01), a previous relationship with the prospective buyer (.056, p < .01), submitting early in the bid process (.009, p < .01), and reduced competition (−.015, p < .01) all increase buyers' acceptance of a price premium.
Across disciplines, substantial research has identified various success determinants in online freelance marketplaces (e.g., [36]; [77]). For example, studies of B2B exchanges and two-sided marketplaces emphasize communication (see Web Appendix A). But at the specific word level, we lack insights into the optimal information or manner of communication ([ 7]). With this initial investigation of how buyers' and freelancers' success might be enhanced by appropriately managing the other party's uncertainty, we postulate, in line with uncertainty reduction ([ 6]) and uncertainty management ([ 9]) theories, that communication that is not completely informative and clear may still be effective. Accordingly, we investigate how buyers' communication can attract freelance bids and how freelancers' communication can determine their bid success and price most effectively, and the results offer both theoretical and practical implications.
First, we advance research on how buyers' communication determines their ability to attract freelancers. Drawing on prior communication research, we identify communication principles that critically relate to receivers' uncertainty, such as relevant task and personal information and the relative concreteness and affective intensity with which this information is communicated ([ 8]; [27]). To entice more freelancers to bid, buyers should carefully formulate their calls for bids to keep them brief. Freelancers' information processing motivation, time, skills, and proficiency likely are limited, so buyers must choose their wording carefully and select from various effective communicative aspects. They can attract a larger pool of bids if they provide moderate degrees of task information in a moderately concrete manner. Offering too little of these features leaves freelancers with too much uncertainty, and dense information provision or being very concrete is too restrictive. If buyers provide greater degrees of personal information or express greater affective intensity in their calls for bids, it reduces the number of service offers they receive. This finding contrasts with uncertainty reduction theory ([ 6]) and B2B research that suggests self-disclosure strengthens buyer–seller cooperativeness ([41]). However, instead of ongoing B2B relationships, our study refers mostly to initial interactions between strangers (in 98% of cases, the freelancer had never worked for the prospective buyer). Evidence obtained from buyer–seller online chats similarly suggests that self-disclosure and emotive expressions are valued only in existing B2B relationships, not in new ones ([46]). Overall, we offer empirical support for communication theorists' suggestions that common communication principles can be purposefully flouted to achieve better conversation outcomes ([23]).
Second, freelancers must keep their bids concise. They too face a trade-off between reducing the buyer's uncertainty and offering overly dense information. In line with research on communication accommodation ([75]) and adaptive selling ([78]), we show that freelancers can improve their bid success by mimicking the prospective buyer's communication. Adding to these research streams, we introduce a contingency perspective that reveals the efficacy of mimicry depends on the degree to which buyers use specific communication elements. In line with accommodation theory and adaptive selling, bid success always improves when freelancers mimic buyers' provision of task information and use of affective intensity. However, in line with uncertainty reduction theory ([ 6]) and expectancy violations ([ 1]) when buyers supply little personal information and are less concrete, freelancers can increase their chances of bid success by diverging and providing more personal information and concreteness.
Third, freelancers often struggle to avoid value traps in which they sell more of their services for less ([76]). Rational buyer expectations should allow high-quality freelancers to charge price premiums ([70]), but the quality of freelance services is unobservable prior to purchase, and rational buyers might refuse to pay any price premium if they feel uncertain and suspect the freelancer may be hiding information ([16]). Therefore, to achieve premiums, freelancers should offer short, appropriately formulated bids. Buyers are more willing to pay a premium to freelancers who mimic their provision of task information, concreteness, and affective intensity, which is in line with communication accommodation theory ([75]) and adaptive selling research ([78]). However, similar to the findings for bid success, freelancers should offer more personal information than buyers, rather than mimicking buyers' provision of such information. In most service settings, a "bad" seller might provide a great product by chance; however, almost by definition, a bad freelancer produces bad service ([36]). This tight coupling between the freelancer and service quality represents a conceptual distinction in our study, which accordingly shows that buyers' willingness to pay a premium increases with more personal information issued by the freelancer.
Our findings offer actionable insights for the millions of buyers and freelancers utilizing online freelance marketplaces, the collective value of which is predicted to reach $2.7 trillion by 2025 ([56]). In detail, being informative and unambiguous may be a common assumption, but it is not an imperative, nor does it always lead to success.
Although 59% of U.S. companies use a flexible workforce to some degree, more than one-third of contracted projects are never completed ([33]). To attract freelancers, buyers should keep their calls for bids succinct. Beyond that recommendation, we offer several tips for formulating calls for bids in Table 5. In particular, a task description with a moderate amount of information helps freelancers anticipate the task without overloading them with details. Due to the relative anonymity of online freelance marketplaces, buyers might assume that freelancers will need to know who they are, but instead, we find that the less buyers describe themselves (to focus on describing the task), the better the outcomes. Relatable and imaginable (rather than abstract) descriptions of the project help freelancers grasp the requirements. However, being excessively concrete becomes prescriptive, which deters freelancers. Using emotion words makes the content of a call for bids relatively more intense. Such intensity can remove ambiguity and make opinions quickly accessible, but we find that calls for bids are more effective if they are formulated relatively impassively. Enthusiastic project descriptions seemingly might raise freelancers' suspicion that the project is too good to be true. Also, offering higher payment might attract a larger pool of freelance bids, as do long- rather than short-term gigs. Finally, more freelancers bid when there are fewer calls for bids in the subsector.
Graph
Table 5. Buying and Selling Services in Online Freelance Marketplaces.
| How to Formulate Calls for Bids to Attract Freelancers |
|---|
| Bad Practice Excerpt | Good Practice Excerpt | Lift in Bids |
|---|
| Specify tasks and skills | "I need a website to showcase the full range of my fitness workouts." | "I need a website designer who can design a WordPress website using a WordPress premium theme." | An increase in task terms from 18% to 29%, resulting in 5% more bids. |
| Avoid personal information | "I have been creating my own classes for almost 10 years now...clients tend to especially love my classes on strength and flexibility. Now I need help setting up my website." | "I am a Fitness Trainer and need help with building my website to showcase my mixed services and home workouts." | A decrease in personal terms from 9% to 4%, resulting in 4% more bids. |
| Be moderately concrete | "I require a professional who is savvy in configuring a stylish website employing a premium theme." | "You should have got very good creative skills but know how to design for web and also know how to include calls to actions within a good design." | An increase in concrete terms from 21% to 26%, resulting in 1% more bids. |
| Avoid being affectively intense | "I have created a fantastic theme but you should be confident and eager about WordPress and help optimize." | "The theme and examples will be provided, but you should also know about WordPress and optimize." | A decrease in affective terms from 11% to 4%, resulting in 4% more bids. |
| How to Formulate Successful Bids and Achieve Price Premiums |
| | Bad Practice Excerpt | Good Practice Excerpt | Lift in Bid Success |
| Mimic task description | "Dear Sir, would love to work for you..." | "Hi Gary, I am happy to help you with your fitness website development and design..." | An increase in task terms from 16% to 25%, resulting in 7% higher bid success and 8% higher price premium. |
| Exceed buyers who supply little personal information | "I am an enthusiastic designer and expert in Web development..." | "I am a WordPress Freelancer with 15 years of work experience..." | An increase in personal terms from 6% to 8%, resulting in 3% higher bid success and 4% higher price premium. |
| Exceed buyers who are not concrete | "I have great skills and plenty of fantastic experience in creating relevant websites..." | "I have worked on several similar projects, designing websites, also using WordPress, including premium themes and I can deliver to a tight schedule..." | An increase in concrete terms from 24% to 30%, resulting in 7% higher bid success (but no effect on price premium). |
| Mimic the buyer's affective intensity | "The content will be creative and fun, attractive, and thoughtful..." | "Website content that I produce will be creative and include original designs..." | A decrease in affective terms from 18% to 6%, resulting in 11% higher bid success and 7% higher price premium. |
10 Notes: Web Appendix S provides the full call for bids and bid examples we used for calculating the degrees of each communicative principle and the corresponding expected lift success. We used the "good practice" call for bids example to devise the bad and good examples for the corresponding freelance bid.
Freelancers are not necessarily natural marketers, but their bid formulations determine their marketability. Existing online reputation systems provide some assistance, but they also create entry barriers to new freelancers who first must earn good overall ratings ([13]). Fortunately, winning gigs and achieving price premiums also depends on freelancers' communication. Table 5 includes advice to help freelancers formulate more successful bids and avoid the value trap. In line with the mantra of adaptive selling, the call for bids provides a starting point in which mimicking the buyer's task information and affective intensity increases freelancers' success—even if they provide few task details or seem very impassive. But freelancers should always offer personal information and be concrete. Even if a buyer does not provide personal information or the call is relatively abstract, freelancers' chances of success and obtaining price premiums increase if their bids contain more personal information and are at least somewhat concrete. The strongest predictor of bid success is a preexisting buyer relationship, so more broadly, freelancers should grow their buyer relations.
In examining theoretically grounded communicative aspects, we offer novel insights into how to manage uncertainty in buyer–freelancer exchanges. Intriguingly, we find that communication approaches that do not aim to minimize uncertainty can be effective. Continued research should investigate this notion further and develop additional insights into the exchange implications of linguistic choices in B2B but also B2C and C2C communication on multisided platforms ([53]). For example, affiliative ([66]) or collaborative terms might affect uncertainty and influence exchanges as well. Arguably, the personal characteristics of buyers and freelancers (e.g., gender, education, experience), channel choices ([50]), different sources of uncertainty ([29]), perceived risks ([25]), and spatial distances between buyers and freelancers also might moderate the efficacy of communication aspects, so additional research should specify their influences. For example, if buyers lack the expertise to specify what they want, they might benefit from more ambiguous calls for bids ([38]). Perhaps buyers' communication or alternative factors that we cannot account for (e.g., underestimation of the amount of work required to fulfill the task) influence the final price they pay, too. Efforts to specify these additional effects also might address some of our more controversial findings, such as the evidence that the number of previously commissioned projects by a buyer relates negatively to the number of freelancers who bid. We posit that experienced buyers might prefer freelancers whom they have hired in the past ([48]). Buyers also might have incurred switching costs or supplier dependencies ([29]). Methodologically, we estimated all the models sequentially, as buyers' calls for bids and their success occur prior to freelancers' bids and their success. But an equilibrium approach that estimates these models simultaneously at the bid level could reflect an alternative way to think about the data structure. The concreteness word list we used ([11]) may also require further refinement to differentiate specific concreteness levels among the word set. Finally, the anonymity and speed of exchanges in online freelance marketplaces may make communication particularly important in this context. A comparative analysis of the influence of uncertainty management efforts across different B2B contexts beyond these marketplaces could offer interesting insights, especially if uncertainty avoidance is a central goal.
Footnotes 1 Hari Sridhar
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Online supplement:https:/doi.org/10.1177/00222429211030841
5 Task and personal information, concreteness, and affective intensity do not comprise an exhaustive list of all the lexical elements that might define relevant information content and manners of communication. Various extensions are thus available for further research. However, we note the primacy of these elements in previous research and therefore prioritize them for this initial effort to establish how the communication principles relate to uncertainty perceptions and exchange outcomes in online freelance marketplaces.
6 More stringent term lists using cutoff levels at 3.5 or 4 strongly correlate (r > .60, p < .01) with the list that uses 3 as a cutoff.
References Afifi Walid A. , Burgoon Judee K.. (2000), " The Impact of Violations on Uncertainty and the Consequences for Attractiveness ," Human Communication Research , 26 (2), 203 – 33.
Ba Sulin , Pavlou Paul A.. (2002), " Evidence of the Effect of Trust Building Technology in Electronic Markets: Price Premiums and Buyer Behavior ," MIS Quarterly , 26 (3) 243 – 68.
Bayer Emanuel , Tuli Kapil R. , Skiera Bernd. (2017), " Do Disclosures of Customer Metrics Lower Investors' and Analysts' Uncertainty but Hurt Firm Performance? " Journal of Marketing Research , 54 (2), 239 – 59.
Benoit Kenneth , Nulty Paul , Barber Pablo , Watanabe Kohei , Lauderdale Benjamin. (2018), "Quanteda: Quantitative Analysis of Textual Data," (accessed September 10, 2021), https://cran.r-project.org/web/packages/quanteda/.
Berger Charles R. (2011), " From Explanation to Application ," Journal of Applied Communication Research , 39 (2), 214 – 22.
Berger Charles R. , Calabrese Richard J.. (1975), " Some Explorations in Initial Interaction and Beyond: Toward a Developmental Theory of Interpersonal Communication ," Human Communication Research , 1 (2), 99 – 112.
7 Berger Jonah , Humphreys Ashlee , Ludwig Stephan , Moe Wendy W. , Netzer Oded , Schweidel David A.. (2020), " Uniting the Tribes: Using Text for Marketing Insight ," Journal of Marketing , 84 (1), 1 – 25.
8 Bradac James J. (2001), " Theory Comparison: Uncertainty Reduction, Problematic Integration, Uncertainty Management, and Other Curious Constructs ," Journal of Communication , 51 (3), 456 – 76.
9 Brashers Dale E. (2001), " Communication and Uncertainty Management ," Journal of Communication , 51 (3), 477 – 97.
Brown Penelope , Levinson Stephen C.. (1987), Politeness: Some Universals in Language Usage. Cambridge, UK : Cambridge University Press.
Brysbaert Marc , Warriner Amy Beth , Kuperman Victor. (2014), " Concreteness Ratings for 40 Thousand Generally Known English Word Lemmas ," Behavior Research Methods , 46 (3), 904 – 11.
Collins Nancy L. , Miller Lynn C.. (1994), " Self-Disclosure and Liking: A Meta-Analytic Review ," Psychological Bulletin , 116 (3), 457 – 75.
Constantinides Panos , Henfridsson Ola , Parker Geoffrey G.. (2018), " Introduction—Platforms and Infrastructures in the Digital Age ," Information Systems Research , 29 (2), 381 – 400.
Crosby Lawrence A. , Evans Kenneth R. , Cowles Deborah. (1990), " Relationship Quality in Services Selling: An Interpersonal Influence Perspective ," Journal of Marketing , 54 (3), 68 – 81.
Derlega Valerian J. , Harris Marian Sue , Chaikin Alan L.. (1973), " Self-Disclosure Reciprocity, Liking, and the Deviant ," Journal of Experimental Social Psychology , 9 (4), 277 – 84.
Dimoka Angelika , Hong Yili , Pavlou Paul A.. (2012), " On Product Uncertainty in Online Markets: Theory and Evidence ," MIS Quarterly , 36 (2), 395 – 426.
Edwards Jeffrey R. , Parry Mark E.. (1993), " On the Use of Polynomial Regression Equations as an Alternative to Difference Scores in Organizational Research ," Academy of Management Journal , 36 (6), 1577 – 1613.
Eisenberg Eric M. , Witten Marsha G.. (1987), " Reconsidering Openness in Organizational Communication ," Academy of Management Review , 12 (3), 418 – 26.
Engelhardt Paul E. , Bailey Karl , Ferreira Fernanda. (2006), " Do Speakers and Listeners Observe the Gricean Maxim of Quantity? " Journal of Memory and Language , 54 (4), 554 – 73.
Farris Paul W. , Bendle Neil , Pfeifer Phillip E. , Reibstein David. (2010), Marketing Metrics: The Definitive Guide to Measuring Marketing Performance. London : Pearson Education.
Ghosh Mrinal , John George. (2005), " Strategic Fit in Industrial Alliances: An Empirical Test of Governance Value Analysis ," Journal of Marketing Research , 42 (3), 346 – 57.
Goffman Erving. (1959), The Presentation of Self in Everyday Life. New York : Doubleday.
Goffman Erving. (2008), Interaction Ritual: Essays on Face-to-Face Behavior. New York : Pantheon Books.
Gooty Janaki , Thomas Jane S. , Yammarino Francis J. , Kim Jayoung , Medaugh Melissa. (2019), " Positive and Negative Emotional Tone Convergence: An Empirical Examination of Associations with Leader and Follower LMX ," The Leadership Quarterly , 30 (4), 427 – 39.
Grewal Dhruv , Gotlieb Jerry , Marmorstein Howard. (1994), " The Moderating Effects of Message Framing and Source Credibility on the Price–Perceived Risk Relationship ," Journal of Consumer Research , 21 (1), 145 – 53.
Grewal Dhruv , Monroe Kent B. , Krishnan Ramayya. (1998), " The Effects of Price-Comparison Advertising on Buyers' Perceptions of Acquisition Value, Transaction Value, and Behavioral Intentions ," Journal of Marketing , 62 (2), 46 – 59.
Grice Herbert P. (1975), "Logic and Conversation," in Syntax and Semantics: Speech Acts , Vol. 3, Cole Peter , Morgan Jerry L. , eds. New York : Academic Press , 41 – 58.
Hamilton Mark A. , Hunter John E.. (1998), " The Effect of Language Intensity on Receiver Evaluations of Message, Source, and Topic, " in Persuasion: Advances Through Meta-Analysis , Allen Mike , Preiss Raymond W. , eds. Cresskill, NJ : Hampton Press , 99 – 138.
Heide Jan B. , Weiss Allen M.. (1995), " Vendor Consideration and Switching Behavior for Buyers in High-Technology Markets ," Journal of Marketing , 59 (3), 30 – 43.
Holtgraves Thomas. (1997), " Styles of Language Use: Individual and Cultural Variability in Conversational Indirectness ," Journal of Personality and Social Psychology , 73 (3), 624 – 37.
Homburg Christian , Klarmann Martin , Staritz Sabine. (2012), " Customer Uncertainty Following Downsizing: The Effects of Extent of Downsizing and Open Communication ," Journal of Marketing , 76 (3), 112 – 29.
Hong Yili , Pavlou Paul A.. (2017), " On Buyer Selection of Service Providers in Online Outsourcing Platforms for IT Services ," Information Systems Research , 28 (3), 547 – 62.
Hong Yili , Shao Benjamin B. M.. (2021), " On Factors that Moderate the Effect of Buyer–Supplier Experience on E-Procurement Platforms ," Production and Operations Management , 30 (4), 1034 – 51.
Hong Yili , Wang Chong , Pavlou Paul A.. (2016), " Comparing Open and Sealed Bid Auctions: Evidence from Online Labor Markets ," Information Systems Research , 27 (1), 49 – 69.
Horton John J. (2017), " The Effects of Algorithmic Labor Market Recommendations: Evidence from a Field Experiment ," Journal of Labor Economics , 35 (2), 345 – 85.
Horton John J. (2019), " Buyer Uncertainty About Seller Capacity: Causes, Consequences, and a Partial Solution ," Management Science , 65 (8), 3518 – 40.
Hosman Lawrence A. (2002), " Language and Persuasion, " in The Persuasion Handbook: Developments in Theory and Practice , Dillard James Price , Pfau Michael , eds. Thousand Oaks, CA : SAGE Publications , 371 – 90.
Humphreys Ashlee , Isaac Mathew S. , Wang Rebecca Jen-Hui. (2020), " Construal Matching in Online Search: Applying Text Analysis to Illuminate the Consumer Decision Journey ," Journal of Marketing Research (published online September 2) , DOI: 10.1177/0022243720940693.
Jap Sandy D. (2007), " The Impact of Online Reverse Auction Design on Buyer–Supplier Relationships ," Journal of Marketing , 71 (1), 146 – 59.
Jones Quentin , Ravid Gilad , Rafaeli Sheizaf. (2004), " Information Overload and the Message Dynamics of Online Interaction Spaces: A Theoretical Model and Empirical Exploration ," Information Systems Research , 15 (2), 194 – 210.
Joshi Ashwin W. (2009), " Continuous Supplier Performance Improvement: Effects of Collaborative Communication and Control ," Journal of Marketing , 73 (1), 133 – 50.
Kanuri Vamsi K. , Chen Yixing , Sridhar Shrihari. (2018), " Scheduling Content on Social Media: Theory, Evidence, and Application ," Journal of Marketing , 82 (6), 89 – 108.
Keleman Mihaela. (2000), " Too Much or Too Little Ambiguity: The Language of Total Quality Management ," Journal of Management Studies , 37 (4), 483 – 89.
Khosarvizadeh Parvaneh , Sadehvandi Nikan. (2011), " Some Instances of Violations and Flouting of the Maxim of Quantity by the Main Character (Barry and Tim) in Dinner for Schmucks ," IPEDR , 26. Singapore: IACSIT Press, http://ipedr.com/vol26/25-ICLLL%202011-L00061.pdf.
Kokkodis Marios , Ipeirotis Panagiotis G.. (2016), " Reputation Transferability in Online Labor Markets ," Management Science , 62 (6), 1687 – 706.
Koponen Jonna Pauliina , Rytsy Saara. (2020), " Social Presence and E-Commerce B2B Chat Functions ," European Journal of Marketing , 54 (6), 1205 – 24.
Krippendorff Klaus. (2013), Content Analysis. An Introduction to Its Methodology , 3rd ed. Thousand Oaks, CA : SAGE Publications.
Lanzolla Gianvito , Frankort Hans T. W.. (2016), " The Online Shadow of Offline Signals: Which Sellers Get Contracted in Online B2B Marketplaces? " Academy of Management Journal , 59 (1), 207 – 31.
Larrimore Laura , Jiang Li , Larrimore Jeff , Markowitz David , Gorski Scott. (2011), " Peer to Peer Lending: The Relationship Between Language Features, Trustworthiness, and Persuasion Success ," Journal of Applied Communication Research , 39 (1), 19 – 37.
Lawrence Justin M. , Crecelius Andrew T. , Scheer Lisa K. , Patil Ashutosh. (2019), " Multichannel Strategies for Managing the Profitability of Business-to-Business Customers ," Journal of Marketing Research , 56 (3), 479 – 97.
Liu Yeyi , Eisingerich Andreas B. , Auh Seigyoung , Merlo Omar , Chun Hae Eun Helen. (2015), " Service Firm Performance Transparency: How, When, and Why Does it Pay Off? " Journal of Service Research , 18 (4), 451 – 67.
Ludwig Stephan , de Ruyter Ko , Friedman Mike , Brüggen Elisabeth C. , Wetzels Martin , Pfann Gerard. (2013), " More Than Words: The Influence on Affective Content and Linguistic Style Matches in Online Reviews on Conversion Rates ," Journal of Marketing , 77 (1), 87 – 103.
Luo Xueming , Tong Siliang , Lin Zhijie , Zhang Cheng. (2021), " The Impact of Platform Protection Insurance on Buyers and Sellers in the Sharing Economy: A Natural Experiment ," Journal of Marketing , 85 (2), 50 – 69.
Ma Zhenfeng , Dubé Laurette. (2011), " Process and Outcome Interdependency in Frontline Service Encounters ," Journal of Marketing , 75 (3), 83 – 98.
Ma Xiao , Hancock Jeffery T. , Mingjie Kenneth Lim , Naaman Mor. (2017), "Self-Disclosure and Perceived Trustworthiness of Airbnb Host Profiles," in CSCW '17: Proceedings of the 2017 ACM Conference on Computer Supported Cooperative Work and Social Computing. New York: ACM, 2397–409.
Manyika James , Lund Susan , Robinson Kelsey , Valentino John , Dobbs Richard. (2015), A Labor Market that Works: Connecting Talent with Opportunity in the Digital Age. San Francisco, CA : McKinsey Global Institute.
McFarland Richard G. , Challagalla Goutam N. , Shervani Tasadduq A.. (2006), " Influence Tactics for Effective Adaptive Selling ," Journal of Marketing , 70 (4), 103 – 17.
Mooi Erik A. , Ghosh Mrinal. (2010), " Contract Specificity and its Performance Implications ," Journal of Marketing , 74 (2), 105 – 20.
Moon Youngme. (2000), " Intimate Exchanges: Using Computers to Elicit Self-Disclosure from Consumers ," Journal of Consumer Research , 26 (3), 323 – 39.
Mullins Ryan , Agnihotri Raj , Hall Zachary. (2020), " The Ambidextrous Sales Force: Aligning Salesperson Polychronicity and Selling Contexts for Sales-Service Behaviors and Customer Value ," Journal of Service Research , 23 (1), 33 – 52.
Packard Grant , Berger Jonah. (2020), " How Concrete Language Shapes Customer Satisfaction ," Journal of Consumer Research , 47 (5), 787 – 806.
Packard Grant , Moore Sarah G. , McFerran Brent. (2018), " (I'm) Happy to Help (You): The Impact of Personal Pronoun Use in Customer–Firm Interactions ," Journal of Marketing Research , 55 (4), 541 – 55.
Palmatier Robert W. , Dant Rajiv P. , Grewal Dhruv. (2007), " A Comparative Longitudinal Analysis of Theoretical Perspectives of Interorganizational Relationship Performance ," Journal of Marketing , 71 (4), 172 – 94.
Pan Lingling , McNamara Gerry , Lee Jennifer J. , Haleblian Jerayr , Devers Cynthia E.. (2018), " Give It to Us Straight (Most of the Time): Top Managers' Use of Concrete Language and its Effect on Investor Reactions ," Strategic Management Journal , 39 (8), 2204 – 25.
Park Sungho , Gupta Sachin. (2012), " Handling Endogenous Regressors by Joint Estimation Using Copulas ," Marketing Science , 31 (4), 567 – 86.
Pennebaker James W. , Boyd Ryan L. , Jordan Kayla , Blackburn Kate. (2015), "The Development and Psychometric Properties of LIWC2015," Austin: University of Texas at Austin, https://repositories.lib.utexas.edu/bitstream/handle/2152/31333/LIWC2015_LanguageManual.pdf.
Pennebaker James W. , Stone Lori D.. (2003), " Words of Wisdom: Language Use Over the Life Span ," Journal of Personality and Social Psychology , 85 (2), 291 – 301.
Pollock Patrick , Lüttgens Dirk , Piller Frank T.. (2019), " Attracting Solutions in Crowdsourcing Contests: The Role of Knowledge Distance, Identity Disclosure, and Seeker Status ," Research Policy , 48 (1), 98 – 114.
Putman Linda , Jones Tricia. (1982), " The Role of Communications in Bargaining ," Human Communications Research , 8 , 262 – 80.
Rao Akshay R. , Monroe Kent B.. (1996), " Causes and Consequences of Price Premiums ," Journal of Business , 69 (4), 511 – 35.
Shannon Claude E. , Weaver Warren. (1949), A Mathematical Model of Communication. Urbana, IL : University of Illinois Press.
Singh Sunil K. , Marinova Detelina , Singh Jagdip. (2020), " Business-to-Business E-Negotiations and Influence Tactics ," Journal of Marketing , 84 (2) 47 – 68.
Singh Jagdip , Sirdeshmukh Deepak. (2000), " Agency and Trust Mechanisms in Consumer Satisfaction and Loyalty Judgments ," Journal of the Academy of Marketing Science , 28 (1), 150 – 67.
Snijders Tom A. , Bosker Roel J.. (2011), Multilevel Analysis: An Introduction to Basic and Advanced Multilevel Modeling. London : SAGE Publications.
Soliz Jordan , Giles Howard. (2014), " Relational and Identity Processes in Communication: A Contextual and Meta-Analytical Review of Communication Accommodation Theory ," Annals of the International Communication Association , 38 (1), 107 – 44.
Sridhar Hari , Mittal Vikas. (2020), " Busting the Value Trap: How B2B Companies Can Increase Sales," BrandExtract (accessed September 10, 2021), https://www.brandextract.com/Insights/Articles/Busting-the-Value-Trap-How-B2B-Companies-Can-Increase-Sales/.
Srivastava Shirish C. , Chandra Shalini. (2018), " Social Presence in Virtual World Collaboration: An Uncertainty Reduction Perspective Using a Mixed Methods Approach ," MIS Quarterly , 42 (3), 779 – 803.
Verbeke Willem , Dietz Bart , Verwaal Ernst. (2011), " Drivers of Sales Performance: A Contemporary Meta-Analysis. Have Salespeople Become Knowledge Brokers? " Journal of the Academy of Marketing Science , 39 (3), 407 – 28.
Weiss Allen M. , Lurie Nicholas H. , MacInnis Deborah J.. (2008), " Listening to Strangers: Whose Responses Are Valuable, How Valuable Are They, and Why? " Journal of Marketing Research , 45 (3), 425 – 36.
Williams M. Lee. (1980), " The Effect of Deliberate Vagueness on Receiver Recall and Agreement ," Communication Studies , 31 (1), 30 – 41.
Zheng Xu , Griffith David A. , Ge Ling , Benoliel Uri. (2020), " Effects of Contract Ambiguity in Interorganizational Governance ," Journal of Marketing , 84 (4), 147 – 67.
Zhou Qiang (Kris) , Allen B.J. , Gretz Richard T. , Houston Mark B.. (2021), " Platform Exploitation: When Service Agents Defect with Customers from Online Service Platforms ," Journal of Marketing (published online February 22), DOI: 10.1177/00222429211001311.
~~~~~~~~
By Stephan Ludwig; Dennis Herhausen; Dhruv Grewal; Liliana Bove; Sabine Benoit; Ko de Ruyter and Peter Urwin
Reported by Author; Author; Author; Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 36- Complaint Publicization in Social Media. By: Golmohammadi, Alireza; Havakhor, Taha; Gauri, Dinesh K.; Comprix, Joseph. Journal of Marketing. Nov2021, Vol. 85 Issue 6, p1-23. 23p. 1 Diagram, 8 Charts, 1 Graph. DOI: 10.1177/00222429211002183.
- Database:
- Business Source Complete
Record: 37- Conducting Research in Marketing with Quasi-Experiments. By: Goldfarb, Avi; Tucker, Catherine; Wang, Yanwen. Journal of Marketing. May2022, Vol. 86 Issue 3, p1-20. 20p. 3 Charts. DOI: 10.1177/00222429221082977.
- Database:
- Business Source Complete
Conducting Research in Marketing with Quasi-Experiments
This article aims to broaden the understanding of quasi-experimental methods among marketing scholars and those who read their work by describing the underlying logic and set of actions that make their work convincing. The purpose of quasi-experimental methods is, in the absence of experimental variation, to determine the presence of a causal relationship. First, the authors explore how to identify settings and data where it is interesting to understand whether an action causally affects a marketing outcome. Second, they outline how to structure an empirical strategy to identify a causal empirical relationship. The article details the application of various methods to identify how an action affects an outcome in marketing, including difference-in-differences, regression discontinuity, instrumental variables, propensity score matching, synthetic control, and selection bias correction. The authors emphasize the importance of clearly communicating the identifying assumptions underlying the assertion of causality. Last, they explain how exploring the behavioral mechanism—whether individual, organizational, or market level—can actually reinforce arguments of causality.
Keywords: quasi-experiments; marketing methods; econometrics
Quasi-experimental methods have been widely applied in marketing to explain changes in consumer behavior, firm behavior, and market-level outcomes. "Quasi-experiment" refers to the use of an experimental mode of analysis and interpretation to data sets where the data-generating process is not itself intentionally experimental ([23]). Instead, quasi-experimental research uses variation that occurs without experimental intervention but is nonetheless exogenous to the particular research setting. Work using quasi-experiments in marketing settings has used events such as weather, geographic boundaries, contract changes, shifts in firm policy, individual-level life changes, and regulatory changes to approximate a real experiment. In each case, an external shock creates a source of exogenous variation that the researcher uses to establish a causal relationship between the variation and the outcome of interest.
Companies also use quasi-experimental methods to understand the consequences of key business actions. For example, [17] analyzed a quasi-experiment where eBay shut all the paid search advertising on Bing during a dispute with Microsoft but lost little traffic. These quasi-experimental results inspired a follow-up field experiment where eBay randomized suspension of its branded paid search advertising and found results consistent with the quasi-experiment. Reflecting the importance of such methods at firms, some companies provide causal inference training for their data scientists ([32]; [81]). The ability to make causal claims is highly valuable in academia and in practice. This article aims to help both marketing scholars and practitioners conduct and evaluate the credibility of quasi-experimental studies.
Quasi-experimental research, as in much work in applied statistics, begins with the equation . The focus is then on whether a change in a single covariate x in the vector of can be demonstrated to cause a change in . This focus often enables the exploration of foundational questions in marketing, because marketers often have data representing the actions of many individual consumers or clients and need to understand the causal relationship between a particular x and y to make decisions about whether and how much x to use.
A marketing article that successfully uses the quasi-experimental econometric approach considers the following nine topics, which are echoed in the structure of this article:
- Research Question: Do We Care Whether x Causes y?
- Data Question: How Can Researchers Find Data with Quasi-Experimental Variation in x?
- Identification Strategy: Does x Cause y to Change?
- Empirical Analysis: How Can Researchers Estimate the Effect of x on y?
- Challenges to Research Design: What if Variation in x Is Not Exogenous?
- Robustness: How Robust Is the Effect of x on y?
- Mechanism: Why Does x Cause y to Change?
- External Validity: How Generalizable Is the Effect of x on y?
- Apologies: What Remains Unproven and What Are the Caveats?
We start by explaining why quasi-experimental scholars may appear obsessed with identification, and how this influences the choice of research question and data setting. Quasi-experiments come in different shades, ranging from an almost completely random exogenous shock to where the treatment assignment is only partly random. We suggest different frameworks to accommodate various levels of evidence depending on the strength of the underlying identification argument. We then turn to the importance of understanding the underlying mechanism behind the causal result. Typically, this means showing that the effect is largest where theory would predict and is smallest where theory would predict a negligible effect. We also emphasize that researchers need to be clear about the external validity of their study and apologize for what remains unconvincing.
Why are quasi-experimental scholars seemingly obsessed with identification? Identification is defined by the challenge that "many different theoretical models and hence many different causal interpretations may be consistent with the same data" ([58], p. 47). However, effective decision making requires an understanding whether a measured relationship is indeed causal.
One way to describe this issue is through the "potential outcomes approach" developed by Jerzy Neyman, Donald Rubin, and others ([86]).[ 5] This approach starts with the insight that for any discrete treatment—which could be an event or explicit policy ( )—each individual has two possible outcomes:
- if the individual i experiences the treatment , and
- if the individual i does not experience the treatment
The difference between the two is the causal effect. The identification problem occurs because a single individual i cannot both receive the treatment and not receive the treatment at the same time. Therefore, only one outcome is observed for each individual at any point in time. The unobserved outcome is called the "counterfactual." The unobservability of the counterfactual means that assumptions are required. The identification problem means that those who experience D, and those who do not, are different in unobserved ways.
Random assignment solves the inference problem, as the "unobserved ways" should not matter ex ante ([31]). [88], p. 13) explain that "if implemented correctly, random assignment creates two or more groups of units that are probabilistically similar to each other on the average." With enough people assigned randomly to one group or another, the only meaningful difference between the groups will be a result of the treatment.
Therefore, random assignment is often called the gold standard of identification ([70], p. 8). [12], p. 11) emphasize that "the most credible and influential research designs use random assignment." That said, we should be clear that field experiments are merely a gold standard for being able to plausibly claim causality, not the gold standard for empirical work ([34]). Indeed, in many marketing situations, experiments are not feasible, appropriate, or affordable ([44]).
Quasi-experimental work, by contrast, is aimed to identify exogenous shocks or events that can approximate random assignment. Given that assignment is not random, a researcher's goal is to make the unobserved ways in which the treatment and control groups differ as untroubling as possible to the researcher and the reader and thereby mimic random assignment as closely as possible.
The first and hardest stage in this process is identifying a question in which marketing scholars, managers, or policy makers actually care whether x causes y. This is difficult because many of the s and s for which we can measure a causal relationship are (unfortunately) uninteresting. Therefore, researchers who do quasi-experimental research do best if they start not with the data or an exogenous shock but instead start by asking themselves, "Suppose I convincingly showed that an increase in x increases —who would care about this substantive issue?
This means that the first stage requires the identification of a causal relationship that would be of interest to marketers or policy makers because their decisions will be usefully informed by a clear understanding of the consequences of a particular action. As marketing technology and practices change, the number of measurable, interesting, and unanswered questions grows. A variety of editorials in this journal and elsewhere focus on how researchers can identify important issues. For example, the January 2021 special issue of the Journal of Marketing was dedicated to finding important marketing research questions, as highlighted in the editorial ([36]. Other editorials that discuss ways to identify important research questions are [59]) and [26] in the Journal of Marketing, [87]) in the Journal of Consumer Research, [96] in Marketing Science, and [50]) in the Journal of Marketing Research.
[12], p. 7) explain that an identification strategy "describe[s] the manner in which a researcher uses observational data or data that is not generated as part of an intentional experiment, to approximate a real experiment." They suggest first thinking of an ideal randomized experiment that can address the research question. This helps the researcher see clearly why an effect may not be identified causally in a nonexperimental setting.
As [75], p. 151) discusses, "Good natural experiments are studies in which there is a transparent exogenous source of variation in the explanatory variables that determine treatment assignment." Unfortunately, there is no universally accepted interpretation of what it means to have a transparent exogenous source of variation. Therefore, [75] (p. 151) emphasizes the importance of clarifying identification assumptions and understanding the institutional setting, stating, "If one cannot experimentally control the variation one is using, one should understand its source." In the marketing context, [84] discusses the dangers of using methods in which the source of the exogenous variation is either poorly understood or only weakly related to the correlation of interest.
Much of the work using quasi-experimental variation in marketing settings uses mundane but easily understood events such as contract changes, regulation, individual-level life changes, or shifts in firm policy that did not occur because of an anticipated effect on the outcome of interest. In some sense, some of the best sources of exogenous variation are mundane: nonmundane sources of variation such as global pandemics or earthquakes tend to be associated with other things happening that make it difficult to establish a clean causal relationship.
Table 1 lists several example quasi-experimental papers published in 2018, 2019, and 2020 in the Journal of Marketing, the Journal of Marketing Research, and Marketing Science. This table also summarizes the source of variation these articles use, spanning contractual changes; ecological variation (e.g., weather); geography; and macroeconomic, individual, organizational, and regulatory changes. It is useful to consider in turn why each of these sources of variation can approximate random assignment.
Graph
Table 1. Examples of Quasi-Experiment Studies in Journal of Marketing, Journal of Marketing Research, and Marketing Science in 2018–2020.
| Quasi-Experimental Variation |
|---|
| General Category | Source | Article | Research Question |
|---|
| Contractual | Timing of American–Orbitz disputes to evaluate the absence of a major airline from a popular aggregator on consumer search | Akca and Rao (2020) | Who has more market power in the airline-aggregator relationships? |
| Timing of the introduction of the New York Times paywall | Pattabhiramaiah, Sriram, and Manchanda (2019) | How does a paywall affect readership and site traffic? |
| Ecological | Variation in the forecast error of the pollen levels | Thomas (2020) | How much does advertising affect purchases of allergy products? |
| Geographical | Discontinuities in the level of advertising at the borders of DMAs | Shapiro (2020) | Does advertising affect consumer choice of health insurance? |
| Discontinuities in the level of political ads at the borders of DMAs | Wang, Lewis, and Schweidel (2018) | How does political advertising source and message tone affect vote shares and turnout rates in 2010 and 2012 Senatorial elections? |
| Individual | Timing of users' adoption of a music streaming service | Datta, Knox, and Bronnenberg (2018) | How does a streaming service affect total music consumption? |
| Variation in national ad exposures due to the local game outcomes | Hartmann and Klapper (2018) | How do Super Bowl ads affect brand purchases? |
| Macroeconomic | Variation in income and wealth due to the recession between 2006 and 2009 | Dube, Hitsch, and Rossi (2018) | Do income and wealth affect demand for private label products? |
| Organizational | Discontinuity in the rounding rule that TripAdvisor uses to convert average ratings into displayed ratings | Hollenbeck, Moorthy, and Proserpio (2019) | How do online reviews affect advertising spending in the hotel industry? |
| Timing of data breach and variation whether customer information was breached in a data breach event | Janakiraman, Lim, and Rishika (2018) | How does a data breach announcement affect customer spending and channel migration? |
| Variation in timing of adoption of front-of-package nutritional labels across categories | Lim et al. (2020) | Do front-of-package nutritional labels affect nutritional quality for other brands in a category? |
| Regulatory | Timing of the Massachusetts open payment law | Guo, Sriram, and Manchanda (2020) | Do payment disclosure laws affect physician prescription behavior? |
| Enforcement of minimum advertisement price policies | Israeli (2018) | What is the effect on violations if firms improve digital monitoring and enforcement of minimum advertised price policies? |
| Timing of India's foreign direct investment liberalization reform in 1991 | Ramani and Srinivasan (2019) | How do firms respond to foreign direct investment liberalization? |
1 Notes: DMA = designated market area.
To find plausibly exogenous variation in timing, it often depends on an argument that the exact timing of a measure is plausibly exogenous. [28] argued that the timing of a dispute between the Associated Press and Google was essentially random as it was influenced by a contract negotiated many years previously, and so the timing could be used to study the effect of the removal of content from news aggregators on downstream news websites.
Generally, within-season variation in weather is plausibly exogenous. For example, [95] uses quasi-experimental variation in actual and expected pollen counts. Key to the identification strategy is the focus on deviations from what was expected by firms.
Work using geographical boundaries often exploits the fact that people who live on either side of a demarcated geographic border are similar enough to be thought of as being randomized across them. For example, by looking at a remote border of Maryland that was geographically isolated from the rest of the state, [ 8] were able to argue that the imposition of sales tax for those who lived on one side of the border was random, relative to those people who lived nearby but just happened to be over the state border.
It is also possible to take leverage of macroeconomic shocks. For example, [40] use the Great Recession as a key source of the variation on household incomes over time. They exploit the within-household variation in private label shares associated with within-household changes in income and wealth. The identifying assumption is that, conditional on all other factors, including an overall trend, within-household changes in income and wealth are as good as randomly assigned or exogenous changes.
Plausibly exogenous variation can also be argued to occur at the individual level. For example, [20] use consumer migration to new locations as a quasi-experiment to study the causal impact of past experiences on current purchases. They argue that while migration is not necessarily random, the precise direction of migration can be, at least with respect to local brand market shares.
Shifts in firm policy and organizational events can also be leveraged as a source of variation. For example, [64] assess the change in customer behaviors between those whose information is breached and those whose information is not. The identification assumption is that the assignment of customers into the data breach group is likely to be random.
Many papers also use the timing of regulatory changes as a source of variation. The argument here is typically that though the imposition of regulation may not be random, the timing of the regulation is. For example, [97] use a change in Massachusetts regulation of home sale listings to identify the effect of information about time on the market on house prices, and [76] use a change in the standardized nutrition labels on food products required by the Nutrition Labeling and Education Act and investigate how the Act changes brand nutritional quality.
This discussion emphasizes that there are many potential sources of exogenous variation that can approximate a randomized experiment. We emphasize that typically the best papers focus on the research question first, and then imagine what the idealized experiment would look like to identify an actual quasi-experiment.
To convince a reader that an identification strategy is valid requires two steps. First, the researcher must explain where the variation they are calling exogenous comes from. This requires institutional knowledge and careful research into the setting. Second, the researcher needs to demonstrate that the relationship between the variation and the outcome of interest is very likely driven by the relationship between x and y and not by some other factor.
To achieve the second requirement, it is useful to think about defending the experiment in terms of the exclusion restriction. Although the term "exclusion restriction" is often used specifically for instrumental variables, it is also a useful concept for other quasi-experiments. The exclusion restriction states that the quasi-experiment only affects y because it affects x.
There are a variety of ways in which the exclusion restriction can fail, and so researchers look for exogenous variation in x that will have no direct effect on y. For example, [91] use wind speed as a quasi-experiment to provide an exogenous driver of posting to a user-generated content site about windsurfing. This allows them to understand the relationship between content creation and the creation of social ties. The argument for the exclusion restriction is that there is no other plausible way that wind could affect the creation of social ties except through content creation. As they mention in the paper, plausible challenges to this exclusion restriction are that windy days could affect friendship formation directly because users meet future online friends at windier surf locations. To address such challenges, the researchers present empirical data to suggest that the social ties that are being formed do not seem to reflect geography.
Another example is [66], which examines the effect of delays in the early part of a banking technology adoption process on ultimate usage. Through a quasi-experiment that provides a source of exogenous variation in delays, they exploit the fact that Germany has a highly regulated system of public holidays and vacations that vary at the state level to prevent freeways from becoming overly congested. This leads to delays in technology adoption in that particular period to customers in one state, and not in others. The exclusion restriction is that there is no other reason that vacations or public holidays in the few days surrounding adoption would affect ultimate usage except through delaying the ability to navigate the security protocols required to sign up for the online banking service. One challenge for the exclusion restriction could be that individuals who sign up for a banking service around public holidays are somehow systematically different from others in terms of their laziness or motivation. To counter this challenge, the researchers present evidence that users are not different along any observable dimension.
The exclusion restriction can also fail because of spillovers between groups that receive the exogenous shock or treatment and those that do not. The assumption that treatment of unit i affects only the outcome of unit i is called the stable unit treatment value assumption (SUTVA) in the treatment literature ([10]; [61]). This is not a trivial assumption. For example, [ 5] use the 2011 Orbitz–American Airlines disputes as an exogeneous event that led to a five-month period in which American fares were not displayed on Orbitz. The authors use this dispute to identify which company was hurt the most in terms of site visits and purchases. The SUTVA requires a valid control group such that the Orbitz–American Airlines disputes have no spillover on that group. As a result, the authors chose not to use airfare- or hotel-booking websites as a control due to the possible spillovers from Orbitz to other websites where customers can purchase. Instead, the authors used consumers' search of Lonely Planet as the control, because Lonely Planet is a travel website that is rarely used for bookings. The underlying idea is that an exclusion restriction cannot hold if the fact that one group was treated may also affect the control group's behavior. The SUTVA is therefore part of an argument that researchers make about an appropriate exclusion restriction.
Importantly, there is no formula for a convincing explanation and defense of the empirical identification strategy in quasi-experiments. Except in cases of random assignment, it is not possible to prove that the identifying assumption is right. Instead, the objective for the authors is to pursue projects only when they can convince themselves (and their readers) that the causal interpretation is more plausible than other possible explanations. It is impossible to prove the validity of a quasi-experiment, such as whether one set of U.S. states serves as a legitimate control group for another or whether the exclusion restriction holds in instrumental variables. The credibility of any quasi-experimental work therefore relies on the plausibility of the argument for causality rather than on any formal statistical test.
After establishing the identification assumption through the underlying framework of an exclusion restriction, the next step is to explore the data and conduct analysis that allows measurement of the effect of interest. This measured causal relationship is what has the potential to inform decision making. We discuss three different regression analysis frameworks using quasi-experiments: difference-in-differences (DID), regression discontinuity, and instrumental variables (IV). At the heart of all these strategies is a similar argument about the validity of the quasi-experiment.
Table 2 outlines eight key steps in the three regression analysis frameworks. As pointed out by [54]) and others, the techniques are very similar in terms of the underlying econometric theory. However, though similar in the conceptual ideas, in terms of practical implementation, presentation, and how the researcher should best reassure their audience about the validity of the technique, there are some differences, which we expand on. The three frameworks differ in the first four implementation steps. We discuss the first four steps for each of the three regression analysis frameworks and highlight the issues in common across the three analysis frameworks in the last four steps. We also emphasize that many excellent papers do not implement each step, and this description is not intended to lead to unproductive dogmatism.
Graph
Table 2. Quasi-Experimental Regression Analysis Frameworks.
| Difference-in-Differences | Regression Discontinuity | Instrumental Variables |
|---|
| Identification | Clarify the source of the shock, provide evidence why the shock can be seen as quasi-experimental, be clear on the identifying assumptions, and be transparent on the potential confoundedness. | Justify the source of the fixed threshold, and whether the assignment to the treatment is determined, either completely or partly, by the value of the predictor on either side of a fixed threshold. | Justify why the IV moves the endogenous covariate as if they are an experiment; explain the exclusion restriction. |
| Raw data | Test whether those who receive the treatment are similar to those who do not; whether the parallel assumptions are satisfied; illustrate the trajectory. | Provide evidence that the threshold is arbitrarily determined and not linked to underlying discontinuities in effects. | Regress the outcome directly on the instrument and show that the instrument has the expected direct effect. |
| Data analysis | Apply difference-in-differences regression framework in Equations 1 and 2 and adapt accordingly for other variations. | Apply regression continuity framework in Equation 3. | Report the first stage and determine whether the instruments are strong. Apply 2SLS in Equations 4 and 5 and conduct relevant tests. |
| Standard errors | Cluster at the level of treatment to account for within-unit correlation of the error term over time. | Use robust standard errors, do not cluster on a discrete variable | Cluster at the level of treatment to account for within-unit correlation of the error term over time. |
| Robustness checks | | Conduct multiple robustness checks. |
| Mechanism checks | | Measure mediator variables or show moderation analysis. |
| External validity | | Discuss the assumptions required to capture the ATE. |
| Apologies and caveats | | Apologize for all that is still unproven and give caveats. |
All of these methods implicitly rely on throwing out variation in the data that is not exogenous. In other words, they involve losing power to support the exogeneity assumption. This means that quasi-experimental work cannot use the R-squared as a useful summary of the appropriateness of the model. [41]) provide some useful evidence. While R-squared or a comparison of log-likelihoods is very useful in many other contexts (e.g., forecasts), benchmarking quasi-experimental analyses against other methods by using the R-squared will be misleading.
A standard DID analysis compares a treatment group and a (quasi-) control group before and after the time of the treatment. The "treatment" is not truly a random experiment but, rather, some "shock." Unlike a simple comparison (or single-difference) analysis, DID methods generate a baseline for comparison between the treatment and the control group. By highlighting the change in the treatment group relative to the control group, DID enables the researcher to control for many of the most obvious sources of heterogeneity across groups.
[47] is an example of a DID paper. The authors examine the impact of privacy regulation on the effectiveness of online advertising. In late 2003 and early 2004, many European countries implemented new restrictions on how firms could collect and use online data. The paper uses data on the success of nearly 10,000 online display advertising campaigns in Europe, the United States, and elsewhere between 2001 and 2008. The authors compare the change in effectiveness of the ad campaigns inside and outside Europe. Therefore, the first difference is the change in the campaign effectiveness, and the second difference is the change in Europe relative to elsewhere. Compared with before the regulation, ad campaigns became 2.8% less effective in Europe after the regulation. In contrast, compared with before the European regulation, ad campaigns became.1% more effective outside of Europe after the European regulation was implemented.
The first step is to clearly lay out the identifying assumptions. [47], p. 63) state that "the identification is based on the assumption that coinciding with the enactment of privacy laws, there was no systematic change in advertising effectiveness independent of the law" and that "the European campaigns and the European respondents do not systematically change over time for reasons other than the regulations." A substantial portion of the paper is devoted to providing empirical evidence regarding whether ( 1) European ad agencies invest less in their ad creatives relative to non-European ad agencies after the laws, ( 2) the demographic profile of the respondents is representative of the general population of internet users, and ( 3) there may have been a change in European consumer attitudes and responsiveness to online advertising separate from the Privacy Directive.
The analysis of consumer attitudes and ad responsiveness is based on a concern about unobservables, specifically whether there are alternative explanations for the measured changes in the attitudes of survey participants toward online advertising that were separate but contemporaneous with the change in European privacy laws. To check for such unobserved heterogeneity, [47]) examine the behavior of Europeans on non-European websites that are not covered by the European Privacy Directive to see if a similar shift in behavior can be observed, and they find evidence that changes in behavior are connected with the websites covered by the law, rather than the people taking the survey. The identification exclusion criterion is further validated by a mirror image of the falsification test by looking at residents of non–European Union (EU) countries who visited EU websites. When residents of non-EU countries visit EU websites, the ads are less effective in the postperiod. In contrast, when residents of these non-EU countries visit non-EU websites, there is no change in effectiveness before and after the EU regulation. Therefore, the results appear to be driven by what happens at EU websites rather than by a difference in how Europeans behave relative to non-Europeans.
The second step is to explore the raw data. Before applying the DID framework, it is important to explore the raw data to assess whether the quasi-experiment appeared to have an effect. For example, when a treatment occurs in the middle of a time series, many papers use a graph that shows that before the treatment occurred, the treatment and control groups were on a similar trend and had similar values; then, after the treatment occurred, the trajectory of the treatment group diverged from the control group.
Researchers should also assess whether their quasi-experimental setting meets the parallel trend assumption while exploring their raw data. This involves demonstrating that behaviors were similar in the period prior to the policy change across the treatment and control groups. Depending on the length of the time period, this can be done by conducting two-sample mean comparisons for each pretreatment period or by running a linear regression and looking at the time trend differences between the control and treatment groups. It is also often ideal to simply plot the raw data to support this point.
Though it is desirable and convincing if the main effect of interest can be seen through descriptive statistics or visualization, we caution that this is not always possible. This may happen because effect sizes are small—as they often are in advertising—or because there is variation in the data that is best addressed using a regression framework.
Although a DID regression can be represented in a 2 × 2 table, it is usually analyzed with regression analysis to allow researchers to control for factors that may change over time and across individuals. The simplest version of this regression is as follows:
Graph
( 1)
where y is the outcome of interest; i represents the individual, firm, or other cross-sectional unit of interest; t represents the time period; and represents the error. The key focus of the DID specification is on , which captures the explanatory power of the crucial interaction term. Usually, researchers add controls to address additional omitted variables concerns, such as an observed covariate that may not affect the treatment and control groups in the same way.
When researchers have access to a panel, it is possible to address this concern directly by observing the same individuals, or the same campaigns, both before and after the timing of the treatment. It is then possible to add fixed effects to control for all individual-level (time-invariant) heterogeneity. Furthermore, if the data set includes more than two time periods, then adding time-specific fixed effects controls for all time-period-specific heterogeneity (across all individuals). With individual and time fixed effects, the DID regression is
Graph
( 2)
where is the individual-level fixed effect and is the time-period fixed effect. The fixed effects mean that the main effect of and drop out because they are collinear with the fixed effects. If possible, it is often desirable to difference out, rather than estimate, the fixed effects to avoid bias due to the incidental parameters problem (e.g., [67]). Most standard statistical packages automatically condition out the individual fixed effects from fixed effects panel data models where possible.[ 6]
Though changes over time are common, DID methods do not require a time-series component. For example, [48] examine the impact of offline advertising restrictions on prices for keyword advertisements. The first difference is the keyword ad prices in states that have restrictions compared with states that do not. The second difference is the keywords that are affected by the restrictions compared with the keywords that are not.
For quasi-experimental analyses that do examine changes over time, another tweak is that quasi-experimental treatment can occur at different times, meaning that individuals are treated at different times and that the variable can change with subscripts i and t. For example, [27] study how a book review posted on Amazon affects sales of that book on Amazon, compared with sales of that book at barnesandnoble.com. Different books are reviewed at different times. Therefore, the treatment here is the review a book receives, and the period occurs at different times for different books. [14], [19], [21], [35]), [49], and [85] explore the effects of variation in treatment timing. The issue is that because a fixed-effects DID estimator is a weighted sum of the treatment effect in each group and at each period, even though the weights sum to one, negative weights may arise when there is a substantial amount of heterogeneity in the treatment effects over time. A related concern has been highlighted by [45], who emphasize the problems that occur when both the treatment effect and treatment variance vary across groups.
This means that researchers should be cautious in summarizing time-varying treatment effects with a homogeneous treatment effect as in the two-way DID framework if there is a substantial timing dimension. To address these issues, researchers have proposed a variety of estimators that allow for a cleaner comparison between the treated group and the control group. Both [21] and [35]) propose new estimands to estimate treatment effects in the presence of heterogeneity across groups and over time.[ 7] Another approach is taken by [93], who discuss corrections that should be applicable in a situation where leads or lags might be expected.
Overall, DID is a powerful tool for helping identify the causal relationships that managers need for effective decision making. It can enable researchers to control for time-invariant individual-level heterogeneity, relying on the assumption that differences in the changes that the treatment and control groups experience over time are driven by the impact of the treatment.
Regression discontinuity is a quasi-experimental technique in which the "experiment" relies on an exogenous arbitrary threshold. As [60], p. 616) put it, "The basic idea behind the RD [regression discontinuity] design is that assignment to the treatment is determined, either completely or partly, by the value of the predictor being on either side of a fixed threshold."
Regression discontinuity may be particularly useful to marketing scholars. [56] argue that many marketing interventions are based on thresholds of real or expected consumer or firm behavior. For example, direct mail companies use the scoring policies for recency, frequency, monetary models. Consumers just above and just below the cutoff should be similar in many dimensions, and their outcomes can be compared to assess the impact of the different mailings.
Similarly, government policies based on firm size can provide a useful identification strategy for marketing scholars. For example, requirements for firms to post calories, undertake layoffs, and provide benefits often depend on the number of employees or other measures of firm size. By comparing firms just above and just below the threshold, it is possible to assess the effect of the policies on firm behavior.
A regression discontinuity design implies that treatment is assigned depending on whether a continuous score crosses a cutoff . The analysis then focuses on whether there is a change in the outcome of interest y in the neighborhood of ([56]). In general, if a threshold is used as the source of the quasi-experiment, particular attention should be devoted to the source of the threshold and providing evidence that the threshold is essentially arbitrary and not likely to be linked to underlying discontinuities in behavior. Any discontinuity in the effect is assumed to be due to the treatment.
This assumption is not always innocuous. Consider a $50 cutoff for receiving a marketing incentive. If the firm promotes the threshold and consumers try to achieve it, then there might be a substantial difference between people who spend $49 and people who spend $51. Those who spend $49 are likely to be unresponsive to the incentive because they did not try to cross the threshold to get the incentive. In contrast, those with exactly $50 in spending might have selectively chosen to spend exactly enough to get an incentive that they planned to use. It is important to address the potential for such concerns directly.
This is reflected in a debate in economics about the effect of thresholds for low birth weight on medical outcomes. In an initial study, [ 6] used the fact that birth weight threshold of 1.5 kg is used to determine whether the newborn receives intensive medical treatment. In a critique of this work, [15] show that the children placed just at the cutoff seem to have significantly worse outcomes than babies on either side of the cutoff. This is evidence against use of this discontinuity for identification. [15] state, "This may be a signal that poor-quality hospitals have relatively high propensities to round birth weights but is also consistent with manipulation of recorded birth weights by doctors, nurses, or parents to obtain favorable treatment for their children" (p. 2119).
Once the researcher has found a regression discontinuity setting, the first step is to explore whether the discontinuity is arbitrary and linked to discontinuities in any other variables. For example, [59] examine the relationship between online reviews and advertising spending in the hotel industry. They exploit the regression discontinuity design of the rounding rule that TripAdvisor uses to convert the average ratings of reviewers into the nearest half or full star (i.e., a rating of 3.74 is shown as 3.5 stars while a rating of 3.75 shown as 4 stars), building on work by [71]. The key identification argument is that the rounding mechanism creates discrete, random variations in perceived quality around the rounding threshold and is independent of a hotel's true quality.
A threat to the arbitrary discontinuity threshold would be that hotels manipulate their average ratings around the rounding thresholds. [59] argue that if there is upward manipulation of ratings, there would be relatively few firms with average ratings just below the thresholds and a clump of firms with average ratings just above the thresholds. They show instead that the density of average ratings is uniform, with neither bumps nor dips above or below the round thresholds. They provide additional empirical evidence that characteristics of the hotels do not differ systematically above or below the threshold. Neither do they observe discontinuities in other key variables such as hotel prices and the number of five-star reviews.
The equation used for regression discontinuity can be written for panel data as
Graph
( 3)
Here is the treatment effect, the parameter of interest. represents covariates. is an indicator function that equals one when and zero otherwise. One final consideration is how to select the appropriate bandwidth for a regression discontinuity design, which is the question of how one decides on the sample to analyze, in terms of how far away the people in the sample are from the threshold where the discontinuity occurs. In general, such decisions have often been rather ad hoc, but there is an emerging literature that can help guide the researcher into thinking about how to take a more conservative approach to selecting bandwidth given the data at hand ([25]). The researcher should also ensure that their results are not sensitive to the choice of bandwidth. As with other quasi-experimental methods, the validity of the method cannot be statistically proven. Therefore, substantial emphasis must be placed on the explanation and defense of the quasi-experiment using raw data.
The quasi-experimental perspective on IVs is somewhat different from the standard treatment in econometrics textbooks, which focuses on simultaneous equations and a more structural approach. The differences relate to justification and interpretation. The quasi-experimental approach emphasizes that the shocks that move the instrument should behave as if they are an experiment. The quasi-experimental approach gives a sense of the sign, significance, and magnitude of the causal effects. The structural approach emphasizes that the shocks should be motivated by an economic model that explains the exclusion restriction. The IV approach used in structural models gives elasticities that can be used to generate counterfactuals outside of the sample. Despite these differences in interpretation, it is important to remember that the underlying mathematics is identical.
The basic idea behind using IVs is that the covariate of interest x contains both useful variation (to identify the causal effect of interest) and less useful variation (that confounds the effect). A good instrument z is strongly correlated with the useful variation but uncorrelated with the confounding variation. In other words, the researcher only uses the variation in x that can be explained by the exogenous shifter z.
The standard two-stage model involves two steps. In the first-stage regression, a fitted value of can be obtained by regressing x on instrument z and covariates :
Graph
( 4)
In the second-stage regression, the IV estimator is obtained by regressing the outcome y on the fitted value of and covariates :
Graph
( 5)
The identification of the effect of x on y relies on the following "reduced form." Inserting the predicted x to the y equation will give Equation 6. Here, is used to highlight that when regressing y directly on instrument z and covariates W, the estimated covariate coefficient is rescaled as .
Graph
( 6)
Therefore, from the quasi-experimental point of view, an instrumental variable can be seen as a treatment that affects the endogenous covariate directly. This means that directly regressing the outcome of interest on the instrument (in one stage) will get the causal effect of interest, but it will not be properly scaled. The purpose of implementing two stages is to scale the treatment effect properly. There are many ways of operationalizing instrumental variables, and this can be a place for highly technical tools. We emphasize the simplest two-stage least squares (2SLS) approach, but the intuition behind the role of instrumental variables as an identification strategy remains regardless of functional form assumptions. Using two stages enables the researcher to disentangle and . In other words, two stages are needed to get the elasticity right, but the experiment happens at the level of the instrument and so, even though the focus is on the relationship between x and y, the intuition on causality happens at the level of the relationship between z and y.
Returning to [91], while the paper adds some additional necessary nuance to the estimation to fit the particular situation, the intuition on causality measures the impact of wind (the instrument ) on social ties (the outcome of interest ). This will be . The relationship of interest, however, is the impact of posts ( ) on social ties ( ), which is measured as .
IV can be a less transparent solution to identifying causal effects compared with the other two analysis frameworks discussed previously (for a detailed discussion, see [84]). The distinction between the relationship of interest ( ) and the direct estimate from the quasi-experiment ( ) means that it is sometimes harder to visualize how the quasi-experimental variation works in IVs.
Transparent communication of IV analysis is difficult for three reasons. First, in contrast to the binary nature of the exogenous variation in DID and regression discontinuity, instruments are often continuous. This makes it more difficult to communicate the intuition for why the variation is exogenous to the potential for omitted variables or simultaneity. The ability to use continuous instruments (and multiple instruments) can also be seen as a strength of IV techniques. They enable a more flexible set of counterfactuals because there are more treatments observed and used in the analysis. For example, while a discrete quasi-experiment on retailer discounts would allow the researcher to compare the impact of a small set of retailer discounts on sales, a continuous instrument for the discounts might allow the researcher to compare a variety of smaller and larger discounts.
Second, weak instruments are a challenge. Instrumental variables techniques are consistent but biased, and this bias can matter even in seemingly large samples ([92]). Weak instruments can lead to incorrect inference in which the bias of the weak instrument dominates the potential bias of the omitted variables.
Knowing the context and the institutional setting can be invaluable in identifying strong IVs. For example, [76] derived their instruments for brand taste and price from the authors' intimate knowledge of the regulation and food industry. There are also recent advances in econometric methods that allow for more accurate presentation of statistical significance when instruments are weak ([68]). As [11] point out, many of the challenges of weak instruments are magnified when authors use multiple instruments to deal with multiple sources of endogeneity. By contrast, a focus on a single endogenous variable with a single source of endogenous variation has attractive statistical properties as well as being more transparent to the reader.
Third, many researchers present IV results with different tests and with different norms. This makes it difficult to read and assess the validity of papers with instruments.
[12], pp. 212–13) provide a sequence of steps to follow in an attempt to standardize practice. In presenting this list, we hope that it does not lead to unproductive dogmatism, and we emphasize that this is just one possible way to communicate the rationale behind a causal interpretation of the results. Still, we hope that in following these steps to the extent possible, marketing scholars can avoid being subject to many of the criticisms highlighted by [84]. The steps are as follows:
- Regress the outcome directly on the instrument. When using IV techniques, it is also desirable to show the reduced form result of regressing the outcome directly on the instrument. Because this is an ordinary least squares regression, it is unbiased. At the very least, the researcher should be confident that the instrument ( ) has the expected direct effect on the outcome ( ).
- Report the first stage. Assess whether the signs and magnitudes of the coefficients make sense.
- Report the F-statistic on the excluded instruments. This helps determine whether the instruments are weak. [92] advise that F-statistics below 10 in case of only one instrument suggest weak instruments, though, as [12], p. 213) note, "Obviously this cannot be a theorem." Similarly, [84] suggests reporting the first stage with and without the instruments to document the incremental impact of the instruments on the R-squared.
- If there are multiple instruments, report the first- and second-stage results for each instrument separately (at least in the appendix) because bias is less likely if there is only one instrument. Presenting the results separately also helps the reader understand the intuition behind the quasi-experiment underlying each instrument—whether the multiple instruments use different variation in increasing the exogenous shift in x. If there are multiple instruments, an overidentification test such as the Sargan–Hansen J can be performed to test whether all instruments are uncorrelated with the 2SLS residuals.[ 8] However, given the difficulty of identifying a robust instrument, it is unusual for researchers to have convincing cases for multiple instruments in a way that leads their regression to be overidentified. In other words, increasingly, standard practice is to focus on one instrument rather than many ([11]).
- Conduct a Hausman test comparing ordinary least squares and instrumental variables. If the results change, reflect on whether they change in a direction that makes sense given the power of the instrument. Do not interpret the results of the Hausman test to prove that the endogeneity problem is irrelevant. As noted by [84], the instrument may not be valid and therefore the test would be uninformative.
- Assess whether there is a weak instrument problem. For example, in a linear model, compare the 2SLS results with the limited information maximum likelihood results. When there is a weak instrument, the two-stage least square estimators are biased in small sample. Limited information maximum likelihood estimators have better small sample properties than 2SLS with weak instruments. If the two estimates are different, there may be a weak instrument problem. Any inconsistency from a small violation of the exclusion restriction gets magnified by weak instruments.
Regardless of which regression analysis framework to employ, presentation of baseline estimates and standard errors, along with a set of robustness checks ([59]) is standard. This typically appears in the form of a regression table with several different specifications. For example, the first column might not include any controls beyond the fixed effects, and the next set of columns might add controls. The economic magnitude of the coefficients should be discussed, both with respect to changes in the covariate of interest and relative to the range and standard deviation of the covariate and dependent variable.
A key issue in quasi-experimental analysis is correlated errors in observations, because the outcome is often observed at a finer level than the treatment. For example, the researcher might observe treatment and control groups for several advertising campaigns over a long time period. For each campaign, the researcher might have data on many individuals per campaign and many time periods per individual; however, the choices of the same individual in many time periods are likely to be correlated. [16] emphasized that failure to control for the correlation between these choices will lead to an overstatement of the effective degrees of freedom in the data, and therefore, standard errors will be biased downward. They suggest clustering standard errors by individual over time to address this issue and provide Monte Carlo evidence that clustering is likely to lead to robust inference.
Similarly, [38] emphasize that if individual responses to the same treatment are likely to be correlated, for example, because of close physical or social proximity, clustering standard errors by groups of individuals is a conservative and useful way to estimate standard errors. Researchers often need to decide on the size of the clusters. For example, in studying ready-to-eat breakfast cereals, is the correct unit the company such as General Mills, the brand such as Cheerios, or the sub-brand such as Honey Nut Cheerios? The answer depends on the data and research question. If the data are at a lower unit level (e.g., individuals) than a treatment that takes place at the firm level, cluster the standard errors at the level of the treatment. A useful perspective on this is provided by [ 2], who remind researchers that the major driver for clustering should be the experimental design rather than simple expectations of correlation. More recently, there has been evidence suggesting that it is undesirable to cluster on the variable that determines whether that observation is subject to the regression discontinuity design (e.g., age). The answer is often instead simply to reduce the bandwidth across which the regression discontinuity is studied ([65]).
Clustered standard errors rely on consistency arguments and large samples. With a small number of clusters, alternative methods are needed, such as those developed by [22], [30], and [53]. For example, [43] investigate consumers' dynamic responses to price promotions in a retail setting that involved randomly assigning ten supermarkets into varying promotion depths. Given that treatment takes place at the store level while the observation is at the consumer level, each consumer's effective contribution to reducing standard error estimates is likely to be lower than in a setting where there is no correlation across observations. However, given the relatively small number of stores/clusters available in this setting, the authors implement the wild bootstrap procedure, as proposed by [22], to correct for downward bias potentially induced in small samples. However, [24] show that even this approach requires rather large assumptions.
A more general point is that quasi-experiments range in how plausible the exogenous variation underlying the paper is, ranging from cases where the allocation is almost completely random to less clear cases where a firm or consumer assignment to treatment or control is partly random and partly an endogenous choice. Perhaps the ideal thought experiment here is [101], whose treatment and control were a pair of kidneys from the same person. [101] finds that in the United States, even identical kidneys from the same donor are received differently depending on the observed number of rejections preceding the recipient in the queue. Most research settings are less favorable. In such settings, it is often useful to combine different approaches in the same paper. For example, [79] combines a DID strategy with counterfeit entry as the treatment with a convincing and high-powered instrument on government regulation.
Still, there will be situations where a compelling exclusion restriction is lacking or the treatment–control allocation appears far from random. If the treatment and control groups are substantially different in the pretreatment or if the treatment appears to be applied based on selected characteristics, the control group is unlikely to be a good proxy for the counterfactual, and the quasi-experiment may be less likely to be valid.
We provide a discussion of three methods that are further steps researchers can take when comparability between the control and treatment groups is violated. They vary in terms of the observed and potentially unobserved differences between the control and treatment groups. Table 3 provides a summary of the frameworks and when to apply them. The table emphasizes that researchers should be cautious about applying matching methods or correction for selection bias on the grounds that there are no plausible exclusion restrictions, because these methods still require the researcher to make an argument about an exclusion restriction. The technical details of matching methods or selection bias correction are different from the three methods described previously, but the idea is similar in nature. The main goal is to bring in additional data to create control and treatment groups that are like those in quasi-experiment studies.
Graph
Table 3. Steps if Researchers Are Worried They Do Not Meet the Exclusion Restriction.
| Propensity Score Matching | Synthetic Control | Selection Bias Correction |
|---|
| Assumptions | Observable control variables are capable of identifying the selection into treatment and control conditions | The counterfactual outcome of the treatment units can be imputed in a linear combination of control units in the absence of treatment. | The unobservables that enter the treatment selection and the outcome are jointly distributed as bivariate normal. |
| Identification | The exclusion restriction can be met conditional on the variables in the match. | The exclusion restriction can be met conditional on the pretreatment outcomes. | There is at least one variable for which a compelling argument can be made for the exclusion restriction in the selection equation. |
| Settings | When matching is done to control the treatment and control pretreatment outcomes on a number of cross-sectional covariates. | When the focus is on the evolution of the outcome and the pretreatment time period has rich data on treatment and control groups. | When the allocation to the treatment condition is not fully random. |
| Caveats | Assess the degree of overlap after matching, and assess sensitivity to potential selection on unobservables. Still need to justify the exclusion restriction. | Harder to interpret the weights used to create the "synthetic control." Still need to justify the exclusion restriction. | Justification of why certain observables only affect treatment selection but not the outcome variable. Still need to justify the exclusion restriction. |
Matching methods, pioneered by [83], have been developed such that the outcomes of the treated are contrasted only against the outcomes of comparable untreated units. Many published articles in marketing have used propensity score matching when comparability between the control and treatment groups is violated. An assumption of propensity score matching is that there are observable control variables capable of identifying the selection into treatment and control conditions. This is not a trivial assumption. It suggests that propensity score matching is only good if the exclusion restriction is met conditional on the variables in the match. Any matching procedure to make the control and treatment more similar in the observables can be seen as a flexible functional form with adding "control variables" to an analysis framework. Propensity score matching requires subject-matter knowledge regarding the role of covariates in the treatment assignment decision and whether the exclusion restriction is satisfied conditional on the covariates. Therefore, we caution against applying matching methods without convincing justification of exclusion restriction.
It is difficult to identify a standard procedure for propensity score matching. We refer to [61] as a good starting point. The general objective of propensity score matching is to estimate a score such that the distribution of all the observed variables and behaviors among the treated units is similar to that among the control units. In this discussion, we consider the set of treated units to be fixed a priori. Four steps are involved in the propensity-score-matching procedure.
First, choose a functional form of the propensity score. The basic strategy uses logistic regression to model the probability of receiving the treatment given a set of observables. Second, measure the distance and apply a matching algorithm. Several possible matching methods are available including, for example, nearest-neighbor matching based on the distance in the estimated propensity score or multiple matching using all controls within some distance from the treated unit. Third, assess the degree of overlap in the distribution of the linearized propensity score after matching. Researchers typically plot and compare the histogram-based estimate of the distribution of the linearized propensity score (logarithm odds ratio) for the treatment and control groups. To inspect the match quality, it is useful to show tables on the distribution of the estimated propensity scores and the mean values of some key variables for the treated and untreated over different propensity score intervals.[ 9] Fourth and finally, calculate the average treatment effect (ATE) with the matched sample using, for example, the DID regression analysis framework discussed previously.
There are at least two caveats regarding propensity score matching. First, the model for the propensity score may be misspecified. In that case, the balance in covariates conditional on the estimated propensity score may not hold, and the credibility of subsequent inferences may be compromised. This calls for a careful discussion on the role of covariates in the treatment assignment decision. Specifically, it is important to provide a discussion of whether the covariates can be considered exogenous to the treatment. Second, regardless of the number of observed covariates used, propensity score matching does not account for the potential selection on unobservables in treatment assignment. It is important to explain why controlling for observables will address concerns with the exclusion restriction or why unobservables are not an issue in treatment assignment.
In some cases, even the closest match may not be close enough. This is particularly relevant when researchers are interested in how an event, regulatory intervention, or firm policy change affects the evolution of the outcome of interest, in contexts where only a modest number of treated units (possibly only a single one) and control units are observed for a large number of periods before and after the event. Two aspects make this setting different from the typical use of the propensity-score-matching method. First, matching is done over the pretreatment outcomes in each period rather than a number of covariates. Second, the number of control units and the number of pretreatment periods can be of similar magnitude. Synthetic controls use a different convex combination of the available control units ([ 3]; [ 4]; [39]). The intuition behind this method is that the created synthetic control unit closely represents the treated unit in all the pretreatment periods and affords time-varying causal inference on the trajectory of the outcome of interest.
Synthetic control has been used in multiple recent studies with quasi-experimental design ([ 1]). For example, [51] analyze the causal effect of industry payment disclosure on physician prescription behavior, [99] assess the impact of mobile hailing technology adoption on drivers' hourly earnings, and [78] study the causal effect of online paywalls on the sales revenues of newspapers.
Like propensity score matching, synthetic control methods are statistically rich, but they do not replace a carefully thought-out exclusion restriction and identification argument. Put differently, if propensity scores or synthetic controls appear to work when the treatment and control group are not similar, it is important to explain why controlling for observables will address issues with the exclusion restriction. In many cases, such explanations are weak and the exclusion restriction is unlikely to hold. Recent work in economics emphasizes this by showing the benefits of combining a synthetic control method with a strong exclusion restriction ([13]).
Many papers written in marketing involve a comparison of potentially different groups that reflect endogenous choices by companies or consumers where the allocation to the treatment condition is not fully random. For example, [46] assess if the introduction of the free mobile app in a business-to-business context increases sales revenues from buyers who adopted the app. In an ideal setting, the company could randomize the treatment, then observe sales from buyers who did not get the app and sales from buyers who did get it. However, this company's app was available to all buyers. Therefore, the buyers' app adoption is not random, and self-selection into the treatment (adoption) group needs to be addressed. Omitted variables that drive strategic app adoption could correlate with the sales from these buyers.
When this happens, it is sometimes useful to estimate a Heckman selection model ([57]), which explicitly models selection into the treatment as a two-step process. As [100], p. 564) pointed out, the exclusion criterion is still key to the identification of the treatment effect of interest in the two-step estimation procedure. Without the exclusion criterion, the effect of the treatment is identified only due to the nonlinearity in the functional form (specifically through the inverse Mills ratio). This may lead to severe collinearity and imprecision in the standard errors. More importantly, without a strong and credible exclusion restriction, identification in this setting is driven by the assumed functional form.
In other words, although the Heckman correction will provide an estimate without an exclusion restriction, that estimate depends entirely on the assumption that the error structure is bivariate normal. When there is an argument for the exclusion restriction, a selection model is helpful. In the absence of the exclusion restriction, even if combined with other techniques such as propensity score matching, the results would be identified off the functional form assumption alone. Put differently, if one of the covariates in the correction equation satisfies the exclusion restriction, then it is the variation in that variable that identifies the control for selection. In contrast, if the covariates in the first step are all also in the second step, then it is only the assumed error structure that identifies the control for selection.
There are both similarities and differences between selection bias correction and instrumental variable approaches. There are also similarities with the control function approach in terms of the importance of functional form assumptions on the errors in the absence of an exclusion restriction. Control functions are not part of the standard quasi-experimental toolkit, so we do not provide a detailed discussion. The selection bias correction approach uses the instrument to control for the effect of unobservables, while the instrumental variable approach attempts to eliminate the threat of endogeneity by only leveraging the useful variation created by the instrument. Yet, the two approaches share the basic idea of using an exclusion criterion (or instrument). Ultimately, both rely on the ability to find an exclusion restriction that creates useful and exogenous variation. This is why we emphasize the importance of identification in quasi-experiments and caution against blindly applying a correction for selection bias without carefully thinking about the identification assumption and providing a justification for why the exclusion restriction holds. Selection bias correction approaches are therefore only useful for causal inference in the presence of a strong credible exclusion restriction.
The specific robustness checks chosen will depend on the exact context. With electronic appendices and increasingly cheap computation, it is possible to show robustness to a large number of alternative specifications. Here, empirical work with quasi-experimental methods differs substantially from research using forecasting models. The aim is not to show one specification (or model) and defend it. Instead, the idea is to show that the sign, significance, and magnitude of the estimate of remain broadly consistent across a vast range of possible models ([59]). Often these robustness checks are dropped from the published version of the article, though they are very useful in the referee process and can end up as part of an online appendix. The following subsections describe some examples of useful robustness checks.
Compare the coefficient of interest in the models with and without controls. For example, if the coefficient changes from 2.5 to 3.5, then this change (+1.0 in this example) is informative about how big the impact of the omitted variables has to be relative to the observed controls for the omitted variables to drive the result. [ 7] provide a method to examine how much the effect of interest changes as controls are added, and then to assess how important the omitted variables would have to be for the treatment effect to disappear. The method is based on Rosenbaum bounds ([37]; [82]). It has been applied in the marketing literature by [73] and extended by [90]. Although the formal method is useful, as discussed in [77], many researchers ([ 9]; [74]) use the more basic insight that there is information in the impact of the controls on the measured effect of interest. This does not mean that results are invalid if the controls do change the estimated effect substantially, but documenting that adding seemingly relevant controls does not change the results can provide further support for the causal interpretation.
Results should not depend on arbitrary choices of functional form. For example, if using a linear probability model, show robustness to logit and probit. The choice between linear probability models and nonlinear models such as logit is widely debated. [12] argue for linear probability models because they are simple to interpret and consistent under a basic set of assumptions. Others argue against them because they are inefficient (and inconsistent if the assumptions are violated). In cases like this, where the literature does not give clear guidance on the choice of model, showing robustness to different choices is optimal.
Researchers often can choose when to start and end the sample. For example, for a treatment that occurs in 2004, researchers should be comfortable that the results are robust to the arbitrary choice of whether the period studied is 2002 to 2006, 2000 to 2008, 1995 to 2015, and so on.
There might be several different dependent variables that relate to the outcome of interest. Showing robustness to these related outcomes increases confidence in results.
Researchers choose whether all the data should be used in the control group, or only a subset of the data that is "close" to the treatment group (e.g., as measured by a propensity score). Researchers can also choose how to define the treatment group.
The idea of a placebo test is to repeat your analysis using a different part of the data set where no intervention occurred. For example, if the quasi-experimental shock happens this year, instead of comparing the difference in the outcome between last year and this year between the control and treatment groups, you can conduct a placebo test by redoing the analysis and compare the difference in the outcome between the control and treatment groups using periods with no intervention shocks. Alternatively, analysis can be conducted on an outcome that should be unrelated to the intervention being studied. The goal is to establish a null effect when there is not supposed to be one.
It is unlikely that every robustness check will yield the same level of significance or the same-sized point estimate as the initial specification. Researchers (and reviewers) should therefore not expect every specification to yield the exact same results. The key is to communicate when the results hold up. This will consequently help inform the reader what drives the statistical power behind the results.
Broadly, quasi-experimental research aspires to identify effects that do not rely on the underlying assumptions outside of the experimental variation. There are many places where that can break down, including functional form assumptions, external validity, and various confounding effects. The focus is on a robust single causal relationship.
The most effective papers typically do not stop with identifying a causal effect and its magnitude. After identifying a likely causal relationship, it is important to assess why x causes y to shift. Understanding mechanisms is often a key goal of social science. There are at least three benefits of establishing mechanisms. First, it provides a rationale for why the effect should exist in the first place. It requires the authors to think about the theoretical contribution of their research more carefully and helps make the argument for causal identification more convincing. Second, identifying mechanisms can help evaluate the benefits and negative consequences of the intervention and identify avenues for course correction, if needed. Third, understanding mechanisms allows for the possibility to extrapolate the findings to other contexts. Research needs to provide guidance on when and why the causal relationship is relevant.
When the data afford a direct measure of mediator variables, mechanisms can be inferred by mediator analysis. To illustrate how quasi-experiments can show process through mediation, we use [52] as an example. They investigate whether a variable compensation scheme increases salespeople's stress, resulting in emotional exhaustion and more sick days, and counteracts the sales benefits companies might expect from variable compensation schemes. In one of their empirical analyses, they use a natural experiment where a company dropped the variable compensation share from 80% to 20% in one of its business units. To test the health state as a possible mediator variable, they were able to measure sick days both before and after the change in the variable compensation share. In the country of study, sick days are strictly regulated by law and require certification by a physician (at the latest on the third day of the leave). Those who take more than three sick days in a given month are more likely to have substantial health problems. They measure the sick days counting after the third sick day in a month.
Combining the DID analysis with mediator analysis, [52] show that the direct effect of the treatment (drop in variable compensation share) on sales performance is significant and negative, and that the indirect effect of the treatment on sales performance via sick days is positive and significant. The mediator analysis suggests that a higher variable compensation share is associated with enhanced sales performance but also with more sick days, which, in turn, reduce the gains to sales performance.
Heterogeneous treatment effects can be used to test behavioral mechanisms. In a quasi-experimental setting, mechanism checks via heterogeneous treatment effects, sometimes referred to as falsification checks, are not simply equal to identifying moderators. They involve identifying which groups would be affected by a certain mechanism that would display the causal effect of interest, and which other groups would not display the causal effect of interest by the proposed mechanism.
Moderation analysis therefore serves a broader purpose by providing an opportunity to help explore the behavioral mechanism. If the effect goes away when theory suggests it should, then this helps identify why it happens. If the effect is larger when theory suggests it should be, then this also helps identify the mechanism. A simple approach is to estimate the effect separately by whether an individual is a member of a group that theory suggests should experience a bigger effect. Formal testing of whether the difference is statistically significant requires a three-way interaction between x, the source of variation, and group membership.
There are many relevant examples in marketing of the use of moderation analyses to demonstrate a mechanism if there is a reason to believe the boundary of underlying process exists or the magnitude of the treatment effect varies by some observables. For example, after showing the European privacy regulation hurt online advertising, [47] ran a falsification check demonstrating that European consumers behaved like Americans when visiting American websites and that American consumers behaved like Europeans when visiting European websites. The paper then explored the mechanism and showed that the regulation especially hurt unobtrusive advertising and advertising on general interest websites, two situations where using data to target advertising is particularly valuable.
Overall, mechanism checks through mediator or moderation analyses are important because they distinguish the goal of the marketing scholar from the marketing practitioner. Marketing practitioners run experiments and analyze data to understand what they should do in the particular situation they are facing. Marketing scholars need to have a broader sense of applicability beyond the specific setting being studied. Mediation and moderation analyses provide an understanding of when a marketing action will and will not lead to the desired behavior. For this reason, marketing papers are more likely to be remembered for the evidence that is shown in support of a theory explaining why the result holds.
The external validity discussion in a paper should recognize the assumptions required for the analysis to capture the ATE across the population of interest, rather than a more local effect that is an artifact of the data sample or the source of quasi-experimental variation. A key concept is the ATE across the entire population. This is the difference in outcomes that would occur by moving the entire population from the control group to the treatment group. However, in some cases, the ATE may not be particularly relevant, because it averages across the entire population and includes units that would never be eligible for treatment ([100], p. 604). For example, we would not want to include millionaires in computing the ATE of a job training program. To address this, the researcher could use the average treatment effect on the treated, which measures the expected effect of treatment for those who actually were in the treatment condition.
One reason why a research setting may fail to be externally valid is if the treated population is unrepresentative ([72]). A concern that will drive whether the treated population is unrepresentative is whether those affected could self-select into and out of the treatment. For example, [28] study a rule change by Google that allowed non–trademark holders to use trademarks in search advertising copy. They study the rule change's effect on user click behavior. In this case, many advertisers did not alter their advertising copy strategy, for a variety of reasons. These advertisers may be systematically different from the advertisers that did change their strategy. Because these advertisers were not forced to change their strategy, we will never know what would have happened if they did. When faced with such issues, it is best to spell out the potential for self-selection and discuss whether it makes the paper more or less relevant. In this case, it would be accurate to say that the researchers captured the effect of a loosening of trademark restrictions, because it is unlikely that a search engine would force its advertisers into using other advertisers' trademarks. However, it would not be accurate to claim that the researchers capture the broader effect of all advertisers using other advertisers' trademarks in their copy.
The treated population may also be unrepresentative if the treatment impacts a subpopulation to change behavior, but not the main population of interest. This means that the measured effect is localized to that subpopulation, and it is referred to in the literature as the local average treatment effect (LATE). For example, in the context of regression discontinuity, the LATE is the average of the treatment effect over the individuals who would have been in the counterfactual condition if the discontinuity threshold were changed. A limitation of regression discontinuity is that the results directly apply only to populations around the threshold. For example, comparing the $49 spend with the $51 spend may be informative about the impact of the marketing incentive on consumers who spend around $50; however, consumers who typically spend a lot more or a lot less might be different. The idea of LATE also has implications for the interpretation of instrumental variables estimates, as any IV estimate is the LATE for the observations in the regression who experienced the kind of variation exploited by the instrument.[10]
More broadly, as with other aspects of quasi-experimental research, the best practice regarding the external validity of results is to clearly lay out the assumptions and limitations. For example, [94] use a quasi-experiment and DID to examine the impact of advertising revenue on the type of content posted on Chinese blogs. While it might be tempting to interpret the results as suggestive of a broader impact of commercial interests on media, they are careful to emphasize the many differences between blogs and other media, between China and the rest of the world, and between the way the bloggers were compensated and other online advertising models. In this way, Sun and Zhu's article explicitly limits the temptation of the reader to extrapolate too much.
An internally valid quasi-experimental estimate can have broader external validity when used to identify relationships such as elasticities and then to use a structural model to identify the counterfactual of interest. In these cases, under the assumption that the model is a useful representation of reality, quasi-experimental methods serve as a complement for, rather than a substitute to, structure. For example, [ 9] use quasi-experimental methods to identify the impact of the automotive brand preferences of parents on the brand preferences of their children. They then use structural methods to estimate the implications for firm strategy. [42] use quasi-experimental variation in health insurance prices to identify price elasticity and then combine this measure with a structural model to estimate the welfare implications of adverse selection. [29] use quasi-experimental variation around set quotas to identify the relationship between commissions and sales, and then use this variation in a structural model to determine optimal compensation schemes.
Overall, effective quasi-experimental research requires an understanding of the underlying assumptions behind any broad interpretation of quasi-experimental results. Quasi-experiments often require a focus on a narrow slice of the data, and therefore, it is important to consider the degree to which the results apply to a broader population.
Any identification strategy relies on a set of assumptions. These assumptions need to be explicit throughout the paper. There are always some tests that cannot be run, for example, due to lack of data. There are always some robustness checks that are weaker than others. There are always some steps from data to interpretation. While apologies do not mean all is forgiven, the objective should be to clarify the boundaries of the claims. Obfuscation is much worse than a clear summary of the identifying assumptions.
As an example, [51] employ a DID research design to study the effect of the payment disclosure law introduced in Massachusetts in June 2009. The research design uses the setting that physicians located in the border counties of Massachusetts and its neighboring states did not have disclosure laws during this period. They lay out the assumptions underlying their estimation:
Our identification of the effect of disclosure legislation relies on the change in new prescriptions by physicians located in Massachusetts (MA) after the policy intervention, relative to their counterparts from "control" states in which no such law existed in the same period.... To assess potential threats to the validity of our research design, we verify if the result was driven by changes in physician payments as a result of the MA disclosure law. If such payment changes were primarily driven by local pharmaceutical reps reallocating their marketing budgets across physicians operating on either side of state borders, this would render the border identification strategy problematic.
([51], p. 517)
This example communicates three distinct points. First, it explains the identification strategy. Second, it details the main threats to the validity of this identification strategy. Third, it describes what they do to address it. These points suggest that effective apologies focus on demonstrating what interpretations are reasonable, and what might be a stretch of the results. The goal is not to show that in all circumstances and every conceivable way the identification is perfect. That is not possible. Instead, the goal is to provide clear bounds on the interpretation. The paper's contribution is then a function of whether it provides new knowledge under this bounded interpretation.
Quasi-experimental techniques are an important tool for marketers. First, marketing scholars need to be able to inform marketing practitioners—both managers and policy makers—about the causal effect to allow practitioners to make superior decisions. Second, the best quasi-experimental papers do not simply prove a causal effect but delve into the underlying mechanism, which is key to marketing scholarship's goal of generalizability. Third, such techniques become more important as the scope and span of marketing practice expands and there are new settings and more varied sources of data that allow their application.
The objective of a quasi-experimental research paper is to answer an interesting and important research question about a causal relationship and provide evidence suggesting the mechanism behind the relationship. The choice of method (DID, regression discontinuity, or instrumental variables) depends on the nature of the quasi-experiment. The framework we present focuses on understanding how exogenous variation helps uncover causal relationships and why specific actions affect behavior. Of course the details of the methods will evolve over time as new research appears. Because marketing scholars are often interested in providing generalizable insights about how marketing actions change the behavior of individual consumers, the quasi-experimental framework is particularly useful. Similarly, firms that want to use those insights benefit. As the availability of detailed data grows and marketing technology changes, these methods will enable marketing scholars to provide assessments of a wide variety of situations in which a particular marketing action is likely to change consumer behavior or market dynamics.
Footnotes 1 Christine Moorman and Harald van Heerde
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Catherine Tucker https://orcid.org/0000-0002-1847-4832
5 In describing these tools and their motivation, we build on numerous books and articles that have covered similar material for economics, policy, and sociology audiences ([12]; [31]; [62]; [75]).
6 For example, the fixed effects specification of Stata's xtreg function uses differences from average values. The fixed effects specifications of Stata's xtlogit and xtpoisson also condition out the individual-level fixed effects.
7 Detailed implementation steps are provided in fuzzydidi and did_multiplegt in STATA and the did package in R.
8 For example, as of 2022, IV estimation can be produced by ivreg2 in STATA. Under i.i.d. error assumption, the command estatoverid provides the Sargan test. When the estimation is done with generalized method of moments (i.e. gmm2s is specified in ivreg2), the test of overidentifying restrictions becomes the Hansen J statistic.
9 The propensity score matching algorithm can be found in multiple statistical packages as of 2022, for example, the PSMATCH2 module in Stata.
Recent work has shown subtleties in interpreting IV results as LATE ([18]).
References Abadie Alberto. (2021), " Using Synthetic Controls: Feasibility, Data Requirements, and Methodological Aspects ," Journal of Economic Literature , 59 (2), 391 – 425.
Abadie Alberto , Athey Susan , Imbens Guido W. , Wooldridge Jeffrey. (2017), " When Should You Adjust Standard Errors for Clustering?" Working Paper 24003, National Bureau of Economic Research, https://www.nber.org/papers/w24003.
Abadie Alberto , Diamond Alexis , Hainmueller Jens. (2010), " Synthetic Control Methods for Comparative Case Studies: Estimating the Effect of California's Tobacco Control Program ," Journal of the American Statistical Association , 105 (490), 493 – 505.
Abadie Alberto , Diamond Alexis , Hainmueller Jens. (2015), " Comparative Politics and the Synthetic Control Method ," American Journal of Political Science , 59 (2), 495 – 510.
Akca Selin , Rao Anita. (2020), " Value of Aggregators ," Marketing Science , 39 (5), 893 – 922.
Almond Douglas , Doyle Joseph , Kowalski Amanda , Williams Heidi. (2010), " Estimating Marginal Returns to Medical Care: Evidence from At-Risk Newborns ," Quarterly Journal of Economics , 125 (2), 591 – 634.
Altonji Joseph , Elder Todd , Taber Christopher. (2005), " Selection on Observed and Unobserved Variables: Assessing the Effectiveness of Catholic Schools ," Journal of Political Economy , 113 (1) , 151 – 84.
Anderson Eric T. , Fong Nathan M. , Simester Duncan I. , Tucker Catherine E.. (2010), " How Sales Taxes Affect Customer and Firm Behavior: The Role of Search on the Internet ," Journal of Marketing Research , 47 (2), 229 – 39.
Anderson Soren T. , Kellogg Ryan , Langer Ashley , Sallee James M.. (2015), " The Intergenerational Transmission of Automobile Brand Preferences ," Journal of Industrial Economics , 63 (4), 763 – 93.
Angrist Joshua D. , Imbens Guido W. , Rubin Donald B.. (1996), " Identification of Causal Effects Using Instrumental Variables ," Journal of the American Statistical Association , 91 (434) , 444 – 55.
Angrist Joshua D. , Kolesár Michal. (2021), " One Instrument to Rule Them All: The Bias and Coverage of Just-ID IV ," Working Paper 29417, National Bureau of Economic Research, https://www.nber.org/papers/w29417.
Angrist Joshua D. , Pischke Jörn-Steffen. (2009), Mostly Harmless Econometrics: An Empiricist's Companion. Princeton, NJ: Princeton University Press.
Arkhangelsky Dmitry , Athey Susan , Hirshberg David , Imbens Guido W. , Wager Stefan. (2021), " Synthetic Difference-in-Differences ," American Economic Review , 111 (12), 4088 – 4118.
Athey Susan , Imbens Guido W.. (2022), " Design-Based Analysis in Difference-in-Differences Settings with Staggered Adoption ," Journal of Econometrics , 226 (1), 62 – 79.
Barreca Alan I. , Guldi Melanie , Lindo Jason M. , Waddell Glen R.. (2011), " Saving Babies? Revisiting the Effect of Very Low Birth Weight Classification ," Quarterly Journal of Economics , 126 (4), 2117 – 23.
Bertrand Marianne , Duflo Esther , Mullainathan Sendhil. (2004), " How Much Should We Trust Differences-in-Differences Estimates? " Quarterly Journal of Economics , 119 (1), 249 – 75.
Blake Thomas , Nosko Chris , Tadelis Steven. (2015), " Consumer Heterogeneity and Paid Search Effectiveness: A Large-Scale Field Experiment ," Econometrica , 83 (1), 155 – 74.
Blandhol Christine , Bonney John , Mogstad Magne , Torgovitsky Alexander. (2022), " When Is TSLS Actually LATE?" Technical Report 29709, National Bureau of Economic Research.
Borusyak Kirill , Hull Peter , Jaravel Xavier. (2022), " Quasi-Experimental Shift-Share Research Designs ," Review of Economics Studies , 89 (1), 181 – 213.
Bronnenberg Bart , Dubé Jean-Pierre , Gentzkow Matthew. (2012), " The Evolution of Brand Preferences: Evidence from Consumer Migration ," American Economic Review , 102 (6), 2472 – 2508.
Callaway Brantly , Sant'Anna Pedro H.C.. (2020), " Difference-in-Differences with Multiple Time Periods ," Journal of Econometrics , 225 (2), 200 – 30.
Cameron Colin , Gelback Jonah , Miller Douglas. (2008), " Bootstrap-Based Improvements for Inference with Clustered Errors ," Review of Economics and Statistics , 90 (3), 414 – 27.
Campbell Roy H.. (1965), " A Managerial Approach to Advertising Measurement ," Journal of Marketing , 29 (4), 1 – 6.
Canay Ivan A. , Santos Andres , Shaikh Azeem M.. (2021), " The Wild Bootstrap with a "Small" Number of "Large" Clusters ," Review of Economics and Statistics , 103 (2), 346 – 63.
Cattaneo Matias D. , Titiunik Rocío , Vazquez-Bare Gonzalo. (2020), " The Regression Discontinuity Design ," in Handbook of Research Methods in Political Science and International Relations , Luigi Curini and Robert Franzese, eds. SAGE Publications , 835 – 57.
Chandy Rajesh K. , Johar Gita Venkataramani , Moorman Christine , Roberts John H.. (2021), " Better Marketing for a Better World ," Journal of Marketing , 85 (3), 1 – 9.
Chevalier Judith , Mayzlin Dina. (2006), " The Effect of Word of Mouth Online: Online Book Reviews ," Journal of Marketing Research , 43 (3), 345 – 54.
Chiou Lesley , Tucker Catherine E.. (2012), " How Does the Use of Trademarks by Third-Party Sellers Affect Online Search? " Marketing Science , 31 (5), 819 – 37.
Chung Doug J. , Steenburgh Thomas , Sudhir K.. (2014), " Do Bonuses Enhance Sales Productivity? A Dynamic Structural Analysis of Bonus-Based Compensation Plans ," Marketing Science , 33 (2), 165 – 87.
Conley Timothy G. , Taber Christopher R.. (2011), " Inference with 'Difference in Differences' with a Small Number of Policy Changes ," Review of Economics and Statistics , 93 (1), 113 – 25.
Cook Thomas D. , Campbell Donald T.. (1979), Quasi-Experimentation: Design & Analysis Issues for Field Settings. Boston: Houghton-Mifflin.
Crayton Ancil. (2020), "Causal Inference for Data Scientists: Econometrics to Machine Learning," (accessed January 26, 2022), https://www.ancilcrayton.com/talk/ct-2020/.
Datta Hannes , Knox George , Bronnenberg Bart. (2018), " Changing Their Tune: How Consumers' Adoption of Online Streaming Affects Music Consumption and Discovery ," Marketing Science , 37 (1), 5 – 21.
Deaton Angus S.. (2009), " Instruments of Development: Randomization in the Tropics, and the Search for the Elusive Keys to Economic Development ," Proceedings of the British Academy , 162 , 123 – 60.
De Chaisemartin Clément , d'Haultfoeuille Xavier. (2020), " Two-Way Fixed Effects Estimators with Heterogeneous Treatment Effects ," American Economic Review , 110 (9), 2964 – 96.
Deighton John A. , Mela Carl F. , Moorman Christine. (2021), " Marketing Thinking and Doing ," Journal of Marketing , 85 (1), 1 – 6.
DiPrete Thomas A. , Gangl Markus. (2004), " Assessing Bias in the Estimation of Causal Effects: Rosenbaum Bounds on Matching Estimators and Instrumental Variables Estimation with Imperfect Instruments ," Sociological Methodology , 34 (1), 271 – 310.
Donald Stephen , Lang Kevin. (2007), " Inference with Difference in Differences and Other Panel Data ," Review of Economics and Statistics , 89 (2), 221 – 33.
Doudchenko Nikolay , Imbens Guido W.. (2016), " Balancing, Regression, Difference-in-Differences and Synthetic Control Methods: A Synthesis ," Working Paper 22791, National Bureau of Economic Research, https://www.nber.org/papers/w22791.
Dubé Jean-Pierre , Hitsch Günter J. , Rossi Peter E.. (2018), " Income and Wealth Effects on Private-Label Demand: Evidence from the Great Recession ," Marketing Science , 37 (1), 22 – 53.
Ebbes Peter , Papies Dominik , Van Heerde Harald J.. (2011), " The Sense and Non-Sense of Holdout Sample Validation in the Presence of Endogeneity ," Marketing Science , 30 (6), 1115 – 22.
Einav Liran , Finkelstein Amy , Cullen Mark. (2010), " Estimating Welfare in Insurance Markets Using Variation in Prices ," Quarterly Journal of Economics , 125 (3), 877 – 921.
Elberg Andres , Gardete Pedro M. , Macera Rosario , Noton Carlos. (2019), " Dynamic Effects of Price Promotions: Field Evidence, Consumer Search, and Supply-Side Implications ," Quantitative Marketing and Economics , 17 (1), 1 – 58.
Gelman Andrew. (2010), " Experimental Reasoning in Social Science ," in Field Experiments and Their Critics , Chap. 7. New Haven, CT : Yale University Press , 185 – 95.
Gibbons Charles E. , Serrato Juan Carlos Suarez , Urbancic Michael B.. (2017), " Broken or Fixed Effects? " Journal of Econometric Methods , 8 (1), 20170002.
Gill Manpreet , Sridhar Shrihari , Grewal Rajdeep. (2017), " Return on Engagement Initiatives: A Study of a Business-to-Business Mobile App ," Journal of Marketing , 81 (4), 45 – 66.
Goldfarb Avi , Tucker Catherine E. (2011a), " Privacy Regulation and Online Advertising ," Management Science , 57 (1), 57 – 71.
Goldfarb Avi , Tucker Catherine E. (2011b), " Search Engine Advertising: Channel Substitution When Pricing Ads to Context ," Management Science , 57 (3), 458 – 70.
Goodman-Bacon Andrew. (2021), " Difference-in-Difference with Variations in Treatment Timing ," Journal of Econometrics , 225 (2), 247 – 77.
Grewal Rajdeep. (2017), " Journal of Marketing Research: Looking Forward ," Journal of Marketing Research , 54 (1), 1 – 4.
Guo Tong , Sriram Srinivasaraghavan , Manchanda Puneet. (2020), " Let the Sunshine In: The Impact of Industry Payment Disclosure on Physician Prescription Behavior ," Marketing Science , 39 (3), 516 – 39.
Habel Johannes , Alavi Sascha , Linsenmayer Kim. (2021), " Variable Compensation and Salesperson Health ," Journal of Marketing , 85 (3), 130 – 49.
Hagemann Andreas. (2019), " Placebo Inference on Treatment Effects when the Number of Clusters Is Small ," Journal of Econometrics , 213 (1), 190 – 209.
Hahn Jinyong , Todd Petra , Van der Klaauw Wilbert. (2001), " Identification and Estimation of Treatment Effects with a Regression-Discontinuity Design ," Econometrica , 69 (1), 201 – 9.
Hartmann Wesley , Klapper Daniel. (2018), " Super Bowl Ads ," Marketing Science , 37 (1), 78 – 96.
Hartmann Wesley , Nair Harikesh S. , Narayanan Sridhar. (2011), " Identifying Causal Marketing Mix Effects Using a Regression Discontinuity Design ," Marketing Science , 30 (6), 1079 – 97.
Heckman James J. (1978), " Dummy Endogenous Variables in a Simultaneous Equation System ," Econometrica , 46 (7/4), 931 – 59.
Heckman James J. (2000), " Causal Parameters and Policy Analysis in Economics: A Twentieth Century Retrospective ," Quarterly Journal of Economics , 115 (1) , 47 – 97.
Hollenbeck Brett , Moorthy Sridhar , Proserpio Davide. (2019), " Advertising Strategy in the Presence of Reviews: An Empirical Analysis ," Marketing Science , 38 (5), 793 – 811.
Imbens Guido W. , Lemieux Thomas. (2008), " Regression Discontinuity Designs: A Guide to Practice ," Journal of Econometrics , 142 , 615 – 35.
Imbens Guido W. , Rubin Donald B.. (2015), Causal Inference for Statistics, Social, and Biomedical Sciences: An Introduction. Cambridge, UK: Cambridge University Press.
Imbens Guido W. , Wooldridge Jeffrey. (2009), " Recent Developments in the Econometrics of Program Evaluation ," Journal of Economic Literature , 47 (1), 5 – 86.
Israeli Ayelet. (2018), " Online MAP Enforcement: Evidence from a Quasi-Experiment ," Marketing Science , 37 (5), 710 – 32.
Janakiraman Ramkumar , Lim Joon Ho , Rishika Rishika. (2018), " The Effect of a Data Breach Announcement on Customer Behavior: Evidence from a Multichannel Retailer ," Journal of Marketing , 82 (2), 85 – 105.
Kolesár Michal , Rothe Christoph. (2018), " Inference in Regression Discontinuity Designs with a Discrete Running Variable ," American Economic Review , 108 (8), 2277 – 2304.
Lambrecht Anja , Seim Katja , Tucker Catherine E.. (2011), " Stuck in the Adoption Funnel: The Effect of Interruptions in the Adoption Process on Usage ," Marketing Science , 30 (2), 355 – 67.
Lancaster Tony. (2000), " The Incidental Parameter Problem since 1948 ," Journal of Econometrics , 95 (2), 391 – 413.
Lee David S. , McCrary Justin , Moreira Marcelo J. , Porter Jack R.. (2021), Valid T-Ratio Inference for IV. Cambridge, MA : National Bureau of Economic Research.
Lim Joon Ho , Rishika Rishika , Janakiraman Ramkumar , Kannan P.K.. (2020), " Competitive Effects of Front-of-Package Nutrition Labeling Adoption on Nutritional Quality: Evidence from Facts Up Front–Style Labels ," Journal of Marketing , 84 (6), 3 – 21.
List John A.. (2011), " Why Economists Should Conduct Field Experiments and 14 Tips for Pulling One Off ," Journal of Economic Perspectives , 25 (3), 3 – 16.
Luca Michael. (2011), " Reviews, Reputation, and Revenue: The Case of Yelp.com," Harvard Business School Working Papers 12-016, Harvard Business School.
Lynch John G.. (1982), " On the External Validity of Experiments in Consumer Research ," Journal of Consumer Research , 9 (3), 225 – 39.
Manchanda Puneet , Packard Grant , Pattabhiramaiah Adithya. (2015), " Social Dollars: The Economic Impact of Customer Participation in a Firm-Sponsored Online Customer Community ," Marketing Science , 34 (3), 367 – 87.
Mayzlin Dina , Dover Yaniv , Chevalier Judith. (2014), " Promotional Reviews: An Empirical Investigation of Online Review Manipulation ," American Economic Review , 104 (8), 2421 – 55.
Meyer Bruce. (1995), " Natural and Quasi-Experiments in Economics ," Journal of Business and Economic Statistics , 12 (2), 151 – 62.
Moorman Christine , Ferraro Rosellina , Huber Joel. (2012), " Unintended Nutrition Consequences: Firm Responses to the Nutrition Labeling and Education Act ," Marketing Science , 31 (5), 717 – 37.
Oster Emily. (2019), " Unobservable Selection and Coefficient Stability: Theory and Evidence ," Journal of Business and Economic Statistics , 37 (2), 187 – 204.
Pattabhiramaiah Adithya , Sriram S. , Manchanda Puneet. (2019), " Paywalls: Monetizing Online Content ," Journal of Marketing , 83 (2), 19 – 36.
Qian Yi. (2008), " Impacts of Entry by Counterfeiters ," Quarterly Journal of Economics , 123 (4) , 1577 –1 609.
Ramani Nandini , Srinivasan Raji. (2019), " Effects of Liberalization on Incumbent Firms' Marketing-Mix Responses and Performance: Evidence from a Quasi-Experiment ," Journal of Marketing , 83 (5), 97 – 114.
Rebecq Antoine. (2020), "Causal Inference Cheat Sheet for Data Scientists," Towards Data Science (April 29), https://towardsdatascience.com/causal-inference-cheat-sheet-for-data-scientists-a1d97b98d515.
Rosenbaum Paul R.. (2002), Observational Studies , 2nd ed. New York : Springer.
Rosenbaum Paul R. , Rubin Donald B.. (1983), " The Central Role of the Propensity Score in Observational Studies for Causal Effects ," Biometrika , 71 (1), 41 – 55.
Rossi Peter E.. (2014), " Even the Rich Can Make Themselves Poor: A Critical Examination of IV Methods in Marketing Applications ," Marketing Science , 33 (5), 621 – 762.
Roth Jonathan , Sant'Anna Pedro H.C. , Bilinski Alyssa , Poe John. (2022), "What's Trending in Difference-in-Differences? A Synthesis of the Recent Econometrics Literature," arXiv preprint arXiv:2201.01194.
Rubin Donald B.. (2005), " Causal Inference Using Potential Outcomes ," Journal of the American Statistical Association , 100 (469), 322 – 31.
Schmitt Bernd H. , Cotte June , Giesler Markus , Stephen Andrew T. , Wood Stacy. (2021), " Our Journal, Our Intellectual Home ," Journal of Consumer Research , 47 (5), 633 –3 5.
Shadish William R. , Cook Thomas D. , Campbell Donald T.. (2002), Experimental and Quasi-Experimental Designs for Generalized Causal Inference. Boston : Houghton-Mifflin Company.
Shapiro Bradley T.. (2020), " Advertising in Health Insurance Markets ," Marketing Science , 39 (3), 587 – 611.
Shin Jiwoong , Sudhir K. , Yoon Dae-Hee. (2012), " When to Fire Customers: Customer Cost-Based Pricing ," Management Science , 58 (5), 932 – 47.
Shriver Scott K. , Nair Harikesh , Hofstetter Reto. (2013), " Social Ties and User Generated Content: Evidence from an Online Social Network ," Management Science , 59 (6), 1425 – 43.
Stock James , Wright Jonathan , Yogo Motohiro. (2002), " A Survey of Weak Instruments and Weak Identification in Generalized Method of Moments ," Journal of Business and Economic Statistics , 20 (4), 518 – 29.
Sun Liyang , Abraham Sarah. (2021), " Estimating Dynamic Treatment Effects in Event Studies with Heterogeneous Treatment Effects ," Journal of Econometrics , 225 (2), 175 – 99.
Sun Monic , Zhu Feng. (2013), " Ad Revenue and Content Commercialization: Evidence from Blogs ," Management Science , 59 (10), 2314 – 31.
Thomas Michael. (2020), " Spillovers from Mass Advertising: An Identification Strategy ," Marketing Science , 39 (4), 807 – 26.
Toubia Olivier. (2022), " Editorial: A New Chapter or a New Page for Marketing Science? " Marketing Science , 41 (1) , 1 – 6.
Tucker Catherine E. , Zhang Juanjuan , Zhu Ting. (2013), " Days on Market and Home Sales ," RAND Journal of Economics , 44 (2), 337 – 60.
Van Heerde Harald J. , Moorman Christine , Page Moreau C. , Palmatier Robert W.. (2021), " Reality Check: Infusing Ecological Value into Academic Marketing Research ," Journal of Marketing , 85 (2), 1 – 13.
Wang Yanwen , Lewis Michael , Schweidel David. (2018), " A Border Strategy Analysis of Ad Source and Message Tone in Senatorial Campaigns ," Marketing Science , 37 (3), 333 – 55.
Wang Yanwen , Wu Chunhua , Zhu Ting. (2019), " Mobile Hailing Technology Value and Taxi Driving Behaviors ," Marketing Science , 38 (5), 733 – 912.
Wooldridge Jeffrey. (2000), Econometric Analysis of Cross Section and Panel Data. Cambridge, MA: MIT Press.
Zhang Juanjuan. (2010), " The Sound of Silence: Observational Learning in the U.S. Kidney Market ," Marketing Science , 29 (2), 315 – 35.
~~~~~~~~
By Avi Goldfarb; Catherine Tucker and Yanwen Wang
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 38- Connecting to Place, People, and Past: How Products Make Us Feel Grounded. By: Eichinger, Isabel; Schreier, Martin; van Osselaer, Stijn M.J. Journal of Marketing. Jul2022, Vol. 86 Issue 4, p1-16. 16p. 1 Diagram, 1 Chart. DOI: 10.1177/00222429211027469.
- Database:
- Business Source Complete
Connecting to Place, People, and Past: How Products Make Us Feel Grounded
Consumption can provide a feeling of groundedness or being emotionally rooted. This can occur when products connect consumers to their physical (place), social (people), and historic (past) environment. The authors introduce the concept of groundedness to the literature and show that it increases consumer choice; happiness; and feelings of safety, strength, and stability. Following these consequential outcomes, the authors demonstrate how marketers can provide consumers with a feeling of groundedness through product designs, distribution channels, and marketing communications. They also show how marketers might segment the market using observable proxies for consumers' need for groundedness, such as high computer use, high socioeconomic status, or life changes brought on by the COVID-19 pandemic. Taken together, the findings show that groundedness is a powerful concept providing a comprehensive explanation for a variety of consumer trends, including the popularity of local, artisanal, and nostalgic products. It seems that in times of digitization, urbanization, and global challenges, the need to feel grounded has become particularly acute.
Keywords: connectedness; alienation; need to belong; groundedness; local; rootedness; terroir; traditional
To be rooted is perhaps the most important and least recognized need of the human soul.
—[47], p. 43).
Dual forces of digitization and globalization have made our social and work lives become increasingly virtual, fast-paced, and mobile, leaving many consumers feeling like trees with weak roots, at risk of being torn from the earth. In response, we observe consumers trying to (re)connect to place, people, and past—to get anchored. Against this backdrop, we propose and test an important driver of consumer behavior that has largely been overlooked in marketing literature: the feeling of groundedness.
We believe that many consumers have a need to feel grounded, which we define as a feeling of emotional rootedness. This feeling emanates from connections to one's physical, social, and historic environment and provides a sense of strength, safety, and stability. Although the concept has received scant attention in prior marketing, consumer behavior, and social psychology research, the feeling of groundedness appears to be a familiar one among lay consumers. For example, we might feel grounded when returning to our birthplace, sitting at our grandparents' kitchen table while enjoying a pie made with apples from their backyard tree and according to a recipe passed down for generations. Similarly, we may have experienced feeling grounded when shopping at the local farmers market or foraging a basket of mushrooms from a nearby forest.
We argue that there are at least three conceptually separable (but in practice often intertwined) sources of feelings of groundedness: connectedness to place, people, and/or past. Collectively, connections to place, people, and past engender feelings of groundedness by "rooting" us in our physical, social, and historic sphere. These connections may be established through many different objects, activities, and types of interactions. In this article, we focus on the role of products in providing customers with a connection to place, people, and past.
Indeed, numerous marketplace examples illustrate increasing consumer demand for products that presumably make them feel more connected and thus grounded: Spearheading a renaissance of artisan, indie, and craft production, for example, locally rooted (micro)breweries have gained substantial market share in recent years. In 2019, craft beer accounted for 13.6% of total beer volume sales—a number that had increased by 4% even as overall U.S. beer volume sales had decreased by 2% ([ 6]). Similarly, sales estimates of local food increased from US$6.1 billion in 2012 to US$8.7 billion in 2015 ([24]; [45]) and farmers markets—which afford a connection to the land and to the people behind the food—are on the rise. In 2014, there were 8,268 farmers markets across the United States: a growth of 180% since 2006 ([24]). Beyond the food industry, online marketplaces such as Etsy connect consumers to handcrafted products and to the craftspeople that sell them. Impressively, Etsy reported 81.9 million users and US$10.3 billion gross merchandise sales worldwide in 2020 ([10]).
This trend in demand for local, personal, and traditional products is surprising when considered against the backdrop of globalization, digitization, and modern society's penchant for technology and innovation. Marketers have begun to capitalize on these shifts in demand—for example, by stocking and promoting local products, encouraging contact with the people who make the products, and highlighting traditional ingredients or production methods. We have recently also observed marketers referring to the concept of groundedness. The Austrian grocery chain BILLA ran a national advertising campaign in fall 2020 referring to the farmers behind their products as "The people who make us grounded" ("Wer uns erdet").
In light of these trends, we contend that products can metaphorically connect us to place, people, and past, and thereby make us feel grounded. For brevity, we hereinafter refer to products that can make consumers feel grounded as "grounding products." We argue that the ability of products to provide a feeling of groundedness will make them more attractive to consumers. We further propose that feeling grounded may contribute to consumer well-being. Groundedness—understood as a feeling of deep-rootedness, having a strong foundation, and being securely anchored—gives consumers feelings of safety, strength, and stability as well as confidence that they can withstand adversity. As such, feelings of groundedness might provide consumers with a sense of happiness, thus adding to their overall well-being.
This work makes several contributions. First, it introduces the feeling of groundedness as a driver of consumer behavior and consumer welfare. Second, it provides an overarching theoretical explanation for a variety of major consumer trends, such as the desire for local, craft, and traditional products. Third, it highlights that consumers experience a feeling of groundedness when products connect consumers to their physical (place), social (people), and historic (past) environment. Fourth, the studies offer various actionable marketing implications for products aimed at helping consumers connect to place, people, and past.
As a personal characteristic, to be "grounded" is a common concept in everyday parlance, easily found in any dictionary. In contrast to everyday parlance, we found groundedness to be a fairly novel and underresearched construct in the literature. There are few direct references to groundedness in the marketing, consumer behavior, or social psychology literature streams. The mentions we did find in other literature (e.g., psychotherapy, environmental or educational psychology) are relatively obscure, only loosely related, or speculative (for an overview of relevant research, see Web Appendix A). For example, educational psychologist [29] writes about "rootedness" and develops a measure of rootedness for college students. However, McAndrew's explanation of rootedness is limited to location. Similarly, environmental psychologists (e.g., [27]; [33]) have studied connectedness to nature, which is also a more limited construct. We found a more closely related conception of groundedness in a psychotherapy doctoral dissertation, where [31], pp. 82–83) describes rootedness in terms of "the personal, social, environmental, and economic anchoring that sees us through tough times. Within rootedness, there is a sense of togetherness, a combination of personal identity and group identity, past and present, and people and places."
In philosophy, [47], p. 43) points to the importance of being rooted. She notes:
A human being has roots by virtue of his real, active, and natural participation in the life of a community, which preserves in living shape certain particular treasures of the past and certain particular expectations of the future. This participation is a natural one, in the sense that it is automatically brought about by place, conditions of birth, profession, and social surroundings.
[12] likewise writes about rootedness in terms of the need to establish roots and feel at home in the world, while [41] refers to a connection to the land as a source of well-being that is undermined by technological forces that separate people from their roots in nature.
In marketing, [42] examine rootedness in the context of community-supported agriculture (CSA), arguing that by connecting consumers to the land and producers, CSA membership may help consumers reconnect to their "material, historical, and spiritual roots" (p. 141). [ 2] also touch on some of the elements, antecedents, and consequences of groundedness, such as community and traditions.
In summary, we believe the idea of groundedness has not been formally developed as a concept, nor have the full scope of the construct and its implications for consumer behavior and marketing been identified. We aim to fill this gap in the literature.
We argue that many consumers have a need to feel grounded, which we define as a feeling of emotional rootedness. The feeling of groundedness results from being metaphorically embedded in one's physical, social, and historical environment. Like the roots of a tree or the foundation of a house, a feeling of groundedness connects a person to their "terroir" (where the French word terroir not only refers to the land per se but also includes its cultural history and human capital [[35]]). Consistent with relevant dictionary definitions—which include being mentally and emotionally stable or firmly established[ 5]—we argue that the feeling of groundedness provides a solid foundation that imparts a sense of strength, safety, stability, and confidence that one can withstand adversity.
Consistent with the idea of "spreading one's roots into the ground," and the literal translation of terroir as "land" or "soil" ([35]), the feeling of groundedness can be obtained from a connection to a physical environment or place. This connection can be physical in the literal sense, as when working with actual, tangible objects that originate in the local environment, or when immersing in the natural environment itself. We find examples of such immersion in, and connection to, the natural sphere in the East Asian tradition of shinrin yoku, or forest bathing ([19]), and the Nordic cultures' idea of outdoor life (Friluftsliv), which, according to [15], p. 3), provides "a biological, social, aesthetic, spiritual and philosophical experience of closeness to a place, the landscape, and the more-than-human world; an experience most urban people today lack." In the same vein, connection to place may be experienced when directly drawing from the earth, as popularly pursued in urban gardening and farming. Indeed, one of [42], pp. 140–41) informants states, "That's what farming actually is [a connection to the earth].... You are working with the living world. It's the connection you give people to the farm." In addition to a physical connection, consumers can also connect to place in a more symbolic sense. They may do so, for example, by consuming locally produced goods, such as a beer from a nearby brewery. Establishing a connection to one's place to feel grounded may have become especially important as a consequence of migration and mobility. For example, a consumer who has recently been relocated to a certain town may particularly desire to consume products local to that town, thus enabling them to build a connection to that place.
Feelings of groundedness can also arise from a connection to one's social environment. Just as the meaning of terroir also includes its human capital ([35]), the idea of a "place" that provides groundedness, such as home, is often strongly shaped by the people and community associated with that place.
In the social psychology literature, the human need for connectedness or belongingness to other people ([ 4]) has been well established. Running counter to that need is the phenomenon of modern-day alienation ([26]). The concept has been revived by marketing scholars to describe alienation of the consumer from the marketplace ([ 1]), and from a product's producer ([46]). Along the same lines, [ 2] observe postmodern consumers' feelings of personal meaninglessness and loss of moorings brought on by globalization and technology, while stressing the importance of identity, home, and community as antidotes to these feelings.
Although the strongest route to groundedness via people might be connecting to one's closest social surroundings (e.g., one's family), we also see customers trying to reestablish a connection to people by means of certain product choices. Both online and offline, consumers may obtain groundedness by buying directly from the producer. At a farmers market, consumers may buy eggs directly from the person who fed the chickens and collected their eggs. On Etsy, online shoppers can order a breakfast mug from the very person who designed and shaped the piece with their own hands; the shopper might even be able to communicate directly with that person and learn how they developed their passion for handicraft. Either way, this enables the customer to get "closer to the creator" ([39]). On the business side, many firms, big and small, try to facilitate connections between customers and the people behind their products: for example, featuring individual producers on the packaging, indicating the name and address of food suppliers, or communicating via the company's founder or chief executive officer ([14]).
The human environment, or terroir, also includes a historical dimension ([35]). We suggest that feelings of groundedness can also be experienced based on a connection to the past. The past provides a foundation of memories, traditions, and cultural values for individuals to be grounded in.
Examples from the marketing literature illustrate how consumption behavior establishes a connection to the past and begets feelings of groundedness. In [42], some respondent quotes suggest that community-supported farms provide not merely a connection with their local physical environment and the people around them but also a symbolic connection to past generations within one's own family (e.g., a connection to ancestors who were farmers). [44], who investigated Nordic consumers' food consumption motives, state that "in the end, it is the caring food-producer who can bring the ubiquitous brand consumption back to where we were before industrialism" (p. 230). Similarly, [ 3] find that visits to local farmers markets allow consumers to "reconnect with their agrarian roots" (p. 567), searching for "food that is embedded in their personal and shared social histories" (p. 564). In the consumer product domain, we see a resurgence of historic brands such as Converse ([23]) and observe companies helping consumers get connected to, or grounded in, the past. For example, firms may purposefully manufacture according to traditional and artisanal methods, such as making things by hand ([13]), or return to using older, often more "natural" materials and ingredients.
Building on this conceptualization, our first prediction is as follows:
- H1: Products that connect consumers to place, people, and past provide consumers with the feeling of groundedness.
Products that connect consumers to place, people, and past frequently differ from other products in more aspects than their affordance of feelings of groundedness. For example, a local, traditional product is probably also more authentic ([32]; [34]). Likewise, products that connect to place, people, and past could be deemed higher quality or costlier to produce. They may be more unique ([25]), or perceived as made with love ([13]). Consumers may feel a stronger brand attachment to such products ([43]). These products may also provide a greater sense of human contact ([38]), brand experience ([ 5]), brand community (e.g., [28]), and sense of nostalgia ([ 9]). Products that provide a feeling of groundedness may also evoke a feeling of being true to oneself (i.e., self-authenticity or existential authenticity; e.g., [ 2]; [16]), a feeling of knowing who one is (self-identity), a general sense of belonging ([ 4]) that is not about feeling grounded and deep-rooted, or a general sense of meaning in life ([21]; [37]; [40])—all of which could increase one's well-being.
While these related constructs are relevant, we argue that they play different conceptual roles than groundedness. First, some constructs—such as product authenticity, product quality, or product uniqueness—are characteristics of products. They logically cannot cast doubt about the existence of groundedness, which is a feeling about the self.
Second, other alternative constructs could be classified not as characteristics of brands but as feelings about brands. For example, brand attachment is a feeling of connection to a brand. In some situations, feeling connected to a brand might be a consequence of a brand's relationship to a place, people, or the past that a consumer longs to feel a connection with. For example, a consumer may be more likely to feel attached to a wine brand from their own region (or to their favorite laptop brand, which may have nothing to do with feeling connected to place, people, or past). However, this feeling of brand attachment is not a feeling about the self. Thus, it cannot be the same as the feeling of groundedness.
A third category of constructs relates to connectedness but is focused on only one of the three sources. For example, nostalgia, as "a sentimental longing or wistful affection for a period in the past,"[ 6] is related to the past but not necessarily people or place. Likewise, these constructs might be alternative explanations for one of the antecedents of groundedness but not groundedness itself. In addition, nostalgia describes a state of longing or affection, but it does not stipulate that this longing has been satisfied by an actual connection to the past. Thus, nostalgia is conceptually more closely related to the need for groundedness than to actually feeling grounded.
Finally, there are some constructs involving feelings about the self that might be driven by similar antecedents or generate similar consequences as the feeling of groundedness; these include feeling true to oneself (i.e., self-authenticity), a sense of belonging that does not involve a feeling of deep-rootedness, a sense of self-identity, and a general sense of meaning in life. Our studies will assess these alternative constructs to groundedness.
Figure 1 depicts our conceptual framework. At the core of this framework and as summarized in H1 is that there are at least three immediate sources of groundedness: connection to the physical environment, or to place; connection to the social environment, or to people; and connection to the historical environment, or to the past.Figure 1 further depicts our hypotheses about the consequences of the feeling of groundedness; in particular, we consider product attractiveness (H2) and consumer well-being (H3) as important outcome variables. We then examine ways in which marketers can leverage groundedness on the basis of marketing-mix elements (H4) and consumer characteristics (H5).
Graph: Figure 1. Conceptual framework.
In our predictions about downstream effects of groundedness, we hypothesize that groundedness increases product attractiveness and, thus, affects consumer choice. In particular, we suggest that products providing a connection to place, people, and past beget feelings of groundedness for the customer and may therefore be more attractive than their competitors that do not. We thus predict that customers will prefer these products and have stronger intent to purchase and higher willingness to pay (WTP). More formally,
- H2: Products' ability to provide consumers with the feeling of groundedness makes those products more attractive to consumers.
Beyond marketplace outcomes, we hypothesize in our predictions that groundedness increases consumer well-being. In particular, we suggest that feeling grounded provides consumers with a sense of strength, stability, safety, and confidence in one's ability to withstand adversity. As such, feelings of groundedness might provide consumers with a sense of happiness, thus adding to their well-being. We find conceptual support for these predictions in the descriptions of [31] and [42]. [31], p. 82) refers to rootedness as providing "a sense of balance, belonging, and fitting to one's place." Further specifying the elements of well-being afforded by groundedness, Ndi (p. 59) says that rootedness is "the ultimate feeling that provides stability, harmony, and happiness among people and their community," whereas a lack of rootedness leaves a person with a sense of meaninglessness, disconnectedness, emptiness, vulnerability, and unhappiness. Building on [41] work in biodynamics, [42], p. 140) also suggest that emotional connections to one's environment "are a primordial source of spiritual sustenance and a foundation of social and personal well-being and, conversely, that psychological and societal unrest are precipitated by technological forces that separate humanity from its roots in nature." Research on constructs related to groundedness also provides indirect, suggestive evidence for our proposition that groundedness increases consumers' well-being. [27], for example, find that connectedness to nature is positively correlated with subjective well-being. We predict the following:
- H3: The feeling of groundedness increases consumers' subjective well-being.
Marketers can use several marketing-mix variables that help connect consumers to place, people, and past and thus make them feel grounded. Marketers can promote the location where the product is made or ingredients are sourced, engage in storytelling about the history of the brand, or introduce the people who produce the products ([14]; [46]). Marketers can design products in a local or traditional style; use local, ethnic, or traditional ingredients; or employ traditional production processes (e.g., in "indie" products). Marketers can also adjust their channels of distribution to help customers connect to place, people, and past. For example, farms and small producers can use farmers markets (vs. supermarkets) that connect consumers with place, people, and past. Retailers can employ traditional store designs or focus their assortments on more traditional products. We propose the following:
- H4: Marketing-mix variables such as communication, product design, and channels of distribution can be designed to increase the feeling of groundedness.
We expect that consumers differ in how important feelings of groundedness are to them. That is, the level of need for connection with place, people, and past, and thus, for groundedness, varies across consumers. We examine three reasons why the need for groundedness might be heightened in certain consumer segments. First, the need for groundedness should be particularly strong when consumers' life and work make it difficult to establish and maintain strong connections with place, people, and past. We suggest that living in large cities (which are often inhabited by people who did not grow up there, are characterized by social anonymity, and tend to showcase modernity) is a predictor of need for groundedness. With regard to work, we expect that performing mostly computerized work, confined to the limits of one's desktop, puts a distance between individuals and other people as well as the physical environment. We consequently argue that computerized work is associated with a stronger need for groundedness.
Second, we propose that the need for groundedness is stronger when consumers' foundations are shaken or connections with place, people, and past are severed or under pressure. We expect this to have been the case, for example, during the COVID-19 pandemic, a global event that indeed disrupted many people's lives. Accordingly, those who the pandemic had more strongly put in a state of flux should have experienced a higher need for groundedness.
Third, we suggest that the need for groundedness will be more prominent for consumers whose more basic needs are satisfied. Respective proxies such as consumers' socioeconomic status (SES) should thus be correlated with their felt need for groundedness. We predict the following:
- H5: The feeling of groundedness is more important to consumers when their work and life do not provide a strong connection to place, people, and past; when life events shake their foundation; or when their basic needs are already sufficiently met.
With a view to robustness and generalizability, we test our predictions in eight experiments and one consumer survey, based on a variety of samples and data collection techniques (students in behavioral labs at universities, online platforms, and professional market research panels, both in the United States and in Europe). For managerial usability, our study paradigms include both consequential outcome measures as well as marketing-relevant factors that can be manipulated or measured. Study 1 provides evidence that groundedness increases product attractiveness in real economic terms using an incentive-compatible measure of WTP. Studies 2a–c show that groundedness has explanatory value above and beyond alternative constructs. These studies also explore how a product's affordance of groundedness depends on the closeness of the consumer's connection to the provenance of the product or the producer of the product. Studies 3 and 4 provide concrete implications for marketing practice by manipulating product design and assortment, showing how demand for traditional versus innovative products is affected by consumers' current need for groundedness, and exploring proxies that might allow managers to assess said need. In Studies 5a and 5b we focus on psychological effects on consumers. Study 5a shows that groundedness has a positive effect on consumer happiness, whereas Study 5b examines the effect of a grounding product on one's feelings of strength, stability, and safety.
Study 1 tests the effect of groundedness on product attractiveness (H2). We do so in a study paradigm that aims to showcase the managerial relevance of the focal effect. Specifically, we exposed participants of a consumer panel to a more grounding "indie" brand of soap versus a less grounding industrial brand and took an incentive-compatible measure of participants' WTP for each product. We separately tested the extent to which the two brands provide a connection to place, people, and past (see Web Appendix B). We also measured a moderator—importance of the product category to the consumer—to provide further insight into the process and strengthen internal validity (e.g., to alleviate any concerns about demand effects). We reasoned that the self-related benefit of groundedness afforded by indie (vs. industrial) brands should be more pronounced when the product category is more central to the self (i.e., more important to the consumer).
An age- and gender-representative sample of 311 Austrian consumers from a professional market research panel participated for monetary compensation (Mage = 41.8 years; 50.2% female; for instructions and stimuli of this and all following studies, see Web Appendices B–F). All participants were exposed to a color picture and verbal description for two bars of soap. An almond-scented soap made by Firm A was always presented on the left. An olive-scented soap from Firm B was always presented on the right. We manipulated which firm was described as indie ("makes high-quality products that are produced in a small and independent craft business") versus industrial ("makes high-quality products that are industrially produced at scale in a large factory").[ 7] Participants indicated their WTP for a bar of soap from both companies separately using an incentive-compatible elicitation method (dual-lottery Becker–DeGroot–Marschak procedure; e.g., [13]). This method provides an incentive-compatible measure of what the product is worth to participants.
Next, participants indicated which soap provided relatively stronger feelings of groundedness by rating agreement with the following two statements (translated from the original German): "When I think of this firm's soap ... I feel deep-rooted and firmly anchored ('grounded')" and "I can firmly feel my feet on the ground." Participants also indicated how well a graphic depicting a human form with branches for arms and a deep, wide root system instead of legs (see Figure 1) represented their emotional state. The three items were measured on a seven-point scale (1 = "true for Firm B," and 7 = "true for Firm A") and were averaged to create a groundedness index (α = .87).[ 8] We captured the importance of the underlying product category to the consumer with a three-item measure (e.g., "The product category 'soap' is very important to me"). All measurement items used in this and subsequent studies, as well as their reliability statistics, are listed in Web Appendices B–F. Unless indicated differently, items are measured on seven-point scales (where "strongly disagree/does not describe my feelings at all/not true of me at all/true for Brand B," etc. is coded as 1, and "strongly agree/describes my feelings very well/very true of me/true for Brand A," etc. is coded as 7).
We ran a repeated-measures analysis of variance (ANOVA) with consumers' WTP in euros as the repeated-measures factor and our indie versus industrial counterbalancing manipulation as the between-subjects factor (for complete results, see Web Appendix B). We find the expected interaction effect (F( 1, 309) = 174.51, p < .001). Follow-up contrast analyses show that participants are willing to pay more for the soap of Firm A if that product is portrayed as an indie (Mindie = €3.29) versus as an industrial (Mindustrial = €1.91; F( 1, 309) = 37.47, p < .001) brand. Likewise, the soap of Firm B is valued more when Firm B is described as an indie (vs. industrial) company (Mindie = €3.12, Mindustrial = €2.11; F( 1, 309) = 20.67, p < .001)—a notable 60% increase in value. For moderation and mediation analyses, we calculated the intraindividual delta WTP (WTPFirm A − WTPFirm B: MFirm A indie = €1.18, MFirm A industrial = −€1.21; F( 1, 309) = 174.51, p < .001).
An ANOVA on the groundedness measure indicates a significant effect: when Firm A is described as indie, participants more strongly declare that Firm A makes them feel grounded (MFirm A indie = 5.15) compared with when Firm A is described as industrial (MFirm A industrial = 2.92; F( 1, 309) = 269.58, p < .001). Mediation analysis ([20], Model 4, 10,000 bootstrap samples) shows that the WTP effect is mediated by feelings of groundedness (indirect effect = 1.24, 95% confidence interval [CI95%]: [.87, 1.67]). A moderation analysis ([20], Model 1) with the delta WTP measure as dependent variable confirms the hypothesis that the indie premium increases as the category importance increases (p < .001; for details, see Web Appendix B). Finally, a moderated mediation analysis ([20], Model 8) shows that this interaction effect is mediated by groundedness: the indirect effect of indie versus industrial on delta WTP through feelings of groundedness is always significant but stronger at high versus low levels of category importance (indirect effect16th percentile = .79, CI95%: [.51, 1.12]; indirect effect50th percentile = 1.13, CI95%: [.77, 1.55]; indirect effect84th percentile = 1.54, CI95%: [1.04, 2.16]; index of moderated mediation = .21, CI95%: [.11,.34]).
Study 1 finds that products making a connection to the past, to people, and to a place make consumers feel more grounded, which increases their WTP. Thus, the result in Study 1 supports H2. The effect is managerially relevant: the more grounding product yielded a notable 60% increase in WTP. In addition, Study 1 shows that the effect is moderated by the importance of the product category. The pattern of moderated mediation, where the indie versus industrial nature of the brand is less important to feelings of groundedness when the product is less important to the consumer's identity, provides further evidence for our process.
A limitation of Study 1 is that indie versus industrial products may differ in more aspects than their ability to provide a feeling of groundedness. For example, an indie brand might provide higher value to consumers by being perceived as more authentic ([32]) and more unique ([25]) than an industrial brand. Further, the description of the indie brand and its production method might give consumers a greater sense of love ([13]), human contact ([38]), attachment ([43]), brand experience ([ 5]), and brand community (e.g., [28]). Or the indie brand might simply be higher quality and costlier to produce. Our mediation and moderated mediation provide initial evidence for the proposed groundedness process, suggesting that these alternative processes are not the only drivers of the effects on WTP. We explicitly address these alternative explanations in Study 2.
One major element of our theory is that the feeling of groundedness afforded by a product results from the connection that product provides to place, people, and past (H1). If products are indeed connectors between customers and their place, people, and past, we should be able to affect groundedness—and product attractiveness (H2)—not just by manipulating the place, people, or past of the product as we did in Study 1 but also by manipulating the place, people, or past of the customer. Thus, in Study 2a, we keep brands and products constant and manipulate how much groundedness a brand is able to provide as a function of a customer characteristic (i.e., customer location), rather than a product characteristic.
We asked 172 students (Mage = 21.9 years; 79.7% female) at a Northeastern U.S. university (n = 89, for a gift voucher and cookies) and an Austrian university (n = 83, for course credit) to imagine that they had just moved to either Karlstad or Umeå in Sweden. We then asked them to choose (using a three-item measure, e.g., "Which of the two craft beers do you choose?") which of two real Swedish craft beer brands, Good Guys Brew from Karlstad and Beer Studio from Umeå, they would purchase on their first night out. Next, participants reported which of the two brands they perceived would make them feel more grounded ("In the situation described, this brand would make me feel deep-rooted," "This brand would make me feel well-grounded," and "In a metaphorical sense: Which of the two craft beers would rather make you feel as illustrated by the following picture?" [showing the picture of a human/tree form with deep roots]; α = .90). All items in this study were captured on seven-point scales where one anchor was the beer from Karlstad and the other anchor the beer from Umeå. We counterbalanced which beer was shown on the left- versus right-hand side. Before the participant location manipulation, we also asked participants to rate the two brands on a relative scale regarding nine product characteristics that might make either product more attractive. Because these were product characteristics that should not have been influenced by the participant's location, and because they were measured before the location manipulation, they did not—and could not—explain our results (for results regarding the control variables in this and all subsequent studies, see Web Appendices C–F). At the end of the study, we captured some information about the participants' relation to beer and to Sweden (e.g., "Have you ever been to Sweden?," "How much do you like beer in general?").[ 9]
A one-way ANOVA shows that participants who moved to Karlstad prefer the Karlstad-based beer significantly more than those who moved to Umeå (MKarlstad = 4.80, MUmeå = 4.14; F( 1, 170) = 6.70, p = .010). Similarly, the Karlstad-based beer provides relatively more groundedness to participants who moved to Karlstad versus Umeå (MKarlstad = 4.29, MUmeå = 3.79; F( 1, 170) = 5.77, p = .017). Groundedness mediates the effect of residence location on preference (indirect effect = .40, CI95%: [.07,.74]; [20], Model 4). For each of the nine alternative constructs, the focal indirect effect via groundedness remains significant when we include the alternative construct as a rival mediator.
Study 2a shows that groundedness drives product attractiveness (H2) when we keep products constant but manipulate the place of the customer. This study highlights that the groundedness effect depends not only on the features of the product but also on the situation of the customer. Managerially, the study shows that local brands are particularly grounding and thus attractive to local consumers. Study 2a manipulated how participants relate to a place that is connected to a focal product, and thus how much groundedness it affords them. Unlike Study 2a, Study 2b capitalizes on participants' existing relationship to a place. Study 2b further addresses alternative constructs to groundedness by measuring them after the focal manipulation.
The week before Christmas, we asked 1,306 Austrian students from a university in Vienna (Mage = 22.8 years; 55.4% female; compensated by a lottery for an iPhone 11 and five €10 gift vouchers, prescreened for having grown up in Austria but outside Vienna and for celebrating Christmas) to imagine they were celebrating Christmas in Vienna this year and looking to buy a Christmas tree at a local market. We then varied between-subjects whether the market's Christmas trees originated from the state the participant grew up in or from a randomly selected other Austrian state. The trees were thus not connected to participants' current place (i.e., where they were studying and buying the tree) but to either the place where they grew up or a third location in Austria. Then, we assessed purchase intent for the Christmas tree using four items (e.g., "I would very much like to buy a Christmas tree at this market"). We next captured feelings of groundedness from purchasing a Christmas tree at that market, using the same three items as in Study 2a. Finally, participants completed two-item measures of alternative constructs (the product's authenticity, uniqueness, quality, love, production costs, sense of human contact, brand experience, feeling of belonging to a brand community, and attachment). In addition, we measured participants' desire to support the producer as a possible alternative explanation. Due to this study's use of multiple items for each construct, we were able to ascertain that groundedness is empirically distinct from the other constructs captured (purchase intent and alternative constructs) using the [11] criterion. We performed the same tests in all subsequent studies with multi-item measures of our dependent variables (see Web Appendices C–F).
Participants are more intent on buying a Christmas tree from the focal market if it is from their own state (Mown place = 5.35) versus another state in the same country (Mother place = 4.95; F( 1, 1,304) = 24.27, p < .001). Further, when the trees originate from participants' own state, participants experience stronger feelings of groundedness than when the trees are from another state (Mown place = 3.39, Mother place = 3.15; F( 1, 1,304) = 8.43, p = .004), which is in line with H1. We do not find significant differences between conditions with regard to the alternative explanations captured (ps >.087). Differences in perceived production costs (Mown place = 4.17, Mother place = 4.30; F( 1, 1,304) = 2.92, p = .088) are marginally significant but run in the opposite direction of the dependent variable. Thus, they are unable to explain our results. Consistent with H2, a mediation model ([20], Model 4) shows that groundedness mediates the treatment effect on purchase intent (indirect effect = .11, CI95% [.03,.18]). For each of the ten alternative constructs, the focal indirect effect via groundedness remains significant when we include the alternative construct as a rival mediator.
Studies 2a and 2b show that a product that connects a consumer to a place they relate to (a city they move to, the state they are from) makes them feel more grounded and is more attractive than a product originating from a specified place they do not relate to (another city or state in the same country). One pertinent question is how much that feeling of groundedness depends on the closeness of the connection to place, people, and past. While the more grounding option in Studies 2a and 2b connects customers to their own ("my") current or past place, the indie brand utilized in Study 1 merely provided a connection to "a" place (and "the" people who made it and "the" past, respectively). Our view is that, ceteris paribus, the depth of groundedness gradually increases with the closeness of the connection. The closer the personal relationship of the customer to the place, people, and past represented by the product, the stronger the connection and thus feelings of groundedness established via the product. We test this prediction in the context of a customer connecting to the people dimension next.
Study 2c addresses whether differences in closeness indeed matter—that is, whether they afford different levels of feelings of groundedness when compared directly. Beyond that, the study isolates connection to people as a potential driver of groundedness (H1).
Two hundred U.K. crowd workers on Prolific (Mage = 33.8 years; 55.0% female; for monetary compensation) were asked to indicate their feelings of groundedness associated with the use of a coffee mug (using the same measure as in Studies 2a and 2b). To sample different levels of personal closeness along the proposed continuum, the producer of the mug was manipulated to be either "an artisan that is personally close to you (e.g., a close friend, relative, partner, etc.)" or "an artisan that is a distant acquaintance of yours (e.g., a colleague from work, a neighbor, a friend of a friend, etc.)." We measured perceived connection to people through the mug using three items (e.g., "Drinking from this mug, I somehow feel a connection to 'my people'"). We used the same control measures as in Study 2b (except for the motivation to provide financial support by purchasing a product, given that there was no purchase in this study).
First, the pattern of results for groundedness and connection to people supports our theorizing about a continuum of closeness and, thus, groundedness: perceived connection to people is significantly higher when the artisan producer is a close other versus when they are merely an acquaintance (Mclose = 4.34, Mdistant = 3.75; F( 1, 198) = 6.63, p = .011). The same is true for feelings of groundedness: participants experience stronger feelings of groundedness when considering the coffee mug produced by an artisan that is a close other versus one that is merely a distant acquaintance (Mclose = 4.14, Mdistant = 3.29; F( 1, 198) = 14.78, p < .001). Further, a mediation model ([20], Model 4) shows that producer closeness mediates the effect on groundedness (indirect effect = .42, CI95%: [.09,.75]). Importantly, for each of the nine alternative constructs, the focal indirect effect via groundedness remains significant when we include the alternative construct as a rival mediator.
Thus, Study 2c shows that being personally closer to one of the sources of groundedness enables consumers to experience stronger feelings of groundedness. More precisely, groundedness is a function of how close the consumer's relationship is to the product's place, people (e.g., the product's producer), or past. As for different routes to groundedness, the study shows that a product's people dimension alone (e.g., its producer) can boost groundedness via a stronger perceived connection to people established by the product. Managerially, the findings are important because marketers can choose the extent to which they highlight the closeness or similarity between customers and producers. In addition, the study highlights that managers may need to search for personally relevant and close sources of groundedness from the perspective of a given target customer.
The next set of studies investigates how the groundedness effect can be leveraged via marketing-mix elements (Studies 3a and 3b) and which types of customers have a particularly high need for groundedness (Studies 3a and 4).
Study 3a focuses on connections to past as a source of groundedness (H1) by manipulating product design (H4). We also examine how the effect of groundedness on product attractiveness (H2) varies across consumers by capturing their chronic need to connect to the past (the higher this need, the stronger the groundedness effect should become). Study 3b manipulates consumers' state need for groundedness and addresses category management considerations by testing how consumers' need for groundedness impacts the preference for traditional versus innovative products.
We showed 223 students in the behavioral laboratory of a large European university (Mage = 23.9 years; 65.5% female; for monetary compensation or course credit) two sets of cutlery (from Brand A and Brand B) side by side, stipulating that they were of comparable price and quality. We manipulated product design to provide more versus less connection to the past by using a more traditional versus modern product design. We manipulated which set of cutlery was presented on the left- versus right-hand side (i.e., as Brand A vs. B). Using adapted versions of the measures in Studies 1 and 2, we asked participants to indicate which of the two brands they would rather purchase, which would make them feel more grounded, and which evoked a stronger connection to the past. Need to connect to the past as a chronic consumer trait—our moderator—was measured in terms of agreement with three items (e.g., "I generally try to see if I can somehow satisfy my desire to [metaphorically] 'connect to the past'").[10]
Our manipulation proved effective: Participants more strongly associate Brand A ( = 7, Brand B = 1) with a connection to the past when Brand A cutlery had a traditional design (MBrand_A_traditional = 5.49, MBrand_A_modern = 1.97; F( 1, 221) = 405.25, p < .001). As expected (H4), we find a significant effect on groundedness—Brand A is perceived to provide more groundedness (relative to Brand B) when Brand A features traditional design (MBrand_A_traditional = 4.39, MBrand_A_modern = 3.65; F( 1, 221) = 18.50, p < .001). For product preference, we find an overall preference for the modern cutlery (MBrand_A_traditional = 3.69, MBrand_A_modern = 4.48; F( 1, 221) = 9.01, p = .003; of course, the fact that traditional products provide a stronger sense of groundedness does not preclude that many people might still prefer a specific set of modern cutlery over a specific set of traditional cutlery, or modern designs over traditional ones in general). More importantly, and as expected (H2), we find a positive effect of groundedness on product preference (b = .61, p < .001), and a positive indirect effect ([20], Model 4) of traditional (vs. modern) design on preference through groundedness (indirect effect = .55, CI95%: [.27,.89]). As one would expect, preference becomes even stronger for the modern cutlery when the groundedness path is controlled for (estimated MBrand_A_traditional = 3.40, estimated MBrand_A_Modern = 4.74).
As anticipated, we find that one's general need to connect to the past significantly moderates purchase preference (p < .001; [20], Model 1). Thus, participants with a low need to connect to the past have a more pronounced preference for the modern cutlery; conversely, participants with a high need to connect to the past show a preference for the traditional cutlery (e.g., at need to connect to past = 1, conditional effect = −2.53, CI95%: [−3.58, −1.48]; at need to connect to the past = 7, conditional effect = 1.29, CI95%: [.002, 2.57]). A moderated mediation analysis ([20], Model 58; see Web Appendix D) shows that traditional design affords a stronger feeling of groundedness, and that groundedness becomes a more important driver of preference as general need to connect to the past increases. In fact, at very low levels of general need to connect to the past, a product's ability to provide feelings of groundedness no longer significantly impacts product preference (e.g., at need to connect to the past = 1, conditional effect = .33, CI95%: [−.05,.71]).
In summary, Study 3a shows that by varying a marketing-mix element (product design) to be more traditional (vs. modern), marketers can affect customer preference via feelings of groundedness. This is because the marketing-mix element directly caters to a source of groundedness (H4).
Study 3b investigates preference for traditional versus innovative products as a direct function of consumers' current need for groundedness and manipulates this need. We also perform a test of how the relative interest in different product categories—traditional versus innovative—is affected by different levels of need for groundedness, pointing to potential boundary conditions of the groundedness effect.
Two hundred crowd workers on Prolific (Mage = 33.4 years; 54.0% female) from the United Kingdom took part in this study for monetary compensation. Participants filled out two ostensibly unrelated surveys. The first manipulated participants' current need for groundedness. Participants in the high-need condition read, "Research has shown that feelings of groundedness can be positive or negative depending on the context and situation we are in." They were then asked to describe a recent situation where feeling grounded was desirable to them because "you metaphorically felt your roots were too loose and weak with respect to your connection to a place, to people, and the past." Conversely, participants in the low-need condition read, "Research has shown that feelings of groundedness can be negative or positive," and were asked to describe a situation where groundedness was undesirable to them because "you metaphorically felt your roots were too dense and strong." After completing the writing task and reporting their current need for groundedness on a version of our three-item groundedness scale, participants were thanked and told they would be forwarded to another study. Here, participants were introduced to two different online stores, presented side by side: one specializing in "the best traditional products" and one specializing in "the best innovative products." We then asked participants to indicate which of the stores they would prefer to shop at on a seven-point scale, with Store A and Store B as anchors. We alternated which of the stores (A vs. B) was presented as traditional versus innovative in our stimuli. We subsequently reversed the Store A versus B preference scores for half the data set, so that the innovative store preference was always anchored at 1 and the traditional store preference was always anchored at 7.
Our manipulation was effective: participants who wrote about a situation where their need for groundedness was high reported experiencing a higher need for groundedness (M = 5.25) than those who wrote about a situation where need for groundedness was low (M = 4.11; F( 1, 198) = 41.19, p < .001). In terms of shopping preferences, participants in the high-need-for-groundedness condition showed a stronger preference for the online store with traditional (vs. innovative) products (M = 4.00) than those in the low-need-for-groundedness condition (M = 3.47; F( 1, 198) = 4.17, p = .043).
Thus, and in line with H4, Study 3b shows that relative interest in purchasing traditional products is higher in situations and contexts where consumers' need for groundedness is high. In situations and contexts where groundedness is less sought after, innovative products become relatively more interesting.
Studies 3a and 3b suggest that groundedness is not equally attractive and relevant to all consumers in all situations. For segmentation purposes, it is important to know which consumers are more likely to have a strong enduring need for groundedness. As predicted in H5, we argue that the feeling of groundedness is more important to consumers when their work and life (e.g., computerized desktop work, living in a large city) do not provide a strong connection to place, people, and past; when certain life events (e.g., the COVID-19 crisis) shake their foundation; or when their basic needs are already sufficiently met (e.g., when they have higher SES). In Study 4, we use a survey to measure these consumer characteristics, along with need for groundedness and preference for products that connect to place, people, and past. The study was conducted in spring 2020, at the beginning of the COVID-19 pandemic and first lockdown. This enabled us to assess the impact of a disruptive life event on the need for groundedness.
An age- and gender-representative sample from a U.S. consumer panel completed this survey for monetary compensation (N = 325; Mage = 45.5 years; 51.1% female). We first measured product preference and need for groundedness: preference for products connected to one's place, people, and past were measured (in random order) using three items each (e.g., "I like to purchase products that connect me to 'my place' ['my people'/'my past'], i.e., my physical [social/historic] environment"). We merged these into one global index of purchase interest. Need for groundedness was measured using a version of our three-item scale, adapted to measure general need for groundedness (e.g., "In general, I want to feel deep-rooted"). We next captured a series of demographic and lifestyle variables.
To assess a potential lack of connection to people, place, and past in consumers' work and social lives, we captured three variables. First, we asked respondents about the type of area they live in (1 = "in the countryside," and 7 = "in a big city"). We hypothesized that living in large cities (which are often inhabited by people who did not grow up there, are characterized by social anonymity, and tend to showcase modernity) is a predictor of need for groundedness. Second, we assessed participants' desktop work using two items (e.g., "During the week [e.g., when being at work] ... I primarily work at the computer"). We expected a positive relationship between desktop work and need for groundedness, because a disproportionate amount of computerized work (while confined to one's desktop) separates individuals from other people as well as the physical environment. A similar logic might apply to people whose job is characterized as "work of the head" (i.e., work that contains many abstract tasks), as opposed to people who perform manual labor ("work of the hands") or work in social jobs ("work of the heart"; [17]). Respondents accordingly indicated which of these three categories their current or most recent job fell into.
Next, to assess a potential link between need for groundedness and a disruptive major life event, we examined perceived impact of the COVID-19 crisis on the consumer's life. We assessed this with a single item ("Due to the current Corona [COVID-19] crisis, I feel that my life is in a state of major change"). Last, we theorized that the need for groundedness should become more prominent when basic needs such as food and shelter are not a concern. Therefore, we tested whether higher SES (measured on a three-item scale [e.g., "I have enough money to buy things I want"]) might be an effective proxy for one's need for groundedness. No other measures were taken.
First, and as expected, we find a significant and positive correlation between one's need for groundedness and purchase intent for products connecting to place, people, and past (r = .57, p < .001). Second, we analyzed the correlations of all proposed indicators with the need for groundedness. In particular, need for groundedness correlates positively with desktop work (r = .26, p < .001), SES (r = .30, p < .001), change experienced as a result of COVID-19 (r = .12, p = .030), and living in a big city rather than the countryside (r = .10, p = .079), but correlates negatively with performing work of the hands (r = −.11, p = .040; for a complete correlation table of this study; see Web Appendix E).
Third, we ran multivariate ordinary least squares (OLS) regressions with all predictor variables on both need for groundedness and purchase intent. For those variables that emerged as significant predictors for both the need for groundedness and purchase intent, we examined whether the need for groundedness mediates the respective effects on purchase intent while entering all other variables as covariates. For conciseness, we report only significant results hereinafter (see Table 1 for details).
Graph
Table 1. Multivariate OLS Regression Models (Study 4).
| Need for Groundedness | Purchase Interest in Products Connected to Place, People, and Past |
|---|
| Unstandardized Coef. | Standardized Coef. | | | Unstandardized Coef. | Standardized Coef. | | |
|---|
| b | SE | β | t | p-value | b | SE | β | t | p-value |
|---|
| (Constant) | 2.983*** | .417 | | 7.157 | <.001 | .682 | .46 | | 1.481 | .140 |
| Living environment | .038 | .034 | .061 | 1.11 | .270 | .08* | .038 | .101 | 2.105 | .036 |
| Desktop work | .143*** | .037 | .248 | 3.892 | <.001 | .24*** | .041 | .324 | 5.905 | <.001 |
| Work of the hands | −.17 | .19 | −.061 | −.9 | .370 | −.228 | .21 | −.064 | −1.091 | .276 |
| (1 = hands, |
| 0 = otherwise) |
| Work of the head | −.281 | .172 | −.117 | −1.64 | .102 | −.699*** | .19 | −.226 | −3.686 | <.001 |
| (1 = head, |
| 0 = otherwise) |
| Change through COVID-19 | .091* | .04 | .123 | 2.302 | .022 | .234*** | .044 | .247 | 5.356 | <.001 |
| SES | .185*** | .037 | .272 | 5.008 | <.001 | .287*** | .041 | .33 | 7.038 | <.001 |
| Age | .01* | .004 | .136 | 2.357 | .019 | .000 | .004 | .005 | .104 | .917 |
| Gender (1 = male, 0 = female) | .004 | .13 | .002 | .034 | .973 | .465** | .144 | .151 | 3.228 | .001 |
| R2 = .162, d.f. = 8, 316 | R2 = .379, d.f. = 8, 316 |
1 *p < .05.
- 2 **p < .01.
- 3 ***p < .001.
- 4 Notes: Mediation models ([20], Model 4): Mediator = need for groundedness, DV = purchase interest; ( 1) IV = SES: indirect effect = .10, CI95%: [.05,.15]; ( 2) IV = desktop work: indirect effect = .07, CI95%: [.03,.12]; ( 3) IV = change through COVID-19: indirect effect = .05, CI95%: [.002,.10].
The multivariate OLS models showed that three predictors remain significant for both the need for groundedness (NG) and purchase intent (PI) when simultaneously including all variables in the model: ( 1) desktop work (NG: b = .14, SE = .04, t(316) = 3.89, p < .001; PI: b = .24, SE = .04, t(316) = 5.91, p < .001), ( 2) SES (NG: b = .19, SE = .04, t(316) = 5.01, p < .001; PI: b = .29, SE = .04, t(316) = 7.04, p < .001), and ( 3) change related to COVID-19 (NG: b = .09, SE = .04, t(316) = 2.30, p = .022; PI: b = .23, SE = .04, t(316) = 5.36, p < .001). Need for groundedness mediates the effect of all three variables on purchase intent (in line with H2; see Table 1 and Web Appendix E).
Our "work of the head" dummy was not significant in the multivariate OLS model. We conclude that the "work of the head/heart/hands" measure was probably too rough and thus unable to adequately detect the important nuances in job characteristics that affect the need for groundedness. We were also surprised that one's living environment did not emerge as a significant predictor for need for groundedness in the multivariate OLS model. A closer look at the data reveals, however, that a disproportionately large number (29.2%) of respondents in our sample indicated living in big cities (i.e., chose the endpoint of the scale). When dichotomizing the measure (i.e., living in big city vs. not), we find the predicted positive effect: people living in a big city have a heightened need for groundedness (see Web Appendix E).
In summary, Study 4 finds that a higher need for groundedness is apparent in consumer profiles characterized by larger societal trends: living in big cities (urbanization), doing desktop work at the computer (digitization), and undergoing major change (such as during the COVID-19 pandemic). Further, groundedness seems to be more relevant for high-SES consumers.
Thus far, we have provided a cohesive picture of groundedness in terms of both triggers (H1) and market-relevant outcomes (H2), as well as ways for marketers to leverage groundedness (H4, H5). In the final two studies, we examine the implications of groundedness for consumers' psychological well-being (H3).
To test our hypothesis that feeling grounded increases consumers' subjective well-being (H3), Study 5a measures happiness as a consequence of attaining groundedness. We also test another managerial manipulation: channel type (H4). Study 5b expands into a broader range of psychological outcomes; as outlined in our conceptual framework, the feeling of groundedness should provide consumers with a sense of strength, stability, safety, and self-confidence. We test these outcomes in the context of using locally grown ingredients and also investigate alternative constructs to groundedness, such as self-authenticity, meaning in life, or sense of identity.
We randomly assigned 190 Austrian students (Mage = 22.5 years; 50.5% female; lab-based, for monetary compensation) to think about shopping at a supermarket or local farmers market. We then asked about their feelings of groundedness; happiness; and being connected to place, people, and past. Happiness was measured using three items (e.g., "In the situation just described, how happy would you feel?"). Feelings of groundedness were measured using our three-item measure. Connection to place, people, and past were captured separately using three items each (e.g., "Having been in the supermarket [to the farmers market] makes me feel connected to my physical/social/historic environment"). The order of the dependent measures (happiness, groundedness), as well as the order of the item blocks capturing connection to place, people, and past, were counterbalanced. Perceived quality and price were measured as control variables.
Channel type has a significant effect on groundedness and happiness. Participants who thought about shopping at the farmers market reported feeling significantly more grounded (Mfarmersmarket = 4.66 vs. Msupermarket = 3.80; F( 1, 188) = 18.19, p < .001) and happier (Mfarmersmarket = 5.32 vs. Msupermarket = 4.87; F( 1, 188) = 7.94, p = .005). Consistent with our theorizing (H4), shopping at the farmers market leads to significantly higher perceived connection to place (Mfarmersmarket = 4.96 vs. Msupermarket = 4.03; F( 1,188) = 15.74, p < .001), people (Mfarmersmarket = 4.58 vs. Msupermarket = 3.45; F( 1, 188) = 24.60, p < .001), and past (Mfarmersmarket = 3.73 vs. Msupermarket = 2.52; F( 1, 188) = 25.47, p < .001). We also find support for serial mediation such that the effect of channel on happiness is mediated, in series, by connection to place, people, and past, and groundedness (for mediation results, see Web Appendix F). All effects remain robust when we enter quality and price as covariates.
Study 5a thus supports our prediction that the feeling of groundedness increases consumers' subjective well-being (H3) while providing converging evidence for H1. Finally, the manipulation of distribution channel (H4) offers an actionable strategy for marketers to leverage groundedness.
In our last study, we employ the context of locally grown ingredients to test a broader range of psychological outcomes of groundedness. We also test the explanatory value of groundedness against alternative constructs that are self-related, such as feelings of self-authenticity or meaning in life.
Three hundred four students from a major European university completed Study 5b's online study for course credit. We excluded 12 participants for failing our reading check, leaving us with a final data set of 292 participants (Mage = 22.3 years; 69.5% female). Participants were asked to think about making apple pie on a Saturday; specifically, a pie with Boskoop apples—their favorite pie-making variety. In addition, they were told that these apples were from either an orchard only 12 kilometers from their home or an orchard 1,200 kilometers from their home. Participants then completed a short survey that measured five downstream psychological outcomes of groundedness using a five-item scale: "I feel truly safe as a person," "I experience a feeling of inner strength," "I feel truly stable," "I have a strong feeling of basic trust and confidence in myself," and "I feel that nothing can stir me up" (α = .89). Afterward, we measured feelings of groundedness using our three-item measure. Finally, participants completed four multi-item measures intended to capture alternative explanations (self-authenticity [e.g., "I feel out of touch with the 'real me'"], meaning in life [e.g., "I have a good sense of what makes my life meaningful"], self-identity [e.g., "I have the feeling that I know who I am"], feeling of belonging [e.g., "I have a feeling of belonging"]).
Participants who considered making apple pie with apples grown close to home scored significantly higher in terms of experiencing the related psychological downstream consequences than those using apples grown far away (Mlocal = 5.08, Mnonlocal = 4.61; F( 1, 290) = 12.22, p = .001). Thus, the apple pie made with local products boosted participants' personal feelings of strength, safety, and stability (for effects on the individual dependent variable items, see Web Appendix F). They also reported significantly stronger feelings of groundedness (Mlocal = 4.65, Mnonlocal = 4.06; F( 1, 290) = 15.20, p < .001). A mediation model ([20], Model 4) shows that the downstream consequences are mediated by feelings of groundedness (indirect effect = .27, CI95%: [.13,.44]). Importantly, the indirect effect via feelings of groundedness on the downstream consequences holds when we add, one at a time, each of the four alternative explanations as a rival mediator.
Study 5b thus confirms positive psychological downstream effects of groundedness (H3) tested in the realm of local products. Products grown closer to the consumer—that is, products that are more strongly connected to one's place—make consumers feel not only more grounded but also stronger, safer, and more stable.
In this research, we have provided systematic evidence that products can provide consumers with feelings of groundedness by giving them a sense of connection to place, people, and past. We do so across nine studies (eight experiments and one survey), both online and in the lab, using different populations (business students, crowd workers on Amazon Mechanical Turk and Prolific, and members of commercial, representative panels) across two continents (total N > 3,000). We have tested our theory for robustness across a variety of product domains, including both disposable and durable consumer goods (food, care products, seasonal products, and tableware), using real brands to strengthen external validity as well as highly controlled stimuli for internal validity. We have provided process evidence via mediation, moderation, and moderated mediation.
This work introduces feelings of groundedness to the marketing literature by identifying these feelings as an important construct for marketing research and systematically examining it as a driver of consumer behavior. While references to groundedness and related constructs can be found in philosophy (e.g., [47]), different domains of psychology (e.g., [29]), and psychotherapy ([31]), the concept of groundedness is new to experimental research in marketing, consumer behavior, and mainstream psychology. Existing research in consumer culture theory has given passing treatment to concepts such as "rooted connections" ([42]) and has definitely been inspirational to this work. However, it has neither discussed nor empirically explored the full concept of groundedness with its antecedents, proxies, boundary conditions, and consequences, which we have aimed to do here.
We also contribute to the growing literature on consumer well-being. [47], p. 43) proposes that "every human being needs to have multiple roots. It is necessary for him to draw well-nigh the whole of his moral, intellectual, and spiritual life by way of the environment of which he forms a natural part." Our work indeed shows that groundedness is related to happiness and a sense of strength, stability, and safety; thus, we propose groundedness as a novel antecedent of these outcomes.
We also theorized about three sources of feelings of groundedness: connections to place, people, and past. Although the three sources are often empirically intertwined, we show that they are theoretically distinct and powerful in fueling consumers' feelings of groundedness. Our analysis further provides rich insight on the nature of these connections by showing that the extent to which products provide feelings of groundedness is a graded function of closeness. That is, a product provides stronger feelings of groundedness when the product's place, people, or past is closer to the consumer. Finally, by identifying the role of groundedness and its sources, we offer an overarching theoretical explanation for major current consumer trends, such as buying local products (connected to place), produced by people we relate to (connected to people), and according to traditional production methods (connected to the past).
Feelings of groundedness are worthy of managers' attention because these feelings have important downstream consequences as shown across our studies. In particular, feelings of groundedness impact consumers' brand preference and WTP. In Study 1, for example, consumers were willing to pay a price premium of about 60% for the product that provided more groundedness.
Our work also provides actionable implications for product and brand management: we give concrete approaches regarding how firms can elicit groundedness by showing consumers their product's connection to place, people, and past. For example, our results in Studies 1, 2a, and 2b show how presenting a product as artisanal or highlighting the local origin of a product can provide feelings of groundedness. In Studies 3a and 3b, we have shown that managers can utilize other marketing-mix elements such as product design or retail assortment and configure them (e.g., as more traditional instead of modern) to provide a stronger connection to the past. Similarly, Study 5a shows that a marketer's choice of distribution channel (e.g., farmers market) has an impact on feelings of groundedness.
In terms of customer targeting, we have pointed out when and for whom groundedness is more important. In particular, we have shown that traditional (vs. innovative) products benefit from situational differences in the need for groundedness (Study 3b). On the level of individual differences, in Study 3a, only consumers with a high chronic need to connect to the past preferred the more traditional cutlery design. Our representative survey (Study 4) further showed a higher need for groundedness among consumers who are particularly affected by large global trends or major disruptive events. These global trends (e.g., digitization, urbanization) and major life events (e.g., the COVID-19 pandemic) make it harder for consumers to feel connected to people, place, and past. From a groundedness perspective, it is not surprising that during the safety- and stability-threatening COVID-19 pandemic, customers returned to the familiar grocery brands consumed with their families while growing up ([ 7]). There are probably multiple drivers for this behavior, but it is likely that consumers chose these products, at least in part, because of the connections to place, people, and past—and thus feeling of groundedness—they provide.
This is the first series of experimental studies investigating feelings of groundedness. As such, many questions remain for future research. With regard to antecedents, for example, we have focused on products as means for consumers to experience feelings of groundedness. However, anecdotal evidence suggests that there are other ways for consumers to feel more connected to place, people, and past and, consequently, more grounded: for example, through services such as genealogy websites, cooking classes, lectures on local history, or yoga and meditation classes providing "grounding" exercises.
The scope of Study 4 has allowed us to identify an initial set of indicators for who has a higher need for groundedness and why, but it is clear there will be additional consumer characteristics and lifestyle variables helpful to marketers in identifying relevant customer segments. For example, people who travel frequently for work and have little chance to connect to their current physical environment may seize opportunities to (re)-connect to place—such as through a local craft beer—to feel more grounded. Likewise, pandemics such as COVID-19 are not the only type of events that can shake a person's foundation. Stressful life events such as separation or loss, starting a new job, or moving homes may cause a higher need to feel grounded. Similarly, the need for groundedness may be subject to seasonal variations. Preliminary insights from our own qualitative explorations suggest that individuals' need for groundedness may be particularly high during the holiday season and other festive occasions, such as Christmas, Thanksgiving, Ramadan, and one's own birthday. Apart from that, interestingly, the need for groundedness appears to be higher during the colder seasons. We believe a more thorough testing of these hypotheses seems promising and would likely have important implications. If the initial signals are correct, for example, studies of scanner or panel data should reveal variations in the demand for products that connect to place, people, and past across the year.
Finally, we have only begun to examine boundary conditions. For example, it seems possible that in some situations strong roots not only provide strength and stability but could also constrain movement, thus giving consumers the feeling of being "stuck" and unable to escape their roots. Imagine growing up on a farm, surrounded by one's family, and doing things day after day in the same way they have traditionally been done by previous generations. A person in this situation will likely feel grounded but might also feel more motivated to break free, move away, or challenge the status quo. If such is the case, too much groundedness might even backfire. Future research might thus enrich the present investigation by focusing on potential downsides of groundedness.
This research introduced feelings of groundedness as a relevant construct for marketing research and consumer behavior. We have demonstrated its importance to marketers by documenting that it increases product attractiveness and that it can be manipulated through a variety of marketing-mix strategies and used for targeting consumer segments prone to a lack of groundedness. We also have shown that groundedness is important to consumer well-being, pointing to important consumer welfare and policy implications. We expect that the importance of this topic to consumers and marketers will only increase as digitization, urbanization, and global migration continue to challenge consumers' connections to place, people, and past.
Footnotes 1 Dhruv Grewal
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: We are grateful for funding from the Austrian Federal Ministry for Education, Science, and Research through the HRSM scheme.
4 Online supplement:https://doi.org/10.1177/00222429211027469
5 See, for example, https://www.merriam-webster.com/dictionary/grounded and https://www.yourdictionary.com/grounded.
6 See https://www.lexico.com/definition/nostalgia.
7 See [30] for a discussion of the effectiveness this type of experimental design and, for example, [8], [18], and [32] for examples of its use.
8 We captured a total of six different items to measure feelings of groundedness in this study but shortened the scale to align with the three-item measure used in the other studies. Results for the six-item scale are fully consistent (see Web Appendix B).
9 We also controlled for participants' perceived awareness of research hypothesis (PARH scale; [36]): one might argue that if a participant had been aware of the research hypothesis, their revealed preferences might have been biased. Results make this possibility unlikely because there is no significant interaction effect between treatment and PARH scale on product preference (p = .288; see [22]).
We counterbalanced whether the need to connect to the past was measured before versus after product presentation and measurement of the dependent variables. Results remain robust when controlling for this.
References Allison Neil K. (1978), " A Psychometric Development of a Test for Consumer Alienation from the Marketplace ," Journal of Marketing Research , 15 (4), 565 – 75.
Arnould Eric J. , Price Linda L.. (2000), " Authenticating Acts and Authoritative Performances: Questing for Self and Community, " in The Why of Consumption: Contemporary Perspectives on Consumer Motives, Goals, and Desires , Ratneshwar S. , Mick D.G. , Huffman C. , eds. London : Routledge , 140 – 63.
Autio Minna , Collins Rebecca , Wahlen Stefan , Anttila Marika. (2013), " Consuming Nostalgia? The Appreciation of Authenticity in Local Food Production ," International Journal of Consumer Studies , 37 (5), 564 – 68.
Baumeister Roy F. , Leary Mark R.. (1995), " The Need to Belong: Desire for Interpersonal Attachments as a Fundamental Human Motivation ," Psychological Bulletin , 117 (3), 497 – 529.
Brakus J. Joško , Schmitt Bernd H. , Zarantonello Lia. (2009) , "Brand Experience: What Is It? How Is It Measured? Does It Affect Loyalty?" Journal of Marketing , 73 (3), 52 – 68.
Brewers' Association (2020), " Brewers Association Releases Annual Growth Report for 2019," press release (April 14), https://www.brewersassociation.org/press-releases/brewers-association-releases-annual-growth-report-for-2019.
Chaudhuri Saabira. (2020), " Comfort Foods Make a Comeback in the Coronavirus Age," The Wall Street Journal (April 24), https://www.wsj.com/articles/shoppers-stock-up-on-comfort-food-amid-pandemic-11587726462.
Dahl Darren W. , Fuchs Christoph , Schreier Martin. (2015), " Why and When Consumers Prefer Products of User-Driven Firms: A Social Identification Account ," Management Science , 61 (8), 1978 – 88.
Davis Fred. (1979), Yearning for Yesterday: A Sociology of Nostalgia. New York : The Free Press.
Etsy (2021), "Investor Relations: Key Figures," (accessed May 18, 2021), https://investors.etsy.com/overview/key-figures/default.aspx.
Fornell Claes , Larcker David F.. (1981), " Evaluating Structural Equation Models with Unobserved Variables and Measurement Error ," Journal of Marketing Research , 18 (1), 39 – 50.
Fromm Erich. (1976), To Have or to Be? London : Continuum.
Fuchs Christoph , Schreier Martin , van Osselaer Stijn M.J.. (2015), " The Handmade Effect: What's Love Got to Do with It? " Journal of Marketing , 79 (2), 98 – 110.
Fuchs Christoph , Schreier Martin , van Osselaer Stijn M.J. , Kaiser Ulrike. (2019), " Making Producers Personal: How Breaking Down the Wall between Producers and Consumers Creates Value," working paper, Technical University of Munich.
Gelter Hans. (2010), " Friluftsliv as Slow and Peak Experiences in the Transmodern Society ," Norwegian Journal of Friluftsliv , 1 – 22.
Gino Francesca , Norton Michael I. , Ariely Dan. (2010), " The Counterfeit Self: The Deceptive Costs of Faking It ," Psychological Science , 21 (5), 712 – 20.
Goodhart David. (2020), Head Hand Heart: The Struggle for Dignity and Status in the 21st Century. London : Allen Lane.
Gunasti Kunter , Ross William T. Jr. (2010), " How and When Alphanumeric Brand Names Affect Consumer Preferences ," Journal of Marketing Research , 47 (6), 1177 – 92.
Hansen Margaret M. , Jones Reo , Tocchini Kirsten. (2017), " Shinrin-Yoku (Forest Bathing) and Nature Therapy: A State-of-the-Art Review ," International Journal of Environmental Research and Public Health , 14 (8), 851 – 99.
Hayes Andrew F. (2013), An Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach. New York : Guilford Press.
Heine Steven J. , Proulx Travis , Vohs Kathleen D.. (2006), " The Meaning Maintenance Model: On the Coherence of Social Motivations ," Personality and Social Psychology Review , 10 (2), 88 – 110.
Larsen Jeff T. , Peter McGraw A.. (2011), " Further Evidence for Mixed Emotions ," Journal of Personality and Social Psychology , 100 (6), 1095 – 1110.
Loveland Katherine E. , Smeesters Dirk , Mandel Naomi. (2010), " Still Preoccupied with 1995: The Need to Belong and Preference for Nostalgic Products ," Journal of Consumer Research , 37 (3), 393 – 408.
Low Sarah A. , Adalja Aaron , Beaulieu Elizabeth , Key Nigel , Martinez Steve , Melton Alex , et al. (2015), Trends in Local and Regional U.S. Food Systems: A Report to Congress. Washington, DC: U.S. Department of Agriculture, Economic Research Service, AP-068.
Maly Ico , Varis Piia. (2016), " The 21st-Century Hipster: On Micro-Populations in Times of Superdiversity ," European Journal of Cultural Studies , 19 (6), 637 – 53.
Marx Karl ([1844] 2007), Economic and Philosophic Manuscripts of 1844. Mineola, NY : Dover.
Mayer F. Stephan , Frantz Cynthia McPherson. (2004), " The Connectedness to Nature Scale: A Measure of Individuals' Feeling in Community with Nature ," Journal of Environmental Psychology , 24 (4), 503 – 15.
McAlexander James H. , Schouten John W. , Koenig Harold F.. (2002), " Building Brand Community ," Journal of Marketing , 66 (1), 38 – 54.
McAndrew Frank. (1998), " The Measurement of 'Rootedness' and the Prediction of Attachment to Home-Towns in College Students ," Journal of Environmental Psychology , 18 (4), 409 – 17.
Meyvis Tom , van Osselaer Stijn M.J.. (2018), " Increasing the Power of Your Study by Increasing the Effect Size ," Journal of Consumer Research , 44 (5), 1157 – 73.
Ndi A. Ebede. (2014), " 5-Factor Rootedness Assessment Model: Toward a New Assessment Model in Psychotherapy, " doctoral thesis, California Institute of Integral Studies.
Newman George E. , Dhar Ravi. (2014), " Authenticity Is Contagious: Brand Essence and the Original Source of Production ," Journal of Marketing Research , 51 (3), 371 – 86.
Nisbet Elizabeth K. , Zelenski John M.. (2013), " The Nr-6: A New Brief Measure of Nature Relatedness ," Frontiers in Psychology , 4 , 1 – 11.
Nunes Joseph C. , Ordanini Andrea , Giambastiani Gaia. (2021), " The Concept of Authenticity: What It Means to Consumers ," Journal of Marketing , 85 (4), 1 – 20.
Rozin Paul , Wolf Sharon. (2008), " Attachment to Land: The Case of the Land of Israel for American and Israeli Jews and the Role of Contagion ," Judgment and Decision Making , 3 (4), 325 – 34.
Rubin Mark. (2016), " The Perceived Awareness of the Research Hypothesis Scale: Assessing the Influence of Demand Characteristics ," Figshare , http://dx.doi.org/10.6084/m9.figshare.4315778.
Sarial-Abi Gülen , Vohs Kathleen D. , Hamilton Ryan , Ulqinaku Aulona. (2017), " Stitching Time: Vintage Consumption Connects the Past, Present, and Future ," Journal of Consumer Psychology , 27 (2), 182 – 94.
Schroll Roland , Schnurr Benedikt , Grewal Dhruv. (2018), " Humanizing Products with Handwritten Typefaces ," Journal of Consumer Research , 45 (3), 648 – 72.
Smith Rosanna K. , Newman George E. , Dhar Ravi. (2016), " Closer to the Creator: Temporal Contagion Explains the Preference for Earlier Serial Numbers ," Journal of Consumer Research , 42 (5), 653 – 68.
Steger Michael F. , Frazier Patricia , Oishi Shigehiro , Kaler Matthew. (2006), " The Meaning in Life Questionnaire: Assessing the Presence of and Search for Meaning in Life ," Journal of Counseling Psychology , 53 (1), 80 – 93.
Steiner Rudolph. (2005), What Is Biodynamics: A Way to Heal and Revitalize the Earth. Fair Oaks, CA : Steiner.
Thompson Craig J. , Coskuner-Balli Gokcen. (2007), " Countervailing Market Responses to Corporate Co-Optation and the Ideological Recruitment of Consumption Communities ," Journal of Consumer Research , 34 (2), 135 – 52.
Thomson Matthew , MacInnis Deborah J. , Whan Park C.. (2005), " The Ties That Bind: Measuring the Strength of Consumers' Emotional Attachments to Brands ," Journal of Consumer Psychology , 15 (1), 77 – 91.
Ulver-Sneistrup Sofia , Askegaard Søren , Kristensen Dorthe Brogård. (2011), " The New Work Ethics of Consumption and the Paradox of Mundane Brand Resistance ," Journal of Consumer Culture , 11 (2), 215 – 38.
U.S. Department of Agriculture (2016), "Direct Farm Sales of Food: Results from the 2015 Local Food Marketing Practices Survey," ACH 12-35 (December 2016), https://www.nass.usda.gov/Publications/Highlights/2016/LocalFoodsMarketingPractices%5fHighlights.pdf.
Van Osselaer Stijn M.J. , Fuchs Christoph , Schreier Martin , Puntoni Stefano. (2020), " The Power of Personal ," Journal of Retailing , 96 (1), 88 – 100.
Weil Simone. (1952), The Need for Roots. New York : Harper Colophon.
~~~~~~~~
By Isabel Eichinger; Martin Schreier and Stijn M.J. van Osselaer
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 39- Consumer Self-Control and the Biological Sciences: Implications for Marketing Stakeholders. By: Zheng, Yanmei; Alba, Joseph W. Journal of Marketing. Jul2021, Vol. 85 Issue 4, p105-122. 18p. 2 Charts. DOI: 10.1177/0022242920983271.
- Database:
- Business Source Complete
Consumer Self-Control and the Biological Sciences: Implications for Marketing Stakeholders
The authors argue that appreciation of the biological underpinnings of human behavior can alter the beliefs and actions of multiple marketing stakeholders in ways that have immense welfare implications. However, a biological perspective often deviates from the lay perspective. The realization of improved welfare depends in part on narrowing this gap. The authors review biological evidence on self-control and report ten empirical studies that examine lay response to biological characterizations of self-control. The authors contrast lay response with scientific understanding and then offer implications of biology—as well as the gap between the scientific and lay perspectives—for policy makers, firms, consumers, marketing educators, and scholars. The authors also identify opportunities for future research. They conclude that marketing scholars can and should play an active role in narrowing the gap between the scientific and lay perspectives in the service of both theory development and human welfare.
Keywords: autonomy; biology; genetics; human welfare; lay beliefs; neuroscience; public policy; self-control
The hope is that when it comes to dealing with humans whose behaviors are among our worst and most damaging, words like "evil" and "soul" will be as irrelevant as when considering a car with faulty brakes, that they will be as rarely spoken in a courtroom as in an auto repair shop....When a car is being dysfunctional and dangerous and we take it to a mechanic, this is not a dualistic situation where (a) if the mechanic discovers some broken widget causing the problem, we have a mechanistic explanation, but (b) if the mechanic can't find anything wrong, we're dealing with an evil car. ([82], p. 611)
One general message that should emerge from these discoveries is tolerance for others—and for ourselves. Rather than blaming other people and ourselves for being depressed, slow to learn or overweight, we should recognize and respect the huge impact of genetics on individual differences. ([73], p. 91)
Society's understanding of human ills is constantly evolving. Tuberculosis is no longer viewed as a lifestyle disease; epilepsy is no longer viewed as evidence of demonic possession. Today, we understand autism as a disruption of circuitry in the "social brain," but three generations ago, a prominent child psychologist claimed that autism resulted from a mother's withholding of affection from her unwanted child. We have slowly experienced a shift in attitudes toward depression and posttraumatic stress disorder, conditions that previously were considered character issues but now are understood as biological syndromes with clear neural etiologies. In each of these instances, a nonbiological account was replaced with a biological account. We argue that biology can similarly advance marketing's contribution to human welfare if it is included as a complement to traditional psychological, anthropological, and economic perspectives on consumption, particularly with respect to the vital topic of self-control. As a behavior, self-control is fundamental to human welfare; as a trait, self-control is central to our evaluations of ourselves and others; as a principle, self-control is woven into the fabric of our legal, political, and religious institutions. In many cases, self-defeating consumption reflects a self-control failure, sometimes abetted by marketing practice.
Our argument takes both sides of the debate coin ([53]): to our research peers, we engage in advocacy; to others, we refute the welfare-defeating policies that stem from indifference to, or rejection of, biological causation. To do so, we consider two biological domains that have produced a tsunami of findings in very recent years: neuroscience and genetics. We argue, however, that biological insights will not translate directly into improved welfare if those insights fail to make an impression on marketing's many stakeholders. We further argue that the road to welfare-enhancing policies will be rocky if the lay public is resistant to the implications of biology. Our task in this regard is to understand laypeople's existing beliefs about biological causation and, moreover, to gauge how those beliefs can be shaped by findings from the biological sciences.
Our presentation is organized accordingly. The first section describes relevant biological findings. The second section summarizes a set of studies that investigate lay response to these biological findings and, in so doing, illustrates the gap between science and lay beliefs. The third section offers implications of the first two sections for marketing's many stakeholders ([54]) and identifies opportunities for future research.
In this section, we discuss biological causation first in terms of neuroscience and then in terms of genetics. Definitive accounts are aspirational, but the current state of the art suffices. When appropriate, we speculate about likely departures from lay beliefs in anticipation of our subsequent empirical investigation.
We focus on three broad realms of human behavior in which the gap between lay beliefs and neuroscience is consequential for human welfare: decision making, compulsion, and impulsivity.
In one sense, the proposition that decision making involves the biological brain is uncontroversial. Neural activity underlies every memory we retrieve and every inference we draw, and disorders that impede mental activity (e.g., Alzheimer's) are understood to be biological conditions. However, both prior research and the studies we present in the next section reveal laypeople's strong dualistic[ 6] tendency to view reasoning and decision making as unconstrained by biology. To address the neuroscience of decision making, we draw on the bottom-up relationship between the limbic system (including the amygdala) and prefrontal cortex (PFC), colloquially known as the seats of emotion and executive control, respectively.
Research in this vein suggests that pure reasoning in everyday life is rare; that is, decision making so routinely involves emotion that the distinction between thought and feeling is a false dichotomy. This assertion can be traced to [19] somatic marker hypothesis. Damasio begins with the uncontroversial premises that emotions are bodily responses to a situation and, furthermore, that associations are formed between features of the situation and those bodily responses. Similar emotions will arise if one reexperiences the situation but also if one merely anticipates the situation ([ 4]). One key feature of Damasio's hypothesis is that these bodily responses can be mentally represented in the brain ([17]), thereby bypassing the body and enabling speedier responses. Put differently, internal representations of those bodily responses encode what a particular outcome would feel like. This information is relayed to the ventromedial PFC (vmPFC), which integrates the emotional signals with response options. A second key feature of the hypothesis is that a decision may be informed by emotional inputs that do not rise to the level of conscious awareness. Regardless, the net positive and negative somatic signals will either determine the decision or bias it.
The somatic marker hypothesis explains why risk assessments may not align with objective cost–benefit calculations ([51]). The hypothesis also provides a process explanation for some system 1 judgments[ 7] ([40]).
This biological account of decision making is pertinent to our argument in two ways. First, because emotion can play a role without ever entering one's consciousness, the pervasive influence of emotion on decision making is likely not apparent to the lay mind, so laypeople may overestimate their capacity to rise above the fray and make reasoned judgments as autonomous beings. Lay and scientific beliefs are especially likely to clash in the case of moral judgment. Consider "moral dumbfounding," in which deeply held moral beliefs are based on an intellectually hollow intuition or feeling of moral rectitude ([32]). This phenomenon comports with [19] own clinical evidence that the extent to which people make socially favored (vs. purely utilitarian) judgments depends on their vmPFC. Patients with damaged vmPFCs make poor personal, social, and moral judgments not because they do not comprehend the objective implications of their judgments but because they are unable to feel the implications of those judgments. We anticipate that laypeople would be especially surprised that the moral judgments that define us are biologically rooted in the tangible brain rather than in the intangible mind.
Second, a biological account of decision making is consistent with the view that moral/social "deviance" exists on a continuum. Over the past two generations, psychiatry has bent to the logic of biological causation and now understands that psychopaths suffer from abnormalities in the limbic system and in the connection between the limbic system and the frontal cortex, resulting in severely reduced empathy and interpersonal connections ([41]). For those of us closer to the mean, our social judgments still vary with the strength of the connection between the limbic system and PFC, as well as with the relative roles played by the utilitarian dorsolateral PFC and the feelings-oriented vmPFC. The difference between a psychopath and an unempathetic person may be a matter of degree rather than a matter of kind ([ 8]). Regarding our subsequent argument regarding human welfare, it is also important to note that structural deficiencies in the brain may be caused by multiple factors. Specifically, structural brain changes can be caused not only by disease and accident but also by less sensational environmental factors that are pervasive and distributed unequally across the population (e.g., poverty).
Marketing has been implicated as complicit in self-defeating acts of compulsive consumption, including substance abuse ([57]) and overeating ([63]). We consider the biological roots of these behaviors.[ 8]
Addiction is a morbidity that neuroscience has come to understand with some certitude (see [49]). We suspect that the public now feels some sympathy for the problem and a willingness to accept its biological causality in the colloquial sense that drugs can "hijack the brain." The lay public may also view the lure of a drug in terms of the intense pleasure it evokes. By contrast, neuroscientists view addiction as a malfunction of the brain's reward circuitry, with a "reward" defined simply as an object or event that induces approach behavior. Different substances may act on different brain loci, but they all increase dopamine concentrations in the striatum. Consumption also creates a conditioned response to the environmental cues that predict dopamine release. Subsequent exposure to these cues induces intense craving ([ 6]). Because these cues are unlikely to be eliminated from the drug user's life, addiction is a chronic disease, as the potential for relapse lurks forever. Thus, relapse is viewed by many experts as a biological syndrome rather than a failure of willpower. The trap of addiction is further exemplified by evidence that long-term substance abuse can effect changes in the brain's reward system, including a possible reduction in dopamine sensitivity. Such an alteration requires more intense consumption to attain the same hedonic outcome ([41]).
Although drug addiction is increasingly receiving lay sympathy, the overconsumption of food is less likely to be attributed by laypeople to a hijacking of the brain. The causes of obesity are varied and complex, but we now have abundant insights into its biological causation and the futility of mere willpower ([10]). Notably, a biological explanation akin to substance abuse is emerging in the form of a "food addiction" model of obesity. The drug addiction analogy works on multiple levels: food consumption has a compulsive element, consumption evokes similar changes in the reward circuitry of the brain ([12]), and the cessation of consumption triggers withdrawal symptoms ([83]). In terms of a crude stop-go model, evidence suggests that overt consumption is determined by the relative signaling strength of competing clusters of neurons, such that consumption is triggered when the "eat" (go) cluster fires more strongly than the "stop" cluster. In a further nod to addiction, the process appears to stimulate the brain's opioid system ([99]). The reward system of obese people tends to be less responsive to dopamine and to have a lower density of dopamine receptors; thus, as with drug addiction, even greater consumption is required to attain sustained pleasure ([41]). In summary, biology's conceptualization of obesity is a far cry from society's character-related attributions that revolve around gluttony.
It has also been argued that obese individuals are victims of a second vicious cycle driven by the prevalent stigma that accompanies being overweight (and in which marketing is again complicit; [30]). Stigma leads to shame and stress. Stress results in higher levels of cortisol, which in turn leads directly to fat deposition and indirectly to a more sensitized food reward system and greater food consumption ([92]).
Our primary emphasis in this research concerns deficits in self-control that are neither temporally bound nor context-specific. Rather, we focus on self-control as an enduring trait that is shaped early (even prenatally) in life.
The granular biological story is intricate, but the bird's-eye view again invokes the hijacking metaphor and the relationship between the amygdala and PFC. Unlike emotion-based decision making, however, trait self-control involves a top-down relationship between the PFC and amygdala. Deficits in trait self-control involve a neural system that is perverted by excess glucocorticoid production ([70]). Glucocorticoid production is an otherwise adaptive response to situational danger; in modern society, however, stress is often chronic, as when it derives from the duress of poverty, physical or emotional abuse, exposure to violence, parental separation, or other adversities ([31]; [43]; [67]). Such sustained psychological stress produces a continual oversupply of glucocorticoids, with damaging outcomes that take root in childhood and persist over the lifetime ([ 7]; [24]; [52]).
The neurological story and its consequences are neatly captured by [82]: childhood adversity in the form of poverty leads to higher basal glucocorticoid levels and/or a more reactive glucocorticoid stress response; thinner frontal cortex with a lower metabolism and less excitable synapses; and poorer frontal functioning in the domains of working memory, emotion regulation, impulse control, and executive decision making. Childhood adversity can atrophy and blunt the functioning of the hippocampus and frontal cortex ([34]) while stimulating the amygdala to enlarge and develop more excitable synapses ([102]). Consequently, childhood adversity empowers the amygdala to inhibit the frontal cortex, which ordinarily should gain the ability to inhibit the amygdala during adolescence. Childhood adversity also damages the dopamine system, leaving the adult more vulnerable to drug and alcohol addiction ([88]). Such damage can also lead to depression, learned helplessness, blunted empathy, and lower prosociality. Thus, it again may be said that the executive control system of individuals with low self-control has been hijacked by their more primitive limbic system ([14]).
Our previous discussion of decision making and compulsive consumption makes it clear that pure reasoning (i.e., emotion-free cognition) not only is rare but also provides little help to those who wish to escape from the acute problems of addiction and obesity. Problematic decision making and compulsive consumption can be due to either brain injury or counterproductive alterations in the brain's reward system in response to consumption. Our major assertion here about impulsivity is that environmental conditions also can adversely influence the brain subtly, but with cumulative effects. We speculate that because the biological account of impulsivity imputes subtle and cumulative effects, it may be elusive to the lay mind. As a result, the lay public may be less sympathetic toward (and surely less informed about) the initial and often irreversible decisions that lead to addiction in the first place, despite lay sympathy regarding the difficulty of escaping addiction. In turn, misconceptions about the initial behavior could translate into a lack of sympathy toward the drug abuser. However, as we shall contend later, it is this biological explanation for why someone would initially travel down a destructive path (i.e., the biological roots of impulsivity) that can provide a basis for public policy intervention.
A complete account of biological causation requires a discussion of genetics, inasmuch as the role of the genome in "psychological" phenomena is now known to be sizable. When genes influence thoughts and behaviors, the brain may be the mediator, but the influence of genes can be appreciated in the absence of mediation, including in the applications of central concern here. Marketing has recognized the relevance of neuroscience ([72]), but the role of heritability has largely been ignored ([20]; [86]). The trajectory of behavioral genetics research is steep, but even in its present state, the field's implications are profound. We highlight the following key points to inform stakeholder implications.
Across a diverse set of traits, including psychological traits, genes have been shown to account for more than 50% of the variance, with the remaining variance being largely unsystematic ([44]; [75]; [93]). More pertinent to our consumption theme, weight is highly heritable.[ 9] Across twin studies and adoption studies, the heritability of weight is estimated to be about 70% ([73], p. 29). The mechanisms by which genes influence weight are still under investigation ([50]), but evidence indicates that body mass index (BMI) is a function of the body's hereditary predisposition to gain fat when food is plentiful. For example, research shows that genes associated with higher BMI (e.g., the FTO gene; [16]) result in greater responsiveness to food cues and decreased feelings of postconsumption satiety. Thus, individuals with a genetic tendency toward a high BMI will have a greater propensity to gain weight and will experience more difficulty losing it.
However, despite these large and robust genetic influences, the lay public is unevenly calibrated regarding its understanding of the genetic contribution to different traits. For example, people vastly underestimate the heritability of weight and school achievement but appear relatively well-attuned to the significant heritability of height and schizophrenia ([73]).
A trait is polygenic if it is controlled by many different genes. A major development in genetic science is the ability to use a person's genetic code (reflected in an individual polygenic score) to predict from birth that person's propensity to exhibit particular dispositions or outcomes. For example, [87] used polygenic scores to predict academic motivation and the Big 5 personality traits. Moreover, the polygenic nature of most traits suggests that genetically influenced traits exist along a continuum. In contrast to outdated medical models of mental disorders, traits such as schizophrenia and autism are now recognized not as categorical syndromes but as existing along a spectrum. The same logic applies to various personality traits (e.g., extraversion), physical states (e.g., BMI), and behavioral tendencies (e.g., addiction); that is, everyone possesses a polygenic score that places them somewhere along the spectrum, and "normal" is defined as a particular range of a characteristic.
Despite the large share of variance explained by genes, the unexplained variance suggests that it is possible, albeit difficult, to swim against the genetic tide (as in the case of weight). It is also important to appreciate the omnipresent gene–environment interactions. Genes and environment combine to produce neurological outcomes. An important example is childhood adversity. The same adverse event does not impact every child equally; some children are genetically more resilient than others ([ 5]). This observation is especially important from a policy perspective if individual polygenic scores can identify the most vulnerable individuals and target intervention accordingly.
Drawing on neuroscience and genetics, we have tried to reinforce our opening observation that understanding can improve when a nonbiological explanation is replaced by a biological explanation. We conclude this section with an illustration of how biology can add explanatory depth to social science. [78] showed how risk factors associated with family structure lead to materialism and compulsive consumption and, further, how the effect of these risk factors can be mediated by the stress they place on the individual. Subsequent research reinforced the importance of stress on consumer behavior and well-being ([62]; [81]). The biology literature reviewed previously adds depth by clarifying the underlying neurological mechanism.
Although enriching a behavioral phenomenon with biology may inspire confidence among social scientists, our presentation thus far is mute regarding whether a biological explanation of social phenomena can alter lay opinion. Along this line, [105] found that social risk factors alone did little to alter people's beliefs that an individual subjected to those risk factors had control over undesirable outcomes that arose subsequently. Consequently, public policies and marketing practices that build on findings such as [78] may not engender public acceptance.[10] Given the importance of public support for the success of policy and practice, we next examine whether biological evidence can alter public attitudes.
Research on lay reactions to neural processes, neural structures, and genetics is sparse. [105] did find that when a consumer's lack of self-control over food consumption was described as attributable to the "short-circuiting" of a brain center via a firm's development and sale of "hyperpalatable foods," participants perceived that the consumer had less personal control over their food consumption. However, it is unclear that laypeople would be similarly forgiving in the absence of a malign external agent or in the presence of different forms of biological causation.
To address these issues and inform policy and marketing practice, we conducted ten studies that examined lay reactions to various biological models (specifically, neural processes, neural structures, and genetics) of self-control. The studies measured participants' perceptions of either ( 1) a protagonist's control over a lapse of self-control or ( 2) their own vulnerability to the same lapse of self-control. The former assesses whether the perceiver has sympathy for the protagonist's plight; the latter assesses whether the perceiver can empathize with the protagonist. Both measures inform public policy because greater sympathy and empathy would presumably lead to greater willingness to support intervention. To simplify the exposition, we initially spotlight results regarding moral transgressions (e.g., shoplifting) among individuals who possess a high belief in free will (the dominant segment of the U.S. population). Table 1 provides a summary of our findings, Web Appendix A contains a summary table of the experimental design and statistical results of the studies, and Web Appendix B contains the full details of each study. We describe the most important takeaways in narrative form.
Graph
Table 1. Summary of Findings.
| Study | Independent Variables | Dependent Measures | Main Findings |
|---|
| 1a | Behavior (eating vs. shoplifting) | Interpretation of the neuroprocess model from three visual depictions Self-likelihood of succumbing to the same (eating/shoplifting) temptation (between 0 and 100)
| Denial of biological causation was transgression-specific. For a moral transgression (i.e., shoplifting), the causal role of neural processes was denied by half of the sample. Acknowledging the causal role of neural processes did not lead to increased perceived self-vulnerability. Perceived self-vulnerability was also transgression-specific, with a lower estimate for the moral transgression.
|
| 1b | Individual differences in religiosity | Interpretation of the neuroprocess model from three visual depictions Self-likelihood of succumbing to the same shoplifting temptation (between 0 and 100) Locus of the ability to act differently (chosen from the tangible brain, other tangible body parts, the intangible mind, or the intangible soul; recoded to form a perceived intangibility score)
| Acknowledging the causal role of neural processes did not lead to an increased perceived self-vulnerability (as in Experiment 1a). Higher religiosity was correlated with a tendency to attribute a superior ability to resist temptation to intangible qualities of the self.
|
| 2 | Trait (intelligence, self-control, integrity vs. empathy) Individual differences in religiosity
| Locus of the differences in [intelligence, self-control, integrity, or empathy] (chosen from the tangible brain, other tangible body parts, the intangible mind, or the intangible soul; recoded to form a perceived intangibility score) | Perceived locus of personality was trait-specific. Virtuous traits (i.e., integrity and empathy) were predominantly viewed as intangible. Higher religiosity was correlated with the tendency to attribute integrity and empathy to the intangible qualities of people; no such association was observed with intelligence and self-control.
|
| 3a | Dependent measure (perceived control vs. self-likelihood) Biological causation (absent vs. present; operationalized as a brain-structure model) Individual differences in free-will beliefs
| The degree of voluntary control the protagonist had over his/her shoplifting behavior on a seven-point scale anchored by "no control" (1) and "complete control" (7) Self-likelihood of succumbing to the same shoplifting temptation (between 0 and 100), followed by questions regarding the locus of the ability to act differently
| Perceived control assigned to a moral transgressor was reduced when a structural neurological deficit was implicated. In contrast, denial of self-vulnerability was pervasive. Rationales primarily invoked intangible qualities of the self and were associated with participants' religiosity (as in Experiment 1b).
|
| 3b | Biological causation (absent vs. present; operationalized as a brain-structure model with a visual image portraying modest neural damage) Individual differences in free-will beliefs
| Self-likelihood of succumbing to the same shoplifting temptation (between 0 and 100), followed by questions regarding the locus of the ability to act differently | Among participants with a high belief in free will, denial of self-vulnerability was replicated when a verbal description of a structural neurological deficit was accompanied by a visual image of modest brain damage. In contrast, participants with a low belief in free will increased perceived self-vulnerability when the verbal model was accompanied by the visual image. The rationales for denial of self-vulnerability were based on intangible qualities of the self and were associated with religiosity (as in Experiments 1b and 3a).
|
| 3c | Behavior (shoplifting, drunk driving, drug use, or verbal abuse) Biological causation (absent vs. present; operationalized as a brain-structure model with two visual images contrasting an extensively-damaged brain with a healthy brain) Individual differences in free-will beliefs
| Same as Experiment 3b | Regardless of belief in free will, perceived self-vulnerability rose when the verbal model was accompanied by two visual images contrasting the damaged brain with a healthy brain. The rationales for the denial of self-vulnerability were based on intangible qualities of the self and were associated with religiosity (as in Experiments 1b, 3a, and 3b).
|
| 3d | Biological causation (absent vs. present; operationalized as a genetic brain-structure model) Individual differences in free-will beliefs
| Same as Experiment 3b | Among participants with a high belief in free will, denial of self-vulnerability was observed, conceptually replicating the results of Experiments 3a and 3b. Participants with a low belief in free will increased perceived self-vulnerability when genetic causation was implicated, conceptually replicating the results of Experiments 3b and 3c. The rationales for the denial of self-vulnerability were based on intangible qualities of the self and were associated with religiosity (as in Experiments 1b, 3a, 3b, and 3c).
|
| 4a | Deliberation (absent vs. present) Biological causation (absent vs. present; operationalized as a neuroprocess model) Individual differences in free-will beliefs
| The degree of voluntary control the protagonist had over his/her food choices on a seven-point scale anchored by "no control" (1) and "complete control" (7) | Among participants with a high belief in free will, biological causation reduced perceived control if the behavior was reflexive but not otherwise. Among participants with a low belief in free will, perceived control was low regardless of biological causation if the behavior was reflexive; in addition, biological causation reduced perceived control if the behavior appeared deliberative.
|
| 4b | Deliberation (absent vs. present) Individual differences in free-will beliefs
| Same as Experiment 4a | The findings from Experiment 4a were closely replicated when genetic causation substituted for neural causation. |
| 5 | Biological causation (no model, neuro-process model, vs. brain-structure model with two visual images contrasting an extensively-damaged brain with a healthy brain) Individual differences in free-will beliefs and political orientation
| Rank order of six policies that address self-control issues from 1 to 6 (reversed coded such that a larger number indicates greater favorability; the ranks of the three biology-based policies were averaged to create an average rank dependent variable) | Among participants with a high belief in free will, support for biology-based interventions was relatively low and unaffected by biological causation when biology was portrayed as a neural process; support increased when biological causation was portrayed as a structural neurological deficit. Among participants with a low belief in free will, support for biology-based interventions was high across conditions. The same patterns were observed when political orientation substituted for belief in free will.
|
Our initial studies (Experiments 1a and 1b) presented participants with a veridical but simplified biological model that described self-control as determined by the relative strength of two competing neural pathways (i.e., "go" vs. "stop"). Participants were asked to consider a protagonist who failed to resist a temptation and to choose one of three visual depictions that best fit their own interpretation of the scenario. The depictions characterized the neural processes as ( 1) correlated with the protagonist's willpower failure but not causal, ( 2) the consequence of willpower failure, or ( 3) the cause of willpower failure. Participants then rated their own likelihood of succumbing to the temptation if the same neural processes took place in their own brain.
We found that half the participants attributed self-control failure to the causal influence of neural processes (i.e., the third depiction). Regardless of their choice of the visual depiction, however, participants assessed their own likelihood of succumbing to the temptation as low. Participants who interpreted the neural model as causal rated their own likelihood of transgressing at less than 20%, which was statistically indistinguishable from the self-assessments provided by participants who interpreted the neural model as noncausal.
Experiment 1a also showed that appreciation of biology is behavior-dependent. When the transgression was amoral (e.g., succumbing to a tempting food), biological causation was embraced more frequently and self-vulnerability was more willingly acknowledged. Together, these results suggest that participants viewed themselves as unconstrained by biology, especially in the case of a moral transgression.
In Experiment 1b and all subsequent studies in which participants were asked to estimate the likelihood that they would fall victim to the same self-control failure as the protagonist, we followed up by asking participants to indicate the root of their superior ability to exert self-control. Participants chose from four options: the tangible brain, other tangible body parts, the intangible mind, or the intangible soul. Across Experiments 1b−3d, most participants chose the intangible mind or soul; in Experiment 1b, 82.0% of the participants did so. These results provide direct evidence of the dualistic nature of lay beliefs regarding self-control.
Experiment 2 followed by directly examining people's view of the roots of self-control. For comparison, it also explored people's views of the roots of intelligence, integrity, and empathy. Participants were asked to indicate the locus of the differences in these traits across the human population by choosing from the tangible brain, other tangible body parts, the intangible mind, or the intangible soul. Results showed that all traits were viewed as predominantly intangible, but with intelligence deemed less intangible (67.4% of participants) than self-control (73.7%), integrity (84.8%), or empathy (81.9%). Notably, more participants perceived the intangible soul to be the locus of integrity and empathy (24.6% and 26.6%, respectively) than the locus of intelligence or self-control (3.9% and 5.4%, respectively). Thus, the morality-tinged traits of integrity and empathy were viewed less secularly than self-control. The strong inclination to characterize each trait as intangible (ranging between 67.4% and 81.9%) is consistent with lay acceptance of mind–body dualism ([ 9]) but contrasts with evidence that human traits have a strong genetic basis ([73]).
Many neuroscientific findings consist of a correlation between a mental phenomenon and brain activity. Because people view their virtues as intangible (see Experiments 1b and 2), they may view the mere presence of neural activity as insufficient evidence of biological causation. In contrast to mere brain activity, we speculated that physical alterations to brain structures would be less easily dismissed.
In Experiments 3a−3d, we described a self-control failure as the result of structural damage in the brain. In Experiment 3a, participants were told of a relationship between childhood stress and a later lack of self-control. We manipulated biological causation by either presenting or withholding scientific physiological evidence (i.e., "Stress leads to overproduction of glucocorticoid hormones in the developing brain, which leads to a thinner and less connected PFC, the part of the brain responsible for self-control"). Participants were then asked to consider a protagonist who was described as having suffered from childhood stress and who had failed to resist the temptation to shoplift. Some participants were asked to assess the protagonist's control over the behavior, with results showing that perceived control significantly declined in the presence of the biological model. Other participants were asked to assess the likelihood that they would succumb to the same temptation under the same neural condition as the protagonist. Results here showed that the biological model had no effect on perceived self-vulnerability. In Experiment 3b, the effect of biology on self-vulnerability remained nonsignificant, even though the verbal description of brain damage was accompanied by a visual image of modest brain damage. Experiment 3d likewise found a null effect of biology when it was described in genetic rather than neurological terms. Only in Experiment 3c, when the verbal description was accompanied by an image of extensive brain damage and an image of a healthy brain for comparison, did participants perceive that they themselves would be vulnerable to biological causation.
Thus, mere references to neural damage lowered participants' perception of the protagonist's personal control, but participants acknowledged their own vulnerability only when they imagined suffering the same extensive brain damage as the protagonist. As in Experiment 1b, a majority of participants (ranging from 78.5% to 81.0% across Experiments 3a−3d) regarded their superior ability to resist the temptation as due to their intangible qualities.
The preceding results also document a divergence between the perceived control and perceived self-vulnerability measures. That is, the amount of control perceived in another (an indicator of sympathy) does not correspond to an assessment of one's own vulnerability (an indicator of empathy). This distinction is important because, as we subsequently discuss, support for treatment and prevention policies should rise when the transgressor is perceived to lack control and fall when people are unwilling to recognize their own vulnerability. Our results suggest that perception of personal discretion is more malleable when the judgment involves others than when it involves the self. This tendency may be due in part to an individual's belief that they possess more free will than others ([76]) and/or confidence in their superior intangible self.
Experiment 4a presented participants with the competing-pathways model used in Experiments 1a and 1b and varied whether the decision to succumb to temptation was made within seconds or after some deliberation. Results showed that deliberation neutralized participants' perceptions of the causal role of biology. This neutralizing effect was replicated when the biological model was based on genetics (Experiment 4b). These findings reveal the lay belief that conscious deliberation provides a route for the intangible self to override a biological process; these findings also corroborate prior research showing that lay perceptions of free will and responsibility are not undermined by neural evidence that decisions are made unconsciously, as long as the decisions are based on the decision maker's own reasons ([65]; see also [79]).
A key element of our argument is that the implications of biology for marketing's stakeholders are dependent on lay beliefs about the role of biology. To illustrate, we conducted a final study (Experiment 5) that examined the policy consequences of neuroscience. Participants were assigned to one of three conditions (no model, neuroprocess model, or brain-structure model). All participants read, "Research has established that excessive stress during childhood strongly inhibits a person's later ability to exhibit willpower and resist temptation." Those in the no-model condition received no further information; those in the neuroprocess model condition were additionally presented with the competing-pathways description used in Experiments 1a, 1b, and 4a; and those in the brain-structure model condition were additionally presented with the brain damage description and visual brain images used in Experiment 3c. Participants were then asked to rank their affinity toward six public policies aimed at addressing problems associated with a lack of self-control. Three policies aligned with the neurological and developmental nature of self-control described in the biology literature reviewed previously. Results showed that support for science-based policies (e.g., funding social programs to prevent excessive stress in childhood) increased when biology was portrayed in terms of structural damage to the brain but not when portrayed in terms of neural processes.
Our results thus far have focused on individuals who hold a high belief in free will. We now turn to how this belief moderates the effect of biological evidence on lay beliefs of self-control (for the measures, see Web Appendix D). Across studies, we found that, in the face of biological evidence of self-control, people with a lower belief in free will perceived themselves to be relatively more vulnerable to committing a transgression (Experiments 3b and 3d) and were less influenced by deliberation when assessing the control possessed by others (Experiments 4a and 4b). Moreover, in terms of policy preferences, people with a lower belief in free will were uniformly inclined to endorse policies consistent with neuroscience (Experiment 5; the same pattern of results was obtained when political conservatism was substituted for belief in free will as the individual-difference variable).
We also found a consistent effect of religiosity, as measured by self-reported attitudes and behaviors regarding religious practices (for the measures, see Web Appendix D). We calculated a perceived intangibility score based on participants' responses to the perceived locus questions (Experiments 1b−3d). In Experiment 2's investigation of traits, we found that religiosity was significantly associated with the perceived intangibility of integrity and empathy but not intelligence or self-control. Across Experiments 1b and 3a−3d, religiosity was significantly associated with the perceived intangibility of the ability to refrain from temptation despite sharing the same neural processes or the same damaged neural structure as the protagonist.
Lay beliefs influence our understanding of the natural world. The teleological view of evolution, for example, has been characterized as a fundamental feature of the human psyche that thwarts the teaching of natural selection ([27]). Closer to home, a lack of scientific understanding can give rise to false beliefs that thwart the diffusion of welfare-enhancing production technologies such as food irradiation and genetic modification ([104]). In the present case of self-control, the results of our studies tell a story of malleable resistance to biological causation. On the one hand, resistance is sizable and entrenched in the lay belief of mind–body dualism. On the other hand, laypeople's perceptions of others' control and their own vulnerability vary with the portrayal of biological causation, the nature of the transgression, the amount of deliberation by the transgressor, and individual differences across the population. These factors explain why dualism persists: biological causation is rarely salient, many transgressions have a moral tone, and a high belief in free will is pervasive. Thus, it is unlikely that society will readily adopt the neuroscientists' perspective that our brains are best viewed as "complex relay points for innumerable inputs, rather than command centers endowed with true self-determination" ([39], p. 5). Nonetheless, the malleability aspect of our findings is informative and instrumental to marketing's stakeholders, due in part to marketing's ability to shape people's beliefs about biology. Thus, we emphasize malleability in the following discussion of the implications of biology and lay understanding.
We endorse calls for boundary-spanning in marketing research ([60]). This approach has many virtues—including, notably, that marketing can influence its external stakeholders while opening new opportunities for the discipline ([54]). The implications developed in this section sit at the intersection of biology and lay beliefs about biology and inform policy makers, firms, consumers, marketing educators, and academics of different disciplines. Table 2 highlights the key implications and future research questions, to which we turn next.
Graph
Table 2. Implications and Sample Research Questions for Marketing Stakeholders.
| Marketing Stakeholders | Implications | Sample Research Questions |
|---|
| Policy makers | Biology can guide policy directly by informing policy makers of the roots of morbidity. Public appreciation of biology can facilitate policy interventions that address the roots of morbidity. Appreciation of biology by policy makers and the lay public can guide policy choices pertaining to prevention versus treatment. Acceptance of biological causation should enhance support for pharmaceutical treatment of substance abuse. Biology will pose challenges to patent law and legislation, with implications for firms and consumers. Concerns over privacy extend to the domain of biology.
| What are the implications of misalignment between policy makers and the lay public? How can the lay public be nudged toward biology-based policies? What are the relative costs and benefits of treatment versus prevention? How will patent law pertaining to biology influence innovation and consumer welfare? What are the implications for consumer autonomy and privacy?
|
| Firms | Firms should consider their vulnerability to changes in lay belief regarding consumer self-control over ill-advised consumption. Deeper appreciation of biological causation has the potential to transform entire industries, including health care and education. Polygenic scores can be used to tailor product offerings to consumers. Public and policy maker acceptance of biology can expand the child welfare industry. Lay understanding of biology may influence the acceptability of pharmaceutically based self-improvement.
| Will biology lead consumers and legal institutions to hold firms to greater account? Will industry alter its practices and profit models in response to insights regarding biological causation? Will consumers alter their choices based on the same insights? Are consumers ready to embrace new pharmaceutical products geared toward self-improvement?
|
| Consumers | Knowledge of one's own biological constitution should guide consumers toward more realistic objectives and less self-criticism. Appreciation of biological causation should guide attempts to alter the behavior of others in ways that increase the likelihood of success and avoid the frustration of failure.
| How does consumers' improved understanding of their biological constitution change their self-control strategies? How does a better understanding of biological causation alter childrearing practices? Will appreciation of the biological basis of psychological interventions (e.g., mindfulness training) enhance their real and perceived efficacy and acceptance?
|
| Marketing educators | A biological perspective can expand the marketing curriculum on consumer theory. Advances in biology invite marketing students to ponder the question of autonomy. Appreciation of biological causation may guide career preferences among marketing students.
| Are educators willing to adjust their curricula? How does biology alter marketing students' understanding of the relative responsibility of firms, consumers, and government? More specifically, how do marketing students interpret the doctrine of caveat emptor in the context of biological causation?
|
| Scholars | Biology should increasingly be incorporated into the causal models of social sciences. Attribution theory needs to accommodate the new scientific understanding of person and situation. Communication of science to the lay public remains an under-investigated issue that fits marketing scholars' strengths.
| How can biology be most effectively communicated to marketing's other stakeholders? How can biology complement traditional theories of self-control?
|
Among marketing's many stakeholders, a principal audience for biology is policy makers. Our preceding discussion indicates that the foremost policy application is the treatment of morbidities, which we discuss first. Then, we note policy implications as they apply to jurisprudence and privacy.
Policy makers exhibit uneven recognition of the role biology plays in self-control. At one end, we sense a growing and relatively uncontroversial acceptance of the role played by biology in imprudent adolescent behavior. As [71] argue, the immature PFCs of teens that prompt risky consumption serve as a basis for regulating such activity. More recently, research has begun to examine the neural effects of vaping on the teenage brain—especially with regards to memory, learning, focus, and impulse control ([33])—which would carry evident regulatory implications if confirmed. In addition, policy makers and social-welfare organizations should include evidence from biology in public service announcements that warn against unwise consumption.
Beyond this protected class of consumers, some specific domains of consumption should also be on policy makers' radar due to their sizable impact on consumer welfare.[11] Our empirical results suggest that acceptance of the causal role of biology in obesity is tentative (see Experiments 1a, 4a, and 4b). Insensitivity to biology in alcohol and substance abuse is reflected in halting utilization of effective pharmaceutical therapies ([29]). For example, in the case of alcohol addiction, 12-step programs that claim to address underlying "character defects" are held in high esteem, despite their uncertain efficacy ([29]). In the case of opioid addiction, underutilization has been attributed in part to skepticism among the lay public, therapists, and the judiciary regarding the appropriateness of a pharmaceutical intervention to address the "underlying" cause of the problem. Some stakeholders maintain that the use of antiaddiction drugs serves only to replace one disorder with another, so they favor abstinence over pharmaceutical intervention, despite the proven success of the latter ([66]; [80]). A biological perspective should shift perceptions of the acceptability and legitimacy of competing interventions.
At the broadest level, consider the case of trait self-control, which exhibits the least appreciation of biology. One analysis holds that personal decisions are the leading cause of premature death and that "individuals have a great deal of control over their own mortality" ([42], p. 1345). The empirical result reinforces the tragedy of poor self-control, but the conclusion regarding control fails to consider the roots of those ill-advised personal decisions. We have argued that trait self-control serves as an antecedent to acute manifestations of self-control failure. We argue here that trait self-control has the broadest policy implications.
Biology shows that trait self-control can be diminished by childhood adversity, and social science shows that self-control in early childhood serves as a predictor of adult health, wealth, and criminality ([59]). More specifically, childhood adversity and the unhealthy lifestyles that may ensue have been associated with a range of morbidities that in turn lead to seemingly intractable economic, racial, and ethnic disparities in welfare and opportunities that are measurable at the level of gross domestic product ([85]). The mediating influence of neurological development is reflected in evidence that adverse childhood experiences hinder the developing brain, as previously discussed in the biology section.
Policy makers at the highest levels of the executive and legislative branches devise policies that influence the developmental trajectory of trait self-control (e.g., via social support). It is likely that many of these high-level decision makers share the lay public's beliefs about personal control. Thus, biology can enlighten policy directly if policy makers gain a deeper understanding of the biological roots of various morbidities. Such enlightenment would illuminate both the need for costly interventions and the nature of those interventions. However, policy makers are moored to the voting public. Changes in policy can be accelerated by public support and can be thwarted by its absence. Beliefs regarding the appropriateness of government action have been associated with lay beliefs about personal control and self-sufficiency ([26]; [103]). Lay support for intervention should rise with lay acceptance of biological causation.
Toward this same end, public confidence would be enhanced by scientific evidence of successful intervention. For example, a PFC already damaged by adverse childhood experiences is not beyond repair ([21]). It is likely that broad-based prevention policies have higher up-front costs than treatment policies, but the public should also be encouraged by the compelling relationship between the earliness of intervention and the return on investment ([36]; [45]; see also [37]).
We ultimately argue that marketers can play a large role in educating the lay public about biological causation. In so doing, marketers can tackle questions that are amenable to research. For example, whereas economic research may examine the returns on prevention versus remediation, marketers can explore the public's attitude toward each, both a priori and following exposure to biology-based explanations. In the face of continued public resistance, future research could examine tactics that nudge the public toward policies that would otherwise raise objections related to personal freedom and thereby overcome the political divide observed in Experiment 5's probe of policy preferences ([91]).
The increasing prominence of bioscience has prompted concern over the demise of personal responsibility, with transgressors asserting, "My brain made me do it." We expect that, in pursuit of social order, society will retain its view of personal responsibility ([28]). However, it is worth contemplating whether there will be a shift in the "reasonable person" criterion applied in the law, which currently treats physical and mental limitations differently. If mental limitations come to be viewed as documentable physical limitations, forbearance could climb.
Beyond the realm of personal responsibility, the rise of bioscience will also lead to ethical and regulatory issues, as exemplified by the attempt to patent human genes ([15]). Although the U.S. Supreme Court has ruled against such patents, the patenting of gene-based risk-analysis methods remains viable at this time. The ultimate disposition of patent law and its implications for innovation and consumer welfare remain open questions.
Policy makers, firms, and consumers have long wrestled with the problem of digital privacy. The problem has been amplified by the scope of "Big Data" and the ability of artificial intelligence to exploit it, with some fearing that autonomy itself is threatened by industry's resultant ability to characterize people's personalities and inclinations at such a granular level that behavior can be accurately predicted and surreptitiously manipulated ([35]). More recently, biometric techniques have provided marketers with additional opportunities to acquire and maintain customers ([22]). Although most forms of these biometric data (e.g., heart rate, eye movements) reside beyond our scope, genetic information resides squarely within it. Whereas some individuals may be quite willing to share their code ([47]), others may be less so. The public generally appears willing to share their genetic code for the purpose of exploring their ancestry but is appropriately concerned at the possibility that the same information could be used as a basis to deny employment or health insurance. Aversion to sharing one's genetic code may be especially high if the code is used to reveal dimensions of one's psychological self, as these dimensions speak to our very essence. Thus, the efforts made by marketers to solve the problem of digital privacy ([18]) should be applied to personal biology to prevent unauthorized collection, sharing, and use. Likewise, the intersection of privacy and autonomy needs to be examined from both a digital and biological perspective ([100]).
Biology is poised to alter the landscape of a variety of businesses. The implications for managers can be distinguished in terms of who is potentially threatened by bioscience and who can leverage it.
Industries that have been complicit in human misery are likely to be threatened by the public's increased understanding of bioscience. Consider again the case of hyperpalatable foods examined by [105]. The lay view of free will allows these firms to place responsibility for obesity and ill health at the feet of the consumer. An addiction model alters the playing field. Forward-looking firms should at least consider the possibility that lay beliefs regarding control and responsibility will shift with advances in biology, and firms should prepare for a more contentious relationship between policy and practice. Firms that market risky products to adolescents may be especially threatened by the public's understanding of bioscience. Analogous to how climate change influences organizational behavior, biology can also have sweeping influences on business practices. Thus, researchers who focus on managerial behavior can assess the extent to which firms are persuaded by the threats of biology and are inclined to adopt forward-looking policies.
The wellness industry, too, faces threats from the public's understanding of bioscience. Industries that have positioned themselves as promoting well-being will need to confront the implications of bioscience for the believability of their claims. For example, the massive weight-loss industry can report little sustained success, as long-term weight reduction is rare ([25]).
Health care more generally could face disruption. The U.S. health care model uses symptoms to diagnose a problem; the health care industry then profits from treatment. Genetics may change this traditional model because polygenic scores may help predict a problem and allow for a different type of intervention (e.g., prevention). Such a fundamental shift would alter not only the patient's clinical experience and health care professional's work experience but also the profit model of health care providers ([48]) and create downstream consequences for the insurance industry.
Finally, the tumultuous education industry could face additional threats from biology. Educational institutions, especially those at the elite level, often tout the success of their students as a competitive differentiator. However, research shows that the superior outcomes of high-quality schools may be based on student selection. Specifically, [73] alleges that, after controlling for genetic effects, achievement is little affected by the quality of schools. Thus, if parents recognize that expensive schools add little value, private institutions, beginning with preschool, could be at risk, as could ancillary businesses that promise to increase the competitiveness of applicants to these institutions.
Future research could examine whether consumer preferences and choices are altered by an increased understanding of biology. For example, in the case of health care, it remains to be seen whether consumers prefer to undertake actions to preempt the occurrence of ailments or to treat ailments as they arise. Moreover, in the case of education, a natural question is whether and how consumers are willing to trade off prestige against actual educational value provided to students.
In terms of leveraging bioscience, much depends on how much consumers are willing to reveal. As we have noted, demands for confidentiality are likely to be especially high for biological characteristics that speak to consumers' essence or reveal their vulnerabilities. Insofar as consumers are willing to reveal polygenic indicators of psychological traits, for example, opportunities emerge for firms that provide a "matching" benefit (e.g., job placement, personal relationships). However, to be a viable contender in the matchmaking business, the genetic indicators need to improve to the point that they outperform traditional profiling methods.
If widespread public appreciation of biology helps strengthen the hand of policy makers, increased funding would benefit providers of child-welfare services in both the public and private sectors. These providers include educational, counseling, and family-support services.
The pharmaceutical industry may be able to address the underlying causality of "psychological" shortcomings by developing and marketing nootropic drugs. Such drugs are now widely available for problems involving anxiety, hyperactivity, and depression. In these cases, the lay public is receptive to biological causation, and the objective of the drug is to move the individual toward normal functioning. For many traits, however, the concern is that the pharmaceutical industry could market nootropic drugs not to enable normal functioning but to enhance the trait above its natural level. [77] found that people are reluctant to improve traits that are deemed essential, particularly when the alteration enhances the individual's natural ability. A question for future research is whether biological causation influences consumers' willingness to alter a trait. We took a tentative step toward addressing this question in the context of self-control (available from the authors). We found that biological causation of self-control increased participants' willingness to use a pharmaceutical to enhance their self-control. Future research needs to examine traits that are even more essential to a person's identity, such as integrity and generosity.
Many of the implications we have described converge on the consumer, but we note several others. First, knowledge of one's own biological constitution should have large effects on one's own behavior. Biology is not destiny, but swimming against the genetic tide is no simple matter. A polygenic score for BMI provides a salient example. Consumers who are not predisposed to a high BMI should feel encouraged to lower their weight if it becomes problematic; however, consumers who are predisposed to a high BMI should understand the nature of their battle. In terms of their time and monetary budgets, such information could guide consumers toward alternative routes to good health ([58]). At a minimum, such self-knowledge should challenge pervasive lay theories about the roles of diet and exercise ([56]).
Second, appreciation of biological causation should influence attempts to alter the behavior of other consumers. Lay theories of self-control have previously been invoked to explain the likelihood that parents will engage in "character-building" policies ([64]), but these lay theories revolve around a belief in the inherent malleability of human self-control. Research in genetics addresses even more fundamental lay beliefs regarding the relative influences of nature and nurture on child development. [73] refers to "the nature of nurture" when describing the potential for misattribution. Consider reading behavior. Parents who read to their children may produce adults who like to read and are good readers, which may be taken as a causal effect of nurture. However, the heritability of traits bolsters the less intuitive role of nature in that parents and children may share a biological affinity for reading. The implication for consumer well-being is that happiness ensues when a parent's good intentions align with a child's dispositions; frustration and conflict may ensue otherwise.
Taken together, a genetic perspective on biology can be beneficial to consumers by guiding them away from instrumental goals that are at odds with their biological predispositions, the result of which should be improved levels of success and contentment and perhaps even a greater sense of authenticity. Further, a genetic perspective prompts less criticism of self and others because failure to achieve a goal can be more appropriately attributed to a biological predisposition (rather than a lack of character); on the flip side, a genetic perspective can also instill a greater sense of accomplishment when success is achieved despite an opposing biological predisposition.
Unique consumer implications of biology can also be derived from neuroscience. An exciting development is the potential contribution that mindfulness can make to consumer and societal well-being ([ 2]). For example, mindfulness meditation has been examined as a weight-loss treatment ([11]; [68]). It would be premature to make definitive statements about the efficacy of mindfulness training as an intervention, but we highlight several implications as they pertain to the present discussion. First, although the training itself can be characterized as a behavioral intervention, biological research has implicated a wide variety of potential neurological mediators ([38]; [89])—again illustrating the potential for biology and social science to inform and enrich each other. Second, regarding our discussion of impulsivity, recent research suggests that mindfulness training may temper emotional reactivity directly by dampening amygdala reactivity or indirectly through its effect on PFC–amygdala connectivity ([46]). Third, dualism appears impervious to this biological evidence for some devotees of meditation, as they are inclined to interpret the evidence in terms of the mind's influence on the physical brain ([39]).
Ample research opportunities arise from the implications of biology on consumer well-being. For example, we find that scientific biological knowledge of self-control can shift lay beliefs of self-control, but its effect is behavior- and segment-dependent. Future research can examine whether the effect of scientific biological knowledge would influence actual self-control behavior and whether that effect would be behavior- or segment-dependent.
In addition, the extent to which knowledgeable parents persist in parenting practices that go against their child's biological dispositions—and how to convince them to reset their expectations—are research questions of consequence. Similarly, it would be worthwhile for future research to investigate the extent to which appreciation of the biological basis of psychological interventions, such as mindfulness training, influences the real and perceived efficacy of the practice, particularly among those who do not hold strong dualistic beliefs and those who may be initially disinclined to adopt the practice.
Marketing education naturally reflects academic marketing research; consequently, the textbook approach to consumer behavior is dominated by the social sciences, particularly psychology. However, just as psychology itself is gravitating toward biology, we argue that marketing education should broaden "consumer theory" to include a biological perspective.
We further maintain that a step in the direction of biology has benefits beyond a more complete understanding of consumers. We increasingly expect marketing students to emerge from their studies with a set of analytic skills. A comprehensive education would also leave students with a refined set of values. Marketing educators often address values under the heading of ethics, but we argue that students should be invited to consider biology and its implications for autonomy. Biology shifts the focus from the extent to which marketing diminishes consumers' autonomy (see the preceding discussion) to the amount of autonomy consumers had in the first place. As such, it goes well beyond the specific domain of ethics. The consumer and social welfare implications could be raised throughout the marketing education curriculum. To incorporate biology successfully into marketing education, marketing educators need not be biologists. The mechanistic details are far less important than the larger perspective of biological causation and its implications.
Finally, marketing educators produce private-sector employees. Students who seek employment in the tobacco industry are aware of the physically addictive nature of nicotine. Students who wish to become brand managers at food companies may be less cognizant of the neurological effects of hyperpalatable foods. Informed students can plan their careers accordingly.
Self-control has been a central concern to scholars of marketing (e.g., [ 1]; [23]; [94]) and associated disciplines (for a motivational view, see [ 3]]; for a cognitive view, see [13]]; for an economic view, see [90]]). Likewise, policy-oriented consumer research has long examined and advocated for messaging strategies that reduce self-defeating consumption. Such research is typically targeted at a specific behavior, such as tobacco, alcohol, or unhealthy food consumption ([69]). Some researchers have taken a broader approach by examining a class of behaviors (e.g., addiction; [49]) or a segment of consumers (e.g., adolescents; [71]). We endorse all these approaches but further argue that self-defeating behavior may have roots in neurological development, emerge across a spectrum of behaviors, and span the lifetime. As consumer researchers embrace the biological view, an important task is to educate both policy makers and the lay public about the biological antecedents of self-defeating behavior. Fortunately, this undertaking leverages marketing's communication and persuasion skills, which have proven effective at altering lay understanding of science ([104]). The endeavor also addresses the appeal for marketing to focus its efforts on matters of relevance and consequence ([61]).
We also argue that this path is paved with opportunities for marketing scholarship. In the realm of communication and persuasion, our research reveals that biological causation runs counter to deep-seated views of self-control and raises very delicate questions about autonomy. How to convey the science and its human welfare and policy implications without prompting reactance is a multifaceted research question.
At a more general level we concur that incorporating biology into marketing's conceptualization of self-control can create new opportunities for theory ([101]). The present research was inspired by the same question that made attribution theory a dominant force in social psychology: On what bases do people draw inferences about the cause of others' behavior? Most research in attribution theory draws a distinction between personal and situational causation, with a focus on the role of intentionality. However, this distinction is silent on the antecedent causes of intentionality. That is, even when a behavior is attributed to the actor, latter-day attribution models have been described as "inert" with respect to characterizing people's beliefs about the underlying reasons for the behavior ([55]).
Revolutions in neuroscience and genetics promise to invigorate this question. Marketing scholars can play a central role not just as observers but as active participants. A first step is for marketing scholars to internalize the view "that our minds are biologically based, rooted in banal physiological processes, and subject to the laws of nature" ([39], p. 3). Once internalized, new meaning is given to the person–situation dichotomy that has dominated attribution theory for 70 years. A biological perspective fundamentally alters the concept of "person" by highlighting the individual's genetic blueprint and the neural processes and structures that drive behavior. A biological perspective fundamentally alters the concept of environment by closing the gap between the biological and psychological forces acting on the individual. Just as malnutrition alters brain development, so too does stress. A biological perspective greatly expands the notion of "situation" and its influences on behavior, including the impact of environment (e.g., adverse childhood experiences) on neural processes and structures, gene–environment correlations (i.e., self-selection of environment based on one's genetic constitution), and the many gene–environment interactions.
We have argued that many of marketing's individual external stakeholders have much to gain from incorporating a biological perspective into their beliefs and practices. At a broader level, we are convinced that adoption of a biological perspective carries important implications for individual well-being, social equality, and national prosperity. Our empirical findings demonstrate, however, that biological causation is neither intuitive nor attractive to the lay mind. However, that gap between lay and scientific understanding is mutable, and we see marketing scholars as being a catalyst for change. We do not underestimate the communication and persuasion challenges ahead, but we also do not underestimate marketing's abilities. A first step is for marketing itself to become knowledgeable and comfortable with the role biology plays in self-control. By doing so, it can prepare itself for the process of change while adding explanatory depth to its long list of impressive behavioral findings.
Beyond marketing's immediate stakeholders, we note the implications of biology for social conduct. A belief in a mechanistic world has been associated with reduced desire for retribution in the realm of justice ([84]). We contend that findings pertaining to justice can generalize to other contexts. A truer understanding of the biological underpinnings of behavior should reduce moral scolding and enhance empathy toward those who exhibit poor self-control and other "failings"—including depression, irresoluteness, social awkwardness, infidelity, and even a lack of empathy—for which the biological and psychological causes are mistakenly dissociated. As the understanding of biological causation increases, so too should comity, mutual understanding, and societal well-being.
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242920983271 - Consumer Self-Control and the Biological Sciences: Implications for Marketing Stakeholders
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242920983271 for Consumer Self-Control and the Biological Sciences: Implications for Marketing Stakeholders by Yanmei Zheng and Joseph W. Alba in Journal of Marketing
Footnotes 1 Deborah MacInnis
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Yanmei Zheng https://orcid.org/0000-0002-8502-6191
5 Online supplement: https://doi.org/10.1177/0022242920983271
6 In accordance with our empirical studies, dualistic thinking refers to the lay belief that the tangible brain is distinct from the intangible mind or soul.
7 System 1 judgments are relatively automatic and based on intuition, in contrast to system 2 judgments, which are more logical and deliberative.
8 It stands to reason that biological causation also underlies a range of dependencies, from sugar and caffeine to gambling and pornography.
9 Heritability is a population statistic defined as the proportion of total variance that is attributable to genetic factors. Heritability does not imply genetic determinism ([96]) or immutability ([74]). As an example of the latter, the single-gene disease phenylketonuria is 100% heritable; however, affected individuals can avoid the disease by eliminating phenylalanine from their diets.
[105] did find that mentioning the impact of the social risk factors on brain development reduced perceptions of personal control, but only when the risk factors explained 100% of the variance in behavior (which is unprecedented) and participants possessed a low belief in free will (which is infrequent).
By one estimate, nearly one in two adults will be obese by 2030 ([97]), with costs in 2016 estimated to be $1.72 trillion—or 9.3% of U.S. gross domestic product ([98]). The costs of drug and alcohol abuse have recently been estimated to hover around $500 billion ([95]).
References Alba Joseph W., Williams Elanor F. (2013), "Pleasure Principles: A Review of Research on Hedonic Consumption," Journal of Consumer Psychology, 23 (1), 2–18.
Bahl Shalini, Milne George R., Ross Spencer M., Mick David Glen, Grier Sonya A., Chugani Sunaina K., et al. (2016), "Mindfulness: Its Transformative Potential for Consumer, Societal, and Environmental Well-Being," Journal of Public Policy & Marketing, 35 (2), 198–210.
Baumeister Roy F., Vohs Kathleen D., Tice Dianne M. (2007), "The Strength Model of Self-Control," Current Directions in Psychological Science, 16 (December), 351–55.
Bechara Antoine, Damasio Antonio R. (2005), "The Somatic Marker Hypothesis: A Neural Theory of Economic Decision," Games and Economic Behavior, 52 (2), 336–72.
Belsky Jay, Pluess Michael. (2009), "Beyond Diathesis Stress: Differential Susceptibility to Environmental Influences," Psychological Bulletin, 135 (6), 885–908.
Berridge Kent C., Robinson Terry E. (2016), "Liking, Wanting, and the Incentive-Sensitization Theory of Addiction," American Psychologist, 71 (8), 670–79.
Black Maureen M., Walker Susan P., Fernald Lia C.H., Andersen Christopher T., DiGirolamo Ann M., Lu Chunling, et al. (2017), "Early Childhood Development Coming of Age: Science Through the Life Course," The Lancet, 389 (10064), 77–90.
Blair R. James R. (2013), "The Neurobiology of Psychopathic Traits in Youths," Nature Reviews Neuroscience, 14 (11), 786–99.
Bloom Paul. (2004), Descartes' Baby: How the Science of Child Development Explains What Makes Us Human. New York: Basic Books.
Brownell Kelly D., Timothy Walsh B., eds. (2017), Eating Disorders and Obesity: A Comprehensive Handbook. New York: Guilford Publications.
Carrière Kimberly, Khoury Bassam, Günak Mia M., Knäuper Bärbel A. (2018), "Mindfulness-Based Interventions for Weight Loss: A Systematic Review and Meta-Analysis," Obesity Reviews, 19 (2), 164–77.
Carter Adrian, Hendrikse Joshua, Lee Natalia, Yücel Murat, Verdejo-Garcia Antonio, Andrews Zane B., et al. (2016), "The Neurobiology of 'Food Addiction' and Its Implications for Obesity Treatment and Policy," Annual Review of Nutrition, 36, 105–28.
Carver Charles S., Scheier Michael F. (1981), Attention and Self-Regulation: A Control-Theory Approach to Human Behavior. New York: Springer-Verlag.
Casey B.J., Somerville Leah H., Gotlib Ian H., Ayduk Ozlem, Franklin Nicholas T., Askren Mary K., et al. (2011), "Behavioral and Neural Correlates of Delay of Gratification 40 Years Later," Proceedings of the National Academy of Sciences, 108 (September), 14998–15003.
Caulfield Timothy, Cook-Deegan Robert M., Kieff F. Scott, Walsh John P. (2006), "Evidence and Anecdotes: An Analysis of Human Gene Patenting Controversies," Nature Biotechnology, 24 (9), 1091–94.
Cecil Joanne E., Tavendale Roger, Watt Peter, Hetherington Marion M., Palmer Colin N.A. (2008), "An Obesity-Associated FTO Gene Variant and Increased Energy Intake in Children," New England Journal of Medicine, 359 (24), 2558–66.
Craig Arthur D. (2002), "How Do You Feel? Interoception: The Sense of the Physiological Condition of the Body," Nature Reviews Neuroscience, 3 (8), 655–66.
Cui Tony Haitao, Ghose Anindya, Halaburda Hanna, Iyengar Raghuram, Pauwels Koen, Sriram S., et al. (2021), "Informational Challenges in Omnichannel Marketing: Remedies and Future Research," Journal of Marketing, 85 (1), 103–20.
Damasio Antonio R. (1994), Descartes' Error: Emotion, Reason, and the Human Brain. New York: Putnam.
Daviet Remi, Nave Gideon, Wind Jerry. (2021), "Genetic Data: Potential Uses and Misuses in Marketing," Journal of Marketing(published online February 10), https://doi.org/10.1177/0022242920980767.
Diamond Adele, Lee Kathleen. (2011), "Interventions Shown to Aid Executive Function Development in Children 4 to 12 Years Old," Science, 333 (6045), 959–64.
Du Rex Yuxing, Netzer Oded, Schweidel David A., Mitra Debanjan. (2021), "Capturing Marketing Information to Fuel Growth," Journal of Marketing, 85 (1), 163–83.
Faber Ronald J., Vohs Kathleen D. (2011), "Self-Regulation and Spending: Evidence from Impulsive and Compulsive Buying," in Handbook of Self-Regulation: Research, Theory, and Applications, 2nd ed.,Vohs Kathleen D., Baumeister Roy F., eds. New York: Guilford Press.
Felitti Vincent J., Anda Robert F., Nordenberg Dale, Williamson David F., Spitz Alison M., Edwards Valerie, et al. (1998), "Relationship of Childhood Abuse and Household Dysfunction to Many of the Leading Causes of Death in Adults: The Adverse Childhood Experiences (ACE) Study," American Journal of Preventive Medicine, 14, 245–58.
Fildes Alison, Charlton Judith, Rudisill Caroline, Littlejohns Peter, Prevost A. Toby, Gulliford Martin C. (2015), "Probability of an Obese Person Attaining Normal Body Weight: Cohort Study Using Electronic Health Records," American Journal of Public Health, 105, e54–e59.
Frank H. Robert. (2013), "Mixing Freedoms in a 32-Ounce Soda," The New York Times (March 23), https://www.nytimes.com/2013/03/24/business/soda-restrictions-and-a-clash-of-two-freedoms.html.
Galli Leonardo M.G., Meinardi Elsa N. (2011), "The Role of Teleological Thinking in Learning the Darwinian Model of Evolution," Evolution: Education and Outreach, 4 (1), 145–52.
Gazzaniga Michael S. (2011), Who's in Charge? Free Will and the Science of the Brain. New York: Ecco.
Glaser Gabrielle. (2015), "The Irrationality of Alcoholics Anonymous," The Atlantic (April), https://www.theatlantic.com/magazine/archive/2015/04/the-irrationality-of-alcoholics-anonymous/386255/.
Grabe Shelly, Ward L. Monique, Hyde Janet Shibley. (2008), "The Role of the Media in Body Image Concerns Among Women: A Meta-Analysis of Experimental and Correlational Studies," Psychological Bulletin, 134, 460–76.
Hackman Daniel A., Farah Martha J., Meaney Michael J. (2010), "Socioeconomic Status and the Brain: Mechanistic Insights from Human and Animal Research," Nature Reviews Neuroscience, 11 (9), 651–59.
Haidt Jonathan. (2001), "The Emotional Dog and Its Rational Tail: A Social Intuitionist Approach to Moral Judgment," Psychological Review, 108 (4), 814–34.
Hamilton Jon. (2019), "How Vaping Nicotine Can Affect a Teenage Brain," NPR (October 10), https://www.npr.org/sections/health-shots/2019/10/10/768588170/how-vaping-nicotine-can-affect-a-teenage-brain.
Hanson Jamie L., Chung Moo K., Avants Brian B., Rudolph Karen D., Shirtcliff Elizabeth A., Gee James C., et al. (2012), "Structural Variations in Prefrontal Cortex Mediate the Relationship between Early Childhood Stress and Spatial Working Memory," Journal of Neuroscience, 32 (23), 7917–25.
Harari Yuval N. (2020), "Rebellion of the Hackable Animals," The Wall Street Journal (May 1), https://www.wsj.com/articles/rebellion-of-the-hackable-animals-11588352123.
Heckman James J. (2007), "The Economics, Technology, and Neuroscience of Human Capability Formation," Proceedings of the National Academy of Sciences, 104, 13250–55.
Heckman James, Pinto Rodrigo, Savelyev Peter. (2013), "Understanding the Mechanisms Through Which an Influential Early Childhood Program Boosted Adult Outcomes," American Economic Review, 103 (6), 2052–86.
Hölzel Britta K., Lazar Sara W., Gard Tim, Schuman-Olivier Zev, Vago David R., Ott Ulrich. (2011), "How Does Mindfulness Meditation Work? Proposing Mechanisms of Action from a Conceptual and Neural Perspective," Perspectives on Psychological Science, 6 (6), 537–59.
Jasanoff Alan. (2018), The Biological Mind: How Brain, Body, and Environment Collaborate to Make Us Who We Are. New York: Basic Books.
Kahneman Daniel. (2011), Thinking, Fast and Slow. New York: Farrar, Straus and Giroux.
Kandel Eric R. (2018), The Disordered Mind: What Unusual Brains Tell Us About Ourselves. New York: Farrar, Straus & Giroux.
Keeney Ralph L. (2008), "Personal Decisions Are the Leading Cause of Death," Operations Research, 56 (6), 1335–47.
Kim Pilyoung, Evans Gary W., Angstadt Michael, Ho S. Shaun, Sripada Chandra S., Swain James E., et al. (2013), "Effects of Childhood Poverty and Chronic Stress on Emotion Regulatory Brain Function in Adulthood," Proceedings of the National Academy of Sciences, 110 (46), 18442–47.
Knopik Valerie S., Neiderhiser Jenae M., DeFries John C., Plomin Robert. (2017), Behavioral Genetics, 7th ed. New York: Macmillan.
Knudsen Eric I., Heckman James J., Cameron Judy L., Shonkoff Jack P. (2006), "Economic, Neurobiological, and Behavioral Perspectives on Building America's Future Workforce," Proceedings of the National Academy of Sciences, 103, 10155–62.
Kral Tammi R.A., Schuyler Brianna S., Mumford Jeanette A., Rosenkranz Melissa A., Lutz Antoine, Davidson Richard J. (2018), "Impact of Short- and Long-Term Mindfulness Meditation Training on Amygdala Reactivity to Emotional Stimuli," NeuroImage, 181, 301–13.
Levy Samuel, Sutton Granger, Ng Pauline C., Feuk Lars, Halpern Aaron L., Walenz Brian P., et al. (2007), "The Diploid Genome Sequence of an Individual Human," PLoS Biology, 5 (10), e254–e54.
Lewis Cathryn M., Vassos Evangelos. (2017), "Prospects for Using Risk Scores in Polygenic Medicine," Genome Medicine, 9 (1), 1–3.
Litt Ab, Pirouz Dante M., Shiv Baba. (2012), "Neuroscience and Addictive Consumption," in Transformative Consumer Research for Personal and Collective Well-Being, Mick David Glen, Pettigrew Simone, Pechmann Cornelia, Ozanne Julie L., eds. New York: Routledge, 523–42.
Locke Adam E., Kahali Bratati, Berndt Sonja I., Justice Anne E., Pers Tune H., Day Felix R., et al. (2015), "Genetic Studies of Body Mass Index Yield New Insights for Obesity Biology," Nature, 518 (7538), 197–206.
Loewenstein George F., Weber Elke U., Hsee Christopher K., Welch Ned. (2001), "Risk as Feelings," Psychological Bulletin, 127, 267–86.
Lupien Sonia J., McEwen Bruce S., Gunnar Megan R., Heim Christine. (2009), "Effects of Stress Throughout the Lifespan on the Brain, Behaviour and Cognition," Nature Reviews Neuroscience, 10 (6), 434–45.
MacInnis Deborah J. (2011), "A Framework for Conceptual Contributions in Marketing," Journal of Marketing, 75 (4), 136–54.
MacInnis Deborah J., Morwitz Vicki G., Botti Simona, Hoffman Donna L., Kozinets Robert V., Lehmann Donald R., et al. (2020), "Creating Boundary-Breaking, Marketing-Relevant Consumer Research," Journal of Marketing, 84 (2), 1–23.
Malle Bertram F. (1999), "How People Explain Behavior: A New Theoretical Framework," Personality and Social Psychology Review, 3, 23–48.
McFerran Brent, Mukhopadhyay Anirban. (2013), "Lay Theories of Obesity Predict Actual Body Mass," Psychological Science, 24 (8), 1428–36.
Meier Barry. (2018), Pain Killer: An Empire of Deceit and the Origin of America's Opioid Epidemic. New York: Random House.
Mensinger Janell L., Calogero Rachel M., Stranges Saverio, Tylka Tracy L. (2016), "A Weight-Neutral Versus Weight-Loss Approach for Health Promotion in Women with High BMI: A Randomized-Controlled Trial," Appetite, 105 (1), 364–74.
Moffitt Terrie E., Arseneault Louise, Belsky Daniel, Dickson Nigel, Hancox Robert J., Harrington HonaLee, et al. (2011), "A Gradient of Childhood Self-Control Predicts Health, Wealth, and Public Safety," Proceedings of the National Academy of Sciences, 108, 2693–98.
Moorman Christine, van Harald J., Heerde C., Moreau Page, Palmatier Robert W. (2019a), "Challenging the Boundaries of Marketing," Journal of Marketing, 83 (5), 1–4.
Moorman Christine, van Harald J., Heerde C., Moreau Page, Palmatier Robert W. (2019b), "JM as a Marketplace of Ideas," Journal of Marketing, 83 (1), 1–7.
Moschis George P. (2007), "Stress and Consumer Behavior," Journal of the Academy of Marketing Science, 35 (3), 430–44.
Moss Michael. (2013), Salt, Sugar, Fat: How the Food Giants Hooked Us. New York: Random House.
Mukhopadhyay Anirban, Yeung Catherine W.M. (2010), "Building Character: Effects of Lay Theories of Self-Control on the Selection of Products for Children," Journal of Marketing Research, 47 (2), 240–50.
Nahmias Eddy, Shepard Jason, Reuter Shane. (2014), "It's OK If 'My Brain Made Me Do It': People's Intuitions About Free Will and Neuroscientific Prediction," Cognition, 133, 502–16.
National Institute on Drug Abuse (2018), "Medications to Treat Opioid Use Disorder," National Institute on Drug Abuse Report.
Noble Kimberly G., Houston Suzanne M., Brito Natalie H., Bartsch Hauke, Kan Eric, Kuperman Joshua M., et al. (2015), "Family Income, Parental Education and Brain Structure in Children and Adolescents," Nature Neuroscience, 18 (5), 773–78.
O'Reilly Gillian A., Cook Lauren, Spruijt-Metz Donna, Black David S. (2014), "Mindfulness-Based Interventions for Obesity-Related Eating Behaviours: A Literature Review," Obesity Reviews, 15 (6), 453–61.
Palmedo P. Christopher, Dorfman Lori, Garza Sarah, Murphy Eleni, Freudenberg Nicholas. (2017), "Countermarketing Alcohol and Unhealthy Food: An Effective Strategy for Preventing Noncommunicable Diseases? Lessons from Tobacco," Annual Review of Public Health, 38, 119–44.
Park Anne T., Leonard Julia A., Saxler Patricia K., Cyr Abigail B., Gabrieli John D.E., Mackey Allyson P. (2018), "Amygdala-Medial Prefrontal Cortex Connectivity Relates to Stress and Mental Health in Early Childhood," Social Cognitive and Affective Neuroscience, 13 (4), 430–39.
Pechmann Cornelia, Levine Linda, Loughlin Sandra, Leslie Frances. (2005), "Impulsive and Self-Conscious: Adolescents' Vulnerability to Advertising and Promotion," Journal of Public Policy & Marketing, 24 (2), 202–21.
Plassmann Hilke, Venkatraman Vinod, Huettel Scott, Yoon Carolyn. (2015), "Consumer Neuroscience: Applications, Challenges, and Possible Solutions," Journal of Marketing Research, 52 (4), 427–35.
Plomin Robert. (2018), Blueprint: How DNA Makes Us Who We Are. Cambridge, MA: MIT Press.
Plomin Robert, DeFries John C., Knopik Valerie S., Neiderhiser Jenae M. (2016), "Top 10 Replicated Findings from Behavioral Genetics," Perspectives on Psychological Science, 11 (1), 3–23.
Polderman Tinca J.C., Benyamin Beben, Leeuw Christiaan A. De, Sullivan Patrick F., Bochoven Arjen Van, Visscher Peter M., et al. (2015), "Meta-Analysis of the Heritability of Human Traits Based on Fifty Years of Twin Studies," Nature Genetics, 47 (7), 702–09.
Pronin Emily, Kugler Matthew B. (2010), "People Believe They Have More Free Will Than Others," Proceedings of the National Academy of Sciences, 107, 22469–74.
Riis Jason, Simmons Joseph P., Goodwin Geoffrey P. (2008), "Preferences for Enhancement Pharmaceuticals: The Reluctance to Enhance Fundamental Traits," Journal of Consumer Research, 35 (3), 495–508.
Rindfleisch Aric, Burroughs James E., Denton Frank. (1997), "Family Structure, Materialism, and Compulsive Consumption," Journal of Consumer Research, 23 (4), 312–25.
Rose David, Buckwalter Wesley, Nichols Shaun. (2017), "Neuroscientific Prediction and the Intrusion of Intuitive Metaphysics," Cognitive Science, 41 (2), 482–502.
Rosenberg Tina. (2016), "Staying Sober After Treatment Ends," The New York Times (February 9), https://opinionator.blogs.nytimes.com/2016/02/09/staying-sober-after-treatment-ends/.
Ruvio Ayalla, Somer Eli, Rindfleisch Aric. (2014), "When Bad Gets Worse: The Amplifying Effect of Materialism on Traumatic Stress and Maladaptive Consumption," Journal of the Academy of Marketing Science, 42 (1), 90–101.
Sapolsky Robert M. (2017), Behave: The Biology of Humans at Our Best and Worst. New York: Penguin.
Schulte Erica M., Smeal Julia K., Lewis Jessi, Gearhardt Ashley N. (2018), "Development of the Highly Processed Food Withdrawal Scale," Appetite, 131, 148–54.
Shariff Azim F., Greene Joshua D., Karremans Johan C., Luguri Jamie B., Clark Cory J., Schooler Jonathan W., et al. (2014), "Free Will and Punishment: A Mechanistic View of Human Nature Reduces Retribution," Psychological Science, 25, 1563–70.
Shonkoff Jack P., Garner Andrew S., Siegel Benjamin S., Dobbins Mary I., Earls Marian F., McGuinn Laura, et al. (2012), "The Lifelong Effects of Early Childhood Adversity and Toxic Stress," Pediatrics, 129, e232–e46.
Simonson Itamar, Sela Aner. (2011), "On the Heritability of Consumer Decision Making: An Exploratory Approach for Studying Genetic Effects on Judgment and Choice," Journal of Consumer Research, 37 (6), 951–66.
Smith-Woolley Emily, Selzam Saskia, Plomin Robert. (2019), "Polygenic Score for Educational Attainment Captures DNA Variants Shared Between Personality Traits and Educational Achievement," Journal of Personality and Social Psychology, 117 (6), 1145–63.
Sweitzer Maggie M., Halder Indrani, Flory Janine D., Craig Anna E., Gianaros Peter J., Ferrell Robert E., et al. (2013), "Polymorphic Variation in the Dopamine D4 Receptor Predicts Delay Discounting as a Function of Childhood Socioeconomic Status: Evidence for Differential Susceptibility," Social Cognitive and Affective Neuroscience, 8 (5), 499–508.
Tang Yi-Yuan, Hölzel Britta K., Posner Michael I. (2015), "The Neuroscience of Mindfulness Meditation," Nature Reviews Neuroscience, 16 (4), 213–25.
Thaler Richard H., Shefrin Hersh M. (1981), "An Economic Theory of Self-Control," Journal of Political Economy, 89 (2), 392–406.
Thaler Richard H., Sunstein Cass R. (2008), Nudge. New Haven, CT: Yale University Press.
Tomiyama A. Janet. (2014), "Weight Stigma Is Stressful. A Review of Evidence for the Cyclic Obesity/Weight-Based Stigma Model," Appetite, 82, 8–15.
Tucker-Drob Elliot M., Briley Daniel A., Engelhardt Laura E., Mann Frank D., Harden K. Paige. (2016), "Genetically-Mediated Associations Between Measures of Childhood Character and Academic Achievement," Journal of Personality and Social Psychology, 111, 790–815.
Urminsky Oleg, Zauberman Gal. (2017), "The Health Consequences of Intertemporal Preferences," in Handbook of Self-Control in Health and Wellbeing, Ridder Denise de, Adriaanse Marieke, Fujita Kentaro, eds. New York: Routledge, 88–99.
U.S. Department of Health & Human Services (2016), "Facing Addiction in America: The Surgeon General's Report on Alcohol, Drugs, and Health," research report (November), https://store.samhsa.gov/product/Facing-Addiction-in-America-The-Surgeon-General-s-Report-on-Alcohol-Drugs-and-Health-Full-Report/SMA16-4991.
Visscher Peter M., Hill William G., Wray Naomi R. (2008), "Heritability in the Genomics Era—Concepts and Misconceptions," Nature Reviews Genetics, 9 (4), 255–66.
Ward Zachary J., Bleich Sara N., Cradock Angie L., Barrett Jessica L., Giles Catherine M., Flax Chasmine, et al. (2019), "Projected US State-Level Prevalence of Adult Obesity and Severe Obesity," New England Journal of Medicine, 381 (25), 2440–50.
Waters Hugh, Graf Marlon. (2018), America's Obesity Crisis: The Health and Economic Costs of Excess Weight. Santa Monica, CA: Milken Institute.
Wei Qiang, Krolewski David M., Moore Shannon, Kumar Vivek, Li Fei, Martin Brian, et al. (2018), "Uneven Balance of Power between Hypothalamic Peptidergic Neurons in the Control of Feeding," Proceedings of the National Academy of Sciences, 115 (40), E9489–E98.
Wertenbroch Klaus, Schrift Rom, Alba Joseph, Barasch Alixandra, Bhattacharjee Amit, Giesler Markus, et al. (2020), "Autonomy in Consumer Choice," Marketing Letters, 31 (4), 429–39.
Williams Lawrence E., Poehlman T. Andrew. (2016), "Conceptualizing Consciousness in Consumer Research," Journal of Consumer Research, 44, 1–21.
Woon Fu L., Hedges Dawson W. (2008), "Hippocampal and Amygdala Volumes in Children and Adults with Childhood Maltreatment-Related Posttraumatic Stress Disorder: A Meta- Analysis," Hippocampus, 18 (8), 729–36.
Yudkin Daniel. (2018), "The Psychology of Political Polarization," The New York Times (November 17), https://www.nytimes.com/2018/11/17/opinion/sunday/political-polarization-psychology.html.
Zheng Yanmei, Bolton Lisa E., Alba Joseph W. (2019), "Technology Resistance: The Case of Food Production Processes," Journal of Public Policy & Marketing, 38 (4), 264–62.
Zheng Yanmei, Van Osselaer Stijn M.J., Alba Joseph W. (2016), "Belief in Free Will: Implications for Public Policy," Journal of Marketing Research, 53 (6), 1050–64.
~~~~~~~~
By Yanmei Zheng and Joseph W. Alba
Reported by Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 40- Consumers and Artificial Intelligence: An Experiential Perspective. By: Puntoni, Stefano; Reczek, Rebecca Walker; Giesler, Markus; Botti, Simona. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p131-151. 21p. 1 Black and White Photograph, 1 Chart. DOI: 10.1177/0022242920953847.
- Database:
- Business Source Complete
Consumers and Artificial Intelligence: An Experiential Perspective
Artificial intelligence (AI) helps companies offer important benefits to consumers, such as health monitoring with wearable devices, advice with recommender systems, peace of mind with smart household products, and convenience with voice-activated virtual assistants. However, although AI can be seen as a neutral tool to be evaluated on efficiency and accuracy, this approach does not consider the social and individual challenges that can occur when AI is deployed. This research aims to bridge these two perspectives: on one side, the authors acknowledge the value that embedding AI technology into products and services can provide to consumers. On the other side, the authors build on and integrate sociological and psychological scholarship to examine some of the costs consumers experience in their interactions with AI. In doing so, the authors identify four types of consumer experiences with AI: ( 1) data capture, ( 2) classification, ( 3) delegation, and ( 4) social. This approach allows the authors to discuss policy and managerial avenues to address the ways in which consumers may fail to experience value in organizations' investments into AI and to lay out an agenda for future research.
Keywords: artificial intelligence; AI; customer experience; technology marketing; privacy; discrimination; replacement; alienation
Not long ago, artificial intelligence (AI) was the stuff of science fiction. Now it is changing how consumers eat, sleep, work, play, and even date. Consider the diversity of interactions consumers might have with AI throughout the day, from Fitbit's fitness tracker and Alibaba's Tmall Genie smart speaker to Google Photo's editing suggestions and Spotify's music playlists. Given the growing ubiquity of AI in consumers' lives, marketers operate in organizations with a culture increasingly shaped by computer science. Software developers' objective of creating technical excellence, however, may not naturally align with marketers' objective of creating valued consumer experiences. For example, computer scientists often characterize algorithms as neutral tools evaluated on efficiency and accuracy ([62]), an approach that may overlook the social and individual complexities of the contexts in which AI is increasingly deployed. Thus, whereas AI can improve consumers' lives in very concrete and relevant ways, a failure to incorporate behavioral insight into technological developments may undermine consumers' experiences with AI.
This article aims to bridge these two perspectives: on one side, we acknowledge the benefits that AI can provide to consumers. On the other side, we build on and integrate sociological and psychological scholarship to examine the costs consumers can experience in their interactions with AI. Exposing the tension between these benefits and costs, we offer recommendations to guide managers and scholars investigating these challenges. In so doing, we respond to the call from the Marketing Science Institute to examine "the role of the human/tech interface in marketing strategy" and to offer more scholarly attention to situations where "customers face an array of new devices with which to interact with firms, fundamentally altering the purchase experience" (Marketing Science [98]).
We begin by offering a framework that conceptualizes AI as an ecosystem with four capabilities. We focus on the consumer experience of these capabilities, including the tensions felt. We then offer more insights into the experience of these tensions at a macro level by exposing relevant and often explosive narratives in the sociological context and at the micro level by illustrating them with real-life examples grounded in relevant psychological literature. Using these insights, we provide marketers with recommendations regarding how to learn about and manage the tensions. Paralleling the joint emphasis on social and individual responses, we make recommendations outlining both the organizational learning in which firms should engage to lead the deployment of consumer AI and the concrete steps they should take to design improved consumer AI experiences. We close with a research agenda that cuts across the four consumer experiences and suggests ideas for how researchers might contribute new knowledge on this important topic.
We conceptualize AI as an ecosystem comprising three fundamental elements—data collection and storage, statistical and computational techniques, and output systems—that enable products and services to perform tasks typically understood as requiring intelligence and autonomous decision making on behalf of humans ([ 3]). These elements are associated with capabilities (i.e., listening, predicting, producing, and communicating). Data collection devices listen in the broad sense of gathering information from different sources; for example, product sensors scan the environment, and wearable devices record physical activity. Algorithms leverage this information to predict; for example, Spotify serves music suggestions through personalized playlists. Finally, output systems produce a response or communicate with consumers, for example by directing a vehicle or responding through consumer interfaces like Baidu's Duer.
To articulate a customer-centric view of AI, we move attention away from the technology toward how the AI capabilities are experienced by consumers. "Consumer experience" relates to the interactions between the consumer and the company during the customer journey and encompasses multiple dimensions: emotional, cognitive, behavioral, sensorial, and social ([19]; [92]). Our framework is built on four experiences that reflect how consumers interact with the four AI capabilities (Figure 1). This experiential perspective helps shed light on the affective and symbolic aspects of technology consumption in addition to the utilitarian and functional ones ([107]). "Data capture" is the experience of endowing individual data to AI, "classification" is the experience of receiving AI's personalized predictions, "delegation" is the experience of engaging in production processes where the AI performs some tasks on behalf of the consumer, and "social" is the experience of interactive communication with an AI partner.
Graph: Figure 1. The consumer AI experience.
For each experience, we identify benefits and costs from a consumer perspective and propose that managers qualify their focus on the former by paying attention to the latter: a data capture experience may serve or exploit consumers, a classification experience may understand or misunderstand them, a delegation experience may empower or replace consumers, and a social experience may connect or alienate them. We next examine each of these experiences, their social science connections, managerial implications, and future research directions.
The listening capability enables AI systems to collect data about consumers and the environment in which they live. We conceptualize the resulting experience as "data capture," which includes the different ways in which data are transferred to the AI. Data can be intentionally provided by consumers, albeit with different degrees of understanding of the process: consumers share data when there is little or no uncertainty about how the data will be used and by whom, or consumers surrender data when this uncertainty is high ([147]). Data can also be obtained by AI from the "shadows" consumers leave behind when they engage in daily activities, as in the case of a shopper perusing a store equipped with facial recognition technology or of an iRobot Roomba creating a map of a residential space ([89]).
The data capture experience provides benefits to consumers because it can make them feel as if they are served by the AI: the provision of personal data allows consumers access to customized services, information, and entertainment, often for free. For example, consumers who install the Google Photos app let Google capture their memories but in return get an AI-powered assistant that suggests context-sensitive actions when viewing photos. Access to customized services also implies that consumers can enjoy the outcome of decisions made by digital assistants, which effectively match personal preferences with available options without having to endure the cognitive and affective fatigue that decision making can entail ([ 4]). Finally, access to customized services offers unprecedented opportunities for self-improvement. Consider one of the projects within Alphabet, in which data from smartphones, genomes, wearables, and ambient sensors are combined to drive personalized health care ([86]).
Despite AI's ability to predict and satisfy preferences, consumers can feel exploited in data capture experiences, mainly because they do not understand AI's operating criteria. This can be attributed to several features of AI. First, the modalities of data acquisition are becoming increasingly intrusive and difficult to avoid. Second, even when consumers intentionally share information, they are not aware of how this information is aggregated over time and across contexts. Finally, data brokers are largely unregulated and often lack transparency and accountability ([59]). As a result, data capture experiences may threaten consumers' ownership of personal data and challenge personal control, that is, the feeling that events are determined by the self rather than by others or by external forces and can be stirred toward desired outcomes ([36]). We examine the consequences of this loss of control next from both a sociological and psychological perspective.
In popular culture, lack of ownership over personal data has been frequently associated with a loss of personal control stemming from technology's threatening potential to enable monitoring of human behavior. Stories such as George Orwell's 1984 or Philip K. Dick's Minority Report envision systems of oppression in which, due to lack of privacy and constant surveillance, people can no longer control their destiny. This dystopian imagination is echoed in sociological scholarship that associates data capture with the rise of a capitalist marketplace in which private information becomes the central form of capital ([157]).
Such dystopian concerns strike a resonant chord when considering Google's move in the early 2000s to transform consumer data from a by-product into an economic asset that formed the basis of a new type of commerce driven by the ability to colonize the consumer's private experience. This commerce contributes to a surveillance marketplace, in which data surplus is "fed into advanced manufacturing processes known as 'machine intelligence' and fabricated into prediction products that anticipate what you will do now, soon, and later" ([157], p. 14, italics in the original). To illustrate the power of this commerce, targeted ads based on personality characteristics inferred from the analysis of Facebook likes in combination with online survey questions can increase conversion rates by about 50% ([102]). In 2018, Facebook's revenues from the sales of such tailored ads was close to $56 billion ([111]).
From the perspective of this narrative, not only are technology companies continually required to find new ways to make monitoring and surveillance palatable to consumers by linking it to convenience, productivity, safety, or health and well-being ([10]), but they must also constantly push the boundaries of what private information consumers should share ([55]) through a complex landscape of notifications, reminders, and nudges intended to initiate behavioral change. Thus, as consumer behavior becomes increasingly retailored to the exigencies of behavioral futures, AI can transform consumers into subjects who are complicit in the commercial exploitation of their own private experience, thereby undermining personal control and promoting the concentration of knowledge and power in the hands of those who own their information.
Data capture experiences are characterized by an underlying tension: consumers recognize that data capture allows AI to serve them through customization, but AI's inherent lack of transparency makes them feel exploited. These feelings of exploitation are fueled by actual and perceived loss of personal control, with important psychological consequences ([17]). The first of such consequences is negative affect, which can turn into demotivation and helplessness. Consider the case of Leila, a sex worker who shielded her identity on her Facebook account and reported being shocked to see some of her regular clients recommended by the "People You May Know" function. According to Leila, "the worst nightmare of sex workers is to have your real name out there, and Facebook connecting people like this is the harbinger of that nightmare." For Leila, like for domestic violence victims or political activists, privacy invasion is not only frightening, it may become a matter of life, death, or time in jail ([73]).
As being in control is a basic need and a precondition of psychological welfare ([93]), the second consequence of loss of personal control may be moral outrage. Consider the case of a German consumer who requested his own data from Amazon and received transcripts of Alexa's interpretations of voice commands, even though he did not own any Alexa devices. The consumer relayed his story to a local magazine, which attempted to identify the consumer whose privacy had been compromised. The magazine staff involved in this experience described it as follows: "[we were able to] navigate around a complete stranger's private life without his knowledge, and the immoral, almost voyeuristic nature of what we were doing got our hair standing on end" ([24]).
The third consequence of loss of personal control relevant to data capture experiences is psychological reactance, a state in which a person is motivated to restore control after a restriction ([21]), which causes more negative evaluations of and hostile behaviors toward the source of the restriction. In marketing, reactance can decrease the likelihood to repurchase and follow recommendations ([48]). Illustrating reactance in AI data capture experience is Danielle, a U.S. consumer who installed Echo devices throughout her home, believing Amazon's claims that they would not invade her privacy. When one of her Alexas recorded a private conversation and sent it to a random number in her address book, Danielle said "I felt invaded" and concluded, "I'm never plugging that device in again, because I can't trust it" ([76]).
In summary, consumers may experience data capture as a form of exploitation: whereas technology companies, firms, and governmental agencies gain financial and political power, consumers lose ownership of their data and feel a loss of control over their lives. As we discuss next, managers should gain a better understanding of feelings of exploitation, as they prevent consumers from seeing the value firms can provide through data capture. This understanding starts at the organizational level and is then translated into decisions about experience design.
A central programmatic task in addressing the issue of consumer exploitation in AI data capture experiences involves determining and enhancing the organization's level of awareness regarding the sociological and psychological costs raised in the previous sections. Companies should strive toward greater organizational sensitivity around consumer privacy and the current asymmetry in the level of control over personal data. For instance, they should use netnographic observation or sentiment analysis to listen empathetically and at scale to consumers who have experienced exploitation in AI data capture experiences. Furthermore, rather than accepting the surveillance society narrative at face value, firms can use these tools to understand when, how, and whether their own data capture experiences play into versus subvert this narrative. Likewise, companies should draw on insights by privacy scholars and activist movements to question their taken-for-granted beliefs. In doing so, for instance, companies could realize that their own view on privacy default settings might differ markedly from that of a vulnerable consumer group and adjust their processes accordingly ([101]).
Organizational learning can also extend beyond the boundaries of the individual firm to encompass other institutions. First, companies could sponsor research aimed at understanding the influence of surveillance society–style thinking on their culture and practice, as well as its negative impact on marketing activities and consumers. Second, companies could adopt a more communal approach to sharing individual organizational learning with other firms, industry associations, educators, and the media. Third, industry groups could collaborate with scholars to create and adopt an algorithm bill of rights for individuals ([77]), which some AI experts have proposed should include a right to transparency, for example, "the right to know when an algorithm is making a decision about us, which factors are being considered by the algorithm, and how those factors are being weighted" ([127]).
Using this organizational learning, organizations should design improved AI data capture experiences. Recent regulations, such as the European Union's General Data Protection Regulation, aim to limit exploitation by making organizations responsible for giving consumers the possibility to opt into specific data collection processes (e.g., cookies) and to ask for greater clarity on how these data are used.
However, as AI becomes more pervasive and ubiquitous, ensuring consumer consent at all steps of the customer journey may result in an overload of choice and information that decreases instead of increases personal control ([81]) and exacerbates the negative affective and behavioral reactions illustrated previously. Interventions related to the way in which options are presented—the choice architecture—can reduce the cognitive and affective costs associated with excessive information and choice ([30]) and thereby give consumers greater control over their data without overloading them.
Among such interventions, including default options has proven especially effective in facilitating decision making as well as influencing specific behaviors ([140]). Because individuals tend to passively accept defaults instead of exercising their right to opt out, the selection of defaults by choice architects may lead to suboptimal outcomes when it does not properly consider preference heterogeneity. The personalization of defaults could mitigate this issue ([137]), and AI itself could assist consumers in the automatic implementation of preferences about how their data are captured and analyzed.
More broadly, organizations can limit consumer exploitation by playing an active role in educating consumers about the costs and benefits entailed in AI data capture experiences. For example, the recently overhauled Google Home app clearly communicates what user data have been stored and why. Understanding the potential for exploitation in data capture experiences is useful not only for managers interested in maximizing the value provided to consumers served by the AI but also for researchers interested in uncovering the sociological and psychological underpinnings of the tension that accompanies this experience.
Future research should investigate how sociocultural forces affect feelings of exploitation in data capture experiences. People from poorer childhood backgrounds have a lower sense of control than those from wealthier ones ([109]), and collective self-construal is associated with a lower desire for choice freedom and control ([ 9]; [100]). Thus, both consumers' socioeconomic status (Research Question 1A, or RQA1; see Table 1) and prevailing cultural norms (RQA2) could influence consumers' propensity to feel and be exploited by AI. Other factors, such as education, political orientation, gender, and race (RQA3) could be examined using an intersectionality lens ([32]).
Graph
Table 1. Consumers and AI Experience: Emerging Research Questions (RQs).
| A: The AI Data Capture Experience |
|---|
| RQA1: How does socioeconomic status influence the likelihood of feeling exploited? |
| RQA2: How do cultural norms influence the likelihood of feeling exploited? |
| RQA3: How does intersectionality normalize or problematize exploitation? |
| RQA4: How does the diffusion of AI affect feelings of exploitation over time? |
| RQA5: How does motivated reasoning shape consumer affective reactions in data capture experiences? |
| RQA6: How does the frequency of data capture affect perceived exploitation over time? |
| RQA7: How are feelings of exploitation influenced by the nature of the data collected (e.g., environmental, behavioral, physiological)? |
| RQA8: How does the physical context of data collection affect the likelihood of feeling exploited? |
| RQA9: Does the experience of data capture depend on the device the consumer is using? |
| RQA10: When and how will consumers sabotage data collection by AI in response to feelings of exploitation? |
| B: The AI Classification Experience |
| RQB1: How do individual differences in awareness of discrimination affect whether a consumer feels misunderstood by AI? |
| RQB2: How do the social classifications inscribed into AI solutions shape consumer behavior and choices? |
| RQB3: How do consumers infer which variables AI is using to make personalized predictions? |
| RQB4: Which types of inferred classifications are more likely to make consumers feel misunderstood? |
| RQB5: How do uniqueness versus belonging motives affect the likelihood of feeling misunderstood? |
| RQB6: How does the nature of the task influence the likelihood of feeling misunderstood? |
| C: The AI Delegation Experience |
| RQC1: How do feelings of being replaced depend on the perceived "humanness" of an activity? |
| RQC2: How does feeling replaced by AI affect the perceived acceptability of various behaviors intended to protect or promote the self? |
| RQC3: As the range of tasks that AI can perform increases over time, how do normative task boundaries around humans versus algorithms shift? |
| RQC4: What specific consumption contexts make delegation to AI more psychologically aversive? |
| RQC5: Do consumers compensate for feelings of being replaced by AI in nonconsumption domains? |
| RQC6: How do instrumental versus symbolic consumption motives determine perceptions of being replaced? |
| RQC7: Is the likelihood of feeling replaced affected by whether consumers focus on consumption outcomes versus processes? |
| RQC8: When and how do consumers respond to threats of replacement by AI by constraining the AI's production capability? |
| D: The AI Social Experience |
| RQD1: How do antibias beliefs affect alienation in social experiences? |
| RQD2: How do cultural differences influence consumer perceptions of social experiences? |
| RQD3: What are the consequences of AI-enabled social experiences for important societal processes such as children's socialization and gender relations? |
| RQD4: When are customers more likely to objectify AI in response to alienation? |
| RQD5: How does the timing of disclosure influence the likelihood of consumer alienation? |
| RQD6: What is the influence of situational characteristics on alienation? |
| RQD7: What is the role of brand equity in reducing or facilitating alienation? |
| E: Interrelationship Between AI Experiences |
| RQE1: How do the ways in which consumers experience data capture influence perceived resource accessibility in a classification experience? |
| RQE2: Does aggressive data capture strengthen or weaken social inclusion? |
| RQE3: Does involving consumers in the validation of assumptions about their preferences shift a classification experience to feel more like a delegation experience? |
| RQE4: Do changes in feelings of control lead to parallel shifts in data capture and delegation experiences? |
| RQE5: Do changes in consumer self-identity concerns lead to parallel shifts in classification and social experiences? |
| RQE6: Are data capture experiences less aversive when demands for data increase together with feelings of empowerment from delegation experiences? |
| F: Unchartered AI Experiences |
| RQF1: How does the learner–AI interaction shape learning experiences and affect student satisfaction, motivation, and learning? |
| RQF2: How does the valence of learning experiences depend on identity relevance and internal attribution of learning outcomes? |
| RQF3: What motivates consumers to have AI-enabled companionship experiences? |
| RQF4: What factors determine whether consumers perceive companionship experiences as deceptive or alienating? |
| RQF5: How do AI solutions that permeate epistemic boundaries between human and machine impact consumer autonomy? |
| RQF6: How does AI perceive and experience the world and marketplace, and how can firms design these experiences effectively? |
Future research should also explore how the cultural cognitive, normative, or regulatory legitimacy of AI changes over time to influence consumer reactions to data capture ([ 1]; [79]), particularly in light of AI's rapid diffusion in the marketplace. For example, researchers could study how and when increasing levels of familiarity with AI may reduce consumer sensitivity toward exploitation (RQA4).
An interesting avenue for future research consists of exploring the role that psychological processes play in interpreting AI data capture experiences as exploitative. For example, researchers could study the role of motivated reasoning ([88]) in shaping consumer affective reactions to data capture experiences (RQA5): strongly held goals may motivate consumers to accept greater risk of exploitation when the AI is seen as a conduit to goal completion, mitigating negative emotional responses.
Other important open questions concern how the source and type of data used by the AI affect its potential to exploit. For example, an AI-enabled device that is constantly listening to biometric data could, over time, become paradoxically less invasive than one that listens only when activated ([142]). Complementing recent scholarship on the consequences of personal quantification ([42]), future research should address how the frequency of data capture (e.g., intermittent vs. continuous) affects perceived exploitation (RQA6). As another example, information collected about the physical environment, such as that acquired by a smart refrigerator, may be less likely to generate feelings of exploitation than information collected about the self, such as that acquired by a fitness tracker (RQA7).
Feelings of exploitation may also differ on the basis of the physical context of consumption (RQA8). Current attempts by companies like Amazon or Google to redefine the family home as a space accessible to corporations rather than a private space may attenuate or exacerbate these feelings. Similarly, physical features of the environment where data collection takes place may differently trigger concerns about exploitation. For example, crowded environments lead to a loss of perceived control, which could decrease willingness to provide data. Concerns about exploitation may also differ on the basis of the device used to interact with AI (RQA9), as research has shown that consumers are more likely to self-disclose when using smartphones versus PCs ([105]).
Finally, when consumers cannot or do not want to take advantage of the benefits of data capture, psychological reactance toward AI may manifest in adversarial user behaviors, as suggested by the experience of Danielle. Future research can explore the factors that lead consumers to respond to feelings of exploitation with behaviors like sabotaging AI by disabling sensors' inputs, intentionally providing false data by creating fake user profiles, or adopting antisurveillance outerwear to confuse the algorithms controlling facial recognition systems (RQA9).
Firms leverage the predicting capability of AI to create ultra-customized offerings and maximize engagement, relevance, and satisfaction ([87]). Sophisticated algorithms consider a wide variety of information, including the characteristics of both current and past consumers. For example, Netflix uses AI to offer personalized movie recommendations based on not only individuals' past viewing history and that of other viewers but also contextual information such as day of the week, time of day, device, and location ([83]). Netflix even uses AI to select videoframe thumbnails that can increase subscribers' likelihood to click on a specific show ([154]). Even though prediction interfaces use individual and contextual information, they often refer to information related to other users either explicitly by mentioning others when framing recommendations (e.g., Amazon noting "customers who bought this also bought") or implicitly by organizing recommendations in terms of communities of users or taste niches (e.g., Amazon Prime drawing attention to movies for "period drama fans"). As consumers are often unaware of the workings of algorithms, they may infer that these recommendations are based on being classified as a certain type of person. Such inferences are amplified by the human tendency for categorical thinking in person- and self-perception ([143]). For example, consumers engage in categorical inference making when they are served behaviorally targeted ads: they attribute the ads they receive to the advertiser labeling them as a person with specific tastes ([136]). We conceptualize the "classification experience" as one in which consumers perceive AI-enabled personalized predictions to be the result of being classified as a certain consumer type.
Classification experiences can be positive because they lead consumers to feel deeply understood either objectively or subjectively. For example, consumer categorizations can be valuable to affirm the self: personalized offers that indicate membership in an aspirational group may help consumers satisfy identity motives when they are perceived as social labels ([136]). Framings based on other users, such as "people who like this also like," make recommendations more persuasive than those based on the product, such as "similar to this item" ([54]), further suggesting that the experience of feeling classified by AI as a certain type of person is often positive. These findings resonate with research demonstrating the psychological benefits of group membership ([124]; [143]). However, classification experiences may also lead consumers to feel misunderstood when they perceive AI as having inaccurately assigned them to a group or as having made biased predictions on the basis of group assignment. At the societal level, classification by AI is linked to a dystopic narrative in which access to resources and freedom is restricted for some groups.
Classification experiences do not exist in a sociological vacuum but are shaped by popular myths. Science fiction stories such as Neill Blomkamp's Elysium have routinely imagined deeply divided police states in which the ruling class draws on algorithms to sustain a regime of inequality and fear. Sociological scholarship on the politics of algorithms ([133]) has also drawn on this popular imagination to theorize AI in the context of rationalization and quantification ([122]), automated inequality ([38]), uneven information landscapes ([43]), and the historical rise of "algorithms of oppression" ([115]) or "weapons of math destruction" ([117]). Emphasizing the intersectionality of race and gender with antisemitism, poverty, unemployment, and social class ([32]), these investigations of AI's potential for social classification are particularly insightful. AI is feared to privilege whiteness and undermine the identity projects of minorities ([39]). This contention is consistent with research on the market (bio)politics of race, which has consistently shown the inherently discriminatory potential of marketized representations of culture and ethnicity, and it is also supported by economic critiques that warn against the monopolization of information by a centralized system ([70]; [121]).
Consider Google's corporate mission to "organize the world's information." From an unequal worlds perspective, such a statement is far from politically neutral; rather, it exemplifies the operation of seemingly benign appeals to data automation and quantification in a market that sanctions the production of biased information. In such an ideological system, the designers of an AI-enabled college admissions software, for instance, may be convinced that AI can help combat human selection bias. However, because "algorithms that rank and prioritize for profits compromise our ability to engage with complicated ideas" ([115], p. 118), the resulting AI experience may not only reduce the complex experiences of targeted marginalized populations to a set of more simplified sociodemographic attributes or stereotypes but it may also unintentionally expose marginalized applicants to racial profiling, misrepresentation, and economic redlining when used by admissions officers. Likewise, problems can arise when banks use AI to decide whether a consumer is worthy of borrowing money. Although algorithms may make the selection process more efficient, they can also systematically exclude consumers who live in a neighborhood with higher credit defaults ([23]). The realization that AI can result in racial and social groups experiencing discrimination is an important backdrop for a psychological analysis of consumers' feelings of being misunderstood.
Classification experiences are characterized by an underlying tension between feeling understood and misunderstood. Consumers can feel misunderstood because of perceived incorrect classification, discriminatory use of classification, or a combination of the two. First, consumers are likely to feel misunderstood when they perceive the identity implied by the AI's output as incorrect, either because it is factually inaccurate or because it is based only on one identity, whereas most individuals identify with a host of personal and social selves ([120]). Identity-based consumer behavior is often the result of a negotiation between belonging and uniqueness motives playing out across this constellation of identities ([29]). In situations where consumers perceive AI predictions to be driven by their membership in a group, uniqueness motives may become relatively more salient. When this happens, group identity appeals may backfire if they are believed to threaten individual agency ([12]). This negative response is especially likely when the consumer perceives the identity assigned to them by the AI as noncentral or dated, as in this excerpt from a Spotify Community post ([60]):
"The recommendations s*ck:
- Listened to a few anime covers, now all my "Discover Weekly" is filled with disgusting covers. I'm trying to "not like" all of them, but it doesn't work .... I've stopped listening to rock years ago and still get rock recommendations."
From this consumer's perspective, the AI used by Spotify seems to have decided that they like anime covers and rock, putting them in a category that they reject or do not see as capturing their multifaceted and evolving self. The consumer is frustrated not only with being misunderstood by the AI, but also with their perceived inability to alter such misunderstanding.
Second, consumers may also feel misunderstood when they fear AI is using a social category in a discriminatory way to make biased predictions about them. This is particularly problematic in contexts where these predictions may enhance consumers' vulnerability because they restrict access to marketplace resources ([74]). For example, fintech companies increasingly use easily accessible digital information such as individuals registering on a webpage to predict their payment behavior and defaults and therefore judge their creditworthiness ([ 8]). Consider this tweet by a software developer, David Heinemeier Hansson (@dhh, November 7, 2019, https://twitter.com/dhh/status/1192540900393705474):
"The @AppleCard is such a f*ing sexist program. My wife and I filed joint tax returns, live in a community-property state, and have been married for a long time. Yet Apple's black box algorithm thinks I deserve 20x the credit limit she does..."
This consumer is frustrated because of the AI's inability to understand the reality of his household's finances, but he is also morally outraged because he thinks that his wife's denial of credit was based on her gender. Perception of vulnerability such as this can have negative effects on the self-concept. This can occur, for example, when minorities whose financial choices are systemically restricted then frame the self as "fettered, alone, discriminated, and subservient" and experience reductions in self-esteem and self-efficacy ([13]).
Consumers can also experience a combination of the two ways of feeling misunderstood mentioned previously: they can be incorrectly assigned to a category and this incorrect assignment can exacerbate existing limitations on choice and freedom for vulnerable consumers. Facial recognition software, for instance, uses AI to identify a person by comparing a target facial signature to databases of known images. The range of applications of such software includes mobile devices (e.g., Apple's Face ID), social media (e.g., Facebook's tagging feature), and physical spaces (e.g., airport customs officials). Whereas a failure of Apple's Face ID to start one's own device may result in frustration, incorrect identification in other applications may result in ethical violations. Consider the open letter to Amazon CEO Jeff Bezos written by the Congressional Black Caucus on the potential danger caused by Amazon's facial recognition tool, Rekognition:
Communities of color are more heavily and aggressively policed than white communities....We are seriously concerned that wrong decisions will be made due to the skewed data set produced by what we view as unfair and, at times, unconstitutional policing practices. ([125])
In a subsequent test, Rekognition indeed incorrectly matched 28 current members of the U.S. Congress with people who had committed a crime, and the false matches were disproportionately for people of color ([134]). In June 2020, Amazon suspended police use of this technology ([47]). We next examine how managers can understand and address the risk of consumers feeling misunderstood.
How does an organization best surface and address accounts of biased treatment? Unlike data capture errors, which may be lagged and hard to correct in real-time, classification errors produce signals soon after they occur. They also happen in very different parts of an organization. For instance, if an AI system has rejected a college applicant due to a biased algorithm, it is likely to assume that such a classification error will almost immediately surface in the college's admissions department and data—data that in turn might be used to structure the next round of applications.
Owing to this data dependency, organizations may not even be aware that a given distribution or algorithm is the result of a classification error. In the case of a college, for instance, classification might be regarded as a natural outcome of the competitive process by those in charge of managing the admissions process. Thus, unlike data capture failings that require the specific attention of software programmers and data scientists, addressing classification errors requires organizations to focus on marketing and consumer-facing departments and to examine whether these departments' databases or, more abstractly, the organizations' taken-for-granted understanding about whom they have served and should serve and why, carry entrenched social and racial biases.
Organizations must thus focus on learning about the specific biases that might be present in their own algorithms and processes to root them out. In the United States, the Algorithmic Accountability Act of 2019 would require companies to assess their AI systems for "risks of 'inaccurate, unfair, biased, or discriminatory decisions' and to 'reasonably address' the results of their assessments" ([97], p. 1). Rather than reacting to a changing regulatory landscape, firms should proactively collaborate with technology experts and thought leaders in computer science, sociology, and psychology to develop and conduct such audits. Firms can then share both their audit processes and outcomes, for example by engaging in lobbying efforts to ensure that regulations passed in the name of consumer welfare include meaningful and technologically appropriate provisions to protect consumers from discrimination.
Organizational learning should be leveraged in the design phase to develop AI classification experiences that minimize consumers' likelihood of feeling misunderstood. Managers could build on the insights gained from listening to consumers who felt they were classified on the basis of narrowly defined identities to experiment with diversifying and broadening the content they provide and to propose products that are dissimilar from the user's preference profile. Indeed, Spotify has launched Taste Breakers, a function that introduces customers to music to which they normally do not listen. Similar attempts at "bursting the bubble" are especially important in light of the possibility that, by optimizing information provision on the basis of past choices, AI both ignores long-term goals that do not reflect short-term behaviors ([ 4]) and increases attitude extremity and polarization ([49]). Firms could also address feelings of being misunderstood by asking consumers to validate AI-based inferences. As greater user participation in the implementation of algorithms increases satisfaction in decision support systems ([151]), periodically offering consumers the opportunity to update the AI's view of the self could similarly reduce potential frustration.
Managers can build on the insights gained from listening to discriminated consumers to design both debiased and antibias AI experiences that foster an inclusive society rather than perpetuate inequality ([62]). To do so, managers should institute protocols that swiftly react to any bias uncovered in regular audits of the AI systems for the presence of discrimination ([156]). Organizations should also diversify their hiring to include more members of social minority groups and ensure that their culture and processes represent diverse viewpoints at all stages of the design of AI classification experiences. For example, advocates for reducing bias in AI have suggested that technology companies must employ more individuals with disabilities to learn how to eliminate disability bias from AI ([31]). The tension between feeling understood and misunderstood in classification experiences represents a learning opportunity not only for managers but also for researchers.
Researchers can unpack the influence of sociocultural factors on classification experiences. Values and ideology may change consumers' interpretation of personalized predictions, as those who are more aware of the sociohistorical context of discrimination by algorithm ([115]) and belong to marginalized groups should also feel more vulnerable to AI's potential to restrict access to resources and freedom (RQB1).
Drawing on research that examines the ways in which powerful institutions define the consumer ([16]), future work should also explore the social classifications that firms routinely inscribe into their AI solutions, such as certain consumers' habits, norms, and preferences. This lens can usefully unearth the existence of ideological blind spots in the models employed by firms and examine the uneven landscapes of experiences and choices that these models produce when consumers are subjected to them (RQB2).
Future research should explore how psychological processes affect the extent to which consumers feel misunderstood in classification experiences. Open questions concern lay beliefs about how organizations create AI classifications (RQB3) and whether certain inferred categorizations are especially likely to induce feelings of being misunderstood (RQB4). For example, research on attributional ambiguity suggests that stigmatized consumers may attribute AI classifications to bias toward their group identity on the part of the algorithm rather than to other causes ([33]).
More generally, feeling misunderstood may be more likely in contexts where consumers value uniqueness over belongingness (RQB5). For example, patients are reluctant to use medical AI due to a sense that it cannot account for their unique characteristics and circumstances as well as human doctors can ([95]). The nature of a task may also have an influence (RQB6): Consumers tend to exhibit greater aversion toward algorithms for subjective tasks, which are based on personal opinions or intuitions, than for objective ones, which are based on quantifiable and measurable facts ([28]). Given that many AI systems learn and predict subjective taste, negative reactions to inferred classification might be especially common.
A "delegation experience" is one in which consumers involve an AI solution in a production process to perform tasks they would have otherwise performed themselves. These tasks can be decisions, such as when Google Assistant, at the consumer's request, calls a hairdresser, matches the consumer and the hairdresser's calendars, and uses a human-like voice to book an appointment. They can also be actions in the digital world, like those performed by Smart Compose, a writing tool that uses AI to help consumers write emails. Finally, they can be actions in the physical world, such as when the Nest Thermostat learns the consumer's temperature preferences and programs itself to fit them.
By not having to engage in the tasks the AI performs on their behalf, consumers in delegation experiences can feel empowered in two distinct ways. First, consumers can spend their time and effort on activities they find more satisfactory and meaningful: they can work less and enjoy the positive effects of leisure ([46]), or they can work better and enjoy greater happiness by delegating extrinsically motivated tasks to AI and keeping intrinsically motivated tasks for themselves ([18]). Second, consumers can focus on activities that are more suitable to their skills and leave to AI those on which they underperform. This way, they can enhance self-efficacy, or the perceived ability to master the environment to produce a desired outcome ([ 5]).
Given the empowering benefits of delegation experiences, managers may be tempted to offer consumers increasingly more opportunities to delegate tasks to AI. However, like the case in which the mere presence of too many choice options can reduce consumers' satisfaction ([81]), the mere presence of too many delegation opportunities may lead to aversive consequences. We next examine this tension between the possibility of AI to both empower and replace consumers both at the societal and individual level.
To analyze the negative aspects of delegation brought about by the possibility of being replaced from a sociological perspective, it is helpful to examine how the heuristics that have guided consumers' interactions with AI tools have been historically understood in popular culture. We draw on widespread science fiction and social science literature that falls into the so-called "transhumanist" genre. From Fritz Lang's Metropolis to Isaac Asimov's I, Robot, and from Mary Shelley's Gothic Frankenstein to James Cameron's Terminator, countless cautionary tales have profiled the dangers of reimagining human capabilities and characteristics through a technological mirror. Specifically, these stories fuel the view that, by transcending human limitations, technology eventually molds into an omnipotent superhuman and subsequently constitutes the ideal of technological perfection—implying new standards.
Critics of this transhumanist perspective ([129], p. 23) have linked AI to "new logics of expulsion" and economic redundancy that arise as AI approaches aging, health, productivity, and other domains through the transhumanist lens of limitless performance rather than standard levels of well-being or productivity. These observers fear that AI solutions will result in significant unemployment, leading to a rapid increase in surplus populations whose AI experience will be their de facto removal from the productive aspects of the social world.
In the social science literature, this superhuman narrative is paralleled in the Computers Are Social Actors and Human Computer Interactions paradigms, according to which the same heuristics used for human interactions are mindlessly applied to computers ([63]; [114]). Since the 1960s, technology companies have periodically imbued the productive aspects of AI technology and machine prototypes with mythic narratives emphasizing that science and technology will eventually accomplish human immortality.
These transhumanist ideas, which emphasize technological progress as an unstoppable force that alters human experience ([71]), have been deeply inscribed in contemporary AI experiences, from the promise that the Roomba vacuum cleaner could perform tasks more effectively than humans to the promise that 23andMe could help in the creation of genetically optimized offspring. However, the transhumanist preoccupation with Promethean aims underlying many contemporary AI experiences also leads to systemic dehumanization ([53]; [64]). For instance, human perception of mastery over the environment depends on not being subject to unilaterally imposed specifications. A world in which our interactions with machines are fueled by transhumanist ideals will endorse a glorification of capitalism's endless creativity while treating destructiveness and human replacement as normal costs of doing business ([131]). Furthermore, an economic obsession with "perfection," "progress," and "efficiency" will promote the rise of the "useless class" ([66]), individuals whose skills are no longer developed or demanded, thus fundamentally eroding democracy and social justice.
Delegation experiences can help consumers feel empowered but can also raise concerns about being replaced. The mere recognition of AI's capability to act as a substitute for human labor can be psychologically threatening for three main reasons. First, people have a strong desire to attribute consumption outcomes to one's own skills and effort ([ 5]; [94]). Research on human–computer interaction has shown that humans often see computers as disempowering because they deprive humans of the sense of accomplishment related to an activity, so much so that humans tend to credit themselves for positive outcomes and blame computers for negative ones ([110]). In contexts where products are crucial to the experience of having an identity as a certain type of person ([124]), delegation experiences may feel tantamount to cheating. In the fishing industry, for example, AI can help anglers be more effective in location and bait decisions. However, in the words of biologist Culum Brown:
It is really getting kind of unfair. If you are going to use GPS to take you to a location, sonar to identify the fish and a lure which reflects light that humans can't even see, you may as well just go to McDonald's and order a fish sandwich. ([40])
Second, outsourcing labor to machines prevents consumers from practicing and improving their skills, which can negatively influence self-worth and contribute to a satisficing tendency by which individuals settle for a level of engagement that is just good enough. Consider the experience of journalist John Seabrook. While composing an email to his son, Seabrook started the sentence "I am p...," intending to write "I am pleased," but resolved to instead accept the suggestion of Google's Smart Compose "I am proud of you." After hitting Tab to accept the suggestion, [132] muses:
What have I done? Had my computer become my co-writer? That's one small step forward for artificial intelligence, but was it also one step backward for my own?...I'd always finished my thought by typing the sentence to a full stop, as though I were defending humanity's exclusive right to writing, an ability unique to our species. I will gladly let Google predict the fastest route from Brooklyn to Boston, but if I allowed its algorithms to navigate to the end of my sentences how long would it be before the machine started thinking for me?
Finally, outsourcing tasks to AI can lead consumers to experience a loss of self-efficacy. Self-efficacy is an antecedent of personal control ([ 5]), and it is heightened when individuals are actively engaged in creative tasks ([35]; [116]). The notion that being productive is a way to feel in control is consistent with findings showing that consumers who experience low control attempt to reestablish it by choosing products that require higher, versus lower, effort to achieve a desired outcome ([34]). In line with this view that delegation can lead to loss of control, drivers involved in GPS-related accidents tend to describe their experience in terms of surrendering control to the machine. Take for instance the tourists who drove their car into the ocean trying to reach an Australian island and recounted that the GPS "told us we could drive down there...It kept saying it would navigate us to a road" ([108]).
The tension between being empowered and replaced is relevant from a managerial perspective because AI designers need to decide how delegation experiences should be designed to protect self-efficacy and self-identity. We next discuss potential recommendations emerging from the sociological and psychological analysis of this tension.
Companies can start by learning how to integrate the human desire for self-efficacy into corporate discourse in two main ways. First, they can collaborate with family scholars, workplace psychologists, and health sociologists to understand the consequences of human replacement by AI. Second, they can engage in conversations with consumers to gain greater insight into which activities they prefer to reserve for themselves versus delegate to AI, and how these preferences shift across consumer, identity, and task. Organizational design and personnel policies can facilitate this learning by ensuring that the insights gained through external collaborations and consumer listening permeate the firm's culture, especially in the more technical functions. For instance, technology firms could hire experts in creativity such as artists, artisans, or chefs into AI-focused experience design roles.
Firms could also learn from organizations that protect, support, and enhance abilities that are conceived as intrinsically "human" and on which individuals remain superior to machines, such as performing complex tasks, adapting to changes, using emotional intelligence, and offering nuanced judgments in unstructured environments ([78]). Thus, collaborations with museums, theaters, and universities' humanities departments can inspire managers to understand how AI can preserve, rather than subvert, traditional human values such as creativity, collaboration, and community ([25]).
The learning achieved in the previous phase should serve as the bedrock on which AI designers decide how to model delegation experiences to protect self-efficacy and self-identity ([94]). Division of labor in production processes can have positive effects on demand if consumers feel they have the competence to make sound decisions about the tasks in which they decide to engage ([52]). Thus, AI can be conceived as a platform to enhance intrinsically human skills and values. In the medical domain, for example, the benefits of AI-powered surgical robots for consumers depend on the way in which the surgeon's input and supervision is designed. Surgical robots are more precise than humans, can make quicker and more reliable diagnoses, and are more democratic and cost-efficient than current systems because they can intervene outside of hospitals. Still, the structure of surgeons' supervision of the robots is central to the success of this technology, both because patients are afraid of being operated on by a machine and because the AI cannot yet outperform human doctors in some critical technical and social skills ([103]).
Given the link between self-efficacy and control, the design of delegation experiences could also consider the extent to which consumers make choices and initiate actions ([27]; [130]). For example, autonomous vehicles should allow consumers to customize peripheral features to avoid perception of a lack of control ([ 4]), and digital assistants in computer games should not be anthropomorphized to preserve players' sense of autonomy ([84]). The classic finding that cracking fresh eggs into a premade Betty Crocker cake mix might be enough to reestablish consumers' self-worth and improve adoption ([99]) still resonates in the context of AI, as the amount of control needed by consumers to reduce a self-efficacy threat can be quite small. For instance, offering users the possibility to correct an algorithm's output, even if only slightly, is enough to increase their likelihood of using the superior, although imperfect, algorithm rather than the preferred, inferior human forecast ([37]).
The extent to which consumers feel replaced by AI is likely shaped by cultural narratives about AI and by the shared understanding of what it means to be productive. Activities that tend to be perceived as if they ought to fall to human skills and competence ([28]) should be more likely to spur feelings of being replaced (RQC1). Consider a self-driving car choosing between stopping and crossing at an intersection versus choosing between swerving and killing one pedestrian or not swerving and killing several pedestrians ([14]): the car's passenger may feel more replaced in the latter case, which involves a moral dilemma, than in the former case, which involves a mechanical decision. Furthermore, feeling replaced by AI may alter the social or moral acceptability of behavior and its likelihood of occurrence (RQC2). For example, self-protective behaviors appear more moral when adopted by autonomous vehicles than by humans ([58]). Perceptions of what ought to fall to human competence may, however, shift rapidly as AI technology advances (RQC3).
Negative reactions to feeling replaced by AI are likely to differ across consumption contexts (RQC4). Future research can explore whether delegation to AI is less threatening in categories where consumers are already familiar with recommendation agents (e.g., entertainment), are less confident in their own preferences (e.g., finance), are open to experimentation (e.g., food), and can trust the AI brand ([82]). As AI encroaches on an ever-expanding set of human activities, researchers could also explore whether feelings of replacement in one domain could motivate consumers to seek control in others (RQC5). For example, will consumers engaged in daily delegation experiences become more controlling in nonconsumption domains, such as politics?
Future research should examine when the psychological processes that lead to the experience of feeling replaced by AI are activated, as well as the consequences of such feelings. For example, is the extent to which individuals perceive delegation experiences as a threat to the self a function of whether consumption is motivated by instrumental or symbolic motives (RQC6)? Preferences for human over robotic labor tend to be stronger in symbolic consumption contexts ([61]), and the same might apply in the case of one's own labor: whereas for most consumers, being replaced by Nest in setting their home's temperature is likely perceived as desirable, for those whose identity is tightly linked to housekeeping, this replacement may be seen as aversive ([94]). A related topic pertains to how a focus on the outcome or on the process differently influences perceptions of delegation experiences (RQC7). Products are means to ends, but the process of consumption, as well as the performative display of skill and knowledge, can often be intrinsically valuable to consumers ([124]). For example, for a person who is nurturing an angler's image, the extent to which AI-driven fishing tools are seen as self-threatening may depend on the reference group's norms about task delegation and the relative importance placed on the outcome (e.g., a bigger catch) or the process (e.g., finding a good location for fishing).
When self-efficacy and control are threatened in delegation experiences, consumers may employ different strategies to restore them, including increasing agency and seeking structure and boundaries ([90]). Thus, future research can explore whether and when consumers who feel replaced opt to constrain the involvement of the AI in production processes (RQC8) to both reaffirm self-efficacy by increasing their own role in these processes and seek structure by physically and/or mentally bounding AI features. This deliberate limitation of the AI is similar to situations in which consumers restrict their experience with smart objects to the most basic and least innovative forms of interaction ([75]).
AI's capability for engaging in reciprocal communication produces what we term a "social experience." We focus on two types of social experiences: when consumers know at the outset that the interaction partner is an AI, such as when using a voice assistant like Apple's Siri, and when they interact with an AI representing an organization without necessarily knowing initially that it is nonhuman, such as when receiving customer service from an automated chatbot. In both cases, consumers have a social interaction with AI as part of a consumption experience in which the end goal is not the AI interaction. We do not focus on two other types of interactions: when consumers are never aware that the interaction partner is a simulated person (because the experience would be perceived as a normal social interaction) and when consumers interact with the AI as an end in itself, as in the case of a robotic pet.
Social experiences are beneficial when consumers can find in AI a vehicle for information exchange that connects them with the firm in a natural way. This often happens when anthropomorphic features are incorporated in AI-enabled products: anthropomorphic cues increase trust toward self-driving cars ([148]) and reduce perceived risk when consumers are in a position of power ([85]), as when they interact with a virtual assistant. More generally, developments in social robotics are making it possible to create comfortable and even emotionally meaningful AI-powered service interactions (Van [144]). Social AI experiences are beneficial also because they can be more efficient, especially in situations where the alternative to AI is not a human interaction but the absence of any interaction: AI provides consumers access to firms through "conversational commerce."
Despite these advantages, social experiences may also alienate consumers. Negative consumer reactions to simulated social interactions can go well beyond the occasional disappointment as these interactions emerge in a rich cultural context where they can easily trigger societal and individual concerns with unbalanced intergroup relations and discrimination.
The sociological starting point for social experiences is the widespread cultural fascination with humanized machines ([ 2]; [67]; [135]), specifically, the preference for machines that emulate the human body and traits. For instance, a well-noted trope in science fiction is the pursuit of the perfect artificial woman ([72]), a male fantasy of a beguiling, seductive, and sexually obliging object ([126]). These female robots or "gynoids" are routinely imagined as "basic pleasure models" in Philip K. Dick's Blade Runner and sex workers in Michael Crichton's Westworld, or they are traded like used cars in Steve de Jarnatt's Cherry 2000.
This cultural preference for humanized AI is amplified by the widespread use of anthropomorphized chatbots and voice assistants in contemporary AI markets. Humans are less open, agreeable, conscientious, and self-disclosing when they interact with AI versus humans ([113]). However, these perceptual barriers can be overcome, and intimate experiences can be accomplished, when AI products feature human characteristics, behaviors, and language, thus ultimately becoming "artificial besties."
Nevertheless, in this narrative, AI companies that strive for greater human touch cannot ignore that AI products and services modeled as "obliging, docile, and eager-to-please [human] helpers" often contribute to the social alienation of particular groups in society ([150], p. 104). Consistent with this finding, from the iconic robot character Maria in Metropolis to Apple's Siri, patriarchal norms and preferences embedded in seemingly benign AI experiences have the potential to engage only certain types of users, such as white men, while alienating others, such as women and racial minorities ([ 2]; [71]; [67]).
From this perspective, an instance such as Siri's earlier programming to answer to users who say, "you're a slut" with "I'd blush if I could" ([123]) would not just be evidence of biases within the male-centric technology sectors and of the fact that AI mirrors the misogyny concealed in language patterns but also diagnostic of the tendency to undermine AI's social and inclusive possibilities. By collapsing dualistic categories such as male versus female, for instance, social experiences could at least partially ease the social isolation brought about by misogynous and racial stereotyping. At the same time, because anthropomorphized AI typically reproduces such dualistic categories to maximize consumer engagement (e.g., men who treat women as assistants, women who are more assistant-like), social experiences have the potential to exclude rather than include and to alienate rather than connect certain groups of consumers.
AI social experiences have the power to bolster consumer–firm relationships but also to alienate consumers. We identify two main types of alienation engendered by AI social experiences. The first type can occur with any failed automated customer service, as exemplified in this exchange between a customer and chatbot, UX Bear ([152]):
Bot: "How would you describe the term 'bot' to your grandma?"
User: "My grandma is dead."
Bot: "Alright! Thanks for your feedback. [Thumbs up emoji]"
This type of alienation may explain consumers' widespread resistance to replacing humans with machines ([28]; [94]). For example, consumers report feelings of discomfort when interacting with "social robots" in service contexts ([106]), and customers' responses in a field study became markedly more negative when they were informed in advance that their interaction partner would not be a human ([96]). The potential of AI to trigger alienation is also evident in the resurgent interest in social connections that are unmediated by technology, such as authentic consumption experiences ([11]) and more personal marketing exchanges ([146]).
The second type of alienation results from AI's failure to interact successfully with specific groups of consumers. For example, the UK government's reliance on AI to handle claims to its social security program led to experiences like that of Danny Brice, who has learning disabilities and dyslexia and describes his attempts to use the automated Universal Credit program as follows ([15]):
I call it the black hole.... I feel shaky. I get stressed about it. This is the worst system in my lifetime. They assess you as a number not a person. Talking is the way forward, not a bloody computer. I feel like the computer is controlling me instead of a person. It's terrifying.
Thus, AI can exacerbate existing barriers that prevent specific social groups from accessing essential social services, reinforcing societal inequity. Another example of how alienating social experiences can feed inequality is chatbots programmed without considering how existing discrimination in society may affect their operation, such as when Tay, a Twitter bot created by Microsoft, began offering white supremacist answers to users soon after its launch, with exchanges like the following ([104]):
User: "What race is the most evil to you?"
Bot: "Mexican and black."
The cultural narratives of oppression and discrimination underlying this example are even more apparent in the context of personal virtual assistants. Journalist Sigal Samuel recounts working on a piece about sexist AI ([128]):
I said into my phone: "Siri, you're ugly." She replied, "I am?" I said, "Siri, you're fat." She replied, "It must be all the chocolate." I felt mortified for both of us. Even though I know Siri has no feelings, I couldn't help apologizing: "Don't worry, Siri. This is just research for an article I'm writing!" She replied, "What, me, worry?"
Alienating social experiences such as this, in which women face societal pressures around their appearance, may lead consumers to denigrate and belittle the AI, similarly to situations in which individuals derogate outgroup members to reaffirm self-esteem following an identity threat ([20]). Dissatisfaction with a voice-enabled device might produce verbal responses that emphasize its artificial and worthless nature. The tendency to objectify others, and women in particular, is well-known ([50]), and it should be stronger when the interaction partner is, in fact, an inanimate entity, however human-like its communication. Indeed, conversational failures lead consumers to express more frustration with AI when it has a female rather than a male voice ([65]). Firms risk translating this denigration of AI into behaviors that reinforce inequality. As technology enables companies to create automated interactions that are more and more like real human interactions, a new set of ethical issues confront both organizations and marketing researchers, as we discuss in the next sections.
To effectively manage AI social experiences, companies should learn how to acknowledge and accommodate the heterogeneity of human interaction styles and needs. To this aim, firms should collect information directly from consumers who have experienced alienation in their interactions with AI. In addition, firms can leverage technology to gauge and measure alienation (operationalized using measures like amount of stress in the customer's voice) in chats with AI service providers to develop generalizable insights about when alienation is most likely to occur. Firms should also interact with psychologists, sociologists, gerontologists, and other experts to learn about both causes and consequences of alienation.
Organizational learning should also ensure that definitions of anthropomorphism do not draw on and calcify harmful stereotypes about social categories and the way they interact. One way to do so is breaking with organizational cultural conventions that idealize AI as a passive and subservient humanized other by involving experts like linguists, critical theorists, and social psychologists who study the subtle ways in which stereotyping affects communication. For example, disseminating information throughout an organization about the potential societal consequences of exposure to subservient female AIs may shift AI designers away from using female names and voices as defaults ([139]).
Using the greater sensitivity emerging from organizational learning activities, firms can improve the design of AI social experiences. As timely and appropriate firm responses can do much to mitigate the harmful consequences of service failure ([68]), firms should work to increase the effectiveness of interactive AI applications to minimize the likelihood of alienation. Research shows that consumers respond positively when AI service providers personalize the interactions, for example by using the customer's name and explaining the reasons for malfunctions ([27]). Relatedly, firms should also ensure easy and swift transitions from AI to human representatives when the interaction becomes difficult or aversive.
To avoid the perpetuation of harmful stereotypes, companies could also strive to develop AI that is less, rather than more, humanlike ([65]), and indeed, software developers have begun investigating the creation of gender-neutral voices ([138]). This requires a radical change in the mindset of many AI designers (and marketing academics), who often take it for granted that anthropomorphism fosters better relationships with customers ([84]). Organizations should also evaluate the potential consequences of using AI for access to basic social services for consumers like Danny. When AI is deployed to provide important welfare services, designers need to recognize the barriers that they can create for specific user groups, even when the technology has satisfied standard performance benchmarks.
Finally, instead of worrying solely about designing to improve human–AI interaction, firms could address alienation by considering how AI design can improve human–human interaction. Firms can design social experiences that help support what [41] call "care assemblages" by connecting individuals to dear ones in ways that are reminiscent of popular social media strategies designed to foster and satisfy consumers' social goals ([41]). Thus, companies could actively shift from understanding AI as a substitute for humans toward understanding AI as an interface that facilitates social connection ([45]).
Consumers vary in the extent to which they hold antibias beliefs and are willing to take action to address bias in society ([80]). Those who are more concerned about AI fostering alienation may be particularly likely to reject the idea that AI can be a true social partner (RQD1). Cultural differences are also likely to influence the extent to which consumers perceive social experiences with AI as alienating (RQD2). Asian consumers feel a stronger connection to both people and things than Western consumers and, as a result, have shaped their social interactions with AI in more personal ways: whereas AI social experiences in the West are mainly utilitarian and involve disembodied personal assistants, those in the East involve human and animal-appearing robots that are assumed to serve and improve society ([ 7]).
If, over time, AI social experiences become commonplace, future research should explore their broader interpersonal and societal consequences (RQD3). Just as the synthetic and unrealistic nature of pornography has been accused of distorting teens' sexual expectations ([119]), AI social experiences might increase the prevalence of sexist language if they trigger female objectification ([65]). Researchers could also build on literature on intergroup relations, such as [69] theory of dehumanization, to investigate the conditions under which objectification of AI is more likely to occur (RQD4).
An information processing perspective could shed light on how AI social experiences are interpreted and evaluated. The timing of disclosure that the interaction partner is, in fact, an algorithm may influence consumer response to social experiences ([96]), similarly to the "change of meaning" that occurs when consumers realize that a message is meant to influence their behavior ([51]). Thus, alienation might be more likely to emerge if consumers question the company's intention behind disclosing the nature of the interaction partner (RQD5). Moreover, research on the effects of disclosure on word of mouth ([141]) and product placement ([26]) shows that situational factors may influence consumer reactions through an effect on cognitive capacity, and researchers can examine how these factors also affect alienation (RQD6).
Future research could also explore the role of brand equity (RQD7). As brand attachment influences consumer expectations and can shield companies from negative appraisals in ambiguous situations ([91]), stronger consumer–brand relationships may also insulate consumers from experiencing interactions with AI as alienating.
We developed a framework to structure our understanding of consumers' interaction with AI by defining and contextualizing the AI data capture, classification, delegation, and social experiences using both sociological and psychological lenses. In this final section, we go beyond these four experiences to identify additional future research questions in two areas: interrelationships between the four experiences and new AI experiences that may emerge along with new capabilities. These additional research questions are also included in Table 1.
Although we discussed the four consumer AI experiences separately, our framework is not intended to suggest that they exist independently. On the contrary, these experiences could be seen as different aspects of the same customer journey and, as such, could influence each other ([92]). An important avenue for future research is to explore where and how consumers' experience with one AI capability directly affects their experience with another AI capability ([56]). For example, whether consumers feel served versus exploited in an AI data capture experience is likely to affect a subsequent AI classification experience. Consumers who feel exploited may be more likely to worry about AI inappropriately using their personal data to regulate access to valued resources (RQE1). Similarly, intrusive data capture requests might foster consumer alienation (RQE2). For instance, students who view an AI-enabled teaching assistant such as Packback.co as overly inquisitive might feel less included in the virtual classroom and less likely to participate in communal activities such as online discussion boards. Future research can also explore whether consumers are more likely to perceive an AI classification as benefiting them when they are asked to validate inferences made by the AI, turning a classification experience into a delegation one (RQE3).
Another avenue for research is related to the identification of additional ways in which AI experiences influence each other by uncovering shared theoretical foundations. For instance, the data capture and delegation experiences share an emphasis on concerns about personal control, as interacting with AI often involves giving up at least some control over personal data and production processes (RQE4). Similarly, classification and social experiences share an emphasis on concerns about self-identity, as interacting with AI often influences inferences about how AI understands the self and feelings of belonging (RQE5). Confirming the relevance of these theoretical perspectives, personal control and self-identity have been recognized as key concerns in the nascent literature on consumer AI ([ 4]; [ 7]; [27]; [130]). A search for shared theoretical foundations may stimulate academic research and help AI designers form a more holistic understanding of consumers' interaction with AI. For example, as consumers come to understand AI as an independent intelligence operating in the marketplace to whom they can delegate tasks and with whom they can interact, marketplace metacognition and social intelligence ([153]) theory can be leveraged to better understand the theories consumers have about how AI "thinks" (its intentions, strategies, etc.) and how these lay theories influence how consumers respond to AI.
An integrated view of the four experiences will also maximize the value consumers see in organizations' investments into AI. Some companies find themselves in a catch-22 situation in which users need to reveal personally sensitive information for the company to provide valuable benefits but are unwilling to do so unless they can first experience such benefits ([59]). Drawing on an integrated understanding of AI consumer experiences, it may be possible to articulate and structure alternative customer journeys. For example, companies could provide an initial basic service requiring limited disclosure of personal information and later offer the possibility to access an upgraded version that requires additional individual data. Thus, demands for data capture could ramp up as the company is able to demonstrate the benefits that delegation brings to consumers (RQE6).
Our framework offers a parsimonious template to conceptualize how consumers navigate the disparate consumption contexts powered by AI, including social media, online shopping, and personal virtual assistants. In doing so, the framework identifies experiences relevant to a large variety of industries and products. However, additional consumer experiences that we did not examine are on the rise in specific industry sectors, and future research can examine both industry-specific experiences stemming from existing capabilities and new experiences stemming from emerging capabilities (Figure 1).
Although we theorized the production capability as leading to a delegation experience, this capability can also be used to develop an AI "learning experience" in the education industry. Educators can facilitate knowledge and skill acquisition by letting AI personalize aspects of the learning process, such as producing tailored content and testing materials. Future research can examine how different aspects of the learning experience affect subjective and objective assessments of educational outcomes (RQF1). For example, the risk of engendering negative feelings of being replaced in delegation experiences may have a parallel in learning experiences: If an AI application makes it more challenging to internalize the outcome of the learning process, learning experiences might decrease satisfaction and motivation. This may be especially likely to occur when the learning content is relevant to one's identity: just like consumers tend to resist automation in identity-relevant consumption domains when it prevents the internal attribution of consumption outcomes ([94]), students may show reactance to AI applications that prevent them from attributing learning to their own talent and effort (RQF2).
Another avenue for future research would be to relax some of our definitional boundaries to include a larger set of consumption contexts. For example, in our discussion of social experiences, we explicitly excluded contexts in which the interaction with AI is the end in itself, such as sex robots and robotic pets, which are increasingly important in the entertainment and health care industries. Such applications of AI's communication capability give rise to an AI "companionship experience" (RQF3). On the one hand, AI companionship experiences are positive because they can provide both cognitive and socioemotional benefits ([22]). On the other hand, they can deceive vulnerable consumers such as the elderly and toddlers into believing the AI has feelings and may be used as substitutes for real human connections ([145]). While the goal of the creation of robot companions is to simulate an interaction with a real living being, future research could explore at what point the potential for deception and substitution becomes damaging (RQF4).
Finally, emerging AI capabilities may create new consumer AI experiences. In the health care sector, nanorobots are being developed to bring AI solutions directly inside the body, and smartphones, fitness trackers, and smart watches provide essential extensions of cognitive and perceptual capabilities. These products give rise to what researchers have called an AI "cyborg experience" ([57]). A cyborg is "a cybernetic organism, a fusion of the organic and the technical forged in particular, historical, cultural practices" ([67], p. 51). Thus, cyborg experiences emphasize hybridity, self-enhancement, and often radical self-modification, requiring future research to reexamine longstanding epistemic boundaries between human and machine ([ 6]). On the one hand, cyborg experiences destabilize human autonomy and control and might fundamentally undermine consumer freedom ([149]). On the other hand, they collapse dualistic categories like man and machine and might promote consumer empowerment and the circumvention of structural inequalities (RQF5). Lastly, cyborg experiences also raise mind-bending but nonetheless intriguing questions about the kinds of consumption experiences that an AI itself might have ([75]). Consider, in this context, that many firms selling on Amazon today no longer market their offerings directly to consumers but to Amazon-controlled algorithms that act on behalf of these consumers. Future research could explore what marketing strategies are most effective when AI is marketing to AI (RQF6).
AI-enabled products promise to make consumers happier, healthier, and more efficient. Consumer-facing AI products and services such as college admissions software, chatbots, and knowledge aggregators have been heralded as forces for good that can make important contributions to problems such as poverty, lack of education, chronic illness, and racial discrimination. For instance, a World Economic Forum discussion on the future of AI argued that "no one will be left behind" ([155]). A key problem with these optimistic celebrations that view AI's alleged accuracy and efficiency as automatic promoters of democracy and human inclusion is their tendency to efface intersectional complexities.
Instead of considering algorithms as neutral tools, AI designers should recognize that their interventions are "inherently political" and interrogate themselves on "the relationship between their design choices, their professional role, and their vision of the good" ([62], p. 26). We hope that our formulation serves as an antidote to the temptation of "technological solutionism" ([112]) and a useful guide to contrast cases in which targeted consumer segments are subjected to biased outcomes as a result of uncritical firm reliance on AI. We therefore end by noting a key role for the American Marketing Association in shaping the way marketers think about using AI ethically. Although some organizations are beginning to create ethical guidelines around AI, such as the Organization for Economic Co-operation and Development's "Principles for AI" ([118]) and the European Commission's "Ethics Guidelines for Trustworthy AI" ([44]), they are not specifically for marketers. The code of conduct of the American Marketing Association currently includes no mention of AI. We recommend the formation of a taskforce of practitioners and academics from different disciplines to evaluate how professional guidelines could acknowledge the new ethical challenges raised for marketers by the growth of AI.
Footnotes 1 Author Contributions Order of authorship was determined by random draw; all authors contributed equally.
2 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
References Acquisti Alessandro, John Leslie J., Loewenstein George. (2012), "The Impact of Relative Standards on the Propensity to Disclose," Journal of Marketing Research, 49 (2), 160–74.
Adam Alison. (1998), Artificial Knowing: Gender and the Thinking Machine. New York : Routledge.
Agrawal Ajay, Gans Joshua S., Goldfarb Avi. (2018), Prediction Machines: The Simple Economics of Artificial Intelligence. Boston : Harvard Business Review Press.
4 André Quentin, Carmon Ziv, Wertenbroch Klaus, Crum Alia, Frank Douglas, Goldstein William, et al. (2018), "Consumer Choice and Autonomy in the Age of Artificial Intelligence and Big Data," Customer Needs and Solutions, 5 (1/2), 28–37.
5 Bandura Albert. (1977), "Self-Efficacy: Toward a Unifying Theory of Behavioral Change," Psychological Review, 84 (2), 191.
6 Belk Russell W. (2019), "Machines and Artificial Intelligence," Journal of Marketing Behavior, 4 (1), 11–30.
7 Belk Russell W., Humayun Mariam, Gopaldas Ahir. (2020), "Artificial Life," Journal of Macromarketing, 20 (10), 1–16.
8 Berg Tobias, Burg Valentin, Gombović Ana, Puri Manju. (2020), "On the Rise of FinTechs: Credit Scoring Using Digital Footprints," Review of Financial Studies, 33 (7), 2845–97.
9 Bernthal Matthew J., Crockett David, Rose Randall L. (2005), "Credit Cards as Lifestyle Facilitators," Journal of Consumer Research, 32 (1), 130–45.
Bettany Shona M., Kerrane Ben. (2016), "The Socio-Materiality of Parental Style: Negotiating the Multiple Affordances of Parenting and Child Welfare Within the New Child Surveillance Technology Market," European Journal of Marketing, 50 (11), 2041–66.
Beverland Michael B., Farrelly Francis J. (2010), "The Quest for Authenticity in Consumption: Consumers' Purposive Choice of Authentic Cues to Shape Experienced Outcomes," Journal of Consumer Research, 36 (5), 838–56.
Bhattacharjee Amit, Berger Jonah, Menon Geeta. (2014), "When Identity Marketing Backfires: Consumer Agency in Identity Expression," Journal of Consumer Research, 41 (2), 294–309.
Bone Sterling A., Christensen Glenn L., Williams Jerome D. (2014), "Rejected, Shackled, and Alone: The Impact of Systemic Restricted Choice on Minority Consumers' Construction of Self," Journal of Consumer Research, 41 (2), 451–74.
Bonnefon Jean-François, Shariff Azim, Rahwan Iyad. (2016), "The Social Dilemma of Autonomous Vehicles," Science, 352 (6293), 1573–76.
Booth Robert. (2019), "Computer Says No: The People Trapped in Universal Credit's 'Black Hole'," The Guardian (October 14), https://www.theguardian.com/society/2019/oct/14/computer-says-no-the-people-trapped-in-universal-credits-black-hole.
Borgerson Janet. (2005), " Materiality, Agency, and the Constitution of Consuming Subjects: Insights for Consumer Research," in North American Advances in Consumer Research, Vol. 32, Menon Geeta, Rao Akshay R., eds. Duluth, MN : Association for Consumer Research, 439–43.
Botti Simona, Iyengar Sheena S. (2006), "The Dark Side of Choice: When Choice Impairs Social Welfare," Journal of Public Policy & Marketing, 25 (1), 24–38.
Botti Simona, McGill Ann L. (2011), "The Locus of Choice: Personal Causality and Satisfaction with Hedonic and Utilitarian Decisions," Journal of Consumer Research, 37 (6), 1065–78.
Brakus J. Joško, Schmitt Bernd H., Zarantonello Lia. (2009), "Brand Experience: What Is It? How Is It Measured? Does It Affect Loyalty?" Journal of Marketing, 73 (3), 52–68.
Branscombe Nyla R., Wann Daniel L. (1994), "Collective Self-Esteem Consequences of Outgroup Derogation When a Valued Social Identity is on Trial," European Journal of Social Psychology, 24 (6), 641–57.
Brehm Jack W. (1966), A Theory of Psychological Reactance. New York : Academic Press.
Broadbent Elizabeth. (2017), "Interactions with Robots: The Truths We Reveal About Ourselves," Annual Review of Psychology, 68, 627–52.
Brown Dalvin. (2019), "AI Bias: How Tech Determines If You Land Job, Get a Loan, or End Up in Jail" USA Today (October 2), https://www.usatoday.com/story/tech/2019/10/02/how-artificial-intelligence-bias-can-work-against-you/2417711001/.
Brown Jennings. (2018), "The Amazon Alexa Eavesdropping Nightmare Came True," Gizmodo (December 20), https://gizmodo.com/the-amazon-alexa-eavesdropping-nightmare-came-true-1831231490.
Brunk Katja H., Giesler Markus, Hartmann Benjamin J. (2017), "Creating a Consumable Past: How Memory Making Shapes Marketization," Journal of Consumer Research, 44 (6), 1325–42.
Campbell Margaret C., Mohr Gina S., Verlegh Peeter W. (2013), "Can Disclosures Lead Consumers to Resist Covert Persuasion? The Important Roles of Disclosure Timing and Type of Response," Journal of Consumer Psychology, 23 (4), 483–95.
Carmon Ziv, Schrift Rom Y., Wertenbroch Klaus, Yang Haiyang. (2020), "Designing AI Systems that Customers Won't Hate," MIT Sloan Management Review, 61 (2), 1–6.
Castelo Noah, Bos Maarten W., Lehmann Donald R. (2019), "Task-Dependent Algorithm Aversion," Journal of Marketing Research, 56 (5), 809–25.
Chan Cindy, Berger Jonah, Boven Leaf van. (2012), "Identifiable but Not Identical: Combining Social Identity and Uniqueness Motives in Choice," Journal of Consumer Research, 39 (3), 561–73.
Chernev Alexander, Böckenholt Ulf, Goodman Joseph. (2015), "Choice Overload: A Conceptual Review and Meta-analysis," Journal of Consumer Psychology, 25 (2), 333–58.
Clegg Alicia. (2020), "How to Design AI that Eliminates Disability Bias," Financial Times (January 26), https://www.ft.com/content/f5bd21da-33b8-11ea-a329-0bcf87a328f2.
Crenshaw Kimberle. (1989), " Demarginalizing the Intersection of Race and Sex: A Black Feminist Critique of Antidiscrimination Doctrine, Feminist Theory and Antiracist Politics," University of Chicago Legal Forum, Vol. 1989, Article 8.
Crocker Jennifer, Major Brenda. (1989), "Social Stigma and Self-Esteem: The Self-Protective Properties of Stigma," Psychological Review, 96 (4), 608–30.
Cutright Keisha M., Samper Adriana. (2014), "Doing It the Hard Way: How Low Control Drives Preferences for High-Effort Products and Services," Journal of Consumer Research, 41 (3), 730–45.
Dahl Darren W., Page Moreau C. (2007), "Thinking Inside the Box: Why Consumers Enjoy Constrained Creative Experience," Journal of Marketing Research, 44 (3), 357–69.
DeCharms Richard. (1968), Personal Causation. New York : Academic Press.
Dietvorst Berkeley J., Simmons Joseph P., Massey Cade. (2016), "Overcoming Algorithm Aversion: People Will Use Imperfect Algorithms If They Can (Even Slightly) Modify Them," Management Science, 64 (3), 1155–70.
Dormehl Luke. (2014 a), "Algorithms Are Great and All, but They Can Also Ruin Lives," Wired (November 19), https://www.wired.com/2014/11/algorithms-great-can-also-ruin-lives/.
Dormehl Luke. (2014 b), "Facial Recognition: Is the Technology Taking Away Your Identity?" The Guardian (May 4), https://www.theguardian.com/technology/2014/may/04/facial-recognition-technology-identity-tesco-ethical-issues.
The Economist (2012), "The One That Didn't Get Away," (June 23), https://www.economist.com/science-and-technology/2012/06/23/the-one-that-didnt-get-away.
Epp Amber M., Schau Hope Jensen, Price Linda L. (2014), "The Role of Brands and Mediating Technologies in Assembling Long-Distance Family Practices," Journal of Marketing, 78 (3), 81–101.
Etkin Jordan. (2016), "The Hidden Cost of Personal Quantification," Journal of Consumer Research, 42 (6), 967–84.
Eubanks Virginia. (2018), Automating Inequality: How High-Tech Tools Profile, Police, and Punish the Poor. New York : St. Martin's Press.
European Commission, "Ethics Guidelines for Trustworthy AI," (accessed September 21, 2020), https://ec.europa.eu/futurium/en/ai-alliance-consultation.
Farooq Umer, Grudin Jonathan. (2016), "Human–Computer Integration," IX Interactions, 6 (November/December), http://interactions.acm.org/archive/view/november-december-2016/human-computer-integration.
Fishbach Ayelet, Choi Jinhee. (2012), "When Thinking About Goals Undermines Goal Pursuit," Organizational Behavior and Human Decision Processes, 118 (2), 99–107.
Fitch Asa. (2020), "Amazon Suspends Police Use of Its Facial-Recognition Technology," The Wall Street Journal (June 10), https://www.wsj.com/articles/amazon-suspends-police-use-of-its-facial-recognition-technology-11591826559.
Fitzsimons Gavan, Lehmann Donald R. (2004), "Reactance to Recommendations: When Unsolicited Advice Yields Contrary Responses," Marketing Science, 23 (1), 82–94.
Flaxman Seth, Goel Sharad, Rao Justin M. (2016), "Filter Bubbles, Echo Chambers, and Online News Consumption," Public Opinion Quarterly, 80 (S1), 298–320.
Fredrickson Barbara L., Roberts Tomi-Ann. (1997), "Objectification Theory: Toward Understanding Women's Lived Experiences and Mental Health Risks," Psychology of Women Quarterly, 21 (2), 173–206.
Friestad Marian, Wright Peter. (1994), "The Persuasion Knowledge Model: How People Cope with Persuasion Attempts," Journal of Consumer Research, 21 (1), 1–31.
Fuchs Christoph, Prandelli Emanuela, Schreier Martin. (2010), "The Psychological Effects of Empowerment Strategies on Consumers' Product Demand," Journal of Marketing, 74 (1), 65–79.
Fukuyama Francis. (2002), Our Posthuman Future. New York : Farrar, Straus, and Giroux.
Gai Phyliss Jia, Klesse Anne-Kathrin. (2019), "Making Recommendations More Effective Through Framings: Impacts of User- Versus Item-Based Framings on Recommendation Click-Throughs," Journal of Marketing, 83 (6), 61–75.
Giesler Markus, Humphreys Ashlee. (2007), " Tensions Between Access and Ownership in the Media Marketplace," in NA–Advances in Consumer Research, Vol. 34, Fitzsimons Gavan, Morwitz Vicki, eds. Duluth, MN : Association for Consumer Research, 696–700.
Giesler Markus, Fischer Eileen. (2018), "IoT Stories: The Good, the Bad and the Freaky," Marketing Intelligence Review, 10 (2), 24–8.
Giesler Markus, Venkatesh Alladi. (2005), " Reframing the Embodied Consumer as Cyborg: A Posthumanist Epistemology of Consumption," in NA–Advances in Consumer Research, Vol. 32, Menon Geeta, Rao Akshay R., eds. Duluth, MN : Association for Consumer Research, 661–69.
Gill Tripat. (2020), "Blame It on the Self-Driving Car: How Autonomous Vehicles Can Alter Consumer Morality," Journal of Consumer Research, 47 (2), 272–91.
Grafanaki Sofia. (2017), "Autonomy Challenges in the Age of Big Data," Fordham Intellectual Property, Media & Entertainment Law Journal, 27, 803.
Grandterr (2019), "Why Are Recommendations So Terrible," Spotify Community forum, https://community.spotify.com/t5/iOS-iPhone-iPad/Why-are-recommendations-so-terrible/td-p/4769866.
Granulo Armin, Fuchs Christoph, Puntoni Stefano. (2020), "Preference for Human (vs. Robotic) Labor Is Stronger in Symbolic Consumption Contexts," Journal of Consumer Psychology (published online July 18), DOI:10.1002/jcpy.1181.
Green Ben, Viljoen Salomé. (2020), " Algorithm Realism: Expanding the Boundaries of Algorithmic Thought," in Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency, January 19–31.
Grudin Jonathan. (2017), From Tool to Partner: The Evolution of Human–Computer Interaction. San Rafael, CA : Morgan and Claypool Publishers.
Habermas Jurgen. (2003), The Future of Human Nature. London : Polity.
Hadi Rhonda, Bock Lauren, Robinson Sandra, Du Jessie. (2020), " When Alexa Lets Us Down: Conversational Failures with Female Artificial Intelligence Lead to Greater Expressed Frustration," working paper.
Harari Yuval N. (2017), Homo Deus: A Brief History of Tomorrow. London : Penguin Publishing.
Haraway Donna. (1985), "A Manifesto for Cyborgs: Science, Technology, and Socialist Feminism in the 1980s," Socialist Review, (80), 65–107.
Hart Christopher W., Heskett James L., Earl Sasser W. Jr. (1990), "The Profitable Art of Service Recovery," Harvard Business Review, 68 (4), 148–56.
Haslam Nick. (2006), "Dehumanization: An Integrative Review," Personality and Social Psychology Review, 10 (3), 252–64.
Hayek Friedrich. (1945), "The Use of Knowledge in Society," The American Economic Review, 35, 519–30.
Hayles N. Katherine. (1999), How We Became Posthuman: Virtual Bodies in Cybernetics, Literature, and Informatics. Chicago : University of Chicago Press.
Hayter Irena. (2017), "Robotics, Science Fiction, and the Search for the Perfect Artificial Woman," The Conversation (October 24), https://theconversation.com/robotics-science-fiction-and-the-search-for-the-perfect-artificial-woman-86092.
Hill Kashmir. (2017), "How Facebook Outs Sex Workers," Gizmodo (October 11), https://gizmodo.com/how-facebook-outs-sex-workers-1818861596.
Hill Ronald, Sharma Eesha. (2020), "Consumer Vulnerability," Journal of Consumer Psychology, 30 (3), 551–70.
Hoffman Donna L., Novak Thomas P. (2018), "Consumer and Object Experience in the Internet of Things: An Assemblage Theory Approach," Journal of Consumer Research, 44 (6), 1178–04.
Horcher Gary. (2018), "Woman Says Her Amazon Device Recorded Private Conversation, Sent It Out to Random Contact," KIRO 7 News (May 25), https://www.kiro7.com/news/local/woman-says-her-amazon-device-recorded-private-conversation-sent-it-out-to-random-contact/755507974/.
Hosanagar Kartik. (2019), A Human's Guide to Machine Intelligence: How Algorithms are Shaping Our Lives and How We Can Stay in Control. New York : Viking.
Hume Kathryn. (2018), "Designing AI to Make Decisions," August 10, 2018, in HBR IdeaCast, produced by Dooe Mary, Nickisch Curt, podcast, https://hbr.org/podcast/2018/08/designing-ai-to-make-decisions.html.
Humphreys Ashlee. (2010), "Semiotic Structure and the Legitimation of Consumption Practices: The Case of Casino Gambling," Journal of Consumer Research, 37 (3), 490–510.
Ivarsflaten Elisabeth, Blinder Scott, Ford Robert. (2010), "The Anti-Racism Norm in Western European Immigration Politics: Why We Need to Consider It and How to Measure It," Journal of Elections, Public Opinion, and Parties, 20 (4), 421–45.
Iyengar Sheena, Lepper Mark. (2000), "When Choice Is Demotivating: Can One Desire Too Much of a Good Thing?" Journal of Personality and Social Psychology, 79 (6), 995–1006.
JWT Intelligence Wunderman Thompson (2016), "Control Shift," (accessed September 22, 2020), https://www.jwtintelligence.com/trend-reports/control-shift/.
Kathayat Vinod. (2019), "How Netflix Uses AI for Content Creation and Recommendation," Medium (September 28), https://medium.com/swlh/how-netflix-uses-ai-for-content-creation-and-recommendation-c1919efc0af4.
Kim Sara, Chen Rocky Peng, Zhang Ke, (2016), "Anthropomorphized Helpers Undermine Autonomy and Enjoyment in Computer Games," Journal of Consumer Research, 43 (2), 282–302.
Kim Sara, McGill Ann L. (2011), "Gaming with Mr. Slot or Gaming the Slot Machine? Power, Anthropomorphism, and Risk Perception," Journal of Consumer Research, 38 (1), 94–107.
Kuchler Hannah. (2020), "Can We Ever Trust Google with Our Health Data?" Financial Times (January 20), https://www.ft.com/content/4ade8884-1b40-11ea-97df-cc63de1d73f4.
Kumar V., Rajan Barath, Venkatesan Rajkumar, Lecinski Jim. (2019), "Understanding the Role of Artificial Intelligence in Personalized Engagement Marketing," California Management Review, 61 (4), 135–55.
Kunda Ziva. (1990). "The Case for Motivated Reasoning," Psychological Bulletin, 108 (3), 480–98.
Kuniavsky Mike. (2010), Smart Things: Ubiquitous Computing User Experience Design. Burlington, MA : Morgan Kauffman Elsevier.
Landau Mark J., Kay Aaron C., Whitson Jennifer A. (2015), "Compensatory Control and the Appeal of a Structured World," Psychological Bulletin, 141 (3), 694–722.
Lee Leonard, Frederick Shane, Ariely Dan. (2006), "Try It, You'll Like It: The Influence of Expectation, Consumption, and Revelation on Preferences for Beer," Psychological Science, 17 (12), 1054–58.
Lemon Katherine N., Verhoef Peter C. (2016), "Understanding Customer Experience Throughout the Customer Journey," Journal of Marketing, 80 (6), 69–96.
Leotti Lauren A., Iyengar Sheena S., Ochsner Kevin N. (2010), "Born to Choose: The Origins and Value of the Need for Control," Trends in Cognitive Sciences, 14 (10), 457–63.
Leung Eugina, Paolacci Gabriele, Puntoni Stefano. (2018), "Man Versus Machine: Resisting Automation in Identity-Based Consumer Behavior," Journal of Marketing Research, 55 (6), 818–31.
Longoni Chiara, Bonezzi Andrea, Morewedge Carey. (2019), "Resistance to Medical Artificial Intelligence," Journal of Consumer Research, 46 (3), 407–34.
Luo Xueming, Tong Siliang, Fang Zheng, Qu Zhe. (2019), "Machines Versus Humans: The Impact of AI Chatbot Disclosure on Customer Purchases," Marketing Science, 38 (6), 937–47.
MacCarthy Mark. (2019), " An Examination of the Algorithmic Accountability Act of 2019," (accessed September 22, 2020), https://ssrn.com/abstract=3615731.
Marketing Science Institute (2018), Research Priorities 2018–2020. Cambridge, MA : Marketing Science Institute.
Marks Susan. (2005), Finding Betty Crocker: The Secret Life of America's First Lady of Food. New York : Simon & Schuster.
Markus Hazel Rose, Schwartz Barry. (2010), "Does Choice Mean Freedom and Well-Being?" Journal of Consumer Research, 37 (2), 344–55.
Martin Kelly D., Murphy Patrick E. (2017), "The Role of Data Privacy in Marketing," Journal of the Academy of Marketing Science, 45, 135–55.
Matz Sandra C., Kosinski Michael, Nave Gideon, Stillwell David J. (2017), "Psychological Targeting as an Effective Approach to Digital Mass Persuasion," Proceedings of the National Academy of Sciences, November, 12714–19.
Max D.T. (2019), "Paging Dr. Robot: A Pathbreaking Surgeon Prefers to Do His Cutting by Remote Control," The New Yorker (September 23), https://www.newyorker.com/magazine/2019/09/30/paging-dr-robot.
Me.me (2020), "Microsoft Creates AI Bot—Internet Immediately Turns It Racist," (accessed September 21), https://me.me/i/damon-daymin-tayandyou-what-race-is-the-most-evil-to-45baf158bb7a40c68b4bbb6c3561f1b3.
Melumad Shiri, Meyer Robert. (2020), "Full Disclosure: How Smartphones Enhance Consumer Self-Disclosure," Journal of Marketing, 84 (3), 28–45.
Mende Martin, Scott Maura L., Doorn Jenny van, Grewal Dhruv, Shanks Ilana. (2019), "Service Robots Rising: How Humanoid Robots Influence Service Experiences and Elicit Compensatory Consumer Responses," Journal of Marketing Research, 56 (4), 535–56.
Mick David G., Fournier Susan. (1998), "Paradoxes of Technology: Consumer Cognizance, Emotions, and Coping Strategies," Journal of Consumer Research, 25 (2), 123–43.
Milner Greg. (2016), "Death by GPS: Are Satnavs Changing Our Brains?" The Guardian (June 25), https://www.theguardian.com/technology/2016/jun/25/gps-horror-stories-driving-satnav-greg-milner.
Mittal Chiraag, Griskevicius Vladas. (2014), "Sense of Control Under Uncertainty Depends on People's Childhood Environment: A Life History Theory Approach," Journal of Personality and Social Psychology, 107 (4), 621–37.
Moon Youngme, Nass Clifford. (1998), "Are Computers Scapegoats? Attributions of Responsibility in Human–Computer Interaction," International Journal of Human–Computer Studies, 49, 79–94.
Moore Elaine, Murphy Hannah. (2019), "Facebook's Fake Numbers Problem," Financial Times (November 17), https://www.ft.com/content/98454222-fef1-11e9-b7bc-f3fa4e77dd47.
Morozov Evgeny. (2013), To Save Everything, Click Here: The Folly of Technological Solutionism. New York : Public Affairs.
Mou Yi, Xu Kun. (2017), "The Media Inequality: Comparing the Initial Human–Human and Human–AI Social Interactions," Computers in Human Behavior, 72, 432–40.
Nass Clifford, Moon Youngme. (2000), "Machines and Mindlessness: Social Responses to Computers," Journal of Social Issues, 56 (1), 81–103.
Noble Safiya Umoja. (2018), Algorithms of Oppression: How Search Engines Reinforce Racism. New York : New York University Press.
Norton Michael I., Mochon Daniel, Ariely Dan. (2012), "The IKEA Effect: When Labor Leads to Love," Journal of Consumer Psychology, 22 (3), 453–60.
O'Neil Cathy. (2016), Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy. New York : Broadway Books.
Organisation for Economic Co-operation and Development, "OECD Principles on AI," (accessed September 21, 2020), https://www.oecd.org/going-digital/ai/principles/.
Owens Eric W., Behun Richard J., Manning Jill C., Reid Rory C. (2012), "The Impact of Internet Pornography on Adolescents: A Review of the Research," Sexual Addiction & Compulsivity, 19 (1/2), 99–122.
Oyserman Daphna. (2009), "Identity-Based Motivation and Consumer Behavior," Journal of Consumer Psychology, 19 (3), 250–60.
Polanyi Michael. (1948), "Planning and Spontaneous Order," The Manchester School, 16 (3), 237–68.
Porter Theodore. (1996), Trust in Numbers. Princeton, NJ : Princeton University Press.
Rawlinson Kevin. (2019), "Digital Assistants like Siri and Alexa Entrench Gender Biases, Says UN," The Guardian (May 21), https://www.theguardian.com/technology/2019/may/22/digital-voice-assistants-siri-alexa-gender-biases-unesco-says.
Reed II Americus, Forehand Mark R., Puntoni Stefano, Warlop Luk. (2012), "Identity-Based Consumer Behavior," International Journal of Research in Marketing, 29 (4), 310–21.
Richmond Cedric L. (2018), "Open Letter to Jeffrey Bezos," Congressional Black Caucus (May 24), https://cbc.house.gov/uploadedfiles/final_cbc_amazon_facial_recognition_letter.pdf.
Rose Steve. (2015), " Ex Machina and Sci-Fi's Obsession with Sexy Female Robots," The Guardian (January 15), https://www.theguardian.com/film/2015/jan/15/ex-machina-sexy-female-robots-scifi-film-obsession.
Samuel Sigal. (2019 a), "10 Things We Should All Demand from Big Tech Right Now," Vox (May 29), https://www.vox.com/the-highlight/2019/5/22/18273284/ai-algorithmic-bill-of-rights-accountability-transparency-consent-bias.
Samuel Sigal. (2019 b), "Alexa, Are You Making Me Sexist?" Vox (June 12), https://www.vox.com/future-perfect/2019/6/12/18660353/siri-alexa-sexism-voice-assistants-un-study.
Sassen Saskia. (2014), Expulsions. Cambridge, MA : Harvard University Press.
Schmitt Bernd. (2019), "From Atoms to Bits and Back: A Research Curation on Digital Technology and Agenda for Future Research," Journal of Consumer Research, 46 (4), 825–32.
Schumpeter Joseph A. (1942), Capitalism, Socialism, and Democracy. New York : Harper.
Seabrook John. (2019), "The Next Word: Where Will Predictive Text Take Us?" The New Yorker (October 14), https://www.newyorker.com/magazine/2019/10/14/can-a-machine-learn-to-write-for-the-new-yorker.
Seaver Nick. (2019), " Knowing Algorithms," in digitalSTS : A Field Guide for Science and Technology Studies, Vertesi Janet, Ribes David, eds. Princeton, NJ : Princeton University Press, 412–22.
Snow Jacob. (2018), "Amazon's Face Recognition Falsely Matched 28 Members of Congress with Mugshots," ACLU (July 26), https://www.aclu.org/blog/privacy-technology/surveillance-technologies/amazons-face-recognition-falsely-matched-28.
Suchman Lucy, Roberts Celia, Hird Myra J. (2011), "Subject Objects," Feminist Theory, 12 (2), 119–45.
Summers Christopher A., Smith Robert W., Reczek Rebecca Walker. (2016), "An Audience of One: Behaviorally Targeted Ads as Implied Social Labels," Journal of Consumer Research, 43 (1), 156–78.
Sunstein Cass R. (2015), Choosing Not to Choose: Understanding the Value of Choice. Oxford : Oxford University Press.
Sydell Laura. (2018), "The Push for A Gender-Neutral Siri," NPR (July 9), https://www.npr.org/2018/07/09/627266501/the-push-for-a-gender-neutral-siri.
Teich David A. (2020), "AI and the Continuation of Gender Bias in Communications," Forbes (May 11), https://www.forbes.com/sites/davidteich/2020/05/11/ai-and-the-continuation-if-gender-bias-in-communications.
Thaler Richard H., Benartzi Shlomo. (2004) "Save More Tomorrow™: Using Behavioral Economics to Increase Employee Saving," Journal of Political Economy, 112 (S1), S164–87.
Tuk Mirjam A., Verlegh Peeter W., Smidts Ale, Wigboldus Daniel H. (2009), "Sales and Sincerity: The Role of Relational Framing in Word-of-Mouth Marketing," Journal of Consumer Psychology, 19 (1), 38–47.
Turkle Sherry. (2008), " Always-On/Always-On-You: The Tethered Self," in Handbook of Mobile Communication Studies, Katz James E., ed. Cambridge, MA : MIT Press.
Turner John C., Reynolds Katherine J. (2011), " Self-categorization Theory," in Handbook of Theories in Social Psychology, Vol. 2, Lange Paul A.M. van, Kruglanski Arie W., Tory Higgins E., eds. 399–417.
Van Doorn, Martin Mende Jenny, Noble Stephanie M., Hulland John, Ostrom Amy L., Grewal Dhruv, ed. (2017), "Domo Arigato Mr. Roboto: Emergence of Automated Social Presence in Organizational Frontlines and Customers' Service Experiences," Journal of Service Research, 20 (1), 43–58.
Van Oost Ellen, Reed Darren. (2010), " Towards a Sociological Understanding of Robots as Companions," in Human–Robot Personal Relationships: Lecture Notes of the Institute for Computer Sciences, Social Informatics, and Telecommunications Engineering, Oost Ellen van, Reed Darren, eds. Berlin : Springer, 59.
Van Osselaer, Stijn M.J., Fuchs Christoph, Schreier Martin, Puntoni Stefano. (2020), "The Power of Personal," Journal of Retailing, 96 (1), 88–100.
Walker Kristen L. (2016), "Surrendering Information Through the Looking Glass: Transparency, Trust, and Protection," Journal of Public Policy & Marketing, 35 (1), 144–58.
Waytz Adam, Heafner Joy, Epley Nicholas. (2014), "The Mind in the Machine: Anthropomorphism Increases Trust in an Autonomous Vehicle," Journal of Experimental Social Psychology, 52 (May), 113–17.
Wertenbroch Klaus, Schrift Rom Y., Alba Joseph W., Barasch Alixandra, Bhattacharjee Amit, Giesler Markus, et al. (2020), "Autonomy in Consumer Choice," Marketing Letters (published online June 28), DOI:10.1007/s11002-020-09521-z.
West Mark, Kraut Rebecca, Chew Han Ei. (2019), I'd Blush if I Could: Closing Gender Divides in Digital Skills Through Education. EQUALS.
Wierenga Berend, Ophuis Peter A.M. Oude. (1997), "Marketing Decision Support Systems: Adoption, Use, and Satisfaction," International Journal of Research in Marketing, 14 (3), 275–90.
Wong Christine. (2019), "Keeping It Real: Preserving the Humanity of Customer Experience in the Age of AI," Futurithmic (April 1), https://www.futurithmic.com/2019/04/01/keeping-it-real-preserving-humanity-customer-experience-in-age-of-ai/.
Wright Peter. (2002), "Marketplace Metacognition and Social Intelligence," Journal of Consumer Research, 28 (4), 677–82.
Yu Allen. (2019), "How Netflix Uses AI, Data Science, and Machine Learning—From A Product Perspective," Medium (February 27), https://becominghuman.ai/how-netflix-uses-ai-and-machine-learning-a087614630fe.
Zhou Bowen. (2020), "Here Are 3 Big Concerns Surrounding AI—and How to Deal with Them," World Economic Forum (February 17), https://www.weforum.org/agenda/2020/02/where-is-artificial-intelligence-going/.
Zou James, Schiebinger Londa. (2019), "AI Can Be Sexist and Racist—It's Time to Make It Fair," Nature, 559 (7714), 324–26.
Zuboff Shoshana. (2019), The Age of Surveillance Capitalism: The Fight for a Human Future at the New Frontier of Power. New York : Profile Books.
~~~~~~~~
By Stefano Puntoni; Rebecca Walker Reczek; Markus Giesler and Simona Botti
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 41- Corrigendum to "Carbon Footprinting and Pricing Under Climate Concerns". Journal of Marketing. Mar2022, Vol. 86 Issue 2, p202-202. 1p. 4 Charts. DOI: 10.1177/00222429221078948.
- Database:
- Business Source Complete
Corrigendum to "Carbon Footprinting and Pricing Under Climate Concerns"
Bertini, Marco, Stefan Buehler, Daniel Halbheer, and Donald R. Lehmann (2020), "Carbon Footprinting and Pricing Under Climate Concerns." Journal of Marketing, (published online July 7, 2020), DOI: 10.1177/0022242920932930.
This article has been revised and republished due to substantial changes to the text of the original article, as published Online First on July 7, 2020.
The article was revised after the authors informed the journal on March 4, 2021 that Proposition 3 in the original Online First article contained a statement on the properties of the comparative statics of the corporate carbon footprint that is inconsistent with the paper's assumptions on demand. This error was discovered by Professor Fabian Herweg (University of Bayreuth, Germany) and communicated to the authors on February 25, 2021.
After assessing the repercussions of the mistake on the remainder of the paper, the authors were invited to submit a revised version to the Journal of Marketing's Editor, Harald van Heerde. The revised paper was then reviewed by the Editor and two new reviewers as well as the Associate Editor who had been involved with the article since the original manuscript was submitted.
Specifically, the revisions lead to the following changes to the originally published article:
- New assumption on demand that in the model section ensures that stronger climate concerns unambiguously reduce the product carbon footprint (Proposition 2) and the corporate carbon footprint (Proposition 3). The revised proofs of Propositions 2 and 3 are based on an approach recommended by Professor Fabian Herweg. The manuscript was updated accordingly in the Abstract, the Introduction, following Propositions 2 and 3, and in the Discussion.
- New Footnote 4 illustrates the ambiguous impact of stronger climate concerns on the profit-maximizing price.
- New Footnote 5 illustrates the possibility of a rebound effect outside of the model. This change is also reflected in the revised "Discussion" section.
- The restatement of Proposition 3 triggered changes in the parts of Propositions 6 and 8 (and their proofs) that rely on comparative statics properties of the corporate carbon footprint.
- The revision requests by the review team resulted in changes to Propositions 4 and 10 (and their proofs).
Due to the number of edits necessary to address the initial mistake and their importance to the article, the journal determined republication of the article would allow readers to follow the article more effectively than a separate notice of the changes. Appended to the end of this republication notice is a watermarked version of the Online First article as published on July 7, 2020, so that interested readers may reference the original version of the article and note changes.
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 42- Despite Efficiencies, Mergers and Acquisitions Reduce Firm Value by Hurting Customer Satisfaction. By: Umashankar, Nita; Bahadir, S. Cem; Bharadwaj, Sundar. Journal of Marketing. Mar2022, Vol. 86 Issue 2, p66-86. 21p. 1 Diagram, 8 Charts, 2 Graphs. DOI: 10.1177/00222429211024255.
- Database:
- Business Source Complete
Despite Efficiencies, Mergers and Acquisitions Reduce Firm Value by Hurting Customer Satisfaction
Most researchers focus on the effect of mergers and acquisitions (M&As) on investor returns and overlook customer reactions, despite the fact that customers are directly impacted by these corporate transformations. Others suggest that in M&A contexts, a dual emphasis of customer satisfaction and firm efficiency is both likely and beneficial. In contrast, the authors demonstrate that M&As not only do not yield a dual emphasis but also cause a decline in customer satisfaction to the extent that they eclipse any gain in firm value from an increase in firm efficiency. A quasiexperimental difference-in-differences analysis and an instrumental variable panel regression provide robust evidence for the dark side of M&As for customers. The authors use the attention-based view of the firm to demonstrate that post-M&A customer dissatisfaction occurs because of a shift in executive attention away from customers and toward financial issues. In line with the related upper echelons theory, they find that marketing representation on a firm's board of directors helps maintain executive attention on customers, which mitigates the dysfunctional effect of M&As on customer satisfaction. This research identifies a negative M&A–customer satisfaction relationship and highlights executive attention to customer issues and marketing leadership as factors that mitigate this negative relationship.
Keywords: American Customer Satisfaction Index; attention-based view; board of directors; customer satisfaction; difference-in-differences; marketing in C-suite; mergers and acquisitions (M&As); upper echelons theory
Firms engage in mergers and acquisitions (M&As) to obtain assets, grow, reduce costs, and stave off competition ([ 6]; [63]). Yet, many M&As fail to generate positive results ([57]; [61]). Although prior research has explained the underperformance of M&As with deal- and firm-related factors, the role of customer reactions has largely been neglected. This is alarming given that customer growth is a key motivation for M&As ([18]) and customers are directly impacted by M&A-based changes to product lines, brands, prices, innovation, and frontline employees.
The sheer enormity of M&A activity (e.g., more than 48,000 deals with a value of $3.7 trillion were transacted globally in 2019, and despite the COVID-19 pandemic, M&A activity declined by only 3% in 2020) suggests that M&As must be rewarding; otherwise, firms would not engage in them. M&As allow firms to reduce prices ([20]) and innovate ([53]), both of which should satisfy customers. Further, M&As enable firms to become more efficient through improvements in scale, scope, and cost savings ([16]; [37]). As a result, M&As are posited to enable a "dual emphasis" in which firms achieve both customer satisfaction and firm efficiency ([62]).
While the link between M&As and firm efficiency is more straightforward, research has not systematically examined the effect of M&As on customer satisfaction. Experimental ([64]) and anecdotal ([65]) evidence suggests that M&As may in fact harm customers. Thus, we question whether M&As actually enable a dual emphasis of firm efficiency and customer satisfaction. Instead, we argue that although M&As might generate firm efficiency, they upset customers considerably, which, in turn, will lower firm value to an extent that any gain in efficiency will be outweighed. We theorize that this is because M&As cause executives to pay more attention to financial issues than to customer ones, which will dissatisfy customers. We contend that marketing representation on the board (MROB) of directors will direct executive attention toward customers during an M&A, which will then lessen a decline in customer satisfaction.
To test our expectation that there is a tension between M&A activity and firm value via competing processes of lower customer satisfaction and higher firm efficiency, we collected data on a panel of firms from 1995 to 2017 from the American Customer Satisfaction Index (ACSI) database. First, we estimated a system of equations to demonstrate that ( 1) M&A activity is associated with a decrease in customer satisfaction, ( 2) M&A activity is associated with an increase firm efficiency, and ( 3) the net effect of a decrease in customer satisfaction and an increase in firm efficiency on firm value is negative. Thus, M&As lower customer satisfaction to the extent that it overshadows any gain in firm value from firm efficiency. Second, to solidly establish a negative effect of M&As on customer satisfaction, we conducted ( 1) a quasiexperimental differences-in-differences (DID) analysis of a treatment group of firms that engaged in M&As and several control groups of firms that did not, ( 2) a conventional panel regression analysis, and ( 3) a long-term analysis. We find strong evidence for a negative M&A–customer satisfaction relationship, which persists for two years post-M&A. Finally, we content-analyzed letters to shareholders to measure executive attention and collected data on MROB. Our instrumental variable panel moderated-mediation analysis provides support for a mediating role of executive attention to customers (vs. finance) and a positive moderating role of MROB.
We contribute to the literature in multiple ways. First, previous research has focused on the effect of M&As on investor returns (e.g., [23]; [44]) and has largely overlooked customer reactions. In fact, a meta-analytic review of 25 years of customer satisfaction research does not report a single result with M&A activity as a driver ([50]). The few studies that have focused on customers (e.g., [62]) have suggested that in M&A contexts, a dual emphasis of customer satisfaction and firm efficiency is both likely and beneficial. In contrast, we demonstrate that M&As not only do not yield a dual emphasis but also cause a decline in customer satisfaction to the extent that they surpass any gain in firm value from an increase in firm efficiency. Although researchers in finance have highlighted the negative ramifications of M&As for acquirers (e.g., [ 2]; [36]), we are the first to empirically establish the negative ramifications of M&As for customers, which we show lowers firm performance.
Second, we verify a negative M&A–customer satisfaction relationship with a DID analysis with multiple control groups. As a result, we add to emerging research (e.g., [25]) on the use of observational inference to document the causal effects of strategic decisions. We also confirm this negative relationship with an instrument variable panel regression with a larger sample and a long-term analysis. Our multimethod approach offers future research a template with which to improve the reliability and validity of findings from secondary research.
Third, to study M&A outcomes, work in finance has relied on the efficient market theory, and work in marketing has relied on the resource-based view (RBV) of the firm. Rather, in a novel direction, we draw on the attention-based view (ABV) of the firm to argue the impact of M&A activity. Thus, we add to recent work on the marketing–finance interface (e.g., [19]) by showing that when faced with boundary-altering strategic decisions, executives tend to focus more on financial issues than on customer issues, which then indirectly lowers performance.
Finally, marketing researchers have typically overlooked board of director composition, despite the fact that marketers on the board help shape a firm's strategic direction ([70]). We address this gap by complementing the ABV of the firm with the upper echelons theory to demonstrate that firms with (vs. without) marketers on their board of directors help channel executive attention to customers (vs. financial issues). This, in turn, helps minimize customers' post-M&A dissatisfaction. Here, we identify marketing leadership's important role in the marketing–finance interface during disruptive strategic transformations such as M&As. As a result, we are the first to incorporate marketing's role on the board into theories about M&As and customer satisfaction.
In terms of our practical contributions, we caution executives against pursuing M&As to gain efficiencies without considering how customers may be harmed. This is because the negative effect of M&As on customer satisfaction lasts for at least two years. In particular, we show that during an M&A, firms that pay greater attention to their customers relative to financial issues experience a 45% reduction in loss of firm value. As a solution, we recommend that firms have at least one marketer on their boards of directors to retain executive attention on customers, which translates into a gain in firm value of 4.28%.
As we show in our literature review in Table 1, we distinguish our research from prior work in four important ways. First, although prior work has investigated the effect of M&As on firm efficiency, we are the first to also consider the effect of M&As on customer satisfaction to determine their overall effect on firm value. Second, we examine the effect of M&As on customer satisfaction with multiple data structures and models across multiple industries and years to make causal inferences. Third, while previous research in finance has overwhelmingly relied on the efficient market hypothesis and those in marketing have relied on the RBV of the firm, we introduce the ABV of the firm to an M&A context. Finally, those who have used upper echelons theory have overlooked the role of marketing leadership in managing M&As and driving customer satisfaction. We address these gaps by proposing and demonstrating that MROB weakens the negative impact of M&As on executive attention to customer (vs. financial) issues. We depict our conceptual framework in Figure 1.
Graph: Figure 1. Conceptual framework.
Graph
Table 1. A Review of the Literature on M&As.
| Research | M&A → Firm Efficiency | M&A → Customer Satisfaction | Net Effect of M&A on Performance | Board Representation | Multiple Industries | Sample Size | Time Period | Analysis | Underlying Theory |
|---|
| Vennet (1996) | Yes | No | No | No | No | 492 | 1988–1993 | Univariate comparison of pre- and post-M&A | Managerial efficiency theory |
| Agrawal, Jaffe, and Mandelker (1992) | Yes | No | No | No | Yes | 937 | 1955–1987 | Event study | Efficient market theory |
| Loughran and Vijh (1997) | Yes | No | No | No | Yes | 947 | 1970–1989 | Event study | Efficient market theory |
| Avkiran (1999) | Yes | No | No | | No | 4 | 1985–1995 | Case study | Traditional merger theory |
| Capron and Hulland (1999) | No | No | No | No | No | 253 | Survey in 1994 | OLS | RBV of the firm |
| Maksimovic and Phillips (2001) | Yes | No | No | No | No | NA | 1972–1992 | OLS | Neoclassical model of firm organization |
| Fuller, Netter, and Stegemoller (2002) | Yes | No | No | No | Yes | 3,135 | 1990–2000 | Event study | Efficient market theory |
| Moeller, Schlingemann, and Stulz (2004) | Yes | No | No | No | Yes | 12,023 | 1980–2001 | Event study | Efficient market theory |
| Homburg and Bucerius (2005) | Yes | No | Yes | No | Yes | 232 | Survey in 2002 | LISREL | RBV of the firm |
| Prabhu, Chandy, and Ellis (2005) | No | No | No | No | No | 157 | 1988–1997 | Error-component regression | Knowledge-based view of the firm |
| Sorescu, Chandy, and Prabhu (2007) | Yes | No | No | No | No | 238 | 1992–2002 | OLS | RBV of the firm |
| Bahadir, Bharadwaj, and Srivastava (2008) | No | No | No | No | Yes | 133 | 2001–2005 | Heckman two-step | RBV of the firm |
| Cummins and Xie (2008) | Yes | No | No | No | No | 150 | 1994–2003 | Malmquist analysis | Corporate control theory |
| Swaminathan, Murshed, and Hulland (2008) | Yes | No | No | No | Yes | 206 | 1990 to 2001 | OLS | RBV of the firm |
| Thorbjørnsen and Dahlén (2011) | No | No | No | No | Yes | around 1,000 | Hypothetical experiments | OLS | Perceived reactance |
| Wiles, Morgan, and Rego (2012) | Yes | No | No | No | Yes | 572 | 1994–2008 | Event study | RBV of the firm |
| Swaminathan et al. (2014) | Yes, only as a moderator | Yes, only as a moderator | No | No | Yes | 429 | 1995–2003 | Random-effects GLS | RBV of the firm |
| Rao, Yu, and Umashankar (2016) | No | No | No | No | No | 1,979 | 1992–2008 | Matching model | RBV of the firm |
| Saboo et al. (2017) | Yes | No | No | CEO background | No | 319 | 1995–2013 | Event study | RBV of the firm |
| Bommaraju et al. (2018) | No | No | No | No | No | 2,512 | Longitudinal survey | DID | Internal marketing theory |
| This article | Yes | Yes | Yes | Yes | Yes | 2,152 | 1995–2017 | SUR, Instrument variable panel regression, DID | ABV of the firm, upper echelons theory |
1 Notes: RBV = resource-based view; ABV = attention-based view; OLS = ordinary least squares; GLS = generalized linear model; DID = difference-in-differences; SUR = seemingly unrelated regression.
Although there is sparse formal research on M&As and customer satisfaction, some work suggests a positive relationship. M&As can expand firms' product portfolios to provide customers with a larger set of choices ([12]) and higher-quality products ([33]). This supports Swaminathan et al.'s (2014) assertion that M&As are associated with higher customer satisfaction. In contrast, other research suggests that M&As may dissatisfy customers. In particular, M&As can result in price increases ([32]) and poor customer service ([60]). For example, the recent sale of DirecTV by AT&T to the private equity firm TPG for a third of the acquired price in 2015 was largely attributed to the loss of dissatisfied customers postacquisition ([34]). Moreover, anecdotes from the ACSI reveal that even two years after M&As, customers are less satisfied than they had been before ([ 1]). In particular, M&As may cause customers to lose access to their favorite brands. A recent survey by PwC shows that as firms become larger after an M&A, they tend to lose grip of their customers' feelings, and, as a result, customer experience suffers ([52]). This is detrimental because customer dissatisfaction lowers firm value and increases firm risk ([21]; [39]; [43]; [50]; [66]). Thus, we expect that M&As will dissatisfy customers.
The strategy literature suggests that a primary motivation for firms to engage in M&As is to gain efficiencies ([37]). M&As increase firm efficiency by spreading fixed costs over more output and eliminating redundancies ([12]). Specifically, M&As result in economies of scale and scope, asset and employee rationalization, a reduction in transaction costs ([14]), and a reallocation of intangible assets ([45]). These extra resources allow firms to reallocate their savings to other valuable projects, which, in turn, increases firm value ([43]). Thus, consistent with prior research, we expect that M&As will increase firm efficiency. This brings us to two competing outcomes of M&A activity:
- H1: M&As are associated with (a) a decrease in customer satisfaction but (b) an increase in firm efficiency.
A logical follow-up question is, what is the total effect of M&As on firm value given our opposing expectations of a decline in customer satisfaction but an increase firm efficiency? We expect that M&As will cause a steeper drop in customer satisfaction than an increase in firm efficiency for the following reasons. First, M&As often result in layoffs to reduce redundancies, which—while beneficial from an efficiency perspective—harms customer experience. The remaining employees that are not laid off are likely to be stressed ([11]), and stressed employees and their dissatisfaction with a major corporate shake-up negatively affect customers and the service they experience ([52]). Second, firms may either change or consolidate procedures such as credit policies, payment terms, and loyalty programs during an M&A to minimize the complexity of managing two separate systems. While these actions may be efficient, customers are likely to see their hard-earned privileges curtailed or, in the extreme, taken away ([64]), which results in relationship uncertainty ([29]). In fact, customers defect even before they know exactly how an M&A will affect them ([42]). Thus, customers who face poorer service and a loss of privileges will feel negatively about their relationship with a post-M&A firm. Third, customer dissatisfaction attracts short sellers, whose trading hurts firm value ([39]). Thus, we expect that a decline in customer satisfaction will be larger than an increase in efficiency, and as a result, firm value will decline. We test this notion in our estimation.
So far, we have argued that although M&As generate efficiencies, their negative impact on customer satisfaction is significant and noteworthy, yet underresearched. Next, we focus on the negative M&A–customer satisfaction relationship and aim to uncover a mechanism that drives this relationship. The marketing literature has often adopted the RBV of the firm view to examine M&A activity (Table 1). This research stream argues that a firm's ability to acquire and deploy marketing resources during an M&A can strengthen performance. Although the RBV provides a valuable strategic lens with which to examine M&A activity, another theoretical process may also be at play. We use the ABV to argue why M&As lower customer satisfaction.
The ABV highlights the importance of executives' information-processing capacity and their distribution of attention. "Attention" refers to as a focus of time and effort with making sense of a firm's environment and how to respond to it ([48]). A key premise is that executives' attention is finite, so they are selective in what they notice and interpret. Further, how they respond to stimuli depends on what they notice and interpret in the first place. Thus, what executives pay attention to affects their resource allocation ([10]), which suggests that executives will invest resources in what they pay attention to at the expense of what they ignore. Further, attention drives executives to match their firms' resources with opportunities in their environment (Vadakeppatt et al. [67]; [72]). We draw on the ABV to argue that M&A activity directs executives' attention away from customers and toward financial issues, which, in turn, reduces the extent of resources allocated toward satisfying customers.
M&As are incredibly expensive, complex, and heavily scrutinized by investors. As a result, executive attention is likely to be diverted to the price of the deals, capital requirements, paying back debt providers, and appeasing investors. In the process, customer experience might be underinvested in or even overlooked. In fact, managers know that there is a trade-off between serving customers and serving shareholders/debtholders such that creating value for one can detract from the other, and vice versa ([58]). For example, H.J. Heinz purchased Kraft Foods for nearly $36 billion in 2015. At the behest of investors, the merged company slashed $1.8 billion in overhead, which included a purge of nearly 2,500 jobs. Then, after Kraft Heinz's post-M&A sales slowed,[ 6] investors pressured it to acquire a large consumer products company to gain market share ([55]). Firms clearly face considerable financial pressure after an M&A, which can cause executives to focus on appeasing investors at the expense of customers.
Further, M&As are often paid for by corporate debt. Recent examples of extensive borrowing for M&As include established companies such as CVS, IBM, Campbell's, Bayer, and Sherwin-Williams ([15]). Debt can turn executives' attention toward loan-servicing obligations, conserving cash rather than investing ([ 3]), and cost cutting ([38]). Debt also limits investments in advertising ([27]) and product quality ([40]). Thus, executives at indebted M&A firms may focus on satisfying debtholders over customers. Therefore, we hypothesize,
- H2: M&As are associated with less executive attention to customers (vs. financial issues), which is associated with lower customer satisfaction.
A key premise that the upper echelons theory ([28]) and the ABV ([48]) share is that the focus of executives' attention drives firm strategy and resource allocation. We use these complementary theories to examine how MROB influences executive attention toward customer-related issues during M&As. If executives pay more attention to, for example, innovation, then they allocate more resources toward innovation-related activities to drive success ([72]; [75]). Similarly, we argue that MROB will direct executive attention toward building organizational resources and processes, directing capabilities, and mobilizing employees to meet customers' needs during M&As.
A firm's board of directors is a key body of leadership at its apex. It is both a governance body and a strategic body that sets a firm's goals and advises executives on how to pursue these goals ([ 9]). While executives are responsible for formulating strategies given a set of objectives, they do not determine these objectives ([24]). Rather, such objectives, which include growth or cost cutting, are usually made at the board level. As a result, a board of directors is heavily involved in the M&A process due to its transformative corporate consequences ([30]). A less researched, but critically important, type of board member is one who has a marketing title. Given their expertise in customer orientation, they provide marketing-related advice to other members on the board and the executive team, which ensures that firm strategies are customer-centric ([70]). We examine how MROB influences the relationship between M&As and customer satisfaction through a shift in executive attention.
The upper echelons theory states that the characteristics of a firm's top leaders influence its strategic decisions and outcomes ([28]). Leaders' backgrounds create a lens through which they view business challenges and determine the strategies needed to address them ([17]). In particular, executive attention is ( 1) channeled toward issues of greater value or legitimacy for the firm, ( 2) evaluated through the lens of an executive's functional role, and ( 3) influenced by the environment ([22]). Given that financial issues dominate executives' attention during M&As, we contend that marketers on the board will serve as "customer attention custodians" to channel resources toward addressing any challenges faced by customers. They will do so by diffusing a customer-focus throughout the organization to mobilize employees to proactively attend to customers that face a disruptive context. Given that firms perform poorly in areas in which their board members have limited expertise ([41]), if there is no MROB, then customer-related issues are likely to be ignored or possibly mismanaged by others ([ 9]; [70]). Thus, we expect that marketers on the board will make customers a part of the conversation M&As largely because they are trained to do so.
Scholars have typically relied on a resource-based perspective when they examine the board of directors' impact on firm strategy (e.g., [ 9]). We contend that the functional role of a board member influences not only whether role-related resources are conferred to the rest of the board and the firm but also what the board member interprets in the environment and encourages others to pay attention to. In other words, we expect that during an M&A, MROB will minimize a depletion of executive attention on customers and marketing-related issues. Therefore,
- H3: The negative effect of M&As on executive attention to customers (vs. financial issues) is smaller when there is (vs. is not) MROB, which is associated with less customer dissatisfaction.
We drew our estimation sample from the ACSI database, which is a credible source for our primary outcome, customer satisfaction ([21]; [43]; [66]). We based our main analysis on a cross-sectional time series data set of 1,359 firm-year observations for 141 firms from 1995 to 2017. To identify the impact of M&As on customer satisfaction, we transformed this panel to a clean four-year rolling-window data structure, which we detail subsequently. Similar to prior research with multimethod studies (e.g., [51]), our sample sizes differ across different data structures and model specifications.
We summarize our variables and data sources in Table 2.
Graph
Table 2. Operationalization of Variables.
| Variable | Measure | Data Source |
|---|
| Firm Value | Number of shares outstanding × Share price | Center for Research in Security Prices |
| Firm Efficiency | Sales/Number of employees | Compustat |
| Customer Satisfaction (CSAT) | ACSI scores, which range from 0 to 100 | ACSI Database |
| M&A Activity | 1 = firm engaged in an M&A in year t, 0 = firm did not | SDC Platinum |
| M&A Count | Number of M&As that a firm engaged in in year t | SDC Platinum |
| Executive Attention to Customers (vs. Finance) | Number of customer-related words/Number of finance-related words in letters to shareholders | Annual Reports (EDGAR) |
| Marketing Representation on the Board (MROB) | Number of board members with a marketing title/Board size.Variables for selection model for MROB: Peer Firm Mean MROB: average number of MROB members for all firms in the focal firm's industry Mean Board Age: average age of board members CMO on TMT: CMO listed among TMT = 1, 0 otherwise Mean Board Tenure: average years board members have served on the focal board Board Size: Total number of board members CEO Duality: CEO holds title of board chair = 1, CEO and board chair are separate = 0 Female Percentage: percent of female board members
| S&P Capital IQ Professional Database |
| Market Share(t) | Firm sales(t)/(Total industry sales at the four-digit SIC-level)(t) | Compustat |
| Advertising/Sales(t) | Advertising expenditures(t)/Sales(t) | Compustat |
| R&D/Sales(t) | R&D Expenditures(t)/Sales(t) | Compustat |
| Firm Size(t) | Natural logarithm of total assets(t) in CSAT models and natural logarithm of employees(t) in other models | Compustat |
| Segments(t) | Natural logarithm of number of different four-digit SIC industries in which the firm operates | Compustat |
| ROA(t) | Operating income before depreciation(t)/Total assets(t − 1) | Compustat |
| Market Growth(t) | Average of four-digit SIC industry year-over-year sales growth over four years preceding year t | Compustat |
| Competitive Intensity(t) | Reciprocal of the Herfindahl–Hirschman index | Compustat |
| Industry ROA(t) | Four-digit SIC-level operating income before depreciation(t)/ Total assets(t − 1) | Compustat |
| Restructuring Charges | The sum of restructuring charges in years t and t − 1 scaled by market capitalization of a firm in year t | Compustat |
| Firm Scope | The number of distinct four-digit SIC business segments that a firm operates in | Compustat/Segment Database |
2 Notes: R&D = research and development; ROA = return on assets; SIC = Standard Industrial Classification.
Similar to prior research (e.g., [43]), when multiple brands were represented in the ACSI database, we averaged their scores to create a firm-level annual score of customer satisfaction, or CSAT. We measured firm efficiency by dividing a firm's annual sales by its number of employees ([ 4]). We measured firm value with market value, or a firm's number of outstanding shares multiplied by its share price, which represents investors' expectations of a firm's profit potential ([19]).
Consistent with previous research (e.g., [54]; [73]), if a firm-year was present in the SDC Platinum database, then we designated that firm-year as having M&A activity (i.e., 100% ownership). If a particular firm-year was not present, then we assumed that this firm did not engage in M&A activity that year, and we used this information to create a group of non-M&A firms. Thus, we coded M&A activity as 1 if a firm engaged in M&A activity that year and 0 if it did not (for our list of M&A firms, see Web Appendix A).
We followed prior research (e.g., [51]) to assess executives' attention directed at theoretically relevant issues with a count of specific types of words from their letters to shareholders. To compile a dictionary of customer-related words, we began with [72]) dictionary of external focus and expanded their list based on a review of popular press announcements of M&As. To compile a dictionary of finance-related words, we reviewed popular press articles, finance M&A papers, and finance textbooks (we present our dictionary in Web Appendix B). We counted the number of words from these two dictionaries and created a ratio, attention to customers (vs. finance), by dividing the total number of customer words by the total number of finance words.
To measure MROB, we created a list of marketing titles in top management based on research by [47] and [70]. We then counted the total number of people with marketing titles on the board and divided this by the total size of the board for each firm-year.
In the CSAT model, we included market share, profitability, advertising intensity, R&D intensity, firm size, number of segments, and market growth ([38]; [56]). In the firm efficiency and firm value models, we included restructuring charges, firm scope, competitive intensity, industry profitability, firm size, market share, firm size, and market growth ([35]).
We used three steps to test our hypotheses. First, we estimated a seemingly unrelated regression (SUR) model to test the effect of M&A activity on customer satisfaction (H1a) and firm efficiency (H1b) and the overall effect of M&A activity on firm value via customer satisfaction and firm efficiency. Second, we tested the negative M&A–customer satisfaction relationship (H1a) with ( 1) a quasiexperimental DID approach ([26]) with multiple control groups, ( 2) an instrumental variable panel regression, and ( 3) a long-term analysis. Third, we implemented a moderated-mediation SUR model to test whether the negative M&A–customer satisfaction relationship is mediated by executive attention to customers (vs. financial issues) (H2) and whether MROB shifts executive attention back toward customers (H3).
We created clean four-year rolling windows to include firms that had no M&A activity two years before an M&A and no M&A activity one year after. This enabled us to isolate the effect of M&A activity without confounding it with previous activity, because the effect of M&As tends to spill over to subsequent years ([68]). For example, in our first window for the M&A group, we included firms that engaged in M&As in 1997 but did not engage in M&As in 1995, 1996, and 1998. In the next window, for the M&A-year of 1998, we included firms that engaged in M&A activity in 1998 but not in 1996, 1997, and 1999. If a firm did not engage in any M&A activity during the four years (e.g., 1995–1998), then we included this firm in a non-M&A group. Overall, we had 119 firms in this sample.
We estimated the following three models: ( 1) the effect of M&As on customer satisfaction (H1a), ( 2) the effect of M&As on firm efficiency (H1b), and ( 3) the effects of customer satisfaction and firm efficiency on firm value. We used the natural logarithmic values of our continuous variables of customer satisfaction, firm efficiency, and firm value to produce elasticities, which enabled us to compare the relative effects of customer satisfaction and firm efficiency on firm value. We included control variables that have been shown to influence CSAT ([38]; [56]), firm efficiency, and firm value ([35]). We winsorized the continuous variables before estimating the model to remove the potential effect of outliers and included fixed effects to account for unobservable firm characteristics. Given that M&A activity may simultaneously affect both customer satisfaction and firm efficiency, we estimated these relationships as a system of equations with SUR. We estimated the following system of equations for firm i at time t:
Graph
(1.1)
Graph
(1.2)
Graph
(1.3)
where the M&A Group variable, j, has a value of 1 for the M&A group and 0 for the non-M&A group, and the Post-M&At variable has a value of 1 in the fourth year of each window (i.e., the post-M&A year). The interaction between M&A Group and Post-M&A has a value of 1 for the M&A firms and a value of 0 for the non-M&A firms in the post-M&A year. Therefore, φ3 (β3) represents the statistical effect of M&As on CSAT (firm efficiency). Finally, Θ1 (Θ2) is the effect of CSAT (firm efficiency) on firm value.
We first present model-free evidence of the effect of M&As on customer satisfaction and firm efficiency. For the M&A firms, customer satisfaction decreases a year after the M&A, whereas for the non-M&A firms, it increases (Figure 2, Panel A). In contrast, for the M&A firms, firm efficiency increases a year after the M&A, whereas for non-M&A firms, it remains steady (Figure 2, Panel B). Further, the average change in CSAT (ΔNon-M&A CSAT(t + 1, t − 1) = .43; ΔM&A CSAT(t + 1, t − 1) = −.14, p < .05) and firm efficiency (ΔNon-M&A Firm Efficiency(t + 1, t − 1) = 13.39; ΔM&A Firm Efficiency(t + 1, t − 1) = 71.83, p < .05) between the two groups are different.
Graph: Figure 2. M&A customer satisfaction and firm efficiency trends.
We present the descriptive statistics and correlations of our variables in Web Appendix C. Our SUR estimation results (Table 3) of Equations 1.1–1.3 demonstrate that M&As are associated with a decrease in customer satisfaction (φ3 = −.010, p < .05; H1a is supported) and an increase in firm efficiency (β3 = .070, p < .01; H1b is supported).
Graph
Table 3. Effect of M&As on CSAT, Firm Efficiency, and Firm Value.
| Dependent Variable | CSATt | Firm Efficiencyt | Firm Valuet |
|---|
| Focal Variables | | | |
| Constant | 3.929*** | .599*** | −1.194 |
| (.024) | (.205) | (1.375) |
| CSATt | | | 2.214*** |
| | (.325) |
| Firm Efficiencyt | | | .838*** |
| | (.062) |
| M&A Group | −.001 | −.067*** | |
| (.004) | (.017) | |
| Post-M&A Year | .005** | .045*** | |
| (.002) | (.015) | |
| M&A Group × Post-M&A Year (H1a–b) | −.010** | .070** | |
| (.004) | (.030) | |
| Covariates | | | |
| Restructuring Charges(t − 1) | | −.263*** | –3.564*** |
| (.095) | (.806) |
| Firm Scope(t − 1) | | .082*** | .009 |
| (.020) | (.048) |
| Competitive Intensity(t − 1) | | .256*** | −.043 |
| (.033) | (.049) |
| Industry Profitability(t − 1) | | −.002** | −.004 |
| (.001) | (.005) |
| Market Share(t − 1) | −.073*** | .350* | −.313 |
| (.012) | (.192) | (.347) |
| Firm Profitability(t − 1) | .169*** | −.451*** | 1.837*** |
| (.019) | (.160) | (.394) |
| Firm Size(t − 1) | .025*** | −.200*** | .493*** |
| (.003) | (.038) | (.077) |
| Market Growth(t) | −.063*** | −.526*** | −.147 |
| (.015) | (.075) | (.211) |
| Advertising/Sales(t − 1) | .021** | | |
| (.010) | | |
| R&D/Sales(t − 1) | −.091*** | | |
| (.016) | | |
| Segments(t − 1) | −.010*** | | |
| (.003) | | |
| Firm fixed effects | Included | Included | Included |
| Time effects | Included | Included | Included |
| Model Information | | | |
| χ2 | χ2(130) = 7,918.48*** | χ2(131) = 61,482.97*** | χ2(131) = 15,867.32*** |
| R2 | .761 | .961 | .865 |
| Number of firms | 119 | 119 | 119 |
| Observations | 2,468 | 2,468 | 2,468 |
- 3 *p < .10.
- 4 **p < .05.
- 5 ***p < .01.
- 6 Notes: We report parameter estimates with bootstrapped standard errors in parentheses.
Given the asymmetric findings of a decline in customer satisfaction but an increase in firm efficiency from M&A activity, we next focus on the net effect of M&As on firm value through customer satisfaction and firm efficiency. From Table 3, we see that the positive association between customer satisfaction and firm value (Θ1 = 2.214, p < .01) is greater than the positive association between firm efficiency and firm value (Θ2 = .838, p < .01). To calculate the net effect of M&A activity on firm value via customer satisfaction and firm efficiency, we used the results from Equations 1.1 and 1.2. On average, the customer satisfaction of the M&A firms was 1.14% lower than the non-M&A firms (φM&A = φ1 + φ3 = −.0114 = −.0012 + −.0102) and the firm efficiency of the M&A firms was.29% higher than the non-M&A firms (βM&A = β1 + β3 = .0029 = −.0668 + .0697). We multiplied the M&A firms' customer satisfaction and firm efficiency elasticities for firm value from Equation 1.3 with the differences between the M&A and non-M&A firms in the post-M&A year from Equations 1.1 and 1.2. Then, we summed the products and found a net effect of −.0243. Therefore, compared with non-M&A firms, M&A firms' value decreased by 2.43% one year after an M&A, and as a result, the net-negative effect of M&As on firm value is due to a decrease in customer satisfaction.
Because we found that, despite gains in firm efficiency, M&As decrease firm value due to a decline in customer satisfaction, we aimed to validate the latter effect more systematically with several approaches. First, we estimated a quasiexperimental DID model with alternate non-M&A firm groups. Second, we created a panel of firms without imposing restrictions on which firms to include (i.e., we included all of the firms from the ACSI database). Third, we tested for the long-term negative effect of M&As on customer satisfaction.
We used with the same four-year rolling window data structure that we previously described. We assigned firms to an M&A treatment group if they engaged in M&A activity in the third year of a four year window and assigned all firms that did not engage in any M&As during those four years to a non-M&A control group (control group 1). For greater reliability, we created two alternative control groups by ( 1) matching the M&A and non-M&A firms on similar predictors of customer satisfaction (control group 2) and ( 2) matching the M&A and non-M&A firms on their propensity to engage in an M&A (control group 3) (for more information, see Tables D.1–D.3 in Web Appendix D). We specified the following model with a fixed-effects error component ([ 8]):
Graph
( 2)
where υi captures unobserved time-invariant firm characteristics. The M&A Group variable, j, has a value of 1 for the treatment group and a 0 for the control group. The Post-M&At variable has a value of 1 in the fourth year of each window (i.e., one year post-M&A). We used a fixed-effects within estimator to eliminate all time-invariant variables, such as υi and M&A Groupj. The interaction between M&A Group and Post-M&A has a value of 1 for the M&A firms and a 0 for the non-M&A firms the year after the M&A. Therefore, β2 is the statistical effect of M&As on CSAT.
We created a conventional panel data setup to test the effect of M&As on customer satisfaction without a four-year rolling-window data restriction; as a result, our sample increased to 2,152 observations for 204 firms, of which 153 engaged in M&A activity. Because this sample includes firms for which there are several years of data (e.g., more than ten years), we used a one-year change in CSAT as our dependent variable. We created two versions of M&A activity: ( 1) a dummy variable that had a value of 1 if a firm engaged in M&A activity in year t and 0 if it did not and ( 2) the natural logarithm of the number of M&As a firm engaged in in year t.
We estimated a selection equation in which our dependent variable was a firm's decision to engage in an M&A (0/1) and our predictors were factors that relate to M&A decisions (e.g., debt-to-equity ratio, competitors' M&A activity) but not to customer satisfaction (see Equation D.3 and Table D.5 in Web Appendix D). Thus, we achieved identification in Equation 3 and our subsequent Equation 4 based on our separation of factors that drive M&A decisions versus those that drive customer satisfaction. Based on this selection model, we included an inverse Mills ratio (IMR) in Equations 3 and 4 and estimated the following model with a fixed-effects within estimator to account for unobservable firm characteristics for firm i in year t:
Graph
( 3)
We investigated the long-term effect of M&As on customer satisfaction with a conventional panel data structure. We computed a change in CSAT from calendar year t + 4 to t as our dependent variable and included firms' M&A activity at t + 1, t + 2, and t + 3 as our independent variables. We included an IMR for each year in the model to control for selection bias. We estimated this panel data model with a fixed-effect within estimator:
Graph
( 4)
In line with DID requirements ([26]), we compared the observable drivers of customer satisfaction between the M&A treatment group and the non-M&A control group two years before the M&A and found that for five of the seven drivers of M&As, the two groups were statistically similar (Table 4, Panel A; we present this graphically in Figures D.1–D.3 in Web Appendix D). We also tested the equality of changes in the drivers of customer satisfaction two years before the M&A through the M&A year and did not find any differences between the two groups. Thus, any distinction in post-M&A customer satisfaction between the two groups was not likely to be caused by firm-level differences, and the parallelness assumption was satisfied for the observable drivers of customer satisfaction. Next, we compared the two-year average customer satisfaction of the M&A and non-M&A groups before the M&A (t = .54, p > .10) and the equality of changes in customer satisfaction two years before the M&A through the M&A year to satisfy the parallelness assumption (t = 1.31, p > .10) and did not find any significant differences (Table 4, Panel B). Finally, the M&A firms' customer satisfaction was lower than the non-M&A firms' customer satisfaction a year after the M&A (t = 2.24, p < .05). We replicated these results for our alternate control groups (Table D.4 in Web Appendix D).
Graph
Table 4. A Comparison Between M&A and Non-M&A Groups.
| A: Comparison of Drivers of Customer Satisfaction Between M&A and Non-M&A Groups |
|---|
| M&A Group(Treatment)(n = 91) | Non-M&A Group(Control)(n = 619) | t-Test of Equality of Means |
|---|
| Comparison of Pre-M&A Averages of Drivers of CSAT | | | |
| Market Share | .15 | .16 | t = .96, p > .10 |
| ROA | .14 | .14 | t = .52, p > .10 |
| Firm Size ($ million) | 44,598.20 | 47,568.90 | t = .29, p > .10 |
| Advertising/Sales | .03 | .04 | t = 2.23, p < .05 |
| R&D/Sales | .04 | .03 | t = −1.14, p > .10 |
| Segments | 1.39 | 1.37 | t = −.24, p > .10 |
| Market Growth | .08 | .05 | t = −2.80, p < .01 |
| Comparison of Pre-M&A Changes in Drivers of Customer Satisfaction | | | |
| Pre-M&A Change in Market Share(t, t − 2) | −.001 | .001 | t = .28, p > .10 |
| Pre-M&A Change in ROA(t, t − 2) | −.009 | −.002 | t = 1.29, p > .10 |
| Pre-M&A Change in Firm Size(t, t − 2) | 1,154.27 | 1,344.63 | t = .12, p > .10 |
| Pre-M&A Change in Advertising/Sales(t, t − 2) | −.004 | −.005 | t = −.29, p > .10 |
| Pre-M&A Change in R&D/Sales(t, t − 2) | −.011 | .007 | t = 1.25, p > .10 |
| Pre-M&A Change in Segments(t, t − 2) | .008 | .004 | t = −.11, p > .10 |
| Pre-M&A Change in Market Growth(t, t − 2) | −.024 | −.008 | t = 1.49, p > .10 |
| B: Comparison of Customer Satisfaction Between M&A and Non-M&A Groups |
| M&A Group(Treatment)(n = 91) | Non-M&A Group(Control)(n = 619) | t-Test of Equality of Means |
| Pre-M&A Periods | | | |
| Customer Satisfaction(t − 2) | 76.1 | 76.5 | t = .39, p > .10 |
| Customer Satisfaction(t − 1) | 76.0 | 76.4 | t = 1.00, p > .10 |
| Average Customer Satisfaction(t − 1, t − 2) | 76.1 | 76.5 | t = .71, p > .10 |
| Change in Customer Satisfaction(t, t − 2) | −.28 | .19 | t = 1.31, p > .10 |
| Post-M&A Period | | | |
| Customer Satisfaction(t + 1) | 75.8 | 77 | t = 2.24, p < .05 |
7 Notes: t denotes the year of the M&A activity.
We present the estimation results of Equation 2 in Table 5. When we estimated Equation 2 with only the post-M&A variable and its interaction with M&A Group, we find that M&As caused a decline in customer satisfaction (β2 = −.635, p < .05; Model 1), which remained consistent with the inclusion of our control variables (β2 = −.754, p < .01; Model 3). We also estimated a model with only control variables (Model 2).
Graph
Table 5. DID Results of M&As and Customer Satisfaction with Multiple Control Groups.
| Dependent Variable: CSATt |
|---|
| Models | (1) Control Group 1:Only M&ATreatment | (2) Control Group 1: Only covariates | (3) Control Group 1:Full Model | (4) Control Group 2 | (5) Control Group 3 |
|---|
| Focal Variables | | | | | |
| Constant | 76.428*** | 76.622*** | 76.411*** | 76.565*** | 77.915*** |
| (.095) | (.605) | (.565) | (.996) | (1.234) |
| Post-M&A Year | .467*** | | .466*** | .356* | .576** |
| (.114) | | (.104) | (.207) | (.262) |
| M&A Group × Post-M&A Year (H1a) | −.635** | | −.754*** | −.622** | −.688** |
| (.316) | | (.290) | (.312) | (.344) |
| Covariates | | | | | |
| Market Share(t − 1) | | –2.315 | –1.991 | –1.285 | –2.561 |
| (1.599) | (1.493) | (2.218) | (1.952) |
| ROA(t − 1) | | 3.682* | 3.762** | 1.362 | –4.577** |
| (1.932) | (1.748) | (2.904) | (1.948) |
| Firm Size(t − 1) | | −.000** | −.000* | −.000 | −.000 |
| (.000) | (.000) | (.000) | (.000) |
| Advertising/Sales(t − 1) | | .351 | .371 | 2.458 | .834 |
| (.779) | (.636) | (5.847) | (1.138) |
| R&D/Sales(t − 1) | | –3.571*** | –3.671*** | –4.424** | –1.157 |
| (1.175) | (1.082) | (1.854) | (2.513) |
| Segments(t − 1) | | .163 | .200 | −.001 | −.188 |
| (.311) | (.284) | (.528) | (.649) |
| Market Growtht | | –1.197 | −.893 | .259 | .597 |
| (1.149) | (1.084) | (2.141) | (1.442) |
| Model Information | | | | | |
| Wald χ2 | χ2(2) = 17.09*** | χ2(7) = 24.62*** | χ2(9) = 57.42*** | χ2(9) = 19.90** | χ2(9) = 15.62* |
| R2 | .01 | .02 | .03 | .03 | .03 |
| Number M&A Firms | 67 | 67 | 67 | 67 | 67 |
| Total Number of Firms | 141 | 141 | 141 | 106 | 98 |
| Observations | 2,840 | 2,840 | 2,840 | 932 | 728 |
- 8 *p < .10.
- 9 **p < .05.
- 10 ***p < .01.
- 11 Notes: CSAT = customer satisfaction; ROA = return on assets; R&D = research and development. We report parameter estimates with bootstrapped standard errors in parentheses.
While we accounted for time-invariant unobservable factors with a firm fixed-effects estimator, we tested whether time-varying unobservable factors altered our inference about the M&A treatment effect. We followed a procedure by [49] and used the PSACALC program in STATA. The result of this procedure suggests that our main analysis performed well because when we matched on time-varying unobservable variables, the coefficient estimate of β2 in Equation 2 only changed from −.75 to −.76.
We analyzed Equation 2 with two alternative control groups and report their results in Table 5. When we only included firms that were similar to the focal firm in terms of predictors of customer satisfaction (control group 2), we find that M&As lowered customer satisfaction (β2 = −.622, p < .05; Model 4). When we used the propensity to engage in M&As scores (control group 3), we still find that M&As caused a decline in customer satisfaction (β2 = −.688, p < .05; Model 5). Thus, we find consistent support for a negative effect of M&As on customer satisfaction (H1a) with two alternative control groups.
We present our estimation results of Equation 3 in Models 1a (with a dummy variable for M&A activity) and 2a (number of M&As) in Table 6. We find that M&A activity lowers customer satisfaction (α1M&A Dummy = −2.438, p < .01; α1M&A Number = −.589, p < .01). Thus, we provide additional support for H1a and show that the negative M&A–customer satisfaction relationship is not sensitive to sampling and modeling approaches. The IMR coefficient is significant (α8 = 1.361, p < .01), which suggests that it is necessary to account for firms' propensity to engage in M&As.
Graph
Table 6. Effect of M&As on Customer Satisfaction Over Time.
| M&A Dummy | M&A Number |
|---|
| Dependent Variables: | Model 1aCSAT(t + 1) − CSAT(t) | Model 1bCSAT(t + 4) − CSAT(t) | Model 2aCSAT(t + 1) − CSAT(t) | Model 2bCSAT(t + 4) − CSAT(t) |
|---|
| Focal Variables | | | | |
| Constant | .634* | 1.515 | .203 | .876 |
| (.363) | (1.077) | (.324) | (.966) |
| M&A(t)/(t + 3) (H1a) | −2.438*** | −2.032** | −.589*** | −.727* |
| (.551) | (.807) | (.203) | (.384) |
| M&A(t + 2) (H1a) | | −2.743*** | | −1.092*** |
| (.755) | | (.329) |
| M&A(t + 1) (H1a) | | .100 | | −.426 |
| (.686) | | (.310) |
| Covariates | | | | |
| Market Share(t)/(t + 3) | −.202 | −1.981 | −.500 | −1.996 |
| (.756) | (2.346) | (.767) | (2.138) |
| ROA(t)/(t + 3) | .947 | 2.194 | .839 | 2.052 |
| (1.594) | (2.808) | (1.603) | (2.815) |
| Firm Size(t)/(t + 3) | −.000*** | −.000*** | −.000*** | −.000*** |
| (.000) | (.000) | (.000) | (.000) |
| Advertising/Sales(t)/(t + 3) | 2.337** | 3.252** | 2.414** | 3.227** |
| (1.175) | (1.645) | (1.173) | (1.608) |
| R&D/Sales(t)/(t + 3) | −1.327 | −5.431*** | −1.367 | −5.765*** |
| (1.146) | (1.954) | (1.139) | (1.928) |
| Segments(t)/(t + 3) | .123 | .686 | .083 | .659 |
| (.212) | (.618) | (.210) | (.620) |
| Market Growth(t + 1)/(t + 4) | −.177 | −1.641 | −.713 | −2.455 |
| (1.005) | (1.934) | (1.006) | (1.993) |
| Decision to M&A IMR(t)/(t + 3) | 1.361*** | 1.172** | .271* | .387* |
| (.333) | (.490) | (.151) | (.232) |
| Decision to M&A IMR(t + 2) | | 1.632*** | | .634*** |
| (.437) | | (.197) |
| Decision to M&A IMR(t + 1) | | −.015 | | .297 |
| (.416) | | (.206) |
| Model Information | | | | |
| F-statistic | F(9, 203) = 4.83*** | F(13, 174) = 3.47*** | F(9, 203) = 3.28*** | F(13, 174) = 3.49*** |
| R2 | .013 | .046 | .007 | .044 |
| Number of firms | 204 | 175 | 204 | 175 |
| Observations | 2,152 | 1,722 | 2,152 | 1,722 |
- 12 *p < .10.
- 13 **p < .05.
- 14 ***p < .01.
- 15 Notes: CSAT = customer satisfaction; IMR = inverse Mills ratio; ROA = return on assets; R&D = research and development. We report parameter estimates with cluster robust standard errors in parentheses. In Models 1a and 2a, the M&A variable, the IMR, and the control variables have the subscript (t), except for the market growth variable, which has the subscript (t + 1). In Models 1b and 2b, the M&A variables and their corresponding IMRs have the subscripts (t + 3), (t + 2), and (t + 1). In these models, the control variables have the subscript (t + 3), except for the market growth variable, which has the subscript (t + 4).
We present the estimation results of Equation 4 in Models 1b (M&A dummy variable) and 2b (number of M&As) in Table 6. The negative impact of M&A activity on customer satisfaction persists for two years (γ1M&A Dummy = −2.032, p < .05; γ2M&A Dummy = −2.743, p < .01; γ1M&A Number = −.727, p < .10; γ2M&A Number = −1.092, p < .01). Thus, we find support for H1a even two years after an M&A.
Given that we have established that M&As lower customer satisfaction with multiple methods, we aimed to test whether this decline is due to a shift in executive attention away from customers and toward financial issues (H2) and whether MROB moderates the M&A–executive attention relationship (H3). We used a four-year rolling window data structure and a SUR modeling approach to test these hypotheses. We present the descriptive statistics for this sample in Table 7, Panels A and B.
Graph
Table 7. Summary Statistics and Correlations.
| A: Summary Statistics | | | |
|---|
| Variable | Observations | Mean | SD |
|---|
| CSAT(t) | 2,468 | 75.93 | 5.26 |
| Executive Attention(t) | 2,468 | .28 | .18 |
| MROB(t) | 2,468 | .02 | .04 |
| M&A Group | 2,468 | .12 | .33 |
| Market Share(t − 1) | 2,468 | .16 | .20 |
| ROA(t − 1) | 2,468 | .14 | .106 |
| Total Assets(t − 1) | 2,468 | 45,271.44 | 213,000 |
| Advertising Intensity(t − 1) | 2,468 | .04 | .08 |
| R&D Intensity(t − 1) | 2,468 | .04 | .09 |
| Ln(Segments)(t − 1) | 2,468 | 1.36 | .47 |
| Market Growth(t − 1) | 2,468 | .04 | .08 |
| B: Correlations |
| Variables | (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | (10) |
| 1. CSAT(t) | | | | | | | | | | |
| 2. Executive Attention(t) | −.10* | | | | | | | | | |
| 3. MROB(t) | .13* | .03 | | | | | | | | |
| 4. M&A Group | −.08* | .03 | .03 | | | | | | | |
| 5. Market Share(t − 1) | .23* | −.11* | .09* | −.03 | | | | | | |
| 6. ROA(t − 1) | .08* | −.10* | −.07* | −.00 | .10* | | | | | |
| 7. Total Assets(t − 1) | −.14* | −.02 | −.01 | −.01 | −.04* | −.15* | | | | |
| 8. Advertising Intensity(t − 1) | .05* | .05* | .07* | −.06* | −.03 | .05* | −.04 | | | |
| 9. R&D Intensity(t − 1) | −.17* | .20* | −.03 | −.00 | −.13* | −.15* | −.04* | .08* | | |
| 10. Ln(Segments)(t − 1) | .14* | .15* | .14* | .01 | .19* | −.01 | −.25* | .06* | .13* | |
| 11. Market Growth(t − 1) | −.07* | .014 | −.14* | .09* | −.09* | .10* | −.05* | −.04 | .13* | .02 |
- 16 *p < .10.
- 17 Notes: CSAT = customer satisfaction; MROB = marketing representation on the board; ROA = return on assets; R&D = research and development.
After collecting data on executive attention and MROB, we had a sample of 122 firms. Arguably, executive attention to customers (vs. finance) is endogenous because executives may strategically pay attention to issues that result in better outcomes, such as customer satisfaction. To address this, we used a latent instrumental variable approach ([31]; [35]). Specifically, we used a binary unobserved instrument to separate the observed endogenous predictor into correlated versus uncorrelated components with an error term in Equation 6.2 (for further details, see Web Appendix E).
It is also plausible that MROB is endogenous such that there are systematic differences between firms that appoint a marketer to their boards and those that do not. We estimated a firm's decision to have MROB ([70]) and used board-related variables (peer firm mean MROB, mean board age, chief marketing officer [CMO] on the top management team [TMT], mean board tenure, board size, chief executive officer [CEO] duality, and female percentage) as our exclusion restrictions. We estimated the following random-effects probit model to produce an MROB IMR to include in our main estimation (for the results of Equation 5, see Table E.1 in Web Appendix E):
Graph
( 5)
To test H2 and H3, we estimated a SUR model with moderated mediation by estimating the effect of an M&A on attention to customers (vs. finance) (Equation 6.1) and the effect of the latent instrumental variable, , on CSAT (Equation 6.2). We included time dummies and firm fixed effects to account for unobservable characteristics and an MROB IMR to control for selection bias. We winsorized our continuous variables and estimated the following models:
Graph
(6.1)
Graph
(6.2)
where β3 captures the impact of M&As on executive attention to customers (vs. finance) a year after the M&A and π4 captures the impact of executive attention to customers (vs. finance) on CSAT, which allows us to test H2. β5 captures the moderating impact of MROB on the relationship between the M&A activity and executive attention to customers (vs. finance), which allows us to test H3.
We present model-free evidence of the relationship between M&A activity and executive attention to customers (vs. finance) in Figure 3. M&A firms experience a decline in executive attention to customers relative to financial issues. For example, from our sample we see that for United Airlines, executive attention to customers (vs. finance) declined 38% because of its acquisition of Continental Airlines and its customer satisfaction declined 8.2%.
Graph: Figure 3. Executive attention to customer versus finance trends.
When we compared a change in executive attention to customers (vs. finance) from two years before an M&A with the year before, the difference between the M&A and non-M&A groups was not significant (t = 1.54, p > .10). In contrast, when we compared a change from a year before the M&A with the year after, the M&A firms experienced a decline in executive attention to customers (vs. finance), whereas the non-M&A firms experienced a slight increase (ΔM&A Attention to Customers [vs. Finance][t + 1, t − 1] = −.03; ΔNon-M&A Attention to Customers [vs. Finance][t + 1, t − 1] = .01, p < .01). For the M&A firms with MROB, they experienced an increase in executive attention to customers (vs. finance) from the year before an M&A to the year after, whereas for those without MROB, they experienced a decrease (ΔM&A with MROB Attention to Customers [vs. Finance][t + 1, t − 1] = .01; ΔM&A without MROB Attention to Customers [vs. Finance][t + 1, t − 1] = −.04, p < .05). Consistent with this trend is the fact that for one of our sample firms, Macy's, it engaged in M&As in 2015 while having MROB. Macy's executive attention to customers (vs. financial issues) increased by 5.86% from the year before the M&As to the year after. Not surprisingly, Macy's did not experience any decline in customer satisfaction during this period.
We present the estimation results for executive attention to customers (vs. finance) (Model 1a) and customer satisfaction (CSAT; Model 1b) in Table 8. We find that M&A activity is associated with lower executive attention to customers (vs. finance) (β3 = −.032, p < .05, Model 1a) and executive attention to customers (vs. finance) is associated with higher customer satisfaction (π4 = 1.423, p < .01, Model 1b). Still, the effect of M&As on customer satisfaction persists with the inclusion of the mediator, executive attention to customers (vs. finance) (π3 = −.764, p < .05, Model 1b), which suggests partial mediation. Its mediating impact persists when we incorporate the MROB interaction terms (Model 2b). Based on Model 2b, the indirect effect of M&As on customer satisfaction through executive attention to customers (vs. finance) is negative and significant (β3[Post-M&A Year and M&A group] × π4[Executive attention to customers (vs. finance)] = −.046, confidence interval = [−.144, −.000]). Thus, in support of H2, customer dissatisfaction from M&As is due, in part, to a shift in executive attention away from customers and toward financial issues.
Graph
Table 8. Executive Attention to Customers (vs. Finance) and MROB.
| Dependent Variables | (1a) Executive Attention to Customers (vs. Finance)t | (1b) CSATt | (2a) Executive Attention to Customers (vs. Finance)t | (2b) CSATt |
|---|
| Focal Variables | | | | |
| Constant | .301*** | 67.024*** | .300*** | 67.218*** |
| (.033) | (.929) | (.032) | (.899) |
| M&A Group | .007 | −.310 | .012 | −.581** |
| (.007) | (.223) | (.009) | (.263) |
| Post-M&A Year | .004 | .452*** | .006 | .461** |
| (.004) | (.157) | (.005) | (.199) |
| M&A Group × Post-M&A Year | −.032** (H2) | −.764** | −.043*** | −.716** |
| (.014) | (.342) | (.016) | (.311) |
| Executive Attention to Customers (vs. Finance) | | 1.423*** | | 1.447*** |
| (.541) | | (.554) |
| Marketing Representation on the Board (MROB) | .756*** | 48.292*** | .838*** | 43.301*** |
| (.094) | (8.202) | (.118) | (8.300) |
| MROB × M&A Group × Post-M&A Year | | | .623*** (H3) | −2.617 |
| | (.241) | (12.285) |
| Covariates | | | | |
| Executive Attention to Customers (vs. Finance) Residuals | | 1.430*** | | 1.486*** |
| (.434) | | (.452) |
| MROB IMR | −.005** | −.396*** | −.006** | −.399*** |
| (.003) | (.048) | (.003) | (.048) |
| MROB × M&A Group | | | −.312 | 24.956*** |
| | (.257) | (7.398) |
| MROB × Post-M&A Year | | | −.130 | −.655 |
| | (.181) | (3.430) |
| Market Share(t − 1) | −.449*** | –3.134*** | −.449*** | −3.229*** |
| (.055) | (1.040) | (.056) | (1.016) |
| ROA(t − 1) | .322*** | 12.470*** | .328*** | 12.105*** |
| (.073) | (1.461) | (.074) | (1.517) |
| Firm Size(t − 1) | .000 | .000*** | .000 | .000*** |
| (.000) | (.000) | (.000) | (.000) |
| Advertising/Sales(t − 1) | .032 | −.124 | .032 | −.129 |
| (.020) | (.573) | (.020) | (.573) |
| R&D/Sales(t − 1) | .024 | −3.888*** | .022 | −3.837*** |
| (.042) | (1.131) | (.043) | (1.129) |
| Segments(t − 1) | .035*** | −.699** | .035*** | −.710** |
| (.013) | (.300) | (.013) | (.302) |
| Market Growtht | −.065** | −2.502*** | −.062** | –2.493*** |
| (.032) | (.731) | (.031) | (.736) |
| Firm fixed effects | Included | Included | Included | Included |
| Time effects | Included | Included | Included | Included |
| Model Information | | | | |
| χ2 | χ2(135) = 9,330.54*** | χ2(137) = 8,150.45*** | χ2(138) = 9,351.80*** | χ2(140) = 8,195.91*** |
| R2 | .791 | .768 | .791 | .769 |
| Number of firms | 122 | 122 | 122 | 122 |
| Total observations | 2,468 | 2,468 | 2,468 | 2,468 |
- 18 *p < .10.
- 19 **p < .05.
- 20 ***p < .01.
- 21 Notes: CSAT = customer satisfaction; ROA = return on assets; R&D = research and development. We report parameter estimates with bootstrapped standard errors in parentheses.
MROB is associated with more executive attention to customers (vs. finance) (β4 = .756, p < .01; Model 1a) and higher customer satisfaction (π5 = 48.292, p < .01; Model 1b). Further, in support of H3, MROB reduces the negative impact of M&As on executive attention to customers (vs. finance) (β5 = .623, p < .01, Model 2a) and executive attention to customers (vs. finance) is associated with an increase in customer satisfaction (π4 = 1.447, p < .01, Model 2b).[ 7]
We reestimated Equations 6.1 and 6.2 by adding an industry-level control variable to capture business-to-business versus business-to-customer membership, and our results do not change (Tables F.1 and F.2). We estimated firm value, firm efficiency, CSAT, and executive attention to customers (vs. finance) as a system of equations (Table F.3). Our effects of interest stay consistent when we use a four-equation model. We find robust support for our hypotheses.
While there has been significant research on customer satisfaction and a stream of research on M&As and financial performance, prior studies have not connected these two streams. We situate our research on this intersection and draw on the two complementary theories of the ABV of the firm and the upper echelons theory to examine the influence of M&A activity on a key, but often overlooked, stakeholder: customers.
Prior marketing strategy research has largely overlooked how disruptive corporate transformations can be for customers. Further, it has overlooked a key pathway between M&A activity and firm value: customer satisfaction. Some empirical work (e.g., [62]) has examined the interplay between M&A activity and customer satisfaction by treating customer satisfaction as a moderator and speculated (but not formally tested) that M&As enable a dual emphasis of firm efficiency and customer satisfaction. In contrast, we show that M&As not only do not enable a dual emphasis but also cause a decline in customer satisfaction to the extent that they outweigh any gain in firm value from firm efficiency. Thus, we add to previous work on firms' dual emphasis (e.g., [43]) but show that M&A activity works against a dual emphasis of firm efficiency and customer satisfaction.
We examine heterogeneity in the decline in customer satisfaction with novel conceptual additions to the M&A and customer satisfaction literature streams: executive attention to customers versus finance and MROB. We address ongoing calls to increase marketing's profile in the C-suite and higher (e.g., [24]; [46]; [70]) by examining how marketing leadership at the top of a firm redirects executive attention to customer issues, which explains differences in customer outcomes of M&As. In doing so, we add to the limited research on marketing presence in the upper echelons (e.g., [ 9]; [70]) by examining its role in channeling executive attention during M&As.
Existing research in marketing has overwhelmingly used the RBV of the firm to examine outcomes of M&As. This view, which emphasizes capabilities, fails to consider executive attention ([75]); however, executive attention is a precursor to resource investments. Further, although the ABV considers executive attention, it has primarily focused on the effect of supply-side (vs. demand-side) factors that influence managerial attention (e.g., [48]; [75]). In contrast, we extend the ABV to study marketing strategy phenomena in general, and a crucial demand-side stakeholder—customers—in particular. This aligns with newer research (e.g., Vadakkepatt et al. 2021) that aligns this theory with customer outcomes.
We contribute work on the marketing–finance interface. We introduce executive attention to customers versus financial issues as a mediator of the relationship between firm strategy (M&As) and a market-based asset (customer satisfaction). We find that during M&As, executives focus on financial issues at the cost of customer issues but that MROB can help mitigate this. Thus, we add a nuanced understanding the role of top leadership in navigating the marketing–finance interface.
We contribute to the literature on firm-level drivers of customer satisfaction (e.g., [50]; [56]) by examining a previously ignored antecedent: M&A activity. By showing that M&As negatively impact customer satisfaction, we shed light on how higher-level strategic actions that are often motivated by shareholder motives can risk the marketing function's most prized asset, its customer relationships. Finally, we add to growing research in marketing on the use of observational inference to document the causal effects of strategic decisions.
M&As have, on average, been shown to produce adverse financial effects. This has been attributed to overpayments as a result of optimism regarding synergies and cost efficiencies. However, we suggest that firms pay a price for dissatisfying customers during the M&A process, and in fact, this effect persists two years post-M&A. This finding is critical given that a recent survey of managers suggests that expanding a firm's customer base is a primary motivation for M&As. Thus, ignoring the dysfunctional effect of M&As on customers has serious long-term financial consequences and is inconsistent with firms' M&A objectives. To demonstrate the financial impact of the M&As due to a decline in customer satisfaction, we compared the firm value of M&A and non-M&A firms due to differences in customer satisfaction and firm efficiency with the estimation results from Table 3. Compared with that of non-M&A firms, the customer satisfaction of M&A firms was 1.14% lower a year after the M&A. In contrast, compared with that of non-M&A firms, the efficiency of M&A firms was.29% higher in the same period. When we incorporated these estimates in the firm value model (Equation 1.3), we found that the value of M&A firms was 2.43% lower than the non-M&A firms a year after the M&A. To calculate a change in firm value, we multiplied the percentage difference in value between M&A and non-M&A firms by the average firm value of the firms one year after an M&A. We find M&A firms' market value was worth $481 million less than that of the non-M&A firms. Although firms may be motivated to pursue an M&A to exploit scale-related synergies that provide cost-benefits, we show that efficiency gains fail to compensate for customer dissatisfaction-related financial losses. Thus, it is critical for managers responsible for M&As and industry consultants to include a consumer impact assessment in their M&A checklists.
Although there are several competing needs that require executives' attention during an M&A process, it is essential for them to allocate some of their attention to customer-related issues. The financial payoff of such attention is meaningful. To demonstrate the impact of executives of M&A firms paying attention to customers despite their tendency to focus on financial issues, we computed the percentage difference in customer satisfaction between M&A firms whose executives pay more attention to customers (vs. finance) (1 SD above the mean) and M&A firms whose executives pay less attention to customers (1 SD below the mean) with the results from Table 8 (Column 2b). Then, we used this percentage difference in customer satisfaction (.46%) to calculate a difference in firm value with the estimates from Table 3. We find that M&A firms that pay more attention to customers relative to financial issues experience 45% reduction in loss in firm value from the M&A (−1.34% vs. −2.43%). Thus, executive attention to customers can help firms significantly reduce M&As' damaging effects on customer satisfaction and firm value.
Moreover, MROB can attenuate a decline in customer satisfaction and, thus, increase firm value. In our data, 27.34% of the firms had MROB at some point during the 1995–2017 sample period. To illustrate, we calculated the firm value impact of adding one marketing title to the board with the estimation results from the moderated-mediation analysis that we report in Table 8. First, we computed the percentage difference in customer satisfaction between M&A firms with no MROB and M&A firms with just one person with a marketing title on the board in the post-M&A year with the results from Column 2b of Table 8, which is 2.85%. We then used this increase in customer satisfaction from MROB to calculate firm value with the estimates from Table 3 and find that the value of a firm with just one person with a marketing title on the board in the post-M&A year was 4.28% higher compared with firms that did not have any MROB. Adding these board positions is not trivial, especially during an M&A process. However, the financial consequence of not having MROB during M&As is severe. Thus, we make the case for marketing's voice in the C-suite, which is an important MSI Tier 1 Research Priority for 2020–2022.
Limiting dissatisfaction from M&As is a complex task, and multiple antecedents, including deal and integration-related factors and firm-level variables that speak to other functions of the firm, should be considered. Recent research has also found that customer satisfaction has a direct positive effect on firm efficiency ([ 7]), and future research could explore this pathway in the context of M&As. Further, our sample size was limited to what the ACSI database of customer satisfaction could provide. Future studies could identify alternative data sets ([39]) to enlarge their sample to extend the time frame of the panel and data frequency to examine changes in satisfaction several years after the M&As. In addition, we study customer satisfaction with the ACSI scores of acquirer firms and not target firms. This seems reasonable given that target firms are subsumed in acquiring firms, so any post-M&A ACSI score should reflect customers of both firms. Still, future research might benefit from assessing changes in satisfaction for the target firm. The challenge is that most Compustat and ACSI data are unavailable for the target firm after it has been acquired. Alternative data sources, which include primary data on customer satisfaction at the business unit level could be a solution. Finally, we empirically show that the ABV of the firm is a viable theoretical mechanism to explain the effect of M&As on customer satisfaction and how MROB moderates this relationship. Still, a change in executive attention is one of many potential pathways from M&A activity to customer satisfaction. Future research could consider how the RBV compares with the ABV in explaining these effects.
Footnotes 1 The authors contributed equally to the article.
2 Raj Venkatesan
3 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
4 The author(s) received no financial support for the research, authorship, and/or publication of this article.
5 Online supplement:https://doi.org/10.1177/00222429211024255
6 Kraft Heinz's customer satisfaction ACSI score declined by five points, which is a 6.3% decrease.
7 Although we did not hypothesize this, we tested the interaction effect between MROB and M&As on customer satisfaction, and the result is nonsignificant (π6 = −2.617, p > .10). Therefore, the interaction between MROB and M&A indirectly affects customer satisfaction through its effect on executive attention to customers (vs. finance), which is evidence of indirect moderated mediation ([74]).
References ACSI (2020), "Key ACSI Findings," (accessed August 25, 2021) , https://www.theacsi.org/about-acsi/key-acsi-findings.
Agrawal Anup , Jaffe Jeffrey F. , Mandelker Geherson N.. (1992), " The Post-Merger Performance of Acquiring Firms: A Re-Examination of an Anomaly ," Journal of Finance , 47 (4), 1605 – 21.
Almeida Heitor , Campello Murillo , Weisbach Michael S.. (2011), " Corporate Financial and Investment Policies when Future Financing Is Not Frictionless ," Journal of Corporate Finance , 17 (3), 675 – 93.
Anderson Eugene , Fornell Claes , Rust Roland T.. (1997), " Customer Satisfaction, Efficiency, and Profitability: Differences Between Goods and Services ," Marketing Science , 16 (2), 129 – 45.
Avkiran Necmi Kemal. (1999), " The Evidence on Efficiency Gains: The Role of Mergers and the Benefits to the Public ," Journal of Banking & Finance , 23 (7), 991 – 1013.
Bahadir Cem S. , Bharadwaj Sundar G. , Srivastava Rajendra K.. (2008), " Financial Value of Brands in Mergers and Acquisitions: Is Value in the Eye of the Beholder? " Journal of Marketing , 72 (6), 49 – 64.
Bhattacharya Abhi , Morgan Neil A. , Rego Lopo L.. (2020), " Customer Satisfaction and Firm Profits in Monopolies: A Study of Utilities ," Journal of Marketing Research , 58 (1), 202 – 22.
8 Bommaraju Raghu , Ahearne Michael , Hall Zachary R. , Tirunillai Seshadri , Lam Son K.. (2018), " The Impact of Mergers and Acquisitions on the Sales Force ," Journal of Marketing Research , 55 (2), 254 – 64.
9 Bommaraju Raghu , Ahearne Michael , Krause Ryan , Tirunillai Seshadri. (2019), " Does a Customer on the Board of Directors Affect Business-to-Business Firm Performance? " Journal of Marketing , 83 (1), 8 – 23.
Bower Joseph L. (1970), Managing the Resource Allocation Process. Boston : Harvard Business School Press.
Brockner Joel , Grover Steven , Reed Thomas , Dewitt Rocki , O'Malley Michael. (1987), " Survivors' Reactions to Layoffs: We Get by with a Little Help for Our Friends ," Administrative Science Quarterly , 32 (4), 526 – 42.
Capron Laurence , Dussauge Pierre , Mitchell Will. (1998), " Resource Redeployment Following Horizontal Acquisitions in Europe and North America, 1988–1992 ," Strategic Management Journal , 19 (3), 631 – 61.
Capron Laurence , Hulland John. (1999), " Redeployment of Brands, Sales Forces, and General Marketing Management Expertise Following Horizontal Acquisitions: A Resource-Based View ," Journal of Marketing , 63 (2), 41 – 54.
Coate Malcolm B. (2005), " Efficiencies in Merger Analysis: An Institutionalist View ," Supreme Court Economic Review , 13 , 189 – 240.
Cohan William D.. (2018), "When Blue Chip Companies Pile on Debt, It's Time to Worry," The New York Times (November 26) , https://www.nytimes.com/2018/11/26/opinion/corporate-debt-bubble-att-ge.html.
Cummins J. David , Xie Xiaoying. (2008), " Mergers and Acquisitions in the US Property-Liability Insurance Industry: Productivity and Efficiency Effects ," Journal of Banking & Finance , 32 (1), 30 – 55.
Dearborn Dewitt C. , Simon Herbert A.. (1958), " Selective Perception: A Note on the Departmental Identifications of Executives ," Sociometry , 21 (2), 140 – 44.
Deloitte (2019), "The State of the Deal: M&A Trends 2019," research report (accessed August 25, 2021) , https://www2.deloitte.com/content/dam/Deloitte/us/Documents/mergers-acqisitions/us-mergers-acquisitions-trends-2019-report.pdf.
Edeling Alexander , Srinivasan Shuba , Hanssens Dominque. (2020), " The Marketing-Finance Interface: A New Integrative Review of Metrics, Methods and Findings and an Agenda for Future Research ," International Journal of Research in Marketing (published online September 19) , https://doi.org/10.1016/j.ijresmar.2020.09.005.
Focarelli Dario , Panetta Fabio. (2003), " Are Mergers Beneficial to Consumers? Evidence from The Market for Bank Deposits ," American Economic Review , 93 (4), 1152 – 72.
Fornell Claes , Morgeson III Forrest V. , Hult G. Tomas M.. (2016), " Stock Returns on Customer Satisfaction Do Beat the Market: Gauging the Effect of a Marketing Intangible ," Journal of Marketing , 80 (5), 92 – 107.
Fu Ruchunyi , Tang Yi , Chen Guoli. (2020), " Chief Sustainability Officers and Corporate Social (Ir)responsibility ," Strategic Management Journal , 41 (4), 656 – 80.
Fuller Kathleen , Netter Jeffrey , Stegemoller Mike. (2002), " What Do Returns to Acquiring Firms Tell Us? Evidence from Firms That Make Many Acquisitions ," Journal of Finance , 57 (4), 1763 – 93.
Germann Frank , Ebbes Peter , Grewal Rajdeep. (2015), " The Chief Marketing Officer Matters! " Journal of Marketing , 79 (3), 1 – 22.
Gill Manpreet , Sridhar Shrihari , Grewal Rajdeep. (2017), " Return on Engagement Initiatives: A Study of a Business to Business Mobile App ," Journal of Marketing , 81 (4), 45 – 66.
Goldfarb Avi , Tucker Catherine E.. (2014), "Conducting Research with Quasi-Experiments: A Guide for Marketers," Working Paper No. 2420920, Rotman School of Management, University of Toronto.
Grullon Gustavo , Kanatas George , Kumar Piyush. (2006), " The Impact of Capital Structure on Advertising Competition: An Empirical Study ," Journal of Business , 79 (6), 3101 – 24.
Hambrick Donald C. , Mason Phyllis A.. (1984), " Upper Echelons: The Organization as a Reflection of Its Top Managers ," Academy of Management Review , 9 (2), 193 – 206.
Homburg Christian , Bucerius Matthias. (2005), " A Marketing Perspective on Mergers and Acquisitions: How Marketing Integration Affects Post Merger Performance ," Journal of Marketing , 69 (1), 95 – 113.
Huang Qianqian , Jiang Feng , Lie Erik , Yang Ke. (2014), " The Role of Investment Banker Directors in M&A ," Journal of Financial Economics , 112 (2), 269 – 86.
Kanuri Vamsi K. , Chen Yixing , Sridhar Shrihari (Hari). (2018), " Scheduling Content on Social Media: Theory, Evidence, and Application ," Journal of Marketing , 82 (6), 89 – 108.
Kim E. Han , Singal Vijay. (1993), " Mergers and Market Power: Evidence from the Airline Industry ," American Economic Review , 83 (3), 549 – 69.
Krishnan Ranjani A. , Joshi Satish , Krishnan Hema. (2004), " The Influence of Mergers on Firms' Product-Mix Strategies ," Strategic Management Journal , 25 (6), 587 – 611.
Lee Edmund , Koblin John. (2021), "AT&T, in Abrupt Turn, Will Shed Media Business in Deal with Discovery," The New York Times (May 17) , https://www.nytimes.com/2021/05/17/business/att-discovery-merger.html.
Lee Ju-Yeon , Sridhar Shrihari , Henderson Conor M. , Palmatier Robert W.. (2015), " Effect of Customer-Centric Structure on Long-Term Financial Performance ," Marketing Science , 34 (2), 250 – 68.
Loughran Tim , Vijh Anand M.. (1997), " Do Long-Term Shareholders Benefit from Corporate Acquisitions? " Journal of Finance , 52 (5), 1765 – 90.
Maksimovic Vojislav , Phillips Gordon. (2001), " The Market for Corporate Assets: Who Engages in Mergers and Asset Sales and Are There Efficiency Gains? " Journal of Finance , 56 (6), 2019 – 65.
Malshe Ashwin , Agarwal Manoj K.. (2015), " From Finance to Marketing: The Impact of Financial Leverage on Customer Satisfaction ," Journal of Marketing , 79 (5), 21 – 38.
Malshe Ashwin , Colicev Anatoli , Mittal Vikas. (2020), " How Main Street Drives Wall Street: Customer (Dis)satisfaction, Short Sellers, and Abnormal Returns ," Journal of Marketing Research , 57 (6), 1055 – 75.
Matsa David A. (2011), " Running on Empty? Financial Leverage and Product Quality in the Supermarket Industry ," American Economic Journal , 3 (1), 137 – 73.
McDonald Michael L. , Westphal James D. , Graebner Melissa E.. (2008), " What Do They Know? The Effects of Outside Director Acquisition Experience on Firm Acquisition Performance ," Strategic Management Journal , 29 (11), 1155 – 77.
Miles Laura , Rouse Ted. (2011), "Keeping Customers First in Merger Integration," (November 3) , https://www.bain.com/insights/keeping-customers-first-in-merger-integration/.
Mittal Vikas , Anderson Eugene W. , Sayrak Akin , Pandu Tadikamalla. (2005), " Dual Emphasis and The Long-Term Financial Impact of Customer Satisfaction ," Marketing Science , 24 (4), 544 – 55.
Moeller Sara B. , Schlingemann Frederik P. , Stulz René M.. (2004), " Firm Size and the Gains from Acquisitions ," Journal of Financial Economics , 73 (2), 201 – 28.
Motis Jrissy , Neven Damien , Seabright Paul. (2006), " Efficiencies in Merger Control ," in European Merger Control: Do We Need an Efficiency Defense? Ilzkovitz F. , Meiklejohn R.. Cheltenham, UK : Edward Elgar , 303 – 18.
MSI (2020), "MSI Research Priorities 2020-2022," (accessed August 25, 2021) , https://www.msi.org/wp-content/uploads/2020/09/MSI-2020-22-Research-Priorities-final.pdf.
Nath Pravin , Vijay Mahajan. (2008), " Chief Marketing Officers: A Study of Their Presence in Firms' Top Management Teams ," Journal of Marketing , 72 (1), 65 – 81.
Ocasio William. (1997), " Towards an Attention-Based View of the Firm ," Strategic Management Journal , 18 (S1), 187 – 206.
Oster Emily. (2019), " Unobservable Selection and Coefficient Stability: Theory and Evidence ," Journal of Business and Economic Statistics , 37 (2), 187 – 204.
Otto Ashley S. , Szymanski David M. , Varadarajan Rajan. (2020), " Customer Satisfaction and Firm Performance: Insights from Over a Quarter Century of Empirical Research ," Journal of the Academy of Marketing Science , 48 (3), 543 – 64.
Panagopoulos Nikolaos G. , Mullins Ryan , Avramidis Panagiotis. (2018) " Sales Force Downsizing and Firm-Idiosyncratic Risk: The Contingent Role of Investors' Screening and Firm's Signaling Processes ," Journal of Marketing , 82 (6), 71 – 88.
Potter John , Sutton Neil. (2019), "CX in M&A: What Consumers Think when Companies Combine," PwC (accessed August 25, 2021) , https://www.pwc.com/us/en/services/consulting/library/consumer-intelligence-series/customer-experience-in-mergers-and-acquisitions.html.
Prabhu Jaideep C. , Chandy Rajesh K. , Ellis Mark E.. (2005), " The Impact of Acquisitions on Innovation: Poison Pill, Placebo, or Tonic? " Journal of Marketing , 69 (1), 114 – 30.
Rao Vithala R. , Yu Yu , Umashankar Nita. (2016), " Anticipated vs. Actual Synergy in Merger Partner Selection and Post-Merger Innovation ," Marketing Science , 35 (6), 934 – 52.
Reed Robert. (2018), "Kraft Heinz Says It Doesn't Need to Make a Big Move, but Investors Know Better," Chicago Tribune (February 23) , https://www.chicagotribune.com/business/ct-biz-kraftheinz-big-buyout-inevitable-robert-reed-0225-story.html.
Rego Lopo L. , Morgan Neil A. , Fornell Claes. (2013), " Reexamining the Market Share–Customer Satisfaction Relationship ," Journal of Marketing , 77 (5), 1 – 20.
Renneboog Luc , Vansteenkiste Cara. (2019), " Failure and Success in Mergers and Acquisitions ," Journal of Corporate Finance , 58 , 650 – 99.
Rubera Gaia , Kirca Ahmet H.. (2017), " You Gotta Serve Somebody: The Effects of Firm Innovation on Customer Satisfaction and Firm Value ," Journal of the Academy of Marketing Science , 45 (5), 741 – 61.
Saboo Alok R. , Sharma Amalesh , Chakravarty Anindita , Kumar V.. (2017), " Influencing Acquisition Performance in High-Technology Industries: The Role of Innovation and Relational Overlap ," Journal of Marketing Research , 54 (2), 219 – 38.
Sikora Martin. (2005), " Consumers Are a Hot Issue for Merging Businesses ," Mergers and Acquisitions , 40 (7), 16 – 8.
Sorescu Alina , Chandy Rajesh K. , Prabhu Jaideep C.. (2007), " Why Some Acquisitions Do Better Than Others: Product Capital as a Driver of Long-Term Stock Returns ," Journal of Marketing Research , 44 (1), 57 – 72.
Swaminathan Vanitha , Groening Christopher , Mittal Vikas , Thomaz Felipe. (2014), " How Achieving the Dual Goal of Customer Satisfaction and Efficiency in Mergers Affects a Firm's Long-Term Financial Performance ," Journal of Service Research , 17 (2), 182 – 94.
Swaminathan Vanitha , Murshed Feisal , Hulland John. (2008), " Value Creation Following Merger and Acquisition Announcements: The Role of Strategic Emphasis Alignment ," Journal of Marketing Research , 45 (1), 33 – 47.
Thorbjørnsen Helge , Dahlén Micael. (2011), " Customer Reactions to Acquirer-Dominant Mergers and Acquisitions ," International Journal of Research in Marketing , 28 (4), 332 – 41.
Thornton Emily , Arndt Michael , Weber Joseph. (2004) "Why Consumers Hate Mergers," Business Week (December 5) , https://www.bloomberg.com/news/articles/2004-12-05/why-consumers-hate-mergers.
Tuli Kapil R. , Bharadwaj Sundar G.. (2009), " Customer Satisfaction and Stock Returns Risk ," Journal of Marketing , 73 (6), 184 – 97.
Vadakkepatt Gautham G. , Arora Sandeep , Martin Kelly D. , Paharia Neeru. (2021), " Shedding Light on the Dark Side of Firm Lobbying: A Customer Perspective ," Journal of Marketing.
Valentini Giovanni. (2012), " Measuring the Effect of M&A on Patenting Quantity and Quality ," Strategic Management Journal , 33 (3), 336 – 46.
Vennet Rudi Vander. (1996), " The Effect of Mergers and Acquisitions on the Efficiency and Profitability of EC Credit Institutions ," Journal of Banking & Finance , 20 (9), 1531 – 58.
Whitler Kimberly A. , Krause Ryan , Lehmann Donald R.. (2018), " When and How Board Members with Marketing Experience Facilitate Firm Growth ," Journal of Marketing , 82 (5), 86 – 105.
Wiles Michael A. , Morgan Neil A. , Rego Lopo L.. (2012), " The Effect of Brand Acquisition and Disposal on Stock Returns ," Journal of Marketing , 76 (1), 38 – 58.
Yadav Manjit S. , Prabhu Jaideep C. , Chandy Rajesh K.. (2007), " Managing the Future: CEO Attention and Innovation Outcomes ," Journal of Marketing , 71 (4), 84 – 101.
Yu Yu , Umashankar Nita , Rao Vithala R.. (2015), " Choosing the Right Target: Relative Preferences for Resource Similarity and Complementarity in Acquisition Choice ," Strategic Management Journal , 37 (8), 1808 – 25.
Zhao Xinshu , Lynch John G. Jr. , Chen Qimei. (2010), " Reconsidering Baron and Kenny: Myths and Truths About Mediation Analysis ," Journal of Consumer Research , 37 (2), 197 – 206.
Zhong Weiguo , Ma Zhiming , Tong Tony W. , Zhang Yuchen , Xie Luqun. (2020), " Customer Concentration, Executive Attention, and Firm Search Behavior ," Academy of Management Journal (published online May 27) , https://doi.org/10.5465/amj.2017.0468.
~~~~~~~~
By Nita Umashankar; S. Cem Bahadir and Sundar Bharadwaj
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 43- Do Backer Affiliations Help or Hurt Crowdfunding Success? By: Herd, Kelly B.; Mallapragada, Girish; Narayan, Vishal. Journal of Marketing. Sep2022, Vol. 86 Issue 5, p117-134. 18p. 1 Diagram, 6 Charts. DOI: 10.1177/00222429211031814.
- Database:
- Business Source Complete
Do Backer Affiliations Help or Hurt Crowdfunding Success?
Crowdfunding has emerged as a mechanism to raise funds for entrepreneurial ideas. On crowdfunding platforms, backers (i.e., individuals who fund ideas) jointly fund the same idea, leading to affiliations, or overlaps, within the community. The authors find that while an increase in the total number of backers may positively affect funding behavior, the resulting affiliations affect funding negatively. They reason that when affiliated others fund a new idea, individuals may feel less of a need to fund, a process known as "vicarious moral licensing." Drawing on data collected from 2,021 ideas on a prominent crowdfunding platform, the authors show that prior affiliation among backers negatively affects an idea's funding amount and eventual funding success. Creator engagement (i.e., idea description and updates) and backer engagement (i.e., Facebook shares) moderate this negative effect. The effect of affiliation is robust across several instrumental variables, model specifications, measures of affiliation, and multiple crowdfunding outcomes. Results from three experiments, a survey, and interviews with backers support the negative effect of affiliation and show that it can be explained by vicarious moral licensing. The authors develop actionable insights for creators to mitigate the negative effects of affiliation with the language used in idea descriptions and updates.
Keywords: backer affiliation; crowdfunding; prosocial; social structure; vicarious moral licensing
Crowdfunding has emerged as a dominant mechanism to harness the power of crowds in raising funds for innovative ideas. Interest in crowdfunding has surged in recent years. Facebook acquired Oculus 3D visualization device, a crowdfunded idea on Kickstarter, for US$3 billion ([14]). Peloton, the highly successful exercise bike, started as a Kickstarter project. The global crowdfunding market is expected to be well over US$40 billion by 2026 ([54]). Brands such as GE ([11]) and Unilever use crowdfunding to spur innovation ([53]), and academic research on the phenomenon and its role in the digital economy is emerging ([ 2]; [12]). Crowdfunding is a form of crowdsourcing in which participants, hereinafter referred to as "backers," are recruited to raise funds for ideas (e.g., [16]; [62]). As some backers fund the same ideas (i.e., "cobacking"), overlaps develop between these backers. These overlaps, called "affiliations," are key building blocks of the community's network structure and have been studied in other crowdsourcing communities (e.g., [48]). In this research, we explore how affiliation, defined as the number of cobacking relationships between potential backers and those who have previously funded the focal idea, might affect the idea's crowdfunding success. We illustrate affiliation in crowdfunding using a stylized example in Figure 1.
Graph: Figure 1. Illustration of affiliation in crowdfunding.
We know that crowd size affects outcomes positively as participants look to anonymous others for cues to decide which ideas to fund, a phenomenon referred to as "herding" (e.g., [66]). Previous research shows that attracting more backers positively impacts crowdfunding outcomes ([24]), an insight that many creators seem to grasp. However, crowd size does not represent the social structure (i.e., the pattern of connections in the community). In crowdfunding, as in other contexts in which shared communal goals exist (e.g., Wikipedia), social structure plays a more prominent role (e.g., [48]; [62]).
Our primary contribution is in showing that while the total number of backers (i.e., crowd size) may positively affect funding behavior and idea success (e.g., [66]), adding backers may not be unilaterally beneficial as the ensuing affiliation between backers negatively affects funding. Our analysis reveals that the negative effect of backer affiliation is above and beyond the positive effect of number of backers (i.e., the herding effect), highlighting the tension between the benefits of adding more backers and the adverse effects of backer affiliation. In other words, while adding a new backer (e.g., the focal backer in Figure 1) may positively affect the focal idea's success, adding this focal backer may not be equally beneficial across the three scenarios in Figure 1 as the degree of affiliation differs. We propose that the affiliation between the focal backer and other backers will influence the amount that the focal backer puts toward the focal idea and, thus, the idea's funding success.
Affiliation is a powerful force because it makes affiliated others' actions lead to changes in one's subsequent behavior (e.g., [35]; [57]). In some contexts, affiliation positively affects behavior as individuals desire to belong and therefore conform to affiliated others' behavior (e.g., [32]). However, in crowdfunding communities where individuals are often motivated by prosocial goals (e.g., [50]), we propose that such affiliation can negatively affect behavior. When individual actions benefit a social cause, seeing affiliated others participate may make individuals feel less of a need to do so, a process referred to as "vicarious moral licensing" (e.g., [13]; [22]; [37]). Thus, we propose that when backers decide whether to fund an idea, they are less likely to do so or more likely to fund a lower amount if affiliated others have already done so.
While affiliations develop in the community through cobacking, creators and backers also engage through nonmonetary actions, thereby driving social interaction. Therefore, to develop further substantive implications about the effect of affiliation, we examine the moderating role of both creator and backer engagement (e.g., [ 4]; [35]). For example, creators communicate with backers through the description of the idea on its homepage, perceived to be an important determinant of an idea's success ([38]; [64]), and by posting updates about progress. Backers engage with the community by sharing ideas on social media. We aim to understand how the effect of affiliation varies due to creator and backer engagement, as they help shed light on the underlying mechanism that drives the effect of affiliation.
We use multiple methods and data sets, including secondary data and experiments, to provide convergent validity to our findings. We also conduct interviews with 6 backers, survey 100 backers, and analyze 572 posts on backer forums to develop insights about the mechanism driving the effect of affiliation on funding outcomes. First, we assemble a comprehensive data set of daily funding for 2,021 new crowdfunded ideas listed on Kickstarter. We study two crowdfunding outcomes: ( 1) the monetary amount of funding received by an idea on any given day and ( 2) whether the idea raises sufficient funds during the funding window to meet or exceed its funding goal. We measure affiliation of an idea on the focal day as the number of cobacking relationships of backers who back on the focal day, with backers who funded until the day before the focal day (e.g., [35]; [41]). We estimate an instrumental variables regression model with fixed effects to assess the impact of affiliation among an idea's backers on the daily funding amount and report results from several robustness analyses. Second, we present results from controlled experiments, where we exogenously manipulate affiliation, and across three experiments, we replicate the negative effect of affiliation on funding, examine the underlying mechanism, and uncover the role of a key moderator. We find that the negative effect is stronger when creators use more communal words—both in the initial description of the idea and in subsequent updates—and when more backers share the idea on social media. Thus, creator and backer engagement may moderate prosocial motives to fund, further validating the proposed licensing mechanism.
We make several contributions. We are the first to show that affiliation among backers affects crowdfunding success in statistically and economically significant ways after controlling for herding and accounting for several alternative explanations. A 10% daily increase in number of backers would lead to an additional 20.2% in funding or an increase of US$83/day (i.e., the herding effect). In contrast, a 10% daily increase in backer affiliation would lead to an 8.7% decrease in funding or a decrease of US$36/day, offsetting the increase due to number of backers by about 43%. Thus, adding backers is good, but if the additional backers increase affiliation, the positive effect of adding these backers is smaller in the scenario where affiliation is high. We isolate vicarious moral licensing as a theoretical mechanism that drives the negative effect of affiliation through experiments. We explore the role of factors related to the idea, the creator, and the backers, all of which interact with affiliation.
Although crowdfunding has emerged as a dominant force for funding new ideas, research on crowdfunding is limited. Most early research focuses on microlending ([34]; [66]) or on crowdfunding platforms for music and journalism (e.g., [ 1]; [ 8]). Topics such as proximity to the deadline ([12]) and the text of content (e.g., [42]) have also garnered attention. Researchers have studied a variety of social factors that influence crowdfunding, in particular, the relationship between creators and individual backers, including the role of offline friendship ([34]), geographic proximity ([ 1]), and social interactions ([28]). We present a summary of representative research in Table 1.
Graph
Table 1. Comparison with Relevant Empirical Research.
| References (Published) | Dependent Variables | Explanatory Variables | Empirical Model Features | Data Context | Experiments | Findings |
|---|
| Our research | Funding | Affiliation Mediator: vicarious moral licensing Moderators: creator and backer engagement
| Fixed-effects log-linear regression Endogeneity (instrument) Robustness checks: probit, logit, and Tobit
| Crowdfunding(Kickstarter, experiments) | Three experiments | Prior backer affiliation decreases funding; the negative effect is due to vicarious moral licensing. This effect is stronger for ideas with communal descriptions, more communal updates, and more backer sharing on social media. |
| Wei, Hong, and Tellis (2021) | Success of funding | Similarity between ideas
| Network similarity Binary regression
| Crowdfunding(Kickstarter) | No | Prior success of similar ideas affects success. Funding performance increases as an idea's novel and imitative characteristics are balanced. The optimal funding level is closer to the level of similar ideas. |
| Netzer, Lemaire, and Herzenstein (2019) | Loan payback | Loan description
| Text analytics Binary regression
| Crowdlending(Prosper) | No | Borrowers who use certain types of words are more likely to default. |
| Dai and Zhang (2019) | Funding time elapsed | Going past deadline Prosocial motivation Creator is individual
| Regression continuity
| Crowdfunding(Kickstarter) | No | Backers might be driven by prosocial motives around deadline following goal pursuit. |
| Kim et al. (2020) | Goal completion | Forward looking Social interactions
| Bayesian IJC method Two-step Counterfactuals
| Crowdfunding (music)(Sellaband) | No | Forward-looking investment behavior as well as contemporaneous and forward-looking social interactions impact share purchases and goal completion. |
| Burtch et al. (2013) | Contributions | Concealment Social norms
| Tobit Endogeneity (IV)
| Crowdfunding(Data set undisclosed) | No | Concealment hurts the likelihood of contribution and contribution. Social norms drive concealment. |
| Agrawal, Catalini, and Goldfarb (2015) | Decision to invest | Geography
| Linear regression Fixed effects
| Crowdfunding (music)(Sellaband) | No | Local backers are not influenced by artist. The effect does not persist past the first investment, indicating the role of search but not monitoring. |
| Burtch et al. (2013) | Contribution frequency | Crowding Funding window Degree of exposure
| Log-linear regression Endogeneity (GMM)
| Crowdfunding (journalism)(Data set undisclosed) | No | Partial crowding-out effect. Backers experience lower marginal utility of giving as the funds become less relevant to the recipient. The funding window and degree of exposure have a positive effect, after publication of the story. |
| Lin, Prabhala, and Viswanathan (2013) | Interest rate, default rate | Friendships
| Probit regression
| Crowdlending(Prosper) | No | Online friendships act as signals of credit quality, increase the probability of funding, lower interest rates, and result in lower ex post default rates—gradation in effects based on roles and identities of friends. |
| Zhang and Liu (2012) | Loan amounts | Crowding
| Hazard model Fixed effects First differences
| Crowdlending(Prosper) | No | Well-funded borrowers attract more funding. Lenders learn from peer decisions and do not mimic. |
1 Notes: IV = instrumental variable, GMM = generalized method of moments; IJC = [26].
In addition to the relationship between creators and backers, there are several ways in which others' actions might inform backers' funding decisions. For example, [66] report that potential lenders assess borrowers' creditworthiness by observing other lenders. They attribute the positive effect of the number of other lenders to herding, wherein crowd size becomes a beacon for others to decide which ideas to fund. This finding might suggest that the mere addition of more supporters unilaterally benefits crowdfunding outcomes as potential backers simply follow other backers. What are some factors that might limit the positive impact of the crowd's behavior on crowdfunding? To answer this question, we note that most research has considered the presence of the anonymous crowd as the cause for a social effect that is generally positive. However, crowd size does not account for an important aspect of networks (i.e., the structure of connections among the community's participants).
Thus, what is missing in extant research is an explicit acknowledgment of social structure beyond crowd size and an exploration of how it impacts crowdfunding outcomes. Social structure arises due to coparticipation in events, in our case, cobacking across ideas, a phenomenon referred to as affiliation (e.g., [17]; [60]). Affiliation, identified as an important phenomenon in the new digital economy dominated by crowdsharing ([15]), is the central focus of our research.
Communities evolve through repeated interactions between members, which give rise to affiliations or overlaps. As affiliations grow, the interconnectivity among backers leads to scaffolding structures that hold the community together through both first- and second-order ties. Affiliations have been studied in interfirm relationships ([58]), board interlocks ([52]), product development (e.g., [35]), and wiki contributions ([48]). Regardless of the context, research suggests that ( 1) individuals notice affiliated others' behavior, ( 2) individuals feel a sense of connectedness and shared identity with affiliated others, and as such, ( 3) affiliated others' actions lead to changes in one's subsequent behavior (e.g., [35]; [57]).
To establish that participants notice affiliated others' behavior when visiting crowdfunding platforms, we ran a pilot study with actual backers prescreened on the basis of their prior crowdfunding behavior. Participants were shown a screenshot of a crowdfunding page created by a web designer. To assess which information captured participants' attention, we used a standard heat-mapping approach for measuring visual attention ([ 5]). Invisible boxes around various pieces of information (e.g., idea title, backer information, idea description) coded visual attention as participants read and clicked on information, as per instructions. We found that many participants read and clicked on backer information, more so than other potentially relevant information such as the number of shares and creator information. Further, of the available backer information, affiliation ranked as highly important (for details, see Web Appendix A). Discussions on crowdfunding message boards and websites, as well as results from a survey that we conducted (discussed in the following sections), further support this idea, suggesting that among all available information, backers do consider affiliated others' behavior as they make funding decisions. Next, to confirm that affiliation affects perceptions of connectedness and shared identity in crowdfunding communities, we ran a pilot study with 150 Amazon Mechanical Turk (MTurk) participants. We find that affiliation significantly increased perceptions of connectedness and shared identity with other backers (see Web Appendix B).
If potential backers notice affiliated others' behaviors and feel a sense of connectedness with these affiliated others, how might affiliated others' behavior influence their own funding decisions? To answer this important question, we next examine crowdfunding platforms and how they differ from noncommunal (i.e., transactional) contexts.
Crowdfunding is a communal endeavor in which individuals collaborate to achieve shared goals, and platforms grow due to members' participation and interactions ([50]). Individuals behave differently in communal contexts than in noncommunal (i.e., transactional) contexts (e.g., [10]; [18]). For example, in communal contexts, individuals are more likely to request help from others, keep track of others' needs, respond to them, and report more positive emotions while doing so ([18]). In crowdfunding, these prosocial goals are reflected in the desire to help others, achieve funding goals, and be part of a community ([19]).
In crowdfunding communities where individuals are often motivated by prosocial goals (e.g., [50]), we propose that seeing affiliated others fund may make individuals feel less of a need to do so, a process referred to as vicarious moral licensing (e.g., [13]; [22]; [37]). Vicarious moral licensing occurs when individuals see affiliated others' actions as satisfying their own goals, which changes their perceived moral imperative and subsequent behavior. For example, learning that affiliated others demonstrate environmentally friendly behavior makes individuals less likely to do so ([37]). It is important to recognize that this effect is not merely akin to strangers' behavior in a crowd (i.e., the bystander effect; e.g., [31]), but that it is those with whom an individual perceives a social connection (i.e., affiliation) that drives the focal effect.
To confirm the importance of affiliated others' behavior and further validate the proposed mechanism, we conducted in-depth interviews of 6 backers, surveyed 100 backers, and coded 572 posts from KickstarterForum.org, the dominant crowdfunding discussion forum (see Web Appendix C). The findings confirmed the prominence of prosocial (i.e., communal) motives on crowdfunding decisions, the importance of affiliated others' behavior on backers' own funding behavior, and the role of vicarious moral licensing. For example, as one interview participant explained, "I look at other funders only to further discover related projects. It's an interesting way to discover—because some people are more involved than you are.... It's interesting to follow that rabbit trail and see, 'Oh, this person supported this, and look at what else they fund.'" Another stated, "You are dealing with finite resources in terms of what you are willing to spend. If you support one thing, I don't know, for me, if I see someone supporting something else, I think, well yeah, they supported that. I'm sure I could find a bunch of other people that support a bunch of other things. I just gave X amount of dollars, whatever amount I have, and I'm not going to be giving any more than that right now."
Although we propose vicarious moral licensing as the mechanism underlying the focal effect of affiliation and initial evidence indicates this to be the case, we acknowledge the complexity of social interactions in crowdfunding. Because these social interactions are likely to be subject to several factors, we consider uniqueness as an alternative explanation for the negative effect of affiliation on funding. Backers may try to identify ideas that have received less funding from affiliated others. By doing so, backers can distinguish themselves from these affiliated others, fulfilling a need for uniqueness (e.g., [59]). In our analysis, we report results from an experiment where we test vicarious moral licensing and uniqueness as potential explanations for the negative effect of affiliation on funding.
Crowdfunding platforms are characterized by contributions from both creators and backers (e.g., [ 4]; [48]) as these interactions create and sustain the community's viability. Therefore, we explore the role of creator and backer engagement in moderating the impact of affiliation on crowdfunding.
Previous research has found that while prosocial goals may be common in crowdfunding platforms (e.g., [50]), an idea's description can further induce prosocial motivation and behavior when it emphasizes communal language ([23]). We propose that the vicarious moral licensing effect (i.e., the negative effect of affiliation on funding) is driven by the communal context and the prosocial behavior it prompts and that this behavior is further heightened by creators describing their ideas with communal words like "together" and asking backers to "partner" with them by providing financial "support" ([47]). As such, ideas described with more (vs. less) communal words will exhibit a stronger negative effect of affiliation on funding outcomes.
Creators can also engage with the backer community by posting updates to highlight their strategic goals and the idea's progress. Updates provide diagnostic information concerning an idea's success (e.g., [ 4]; [35]). Updates might draw backers' attention to the idea's characteristics and evolution, and lessen attention toward cobackers and affiliation. Consistent with the vicarious moral licensing mechanism, we expect updates that use more (vs. less) communal words to strengthen affiliation's negative effect. As such, we estimate the moderating effects of communal words in the creator's updates.
We also explore how backers' engagement might moderate the affiliation effect by exploring the role of social media sharing of the focal idea by backers. While sharing behavior on social media could have several motivations, altruism is perceived as a primary motivator, and others seeing the shares likely view them as such ([29]; [33]). We expect that such sharing heightens funders' prosocial motives and vicarious moral licensing, further strengthening the negative effect of affiliation. Next, we describe our data and methodology.
We employed a multimethod approach to investigate the phenomenon. We collected and analyzed two types of data: observational data from a crowdfunding platform and experimental data from lab settings. We begin by describing the observational data, the empirical model and identification strategy, the results, and robustness checks. Then, we describe three experiments in which we identify the primary effect in a controlled setting and shed light on the mechanism underlying the primary effect and its moderator. The first experiment demonstrates the negative effect of affiliation on funding behavior. The second experiment validates vicarious moral licensing as an underlying mechanism and rules out uniqueness as one potential alternative explanation. The third experiment examines how the idea's description moderates the effect of affiliation.
We collected data on Kickstarter, the world's largest and most prominent crowdfunding platform. We utilized a web crawler to visit the new ideas page listed on Kickstarter beginning December 18, 2013. From that day and every subsequent day of data collection, the crawler visited the pages of the ideas that were started on the first day of the crawl, in addition to all the ideas that were started on the subsequent days. We stopped the crawler after 37 days, giving us data on 2,021 new ideas. We acknowledge that our research's funding constraints affected the number of days, but we went one week past the most common deadline of 30 days. We note that while some ideas in our sample received funding after data collection stopped, our results are robust to this truncation.[ 7]
For the data collection, the crawler began with ideas that started receiving funds on the day of the crawl, and it identified every backer who funded the focal idea, the funded amount, and the calendar date. The crawler then visited every backer's history and collected information on all the other ideas that the backer had funded in the past. At the time of data collection, Kickstarter made all backers visible to all prospective backers. The list of backers on Kickstarter was available by clicking the "community" link that prominently appears on the focal idea's web page.[ 8] This process allowed us to construct the network, giving us the structure of relationships to calculate affiliation. The crawler also collected other relevant information from the page, including the idea's description, number and text of updates, and the number of Facebook shares of the idea to measure backer engagement.
Our unit of analysis for the daily amount funded is an idea-day, and our final sample had 32,438 observations at the idea-day level. This specification makes the most sense because, for a data set with idea-day-backer as the unit of analysis, the funded amount (for an idea on a day) takes zero values for over 99.9% of observations, making such a specification noninformative. Next, we describe the key measures.
Consistent with prior literature ([ 1]; [ 8]), our funding success measure is the amount of funding received by an idea on any given day. Across all crowdfunding platforms, this measure is always easily and prominently visible on the idea's webpage. Subsequently, we show that our results are robust to other measures of success.
Consistent with prior literature (e.g., [35]; [41]), we posit that two backers are affiliated if they have funded at least one common idea on the platform and are not affiliated if all the ideas that they have funded are mutually exclusive. Thus, the backer affiliation for a focal idea on a focal day is the number of cobacking relationships between those backers who fund the focal idea on the focal day and all backers who have funded the focal idea at any time before the focal day.
Consider a backer of a focal idea who funds the focal idea on the focal day. Consider another backer of the focal idea who funds the focal idea any time before the focal day. A cobacking relationship exists between these two backers if they have both funded one idea (other than the focal idea) any time before the focal day. One cobacking relationship represents one unit of affiliation. Affiliation increases both with the number of backers who coback and with the number of cobacked ideas.
To elaborate, consider the following examples. In each example, idea i is launched on day t = 1, say December 13. Further, Jack funds idea i on December 17 (t = 5), and the goal is to calculate affiliation as of December 17 (t = 5).
Example 1: Tom funds idea i on December 13. Also, before December 17, Jack and Tom both fund another idea j. As there is one cobacking relationship (that between Jack and Tom for cobacking idea j), Affiliationi, t = 5 = 1.
Example 2: Tom funds idea i on December 13. Jack and Jill both fund idea i on December 17. Before December 17, Jack and Tom both fund another idea j. Furthermore, before December 17, Jill and Tom both fund another idea k. As there are two cobacking relationships (those between Jack and Tom for cobacking idea j, and between Jill and Tom for cobacking idea k), Affiliationi, t = 5 = 2.
Example 3: Tom funds idea i on December 13. Jane funds idea i on December 14. Before December 17, Tom, Jack, and Jane funded another idea j. As there are two cobacking relationships (those between Jack and Tom for cobacking idea j, and between Jack and Jane for cobacking idea j), Affiliationi, t = 5 = 2.
We present summary statistics for Kickstarter in Table 2. The median number of backers who fund an idea in a day is 1, and most ideas have only a few backers. When a backer funds an idea, the median number of past backers of that idea is 10 backers (i.e., the median of the variable cumulative number of backers funding idea i before day t is 10). In the six months preceding data collection, 82% of backers in our data had not funded any idea on Kickstarter. Thus, the odds of having to remember multiple cobacking relationships are relatively low. Most importantly, the median value of affiliation is zero, and the mean is 3.3. In other words, a large majority of backers in our data must process a very small amount of information to infer affiliation. Our measure of affiliation reflects a more nuanced and disaggregated conceptualization of affiliations than the number of "cobacked ideas" or the number of "common backers." Other measures are likely sparser than our measure. Subsequently, we show that our results are robust to alternate measures of affiliation.
Graph
Table 2. Summary Statistics for Kickstarter.
| Variable | Mean | SD | Min | Median | Max |
|---|
| Amount of funding of idea i (in $) on day t | 409.34 | 4,764.72 | 0 | 0 | 593,731 |
| Backer affiliation of idea i by day t − 1 (Affilit − 1) | 3.33 | 10.66 | 0 | 0 | 477 |
| Cumulative number of backers funding idea i by day t − 1 (CumBackersit − 1) | 53.85 | 357.87 | 0 | 10 | 17,018 |
| Number of backers funding idea i on day t − 1 (Backersit − 1) | 6.71 | 142.99 | 0 | 1 | 17,010 |
| Cumulative number of updates by creator of idea i by day t − 1 (CumUpdatesit − 1) | 8.82 | 23.64 | 0 | 0 | 524 |
| Proportion of funding goal of idea i achieved by day t − 1 (PropGoalit − 1) | .49 | 2.14 | 0 | .13 | 132.63 |
| Proportion of funding duration of idea i completed by day t − 1 (PropDurationit − 1) | .31 | .24 | 0 | .27 | .97 |
| Closeness centrality of idea i as of day t − 1 | 5.67 × 10 − 9 | 5.01 × 10 − 8 | 3.49 × 10 − 11 | 2.50 × 10 − 10 | 2.2 × 10 − 6 |
| Betweenness centrality of idea i as of day t − 1 | 1,258.07 | 8,698.49 | 0 | 0 | 335,213 |
| Eigenvector centrality of idea i as of day t − 1 | .002 | .03 | 0 | 2.50 × 10 − 9 | 1 |
| Last week (1 if day t is in the last week of funding of idea i, 0 otherwise) | .08 | .28 | 0 | 0 | 1 |
| Cumulative number of communal words in updates by creator of idea i by day t − 1 (CommunalUpdatesit − 1) | 20.17 | 1,696.27 | 0 | 0 | 230,232 |
We measure the creator's engagement using the number of creator's updates on the idea page and separately measure the level of communal content in each update. To code communal words, we created a dictionary to capture words that reflect the use of communal language. For this, we asked two graduate research assistants to read descriptions of a random sample of 100 ideas (from our data) and identify words that reflected a "communal" idea while coding each description on whether it was communal. We provided the Merriam-Webster definition of "communal" ("of or relating to a community") to the two coders along with synonyms from a thesaurus. Then, we cross-verified these words with LIWC's category for "affiliation," comprising 248 words ([46]). Communal words that appear at least once in our corpus are member, team, group, groups, family, friends, affiliation, affiliate, relation, connection, alliance, relationship, partner, partners, partnership, link, merge, cooperate, cooperation, together, join, thanks, thank you, appreciate, our, and we. We measure backer engagement as the number of Facebook shares of the idea by backers, which we collected when the web crawler visited an idea's webpage.
To explore model-free evidence, we present summary statistics about three regimes of the distribution of the amount of daily funding achieved for Kickstarter in Table 3: ( 1) idea-day-specific observations when there is no funding, ( 2) when the daily funding is positive but does not exceed the mean level in the data ($409.34), and ( 3) when the daily funding exceeds the mean level. Backer affiliation is highest when ideas do not receive any funding and lowest when ideas achieve the highest funding. The measure of backer affiliation for an idea on a given day is based on cobacking relationships of backers that fund the idea on that specific day with backers who funded before that day. If no backer funds on a specific day, the affiliation measure for that day is zero. The measure is not cumulative, and it does not increase over time. Thus, there is model-free evidence for the negative effect of affiliation on funding outcomes. We collected similar data from another crowdfunding platform, Indiegogo, which we use in the robustness analysis. Additional details about the Kickstarter data and summary statistics for the Indiegogo data appear in Web Appendix D. We illustrate affiliation in Figures W1–W5 and the sample's network structure and growth in Figures W6–W8 in Web Appendix E. We estimate the primary empirical model on Kickstarter data.
Graph
Table 3. Means of Backer Affiliation and Other Time-Varying Covariates at Different Levels of Daily Funding (Kickstarter).
| Variable | Amount Funded (yit) = 0 | Amount Funded 0 < (yit) ≤ $409.34 | Amount Funded (yit) > $409.34 |
|---|
| Number of observations | 17,505 | 11,071 | 3,863 |
| Proportion of all observations | 53.96% | 34.13% | 11.91% |
| Amount of funding of idea i (in $) in day t | 0 | 109.29 | 3,214.89 |
| Backer affiliation of idea i by day t − 1 (Affilit − 1) | 4.22 | 2.26 | 1.79 |
| Cumulative number of backers funding idea i by day t − 1(CumBackersit − 1) | 22.65 | 44.23 | 222.89 |
| Cumulative number of updates by creator of idea i by day t − 1 (CumUpdatesit − 1) | 6.21 | 10.54 | 15.76 |
| Proportion of funding goal of idea i achieved by day t − 1 (PropGoalit − 1) | .26 | .53 | 1.69 |
| Proportion of funding duration of idea i completed by day t − 1 (PropDurationit − 1) | .34 | .29 | .27 |
| Closeness centrality of idea i as of day t − 1 | 5.68 × 10−9 | 4.56 × 10−9 | 9.49 × 10−9 |
| Betweenness centrality of idea i as of day t − 1 | 473.99 | 1,178.21 | 5,974.58 |
| Eigenvector centrality of idea i as of day t − 1 | .001 | .000 | .012 |
| Last week (1 if day t is in the last week of funding of idea i, 0 otherwise) | .10 | .07 | .06 |
| Cumulative number of communal words in updates by creator of idea i by day t − 1 (CommunalUpdatesit − 1) | .69 | 1.93 | 199.38 |
Following [ 8] and [66], our primary dependent variable (yit) is the monetary funding received by an idea i (i = 1,...N) on day t (t = 1,...Ti). As a starting point, we incorporate backer affiliation and several controls in a fixed-effects regression model as follows:
Graph
( 1)
To account for nonnegativity, we log-transform all variables that are not proportions. For variables that can take zero values, we take the logarithm of the variable added to.001. Replacing this constant with other constants does not affect our results. Estimating the model without taking logarithms of any variable gave us consistent results.
To control idea-specific confounding factors such as inherent differences in idea quality, the novelty of idea description, creator expertise, and so on, we employ idea-specific fixed effects αi, a vector of 2,021 elements for the Kickstarter data set. We incorporate fixed effects for each day in the idea's funding window to control temporal patterns in funding and changes in the Kickstarter environment over time. These are denoted by the vector αt. Error terms are assumed normally distributed and clustered at the idea level.
Our key independent variable is Affilit − 1. This is the number of cobacking relationships between those backers who fund idea i on day t − 1 and all backers who have funded this idea before day t − 1. Subsequently, we report robustness checks to alternate measures of backer affiliation. Although our fixed-effects specification controls for confounds at the idea level and the day level, we need to control idea-specific factors that are time varying. Chief among these is the amount of funding received by the focal idea on day t − 1 ([ 8]), enabling us to control those time-varying idea-specific unobservables, which may be serially correlated (e.g., word of mouth about the idea) and to attenuate serial correlation among the residuals. This also accounts for the alternate explanation that affiliation on day t − 1 affects funding on day t − 1, but not on day t. By incorporating the lagged measure of funding, we can account for all factors that affect funding until the day t − 1.
We next discuss other time-varying idea-specific controls. First, the number of affiliations among backers is correlated with the number of backers. There can be no backer affiliations without backers; more backers could result in more possibilities for affiliation. To control for the possibility that the number of backers drives the effect of affiliation on funding, we include CumBackersit − 1, the cumulative number of backers funding idea i by day t − 1, as a control variable. Also, to the extent that ideas with more backers attract more funding ([66]), this serves as a measure for herding behavior. Second, creators communicate with backers via updates, a means to elevate idea visibility and signal effort ([12]). To understand how creator actions might drive funding, we include CumUpdatesit − 1, the cumulative number of updates by the creator of the idea i by day t − 1. In addition, CommunalUpdatesit − 1 is the number of communal words contained in the updates.
Third, the funding window of an idea influences its funding outcomes. Ideas receive more funding in the later stages of the funding window as the funding deadline nears (e.g., [12]; [31]). To account for this, we include the duration of the funding window completed for the idea (PropDurationit − 1) as a proportion of the total funding window (typically 30 days). Furthermore, ideas receive greater funding as they get closer to meeting their funding goals ([12]). Although daily fixed effects account for temporal variations in funding, they might not capture the effect of proximity to the funding goal. Therefore, we include PropGoalit − 1, the proportion of the funding goal of the idea that has been achieved until day t − 1, and LastWeekit − 1, a dummy variable for whether the observation belongs to the last week of the funding window.
Finally, structural measures of network centrality might affect the outcome. Because these measures capture the extent of social capital that accrues to ideas due to being associated with certain backers, we want to control for the effects of these measures. We compute and include three of the most widely used network measures in marketing (e.g., [35]; [48]; [58]), (Networkit − 1): closeness centrality, betweenness centrality, and eigenvector centrality of idea i on day t. Closeness centrality in our context is how close the focal idea is from all the backers (connected and not connected) in the network, betweenness centrality is the extent to which the focal idea lies on the common paths between all pairs of backers in the network, and eigenvector centrality is the extent to which the focal idea's backers are prolific in backing other ideas. We computed both bipartite and single-mode network variants of each of these measures.[ 9] Given the high correlation across the bipartite and single-mode versions of each measure, we included in the model the version of each measure that leads to a more significant improvement in R2. As shown in the correlation matrix of all variables (Table 4), these variables are not highly correlated with our measure of affiliation, suggesting that affiliation captures the network's unique structural properties based on counts of overlaps. To assess interaction effects, we interact affiliation with CommunalUpdatesit − 1[10] and with the number of Facebook shares of the idea by backers (FBSharesi).
Graph
Table 4. Pairwise Correlation Coefficients of All Variables (Kickstarter).
| Variable | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 |
|---|
| 1. Amount of funding of (in $) in t | 1 | | | | | | | | | | |
| 2. Backer affiliation i by t − 1 | −.07 | 1 | | | | | | | | | |
| 3. Cum. number of backers funding by t − 1 | .13 | .14 | 1 | | | | | | | | |
| 4. Cum. number of updates by creator i by t − 1 | .13 | .15 | .18 | 1 | | | | | | | |
| 5. Amount of funding (in $) in t − 1 | .49 | −.11 | .17 | .15 | 1 | | | | | | |
| 6. Proportion of funding goal achieved by t − 1 | .15 | .00 | .12 | .20 | .19 | 1 | | | | | |
| 7. Proportion of funding duration completed by t − 1 | −.11 | .37 | .05 | .30 | −.19 | .05 | 1 | | | | |
| 8. Closeness centrality as of t − 1 | .01 | −.07 | −.00 | −.04 | .14 | −.01 | −.12 | 1 | | | |
| 9. Betweenness centrality i as of t − 1 | .09 | .37 | .23 | .29 | .07 | .07 | .32 | −.07 | 1 | | |
| 10. Eigenvector centrality as of t − 1 | .05 | .04 | .33 | .09 | .07 | .04 | .00 | .07 | .09 | 1 | |
| 11. Last week (1 if day t is in the last week of funding) | −.05 | .14 | .02 | .15 | −.05 | .05 | .61 | .03 | .01 | .00 | 1 |
| 12. Cum. no. of communal words in updates by t − 1 | .02 | .02 | .26 | .01 | .03 | .00 | −.00 | .01 | .00 | .09 | −.00 |
2 Notes: We take logarithms of all variables, which are not proportions. For variables that can take zero values, we take the logarithm of the variable added to.001. All variables pertain to the focal idea. Coefficients with p < .05 are in boldface.
We use lags of all covariates because information about the focal day is not updated in real time and is unavailable until the following day.[11] We present the correlation matrix of all variables in Table 4; most correlations are less than.3, allaying multicollinearity's ill effects.
We first discuss how our work is different from the peer effects literature and then explain our identification strategy. The prototypical problem in the marketing literature on the identification of peer effects (e.g., [40]) is to estimate the likelihood of agent A adopting a product (e.g., buying an online game) under the knowledge that agent B (a self-identified "friend" or influencer) has already adopted that same product. The herding literature has conclusively documented positive peer effects across various consumer contexts (e.g., [57]; [66]). If there are positive effects due to the size of the crowd or the number of peers, that would be equivalent to herding, not our study's primary focus. In other words, our main objective is not to estimate how backer A will fund the focal idea if another backer B has previously funded it. We account for herding in our model by incorporating the prior number of backers of the focal idea as a control. Instead, our objective is to study the effect of affiliation, which is formed when two backers back an idea that is not the focal idea. Affiliation arises in collaborative contexts (e.g., board interlocks, product development teams) rather than common product purchases. We note that this is a key difference of our article from other contexts. In addition, our interest is less in modeling agent behavior (e.g., an individual's rating of a product in [57]) than in modeling product success (i.e., funding success of an idea). We next address three main issues that could confound identifying the causal effect of affiliation on the focal idea's funding.
Idea-specific characteristics that are not observable to the researcher could be correlated with our affiliation measure and affect the focal idea's funding. Perhaps highly affiliated backers are attracted to ideas with high (or low) unobserved quality. The inability to control for quality dimensions might induce an upward (or downward) bias in our estimate of the effect of affiliation. Following [40] and [57], we incorporate idea-specific fixed effects. These effectively control for all idea-specific factors that might be correlated with affiliation. Next, there could be time-varying factors across the funding window that might be correlated with affiliation and funding. For example, affiliation and funding are both likely to be low in the first few days of funding. We control for all day-specific trends by incorporating day fixed effects. Finally, the presence of idea-specific time-varying factors cannot be ruled out. We control for funding received by the focal idea on day t − 1. As mentioned previously, this approach enables us to control those time-varying idea-specific unobservables, which may be serially correlated, and to attenuate serial correlation among the residuals.
In the context of peer influence, simultaneity implies that not only can the influencer influence the focal individual but the focal individual could also affect the influencer's actions, leading to an upward bias in the estimate of peer effects. In our context, affiliations formed on a focal day may affect the focal idea's funding. Simultaneously, the focal idea's funding on a focal day also affects affiliation formation on that day. Following recent literature (e.g., [45]), we use the lagged measures of affiliation in the model. While affiliation before the focal day can affect funding on the focal day, the reverse is not possible.
Backers with similar preferences may be more likely to behave similarly. In such a scenario, the effect of prior affiliation on subsequent funding of the focal idea might manifest these common preferences. The literature on consumer peer effects has used consumer-specific fixed effects to deal with this. However, crowdfunding is different from consumer contexts in that while consumers buy (and evaluate) several products, backers typically fund very few ideas on a platform.
Moreover, unlike crowdfunding, consumer contexts generally focus on the individual more than collective action ([50]). So, backer-specific fixed effects are econometrically infeasible to estimate for both the researcher and the platform. Instead, we first include the cumulative number of backers and several other network measures as controls. Next, we note that controlling for lagged funding of the idea also controls backer characteristics that have affected funding before the focal day.
Finally, we include an instrument for affiliation. If our measure of affiliation is correlated with the error term in Equation 1, its coefficient could be biased. In our primary analysis, we use an observed instrument to estimate a two-stage least-squares instrumental variable regression model. As [49], p. 4) mentions, the ideal solution for endogeneity is to conduct an experiment where the endogenous variable is uncorrelated with the construction's dependent variable. Therefore, we ran controlled experiments, which we explain subsequently, where participants were randomly assigned to different affiliation levels, creating exogenous variation.
For the primary instrumental variable approach, we follow recent research (e.g., [20]; [51]) that uses instruments based on agent behavior in categories (or firms) different from the focal category (or a firm). Following this approach, we use the mean (across ideas) of affiliations on day t − 1 of all ideas in our Kickstarter data, which are in a category different from that of the focal idea as the primary instrument for Affilit − 1 in Kickstarter. For example, for an observation about an idea on movies on December 22, this instrument is the mean of affiliations on December 22 of all ideas in our data that are not in the movies category. This instrument is correlated with Affilit − 1 (correlation = .16).
Conceptually, this instrument is appealing because of the interdependencies across different parts of the global affiliation network on Kickstarter (i.e., the affiliation network across all ideas seeking funding concurrently), thus satisfying the relevance criterion. However, because most backers only back one idea (i.e., affiliation is sparse), the mean affiliation across ideas in other categories is very unlikely to be related to the unobserved component of the focal idea's funding outcome in Equation 1 providing the basis for identification. Further, a category-level measure of affiliation should remain unaffected by idea-level factors, especially if the idea is from a different category. A category-level measure should not correlate strongly with idea-day-level idiosyncratic shocks from another category, thus meeting the exclusion criterion. The first-stage equation is specified as
Graph
( 2)
The R2 for the first stage regression without the instrument (i.e., assuming that λ1 = 0) is.365 and with the instrument is.385, showing that the instrument's addition improves the in-sample model fit. The estimate of λ1 is.42 (p < .01). The corresponding F-statistic for the F-test of excluded instruments is 879.83, far exceeding the threshold value of 10 ([55], p. 522). The large value of the Anderson–Rubin statistic (F( 1, 28,300) = 298.41) rejects the null hypothesis that the instrument is weak. We show in robustness analyses that the estimates are consistent across the use of alternative instruments. We also instrument for the interaction of affiliation and the number of communal words contained in the updates (CommunalUpdatesit − 1). Following [44], the instrument for this interaction variable is the interaction of the instrument for affiliation and CommunalUpdatesit − 1. We do not instrument for the interaction of affiliation and the number of Facebook shares, because the Durbin–Wu–Hausman test of the hypothesis that this regressor is exogenous could not be rejected (χ2 = .055, p > .1). Furthermore, the sharing activity of a specific idea on a social media platform other than Kickstarter is conceptually independent of its funding outcome on Kickstarter.
First, we present the parameter estimates of the instrumental variable regression models estimated on the Kickstarter data and then discuss robustness checks. We present estimates of five models, with and without instruments, and the sequential addition of interactions in Table 5. M1–M4 do not have interaction effects, and while M1 ignores endogeneity, M2, M3, and M4 correct for it and show that the results are robust to different instruments. The results from the full model specified in Equation 1 are reported in M5, which we discuss next.
Graph
Table 5. Coefficient Estimates of the Fixed-Effects Regression Model of Daily Funding of Ideas on Kickstarter.
| Variable | M1 | M2 | M3 | M4 | M5 (Final Model) |
|---|
| Backer affiliation of idea i by day t − 1 (Affilit − 1) | −.04***(.01) | −.86***(.06) | −1.88***(.41) | −.80***(.06) | −.87***(.06) |
| Cumulative number of backers funding idea i by day t − 1 (CumBackersit − 1) | .15**(.06) | 1.92***(.15) | 4.13***(.90) | 1.79***(.15) | 2.02***(.16) |
| Cumulative number of updates by creator of idea i by day t − 1 (CumUpdatesit− 1) | −.05**(.02) | −.06**(.03) | −.07*(.04) | −.06**(.03) | −.05*(.03) |
| Amount of funding of idea i (in $) on day t − 1 | −.03(.02) | .01(.02) | .07**(.03) | .01(.02) | .01(.02) |
| Proportion of funding goal of idea i achieved by day t − 1 | −.09(.06) | −.23***(.07) | −.39**(.10) | −.22**(.07) | −.17**(.07) |
| Proportion of funding duration of idea i completed by day t − 1 | −.27(.99) | 1.12(1.10) | 2.84*(1.65) | 1.02(1.09) | .63(1.10) |
| Closeness centrality of idea i as of day t − 1 | 2.54(3.12) | 2.43(3.54) | 8.46(5.30) | 1.99(3.48) | 2.71(3.54) |
| Betweenness centrality of idea i as of day t − 1 | −.02**(.01) | −.02**(.01) | −.03*(.02) | −.02**(.01) | −.02*(.01) |
| Eigenvector centrality of idea i as of day t − 1 | −.74(.82) | −4.39*(2.33) | −8.95*(4.66) | −4.13*(2.21) | −4.40*(2.67) |
| Last week (1 if day t is in the last week of funding of idea i, 0 otherwise) | .25(.21) | .13(.24) | .03(.35) | .14(.24) | .09(.25) |
| Cumulative number of communal words in updates by creator of idea i by day t − 1 (CommunalUpdatesit − 1) | −.04(.03) | −.02(.03) | −.01(.04) | −.02(.03) | −.29*(.15) |
| Interactions of Backer Affiliations | | | | | |
| Affilit − 1 × CommunalUpdatesit − 1 | | | | | −7.68**(3.80) |
| Affilit − 1 × Number of Facebook shares of idea i | | | | | −.006***(.001) |
| Fixed effects for each idea i | Yes | Yes | Yes | Yes | Yes |
| Fixed effects for each day t | Yes | Yes | Yes | Yes | Yes |
| Instrument for Affilit − 1 | No | Yes | Constra | Other | Yes |
- 3 *p < .10.
- 4 **p < .05.
- 5 ***p < .01.
- 6 Notes:"Constra" refers to [ 7] measure of constraint of the focal idea. "Other" instrument refers to the instrument constructed from Indiegogo data.
We find that affiliation among backers has a consistent negative effect on the funding of ideas on Kickstarter (β = −.87, p < .01). This effect persists despite the inclusion of idea-specific fixed effects, daily fixed effects, controlling for lagged funding, and the prior number of backers of the idea. We corroborate extant findings on herding (e.g., [66]) and additionally show that affiliation plays a key role and that its effect is negative.
Concerning the moderators, the creator's engagement measured as using communal words in updates further strengthens the negative effect of affiliation, perhaps because of a heightened licensing effect (β = −7.68, p < .01). For backer engagement, we find the negative effect of affiliation is stronger as backer engagement, measured as the number of Facebook shares of the idea by backers, increases (β = −.006, p < .01). One explanation of this is that while individuals share on Facebook for various motives, the primary motivation is prosocial, and others seeing the shares likely see them as such, strengthening the vicarious moral licensing effect ([29]).
Concerning control variables, the greater the number of backers of an idea before the focal day, a measure of herding, the more funding the idea will attract on the focal day (β = 2.02, p < .01). This indicates that the total number of backers for an idea may act as a signal of its quality or potential worthiness, a finding that is consistent with prior research (e.g., [34]). The current research replicates this effect and demonstrates that social structure influences behavior beyond the herding effect. Moreover, this theory supports our contention that affiliation, measured by cobacking, drives the negative effect, not herding. We also find that the total number of updates posted by the creator has a negative effect on crowdfunding success (β = −.05, p < .10), although this effect is not significant across all model specifications. The effect of the proportion of the funding goal which was achieved on the previous day is negative (β = −.17, p < .05), perhaps suggesting a preference to fund underfunded ideas. For network centrality measures, we find that betweenness (β = −.02, p < .05) and eigenvector centrality (β = −4.40, p < .05) have a negative effect on funding. The negative effects of these second-order network measures, compared with the positive effect of number of backers (proxy for first-order network effect), highlight the complexity in flow of information on the network and are consistent with findings from prior studies (e.g., [35]). This is perhaps because these measures indicate the backers' ability to identify and fund salient opportunities, or access to information from their overall networks about idea quality based on indirect ties across the whole network, not just direct ties. Thus, the effects also highlight the importance of distinguishing direct and indirect aspects of how networks operate in community contexts.
To ensure that outliers are not driving our results, we estimate the main model (M5) after dropping the top 10th percentile of observations (which have affiliation values greater than 7), yielding a significant and negative estimate of the affiliation coefficient (β = −.87, p < .01). We find a similar negative effect in models estimated on various subsets of the data. To investigate if specific categories of ideas drive our results, we estimate the model separately for each category's ideas. We find a negative effect of affiliation for 11 out of 12 categories, with the most negative effect of affiliation in the ideas from the photography and technology categories. Our estimate of affiliation's effect is negative but not statistically significant for the "dance" category, which accounts for just 21 out of 2,021 ideas in our data.
We conducted several robustness analyses. First, we estimated the model on Indiegogo data; the results are quite consistent (see Web Appendix F). Second, we estimated the model on Kickstarter data using three alternative sets of instruments: discrete latent instrumental variables, an instrument constructed using affiliation from another platform, and a network-based instrument (see Web Appendix G). Third, we estimated probit, logit, and Tobit models of funding success and checked the robustness of our results to two alternative measures of affiliation (Web Appendix H). All analyses show that our results are robust. Next, we report three experiments in which we probe the effect of affiliation, the underlying process, and a moderating factor to further validate our empirical model.
In the first experiment, we demonstrate the negative effect of affiliation in a controlled experimental setting. In the second experiment, we validate vicarious moral licensing as an underlying mechanism and rule out uniqueness as one potential alternative explanation. In the third experiment, we show how the idea's description might moderate the effect of affiliation.
We conducted Experiment 1 on MTurk with 200 North American residents[12] (Mage = 35.26 years; 49.8% women; 42.6% have previously funded a crowdfunding idea). We presented participants with two ideas seeking funding (both real ideas from Kickstarter; see Web Appendix I). First, participants saw a screenshot of a website created by a graphic designer to look like an idea page on a real crowdfunding platform (e.g., [ 8]; [65]).
Consistent with prior research, participants were given money beyond study payment, creating an incentive-compatible dependent measure ([21]; [39]). Participants were told, "As part of this study, you will receive a $2 bonus. You can use some or all of this money to fund this project." They were then asked how much they would give toward the idea on a nine-point scale with dollar amounts in $.25 intervals, ranging from $0 to $2.00. If participants chose "$0" and opted to keep the full bonus, they were then forwarded to the end of the survey and were paid the original MTurk fee as well as the $2 bonus. If participants used any of their bonus to fund the first idea, they were included in our primary analyses. Ninety-three participants opted not to fund the first idea, leaving us with 107 participants. Four participants were removed who indicated that they had a child affected by autism, the focus of one of the two ideas, and were inclined toward funding but would opt to put the money toward helping their child. All participants completed the dependent measures. Two participants were removed for spending less than a second on the manipulation, leaving us with 101 participants. Participants then saw a screenshot of a second website designed to look like an idea on a crowdfunding platform (for details, see Web Appendix I).
The screenshot included idea information and a list of recent backers shown on the screen's right side. Participants in the high-affiliation condition saw a high overlap in the number of backers across the two ideas. Participants in the control-affiliation condition saw the same number of backers, but the names on the two lists did not overlap. A manipulation check confirmed the effectiveness of the manipulation. All participants who funded the first idea were told that they would receive an additional $2 bonus to keep or use to fund the second idea. Their decision on a nine-point scale ranging from $0 to $2.00 in $.25 intervals served as the outcome. At the end of the study, participants were given the money that they chose to keep as a bonus, and the remainder (i.e., what they chose to fund each of the ideas) was put toward each crowdfunding idea. Finally, participants responded to a set of demographic measures (e.g., age, gender, whether they had previously funded an idea on a crowdfunding platform). A one-way analysis of variance showed a significant effect of affiliation on the funding of the second idea (F( 1, 99) = 4.05, p < .05). As we expected, those in the high-affiliation condition funded less than those in the control-affiliation condition (Mhigh = 4.27, SD = 2.27 vs. Mcontrol = 5.27, SD = 2.70). Of the $2 bonus, those in the high-affiliation condition chose to fund $.82 toward the focal idea, while those in the control-affiliation condition chose to fund $1.07, on average.
The first experiment confirmed the negative effect of affiliation in the lab setting, validating our primary empirical finding that affiliation negatively affects crowdfunding success.
In the second experiment, we measured two potential mediators in an attempt to document "a" mediating process (i.e., the mediating process given our stimuli and procedures) as opposed to "the" mediating process (i.e., a single mediating process that is operative across all crowdfunding contexts; e.g., [ 6]). We propose vicarious moral licensing as a mechanism for the negative impact of affiliation on funding and test need for uniqueness as an alternative mechanism ([59]).
We conducted the study on MTurk with 228 North American residents (Mage = 39.57 years; 54.4% women; 38.2% had previously funded an idea on an online crowdfunding platform). All participants spent adequate time on the manipulation. Three participants did not complete the dependent measures, resulting in an effective sample of 225 participants. Participants were told to imagine that they had $50 and were asked to choose one idea to fund from a set of four real ideas seeking funding on Kickstarter and across categories (e.g., technology, nonprofits, arts/film); details appear in Web Appendix I. After this decision, they read about a second idea that they were told is seeking funding. Those in the high-affiliation condition were told that many of the backers who funded the first idea they chose also funded the focal idea. Those in the control affiliation condition were provided no information about other backers' funding decisions. A pretest confirmed the effectiveness of the manipulation (see Web Appendix I). Next, participants responded to two items to capture vicarious moral licensing ("Based on the funding behavior of cobackers, I do not feel the need to fund [focal idea]" and "Based on the funding behavior of cobackers, I do not feel obligated to fund [focal idea]"; M = 4.10, SD = 1.48; r = .72) and two items to capture uniqueness ("If I funded [focal idea], my decision to fund would say a lot about me as a unique individual" and "If I funded [focal idea], it would help me stand out from the crowd"; M = 3.68, SD = 1.45; r = .81).
Next, we asked participants how much money they would pledge toward funding the subsequent focal idea (range: $0–$5,000, the total needed to hit the focal idea's funding goal). Consistent with prior research and our empirical model, we log-transformed funding ([36]). Finally, participants completed demographic questions.
As expected, we found a negative effect of affiliation on funding (F( 1, 223) = 4.29, p < .04) such that those in the high-affiliation condition reported a lower funding amount than those in the control condition (Mhigh = 3.17, SD = 2.49 vs. Mcontrol = 3.82, SD = 2.24) or in raw numbers (Mhigh = $256.96, SD = $674.40 vs. Mcontrol = $339.88, SD = $875.90). A one-way analysis of variance showed a significant effect of affiliation on the licensing measure (F( 1, 223) = 3.89, p = .05). As we expected, those in the high-affiliation condition agreed more with the licensing measure, indicating less need to fund than those in the control condition (Mhigh = 4.29, SD = 1.57 vs. Mcontrol = 3.90, SD = 1.37). However, there was no significant effect of affiliation on uniqueness (Mhigh = 3.56, SD = 1.52 vs. Mcontrol = 3.80, SD = 1.36; F( 1, 223) = 1.53, p = .22). We then assessed the indirect effects of the two mediators on funding. The results indicate that licensing was a significant mediator (95% confidence interval does not include 0: [−.4423, −.0003]), but uniqueness was not (95% confidence interval: [−.5479,.1142]).
In this experiment, we replicated the negative effect of affiliation and uncovered vicarious moral licensing as an underlying mechanism. Although we did not find an effect of affiliation on uniqueness in this study, we note that uniqueness may operate more strongly for some ideas and some individuals, providing an interesting avenue for future research on crowdfunding ([59]).
In Experiment 3, we explored the role of a moderator: how the creator describes the idea. We theorized that the negative effect of affiliation occurs in a crowdfunding context, at least partly due to its communal nature and how the ideas are presented to potential backers. We conducted the third experiment on MTurk with 206 North American residents (Mage = 38.81 years; 46.1% women; 42.2% have previously funded an idea on an online crowdfunding platform). All participants completed the dependent measures. Three participants who spent less than one second reading the manipulation were removed, resulting in N = 203. We manipulated two factors between participants: ( 1) affiliation (high vs. control) and ( 2) idea description (more vs. less communal).
As in Experiment 2, participants read about an idea currently seeking funding on Kickstarter and were told to imagine that they had funded this idea (see Web Appendix I). We used the same manipulation of affiliation as in Experiment 2. Those in the high-affiliation condition were told that many backers who funded the first idea they chose also funded the focal idea. Those in the control-affiliation condition were not provided any information about other backers' funding decisions. Participants then read about diveLIVE, a technology that allows divers to talk underwater while streaming live video to the internet. diveLIVE, the focal idea, was described as more or less communal with small changes (e.g., "Let's learn about the oceans" vs. "This product uses technology to take videos of the oceans").
Next, participants indicated how much money they would pledge toward diveLIVE, the focal idea (range: $0–$20,000, the total needed to hit the focal idea's funding goal). Consistent with prior research, our empirical model, and Experiment 2, we log-transformed funding ([36]) for analysis but provide results in raw numbers for ease of interpretation. Finally, participants completed demographic questions.
We found evidence for a main effect of idea description (F( 1, 199) = 13.86, p < .01) consistent with prior research, which finds that ideas described as more communal tend to be more successful than those described as an investment opportunity ([ 3]). More importantly, we found an interaction between the two manipulated factors (F( 1, 199) = 5.84, p < .02). As we expected, when the idea was described as more communal, those in the high-affiliation condition reported lower funding than those in the control-affiliation condition (Mhigh = $2,155.32, SD = $3,998.08 vs. Mcontrol = $4,073.04, SD = $5,316.08; t(199) = 2.09, p < .04). When the idea was described as less communal, there was no effect of affiliation on funding (Mhigh = $2,572.33, SD = $4,933.01 vs. Mcontrol = $1,868.68, SD = $4,039.03; t(199) = −1.34, p = .18; see Figure W9 in Web Appendix I). The third experiment established that the negative effect of affiliation is stronger when creator's use more communal words in the description of the idea.
As discussed previously, we find a negative moderating effect of the number of communal words in updates posted by creators. To validate the third experiment with converging evidence, we returned to our secondary data to examine how the number of communal words in the idea description influenced the relationship between affiliation and funding behavior across thousands of crowdfunding ideas (e.g., [42]). This would establish how the use of communal words in creator's updates as well as in the idea's description would influence the effect of backer affiliation and highlight the importance of the communal mechanism. We used the same text dictionary that we created for coding communal words in updates and coded the description of every idea in our sample. The median number of communal words in an idea description is 3 (M = 6.1). We then created two subsets of our data based on a median split of the number of communal words used in describing the idea. We estimated the model separately on each subset and find that the coefficient of affiliation is less negative for ideas described using three or fewer communal words (M = −.92, SE = .09) than for ideas described using four or more communal words (M = −1.25, SE = .15). Replacing the number of communal words in this analysis with the ratio of the number of communal words to the total number of words does not affect this result, nor does splitting the data on the basis of the average number of communal words instead of the median. Finally, the effect of affiliation is less negative for ideas with no communal words than for ideas with at least one communal word. This provides real-world evidence for the role of idea description on the relationship between affiliation and funding behavior, validating our theory and experimental evidence.
In summary, these findings further support our reasoning that the negative effect of affiliation is driven, at least in part, by the communal nature of crowdfunding and the prosocial mindset that it prompts ([50]). When an idea is described as more communal, these prosocial goals are exacerbated, leading potential backers to feel that they do not need to fund the idea because these affiliated others are funding it (e.g., [37]). However, when an idea is described as less communal, this effect is mitigated. Next, we discuss our results and develop implications for theory and practice.
We establish a negative effect of affiliation on the crowdfunding success of ideas using a large empirical study and then validating the effect through experiments. We provide preliminary insights into the role of vicarious moral licensing as the underlying mechanism for this effect and investigate the moderating role of creator and backer engagement. The licensing effect and its role in reducing backers' perceived obligation to fund ideas could make backers less likely to fund or fund with less money if they opt to fund, both of which could explain the negative effect at the idea level. We begin with a focus on the novel contribution of our finding concerning affiliation, discuss the economic implications of our results, and identify the primary contributions of our research and how it paves the way for future research.
The negative effect of affiliation among backers in crowdfunding is distinct from and in addition to the positive effect of herding due to the crowd's size shown in prior research (e.g., [66]). We establish an inherent tension between the positive effect of crowd size and the negative effect of backer affiliation in crowdfunding. Thus, we show that, in addition to relying on crowd size, backers make inferences based on the behavior of affiliated others in a crowdfunding context. A 10% daily increase in number of backers leads to an additional 20.2% in funding or an increase of US$83/day (i.e., the herding effect). In contrast, a 10% daily increase in backer affiliation leads to an 8.7% decrease in funding or a decrease of US$36/day, offsetting the increase due to number of backers by 43%. Our results concerning affiliation are both statistically and economically meaningful and highlight the need to recognize the tension between increasing the number of backers and limiting the ill effects of affiliation.
Interestingly, Kickstarter stopped disclosing the prior backers' list on an idea's page as of the time of writing this article. This policy change is consistent with our results. If backer identities remain unknown, potential backers cannot infer affiliation, and therefore ideas cannot be negatively impacted by backer affiliation. Other crowdfunding platforms should reevaluate disclosure policies about past backers of an idea or perhaps reconsider whom they show at the top of their backer lists.
So how might creators mitigate the negative effects of affiliation? The moderation effects from our results provide actionable insights for creators seeking crowdfunding from potential backers and considering what platforms to pursue. Our results concerning the interaction between affiliation and creator engagement show that creators can subdue the negative effects of affiliation by carefully crafting the idea description and updates, avoiding communal language.
Further, while it appears that encouraging backers to share the idea on social media might be counterproductive because it strengthens affiliation's negative effect, the impact is small and should not be a major concern. The change in the marginal effect of affiliation as sharing by backers increases is small, indicating that change in backers' engagement, while statistically significant, does not have a meaningful effect on crowdfunding. Doubling the number of Facebook shares from its mean of 79 to 148 strengthens the negative effect of affiliation by.42% and translates to a decline of US$1.72/day.
We developed recommendations for creators and examples of best practices from our data set (see Table 6). For example, creators should focus on the idea's inherent purpose and objective value in its description and avoid using too much communal language (e.g., cooperate, partner, support) in the idea description and updates. Overall, we recommend that platforms educate creators on how best to structure communication with backers and guide creators in meeting their goals. Backers could perhaps learn to interpret such updates better and use the information provided by the backer to qualify what they infer from the community.
Graph
Table 6. Actionable Outcomes for Managers Recommendations for Idea Descriptions and Updates.
| Finding | Recommendations for Creators | Examples from Kickstarter Data |
|---|
| Interaction between affiliation and communal words in idea description | Focus on the idea's inherent purpose as opposed to a focus on community. | The Drone PocketIdea Description: "The world's first multicopter that'spowerful enough to carry a high-quality action camera and folds up smaller than a 7 in tablet."Key technology features outlined prominently on idea's home page.Total Amount Raised: $929,212Pegasus Touch Laser SLA 3D PrinterIdea page includes recent press articles with links that highlight idea's featuresTotal Amount Raised: $819,535 |
| Use noncommunal words (e.g., "you" vs. "we") in idea description. | The Floyd Leg"The Floyd Leg gives you the framework to take ownership of your furniture by allowing you to create a table from any flat surface" (emphasis added)Total Amount Raised: $256,273 |
| Avoid thanking backers too much in idea description, as it can make the project appear needy. | "First off, I want to say thanks for checking out of project. Every single person that takes the time to look at our project means the world to us." |
| Do not describe idea with overemphasis on communal language (e.g., "support," "team"). | "As we approach Thanksgiving, I continue to be thankful for the patience and support that the unsung backers have shown with our team." |
| Interaction between affiliation and communal words in updates | Do not show too much appreciation via updates for funding as it is progressing. | "Thanks to all of you who pledged for this campaign. We really appreciate your continued support." |
| Minimize communal language (e.g., partner) in updates. | "Your first duty as partners with us on this project; should you choose to accept..." |
Our results about the mechanism provide insights on how platforms and creators should engage with backers. Research has shown that licensing is a nonconscious effect and can be mitigated by making individuals aware of their behavior ([27]). Particularly in this type of vicarious moral licensing, highlighting individuals' uniqueness and independent identity may also mitigate the negative effect of affiliation on funding ([30]; [37]; [43]). If creators expect high overlap among backers, they could describe their ideas using less communal language, thereby lowering the licensing effect. Our results suggest that vicarious licensing might overwhelm other relevant idea information, potentially leading to suboptimal backer decisions. In line with our findings, backers might, in some cases, pay more attention to signals from affiliated others rather than from the whole crowd.
For crowdfunding platforms, our findings provide a rationale for why there might be room for new crowdfunding platforms to thrive and grow. Although several crowdfunding platforms have flourished in the past decade, Kickstarter, Indiegogo, and GoFundMe have arguably dominated the market. Other once-popular platforms, such as Sellaband and PledgeMusic, have failed. Large platforms with millions of backers might pose high entry barriers to new entrants. However, our findings point to one source of competitive advantage for newer platforms: negative affiliation effects are more likely to occur in well-established platforms with large backer communities. Strategically building diverse and unaffiliated communities of backers might confer a competitive advantage to new platforms. Our results show that this can be achieved by expanding the number of categories of ideas, as affiliation's negative effect may be mitigated as backers of ideas across different categories may be less likely to coback ideas. The failure of category-specific platforms such as Sellaband (music), and the relative success of platforms hosting diverse ideas, such as Kickstarter, provides support for this reasoning. Second, platforms allocating marketing resources across existing and new backers (e.g., allocating social media spending across established markets such as Los Angeles and new markets such as Lima) could perhaps view our results as a reason to divert resources away from backer-dense markets. Third, platforms that provide backer information may also want to use algorithms that promote unaffiliated (vs. affiliated) backers, for example, by highlighting first-time backers. Finally, drawing on our results about creator engagement, we recommend that platforms educate creators on how to design better backer communication.
Insights from our study are relevant to other types of crowdsourcing platforms as well. For example, participants on LEGO's Ideas, which focuses on ideation, and SeedInvest, which helps raise equity, could mitigate the negative effects of affiliation, for example, by describing initiatives as less communal and by posting updates with less communal language. Our findings are also applicable to crowdfunding contests (e.g., [ 9]; [25]), where participants could be encouraged to vote across categories to reduce coparticipation and help them break away from the adverse effects of groupthink.
We highlight several areas of inquiry for future research. Reward structures could impact the role of affiliation in crowdfunding and thus merit attention (e.g., [56]). Fake reviews have been investigated in the online context (e.g., [67]), and it would be interesting to explore the veracity of idea descriptions and creator updates. In addition to affiliation, which we study, other network characteristics such as clans and core–periphery structures ([60]) could explain the nature of information flow across affiliation structures.
As interest in crowdfunding increases, interesting research questions continue to emerge. We believe that our research explores important questions concerning crowdfunding that involve backer affiliation and community structure, and we hope to lay the foundation for future studies in the domain.
sj-pdf-1-jmx-10.1177_00222429211031814 - Supplemental material for Do Backer Affiliations Help or Hurt Crowdfunding Success?
Supplemental material, sj-pdf-1-jmx-10.1177_00222429211031814 for Do Backer Affiliations Help or Hurt Crowdfunding Success? by Kelly B. Herd, Girish Mallapragada, and Vishal Narayan in Journal of Marketing
Footnotes 1 The authors contributed equally to this research and are listed randomly.
2 Hari Sridhar
3 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
4 The author(s) received no financial support for the research, authorship, and/or publication of this article.
5 Girish Mallapragada https://orcid.org/0000-0001-7425-9793 Vishal Narayan https://orcid.org/0000-0001-7230-3084
6 Online supplement:https://doi.org/10.1177/00222429211031814
7 Results for this analysis are available from the authors upon request.
8 On many platforms, such information is readily available on the idea's home page (e.g., https://gogetfunding.com, https://www.piggybackr.com).
9 Clustering coefficient, a related network measure, deserves mention. A node's clustering coefficient is the proportion of nodes in its neighborhood that are connected to each other ([61]). In our context, the clustering coefficient of an idea would be the proportion of existing backers that have other cobacking relationships with each other.
We transformed this variable, such that the transformed variable is 0 if Affilit − 1 × CommunalUpdatesit − 1 = 0 and the transformed variable is 1 if Affilit − 1 × CommunalUpdatesit − 1 > 0. Because Affilit − 1 × CommunalUpdatesit − 1 = 0 for 93.3% of all observations, this transformation does not significantly change the original variable but offers the advantage of reduced multicollinearity. The variance inflation factor of the transformed variable is 1.9.
Our results are robust to the inclusion of four additional controls: (1) a dummy for whether the funding goal has been met; (2) the number of backers of the focal idea on the focal day; (3) "backer propensity," a measure of how likely backers are to fund the project; and (4) "backer longevity," a measure of how long backers have been active on the platform.
Because this study involved real pay, we restricted participants to exclude those who had completed over 1,500 studies, as these "professional" MTurkers may behave in significantly different ways, particularly when monetary bonuses are included ([63]).
References Agrawal Ajay , Catalini Christian , Goldfarb Avi. (2015), " Crowdfunding: Geography, Social Networks, and the Timing of Investment Decisions ," Journal of Economics & Management Strategy , 24 (2), 253 – 74.
Allen B.J. , Chandrasekaran Deepa , Basuroy Suman. (2018), " Design Crowdsourcing: The Impact on New Product Performance of Sourcing Design Solutions from the 'Crowd' ," Journal of Marketing , 82 (2), 106 – 23.
Allison Thomas H. , Davis Blakley C. , Short Jeremy C. , Webb Justin W.. (2015), " Crowdfunding in a Prosocial Microlending Environment: Examining the Role of Intrinsic Versus Extrinsic Cues ," Entrepreneurship Theory and Practice , 39 (1), 53 – 73.
Bayus Barry L. (2013), " Crowdsourcing New Product Ideas Over time: An Analysis of the Dell IdeaStorm Community ," Management Science , 59 (1), 226 – 44.
Berger Christoph , Winkels Martin , Lischke Alexander , Höppner Jacqueline. (2012), " GazeAlyze: a MATLAB Toolbox for the Analysis of Eye Movement Data ," Behavior Research Methods , 44 (2), 404 – 19.
Buechel Eva C. , Janiszewski Chris. (2013), " A Lot of Work or a Work of Art: How the Structure of a Customized Assembly Task Determines the Utility Derived From Assembly Effort ," Journal of Consumer Research , 40 (5), 960 – 72.
Burt Ronald S. (1992), Structural Holes: The Social Structure of Competition. Boston : Harvard University Press.
Burtch Gordon , Ghose Anindya , Wattal Sunil. (2013), " An Empirical Examination of the Antecedents and Consequences of Contribution Patterns in Crowd-funded Markets ," Information Systems Research , 24 (3), 499 – 519.
Camacho Nuno , Nam Hyoryung , Kannan P.K. , Stremersch Stefan. (2019), " Tournaments to Crowdsource Innovation: The Role of Moderator Feedback and Participation Intensity ," Journal of Marketing , 83 (2), 138 – 57.
Clark Margaret S. , Mils Judson. (1993), " The Difference Between Communal and Exchange Relationships: What It Is and Is Not ," Personality and Social Psychology Bulletin , 19 (6), 684 – 91.
Cowley Stacy. (2016), "Global Brands, Taking Cue from Tinkerers, Explore Crowdfunding," The New York Times (January 6), https://www.nytimes.com/2016/01/07/business/global-brands-taking-cue-from-tinkerers-explore-crowdfunding.html.
Dai Hengchen , Zhang Dennis J.. (2019), " Prosocial Goal Pursuit in Crowdfunding: Evidence from Kickstarter ," Journal of Marketing Research , 56 (3), 498 – 517.
Decety Jean , Grèzes Julie. (2006), " The Power of Simulation: Imagining One's Own and Other's Behavior ," Brain Research , 1079 (1), 4 – 14.
Durbin Joe. (2017), "The Oculus Acquisition May Cost Facebook $3 Billion, Not $2.3 Billion," Upload (January 19), https://uploadvr.com/oculus-acquisition-3-billion/.
Eckhardt Giana M. , Houston Mark B. , Jiang Baojun , Lamberton Cait , Rindfleisch Aric , Zervas Georgios. (2019), " Marketing in the Sharing Economy ," Journal of Marketing , 83 (5), 5 – 27.
Fan Tingting , Gao Leilei , Steinhart Yael. (2020), " The Small Predicts Large Effect in Crowdfunding ," Journal of Consumer Research , 47 (4), 544 – 65.
Faust Katherine. (1997), " Centrality in Affiliation Networks ," Social Networks , 19 (2), 157 – 91.
Fiske Alan P. (1992), " The Four Elementary Forms of Sociality: Framework for a Unified Theory of Social Relations ," Psychological Review , 99 (4), 689 – 723.
Gerber Elizabeth M. , Hui Julie. (2014), " Crowdfunding: Motivations and Deterrents for Participation ," ACM Transactions on Computer-Human Interaction , 20 (6), 24 – 32.
Germann Frank , Ebbes Peter , Grewal Rajdeep. (2015), " The Chief Marketing Officer Matters! " Journal of Marketing , 79 (3), 1 – 22.
Goenka Shreyans , Van Osselaer Stijn M.J.. (2019), " Charities Can Increase the Effectiveness of Donation Appeals by Using a Morally Congruent Positive Emotion ," Journal of Consumer Research , 46 (4), 774 – 90.
Goldstein Noah J. , Cialdini Robert B.. (2007), " The Spyglass Self: A Model of Vicarious Self-Perception ," Journal of Personality and Social Psychology , 92 (3), 402 – 17.
Hong Yili , Hu Yuheng , Burtch Gordon. (2018), " Embeddedness, Prosociality, and Social Influence: Evidence from Online Crowdfunding ," MIS Quarterly , 42 (4), 1211 – 24.
Hou Rui , Li Leiming , Liu Bingquan. (2020), " Backers Investment Behavior on Explicit and Implicit Factors in Reward-Based Crowdfunding Based on ELM Theory ," PLoS One , 15 (8), e0236979.
Hurst Samantha. (2017), " Motorola Announces: Indiegogo-Collaborated 'Transform the Smartphone' Challenge Finalists Have Been Selected," Crowdfund Insider (February 9), https://www.crowdfundinsider.com/2017/02/95919-motorola-announces-indiegogo-collaborated-transform-smartphone-challenge-finalists-selected/.
Imai Susumu , Jain Neelam , Ching Andrew. (2009), " Bayesian Estimation of Dynamic Discrete Choice Models ," Marketing Science , 77 (6), 1865 – 99.
Khan Uzma , Dhar Ravi. (2006), " Licensing Effect in Consumer Choice ," Journal of Marketing Research , 43 (2), 259 – 66.
Kim Chul , Kannan P.K. , Trusov Michael , Ordanini Andrea. (2020), " Modeling Dynamics in Crowdfunding ," Marketing Science , 39 (2), 339 – 65.
Kim Cheonsoo , Yang Sung-Un. (2017), " Like, Comment, and Share on Facebook: How Each Behavior Differs from the Other ," Public Relations Review , 43 (2), 441 – 49.
Kouchaki Maryam. (2011), " Vicarious Moral Licensing: The Influence of Others' Past Moral Actions on Moral Behavior ," Journal of Personality and Social Psychology , 101 (4), 702 – 15.
Kuppuswamy Venkat , Bayus Barry L.. (2017), " Does My Contribution to Your Crowdfunding Project Matter? " Journal of Business Venturing , 32 (1), 72 – 89.
Leary Mark R. (2010), " Affiliation, Acceptance, and Belonging: The Pursuit of Interpersonal Connection, " in Handbook of Social Psychology , Vol. 2, 5th ed. Gilbert D.T. , Lindzey G. , Fiske S.T. , eds. Hoboken, NJ : John Wiley & Sons.
Li Gen , Wang Jing. (2019), " Threshold Effects on Backer Motivations in Reward-Based Crowdfunding ," Journal of Management Information Systems , 36 (2), 546 – 73.
Lin Mingfeng , Prabhala Nagpurnanand R. , Viswanathan Siva. (2013), " Judging Borrowers by the Company They Keep: Friendship Networks and Information Asymmetry in Online Peer-to-Peer Lending ," Management Science , 59 (1), 17 – 35.
Mallapragada Girish , Grewal Rajdeep , Lilien Gary. (2012), " User-Generated Open Source Products: Founder's Social Capital and Time to Product Release ," Marketing Science , 31 (3), 474 – 92.
Matthews William John , Gheorghiu Ana I. , Callan Mitchell J.. (2016), " Why Do We Overestimate Others' Willingness to Pay? " Judgment and Decision Making , 11 (1), 21 – 39.
Meijers Marijn H.C. , Noordewier Marret K. , Verlegh Peeter W.J. , Zebregs Simon , Smit Edith G.. (2019), " Taking Close Others' Environmental Behavior into Account When Striking the Moral Balance? Evidence for Vicarious Licensing, Not for Vicarious Cleansing ," Environment and Behavior , 51 (9/10), 1027 – 54.
Moradi Masoud , Dass Mayukh. (2019), " An Investigation into the Effects of Message Framing on Crowdfunding Funding Level ," Journal of Electronic Commerce Research , 20 (4), 238 – 54.
Morewedge Carey K. , Zhu Meng , Buechel Eva C.. (2019), " Hedonic Contrast Effects are Larger when Comparisons are Social ," Journal of Consumer Research , 46 (2), 286 – 306.
Nair Harikesh S. , Manchanda Puneet , Bhatia Tulikaa. (2010), " Asymmetric Social Interactions in Physician Prescription Behavior: The Role of Opinion Leaders ," Journal of Marketing Research , 47 (5), 883 – 95.
Narayan Vishal , Kadiyali Vrinda. (2016), " Repeated Interactions and Improved Outcomes: An Empirical Analysis of Movie Production in the U.S.," Management Science , 62 (2), 591 – 607.
Netzer Oded , Lemaire Alain , Herzenstein Michal. (2019), " When Words Sweat: Identifying Signals for Loan Default in the Text of Loan Applications ," Journal of Marketing Research , 56 (6), 960 – 80.
Newman Kevin P. , Brucks Merrie. (2018), " The Influence of Corporate Social Responsibility Efforts on the Moral Behavior of High Self-Brand Overlap Consumers ," Journal of Consumer Psychology , 28 (2), 253 – 71.
Papies Dominik , Ebbes Peter , van Heerde Harald J.. (2017), " Addressing Endogeneity in Marketing Models, " in Advanced Methods for Modeling Markets , Wieringa J.E. , Pauwels K.H. , Leeflang P.S.H. , Bijmolt T.H.A , eds. Berlin : Springer.
Park Eunho , Rishika Rishika , Janakiraman Ramkumar , Houston Mark B. , Yoo Byungjoon. (2018), " Social Dollars in Online Communities: The Effect of Product, User, and Network Characteristics ," Journal of Marketing , 82 (1), 93 – 114.
Pennebaker James W. , Boyd Ryan L. , Jordan Kayla , Blackburn Kate. (2015), The Development and Psychometric Properties of LIWC2015. Austin : University of Texas.
Pietraszkiewicz Agnieszka , Soppe Birthe , Formanowicz Magdalena. (2017), " Go Pro Bono: Prosocial Language as a Success Factor in Crowdfunding ," Social Psychology , 48 (5), 265 – 78.
Ransbotham Sam , Kane Gerald C. , Lurie Nicholas H.. (2012), " Network Characteristics and the Value of Collaborative User-Generated Content ," Marketing Science , 31 (3), 387 – 405.
Rossi Peter E. (2014), " Even the Rich Can Make Themselves Poor: A Critical Examination of IV Methods in Marketing Applications ," Marketing Science , 33 (5), 655 – 72.
Simpson Bonnie , Schreier Martin , Bitterl Sally , White Katherine. (2021), " Making the World a Better Place: How Crowdfunding Increases Consumer Demand for Social-Good Products ," Journal of Marketing Research , 58 (2), 363 – 76.
Sridhar Shrihari , Germann Frank , Kang Charles , Grewal Rajdeep. (2016), " Relating Online, Regional, and National Advertising to Firm Value ," Journal of Marketing , 80 (4), 39 – 55.
Srinivasan Raji , Wuyts Stefan , Mallapragada Girish. (2018), " Corporate Board Interlocks and New Product Introductions ," Journal of Marketing , 82 (1), 132 – 48.
Stalder Hannah , Stenson Sarah-Jane. (2016), " Unilever Foundry and Indiegogo Launch Crowdfunding Partnership To Accelerate Innovation," Unilever (March 8), https://www.unileverusa.com/news/press-releases/2016/unilever-foundry-and-indiegogo-launch-crowdfunding-partnership.html.
Statista (2021), " Crowdfunding - Statistics & Facts," (accessed May 5, 2021), Available at: https://www.statista.com/topics/1283/crowdfunding/.
Stock James H. , Wright Jonathan H. , Yogo Motohiro. (2002), " A Survey of Weak Instruments and Weak Identification in Generalized Method of Moments ," Journal of Business & Economic Statistics , 20 (4), 518 – 29.
Sun Yacheng , Dong Xiaojing , McIntyre Shelby. (2017), " Motivation of User-Generated Content: Social Connectedness Moderates the Effects of Monetary Rewards ," Marketing Science , 36 (3), 329 – 37.
Sunder Sarang , Kim Kihyun Hannah , Yorkston Eric A.. (2019), " What Drives Herding Behavior in Online Ratings? The Role of Rater Experience, Product Portfolio, and Diverging Opinions ," Journal of Marketing , 83 (6), 93 – 112.
Swaminathan Vanitha , Moorman Christine. (2009), " Marketing Alliances, Firm Networks, and Firm Value Creation ," Journal of Marketing , 73 (5), 52 – 69.
Tian Kelly T. , Bearden William O. , Hunter Gary L.. (2001), " Consumers' Need for Uniqueness: Scale Development and Validation ," Journal of Consumer Research , 28 (1), 50 – 66.
Wasserman Stanley , Faust Katherine. (1999), Social Network Analysis: Methods and Applications. Cambridge, UK : Cambridge University Press.
Watts Duncan J. (1999), " Networks, Dynamics, and the Small-World Phenomenon ," American Journal of Sociology , 105 (2), 493 – 527.
Wei Yanhao "Max" , Hong Jihoon , Tellis Gerard J.. (2021), " Machine Learning for Creativity: Using Similarity Networks to Design Better Crowdfunding Projects ," Journal of Marketing (published online March 9), https://doi.org/10.1177/00222429211005481.
Wessling Kathryn Sharpe , Huber Joel , Netzer Oded. (2017), " MTurk Character Misrepresentation: Assessment and Solutions ," Journal of Consumer Research , 44 (1), 211 – 30.
Xiang Diandian , Zhang Leinan , Tao Qiuyan , Wang Yonggui , Ma Shuang. (2019), " Informational or Emotional Appeals in Crowdfunding Message Strategy: An Empirical Investigation of Backers' Support Decisions ," Journal of the Academy of Marketing Science , 47 (6), 1046 – 63.
Younkin Peter , Kuppuswamy Venkat. (2017), " The Colorblind Crowd? Founder Race and Performance in Crowdfunding ," Management Science , 64 (7), 3269 – 87.
Zhang Juanjuan , Liu Peng. (2012), " Rational Herding in Microloan Markets ," Management Science , 58 (5), 892 – 912.
Zhao Yi , Yang Sha , Narayan Vishal , Zhao Ying. (2013), " Modeling Consumer Learning From Online Product Reviews ," Marketing Science , 32 (1), 153 – 69.
~~~~~~~~
By Kelly B. Herd; Girish Mallapragada and Vishal Narayan
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 44- Do Marketers Matter for Entrepreneurs? Evidence from a Field Experiment in Uganda. By: Anderson, Stephen J.; Chintagunta, Pradeep; Germann, Frank; Vilcassim, Naufel. Journal of Marketing. May2021, Vol. 85 Issue 3, p78-96. 19p. 1 Diagram, 4 Charts, 3 Graphs. DOI: 10.1177/0022242921993176.
- Database:
- Business Source Complete
Do Marketers Matter for Entrepreneurs? Evidence from a Field Experiment in Uganda
Promoting growth by differentiating products is a core tenet of marketing. However, establishing and quantifying marketing's causal impact on firm growth, while critical, can be difficult. This article examines the effects of a business support intervention in which international professionals from different functional backgrounds (e.g., marketing, consulting) volunteered time to help Ugandan entrepreneurs improve growth. Findings from a multiyear field experiment show that entrepreneurs who were randomly matched with volunteer marketers significantly increased firm growth: on average, monthly sales grew by 51.7%, monthly profits improved by 35.8%, total assets increased by 31.0%, and number of paid employees rose by 23.8%. A linguistic analysis of interactions between volunteers and entrepreneurs indicates that the marketers spent more time on product-related topics than other volunteers. Further mechanism analyses indicate that the marketers helped the entrepreneurs focus on premium products to differentiate in the marketplace. In line with the study's process evidence, firms with greater market knowledge or resource availability benefited significantly more than their peers when matched with volunteer marketers. As small-scale businesses form the commercial backbone of most emerging markets, their performance and development are critically important. Marketers' positive impact on these businesses highlights the need for the field's increased presence in emerging markets.
Keywords: differentiation; emerging markets; firm growth; marketing–entrepreneurship interface; premium products; randomized controlled field experiment; volunteer marketers
Most of the businesses are too small and utterly undifferentiated from the many others. —[ 6], p. 218) on entrepreneurial businesses in emerging markets
What role, if any, do marketing professionals play in improving the world? We propose that marketers help firms grow profitably, and their positive effects can be tremendous, especially when considering entrepreneurial firms in emerging markets. Flourishing entrepreneurs create jobs and wealth and help improve overall living standards ([ 2]; [ 6]; [14]; [47]). In the words of [20], p. 196), "Entrepreneurship is one of the most effective means to alleviate poverty in developing countries."
Entrepreneurs are ubiquitous in emerging markets ([22]). In 2010, more than 31% of the adult population in Uganda, the setting for our study, was either starting a business or running a business less than four years old ([29]). However, many emerging-market entrepreneurs struggle to make ends meet, and their firms' growth rates are low ([26]; [31]), stifling the positive impact they could have on society ([20]). As [ 6] assert, the low growth rates seem to result from most businesses being "utterly undifferentiated" and failing to attract customer interest.
Marketing helps firms differentiate by attempting to answer the question, "Why should the customer buy from the firm and not elsewhere?" (see, e.g., [10]; [33], p. 5). Thus, we examine whether entrepreneurs in emerging markets can benefit from marketers' help. As Figure 1 shows, we conducted a randomized controlled field experiment with 930 entrepreneurs to examine a virtual business support intervention in which international professionals from different functional backgrounds volunteered their time supporting Ugandan entrepreneurs via Skype video conferencing, mobile calls, emails, WhatsApp, and so on. We partnered with a nonprofit, Grow Movement, to recruit international professionals from more than 60 countries to engage in the volunteer activity.
Graph: Figure 1. Timeline and data collection.
When recruiting the professionals, Grow Movement did not focus on specific functional backgrounds; rather, the organization recruited volunteers from multiple areas with substantial business experience and time to work with an entrepreneur. Marketers made up the largest group: 26% of the volunteers. Business professionals from consulting and other functional backgrounds were also included. After being randomly assigned to the control group (n = 400) or the treatment group (n = 530), the entrepreneurs receiving the intervention were randomly matched with volunteers. The result was three exogenously determined groups of 136, 122, and 272 treated entrepreneurs working with volunteers from "marketing," "consulting," and "other" backgrounds, respectively. Each entrepreneur–volunteer pair worked virtually for two to six months to improve business performance.
Our study shows the intervention was effective, especially for entrepreneurs collaborating with volunteer marketers. Compared with the control group, firms matched with volunteer marketers increased monthly sales by 51.7%. The firms also achieved 35.8% higher profits than control firms and increased total assets by 31.0% and employees by 23.8%. Importantly, based on a standardized outcome index, only the firms matched with volunteer marketers experienced significant firm growth compared with the control group.[ 6]
Mechanism evidence suggests that the volunteer marketers tended to help entrepreneurs differentiate their businesses by focusing on the goods or services they offer.[ 7] A linguistic analysis of the meetings and interactions between volunteers and entrepreneurs indicates that the marketers spent significantly more time on product-related topics than volunteers from other functional areas. Moreover, an intermediate outcome analysis shows that entrepreneurs collaborating with volunteer marketers increased average product price, contribution, markup percentage, and value add compared with those in the control group, indicating that the firms offered more premium products after the intervention than before ([10]; [13]). In addition, we find that these premium product proxies (e.g., price) mediate volunteer marketers' effect on firm growth.
We also investigated heterogeneous treatment effects. In particular, international volunteers are unlikely to have local-market knowledge, a prerequisite for developing business differentiation ([43]), and firms require resources to deploy differentiation efforts ([30]). Accordingly, our results show that emerging-market entrepreneurs with greater ex ante market knowledge or resource availability gain the most from working with a volunteer marketer.
Our study is the first field experiment examining whether and how volunteer marketers help emerging-market entrepreneurs grow their businesses. By addressing our two research questions (i.e., the main effect and its mechanism), we add to the literature in marketing, entrepreneurship, and development economics. We advance understanding of the effectiveness of business support services, including new ways of designing virtual collaborations leveraging technology and enhancing access for emerging-market entrepreneurs. We hope the study assists organizations such as the United Nations and multinationals such as Unilever or Procter & Gamble in designing future business support services for emerging markets.
While promoting firm growth by differentiating products is a core marketing tenet, establishing and quantifying marketing's causal impact on growth is nontrivial ([ 9]). Our study causally identifies marketers' positive impact on emerging-market entrepreneur firm growth, thereby adding to the entrepreneurship literature (e.g., [36]; [56]) and research on marketing's influence within the firm (e.g., [25]; [53]).
In addition, while it may seem obvious that marketing professionals focus on differentiation and premium products, this approach may be counterintuitive in emerging markets, where consumers have limited disposable income. If emerging-market consumers can only afford inexpensive, low-quality products, premium products are likely to fail. Our study indicates that this assumption is incorrect. We show that emerging-market entrepreneurs can successfully offer premium products well-aligned with their customers' needs and wants. Thus, we provide support for [34] observation that low-income consumers in emerging markets desire premium products (see also [ 5]]). Our finding also responds to calls for research on how to operate in emerging markets ([42]).
Finally, our heterogeneous treatment effects provide guidance on which emerging-market entrepreneurs marketing interventions should target (i.e., those with greater ex ante market knowledge or resource availability). Many economists believe that emerging-market entrepreneurs fail to flourish largely due to resource constraints (e.g., [58]). Our results confirm that more resources help. However, our results also suggest that emerging-market entrepreneurs may require guidance to use available resources effectively.
Many people in emerging markets start businesses ([22]). Due to limited employment opportunities, the businesses are typically necessity-driven, created for survival rather than to address a clearly identified market opportunity. Most of the businesses are small and undifferentiated and cannot grow beyond subsistence. Many emerging-market entrepreneurs' products closely resemble other products, making it difficult to succeed and grow ([ 6]). When emerging-market firms fail to grow, gainful employment and its positive effects also stagnate ([11]; [31]).
All else equal, emerging-market entrepreneurs who operate growing businesses enjoy enhanced income and greater purchasing power. The entrepreneurs' families are able to afford quality food, education, and health care and are generally less concerned about meeting basic needs. Their employees benefit through increased wages and job stability. Stable jobs enable employees to access savings accounts and loans to purchase products such as stoves and refrigerators, which can significantly increase quality of life. Emerging-market governments and societies also benefit from growing entrepreneurial businesses, as the firms typically pay higher taxes, and the additional income can be used to enhance regulations and infrastructure (e.g., transportation, sewers, freshwater systems).
Research has shown that entrepreneurship is one of the most effective means of alleviating poverty in emerging markets ([ 6]; [20]; [47]). Scholars also suggest that businesses must clearly identify opportunities in their markets and stand out from the crowd (i.e., be sufficiently differentiated) to grow ([ 6]; [31]). Differentiation opportunities abound in emerging markets (e.g., [38]), but entrepreneurs must identify and implement them. Unfortunately, significant gaps remain in emerging-market entrepreneurs' business education and knowledge quality and relevance ([ 2]; [ 8]; [31]; [38]).
We suggest that, as a possible solution, experienced professionals could volunteer time to guide emerging-market entrepreneurs. Specifically, we suggest that virtually connecting emerging-market entrepreneurs with experienced professionals from advanced markets could facilitate differentiation. Given their functional backgrounds and experience, we believe volunteer marketers should be particularly effective for helping the entrepreneurs identify and implement viable differentiation strategies, as marketing helps firms discover market needs and customer groups, target appropriate customers, and position products so customers recognize them as distinct from others ([33], p. 5).
A recent study by [ 3] examines how remote volunteers help emerging-market entrepreneurs "pivot" their business model (broadly defined; see [44]]), thereby helping them improve their firms' sales. That study is based on the same business support intervention and data gathering as our study. However, there are key distinctions between their study and ours. First, we focus on isolating the specific impact of marketing volunteers (vs. volunteers in general) as well as how marketing volunteers help emerging-market entrepreneurs become more differentiated by offering premium products. Neither of these aspects (i.e., main effect and mechanism differences) are considered in Anderson, Chintagunta, and Vilcassim. Second, we include multiple outcome measures (e.g., profits, assets, employees, firm growth indices) beyond just sales, which is the focal outcome considered in Anderson, Chintagunta, and Vilcassim. Third, our mediation and text analyses in support of the mechanism are unique and add further distinction. Fourth, our article's interaction analyses are novel given our use of multiple business-level moderators as well as our examination of nonlinear relationships. As a result, our study provides more fine-grained information for governments, nongovernmental organizations (NGOs), researchers, and multinationals on the types of businesses and volunteers likely to lead to greater differentiation and firm growth. The two studies should therefore be viewed as complementary.
Marketing and entrepreneurship are two key responsibilities of any young firm ([17]). However, research on the combination and interaction of marketing and entrepreneurship is sparse (e.g., [36]; [39]; [56]) and suggests competing insights. [15] hints at incompatibilities between marketing and entrepreneurship, arguing that market-oriented entrepreneurial firms (i.e., those in which marketing flourishes [[32]]) fail to innovate because they are preoccupied with the market ([36]; [39]). In contrast, [56] argue that marketing significantly supports the entrepreneurship process (see also [36]). Although they do not test their predictions empirically, Webb et al. propose marketing activities and entrepreneurship processes are positively and reciprocally related.
The archetypal entrepreneurship process has five stages ([12]. The process begins with ( 1) entrepreneurial alertness, which leads to ( 2) recognizing an opportunity, followed by ( 3) innovation, ( 4) opportunity exploitation, and ( 5) enhanced performance. [56] propose that marketing—in particular an entrepreneurial firm's market orientation and marketing-mix skills—positively influences the five steps and enhances performance. The theory implicitly assumes that entrepreneurs, either themselves or through employees, have access to marketing capabilities. However, the assumption is less likely to apply to emerging-market entrepreneurs than those in advanced markets.
Research has shown that emerging-market entrepreneurs employ "sporadic and rudimentary" marketing efforts ([38], p. 49) and lack marketing knowledge and related skills ([ 2]; [31]). Most emerging market entrepreneurial ventures have few employees ([38]), and the workforce cannot compensate for the entrepreneur's lack of marketing knowledge. Thus, emerging markets are less likely to experience the positive interaction between marketing and entrepreneurship that [56] propose. However, we argue that virtual access to professionals with marketing backgrounds could help emerging-market entrepreneurs address their capability gap.
Extant research indicates that emerging-market entrepreneurs can acquire general marketing capabilities by attending broad, in-class marketing courses ([ 2]). We propose that emerging-market entrepreneurs can also acquire the skills by collaborating with an experienced volunteer from an advanced market. In contrast to group-based marketing principles courses ([ 2]), one-on-one collaborations deal directly with each entrepreneur's unique products and business challenges. Thus, regularly interacting with an experienced volunteer marketer may be more applicable to entrepreneurs than general classroom training ([14]; [38]).
Depending on their functional backgrounds, volunteers likely emphasize different business practices during their collaborations with entrepreneurs. Volunteers naturally bring their own experiences to interactions with entrepreneurs (e.g., [21]), and even when business professionals operate outside their primary functional area, past learning and conditioning affects their thinking ([55]) and leads them toward familiar solutions ([35]). Kaplan's Law states that individuals rely on familiar "tools" ([28]); thus, we expect volunteer marketers to focus on their marketing expertise during their interactions with entrepreneurs. Likewise, we expect volunteers with other backgrounds to focus on their unique skills.
Marketing education and professional development emphasizes identifying demand-increasing opportunities (e.g., [19]; [57]). Most other business functions focus on throughput. The finance, legal, and accounting functions, for example, focus internally on improving firm efficiency ([23]). A significant body of research indicates that marketers recognize market-based opportunities (e.g., [54]; [59]) and help firms differentiate ([33], p. 5; [48]). Marketers say that they keep differentiation strategies at the top of their minds (e.g., [51]). Volunteer marketers should thus be well suited and eager to help emerging-market entrepreneurs differentiate and address one cause of their low growth rates ([ 6]). Therefore, we expect emerging-market entrepreneurs to exhibit improved performance and grow their firms after interacting with volunteer marketers.
Firms often make product changes and attempt to align better with target customers' needs and wants to become more differentiated ([30], p. 628). Indeed, [43] argues that firms frequently aim to distinguish themselves from their rivals by offering differentiated products. Moreover, the emerging-market context makes it difficult for entrepreneurs to differentiate on characteristics other than product. That is, their businesses tend to be local, so differentiation tactics relying on adding new channels or advertising and promotion are less accessible. Thus, ceteris paribus, we expect volunteer marketers to focus on product-related differentiation during collaborations with emerging-market entrepreneurs.
That said, firms can use several approaches to differentiate their products ([16]), and it is not clear, a priori, which tactic emerging-market entrepreneurs working with volunteer marketers would use. Therefore, we set up our experimental design and data collection so we could explore the approaches that entrepreneurs pursued.
Studying volunteer marketers' impact on emerging-market entrepreneurs' differentiation and growth is challenging. No databases record both firm growth indicators (e.g., sales) over time for the same set of entrepreneurs and the functional backgrounds of volunteer business professionals working with the entrepreneurs. Moreover, exogeneous variation in entrepreneur exposure to the volunteers would be needed to overcome omitted variables bias (e.g., unobserved alternative factors driving firm growth) and reverse-causality concerns (e.g., substantial firm size as a prerequisite for attracting assistance). In addition, obtaining a relevant panel data set may still not solve potential bias from self-selection by entrepreneurs (i.e., varying motivations for choosing to receive assistance) and volunteers (i.e., different preferences for choosing firms to work with). We therefore conducted a two-year field experiment (see Figure 1) in which 930 Ugandan entrepreneurs were randomized into a control group (n = 400) and a treatment group (n = 530). We also randomly matched the treated firms with volunteer business professionals from different functional backgrounds.
From January to August 2015, we followed multiple steps to obtain a representative sample of emerging-market entrepreneurs running small firms in Uganda.[ 8] First, a team of 15 enumerators went door-to-door across greater Kampala, systematically covering all business hubs, marketplaces, and commercial zones. We conducted a recruitment survey of every entrepreneur who could speak conversational English, operated their firm from a physical structure, and was interested in receiving assistance from a volunteer business professional. The survey contained questions on entrepreneur and business characteristics for screening or to be used as controls in our main analysis. Our sampling frame includes the 4,043 entrepreneurs who completed the recruitment survey.
We then implemented an "established firm" scorecard, ranging from 0 to 100 points, using nine proxies from the recruitment survey: business premises, upfront investment, full-time staff, internal affairs organization, new activities and processes, business and formal education, prior corporate experience, exposure to other countries, and external ecosystem awareness. We ranked the 4,043 entrepreneurs using the scorecard and proceeded with the top 1,500 firms.[ 9] We attempted a baseline survey of the entire group; however, only 1,254 entrepreneurs completed the 90-minute site visit and audit. The survey contained business background questions, detailed financial data (e.g., sales, profits, assets, employees), and product data (e.g., descriptions, prices, costs, markups). Finally, our partner invited the qualifying 1,254 entrepreneurs to a one-on-one interview where they received details about the business support service. Our partner used the registration meeting as an additional eligibility screen and approved 930 entrepreneurs, which formed our sample. The sample includes a broad mix of firms, with business-to-consumer retailers and service providers being the most common. (For a summary of firms by industry, see Web Appendix 1.)
All 930 firms were randomly assigned to a control group (n = 400) or a treatment group (n = 530). Each treated firm was randomly matched one-to-one with a unique volunteer business professional. The randomization process was done by computer, so differences across groups were due to chance.
Two independent experts coded volunteers' background variables after the study finished using their curriculum vitae, LinkedIn profiles, and partner administrative data. The coders did not have access to entrepreneur or firm data. Volunteers' primary functional backgrounds refer to the business area or specialization in which they spent the majority of their career until project participation. The interrater reliability for coding functional backgrounds was 89.8%; all discrepancies were resolved through discussion. Background data were missing or insufficient for 38 volunteers. The 530 functional backgrounds were coded into ten areas: marketing and sales (n = 136), consulting and advisory (n = 122), finance and accounting (n = 84), strategy and general management (n = 48), engineering and research and development (n = 39), operations and supply chain (n = 23), entrepreneurs and owners (n = 18), human resources (n = 14), legal (n = 8), and unknown (n = 38).
All entrepreneurs and volunteers, as well as the partner's intervention managers, were blind to the experiment. We permitted no one to switch volunteers or entrepreneurs, and we controlled all matching steps and dyad formation. Thus, self-selection did not occur and the assignment of volunteers to treated firms was exogenously determined. This randomized matching (of volunteers and entrepreneurs) enabled us to construct treatment groups based on functional backgrounds. We set the group size minimum at 100 firms to provide sufficient statistical power and thus divided our study sample of 930 firms into four experimental groups: ( 1) treatment 1 (or marketers), which includes the 136 entrepreneurs exposed to a marketing/sales volunteer; ( 2) treatment 2 (or consultants), which includes the 122 entrepreneurs exposed to a consulting/advisory volunteer; ( 3) treatment 3 (or other professionals), which includes the 272 entrepreneurs exposed to volunteers from one of the remaining functional areas (e.g., finance, engineering, strategy, operations); and ( 4) control, which includes the 400 entrepreneurs who did not receive the intervention during the two-year study.
The identification approach enables us to isolate marketing volunteers' effect on firm growth and product differentiation. It is aligned with our research objective of understanding the relationship between volunteer marketers and emerging-market entrepreneurs.
Our intervention exposed each Ugandan entrepreneur to a volunteer in a different country and let the dyad work together for two to six months to improve firm performance. The collaborations were virtual, with every entrepreneur–volunteer interaction, sometimes multiple per week, happening via Skype video conferencing, mobile calls, and text messages. Many dyads leveraged other virtual productivity tools, such as email, Google Docs, Dropbox, and WhatsApp. Our partner, Grow Movement, provided in-country intervention managers to facilitate introductions and ensure that collaborations continued on schedule but otherwise did not intervene. The partner maintained an online project management system allowing volunteers to enter goals, track milestones, and record interaction details at biweekly intervals. Outside its basic structure, the intervention was open-ended (i.e., the volunteers had the discretion to guide the project and tailor the topics, assignments, and activities as they saw fit). Web Appendix 2 provides examples of typical entrepreneurs in the sample and their products.
The 530 volunteers approved to participate in the project initially applied online via the Grow Movement website. Our partner subsequently interviewed and vetted them to ensure we matched only committed volunteers with entrepreneurs. The volunteers had to demonstrate substantial business experience and convince Grow Movement they were willing to work with a Ugandan entrepreneur for multiple months to improve business performance. The partner did not implement prerequisites or quotas regarding volunteers' functional backgrounds. The intervention included business professionals from nearly every continent (see Web Appendixes 3 and 4). Volunteers represented more than 60 countries, with the largest number coming from the United Kingdom (28%), India (10%), the United States (9%), Germany (4%), Italy (4%), Canada (4%), Australia (3%), and Spain (3%).
The intervention featured a relatively high take-up rate, as 88% of treated entrepreneurs completed at least one of the two-week modules, each of which included multiple interactions with a volunteer (for a breakdown by treatment group, see Web Appendix 5). The first two-week module entailed arranging logistics with an intervention manager, scheduling a two-hour Skype call with the matched volunteer, traveling to a field office or internet café to hold the call, completing multiple assignments (e.g., problem identification, product details, financials, market research, goal setting), and communicating with the professional via follow-up calls, texts, and emails. Intervention compliance was relatively high. The typical collaboration lasted about 2.5 months, with the average number of completed modules varying by group (marketers = 5.04, consultants = 5.98, other professionals = 5.60).[10] However, entrepreneurs reported completing more modules (around eight in total) than were recorded in our partner's system, likely making the compliance estimate a lower bound.
Our study's intervention phase lasted roughly one year, from August 2015 to July 2016. To allow a two-year gap for potential growth from pre- to postintervention data collection, we implemented our end-line survey in May 2017. An independent auditor conducted the survey at each entrepreneur's business location under the supervision of an Innovations for Poverty Action (IPA) research manager (the Uganda office of IPA hosted our study and provided research support). Questions closely mirrored those in the baseline survey to ensure that auditors collected the same financial data (e.g., sales, profits, assets, employees) and product differentiation data (e.g., descriptions, prices, costs, markups) pre- and postintervention. We used an electronic survey tool to collect firm financial data and followed a standard aggregation, anchoring, and adjustment methodology to obtain plausible and precise estimates on key outcomes such as sales and profits ([ 4]). Our team leaders, field manager, and research manager took several rigorous auditing and verification steps to ensure that every survey was complete and accurate.[11]
Our study aims to learn whether and how volunteer marketers help emerging-market entrepreneurs improve their business performance and size. Firm growth is the main outcome of interest. We define firm growth conceptually as an increase in a firm's sales, profits, assets, or employees. We measure firm growth operationally using several indicators and two overall indices. We use aided-recall and iterative anchored-adjusted approaches to measure monthly sales and profits ([ 4]). Drawing on these measures, we constructed four composites of monthly sales and profits: ( 1) a winsorized sales composite (average of the aided-recall and anchored-adjusted sales measures after winsorizing each 1%), ( 2) an inverse-hyperbolic-sine (IHS)-transformed sales composite (average of the aided-recall and anchored-adjusted sales measures after IHS-transforming each), ( 3) a winsorized profits composite (average of the aided-recall and anchored-adjusted profit measures after winsorizing each 1%), and ( 4) an IHS-transformed profits composite (average of the aided-recall and anchored-adjusted profit measures after IHS-transforming each). Moreover, we use an iterative approach to measure the current value of all firm assets and the number of employees, again constructing four composites: ( 1) a winsorized (1%) assets composite, ( 2) an IHS-transformed assets composite, ( 3) a winsorized (1%) employees composite, and ( 4) an IHS-transformed employees composite.
Finally, we constructed two indices of firm growth. For the first index, we used the following 12 measures: ( 1) aided-recall sales winsorized, ( 2) anchored-adjusted sales winsorized, ( 3) aided-recall sales IHS-transformed, ( 4) anchored-adjusted sales IHS-transformed, ( 5) aided-recall profits winsorized, ( 6) anchored-adjusted profits winsorized, ( 7) aided-recall profits IHS-transformed, ( 8) anchored-adjusted profits IHS-transformed, ( 9) assets winsorized, (10) assets IHS-transformed, (11) employees winsorized, and (12) employees IHS-transformed. We standardized each of these 12 measures (control group as the base) and then computed the average of these values to construct the overall Firm Growth Index 1 outcome variable. For the second index, we used the following eight composite measures: ( 1) winsorized sales composite, ( 2) IHS-transformed sales composite, ( 3) winsorized profits composite, ( 4) IHS-transformed profits composite, ( 5) winsorized assets composite, ( 6) IHS-transformed assets composite, ( 7) winsorized employees composite, and ( 8) IHS-transformed employees composite. We again standardized each of these eight composite measures (control group as the base) and then computed the average of these values to construct the overall Firm Growth Index 2 outcome variable. This second index measure is the main dependent variable used in our additional analyses (i.e., intermediate effects and interaction effects). Combining the outcomes into an index better represents the construct by capturing all relevant dimensions, improving statistical power to detect effects in the same direction, and guarding against multiple hypothesis testing (e.g., [14]). Web Appendix 6 provides additional details for each firm growth indicator and index.
Given that we randomly assigned entrepreneurs to experimental groups, we estimate the effect of exposure to a volunteer business professional as the difference in average outcomes for the treatment and control firms at end line using an intention-to-treat regression:
Yi=α+β1Marketeri+β2Consultanti+β3OtherProfessionali+∑γsdi.s+δYi, b+∊i.1
Yi is the dependent variable (i.e., firm growth) for firm i at end line. Marketeri is a treatment dummy variable indicating whether a firm is randomly assigned to the marketing intervention and matched with a marketing volunteer. Consultanti is a treatment dummy variable indicating whether a firm is randomized into the consultant intervention group and matched with a consulting volunteer. OtherProfessionali is a treatment dummy variable indicating whether a firm is randomized into the other professional intervention group and matched with a nonmarketing or nonconsulting volunteer.[12] di.s comprises control variables measured preintervention, including 10 controls for baseline entrepreneur characteristics (gender, age, ethnicity, marital status, children, education level, business program, prior salaried job, previous ownership experience, and commitment), 15 controls for baseline business characteristics (founder, operating years, start-up capital, formal loans, separation of business–personal affairs, days open per week, sales frequency, business premises, location, registration, size, business practices, product competition, business-to-business customers, and markets outside neighborhood), and 10 industry fixed effects based on two-digit Standard Industrial Classification codes. We include the controls to improve estimate precision and account for any group imbalances due to attrition or spurious correlations in interaction analyses. Equation 1 also controls for the baseline value of the dependent variable, Yi, b (whenever this outcome was measured at baseline).[13] Robust standard errors are reported in all regression specifications. Because the dependent variable is continuous (e.g., sales, profits, assets, employees), we estimate Equation 1 via an ordinary least squares regression.
In our sample, 70% of the firms are run by the founder and, on average, have been in operation for nearly four years and are open 6.5 days per week. The firms are fairly formalized, with 74% maintaining separate business and personal affairs, 13% having received a financial institution loan, and 22% being formally government-registered. The average firm in the sample operates from a small stand-alone shop or larger physical premises, is located in a busy area, has monthly sales of 4.4 million UGX (∼$1,190[14]), has monthly profits of 673,000 UGX (∼$184), owns assets valued at 14.4 million UGX (∼$3,950), and employs 1.7 paid staff (excluding the owner).
Female entrepreneurs make up 40% of the sample, and 99% are local Ugandans. The typical entrepreneur is 31 years old, has 2.3 children, and has completed at least high school. On average, 55% have engaged in a prior business development program (e.g., training course), 54% are married, and 46% previously owned a business. Web Appendix 7 displays summary baseline statistics for our full sample of 930 firms.
Our experimental groups are reasonably balanced on preintervention covariates (i.e., randomization was successful; see Web Appendix 7). Out of 120 t-tests, we find six statistically significant differences in means, which would be expected by chance. Nonetheless, we control for entrepreneur and business characteristics in all regression analyses to account for group imbalances on observables.[15] We perform attrition and survival checks but do not detect differential effects among groups (see Web Appendix 9).
Given that the experimental groups do not differ in attrition or failure, our subsequent analysis includes the full sample of survivors with complete end-line surveys and key data (n = 605). We also followed the standard conservative approach for dealing with nonsurvivors in small firm studies suggested by [ 2] and rerun each analysis with nonsurvivors, obtaining qualitatively similar results.
Figure 2 provides model-free evidence for volunteer marketers' impact on firm growth. The control group decreased on the raw index measure (−.030 SD) from baseline to end line. The average change in growth for the marketer treatment group is positive (.123 SD) and significantly larger than for the control group (p =.042). We see a similar pattern of positive growth effects across our outcome measures: change in monthly sales, monthly profits, total assets, and paid employees is greater for firms exposed to a volunteer marketer than for control firms. We also plotted the four experimental groups' cumulative distribution functions for the firm growth index, which show a rightward shift for treated firms. In particular, across the distribution, it appears that entrepreneurs matched with a volunteer marketer achieved the most growth compared with the control group (see Web Appendix 10).
Graph: Figure 2. Volunteer marketers' main effects on firm growth.*p <.10.**p <.05.Notes: The y-axis represents the pre-to-post change in Firm Growth Index 2. Error bars = ±1 SE.
Table 1 presents our regression results for the volunteers' effect on firm growth. Our findings from the intention-to-treat analysis are consistent with the model free evidence. Across the outcome measures, we see significant positive main effects for the marketer treatment group (for full details, see Web Appendix 11).
Graph
Table 1. Volunteer Marketers' Main Effects on Firm Growth.
| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | (10) |
|---|
| | Monthly Sales | Monthly Profits | Total Assets | Total Employees | Firm Growth |
|---|
| | Levels: UGX | Logs: IHS | Levels: UGX | Logs: IHS | Levels: UGX | Logs: IHS | Levels: UGX | Logs: IHS | Index 1 | Index 2 |
|---|
| Treatment 1: offered marketer (yes = 1) | 2,311.757** | .245* | 292.912* | .559** | 4,386.521* | .216** | .454* | .162* | .187** | .189*** |
| (910.151) | (.133) | (158.327) | (.223) | (2,368.823) | (.105) | (.252) | (.091) | (.075) | (.072) |
| Treatment 2: offered consultant (yes = 1) | 1,210.990 | .153 | 1.116 | −.047 | 2,894.067 | .177 | .094 | .071 | .066 | .064 |
| (979.500) | (.132) | (142.183) | (.264) | (2,405.527) | (.133) | (.234) | (.089) | (.073) | (.071) |
| Treatment 3: offered other professional (yes = 1) | 970.828 | .219** | 101.796 | .302 | 2,230.143 | .103 | .108 | .104 | .076 | .076 |
| (718.131) | (.099) | (130.877) | (.217) | (1,857.546) | (.093) | (.176) | (.070) | (.060) | (.058) |
| Baseline value of dependent variable included | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| 15 business controls included | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| 10 entrepreneur controls included | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| 10 industry fixed effects included | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| R-squared | .375 | .425 | .312 | .148 | .404 | .425 | .525 | .469 | .428 | .465 |
| Sample size: total | 605 | 605 | 605 | 605 | 605 | 605 | 605 | 605 | 605 | 605 |
| Control: mean of dependent variable | 4,475.372 | 8.290 | 818.983 | 6.244 | 14,167.098 | 9.531 | 1.903 | .996 | −.013 | −.012 |
| Control: standard deviation of dependent variable | 7,625.873 | 1.251 | 1,254.108 | 2.246 | 23,190.844 | 1.227 | 2.828 | .955 | .700 | .717 |
- 40022242921993176 *p <.10.
- 50022242921993176 **p <.05.
- 60022242921993176 ***p <.01.
- 70022242921993176 Notes: Robust standard errors are in parentheses. Firm growth values in levels (sales, profits, assets) are listed as Ugandan Shillings (UGX) in thousands. Firm Growth Index 1 is the average of the 12 standardized measures of sales, profits, assets, and employees. Firm Growth Index 2 is the average of the eight standardized composites of sales, profits, assets, and employees.
We find that entrepreneurs who were matched with a volunteer marketer, on average, increased in size on multiple growth indicators. Table 1 shows monthly sales increased by 2,311,757 UGX (51.7% or.30 SD), monthly profits by 292,912 UGX (35.8% or.23 SD), total assets by 4,386,521 UGX (31.0% or.19 SD), and paid employees by.45 (23.8% or.17 SD) for marketer treatment group firms compared with control group firms. We also include the respective changes in logs (based on the IHS-transformed measures) in Table 1 for each growth indicator. Although firm growth measures commonly feature large standard errors in emerging market business studies ([37]; [38]), we find consistent coefficient magnitudes across our eight indicators (32.5% average effect size across columns 1–8 of Table 1).
Most importantly, our overall firm growth indices are positive and significant. Table 1 shows a firm growth index effect of.187 to.189 standard deviations for volunteer marketers, 2.95 times greater than that for consultants (.064 SD) and 2.49 times greater than that for other professionals (.076 SD). Taken together, the regression analysis finds a positive and meaningful treatment effect for the marketing intervention. For example, a 292,912 UGX ($80) increase in monthly profits (i.e., the marketer treatment effect in column 3) would enable the average firm in our sample to substantially expand its business premises, especially given that mean rent at baseline was 341,136 UGX per month. Moreover, as per Table 1 (column 5), growing total assets by 4,386,521 UGX ($1,200) is equivalent to a 67% rise in stock and inventory. Such working capital gains can fuel the sales engine of a small emerging-market business. Overall, the main-effect results suggest that entrepreneurs exposed to a marketer tended to grow their firms more than those who did not receive any intervention.[16]
We obtain a similar pattern of main effect results using the following alternative specifications: excluding control variables, selecting control variables via Lasso, including nonsurvivor firms, and designating the marketer treatment as the excluded base group. The main effect also continues to hold when we use difference-in-differences approaches instead of the analysis of covariance model specified in Equation 1. We further support our findings using a bounding exercise to examine attrition, where lost control group firms are assigned the treated firms' average growth values. Web Appendix 12 shows these robustness checks.
Web Appendix 13 presents additional robustness checks. The regression results show that the marketer treatment effects continue to hold, with coefficients similar to those in Table 1, when consultants and other professionals are collapsed into a single treatment group labeled nonmarketers. Critically, this lends support to the exogeneity of the marketer treatment dummy (i.e., the randomized matching of entrepreneurs and volunteers) as the effects remain similar. The nonmarketer treatment dummy variable is significant for the sales outcomes, which is consistent with [ 3] findings.
We argued that volunteer marketers help emerging-market entrepreneurs differentiate, a trait that many entrepreneurs lack and a key reason that they fail or stagnate ([ 6]). Moreover, we predicted that volunteer marketers would focus specifically on product-related differentiation strategies. However, we noted that firms can take different routes to product differentiation ([16]), and it is not clear how the entrepreneurs exposed to volunteer marketers would proceed. Thus, we set up our experimental design and data collection so we could analyze the entrepreneurs' approaches. In what follows, we present the insights from these analyses.
As we have described, the volunteers were encouraged to use Grow Movement's online project management system to summarize the topics they discussed in each entrepreneur meeting. All summaries were provided in English and saved in the partner's database. On average, 71.5% of volunteer marketers, 70.6% of volunteer consultants, and 69.1% of other volunteers provided written summaries. The entry rates were not significantly different (p >.55). We also examined average entry length; marketers averaged 959 words (SD = 1,413), consultants averaged 1,163 words (SD = 1,518), and other professionals averaged 915 words (SD = 1,380). The three groups did not significantly differ in average words used (p >.18).
Words and text provide information about their author ([52]), and analysts can aggregate text across authors to study larger groups. Because grouping individuals on the basis of shared characteristics can provide insight into their similarities and differences ([ 7]), we first organized all session summary text by treatment group. We then used topic modeling to identify underlying themes and general topics discussed during the intervention and differences in the extent to which each treatment group focused on topics. We used structural topic modeling (STM) for the analysis, removing stop words and employing stemming ([ 7]). We also removed all names. We employed the "stm: R package" developed by [45] for our analysis. No clear guidance is available for selecting an optimal number of topics for STM analysis ([ 7]). However, the semantic coherence measure of our data was highest when we set topics at K = 6. Thus, combining the statistical measure results with researcher judgment ([ 7]), we used K = 6 topics. Table 2 presents the topics extracted from the text, along with the words most likely to be present for each ([45]).
Graph
Table 2. Linguistic Analysis Insights.
| (1) | (2) | (3) | (4) | (5) | (6) |
|---|
| Topic 1: | Topic 2: | Topic 3: | Topic 4: | Topic 5: | Topic 6:Intervention Logistics |
|---|
| Proactive Behavior | Customer and Market | Business in Uganda | Products and Performance | Understanding the Firm |
|---|
| Highest probability words (in descending order) | get | custom | uganda | product | busi | client |
| will | discuss | will | shop | session | session |
| can | market | creat | sale | discuss | call |
| also | client | can | month | plan | progress |
| talk | new | busi | profit | understand | time |
| ask | servic | time | cost | manag | email |
| now | increas | page | new | cash | week |
| Examples of text | Talked to me before final decision with the loan shark | Concluded that better customer service would help | Conduct research on payment options | Has a good handle on profit and loss. Needs to focus on marketing | Understand business and what main challenges are | He had not received email |
| Started using email | Discussed the competitive analysis | Enabling Uganda's vulnerable youth | Products that are often wasted | Conducted cash flow and profitability analysis | We agreed on a time for next call |
| Will get a large dryer | What are the market needs? | Opening a bank account for business | Introduce a new line of products | Discussed revenue and cost | The session was cancelled |
| Text Devoted to Topic by Treatment Group | | | | | | |
| Treatment 1: marketer | 12% | 17% | 5% | 18% | 42% | 6% |
| Treatment 2: consultant | 15% | 22% | 3% | 10% *** | 43% | 8% |
| Treatment 3: other professional | 14% | 18% | 4% | 12% ** | 42% | 10% *** |
- 80022242921993180 *p <.10.
- 90022242921993180 **p <.05.
- 100022242921993180 ***p <.01.
- 110022242921993180 Notes: Marketers devoted significantly more text to topic 4 (18%) than consultants (10%) and the other professionals (12%). In addition, marketers devoted significantly less text to topic 6 (6%) than the other professionals (10%). Text devoted to the six topics does not significantly differ between the consultants and other professionals.
Across the three treatment groups, volunteers devoted similar amounts of text to the six topics when creating their session summaries, with one notable exception. Volunteer marketers devoted significantly more text to topic 4, which relates to products, than consultants and other professionals. Topic 4 captures text such as "She has a good handle on the profit and loss side of business. To grow the business [she] will need to focus on marketing [her products better]" and "She has visited three supermarkets [so far]. They are telling her that they want her product delivered hot and have their own display." Other text includes "Are there products that are often wasted and not sold?," "Are there products that take a lot of time to make?," and "[I advised her to] introduce a new line of products."
In particular, volunteer marketers devoted, on average, 18% of their text to topic 4. Consulting volunteers devoted, on average, 10% of their text to topic 4, while other volunteers devoted, on average, 12% to topic 4. The differences are statistically significant, with marketers being greater than consultants (p =.006) and than other volunteers (p =.015).
Topic 4 captures text devoted to products, including their performance, which resonates with customers. Thus, consistent with our prediction, our STM results suggest that volunteer marketers aimed to help entrepreneurs differentiate through product-focused approaches. However, the STM results do not offer insights into how entrepreneurs supported by volunteer marketers changed their products. Nevertheless, these results indicate further analyses pertaining to emerging-market entrepreneurs' products are warranted.
It is also noteworthy that marketers did not devote more text to topic 2 (which captures text devoted to customers and the market) than the other volunteers. This finding suggests that customer and market-related topics—aside from product-related discussions captured by topic 4—were equally covered across the treatment groups. Thus, this offers further evidence that volunteer marketers' product focus was a key driver of their positive impact, again suggesting that additional analyses of emerging-market entrepreneurs' products are worthwhile.
According to [43], to effectively differentiate products, firms must provide some unique and meaningful value. [43] also argues firms that differentiate are frequently able to charge a premium price for their products, not just to compensate for potentially higher costs but also to achieve higher margins. Notably, differentiation has been found to reduce customers' price sensitivity and to enable the firm to earn a price premium (e.g., [48]). That said, emerging-market entrepreneurs may also try to differentiate their products by offering lower prices (e.g., [ 5]). Against this backdrop, we analyzed the marketers' effect on four proxies to assess whether and how entrepreneurs differentiate their products: ( 1) price per unit, ( 2) contribution per unit, ( 3) markup percentage per unit, and ( 4) enhancement of products. Web Appendix 14 provides details on measurement of the product differentiation proxies. The regression results in Table 3 demonstrate volunteer marketers' impact on emerging-market entrepreneurs' product differentiation efforts (for full details, see Web Appendix 15).
Graph
Table 3. Volunteer Marketers' Effects on (Intermediate) Product-Related Outcomes.
| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | (10) |
|---|
| Product Price | Product Unit Contribution | Product Markup | Product Enhancement | Premium Product Index | Firm Growth Index 2 |
|---|
| Treatment 1: offered marketer (yes = 1) | 46.944* | 23.356** | .153** | .103* | .254*** | | | | | |
| | (27.023) | (11.174) | (.069) | (.054) | (.086) | | | | | |
| Treatment 2: offered consultant (yes = 1) | 23.734 | 16.096 | .021 | −.041 | .065 | | | | | |
| | (28.624) | (13.676) | (.083) | (.056) | (.094) | | | | | |
| Treatment 3: offered other professional (yes = 1) | –21.473 | –7.971 | .025 | −.066 | −.087 | | | | | |
| | (15.763) | (6.869) | (.059) | (.045) | (.054) | | | | | |
| Product price: average per unit (UGX in 1000s) | | | | | | .00067*** | | | | |
| | | | | | | (.00017) | | | | |
| Product unit contribution: average per unit (UGX in 1000s) | | | | | | | .00144*** (.00039) | | | |
| Product markup: average per unit (%) | | | | | | | | −.05885 | | |
| | | | | | | | | (.04261) | | |
| Product enhancement: changed output and added value (yes = 1) | | | | | | | | | .09486* | |
| | | | | | | | | | (.05224) | |
| Premium Product Index (Average of Standardized Measures) | | | | | | | | | | .17089*** (.04494) |
| Baseline value of dependent variable included | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| 15 business controls included | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| 10 entrepreneur controls included | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| 10 industry fixed effects included | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| R-squared | .496 | .401 | .311 | .176 | .418 | .495 | .496 | .460 | .463 | .483 |
| Sample size: total | 605 | 605 | 605 | 605 | 605 | 605 | 605 | 605 | 605 | 605 |
| Control: mean of dependent variable | 80.604 | 31.058 | 4.154 | .309 | .002 | −.012 | −.012 | −.012 | −.012 | −.012 |
| Control: standard deviation of dependent variable | 201.888 | 85.062 | .716 | .463 | .682 | .717 | .717 | .717 | .717 | .717 |
| Premium product measure (normalized 0–1): coefficient | | | | | | 1.340 | 1.220 | −.312 | .095 | 1.406 |
- 120022242921993180 *p <.10.
- 130022242921993180 **p <.05.
- 140022242921993180 ***p <.01.
- 150022242921993180 Notes: Columns 1–5 show the effects on different measures of product outcomes. Columns 6–10 show the relationship between each product measure and overall firm growth outcomes (i.e., Firm Growth Index 2, the average of the eight standardized composites of sales, profits, assets, and employees). Robust standard errors are in parentheses. Firm growth values in levels (price and contribution per unit) are listed as Ugandan shillings (UGX) in thousands.
We find a 58.2% increase (β1 = 46.94) in average price per unit for firms in the marketer treatment group versus the control group. Moreover, we find that the average unit contribution increased by 75.2% (β1 = 23.36) for firms exposed to volunteer marketers relative to firms receiving no intervention. In addition, compared with control group firms, marketer treatment group firms improved markups by 15.3% on average, and 33.3% more of the firms (β1 =.103) enhanced their products. These results suggest that volunteer marketers indeed helped emerging-market entrepreneurs differentiate their products. The results also suggest that emerging-market entrepreneurs started offering more premium products—defined as products that demand "higher prices" and that "provide greater value to consumers" (e.g., [13])—after the marketing intervention. We also examined volunteer marketers' impact on changes in outcomes not related to product differentiation (e.g., firm operational or financial capabilities) but do not find significant effects, providing some evidence against alternative mechanism explanations.
To address noisy measurement issues, we also tested volunteer marketers' effect on a product index (referred to as "premium product index"), constructed by averaging the four standardized product differentiation proxies. As Table 3 shows, marketer group firms achieve a.254-standard-deviation increase for the overall premium product index compared with those in the control group, a roughly 37% increase. By contrast, we observe no significant change in the premium product index or the four product differentiation proxies for consultant and other professional group firms.
In terms of the substantive impact for entrepreneurs who were paired with a volunteer marketer, on average, their per-unit prices increased by 46,944 UGX ($12.84, or a 58.2% increase relative to control firms), and their unit contribution increased by 23,356 UGX ($6.39, or a 75.2% increase relative to control firms). These increases represent meaningful effect sizes for entrepreneurs selling in a marketplace where most customers are earning $5–$12 per day.
We next examined the relationship between the product differentiation proxies and firm growth. The general pattern of results suggests a positive and significant correlation between product differentiation and firm growth (see Table 3). We also tested whether product differentiation mediates volunteer marketers' effect on firm growth using [24] PROCESS Model 4. The indirect effect of the marketer treatment on our main firm growth index—through the premium product index—is positive and significant (i.e., a × b =.04; 95% confidence interval based on 10,000 bootstrap samples = [.01,.08]; see Web Appendix 16). Thus, entrepreneurs exposed to volunteer marketers not only created more premium products with higher prices, unit contributions, and markups but also were successful at selling these products, as indicated by their increased sales and profits. We repeated the mediation analysis for the consultant treatment and other professional treatment groups. Neither of the indirect effects was significant, indicating that product differentiation does not mediate the firm growth effects for these groups.
Taken together, the results support our predictions that volunteer marketers help emerging-market entrepreneurs improve product differentiation. Interestingly, the focus seems to be on selling more "premium" products, which is somewhat counterintuitive given the low disposable incomes of consumers in these markets. This analysis uncovers at least one (new) process through which the marketing intervention leads to firm growth.
Next, we analyzed interaction effects to determine which types of firms volunteer marketers help most. In particular, given the findings from the mechanism analysis thus far, the marketing intervention should be more effective for businesses better equipped for product differentiation. This raises the question, what makes a firm better equipped for a product differentiation–focused marketing intervention? [40] show that a firm's marketing capabilities and market orientation combine and interact and are akin to interconnected assets ([50]). Intelligence generation and dissemination are key components of a firm's market orientation ([27]). In turn, market knowledge is an important outcome of the two and helps firms understand customer preferences and competitor positions, which should enhance differentiated product development. Thus, we expect entrepreneurs with greater market knowledge to benefit more from the marketing intervention.
Moreover, the marketing intervention enables and is akin to benchmarking (e.g., [54]), a learning process by which the entrepreneur tries to identify best practices from the volunteer marketer. That said, the benchmarking literature has shown that firms with greater resources are better equipped to act on benchmarking insights (e.g., [ 1]). Indeed, greater resources (e.g., money, time) should assist firms in delivering products to market and improve their deployment of premium, differentiated products. Thus, we expect entrepreneurs with greater resources to benefit more from the marketing intervention.
We used three business characteristics to capture each firm's market knowledge (i.e., local market experience, demand tracking system, and diverse customers) and resource availability (i.e., start-up capital, business partners, and cash reserves). We provide details on measuring the characteristics in Web Appendixes 17 and 18. We created two composites for each construct (normalized 0–1 and median split, with 0 = lower and 1 = higher) and separately examined heterogeneity in the volunteer marketers' treatment effect.
Table 4 (columns 1–5) presents interaction regressions based on a firm's ex ante market knowledge using the composite measures and all three dimensions. We observe positive firm growth effects for entrepreneurs exposed to volunteer marketers when the businesses have greater market knowledge. In particular, the marketer interaction coefficient is large, with a 2.71-standard-deviation firm growth increase. Interpreted differently, a 33% market knowledge composite increase (i.e., obtaining the maximum score on one of three dimensions) leads to a.904-standard-deviation gain in overall firm growth.
Graph
Table 4. Heterogeneity in Volunteer Marketers' Interaction Effects on Firm Growth.
| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | (10) | (11) |
|---|
| Firm Growth Index 2 | Combined Model |
|---|
| MK Analysis | RA Analysis | MK Coef. | RA Coef. |
|---|
| (MK or RA) × (Marketer) | 2.713*** | .266* | 1.681** | .898** | 1.048** | 3.565** | .318** | 1.126* | 3.357** | 2.083** | 2.457*** | 3.261** |
| | (.976) | (.146) | (.701) | (.438) | (.530) | (1.687) | (.141) | (.655) | (1.590) | (.992) | (.926) | (1.619) |
| Treatment 1: offered marketer (yes = 1) | .206*** | .187*** | .198*** | .211*** | .184*** | .194*** | .197*** | .175** | .186*** | .206*** | .213*** |
| | (.070) | (.072) | (.069) | (.076) | (.071) | (.071) | (.072) | (.071) | (.068) | (.070) | (.069) |
| (MK or RA) × (Consultant) | 1.337** | .156 | −.149 | .588 | .699** | 2.000* | .031 | 1.289 | .911 | .728 | 1.061* | 1.478 |
| | (.628) | (.136) | (.378) | (.416) | (.344) | (1.106) | (.141) | (.898) | (1.768) | (.640) | (.617) | (1.034) |
| Treatment 2: offered consultant (yes = 1) | .053 | .061 | .049 | .053 | .062 | .081 | .075 | .064 | .066 | .063 | .061 |
| | (.069) | (.071) | (.071) | (.069) | (.070) | (.071) | (.072) | (.073) | (.071) | (.071) | (.069) |
| (MK or RA) × (Other professional) | 1.496** | .222** | −.340 | .503 | .864*** | 2.778*** | .055 | 1.541*** | .631 | 1.193 | 1.135** | 2.631*** |
| | (.655) | (.110) | (.703) | (.423) | (.271) | (1.067) | (.102) | (.506) | (1.051) | (.759) | (.569) | (.998) |
| Treatment 3: offered other professional (yes = 1) | .072 | .077 | .070 | .061 | .083 | .061 | .077 | .073 | .071 | .069 | .057 |
| | (.057) | (.058) | (.058) | (.055) | (.057) | (.053) | (.057) | (.055) | (.057) | (.055) | (.054) |
| MK: compositea | −.416 | | | | | | | | | | −.215 | |
| | (.460) | | | | | | | | | | (.428) | |
| MK: high (yes = 1; median split) | | −.036 | | | | | | | | | | |
| | | (.074) | | | | | | | | | | |
| MK dimension: local market experiencea | | | .459 | | | | | | | | | |
| | | | (.288) | | | | | | | | | |
| MK dimension: demand tracking systema | | | | −.075 | | | | | | | | |
| | | | | (.232) | | | | | | | | |
| MK dimension: diverse customersa | | | | | −.574** | | | | | | | |
| | | | | | (.252) | | | | | | | |
| RA: compositea | | | | | | −.804 | | | | | | −.842 |
| | | | | | | (.836) | | | | | | (.794) |
| RA: high (yes = 1; median split) | | | | | | | .039 | | | | | |
| | | | | | | | (.068) | | | | | |
| Dimension: start-up capitala | | | | | | | | −.524 | | | | |
| | | | | | | | | (.350) | | | | |
| Dimension: business partnersa | | | | | | | | | −.645 | | | |
| | | | | | | | | | (1.063) | | | |
| Dimension: cash reservesa | | | | | | | | | | −.185 | | |
| | | | | | | | | | | (.575) | | |
| R-squared | .493 | .474 | .491 | .475 | .483 | .488 | .470 | .485 | .477 | .482 | .515 |
| Sample size: total | 605 | 605 | 599 | 604 | 605 | 605 | 605 | 589 | 605 | 605 | 605 |
| Control: mean of dependent variable | −.012 | −.012 | −.012 | −.008 | −.012 | −.012 | −.012 | −.014 | −.012 | −.012 | −.012 |
| Control: standard deviation of dependent variable | .717 | .717 | .719 | .715 | .717 | .717 | .717 | .722 | .717 | .717 | .717 |
- 160022242921993180 *p <.10.
- 170022242921993180 **p <.05.
- 180022242921993180 ***p <.01.
- 190022242921993180 a Normalized 0–1; mean-centered.
- 200022242921993180 Notes: MK = market knowledge; RA = resource availability. 15 business controls, 10 entrepreneur controls, and 10 industry fixed effects are included in all regressions. To avoid duplication, the "start-up capital" control is dropped from the resource availability regressions in columns 6–8. Robust standard errors are in parentheses. Firm Growth Index 2 is the average of the eight standardized composites of sales, profits, assets, and employees.
Likewise, marketers' impact on firm growth is greater for entrepreneurs with more resource availability. As shown in Table 4 (columns 6–10), firms matched with volunteer marketers realize a 3.57-standard-deviation gain when their resource availability is highest (i.e., 1 on the normalized composite). The positive firm growth effects persist whether the composite measure is normalized or split at the median, as well as for each of its three dimensions. We note that when all interaction terms are included in the same model (column 11 in Table 4), the results are substantively similar.[17]
We also explored nonlinearities in the relationship between market knowledge and firm growth to delve deeper into heterogeneity. Web Appendix 19 summarizes the regression results when we include the continuous market knowledge measure (normalized 0–1 and mean-centered) and its squared term interacted with our treatment dummy variables. The positive impact on firm growth persists when businesses increase in market knowledge and are matched with a volunteer marketer. Moreover, we find a positive and significant squared term (7.03), which suggests that the relationship is nonlinear. We plot the predicted values from the regression in Figure 3 to highlight differences between the marketing treatment and control groups. For marketing treatment firms, we observe a convex relationship as market knowledge increases from the left tail (−.159) to the right tail (+.292) of its distribution. The plot shows that most of the interaction effect occurs toward the right tail, where market knowledge is highest and separation from the control group distribution is greatest.
Graph: Figure 3. Market knowledge and nonlinear firm growth effects.Notes: The predicted values of firm growth (p-hat) are obtained following a nonlinear interaction analysis that regresses Firm Growth Index 2 onto the continuous measures of market knowledge and its squared term as well as the interactions of both variables with each of the treatment dummies (and the full set of controls). For complete results, see Web Appendix 19. For display purposes, 2.5% of the distribution's right tail is truncated in the figure. Error bars = ±1 SE.
To better understand the pattern, we also divided the market knowledge composite into terciles and see a similar nonlinear relationship (see Web Appendix 19). Thus, these results suggest that only businesses with high market knowledge appear to see a large and increasing positive effect on firm growth when exposed to a marketer.
We also explored treatment heterogeneity and nonlinear firm growth effects for resource availability. Web Appendix 20 summarizes the regression results when we include the continuous resource availability measure (normalized 0–1 and mean centered) and its squared term interacted with our treatment dummy variables. The positive firm growth effect persists when businesses increase in resource availability and are matched with a marketer. However, the negative and significant squared term (−14.19) suggests that the relationship is again nonlinear. We plot the predicted values from the regression in Figure 4 to highlight the differences between the marketing treatment and control groups. For marketing treatment firms, we observe a concave relationship as resource availability increases from the left tail (−.057) to the right tail (+.291) of its distribution. The plot shows that the interaction effect occurs mainly toward the mid- to right tail as resource availability increases.
Graph: Figure 4. Resource availability and nonlinear firm growth effects.Notes: The predicted values of firm growth (p-hat) are obtained following a nonlinear interaction analysis that regresses Firm Growth Index 2 onto the continuous measures of resource availability and its squared term as well as the interactions of both variables with each of the treatment dummies (and the full set of controls). For complete results, see Web Appendix 20. For display purposes, 2.5% of the distribution's right tail is truncated in the figure. Error bars = ±1 SE.
To further examine the nonlinear relationship, we also divided the resource availability composite into terciles and again obtain similar results (see Web Appendix 20). These findings indicate that only businesses with high resource availability appear to see a large and slightly decreasing positive effect on growth when exposed to a marketer.
Interest in the effects of business support interventions on firm and economic growth in emerging markets has risen over the past decade. Researchers have suggested that entrepreneurship, in particular, can be a catalyst for growth ([14]; [20]). However, scholars have also pointed out a need for research determining which business skills are impactful, and for whom, and for work examining the process through which interventions enhance firm performance (e.g., [38]).
Our results, based on a randomized controlled field experiment with 930 entrepreneurs in Uganda, indicate that volunteer marketers significantly and positively impact the entrepreneurs' firm growth by 32.5% on average, as measured in monthly sales and profits, total assets, and paid employees.
Our theory and mechanism analyses indicate that volunteer marketers are effective because they help the entrepreneurs differentiate, a capability many desperately lack ([ 6]). Process evidence suggests that entrepreneurs matched with volunteer marketers create more premium products that resonate with target customers. Finally, our evidence based on interaction effects provides insight into which types of businesses benefit most from a volunteer marketer—namely, those with greater ex ante market knowledge or resources.
Governmental organizations and NGOs invest billions in business support interventions to fight poverty in emerging markets each year ([14]). Researchers debate whether the aid is beneficial (e.g., [18]; [46]; [49]). Our study focuses on a basic, concrete question: Can marketers help small-scale entrepreneurs in Uganda grow their businesses? If yes, marketers could partially alleviate Uganda's pervasive poverty (e.g., [31]). As [20], p. 196) point out, "Increasing the...quality of entrepreneurs is probably one of the most helpful ways to reduce poverty because it creates employment and boosts the innovation and economic empowerment of individuals in poor countries with extremely high unemployment rates."
Many emerging-market entrepreneurs struggle and fail to grow because they are "utterly undifferentiated" ([ 6]). We find that marketers can be especially effective as volunteers because they help entrepreneurs differentiate.
We therefore offer governmental organizations and NGOs an accessible recommendation for future business support interventions in emerging markets. We hope our findings will earn marketers a seat at the policy table with organizations such as the World Bank, International Monetary Fund, and United Nations, which invest heavily in business and entrepreneurship programs every year. Our results suggest that the organizations should consider how marketers and marketing tools can be integrated into solutions for stimulating firm growth.
Many economists believe that emerging-market entrepreneurs often fail to thrive due to resource constraints (e.g., [58]). While our results confirm that resources help entrepreneurs succeed, we find that resources alone may not be enough. Emerging-market entrepreneurs may also need guidance from experienced business professionals, particularly marketers, to use their available resources.
Our partner, Grow Movement, estimates that each of its entrepreneur–volunteer collaborations costs $450–$500 when run at a large scale in a single country, where fixed costs can be spread across units. These costs compare favorably to other business support interventions in emerging markets (e.g., [14]; [38]), suggesting that governmental organizations and NGOs would be willing to support the costs. In fact, several business schools and NGOs have recently started incorporating versions of our "remote coaching" intervention into their programs with a focus on matching entrepreneurs with marketing practitioners. In addition, multinationals in developed markets could participate in future remote marketing coaching interventions such as ours. In short, we envision multinationals enabling their interested marketers to spend a few hours a week remotely coaching an emerging-market entrepreneur. This endeavor, we believe, could be a win-win for the entrepreneurs and the multinationals: the entrepreneurs' businesses would likely grow, and the multinationals would likely have more satisfied employees, accrue corporate social responsibility–related benefits, and learn about opportunities (and threats) in emerging markets.
The marketing literature has largely neglected entrepreneurial firms, which is surprising given the important role such companies play across all markets (e.g., [36]; [56]). Likewise, the entrepreneurship literature has largely ignored marketing, which is equally surprising, as some have argued that "marketing is the home for the entrepreneurial process" ([41], p. 247). Although marketing and entrepreneurship are two key business responsibilities ([17]), researchers have done little to understand how the two interact ([56]). Our study offers evidence that marketing and entrepreneurship blend especially well in emerging markets. The insight adds to the literature on marketing's influence within the firm (e.g., [25]; [53]), suggesting that emerging-market entrepreneurs benefit from marketing knowledge and skills.
We hope that entrepreneurs in emerging markets take note of our findings and consider either acquiring marketing skills or hiring marketers. Marketers could consider partnering with entrepreneurial firms as volunteers or paid employees. Finally, we hope that emerging-market entrepreneurs and marketers note our finding that premium products can be successful in emerging markets. Thus, we add to the emerging literature on low-income consumers' preferences in emerging markets (e.g., [ 5]; [34]).
Our study is not without limitations, some of which provide opportunities for future research. Although our study was conducted over two years, longer than many prior business-support-intervention studies, its long-term implications are not obvious. For example, we cannot say with certainty that the treated entrepreneurs will continue using the marketing capabilities they acquired during the intervention. Although we show that the entrepreneurs significantly changed their products, which bodes well for long-term effects ([38]), future intervention studies might measure outcomes over longer periods.
We randomly assigned volunteers to entrepreneurs as part of our experimental setup. Thus, we did not match volunteers and entrepreneurs on the basis of their backgrounds. However, more technical businesses, for example, might benefit from a volunteer with an engineering background. Entrepreneurs and volunteers might also match well on the basis of demographics such as gender or age. Future research should explore matching-related questions.
Finally, some economists (e.g., [18]) and organizations (e.g., the American Enterprise Institute) are skeptical of or oppose foreign aid. Some suggest that foreign aid is often focused on recipients' material well-being without addressing underlying issues such as corrupt governments and individual rights suppression. These concerns are serious and valid; however, evidence suggests that flourishing entrepreneurship translates to positive long-term net effects in developing economies (e.g., [20]). We hope future research continues to explore ways in which marketers can play a role in "doing good" in the economies and societies of emerging markets, thereby contributing to a better world.
Supplemental Material, sj-docx-1-jmx-10.1177_0022242921993176 - Do Marketers Matter for Entrepreneurs? Evidence from a Field Experiment in Uganda
Supplemental Material, sj-docx-1-jmx-10.1177_0022242921993176 for Do Marketers Matter for Entrepreneurs? Evidence from a Field Experiment in Uganda by Stephen J. Anderson, Pradeep Chintagunta, Frank Germann and Naufel Vilcassim in Journal of Marketing
Footnotes 1 Sundar Bharadwaj
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by grants from: the UK Department for International Development (DFID) and Economic and Social Research Council's (ESRC) joint Growth Research Program, Deloitte Institute for Innovation and Entrepreneurship (DIIE), the John A. and Cynthia Fry Gunn Faculty Scholar award (Stanford), London Business School (LBS) and the LBS Leadership Institute (LI), Private Enterprise Development in Low-Income Countries (PEDL), the Chicago Booth School of Business Social Enterprise Initiative (SEI), the Chicago Booth Initiative on Global Markets (IGM), the Chicago Booth Kilts Center for Marketing, the London School of Economics Marshall Institute, the Stanford Graduate School of Business, the Mendoza College of Business at Notre Dame, the Stanford Institute for Innovation in Developing Economies (SEED), and the Stanford Center on Global Poverty and Development (SCGPD).
4 Stephen J. Anderson https://orcid.org/0000-0003-1945-1963
5 Online supplement https://doi.org/10.1177/0022242921993176
6 We included the volunteer professionals from other functional backgrounds in our analysis primarily to understand the theoretical mechanism allowing marketers to help entrepreneurs grow.
7 Hereinafter, we use "products" in reference to tangible, physical goods (e.g., donuts, shampoo), intangible, nonphysical services (e.g., breakfast delivery, hair cutting), and combined offerings ([30], p. 266).
8 As noted previously, our data gathering was the same as in the [3] study. Although our research questions, study designs, and empirical analyses differ, we repeat some of the sampling and measurement descriptions here for transparency and completeness.
9 The screening step was in line with our partner's program requirement to work with operational firms committed to and ready to receive a business support service. Screening or targeting approaches have become common in government and NGO programs aiming to allocate scarce resources for stimulating firm or economic growth (e.g., [2]). The screening step influences the population to which our results generalize but not our causal effects.
Take-up rates did not differ between the marketers group and the other professionals group (p =.137) or the consultants group (p =.847). Compliance levels did not differ from marketers to other professionals (p =.246), but consultants completed more modules than marketers (p =.099).
Team leaders, a field manager, and a research manager supervised our field team and reviewed data daily. Outliers, anomalies, and data entry errors were immediately clarified with the enumerator or entrepreneur. Additional auditors, blind to the research design and firms, cross-checked a random set of 10% of the surveys with the entrepreneurs daily. The field manager and/or team leaders conducted on-site business audits to verify flagged responses. After all data from a survey round had been collected, the research manager in Uganda verified the accuracy of all outliers and anomalies, with particular attention paid to sales, profit, asset, and employee estimates, by visiting the entrepreneur and conducting an additional audit of financial information and cross-checking flagged variables. The same steps were taken for each completed survey.
In our interaction analysis, Equation 1 includes the pretreatment theoretical variables of interest (i.e., market knowledge and resource availability) and interaction terms, one for each interaction between the treatment dummy and theoretical variables.
The analysis of covariance specification can increase statistical power compared with a difference-in-differences model when measures are noisy and low autocorrelation exists between the baseline and end-line dependent variable values, a common condition for small firm outcomes such as sales and profits in emerging markets ([37]).
We use a currency conversion rate of US$1 = 3,656 UGX (as per http://www.xe.com on October 31, 2017; the midpoint of our end-line surveying period).
We perform the same randomization checks with the full sample at end line in Web Appendix 8. The F-test is not significant for any of the three group comparisons. We find only eight statistically significant differences across the 120 t-tests.
Considering the firm growth indices, we cannot reject the null hypothesis of equal coefficients between firms in the marketer and consultant treatment groups (p =.161 and p =.192, respectively) or between the marketer and other professional treatment groups (p =.139 and p =.169, respectively). However, our goal is to examine volunteer marketers' effects on business growth (instead of the differences among treatment groups), and we therefore focus on the marketer effects in our discussion.
An additional analysis (not reported) indicates that high market knowledge and high resource availability combined result in a growth effect of.431 standard deviations (p =.006) for firms exposed to a volunteer marketer, which is greater than knowledge alone (.266 SD) or resources alone (.318 SD). This effect suggests a synergistic relationship between knowledge, which can help develop differentiated products, and resources, which can help deploy products in the market.
References Anand Gurumurthy, Kodali Rambabu. (2008), "Benchmarking the Benchmarking Models," Benchmarking: An International Journal, 15 (3), 257–91.
Anderson Stephen J., Chandy Rajesh, Zia Bilal. (2018), "Pathways to Profits: The Impact of Marketing vs. Finance Skills on Business Performance," Management Science, 64 (12), 5559–83.
Anderson Stephen J., Chintagunta Pradeep, Vilcassim Naufel. (2021), "Connections Across Markets: Stimulating Business Model Innovation and Examining the Impact on Firm Sales in Uganda," working paper.
Anderson Stephen J., Lazicky Christy, Zia Bilal. (2021), "Measuring the Unmeasured: Aggregating, Anchoring, and Adjusting to Estimate Small Business Performance," working paper.
Arunachalam S., Bahadir S. Cem, Bharadwaj Sundar G., Guesalaga Rodrigo. (2020), "New Product Introductions for Low-Income Consumers in Emerging Markets," Journal of the Academy of Marketing Science, 48 (5), 914–40.
Banerjee Abhijit, Duflo Esther. (2011), Poor Economics: A Radical Rethinking of the Way to Fight Global Poverty. New York: Public Affairs.
Berger Jonah, Humphreys Ashlee, Ludwig Stephan, Moe Wendy W., Netzer Oded, Schweidel David A. (2020), "Uniting the Tribes: Using Text for Marketing Insight," Journal of Marketing, 84 (1), 1–25.
Bloom Nicholas, Eifert Benn, Mahajan Aprajit, McKenzie David, Roberts John. (2013), "Does Management Matter? Evidence from India," Quarterly Journal of Economics, 128 (1), 1–51.
Boulding William, Staelin Richard. (1995), "Identifying Generalizable Effects of Strategic Actions on Firm Performance: The Case of Demand-Side Returns to R&D Spending," Marketing Science, 14 (3), 222–36.
Boulding William, Lee Eunkyu, Staelin Richard. (1994), "Mastering the Mix: Do Advertising, Promotion, and Sales Force Activities Lead to Differentiation?" Journal of Marketing Research, 31 (2), 159–72.
Bruton Garry D., Ketchen David J.Jr, Ireland R. Duane. (2013), "Entrepreneurship as a Solution to Poverty," Journal of Business Venturing, 28 (6), 683–89.
Bygrave William D., Hofer Charles W. (1992), "Theorizing About Entrepreneurship," Entrepreneurship Theory and Practice, 16 (2), 13–22.
Caldieraro Fabio, Kao Ling-Jing, Cunha MarcusJr. (2015), "Harmful Upward Line Extensions: Can the Launch of Premium Products Result in Competitive Disadvantages?" Journal of Marketing79 (6), 50–70.
Campos Francisco, Frese Michael, Goldstein Markus, Iacovone Leonardo, Johnson Hillary C., McKenzie David, Mensmann Mona. (2017), "Teaching Personal Initiative Beats Traditional Training in Boosting Small Business in West Africa," Science, 357 (6357), 1287–90.
Christensen Clayton M. (1997), The Innovator's Dilemma. Cambridge, MA: Harvard Business School Press.
Dickson Peter R., Ginter James L. (1987), "Market Segmentation, Product Differentiation, and Marketing Strategy," Journal of Marketing, 51 (2), 1–10.
Drucker Peter. (1954), The Practice of Management. New York: Harper & Row.
Easterly William. (2014), The Tyranny of Experts: Economists, Dictators, and the Forgotten Rights of the Poor. New York: Basic Books.
Fleit Caren, Morel-Curran Brigitte. (2012), The Transformative CMO. Los Angeles: The Korn/Ferry Institute.
Frese Michael, Gielnik Michael M., Mensmann Mona. (2016), "Psychological Training for Entrepreneurs to Take Action: Contributing to Poverty Reduction in Developing Countries," Current Directions in Psychological Science, 25 (3), 196–202.
Friedrichs David O. (1987), "Bringing Ourselves Back In: The Reflexive Dimension in Teaching a Humanist Sociology," Teaching Sociology, 15 (1), 1–6.
Gollin Douglas. (2002), "Getting Income Shares Right," Journal of Political Economy, 110 (2), 458–74.
Hambrick Donald C., Mason Phyllis A. (1984), "Upper Echelons: The Organization as a Reflection of Its Top Managers," Academy of Management Review, 9 (2), 193–206.
Hayes Andrew F. (2018), Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Perspective, 2nd ed. New York: Guilford Press.
Homburg Christian, Workman John P.Jr, Krohmer Harley. (1999), "Marketing's Influence Within the Firm," Journal of Marketing, 63 (2), 1–17.
Hsieh Chang-Tai, Klenow Peter J. (2014), "The Life Cycle of Plants in India and Mexico," Quarterly Journal of Economics, 123 (3), 1035–84.
Jaworski Bernard J., Kohli Ajay K. (1993), "Market Orientation: Antecedents and Consequences," Journal of Marketing, 57 (3), 53–70.
Kaplan Abraham. (2017), The Conduct of Inquiry: Methodology for Behavioural Science. Abingdon, UK: Routledge.
Kelley Donna, Bosma Niels, Amorós José E. (2011), "Global Entrepreneurship Monitor 2010 Executive Report," Global Entrepreneurship Monitor, https://www.gemconsortium.org/report/gem-2010-global-report.
Kerin A. Roger, Hartley Steven W. (2017), Marketing, 13th ed. New York: McGraw-Hill.
Kiranda Yusuf, Walter Max, Mugisha Michael. (2017), Reality Check: Employment, Entrepreneurship, and Education in Uganda. Kampala: Konrad-Adenauer Stiftung.
Kohli Ajay K., Jaworski Bernard J. (1990), "Market Orientation: The Construct, Research Propositions, and Managerial Implications," Journal of Marketing, 54 (2), 1–18.
Kotler Philip, Keller Kevin Lane. (2016), Marketing Management, 15th ed. London: Pearson.
Mahajan Vijay. (2016), Rise of Rural Consumers in Developing Countries: Harvesting 3 Billion Aspirations. New York: SAGE Publications India.
March James G., Simon Herbert A. (1958), Organizations. New York: John Wiley & Sons.
Matsuno Ken, Mentzer John T., Özsomer Ayşegül. (2002), "The Effects of Entrepreneurial Proclivity and Market Orientation on Business Performance," Journal of Marketing, 66 (3), 18–32.
McKenzie David. (2012), "Beyond Baseline and Follow-Up: The Case for More T in Experiments." Journal of Development Economics, 99 (2), 210–21.
McKenzie David, Woodruff Christopher. (2014), "What Are We Learning from Business Training and Entrepreneurship Evaluations Around the Developing World?" World Bank Research Observer, 29 (1), 48–82.
Merlo Omar, Auh Seigyoung. (2009), "The Effects of Entrepreneurial Orientation, Market Orientation, and Marketing Subunit Influence on Firm Performance," Marketing Letters, 20 (3), 295–311.
Morgan Neil A., Vorhies Douglas W., Mason Charlotte H. (2009), "Market Orientation, Marketing Capabilities, and Firm Performance," Strategic Management Journal, 30 (8), 909–20.
Morris Michael H., Paul Gordon W. (1987), "The Relationship Between Entrepreneurship and Marketing in Established Firms," Journal of Business Venturing, 2 (3), 247–59.
Narasimhan Laxman, Srinivasan Kannan, Sudhir K. (2015), "Marketing Science in Emerging Markets," Marketing Science, 34 (4), 473–79.
Porter Michael E. (1980), Competitive Strategy: Techniques for Analyzing Industries and Competitors. New York: The Free Press.
Ries Eric. (2011), The Lean Startup: How Today's Entrepreneurs Use Continuous Innovation to Create Radically Successful Businesses. New York: Crown Business.
Roberts Margaret E., Stewart Brandon M., Tingley Dustin. (2017), "STM: R Package for Structural Topic Models," Journal of Statistical Software, 91 (2), 1–40.
Sachs Jeffrey. (2005), The End of Poverty: Economic Possibilities for Our Time. New York: Penguin Press.
Schumpeter Joseph A. (1934), The Theory of Economic Development. Cambridge, MA: Harvard University Press.
Sharp Byron, Dawes John. (2001), "What Is Differentiation and How Does It Work?" Journal of Marketing Management, 17 (7–8), 739–59.
Singer Peter. (2009), The Life You Can Save. New York: Random House.
Teece David J., Pisano Gary, Shuen Amy. (1997), "Dynamic Capabilities and Strategic Management," Strategic Management Journal, 18 (7), 509–33.
The CMO Survey (2019), "Highlights and Insight Report—August," (accessed November 19, 2019), https://cmosurvey.org/results/august-2019/.
Tirunillai Seshadri, Tellis Gerard J. (2014), "Mining Marketing Meaning from Online Chatter: Strategic Brand Analysis of Big Data Using Latent Dirichlet Allocation," Journal of Marketing Research, 51 (4), 463–79.
Verhoef Peter C., Leeflang Peter S.H. (2009), "Understanding the Marketing Department's Influence Within the Firm," Journal of Marketing, 73 (2), 14–37.
Vorhies Douglas W., Morgan Neil A. (2005), "Benchmarking Marketing Capabilities for Sustainable Competitive Advantage," Journal of Marketing, 69 (1), 80–94.
Waller Mary J., Huber George P., Glick William H. (1995), "Functional Background as a Determinant of Executives' Selective Perception," Academy of Management Journal, 38 (4), 943–74.
Webb Justin W., Ireland R. Duane, Hitt Michael A., Kistruck Geoffrey M., Tihanyi Laszlo. (2011), "Where Is the Opportunity Without the Customer? An Integration of Marketing Activities, the Entrepreneurship Process, and Institutional Theory," Journal of the Academy of Marketing Science, 39 (4), 537–54.
Whitler Kimberly A., Krause Ryan, Lehmann Donald R. (2018), "When and How Board Members with Marketing Experience Facilitate Firm Growth," Journal of Marketing, 82 (5), 86–105.
Yunus Muhammad. (2007), Banker to the Poor: Micro-Lending and the Battle Against World Poverty. New York: Public Affairs.
Zhou Kevin Zheng, Yim Chi Kin, Tse David K. (2005), "The Effects of Strategic Orientations on Technology and Market-based Breakthrough Innovations," Journal of Marketing, 69 (2), 42–60.
~~~~~~~~
By Stephen J. Anderson; Pradeep Chintagunta; Frank Germann and Naufel Vilcassim
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 45- Do Nudges Reduce Disparities? Choice Architecture Compensates for Low Consumer Knowledge. By: Mrkva, Kellen; Posner, Nathaniel A.; Reeck, Crystal; Johnson, Eric J. Journal of Marketing. Jul2021, Vol. 85 Issue 4, p67-84. 18p. 1 Diagram, 1 Chart, 2 Graphs. DOI: 10.1177/0022242921993186.
- Database:
- Business Source Complete
Do Nudges Reduce Disparities? Choice Architecture Compensates for Low Consumer Knowledge
Choice architecture tools, commonly known as nudges, powerfully impact decisions and can improve welfare. Yet it is unclear who is most impacted by nudges. If nudge effects are moderated by socioeconomic status (SES), these differential effects could increase or decrease disparities across consumers. Using field data and several preregistered studies, the authors demonstrate that consumers with lower SES, domain knowledge, and numerical ability are impacted more by a wide variety of nudges. As a result, "good nudges" designed to increase selection of superior options reduced choice disparities, improving choices more among consumers with lower SES, lower financial literacy, and lower numeracy than among those with higher levels of these variables. Compared with "good nudges," "bad nudges" designed to facilitate selection of inferior options exacerbated choice disparities. These results generalized across real retirement decisions, different nudges, and different decision domains. Across studies, the authors tested different explanations of why SES, domain knowledge, and numeracy moderate nudges. The results suggest that nudges are a useful tool for those who wish to reduce disparities. The research concludes with a discussion of implications for marketing firms and segmentation.
Keywords: defaults; financial literacy; nudges; numeracy; socioeconomic status
Choice architecture can powerfully impact decisions and improve welfare. Firms have adopted choice architecture changes that have increased retirement savings, increased environmentally friendly purchases, increased the number of premium features consumers buy when purchasing an automobile, and influenced other types of consumer decisions ([17]; [27]; [40]; [75]).
But who does choice architecture influence most? Do changes to the choice environment impact some consumers more than others? We examined two related sources of heterogeneity in nudge effects, testing whether domain-specific skills and knowledge moderate nudge effects and whether socioeconomic status (SES) moderates nudge effects. We hypothesized that choice architecture can reduce choice disparities by having the largest impact on consumers with low SES and low levels of domain knowledge and skill.[ 6] Though choice architecture is inherent to online retail, many firms might not consider how choice architecture tools impact different consumers to different degrees, potentially reducing or exacerbating inequities between them. Knowledge of factors that make consumers more susceptible to choice architecture effects can allow firms and policy makers to use choice architecture more effectively to achieve the impact they want ([74]).
Choice architecture describes how a change in the structure of a choice influences behavior without significantly altering economic incentives or what consumers know about each option ([40]; [75]). Choice architecture can be manipulated to change the decisions that consumers make; these manipulations are often called "nudges" ([45]; [75]).
Nudges are inexpensive and cost effective for firms and governments ([ 7]). Perhaps for this reason, they have gained tremendous popularity among policy makers and marketers ([ 2]; [ 7]). Over 200 "nudge units" currently exist around the world across private and public sectors ([ 2]). Marketing research has examined how choice architecture tools such as defaults and sorting alter consumer behavior (e.g., [22]; [27]; [40]). All marketing managers must make decisions about choice architecture. For example, retailers choose which products to display first and whether to use defaults to automatically select a shipping option, insurance, or product add-on ([72]; [75]). These decisions impact purchases and consumers' subsequent wealth, health, and well-being.
Choice architecture is often used to facilitate choices that benefit consumers, firms, or both. For example, one auto manufacturer benefited both consumers and itself by changing the default car specifications on their website. Though it had previously defaulted consumers into basic, stripped-down car models, it found that changing defaults to tailor them to different types of customers increased firm profits while also benefiting consumers ([27]). Though nudges are frequently designed to help consumers, they can sometimes increase firm profits while decreasing consumer welfare. Nudges that harm consumers have been referred to as "bad nudges," "dark patterns," or "evil nudges" ([30]; [50]; [73]), in contrast to "good nudges" that benefit consumers. We examine whether "bad nudges" exacerbate choice disparities relative to "good nudges" by impacting low-SES and low-knowledge consumers most.
There are many types of nudges, including defaults, sorting, partitioning, and several nudges that reduce the complexity or number of attributes or options ([15]; [16]; [20]; [22]; [39]; [40]; [49]; [70]). We examine three types of choice architecture: defaults, sorting, and changes to the number of options. Defaults, a type of nudge that preselects one option but allows consumers to easily opt out of the preselected option, have been called "unquestionably the most prominent...[nudge], across all domains of application" ([45], p. 27). Another nudge, called sorting, organizes options in a systematic way. For example, many products can be sorted by price, consumer rating, total cost, sales volume, or other attributes ([22]; [42]; [49]). Another form of choice architecture, which reduces the number of options presented to consumers, can improve decision making, reduce regret, and decrease the likelihood that consumers will defer their decision by choosing nothing ([10]; [16]; [40]).
Previous research on nudges has typically focused on the overall effect of a nudge averaged across all individuals. For example, preselecting cars with premium features as the default increased the automobile purchase price by $1,500 on average ([27]) and opting people into retirement contributions resulted in large overall effects on enrollment ([17]). Other investigations have focused on the average cost effectiveness of nudges ([ 7]) or the impact of other nudges (e.g., sorting or changing the number of options) on the average consumer (e.g., [49]; [64]). Though these nudges have large impacts on average, it is unclear who they benefit most or whether they reduce or exacerbate inequities across consumers.
Yet it is important to consider the heterogeneous impact of nudges rather than only the average effect collapsing across all consumers. Some scholars have suggested that nudges may affect the rich more than the poor ([61]). [61] argues that because structural factors impede the autonomy of vulnerable low-SES consumers, high-SES consumers will change their behavior when nudged, whereas low-SES consumers will be "nudge-proof" due to their lack of autonomy. A different prominent account suggests that scarcity and low income influence decision making by increasing time and attention on a focal task at the expense of tasks and decisions that are secondary or require thinking about the future ([69]). This might reduce the effect of nudges if heightened time and attention on focal decisions increase motivation and accuracy. In contrast, we predicted that nudges would impact consumers with low SES and less domain-specific knowledge and skills more than those with higher levels of these characteristics for other reasons (detailed subsequently and in Figure 1). Thus, we hypothesized that interventions encouraging the selection of the best option should reduce choice disparities between consumers who differ in SES, domain knowledge, and numeracy. We tested these predictions across a wide variety of contexts and nudges.
Graph: Figure 1. Diagram of our theoretical framework explaining who is more susceptible to choice architecture and why.Notes: Consumers lower in SES and choice-relevant skills (e.g., numeracy) are impacted more by nudges. The model suggests that the SES moderator is explained by choice-relevant skills and knowledge, which moderates nudge effects partly because of anxiety, preference construction, and decision uncertainty. The relationships depicted by the dark gray arrows were the key relationships in our conceptual framework that we examined in primary analyses, and the light gray arrows were also supported by our data.
We focused on the moderators of SES, numeracy, and domain knowledge for several reasons. We focused on SES partly because it is easy to measure and use for segmentation ([11]; firms often have this information about their customers) and partly because SES strongly influences consumer behavior ([13]; [23]; [34]). We also focused on SES because previous research on choice architecture has largely neglected how effects of nudges differ across different levels of SES, and because reducing SES inequities is a major goal for many policy makers and firms. Firms and policy makers serve individuals with varying levels of SES; our investigation can help them estimate which consumers their nudges will impact most. Furthermore, SES has robust positive associations with numeracy, domain knowledge, and anxiety ([ 3]; [48]; [71]), which, in our view, shape susceptibility to nudges.
We examined numeracy and domain knowledge as focal moderators because these constructs play major roles in consumer decision making ([28]; [53]) and are useful for theory building. As we explain in the following section, these variables, along with anxiety, decision uncertainty, and preference construction, determine the extent to which choice architecture influences decisions according to our account.
Understanding heterogeneous effects of nudges could help firms by allowing them to target specific consumer segments, which could make nudges more effective. Furthermore, scholars have suggested that understanding heterogeneity would provide insight into why nudges often have smaller effects when applied at scale ([74]).
Socioeconomic disparities pervade consumer behavior. SES influences what products and brands consumers buy, how they access credit, and how they are treated in some stores, among other impacts ([13]; [23]; [34]). Consumers with lower SES and education (as well as the elderly) are often more vulnerable to marketing scams and manipulations ([34]; [43]). In addition, there are wide gaps between high-SES and low-SES consumers in terms of how much money they have in stocks, retirement savings, credit card debt, payday loan debt, and other assets or liabilities, which can greatly influence present behavior and future wealth ([ 8]; [23]).
Lower SES is associated with lower levels of numeracy and financial literacy ([ 3]; [48]; [71]). These skills play a role in nearly every type of consumer decision (e.g., [28]; [71]), and the discrepancy in these skills between low-SES and high-SES consumers can lead to disparate decisions and outcomes. The experience of scarcity that accompanies low SES sometimes narrows attention on a focal decision and influences time allocation, which can impact decisions ([69]).
Numeracy is the ability to process and use basic numerical concepts; make quantitative estimations; and use probabilities, percentages, and ratios ([58]). In the context of consumer decision making, innumerate people cannot calculate unit prices, use percentages to calculate discounts, compute interest, or even estimate a tip accurately ([28]; [53]; [63]). Broadly, numerate individuals often make better consumer and health decisions, especially when these decisions involve numbers, calculations, prices, or financial information ([59]). Numeracy refers to the ability to use and process numbers, which is distinct from other traits such as self-efficacy, math emotions (e.g., math anxiety), uncertainty, and subjective numeracy ([57]; [58]; [71]).
Financial literacy is the knowledge of basic financial concepts, operations, and facts. It is an important skill used to make financial decisions as well as decisions involving product prices and attributes ([19]; [32]). Financially literate consumers are less likely to overspend and are more likely to save for retirement, invest in stocks, comparison shop, and pay off their full credit card balance ([19]; [32]; [47]). Though financial literacy is associated with a wide range of consumer behaviors, financial literacy training only weakly influences financial knowledge, and any effects dissipate quickly ([25]).
Numeracy and financial literacy impact consumer behavior partly because consumers with lower numeracy and financial literacy experience greater anxiety and decision uncertainty when dealing with numbers and financial decisions ([58]; [71]). In addition, rather than retrieve stable preferences from memory, uncertain consumers construct their preferences on the fly more often than do consumers with higher certainty ([35]). As a result, their preferences are more labile; they are more reliant on effort-reducing heuristics; and they are more impacted by defaults, the status quo, and changes in the number of options ([16]; [36]; [37]; [68]). In other words, consumers with lower numeracy and financial literacy feel more uncertainty and anxiety; thus, we expected that in the context of nudges they would rely on strategies such as choosing the default option or first option presented. Furthermore, we hypothesized that low-SES consumers would be more impacted by nudges because they score lower in relevant skills such as numeracy and because they experience more anxiety when making decisions (Figure 1).
Although the constructs of decision uncertainty, preference construction, anxiety, subjective knowledge, and the three focal moderators (SES, numeracy, and domain knowledge) are all associated with one another, they are distinct ([57]; [58]; [71]). Furthermore, they differ from general intelligence and other types of confidence (e.g., general self-efficacy vs. search confidence; [25]; [55]). These constructs have clear discriminant validity; for example, objective numeracy and domain knowledge are types of objective knowledge or skill, which differ from subjective beliefs about ability (e.g., subjective numeracy, confidence, uncertainty; [29]; [58]). Many people have high confidence in their numeric abilities despite low objective numeracy or vice versa, either of which can lead to harmful financial and health outcomes ([58]). In addition, subjective confidence and anxiety differentially predict memory and evaluations ([57]). Decision anxiety can also impact performance independent of objective numeracy. For example, people can be anxious about disconfirming negative stereotypes despite high objective ability. This can create a self-fulfilling prophecy, because the feeling of stigmatization can increase anxiety and negatively impact decisions ([76]).
Although previous choice architecture research has typically focused on the average effects on consumers, some individual difference moderators of choice architecture effects have been identified. However, nearly all of these moderators have been tested for only a single type of nudge within a single domain. Next, we summarize the research about choice architecture moderators that is most relevant to our hypotheses.
Very little previous research has examined whether the impact of nudges is moderated by SES. For other marketing manipulations such as scams, previous research has suggested that vulnerable consumers (e.g., those who are elderly or less educated) are sometimes targeted and impacted to a greater extent ([33]; [34]; [43]). Within the context of nudges, recent unpublished papers have found that automatic retirement contributions increased savings more for younger and lower-income individuals than others ([ 9]; [18]). This conflicts with other scholars who have made theoretical claims that low-SES individuals are less nudgeable ([61]). Other work has provided mixed evidence about whether low- or high-income individuals are more impacted by different nudges such as framing ([26]; [31]; [69]). Clearly, more research is needed to test these opposing claims across a wide variety of nudges and contexts.
Some theorists have previously suggested that people with more expertise or knowledge might be less impacted by choice architecture. For example, [12] claimed that the aim of policy nudges is to create large benefits for those who have lower expertise and make errors, with minimal impact on more rational or expert decision makers. In other words, consumers with more knowledge or expertise may be less impacted by nudges. However, this claim has received very little empirical attention. One investigation found no default effect in the environmental domain among a sample of environmental economists ([46]). However, the study did not measure experience, examine moderators, or include a control group of people with low experience, so it is difficult to draw conclusions from it. Another investigation found no effect of experience or education on default effects ([38]). There has been some relevant previous research on numeracy. [59] found that numerate people are less impacted than innumerate people by manipulations that present numbers as frequencies rather than probabilities, while [14] found the opposite in the context of Bayesian reasoning. Prior research has not examined whether financial literacy and numeracy moderate effects of defaults, sorting, or other choice architecture tools. Clearly, the present studies are needed to clarify these relationships.
Across six studies, we tested whether nudges have larger impacts on low-SES consumers and those with lower numeracy and domain knowledge. In Study 1, we demonstrate these effects in the context of consumer financial decisions such as selecting which credit card to acquire. In Study 2, we show that the findings of Study 1 generalize across different consumer decision contexts (consumer sustainability decisions, consumer financial decisions, and retail product choices) and different types of choice architecture (interventions that sort options, preselect a default, and reduce the number of options, specifically). In Study 3, we used data from individuals whose employers by default automatically enrolled them into a retirement plan, testing whether consumers with lower SES and domain knowledge were more likely to accept the default enrollment according to self-reported decisions in this high-stakes real-life context. In Study 4, we examined whether the effects of domain knowledge and SES generalize to a vastly different domain: consumer health decisions in the context of COVID-19. Finally, in Study 5 and a supplemental study in the Web Appendix, we conceptually replicated Study 1 while addressing alternative explanations and examining proposed mediators.
We preregistered all studies at aspredicted.org, except Study 3, which used an existing data set. To eliminate the file drawer problem for this research, we report all studies that we conducted and all preregistered analyses for each study. Data, preregistrations, and analysis scripts are available at https://osf.io/a7b32/?view%5fonly=f4df788f178844f6b26e5274a9cbdab1, with the exception of Study 3, which was from a syndicated panel that we do not have permission to share. Across studies, we sought converging evidence for our hypotheses.
In Study 1, participants made five consumer financial decisions. For each decision, they were randomly assigned to a good-default, bad-default, or no-default condition. We hypothesized that good defaults would benefit consumers with low SES, low financial literacy, and low numerical ability more than consumers with high SES, financial literacy, and numerical ability. The Study 1 hypotheses, sample size, and analysis plan are available at https://aspredicted.org/blind.php?x=x547ih.
We requested 450 participants from ROIRocket. Participants (53.1% female; Mage = 50.2 years) were given $1 upon completion of the study. ROIRocket provides a population inexperienced with academic surveys (median of two previous academic surveys; see the Web Appendix), and substantially less experienced than participants on MTurk. To increase statistical power to attain SES effects and ensure that we had enough SES variability, we requested that ROIRocket oversample people who did not finish high school as well as people with advanced degrees (in Study 1 only). ROIRocket provided us with far more participants than requested (N = 825). We included in primary analyses all 825 participants who finished the study.[ 7]
After the consent process, participants made five focal decisions. These decisions are displayed in Table 1. For example, one decision asked participants whether they would repay interest on a high-interest credit card or lower-interest card if they had equal debt on both cards (a common task similar to [ 5]]). Participants were asked to select the option that had the largest total monetary benefits minus costs. These five questions each had a mathematically correct option that would save the most money if it were a real-life decision.
Graph
Table 1. Questions and Answer Options Used in Study 1.
| Questionsa | Optionsb | Good Default | No Default | Bad Default |
|---|
| Imagine you want a new credit card. Imagine also that you make purchases totaling a few hundred dollars each month and always pay just the minimum payment on your credit card (you will always continue doing this each month in the future). You are pre-approved for these three cards. Given this scenario, choose the best credit card considering monetary costs and benefits. | Surge Card (15% APR, no cash back) Trek Card (25% APR, 2% cash back) Journey Card (20% APR, 1% cash back)
| 72% | 66% | 52% |
| Imagine you want a new credit card. Imagine also that you make many purchases each month and always pay off your full balance on the credit card before you accrue interest (you will continue to pay off your full balance each month in the future in this manner). Given this scenario, choose the best card given monetary costs and benefits. | Ascent Card (15% APR, no cash back) Midnight Card (25% APR, 2% back) Trust Card (20% APR, 1% back)
| 66% | 64% | 50% |
| Imagine you have debt on two credit cards with the same bank and have money that you would like to use to pay off this debt. Both cards have balances of more than $500. One card has an interest rate that is twice as high as the other. (Assume your choice won't impact your motivation to make future payments.) | Pay off $500 on higher interest card Pay off $250 on both Pay off $500 on lower interest card
| 76% | 65% | 60% |
| Imagine your employer matches up to 8% if you contribute from your pay checks to your retirement account. Which of these do you choose? (Assume you have 3 years' worth of your new job's salary saved and plan to retire at 65 and live to 85) | Contribute nothing Contribute 2% Contribute 6%
| 86% | 75% | 69% |
| Imagine you have a $425 balance on your credit card, due tomorrow. You have thousands of dollars that you don't need for any other expenses. | Make min payment Pay whole balance Pay $100
| 89% | 85% | 80% |
- 40022242921993180 a The questions presented in this table are abbreviated; for exact text, see the Web Appendix.
- 50022242921993180 b The options presented in italics are the correct answers.
- 60022242921993180 Notes: The percentages listed are the percentages who chose the correct answer. Overall, across each item, accuracy was significantly higher in the no-default condition compared with the bad-default condition and significantly higher in the good-default condition compared with the no-default condition. APR = annual percentage rate.
For each question, participants were randomly assigned to one of three default conditions. In the no-default condition, no answer was preselected. In the good-default condition, the correct option (which would save the consumer the most money) was preselected. In the bad-default condition, an incorrect (i.e., more costly) option was preselected. Participants in the good- and bad-default conditions were told, "An option has been pre-selected for you. You may keep that selection or switch to another option." Because the default condition was randomly determined for each question, participants received different conditions for different questions. We used this design to increase power ([52]).
After making the five focal decisions, participants completed measures designed to assess their predictions about how much they were influenced by the defaults. Two questions asked them how likely they thought they would be to get a focal consumer financial decision correct if ( 1) the correct answer was preselected or ( 2) if an incorrect answer was preselected.
Then, participants completed measures of the factors we predicted would moderate nudge effects—financial literacy, numeracy, and SES. They also completed exploratory measures of agreeableness, need for cognition, self-reported credit score, and self-reported patience (for text of all measures, see the Web Appendix). To assess financial literacy, we used a common scale ([25]) that asked participants multiple choice questions about common financial instruments and techniques such as stocks, 401(k)s, and diversification (α =.85). We measured numeracy with 11 questions (α =.87) that assessed understanding of probability, frequency, and percentages ([44]). Following previous research and American Psychological Association recommendations for measuring and conceptualizing SES ([ 1]; [62]), the SES measure included three components: education level, occupation status, and income. As in previous SES research, we standardized and averaged the three components for analysis ([ 1]). The measure had high internal consistency (α =.78). Factor analyses indicated that the SES, financial literacy, and numeracy items loaded on three separate factors as expected (Web Appendix Tables A1 and A2; oblimin rotation was used).
We included measures of agreeableness and need for cognition in this study to address alternative explanations that agreeable personalities or desires for elaborative thought (rather than SES and domain knowledge) might explain differences in default effects across people. We also measured the total time participants spent completing the study (log-transformed as preregistered), which served as a proxy for overall survey engagement.
In addition, we included assumption check items to ensure that the correct answers were best for a wide variety of people, including those with low SES and few liquid assets (details in the Web Appendix). After responding to the main measures in Study 1 but before reporting demographics, participants made three consumer decisions with no correct answer, so that we could ensure that the focal moderators generalize beyond the context of questions with a correct answer. The three items asked participants to choose which flight insurance option to buy; which laptop computer to buy based on price, image, and consumer reviews; and which painkiller to purchase based on price and brand. Each item had three options and participants saw each item with one of the three options preselected or with no option preselected. Results from these questions were not included in the primary analyses mentioned in the preregistration, so we label these analyses as exploratory and report them separately from primary analyses.
Participants also reported demographics and how many past studies they had completed. We included an attention check to ensure that effects were robust when accounting for people who rushed through the survey. The attention check asked them to select a particular answer for a fake question added in the middle of the financial literacy scale. Following our preregistration, we included all participants, including those who failed the attention check, in primary analyses (though all effects remained significant when excluding attention check failures).
In each study, we analyzed results using binomial generalized mixed effects models. We estimated decision accuracy (1 = correct, 0 = incorrect) as the dependent variable and treated participants as random factors to properly model variance across people ([ 6]). As preregistered, the models in studies with three default conditions included a contrast-coded default condition term (1 = good default, 0 = no default, −1 = bad default) and the orthogonal contrast. All models contained item fixed effects that accounted for variation in difficulty across different questions. We tested hypothesized moderators of default effects and standardized the moderating variables.
Defaults strongly influenced decisions on average. Participants in the bad-default condition answered 62% of items correctly (choosing the most advantageous option) compared with 71% in the no-default condition and 78% in the good-default condition (z = 10.73, Exp(B) = 1.62, p <.001). Simple effects tests indicated that the difference between the no-default and good-default conditions was sizable (z = 4.39, Exp(B) = 1.62, p <.001), as was the difference between the no-default and bad-default conditions (z = −4.63, Exp(B) =.61, p <.001).
As predicted, there was an SES × default condition interaction, such that default effects were larger among lower-SES consumers than higher-SES consumers (z = −3.64, Exp(B) =.83, p <.001; Figure 2, Panel A). Simple effects tests indicated that default effects were over 2.2 times larger for people in the bottom half of the SES distribution compared with the top half. SES was weakly correlated with survey engagement (r =.03), and its interaction with default condition was robust when controlling for engagement (z = −3.65, p <.001).
Graph: Figure 2. Default effects were larger among consumers lower in SES, lower in financial literacy, and lower in numeracy.Notes: The bad-default, no-default, and good-default conditions are depicted by dashed, dotted, and solid lines, respectively. Good nudges reduced disparities, as depicted by the small difference between consumers low and high in each variable in the good-default condition (shallow solid line) compared with the no-default and bad-default conditions (steeper dotted and dashed lines). Shaded regions depict ±1 SE. The histograms along the x-axis depict the distribution of each moderator.
As we predicted, there was a large financial literacy × default condition interaction (z = −6.32, Exp(B) =.75, p <.001; Figure 2, Panel B). Participants lower in financial literacy were impacted by defaults more than participants higher in financial literacy. This interaction remained significant when controlling for SES, numeracy, and their interactions with default condition (z = −2.41, Exp(B) =.86, p =.016).
There was also a numeracy × default condition interaction (z = −6.83, Exp(B) =.74, p <.001; Figure 2, Panel C), such that those with lower numerical ability were impacted by defaults more than those with higher numerical ability. This interaction remained significant when controlling for SES, financial literacy, and their interactions with default condition (z = −3.63, Exp(B) =.82, p <.001). This implies that the interactions with numeracy and financial literacy were at least partly independent effects.
Our conceptual framework (Figure 1) suggests that financial literacy and numeracy account for the SES × default condition interaction. Consistent with this, the SES × default condition interaction was greatly reduced when we controlled for numeracy, financial literacy, and their interactions with default condition (z = −.46, Exp(B) =.97, p =.648). In the Web Appendix, we show that mediation models were also consistent with this idea that numeracy and financial literacy account for the moderating effects of SES on default effects.
We also conducted exploratory analyses of three consumer choice questions with no correct answer. We included these items to examine whether the key results (that consumers lower in SES, numeracy, and domain knowledge are more impacted by defaults) generalized beyond the context of questions with a correct answer. Participants with lower SES were more likely to retain default options on average (z = −3.22, Exp(B) =.81, p =.001). In addition, those with lower financial literacy were more likely to retain the default options (z = −4.36, Exp(B) =.79, p <.001), as were those with lower numeracy (z = −4.85, Exp(B) =.76, p <.001). This suggests that participants with low SES, low financial literacy, and low numeracy are more likely to choose default options and that our key findings are not simply the result of participants with low SES, low financial literacy, and low numeracy having less access to correct answers.
Participants predicted that defaults would have little, if any, impact on their decision accuracy. We asked participants two questions in which they reported how likely they thought it was that they would answer a focal consumer financial decision question correctly ( 1) if the correct answer was preselected and ( 2) if an incorrect answer was preselected (in each case, they were asked to assume they were not told whether the default option was correct).
Participants thought their likelihood of answering correctly would be 65% if assigned to a good default and 64% if assigned to a bad default (t(824) = 1.84, p =.066). Financial literacy, numeracy, and SES were not significantly associated with participants' predictions of how much they would be impacted by defaults (see the Web Appendix; if anything, more numerate consumers thought they would be impacted more by defaults, though they were actually less impacted). Interestingly, participants were not overconfident on average; they were simply miscalibrated about default effects. They greatly underestimated how accurate they would be when assigned to a good default (estimates = 65%, reality = 78%) and were close to reality when regarding bad defaults (estimates = 64%, reality = 62%).
We preregistered the following three robustness tests. In the first, we wanted to control for how engaged participants were with the study (assessed via the log-transformed time they spent completing it). In the second, we controlled for agreeableness and need for cognition.[ 8] In the third, we excluded participants who failed the attention check. All three focal moderators remained significant and similar in size across all of these robustness tests (all zs < −3, all ps <.001; for further details, see the Web Appendix).
As we predicted, consumers who had lower SES, lower financial literacy, and lower numeracy were more impacted by defaults than consumers who had higher SES, higher financial literacy, and higher numeracy. In other words, good defaults were an equalizer that helped reduce the differences in decision quality between consumers with low versus high SES, numeracy, and financial literacy. Interestingly, participants seemed largely unaware of the impact of defaults. They did not anticipate that defaults would influence their behavior, nor did consumers lower in SES, financial literacy, or numeracy predict they would be impacted more. It is worth noting that we oversampled people with very low or very high education in Study 1. Although this increased statistical power, the sample was different from the general population. In subsequent studies, we use more balanced samples (with no oversamples), and in Study 3 we use a more representative stratified random sample of U.S. households.
In a supplemental study, we addressed alternative explanations for the effects found in Study 1, namely that effects of financial literacy and numeracy might be explained by participants who were not understanding the questions, not paying attention, or not conscientious (see Web Appendix). In this supplemental study, we replicated these key interactions from Study 1 and showed that these were robust even when controlling for comprehension of the decision questions and individual differences in conscientiousness. This suggests that people lower in numeracy and domain knowledge are impacted more by defaults, that these effects are replicable, and that they are not attributable to low conscientiousness or poor comprehension.
Study 1 highlights how default effects are moderated by differences in financial literacy, numeracy, and SES. In Study 2, we wanted to examine whether these results generalize across three different types of nudges in three decision-making contexts with incentives for accuracy.
Study 2 was designed primarily to test whether the results observed in Study 1 generalize across different types of nudges and across different consumer contexts. In addition, we added incentives for half of the decisions to examine whether incentives moderate the effects observed in Study 1. We expected that moderators observed in Study 1 would generalize across the three nudge types, across the three contexts, and across incentivized and nonincentivized decisions. We also included a measure of general fluid intelligence to isolate domain knowledge from general intelligence. We preregistered sample size, predictions, and analyses at https://aspredicted.org/blind.php?x=v3ci5q and report all preregistered analyses.
ROIRocket respondents (N = 428; 51.6% female; Mage = 53.2 years) participated in exchange for a fixed payment of $1 and a $2 bonus if they answered one of the focal consumer financial decisions correctly. In this and all subsequent studies, participants who had completed any of our previous studies were not allowed to participate.[ 9]
The procedure was similar to Study 1, but with three different types of nudges and with decisions that spanned three different contexts. Participants answered six focal questions with mathematically correct answers. The three contexts were retail product choices, consumer financial decisions, and consumer sustainability decisions. The two retail product choices involved choosing a computer with or without insurance, and food with the lowest price per ounce. The consumer financial decisions were slightly altered versions of the debt repayment and retirement questions used in Study 1. The consumer sustainability decisions involved choosing window insulation that would maximize total savings and choosing lightbulbs with the lowest unit price. Participants were asked to choose the item with the lowest average monetary costs and were incentivized to choose these options for half the questions. The Web Appendix provides the full text of each question. The three types of nudges were defaults, sorting, and number of options. The default manipulation was similar to Study 1 but with only the good-nudge and no-nudge conditions (because these often have higher ecological validity),[10] the sorting manipulation varied whether options were ordered from best to worst ("good sort") or randomly ("no sort"), and the number of options manipulation varied whether ten options were presented ("many options") or only two of the best options ("few options"), following [67]. All sorting and default questions had ten options.
The design was thus a 3 (context: retail product choices, consumer financial decisions, consumer sustainability decisions) × 3 (nudge type: defaults, sorting, number of options) × 2 (nudge condition: good nudge, no nudge) × 2 (incentive: $2, $0) experimental design. The questions were organized in three blocks in counterbalanced order corresponding to different contexts and nudge types.[11] The first three questions were incentivized for some participants and the last three questions were incentivized for others.
Following the six focal decisions, participants completed the same measures of financial literacy as in Study 1 and a shortened three-item version of the numeracy measure ([65]) to reduce the length of the survey. To isolate domain knowledge effects (financial literacy) from general intelligence, we included a measure of general fluid intelligence called number series ([51]). The measure asked participants to answer six questions that involved completing a pattern of numbers such as "23, 26, 30, 35, __" (correct answer: 41). Then, participants completed the three-item of measure of SES described in Study 1. Finally, participants reported their credit score range, completed the attention check item, completed a measure of time preferences (see the Web Appendix), and reported their age and gender.
On average, the nudges had their intended effects. We estimated accuracy in binomial mixed-effects models as a function of nudge condition (contrast-coded), with the rest of the model the same as in Study 1. Accuracy was higher when good nudges were used (M = 56%) compared with no nudge (M = 42%; z = 7.49, Exp(B) = 1.87, p <.001). These effects were strong for the default and number of options nudges but nonsignificant for sorting (Mgood default = 55%, Mno default = 40%; Mfew options = 68%, Mmany options = 43%; Mgood sort = 46%, Mno sort = 43%).
Nudge effects were moderated by SES such that they impacted low-SES participants more than high-SES participants (z = −2.92, Exp(B) =.77, p =.004). That is, nudges designed to facilitate selection of the best option reduced choice disparities by helping low-SES consumers more than high-SES consumers. Consistent with our framework (Figure 1), when we included financial literacy and the financial literacy × nudge condition interaction in the model, the SES × nudge condition interaction was no longer significant (χ2( 2, n = 413) = 1.87, p =.170). The SES × nudge condition interaction was not significantly moderated by nudge type (χ2( 2, n = 428) = 2.74, p =.255) or decision context (χ2( 2, n = 428) = 4.09, p =.129). It was also robust when controlling for survey engagement (z = −2.91, p =.004), and SES was very weakly correlated with survey engagement (r =.02).
As we predicted, nudges had more impact on consumers with lower financial literacy than those with higher financial literacy (z = −2.42, Exp(B) =.80, p =.015). Figure 3 shows the robust effects of financial literacy across studies. These financial literacy × nudge condition interactions were not significantly moderated by the type of nudge (χ2( 2, n = 428) =.47, p =.790) or by the decision context (χ2( 2, n = 428) = 3.06, p =.216).
Graph: Figure 3. Forest plot conveying the three moderators of nudge effects across studies.Notesz: The effects were relatively consistent and robust across studies, though numeracy and SES had nonsignificant effects in one study each. Study 3 is omitted because it used a different dependent variable (self-reports of whether participants retained default retirement options).
Unlike in Study 1 and all of our subsequent studies, numeracy did not moderate the impact of nudges (z = −.27, Exp(B) =.97, p =.785). In the Web Appendix, we explore different reasons for this difference, concluding that this is partly attributable to low reliability and lower validity on the three-item numeracy scale in Study 2 (α =.53) compared with the longer and more sensitive numeracy measure used in Study 1 and subsequent studies (Study 1: α =.87). The relationship between numeracy and nudge effects was not moderated by the type of nudge (χ2( 2, n = 428) =.57, p =.751) or by the decision context (χ2( 2, n = 428) = 1.57, p =.457).
We predicted that general intelligence would also moderate default effects but that it would not fully account for the financial literacy effect. Contrary to our expectations, consumers who scored higher on the measure of general fluid intelligence were not significantly less susceptible to nudges (z = −1.41, Exp(B) =.87, 95% confidence interval [CI] = [.72, 1.05], p =.157). The financial literacy × nudge condition and SES × nudge condition interactions remained significant when controlling for fluid intelligence (both zs = −2, ps <.05). This finding suggests that financial literacy and other forms of domain-specific knowledge likely influence nudge effects more than general fluid intelligence.
We controlled for survey engagement, which did not appreciably change the interactions of condition with financial literacy or SES (both zs < −2, both ps <.05).
The incentive manipulation did not significantly influence accuracy (Mincentivized = 51%, Mnonincentivized = 48%; z = 1.35, Exp(B) = 1.13, p =.178), though it did increase the amount of time participants spent on the questions (Mincentivized = 126 seconds, Mnonincentivized = 87 seconds; t(2,132.01) = 4.23, b =.11, p <.001). On average, the nudges increased accuracy about four times more than a $2 incentive. The key interactions were not any smaller for the incentivized questions than the nonincentivized questions (see the Web Appendix).
Consistent with Study 1, SES and financial literacy each moderated the effects of nudges in Study 2. These effects were present even though decisions were incentivized. It is not surprising that the effect of financial literacy was not moderated by the decision context, because the decisions we examined in Study 2 all involved numbers, prices, financial information, or calculations. As mentioned previously, financial literacy and numeracy are useful across many contexts of consumer choice because they are used to compare prices and quantities, calculate unit prices, and calculate cost effectiveness and long-term savings ([28]; [63]). Thus, we did not expect context to moderate effects of financial literacy in Study 2.
Although the results of Studies 1 and 2 demonstrate important and consistent effects, it is not yet clear whether the results generalize to high-stakes, real-life decisions. Therefore, in Study 3, we use data about Americans' self-reported retirement investment choices. We examine whether defaults influence low-SES consumers more than high-SES consumers in this context.
In Study 3, we acquired (self-reported) data about Americans' retirement investment decisions. We examined a sample of consumers who work for companies that set defaults by automatically enrolling employees into retirement contributions. Respondents were asked whether they opted out of the default contribution amount and default investment allocation set by their company. We predicted that consumers lower in SES and financial knowledge would be more likely to choose the default options than those with higher SES and financial knowledge.
The secondary data we used consisted of stratified random samples of U.S. households. The panel, Strategic Business Insights (SBI) MacroMonitor, is a syndicated panel that asks respondents questions about their financial decisions and demographics. The panel is conducted with different households every other year. We were given access to four different samples from the panels that were conducted in 2010, 2012, 2014, and 2016, respectively.
Our primary interest was in three questions that asked respondents whether they accepted or rejected their employer's default options in real retirement decisions. Specifically, respondents were asked whether their current employer automatically enrolled them into a retirement plan (753 indicated yes, 3,580 indicated no, and the rest selected "does not apply" because they were retired or unemployed). Following this, respondents who had answered "yes" were asked two questions assessing ( 1) whether they kept the default contribution percentage and ( 2) whether they kept the default investment allocation. Of those who reported they were automatically enrolled, 48% indicated they accepted the default investment allocation, whereas 52% opted out and chose a different allocation. For the default contribution amount question, 45% indicated they had accepted the default contribution amount, whereas 55% opted out. We analyzed default selection (1 = chose default option, 0 = opted out of default) in binomial generalized mixed models as a function of the question (allocation or amount) and hypothesized predictors.
We examined measures of SES and financial sophistication. The SES measure followed the preregistered measure used in Study 2 as closely as possible (education, income, and occupation, standardized and combined; for details, see the Web Appendix).
Self-reported financial sophistication was analyzed using the following two measures, consistent with previous research that used the SBI MacroMonitor data ([54]). The self-reported financial sophistication item asked participants to rate their agreement with the statement "I consider myself a sophisticated investor" (1 = "mostly disagree," and 4 = "mostly agree"). The financial experience item asked respondents whether they handle their household's financial investments. Other items were assessed in the survey, including gender, age, marital status, number of children, U.S. census region, religion, race, hours worked per week, and risk aversion.
We first tested whether low-SES individuals were more likely to choose the default options. Participants with lower SES were more likely to accept the default options (z = −5.71, Exp(B) =.33, p <.001).
We computed a model estimating default choices as a function of self-reported financial sophistication and financial experience. Individuals with lower financial sophistication were more likely to accept the default option (z = −5.62, Exp(B) =.40, p <.001), as were those with lower financial experience (z = −2.88, Exp(B) =.66, p =.004). These effects are broadly consistent with Studies 1 and 2, though SBI used measures of financial sophistication that differed from the financial literacy scale we used in the experiments we designed.
We also conducted a robustness test in which we controlled for all the covariates listed in the "Method" subsection. This was designed to address alternative explanations that the effects of SES and financial sophistication were actually explained by differences in any of these other variables. When adjusting for these covariates, the effects of SES, financial sophistication, and investment experience remained significant (all zs < −3, ps <.01). SES and financial sophistication influenced both default questions individually (Web Appendix).
The results of Study 3 demonstrate that consumers with low SES and low financial sophistication are more likely to retain default options, even in self-reports of their high-stakes retirement decisions. This is consistent with working papers that found larger effects of automatic enrollment for younger and low-income individuals compared with older and high-income individuals ([ 9]; [18]). It is worth noting that typical default enrollment rates of 3% and 6% are likely insufficient for many people, and it is possible that some respondents who opted out chose higher amounts in Study 3. Therefore, we cannot infer that automatic enrollment improved decisions.
Though Studies 1–3 suggest the results generalize to many important decisions, most of the decisions we examined were consumer decisions with prices or financial elements. In Study 4, we demonstrate generalizability further by examining a dramatically different context of health decisions in the early stages of the COVID-19 pandemic.
In Study 4, we aimed to generalize our results from Studies 1–3 to questions about optimal behavior during the COVID-19 pandemic. We hypothesized that participants with lower SES, numeracy, and health literacy would be impacted more by defaults in this context. We also tested whether domain-specific health knowledge moderated nudge effects more than less relevant financial knowledge. Thus, unlike in the previous studies, we did not predict financial literacy would moderate default effects, because it is less relevant for health decisions. Instead, we predicted that health literacy would moderate default effects. We preregistered the sample size, hypotheses, and analyses at https://aspredicted.org/blind.php?x=an4kx6.
Participants from ROIRocket completed the experiment in exchange for $.50 (N = 305; 50.8% female; Mage = 52.0 years). This experiment was conducted in April 2020 while much of the United States was under restrictions designed to slow the spread of COVID-19.
Participants answered four questions about how they would respond to different scenarios in the context of COVID-19. The four questions, respectively, asked participants whether they would wear a mask in public, how they would disinfect surfaces, what they should do if they have an upset stomach and runny nose, and how long to wait before touching packages delivered to the door (for full text, see the Web Appendix). Participants were told to follow Centers for Disease Control and Prevention guidelines and assume that those guidelines were all correct. Answers were coded for accuracy (1 = correct, 0 = incorrect). For each question, participants were assigned to either the no-default or good-default condition. We did not include a bad-default condition, because it could spread misinformation about COVID-19. The questions assigned to each condition were counterbalanced, and participants received two questions in each condition.
Following these four questions, participants completed measures of numeracy, financial literacy, health literacy, SES, other demographics, and an attention check. We used a longer nine-item numeracy measure in Study 4 ([44]), because we suspected that the null numeracy result in Study 2 was due to low reliability of the three-item measure. The numeracy measure included two subscales consisting of health numeracy questions (six items) and general numeracy questions about lotteries (three items), respectively. The health literacy measure included items such as interpreting "drug facts" from a medicine label (see Web Appendix). The other measures (financial literacy, SES, and attention check) were the same as in Study 2.
On average, good defaults increased accuracy compared with the no-default condition. Accuracy was significantly lower in the no-default condition (M = 64%) compared with the good-default condition (M = 72%; z = 3.35, Exp(B) = 1.58, p <.001).
As we predicted, consumers with lower SES were more impacted by defaults, as indicated by the SES × default condition interaction (z = −2.31, Exp(B) =.73, p =.021). The default effect was over four times larger among consumers with below-average SES compared with those with above-average SES. This effect was no longer significant when we added numeracy to the model (z = −1.43, Exp(B) =.81, p =.151), consistent with Figure 1.
Overall, numeracy moderated default effects: less numerate participants were more impacted by defaults than numerate participants (z = −2.57 Exp(B) =.71, p =.010). To examine whether domain-specific health numeracy impacted decisions more than general numeracy, we also separately examined subscales that assessed health numeracy and general numeracy, respectively. Health numeracy significantly moderated the default effects, such that those with lower health numeracy exhibited larger default effects (z = −2.83, Exp(B) =.79, p =.004). In contrast, the general numeracy subscale did not significantly moderate default effects (z = −1.57, Exp(B) =.81, p =.117).
We predicted that same-domain (health) knowledge would influence default effects more than other-domain knowledge (e.g., financial literacy). Consistent with this, financial literacy did not significantly moderate the default effects (z =.55, Exp(B) = 1.02, 95% CI = [.83, 1.40], p =.582). Note that one cannot conclude from a nonsignificant result that the moderating effect of financial literacy is zero. However, the 95% CI includes only small positive or negative effects that are smaller than the moderating effects of numeracy and SES (for Bayes factor analyses, see the Web Appendix).
Although we expected health literacy to significantly moderate default effects, this result was only marginal (z = −1.81, Exp(B) =.79, 95% CI = [.60, 1.02], p =.070). As detailed in the Web Appendix, we suspect that this health literacy result was marginal and smaller than expected because nearly all participants scored very high on the measure (giving us low power due to the low variability). Health literacy was weakly correlated with SES (r =.13).
We preregistered two robustness tests that excluded attention check failures and adjusted for survey engagement, respectively. All significant interactions with default condition remained significant in these robustness tests (all zs < −2, ps <.05).
We used a bootstrapped mediation model with 5,000 resamples ([60]) to examine whether consumers with low SES are more nudgeable because they are less numerate (see Figure 1). There was a significant indirect effect consistent with the proposed path from lower SES to lower numeracy to larger default effects (indirect effect = −.07, 95% CI = [–.13, −.02]). The effect of SES on the size of default effects was reduced when numeracy was added to the model (from c = −.19, 95% CI = [–.31, −.06] to c1 = −.11, 95% CI = [−.25,.03]), consistent with our predictions. An alternative mediation possibility is that SES influences nudges by causing consumers to allocate time differently ([69]). Contrary to this possibility, SES was not associated with time spent on these questions (z = −.19, Exp(B) =.99, p =.847), and there was no indirect effect of SES on default effects through decision time in a parallel mediation model (ab =.00, 95% CI = [−.01,.01]).
The results of Study 4 replicate and extend the results of previous studies to the context of COVID-19 health decisions. Low-SES people benefited disproportionately from nudges even in the context of questions about COVID-19. Low-SES people are disproportionately affected by COVID-19 and thus have the most to gain from interventions that help them. Mediation models were consistent with our framework in which low-SES individuals are more impacted by nudges, not because they allocate time differently but because of differences in domain-specific skills. In Study 5, we test the remainder of our conceptual diagram in sequential mediation models.
Study 5 had two purposes. First, we generalized our results across two different samples, including a sample of Master of Business Administration (MBA) students at an elite university. This would ensure that our findings generalized beyond a sample with relatively low financial knowledge. Second, we tested the proposed mediation model displayed in Figure 1 about why financial literacy, numeracy, and SES moderate default effects. Consumers with low numeracy and financial literacy experience greater uncertainty and anxiety when facing consumer decisions involving numbers or math ([71]). In turn, anxiety and uncertainty likely increases susceptibility to default effects (e.g., [36]). We tested these proposed paths with mediation models. In addition, we hypothesized that financial literacy, numeracy, and SES would moderate default effects, replicating the results of our previous studies. We preregistered the sample size, hypotheses, and analyses at https://aspredicted.org/blind.php?x=4yz385.
We requested and preregistered a sample of 200 participants from ROIRocket and an estimated 100 MBA students. All participants received a $2 bonus if they answered one randomly selected focal financial question correctly. ROIRocket participants also received $1 fixed payment, whereas MBA students received points for a minor class assignment. The ROIRocket sample was more diverse and older (n = 212; 50.9% male; median age = 54 years) than the MBA sample (n = 75; 61.3% male; median age = 29 years). The MBA students had higher financial literacy and numeracy compared with ROIRocket participants (financial literacy questions answered correctly: MMBA = 87%, MROIRocket = 64%; numeracy questions answered correctly: MMBA = 90%, MROIRocket = 53%; both ts > 3, ps <.001).
The procedure was similar to Study 1, except for the following differences. All five decisions were incentive compatible, and one retail product choice (of laptops with different insurance options) was added (also used in Study 2).[12] We also examined potential mediators by assessing perceived uncertainty, decision anxiety, and preference construction; we suspected that each of these three variables partially accounts for the effects of domain-specific skills on default effects (as described previously). The three factors, though correlated, had discriminant validity (see the Web Appendix) and have also been differentiated in previous research ([57]; [58]; [71]).
We used the same model structure as in Study 1. Participants were more likely to choose the correct answer in the good-default condition (M = 63%) than in the no-default (M = 60%) and bad-default conditions (M = 56%; z = 3.23, Exp(B) = 1.33, p =.001).
Unlike in the previous studies, SES did not significantly moderate the effects in Study 5 (z =.12, Exp(B) = 1.01, p =.901).[13] SES was not correlated with survey engagement either (r =.00).
As we predicted, consumers with lower financial literacy were more impacted by defaults as in the previous studies (z = −4.21, Exp(B) =.70, p <.001; Figure 3). The default effect was over five times larger among consumers with below-average financial literacy compared with those above-average financial literacy. When we controlled for the numeracy × default condition interaction, the financial literacy interaction remained significant.
Participants low in numeracy were also more impacted by defaults as indicated by the numeracy × default condition interaction (z = −2.81, Exp(B) =.78, p =.005).
Financial literacy and numeracy moderated the default effects even when adjusting for survey engagement (and when adjusting for MBA vs. ROIRocket participants; all zs < −2.5, ps <.01). When we excluded attention check failures, the interaction with financial literacy remained similar in size, though the interaction with numeracy reduced slightly and was marginal (financial literacy: z < −2, p <.01; numeracy: z = −1.84, Exp(B) =.84, p =.065).
We conducted mediation models with 5,000 bootstrapped resamples to examine the proposed mediation paths displayed in Figure 1 ([60]). The first models examined the paths from SES to numeracy to the three possible mediators (anxiety, preference construction, and decision uncertainty) to larger default effects. When examining these three mediators in parallel, there was a significant indirect effect through anxiety, consistent with partial mediation through anxiety (ab = −.01, 95% CI = [−.027, −.001]). This reflected a positive relationship between SES and numeracy (b =.39, 95% CI = [.28,.50]), a negative relationship between numeracy and anxiety (b = −.29, 95% CI = [−.42, −.16], and a positive relationship between anxiety and larger default effects (b =.11, 95% CI = [.04,.18]). (A second indirect effect through preference construction was significant when examined without the other two mediators, but not in a parallel mediation model with the other two mediators. There was no significant indirect effect through uncertainty, contrary to expectation.) The analogous indirect effects through financial literacy rather than numeracy revealed very similar results (see the Web Appendix). Although there was no direct effect of SES on the size of default effects in Study 5 (unlike the previous studies), we nonetheless found support for the proposed indirect effect through numeracy and anxiety. This is consistent with our conceptual framework, though, like any mediation analysis, it should be interpreted with caution because mediation analyses cannot conclusively determine whether a mediator causes an effect.
In Study 5, we examined whether the moderators of default effects observed in Studies 1–4 would generalize to a markedly different sample (MBA students). Consumers with lower financial literacy and numeracy were more impacted by defaults, and the mediation model was consistent with our theoretical explanation (see Figure 1) of these default effect moderators.
Across several studies, nudges not only influenced decision making on average but also influenced choice disparities across consumers. Low-SES consumers were impacted more by nudges, meaning that nudges that facilitated selection of a good option benefited them more than high-SES consumers. Domain knowledge and numeracy also moderated the effects of nudges: Consumers with less domain knowledge and lower numeracy were impacted more by nudges compared with those with more domain knowledge and higher numeracy.
These results generalized across a wide variety of consumption contexts. In addition, the effects were sizable. Across studies, nudges typically had two to five times greater impact among consumers with below-average SES, domain knowledge, and numeracy compared with consumers with above-average SES, domain knowledge, and numeracy. These results remained strong in incentivized decisions and across a series of preregistered robustness tests in which we adjusted for survey engagement, attention check failures, and alternative explanations of our results.
In our studies, we sought to use decisions in which one option was best for essentially all consumers (even those with low SES and few liquid assets). The results of Study 1 were consistent with this assumption. We provided participants with the outcomes of options in Study 1 based on their actual age and liquid assets and asked them which outcome would leave them better off (see the Web Appendix). The vast majority of consumers, including those with low SES and few liquid assets, selected the options facilitated by the good nudges of saving more for retirement and making a full credit card payment as more beneficial than the other options.
Because we tested the moderators of nudges across several contexts, it was possible to examine whether domain-specific skills and knowledge drive these effects. Financial literacy moderated nudge effects in the context of consumer financial decisions but not COVID-19 health decisions. In the context of the COVID-19 health decisions, health numeracy significantly moderated default effects, whereas general numeracy was not a significant moderator. These findings provide evidence that skills and knowledge moderate the effects of nudges primarily in the particular contexts in which those skills and knowledge are relevant.
Nudges have become pervasive in marketing firms and policy circles because of their low costs and large average impact ([ 7]). Our results demonstrate that, beyond improving decisions on average, good nudges can reduce disparities. Because nearly every standard of ethics endorsed by governments and corporations places value on equality and reducing inequities (e.g., [66]), this provides a strong reason to use nudges.
In addition, our findings have implications for nearly any marketing manager or online retailer. Choice architecture is an unavoidable aspect of online retail. For example, retailers must present products in some order, whether they ultimately choose to present products with highest ratings, lowest prices, highest sales volumes, or highest profit margins first ([72]; [75]). At checkout, retailers can choose to set the default to be the product with no insurance, no add-ons, and the least expensive shipping option, or other options can be preselected that might increase revenue. The results of the present studies suggest that these choice architecture decisions not only impact consumers' choices on average but can help reduce choice disparities. Many marketing managers try to reduce inequity and invest in expensive efforts to do so ([41]). For example, some marketing firms reduce their prices for the poor or offer financial assistance to expand access to their products and reduce inequities. Because nudges are low-cost interventions and can promote options in the mutual best interest of consumers and firms ([ 7]), the present results suggest that nudges may be an inexpensive alternative way for firms to help the poor.
The present results also suggest that policy makers and firms need to carefully monitor the impact of their choice architecture tools on different segments of the population. Scholars have recently argued that it is important that researchers and policy makers understand heterogeneous effects of nudges across people ([74]). This can allow policy makers to design interventions that are effective even if they do not impact all consumer segments or groups of the population. Heterogeneity in nudge effects might also partly explain why some nudges have had smaller effects when applied and implemented at scale by policy makers or firms than when examined by researchers ([ 4]; [21]). In addition, our replications of the key results across studies and contexts addresses recent calls for researchers to replicate nudge effects ([ 4]) and examine effects of the context ([74]) to make results more useful to practitioners.
Similarly, understanding heterogeneity can help marketing firms and retailers target consumer segments that would be most impacted by nudges. For example, nudges that present options with lowest unit prices first might increase purchases among low-knowledge consumers who are less familiar with the brand more than high-knowledge consumers. In some cases, managers who ignore heterogeneity in nudge effects might underestimate the effectiveness of nudges if, for example, the low-knowledge consumers most impacted by nudges include many new customers who will continue to purchase the brand in the future. In other cases, managers who are unaware of this heterogeneity might overestimate nudge effects if, for example, nudges influence one-time purchases from low-knowledge consumers rather than high-knowledge repeat customers with greater customer lifetime value. Because low-SES consumers are most impacted by nudges, this may suggest that nudges will be less successful among luxury retailers and anyone with high-SES clientele, compared with retailers catering to low-SES clientele.
When only a one-size-fits-all nudge is available, our results suggest that policy makers should focus on the needs and potential benefits for low-SES and low-knowledge citizens when deciding which option to facilitate with the nudge. Nudges have less influence on high-SES and high-knowledge individuals, and it is reasonable to focus on the policies that will benefit those most impacted. For example, if one health care plan is optimal for low-SES people while another option is better for high-SES people (and only a one-size-fits-all nudge is possible), choice architects should prioritize the needs of low-SES individuals when choosing the default option.
Although we think choice architecture interventions are important tools that can reduce disparities, they should not be the only tools used to address them. Many disparities are systemic and deeply entrenched for historical, societal, or macroeconomic reasons and require interventions that change laws or elements of the macroeconomy ([24]; [45]). In addition, interventions that use incentives or provide new information can be an effective supplement to nudges ([45]). Nudges can be part of a solution that reduces disparities, but they are not enough by themselves.
Across our studies, we found that the moderating effect of SES was consistent in a wide variety of contexts including consumer product choices that contained no calculations and no correct answer. Of course, it is possible that these effects do not generalize to decisions in every context. The findings might not generalize to cases in which the nudged behavior is deeply constrained ([61]). For example, healthy eating nudges might be ineffective if many low-SES consumers live in food deserts, where healthy food is difficult to obtain or expensive. It is also possible that the moderating effects would be smaller or absent for decisions in which knowledge is irrelevant or in which numbers, calculations, and ambiguity are absent. Similarly, if low-SES consumers have strong preferences and more expertise than others within a particular domain, the effect in which nudges impact low-SES consumers most might not generalize to that domain.
Future research should also examine the mechanisms underlying the nudge moderators in more detail. Mediation models were consistent with our framework in which low-SES consumers are more impacted by nudges because they score lower in domain-specific skills such as numeracy (not because they allocate time or attention differently; cf. [69]). There were also indirect effects through anxiety in Study 5 (but not through uncertainty). Future investigations could expand on this by manipulating psychological processes and by examining different forms of confidence and uncertainty (e.g., [25]). Moreover, future work should examine the extent to which subjective rather than objective knowledge accounts for the effects. It is possible that people with low subjective knowledge would be greatly impacted by nudges even if they have high objective knowledge. Future work could also examine why anxiety plays a role in the differential nudge effects. For example, low-income individuals often feel stigmatized and anxious about confirming a negative ability stereotype (e.g., [76]), which might account for any effects of anxiety on nudge effects.
Of course, though we manipulated choice architecture, we cannot conclude that SES, financial literacy, or numeracy caused consumers to be less susceptible to nudges, because we did not manipulate these variables. Some researchers have manipulated temporary scarcity or perceived social class (e.g., [69]). However, we would not expect these manipulations to increase nudge effects because they do not operate through our proposed mechanisms of financial literacy, numeracy, and anxiety.
When signing copies of his book Nudge, Richard Thaler often writes "Nudge for good," encouraging readers to use nudges to benefit people rather than to increase profits at the expense of consumer welfare. The present investigation suggests that "nudging for good" not only helps consumers overall but also reduces inequities. The implications are clear for anyone interested in reducing inequities: nudge for good.
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921993186 - Do Nudges Reduce Disparities? Choice Architecture Compensates for Low Consumer Knowledge
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921993186 for Do Nudges Reduce Disparities? Choice Architecture Compensates for Low Consumer Knowledge by Kellen Mrkva, Nathaniel A. Posner, Crystal Reeck and Eric J. Johnson in Journal of Marketing
Footnotes 1 Dilip Soman
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Grants from the Alfred P. Sloan Foundation (G-2018-11114) and the National Science Foundation (UV-GA11247-155597) supported this research.
4 Kellen Mrkva https://orcid.org/0000-0002-6316-5502 Eric J. Johnson https://orcid.org/0000-0001-7797-8347
5 Online supplement https://doi.org/10.1177/0022242921993186
6 By "choice disparities," we mean large gaps in the decisions of people with low compared with high levels of a trait (e.g., SES). For example, low-SES consumers are much less likely to buy in bulk ([56]).
7 We included all participants in primary analyses because we preregistered that we would include all who finished the study. All significant effects remained significant when including only the 450 requested (i.e., the first 450 to finish; see the Web Appendix).
8 In these first two robustness tests and all other robustness tests in any experiment that involved adding a covariate, we also controlled for the covariate × nudge condition interaction.
9 In each study, ROIRocket provided a larger sample than requested to account for potential dropouts (e.g., 428 rather than 400 in Study 2), though we analyzed all participants who finished the study, as preregistered.
We sampled the good-nudge and control conditions in Study 2 (without bad-nudge conditions), partly because they likely have higher ecological validity (e.g., sorting from worst value to best value in online retail is likely uncommon) and partly because a "bad sort" could make choices easier than a random arrangement of options if consumers notice the pattern.
For example, some participants received two retail purchase questions followed by two consumer financial decisions followed by two consumer sustainability decisions. A Latin square was used to counterbalance the context and nudge type across participants.
The questions used the same wordings as in the supplemental study (see Web Appendix section 3 for the wordings), which were slightly different from the question wordings used in Study 1.
We suspect one reason for this is that the Study 5 sample was much less diverse than the sample in previous studies. Having larger numbers of participants at each end of the SES continuum (as we had in previous studies and especially Study 1) provides much more power to detect effects ([52]).
References Adler Nancy E., Epel Elissa, Castellazzo Grace, Ickovics Jeannette. (2000), "Relationship of Subjective & Objective Social Status with Psychological and Physiological Functioning: Preliminary Data in Healthy, White Women," Health Psychology, 19 (6), 586–92.
Afif Zeina, Islan William W., Calvo-Gonzalez Oscar, Dalton Abigail. (2018), "Behavioral Science Around the World: Profiles of 10 Countries,"eMBeD brief, World Bank Group.
Al Bahrani, Abdullah, Weathers Jamie, Patel Darshak. (2019), "Racial Differences in the Returns to Financial Literacy Education," Journal of Consumer Affairs, 53 (2), 572–99.
Al-Ubaydli O., Lee Min Sok, List John A., Mackevicius Claire L., Suskind Dana. (2020), "How Can Experiments Play a Greater Role in Public Policy? Twelve Proposals from an Economic Model of Scaling," Behavioural Public Policy, 5 (1), 1–48.
Amar Moty, Ariely Dan, Ayal Shahar, Cryder Cynthia E., Rick Scott I. (2011), "Winning the Battle but Losing the War: The Psychology of Debt Management," Journal of Marketing Research, 48 (Special Issue), S38–50.
Bates Douglas, Maechler Martin, Bolker Ben, Walker Steven, Christensen Rune Haubo Bojesen, et al. (2015), "Package 'Lme4,'" Convergence, 12 (1), 2.
Benartzi Shlomo, Beshears John, Milkman Katherine L., Sunstein Cass, Thaler Richard, Shankar Maya. (2017), "Should Governments Invest More in Nudging?" Psychological Science, 28 (8), 1041–55.
Bernheim Douglas. (1998), "Financial Illiteracy, Education and Retirement Saving," Living with Defined Contribution Pensions 3868.
Beshears John, Choi James J., Laibson David, Madrian Brigitte C., Wang Sean Yixiang. (2016), "Who Is Easier to Nudge?" Working Paper 401, NBER.
Bhargava Saurabh, Loewenstein George, Sydnor Justin. (2017), "Choose to Lose: Health Plan Choices from a Menu with Dominated Option," Quarterly Journal of Economics, 132 (3), 1319–72.
Brown-Johnson Cati G., England Lucinda J., Glantz Stanton A., Ling Pamela M. (2014), "Tobacco Industry Marketing to Low Socioeconomic Status Women in the USA," Tobacco Control, 23 (2), e139–46.
Camerer Colin, Issacharoff Samuel, Loewenstein George, O'Donoghue Ted, Rabin Matthew. (2003), "Regulation for Conservatives: Behavioral Economics and the Case for 'Asymmetric Paternalism,'" University of Pennsylvania Law Review, 151 (3), 1211–54.
Cervellon Marie-Cécile, Poujol Juliet F., Tanner J.F.Jr. (2019), "Judging by the Wristwatch: Salespersons' Responses to Status Signals and Stereotypes of Luxury Clients," Journal of Retailing and Consumer Services, 51, 191–201.
Chapman G.B., Liu J. (2009), "Numeracy, Frequency, and Bayesian Reasoning," Judgment and Decision Making, 4 (1), 34–40.
Cheema Amar, Soman Dilip. (2008), "The Effect of Partitions on Controlling Consumption," Journal of Marketing Research, 45 (6), 665–75.
Chernev Alexander, Bockenholt Ulf, Goodman Joseph. (2015), "Choice Overload: A Conceptual Review and Meta-Analysis," Journal of Consumer Psychology, 25 (2), 333–58.
Choi James J., Laibson David, Madrian Brigitte, Metrick Andrew. (2004), "For Better or for Worse: Default Effects and 401(k) Savings Behavior," in Perspectives on the Economics of Aging, Wise David A., ed. Chicago: University of Chicago Press, 81–121.
Choukhmane Taha. (2021), "Default Options and Retirement Saving Dynamics," working paper, Massachusetts Institute of Technology.
Danes Sharon, Huddleston-Casas Catherine, Boyce Laurie. (1999), "Financial Planning Curriculum for Teens," Journal of Financial Counseling and Planning, 10 (1), 26.
Dellaert Benedict G.C., Haubl Gerald. (2012), "Searching in Choice Mode: Consumer Decision Processes in Product Search with Recommendations," Journal of Marketing Research, 49 (2), 277–88.
DellaVigna Stefano, Linos Elizabeth. (2020), "RCTs to Scale: Comprehensive Evidence from Two Nudge Units," working paper, University of California, Berkeley.
Diehl Kristin. (2005), "When Two Rights Make a Wrong: Searching Too Much in Ordered Environments," Journal of Marketing Research, 42 (3), 313–22.
Eisenberg-Guyot Jerzy, Firth Caislin, Klawitter Marieka, Hajat Anjum. (2018), "From Payday Loans to Pawnshops: Fringe Banking, the Unbanked, and Health," Health Affairs, 37 (3), 429–37.
Feitsma Joram Nanne Pieter. (2018), "The Behavioural State: Critical Observations on Technocracy and Psychocracy," Policy Sciences, 51 (3), 387–410.
Fernandes Daniel, Lynch John G.Jr, Netemeyer Richard G., (2014), "Financial Literacy, Financial Education, and Downstream Financial Behaviors," Management Science, 60 (8), 1861–2109.
Fishbane Alissa, Ouss Aurelie, Shah Anuj K. (2020), "Behavioral Nudges Reduce Failure to Appear for Court," Science, 370 (682), 1–10.
Goldstein Daniel G., Johnson Eric J., Herrmann Andreas, Heitmann Mark. (2008), "Nudge Your Customers Toward Better Choices," Harvard Business Review (December), https://hbr.org/2008/12/nudge-your-customers-toward-better-choices.
Graffeo Michele, Polonio Luca, Bonini Nicolao. (2015), "Individual Differences in Competent Consumer Choice: The Role of Cognitive Reflection and Numeracy Skills," Frontiers in Psychology, 6, 844.
Hadar Liat, Sood Sanjay, Fox Craig R. (2013), "Subjective Knowledge in Consumer Financial Decisions," Journal of Marketing Research, 50 (3), 303–16.
Hansen P.G., Jespersen A.M. (2013), "Nudge and the Manipulation of Choice: A Framework for the Responsible Use of the Nudge Approach to Behavior Change in Public Policy," European Journal of Risk Regulation, 4 (1), 3–28.
Hershfield Hal E., Shu Stephen, Benartzi Shlomo. (2020), "Temporal Reframing and Participation in a Savings Program: A Field Experiment," Marketing Science, 39 (6), 1039–51.
Hilgert Marianne A., Hogarth Jeanne M., Beverly Sondra G. (2003), "Household Financial Management: The Connection Between Knowledge and Behavior," Federal Reserve Bulletin, 89 (7), 309–22.
Hill Ronald Paul. (1995), "Researching Sensitive Topics in Marketing: The Special Case of Vulnerable Populations," Journal of Public Policy & Marketing, 14 (1), 143–48.
Hill Ronald Paul, Sharma Eesha. (2020), "Consumer Vulnerability," Journal of Consumer Psychology, 30 (3), 551–70.
Hoeffler Steve, Ariely Dan. (1999), "Constructing Stable Preferences: A Look into Dimensions of Experience and Their Impact on Preference Stability," Journal of Consumer Psychology, 8 (2), 113–39.
Huh Young E., Vosgerau Joachim, Morewedge Carey K. (2014), "Social Defaults: Observed Choices Become Choice Defaults," Journal of Consumer Research, 41 (3), 746–60.
Hutchinson J. Wesley, Alba Joseph W. (1991), "Ignoring Irrelevant Information: Situational Determinants of Consumer Learning." Journal of Consumer Research, 18 (3), 325–45.
Johnson Eric J., Bellman Steven, Lohse Gerald L. (2002), "Defaults, Framing and Privacy: Why Opting In ≠ Opting Out," Marketing Letters, 13 (1), 5–15.
Johnson Eric J., Goldstein Daniel. (2003), "Do Defaults Save Lives?" Science, 302 (5649), 1338–39.
Johnson Eric J., Shu Suzzane B., Dellaert Benedict G.C., Wansink Brian. (2012), "Beyond Nudges: Tools of a Choice Architecture," Marketing Letters, 23 (2), 487–504.
Kotler Philip, Hessekiel David, Lee Nancy. (2012), Good Works! Marketing and Corporate Initiatives that Build a Better World and the Bottom Line. Hoboken, NJ: John Wiley & Sons.
Lamberton Cait Poynor, Diehl Kristin. (2013), "Retail Choice Architecture: The Effects of Benefit-and Attribute-Based Assortment Organization on Consumer Perceptions and Choice," Journal of Consumer Research, 40 (3), 393–411.
Langenderfer Jeff, Shimp Terence A. (2001), "Consumer Vulnerability to Scams, Swindles, and Fraud: A New Theory of Visceral Influences on Persuasion," Psychology & Marketing18 (7), 763–83.
Lipkus Isaac M., Samsa Greg, Rimer Barbara K. (2001), "General Performance on a Numeracy Scale Among Highly Educated Samples," Society for Medical Decision Making, 21 (1), 37–44.
Loewenstein George, Chater Nick. (2017), "Putting Nudges in Perspective," Behavioural Public Policy, 1 (1), 26–53.
Löfgren Åsa, Martinsson Peter, Hennlock Magnus, Sterner Thomas. (2012), "Are Experienced People Affected by a Pre-Set Default Option—Results from a Field Experiment," Journal of Environmental Economics and Management, 63 (1), 66–72.
Lusardi Annamaria. (2008), "Financial Literacy: An Essential Tool for Informed Consumer Choice?" Working Paper No. 14084, NBER.
Lusardi Annamaria, Michaud Pierre-Carl, Mitchell Olivia S. (2013), "Optimal Financial Knowledge and Wealth Inequality," Working Paper No. 18669, NBER.
Lynch John G.Jr, Ariely Dan. (2000), "Wine Online: Search Costs Affect Competition on Price, Quality, and Distribution," Marketing Science, 19 (1), 1–104.
Mathur Arunesh, Acar Gunes, Friedman Michael J., Lucherini Elena, Mayer Jonathan, Chetty Marshini, Narayanan Arvind. (2019), "Dark Patterns at Scale: Findings from a Crawl of 11K Shopping Websites," in Proceedings of the ACM on Human-Computer Interaction, Vol. 19 (CSCW), 1–32.
McArdle John J. (2015), "Adaptive Testing in Aging Populations," in The Encyclopedia of Adulthood and Aging, Whitbourne Susan Krauss, ed. Chichester, UK: John Wiley & Sons, 1–6.
McClelland Gary H. (2000), "Increasing Statistical Power Without Increasing Sample Size," American Psychologist, 55 (8), 963–64.
Mitchell Vincent-Wayne, Lennard David, McGoldrick Peter. (2003), "Consumer Awareness, Understanding and Usage of Unit Pricing," British Journal of Management, 14 (2), 173.
Mrkva Kellen, Johnson Eric J., Gächter Simon, Herrmann Andreas. (2020), "Moderating Loss Aversion: Loss Aversion Has Moderators, but Reports of Its Death Are Greatly Exaggerated," Journal of Consumer Psychology, 30 (3), 407–28.
Netemeyer Richard G., Warmath Dee, Fernandes Daniel, Lynch John G.Jr,. (2018), "How Am I Doing? Perceived Financial Well-Being, Its Potential Antecedents, and Its Relation to Overall Well-Being," Journal of Consumer Research, 45 (1), 68–89.
Orhun A. Yeşim, Palazzolo Mike. (2019), "Frugality Is Hard to Afford," Journal of Marketing Research, 56 (1), 1–17.
Peters Ellen, Bjalkebring Par. (2015), "Multiple Numeric Competencies: When a Number Is Not Just a Number," Journal of Personality and Social Psychology, 108 (5), 802–22.
Peters Ellen, Tompkins Mary Kate, Knoll Melissa A.Z., Ardoin Stacy P., Shoots-Reinhard Brittany, Meara Alexa S. (2019), "Despite High Objective Numeracy, Lower Numeric Confidence Relates to Worse Financial and Medical Outcomes," PNAS, 116 (39), 19386–391.
Peters Ellen, Västfjäll Daniel, Slovic Paul, Mertz C.K., Mazzocco Ketti, Dickert Stephan. (2006), "Numeracy and Decision Making,"Psychological Science, 17 (5), 407–13.
Preacher Kristopher J., Hayes Andrew F. (2008), "Asymptotic and Resampling Strategies for Assessing and Comparing Indirect Effects in Multiple Mediator Models," Behavior Research Methods, 40 (3), 879–91.
Roberts Jessica L. (2018), "Nudge-Proof: Distributive Justice and the Ethics of Nudging," Michigan Law Review, 116 (6), 1045–66.
Saegert Susan C., Adler Nancy E., Bullock Heather E., Cauce Ana Mari, Liu William Ming, Wyche Karen F. (2006), "APA Task Force on Socioeconomic Status (SES)," American Psychological Association, research report, https://www.apa.org/pi/ses/resources/publications/task-force-2006.pdf.
Santana Shelle, Thomas Manoj, Morwitz Vicki G. (2020), "The Role of Numbers in the Customer Journey," Journal of Retailing, 96 (1), 138–54.
Scheibehenne Benjamin, Greifeneder Rainer, Todd Peter. (2010), "Can There Ever Be Too Many Options? A Meta-Analytic Review of Choice Overload," Journal of Consumer Research, 37 (3), 409–25.
Schwartz Lisa M., Woloshin Steven, Black William C., Gilbert Welch H. (1997), "The Role of Numeracy in Understanding the Benefit of Screening Mammography," Annals of Internal Medicine, 127 (11), 966–72.
Schwartz Mark S. (2005), "Universal Moral Values for Corporate Codes of Ethics," Journal of Business Ethics, 59 (1/2), 27–44.
Sela Aner, Berger Jonah, Liu Wendy. (2008), "Variety, Vice, and Virtue: How Assortment Size Influences Option Choice," Journal of Consumer Research, 35 (6), 941–51.
Sengupta Jaideep, Johar Gita V. (2001), "Contingent Effects of Anxiety on Message Elaboration and Persuasion," Personality and Social Psychology Bulletin, 27 (2), 139–50.
Shah Anuj K., Mullainathan Sendhil, Shafir Eldar. (2012), "Some Consequences of Having Too Little," Science, 338 (6107), 682–85.
Sharif Marissa A., Shu Suzanne B. (2017), "The Benefits of Emergency Reserves: Greater Preference and Persistence for Goals That Have Slack with a Cost," Journal of Marketing Research, 54 (3), 495–509.
Skagerlund Kenny, Lind Therese, Stromback Camilla, Tinghog Gustav, Västfjäll Daniel. (2018), "Financial Literacy and the Role of Numeracy: How Individuals' Attitude and Affinity with Numbers Influence Financial Literacy," Journal of Behavioral and Experimental Economics, 74, 18–25.
Soman Dilip. (2015), The Last Mile: Creating Social and Economic Value from Behavioral Insights. Toronto: University of Toronto Press.
Soman Dilip, Cowen Daniel, Kannan Niketana, Feng Bing. (2019), "Seeing Sludge: Towards a Dashboard to Help Organizations Recognize Impedance to End-User Decisions and Action," Research Report Series: Behavioural Economics in Action at Rotman, https://ssrn.com/abstract=3460734.
Soman Dilip, Hossain Tanjim. (2020), "Successfully Scaled Solutions Need Not Be Homogenous," Behavioural Public Policy, 5 (1), 1–10.
Thaler Richard H., Sunstein Cass R. (2009), Nudge: Improving Decisions About Health, Wealth, and Happiness. London: Penguin.
Tine Michele, Gotlieb Rebecca. (2013), "Gender-, Race-, and Income-Based Stereotype Threat: The Effects of Multiple Stigmatized Aspects of Identity on Math Performance and Working Memory Function," Social Psychology of Education, 16 (3), 353–76.
~~~~~~~~
By Kellen Mrkva; Nathaniel A. Posner; Crystal Reeck and Eric J. Johnson
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 46- Do Promotions Make Consumers More Generous? The Impact of Price Promotions on Consumers' Donation Behavior. By: Zhang, Kuangjie; Cai, Fengyan; Shi, Zhengyu. Journal of Marketing. May2021, Vol. 85 Issue 3, p240-255. 16p. 1 Diagram, 2 Charts, 1 Graph. DOI: 10.1177/0022242920988253.
- Database:
- Business Source Complete
Do Promotions Make Consumers More Generous? The Impact of Price Promotions on Consumers' Donation Behavior
Despite growing concerns regarding the increasing consumerism related to promotions, this research documents a positive effect of price promotions on consumers' donation behavior. Specifically, the authors propose that price promotions increase consumers' perceived resources, which in turn increase consumers' donation behavior. A series of seven studies, combining field and experimental data, provide converging support for this proposition and its underlying mechanism of perceived resources. Furthermore, the authors show that the positive effect of price promotions on consumers' donation behavior is attenuated when consumers focus on the amount of money spent (rather than saved), when consumers feel they have overspent their budget, and when the monetary savings cannot be realized immediately. Finally, the authors show that this effect is stronger when donation solicitation occurs immediately after the price promotion (vs. after a delay). This research documents a novel behavioral consequence of price promotions and uncovers a mechanism by which price promotions can lead to positive social consequences and contribute to a better world.
Keywords: donation behavior; perceived resources; price promotion
As an effective strategy for attracting consumers and increasing sales, price promotions are undoubtedly one of the most important marketing tools. In the last decade, price promotion events, such as Black Friday and Cyber Monday in Western countries as well as Alibaba's Double 11 Singles' Day in China, have become increasingly popular and yielded record-setting sales for the companies involved. For example, in 2019, U.S. consumers spent $7.4 billion on Black Friday and $9.4 billion on Cyber Monday ([35]). In the same year, the Chinese e-commerce giant Alibaba achieved a sales volume of more than $38.3 billion on Singles' Day alone ([33]).
Despite the large sales volume achieved in price promotion events, some criticize the events for promoting consumerism and materialism. For example, the media has portrayed Black Friday as "America's greediest holiday" ([25]). As a result, several retailers and organizations call for boycotts on Black Friday and Cyber Monday sales ([22]), arguing that Black Friday not only ruins the spirit of Thanksgiving but also decreases social welfare by making consumers stingier and more selfish. An important question thus arises: Can price promotions lead to any positive social consequences? In the current research, we aim to address this question in the context of donation behavior.
While a large body of research has examined the effect of price promotions on firm performance and consumers' purchasing behaviors, the existing research is relatively silent on whether and how price promotions have important social consequences. In this research, we demonstrate that price promotions actually can have a positive impact on consumers' donation behavior via an increase in consumers' perceived resources. A series of seven studies, combining field and experimental data, offer converging support for this proposition. Providing support for the perceived resources mechanism, we show that the effect of price promotions on donation behavior is moderated by whether consumers focus on the money spent rather than saved, whether consumers feel they have overspent their budget, and whether the monetary savings can be realized immediately. Furthermore, we show that the increase in perceived resources dissipates over time, and thus, the positive effect of price promotions on donation behavior is stronger when donations are solicited immediately after the price promotion. We also demonstrate the external validity of this effect in two field surveys among actual shoppers as well as two field experiments.
This research makes important theoretical and practical contributions. From a theoretical perspective, our findings contribute to the price promotion literature by shedding light on the positive social consequences of price promotions. To the best of our knowledge, this is the first research to examine how price promotions can facilitate consumers' donation behavior. While the cause-related marketing literature has studied promotions and donation behavior together, it has focused narrowly on charitable donations as the promotion (e.g., [56]; [59]). By contrast, the present research focuses on donation behavior as a consequence of promotions. Furthermore, while researchers have examined various factors that influence consumers' donation behavior, price promotion (to the best of our knowledge) has never been considered one of those factors. This research thus adds to the donation behavior literature by identifying price promotion as a novel situational factor that can drive consumers' donation behavior.
From a practical perspective, this research provides pertinent and actionable implications for both charitable organizations and firms. Charitable organizations can optimize their campaigns by choosing strategic targets and timing for donation solicitations—specifically, charities should target consumers who are taking part in promotions (because this group of consumers is easy to identify and is more likely to donate than the general population) and should solicit donations immediately after a price promotion event. For firms, price promotions may be great opportunities to raise funds for charitable causes, in the spirit of corporate social responsibility, by soliciting donations right after consumers have made their purchases. In traditional cause-related marketing practices, consumers might doubt a firm's prosocial motivation because its charitable donations depend on whether consumers purchase the products ([18]; [24]). Our findings suggest that firms can overcome this negative inference and enhance their corporate social responsibility image by soliciting donations after consumers make their purchases.
In the following section, we review the relevant literature and derive our hypotheses for the mechanism by which price promotions can influence consumers' donation behavior. Finally, we articulate implications of our findings in the "General Discussion" section.
The effects of price promotions have been studied extensively from both the marketer and consumer perspectives. From the marketers' perspective, the most obvious benefit of a price promotion is that it can increase sales ([11]). This increase of sales caused by price promotions is attributed to two major mechanisms: purchase acceleration, in which consumers purchase promoted products sooner or in larger quantities, and brand switching, in which consumers switch to promoted brands of higher quality ([ 8]). However, price promotions may also entail potential long-term risks for firms. Specifically, stockpiling may preempt future sales ([ 1]); frequent promotion events may increase consumers' price sensitivity ([45]) and may also undermine firms' brand equity ([65]).
From the consumers' perspective, they can benefit from price promotions in several ways. First, consumers can derive utilitarian benefits, such as monetary savings and the opportunity to upgrade to higher-quality products ([11]; [15]). Second, consumers can derive hedonic benefits, such as happiness or enjoyment, from saving money ([ 6]; [15]; [46]). Third, by participating in price promotions, consumers may construct a positive self-perception as a smart shopper ([51]).
Price promotions may also induce certain negative responses from consumers. For example, consumers may make negative inferences about a discounted product (e.g., that it must be low-quality; [17]; [48]). As a result, some consumers might be reluctant to make purchases during price promotions ([ 5]). In line with this notion, [12] showed that the sales volumes of indulgent products (e.g., ice cream) actually decrease when there is a small price discount (e.g., <10% off). In addition, prior research suggests that price promotions can reduce perceived product efficacy ([53]) and negatively influence consumers' enjoyment of postpurchase consumption ([40]). Moreover, a price promotion can prompt unplanned or impulsive purchases ([31]), which can induce negative emotions such as guilt ([46]). More recently, [52] showed that mere exposure to price promotions can cause consumers to act more impatiently in unrelated domains.
Although robust research has studied the impact of price promotions on firms' performances, consumers' purchasing behaviors, and consumption experiences, there is a gap in the literature on the social consequences of price promotions. In this research, we aim to address this gap by examining whether and how price promotions influence consumers' donation behavior.
Prior research has suggested that consumers' donation behavior can be driven by both individual and situational antecedents. In terms of individual factors, moral identity (e.g., [41]; [49]; [63]), gender identity ([64]), and self-construal ([21]) have been studied as important predictors of donation behavior.
In addition, prior research has examined various situational factors that influence donation behavior, such as positive mood (e.g., [32]; [44]), discrete emotions such as guilt and love (e.g., [ 7]; [14]), mortality salience (e.g., [13]), and social norms (e.g., [60]). In the current research, we contribute to the donation behavior literature by identifying price promotion as a novel situational antecedent of donation behavior.
In this research, we examine the effect of price promotions on consumers' donation behavior. Specifically, we propose that price promotions can increase consumers' perceived resources, and the perception of greater resources, in turn, can increase consumers' donation behavior. We elaborate on our conceptual framework to derive our hypotheses in the following subsection.
In this research, we define "perceived resources" as consumers' perception of their current monetary resources. Such a perception can be influenced by both objective factors (e.g., one's actual monetary resources) and subjective factors (e.g., the subjective experience of saving money). Price promotions can work on both levels.
First, price promotions can offer actual monetary savings. If consumers have already mentally budgeted their purchase expenses, then a price promotion (e.g., a discount on the promoted product or an offer for more of the same product for free) reduces consumers' expenses, thus freeing up monetary resources for other purposes ([11]; [15]). This is especially true for unexpected promotions (e.g., surprise coupons). In line with this notion, prior research suggests that consumers perceive unexpected savings from buying products on sale as windfall gains ([ 4]; [31]). This phenomenon is also referred to as a "psychological income effect" ([31]). Thus, the actual monetary savings from price promotions should prompt consumers to perceive an increase in resources.
Second, price promotions can increase consumers' subjective experience of saving money, which is derived from a comparison with a reference price ([27]; [43]; [57]). For example, transaction utility theory ([57]) suggests that consumers gain transaction utility when they compare the price they pay with a reference price. In price promotions, the regular price is usually offered as a reference price, causing consumers to perceive that they are saving money. Consistent with transaction utility theory, prior research has shown that consumers perceive savings in price promotions when they compare the discounted price with the higher original price ([10]). Various marketing efforts (e.g., strategic display of a higher manufacturer's suggested price) can also increase consumers' perceived savings in price promotions ([37]). In no-promotion situations, in contrast, an external reference price is usually not available, so there are no salient perceived savings. Therefore, compared with no-promotion situations, participating in price promotions can make consumers believe they are saving money. This subjective experience of saving money can also increase consumers' perceived resources. Synthesizing these two factors, we propose that price promotions can boost consumers' perceived resources.
The amount of resources that consumers have available to allocate is an important driver of donation behavior ([ 9]; [20]; [54]). Because donation behavior requires consumers to direct resources away from the self and toward others, consumers should be more (vs. less) likely to engage in donation behavior when their actual resources are abundant (vs. scarce). In line with this notion, [36] conducted a large-scale test and found an overall positive relationship between social class and donation behavior. Similarly, [ 3] used a field experiment to show that rich (vs. poor) households are more likely to behave prosocially by returning "misdelivered" envelopes. Nevertheless, the literature contains mixed empirical evidence on how a consumer's actual resources affect their donation behavior. For example, [47] found that people from lower socioeconomic classes or with less power are more motivated to help others in need.
In this research, we focus on the amount of perceived (not actual) resources, which has a more established positive relationship with donation behavior ([28]; [50]; [62]). For example, [28] and [62] showed that, regardless of actual financial resources, people who perceive their financial situation as more abundant (vs. scarce) are more generous in their donations. In another stream of literature on the effects of resource scarcity on consumer behavior, [50] showed that perceived resource scarcity can trigger a competitive orientation and reduce consumers' likelihood of donating to charities. Similarly, [42] suggested that a perceived resource deficiency causes consumers to donate less to charities. These findings lead to our prediction that an increase in perceived resources should promote donation behavior.
Note that the donation behavior literature distinguishes between donation rate and donation amount. The existing literature on the relationship between resources and donation behavior has documented a consistently positive impact of resources on donation amount and an inconsistent impact of resources on donation rate (for a review, see [61]]). In the current research, we examine donation behavior by analyzing both the donation rate and donation amount.
In this research, we define perceived resources as consumers' perception of their current monetary resources. Building on the aforementioned discussion, we propose that price promotions can increase consumers' perceived resources, and the greater perceived resources, in turn, increase consumers' donation behavior. More formally,
- H1: Price promotions increase consumers' donation behavior.
- H2: The positive effect of price promotions on donation behavior is mediated by an increase in perceived resources.
As discussed previously, the positive effect of price promotions on perceived resources can come from two sources: actual monetary savings and a subjective experience of saving money. The relative importance of these two driving forces may differ across situations. Specifically, consumers experience actual monetary savings when they have already decided to make the purchase and the promotion comes as a surprise. In such cases, consumers perceive the saved money as windfall gains. However, the subjective experience of saving money depends on the comparison with a salient reference price. Thus, when a promotion is unexpected, consumers may experience a greater increase in perceived resources due to both actual monetary savings and a subjective experience of saving money. By contrast, when a promotion is expected, consumers do not benefit objectively from actual monetary savings, but they may still experience a subjective experience of saving money. In the current research, we examine both expected and unexpected promotions. Furthermore, because the definition of perceived resources in this research pertains to consumers' perception of their current monetary resources, consumers are less likely to experience an increase in their perceived resources when the monetary savings from the promotion cannot be realized immediately.
Thus, we propose that the positive effect of price promotions on consumers' donation behavior should be attenuated ( 1) when consumers are made to think about how much money they have spent rather than saved, ( 2) when consumers feel they have overspent their budget, and ( 3) when the monetary savings from the promotion cannot be realized immediately. Next, we discuss how each of these factors influences consumers' perceived resources in more detail. Of course, there are probably other conceptually relevant moderators that we do not explore here, and we encourage future research to test other possibilities.
Consumers have greater perceived resources after participating in price promotions because monetary savings from promotions are often more salient to consumers ([10]; [37]). Following this logic, the salience of monetary savings should be disrupted if consumers are prompted to focus on the money they have spent on their purchases. In such cases, consumers' perceived resources are less likely to increase because the subjective experience of saving money is weaker. Instead, consumers may realize that their current monetary resources have been diminished by the recent purchases.
- H3: The positive effect of price promotions on consumers' donation behavior is attenuated when consumers are made to think about how much money they have spent (vs. saved) on their purchases.
As discussed previously, consumers' subjective experience of saving money in price promotions depends on the comparison with a higher reference price (e.g., regular price). However, such subjective perception can be influenced by a different reference price: mental budget. Specifically, consumers tend to track their expenses against their mental budget ([30]; [55]). When consumers feel that they have overspent their budget, their subjective experience of saving money is reduced as they realize that they have spent more money than they should. In such situations, their perceived resources are less likely to increase.
- H4: The positive effect of price promotions on donation behavior is attenuated when consumers feel that they have (vs. have not) overspent their budget.
As we have mentioned, the monetary savings from promotions cause consumers to perceive an increase in their current monetary resources. However, the savings from price promotions are not always immediate. For example, stores often issue rebates or rebate coupons that consumers cannot use until their next purchase. In such situations, consumers' current expenditures are unaffected by the promotion, so their perceived resources are unlikely to increase until they realize those monetary savings in the next purchase. Therefore, the positive effect of price promotions on consumers' donation behavior should be attenuated when the monetary savings cannot be realized immediately.
- H5: The positive effect of price promotions on donation behavior is attenuated when the monetary savings cannot be realized immediately (e.g., future rebates).
In addition to the aforementioned three moderators that influence the relationship between price promotions and consumers' perceived resources, we identify another managerially relevant moderator that can influence the relationship between perceived resources and donation behavior.
After consumers participate in a price promotion, their increase in perceived monetary resources—and subsequent increase in donation behavior—should diminish over time for two main reasons. First, if consumers actually saved money in the promotion, they may spend that money on other consumption alternatives in the near future. Second, consumers may gradually adapt to the subjective experience of saving money as time passes after the promotion event ([26]; [58]). Thus, the positive impact of the increase in perceived resources on donation behavior should dissipate with a delay between the price promotion event and donation solicitation.
- H6: The positive effect of price promotions on donation behavior is stronger when the donation is solicited immediately after the price promotion and is attenuated when there is a delay.
We tested our hypotheses in seven studies (for an overview of our conceptual framework and studies, see Figure 1). Study 1 provided initial field evidence for the positive effect of price promotions on consumers' donation behavior (H1). Study 2 manipulated the presence as well as the magnitude of price promotions, provided causal evidence, and offered process support by examining the mediating role of perceived resources (H2). Study 3 provided further causal evidence in a field experiment (H1). Study 4 provided further support for the perceived resources mechanism by examining whether the effect is attenuated when consumers focus on the amount of money they spent in the promotion (H3). Study 5 examined the moderating role of budget overspending (H4). Study 6, another field experiment, examined the moderating role of the immediacy of the savings (H5). Finally, Study 7 tested the moderating role of the time interval between the price promotion and donation solicitation (H6). Across studies, we applied consistent outlier exclusion criteria (i.e., ±3 SD from the mean) in our analyses and clearly stated any additional exclusion criteria in each study (for exclusion details across studies, see Web Appendix 7). Studies 2 and 5 also examined the impact of price promotions on consumers' feelings (for supplementary analyses on these measures, see Web Appendix 5). Across the seven studies, using both field and experimental data, our findings converged to establish the robustness and external validity of the effect and to provide support for the underlying mechanism of perceived resources.
Graph: Figure 1. Overview of conceptual framework and studies.
Alibaba's Double 11 sales event is the largest price promotion event in China, and its online retail website, Tmall, offers big price promotions on November 11 (hence the name "Double 11"). In Study 1, we used this opportunity to collect field evidence on the impact of price promotions on consumers' charitable behaviors. Actual donation data from two charitable organizations provided preliminary support for our hypothesis (see Web Appendix 1). To further examine the impact of the Double 11 sales event on charitable behaviors, we surveyed individual shoppers who participated and examined whether their spending during the Double 11 sales event positively predicted their charitable behaviors in the subsequent week.
One hundred thirty-five shoppers who participated in the 2017 Alibaba's Double 11 sales event (44.4% female; all ≥18 years old) were recruited from an online panel in China on November 18 to complete this study in exchange for monetary compensation. Shoppers were told that the study aimed to understand consumers' shopping experience, and they completed a few ostensibly unrelated surveys. In the first survey, they were asked to indicate how much money (in RMB) they actually spent on their purchases during the sales event on a 22-point scale (1 = 0–200, 2 = 201–500,..., 21 = 40,001–50,000, and 22 = >50,000). Next, shoppers were asked to rate how much money they perceived they had saved during the sales event on a seven-point scale (1 = "very little money saved," and 7 = "a lot of money saved").
The second survey ostensibly sought to understand the shoppers' daily life experiences in the one week preceding the survey (i.e., the week after the sales event). Specifically, shoppers were asked whether they had engaged in each of 13 types of behavior in the week of November 12–18 (1 = yes, 0 = no; for the list of behaviors, see the Appendix). The responses to these 13 items were summed to form a composite charitable behaviors index. Shoppers were also asked whether they would be interested in engaging in those behaviors in the future. The responses to these 13 items were averaged to form a behavioral intention index (α =.92). Finally, shoppers completed basic demographic questions including their age range and gender; these measures were included in all subsequent studies and will not be mentioned again.
A linear regression analysis revealed that the amount spent during the sales event had a significant, positive effect on perceived savings (b =.11, t(133) = 3.20, p =.002). In other words, the more money shoppers spent during the Double 11 sales event, the more money they perceived they had saved.
In a linear regression, the amount spent during the event had a significant positive correlation with engagement in charitable behaviors (b =.26, t(133) = 2.89, p =.004). A mediation analysis ([29], PROCESS Model 4 with 5,000 bootstrap samples) revealed that perceived savings mediated this effect (indirect effect =.11, SE =.05, 95% confidence interval [CI] = [.03,.22]).
Similarly, a linear regression analysis revealed a significant positive correlation between the amount spent during the Double 11 sales event and shoppers' intention to engage in charitable behaviors in the future (b =.07, t(133) = 3.29, p =.001). Again, a mediation analysis ([29], PROCESS Model 4 with 5,000 bootstrap samples) revealed that perceived savings mediated the effect (indirect effect =.02, SE =.01, 95% CI = [.01,.05]).
As a robustness check, we conducted the same analyses with age and gender as covariates. Notably, controlling for age and gender did not alter the interpretation or level of significance of our results in this or any of the subsequent studies.
In summary, Study 1 found that shoppers' spending during the Double 11 sales event positively predicted both their charitable behaviors in the week following the event and their intention to engage in charitable behaviors in the future. Furthermore, perceived savings from the price promotions mediated the effect of shoppers' spending on their charitable behaviors. One limitation of such field data is that it can provide only correlational evidence. For example, one could argue that shoppers who spent more during the Double 11 sales event happened to be more interested in various forms of charitable behaviors in general. In the next studies, we intend to provide causal evidence for the effect of price promotions on consumers' charitable behavior.
In Study 2, we aimed to provide causal evidence for the impact of price promotions on consumers' donation behavior by manipulating not only the presence but also the magnitude of the price promotion. Furthermore, we examined the mediating role of perceived resources.
Two hundred participants in the United States (37.5% female; Mage = 34.7 years) were recruited from Amazon's Mechanical Turk to complete this study in exchange for monetary compensation. Participants were randomly assigned to one of four conditions (no promotion vs. 10% off vs. 50% off vs. 50% off with double spending).
All participants were asked to imagine that they were visiting a shopping mall. In the 50%-off (10%-off) condition, participants were told that the shopping mall was having a 50% off (10% off) sales event, and they spent $500 on purchases that originally cost $1,000 ($560). In the no-promotion (control) condition, participants were simply told that they spent $500 on purchases. Finally, in the 50%-off-with-double-spending condition, participants were told that they spent $1,000 on purchases that originally cost $2,000. Because participants in the 50%-off-with-double-spending condition saved the most money, we expected that they would experience even greater perceived resources and hence would donate more to charities, relative to participants in the 10%-off and 50%-off conditions.
Subsequently, all participants were told that as they walked out of the shopping mall, they noticed that United Way was raising money for the Hurricane Florence Relief Fund to support local communities in South Carolina, North Carolina, Virginia, and the surrounding areas affected by Hurricane Florence. Participants indicated the amount they were willing to donate to this charity fund. Next, participants responded to three items regarding their perceived resources after making their purchases in the shopping mall: ("After making the purchase, I feel that I have saved a lot of money," "After making the purchase, I feel that I have more resources at hand," and "After making the purchase, I feel that my resources are sufficient"; 1 = "strongly disagree," and 7 = "strongly agree"). Responses to these three statements were averaged to form a perceived resources index (α =.88). Finally, participants responded to two items regarding whether Hurricane Florence had affected them personally and whether it had affected their family and/or close friends (1 = "not at all," and 7 = "very much"); controlling for these two items in our subsequent analyses did not alter the interpretation or level of significance of the results (reported next).
To better understand the donation behavior behaviors in our experiments, we analyze both the donation rate and the average donation amount in each condition. In line with prior research (e.g., [63]; [64]), we report the average donation amount across all participants (i.e., including those who indicated a zero donation amount) in the article. For completeness, we also report the average donation amount among only those who donated (i.e., excluding those who indicated a zero donation amount) in Web Appendix 6.
The percentage of participants who were willing to make a donation (i.e., who indicated a nonzero donation amount) did not significantly differ across conditions (Wald χ2( 3) = 3.73, p =.29). Participants in the no-promotion condition (65.31%) exhibited a directionally lower likelihood of donating than participants in the 10%-off condition (76.00%), 50%-off condition (66.67%), and 50%-off-with-double-spending condition (80.00%). The difference between the no-promotion condition (65.31%) and the pooled promotion conditions (74.17%) was not significant (Wald χ2( 1) = 1.57, p =.21).
Given the large variance in the donation amount, we first identified and removed two outliers that were three standard deviations above or below the mean. In addition, the donation amount was positively skewed (skewness = 2.56, SE =.17). Thus, we log-transformed the donation amount after adding 1 to each score to include zeros in the analysis. The pattern of results remained the same regardless of whether we log-transformed the donation amount; for ease of interpretation, we report the untransformed means. An analysis of variance (ANOVA) revealed a significant main effect of the price promotion (F( 3, 194) = 3.88, p =.01, =.06), and a trend analysis confirmed a linear trend (F( 1, 194) = 10.94, p =.001, =.05). As Figure 2 shows, participants in the 50%-off condition (M50% = $14.92, SD = $22.09) indicated a greater donation amount than participants in the no-promotion condition (Mno promo = $7.54, SD = $11.44; F( 1, 194) = 2.89, p =.09, =.02). More interestingly, participants in the 50%-off-with-double-spending condition indicated an even greater donation amount (M50%, double spend = $23.86, SD = $30.29) than both participants in the 50%-off condition (M50% = $14.92, SD = $22.09; F( 1, 194) = 2.99, p =.09, =.02) and participants in the no-promotion condition (Mno promo = $7.54, SD = $11.44; F( 1, 194) = 11.52, p <.001, =.06). The donation amount in the 10%-off condition (M10% = $11.37, SD = $13.52) fell in the middle and did not differ significantly from the no-promotion condition (F( 1, 194) = 2.01, p =.16, =.01), though the pattern was consistent with our expectation. Furthermore, a planned contrast between the no-promotion condition and the pooled promotion conditions was significant (Mno promo = $7.54, SD = $11.44 vs. Mpromo pooled = $16.74, SD = $23.49; F( 1, 194) = 6.99, p =.01, =.03).
Graph: Figure 2. Donation amount and perceived resources as a function of price promotion (Study 2).†p <.10.**p <.01.Notes: Error bars denote ±1 SE.
An ANOVA on the perceived resources index also revealed a significant main effect of the price promotion (F( 3, 194) = 29.51, p <.001, =.31), and a trend analysis confirmed a linear trend (F( 1, 194) = 79.81, p <.001, =.29). As shown in Figure 2, participants in the 50%-off condition (M50% = 4.82, SD = 1.76) reported greater perceived resources than participants in the no-promotion condition (Mno promo = 2.57, SD = 1.23; F( 1, 194) = 53.78, p <.001, =.22). Interestingly, participants in the 50%-off-with-double-spending condition (M50%, double spend = 5.29, SD = 1.47) reported even greater perceived resources than both participants in the 50%-off condition (M50% = 4.82, SD = 1.59; F( 1, 194) = 2.39, p =.12, =.01) and participants in the no-promotion condition (Mno promo = 2.57, SD = 1.23; F( 1, 194) = 77.79, p <.001, =.29). Participants in the 10%-off condition (M10% = 4.28, SD = 1.59) also reported greater perceived resources than participants in the no-promotion condition (Mno promo = 2.57, SD = 1.23; F( 1, 194) = 30.33, p <.001, =.14). Furthermore, a planned contrast between the no-promotion condition and the pooled promotion conditions was also significant (F1, 194; Mno promo = 2.57, SD = 1.23 vs. Mpromo pooled = 4.80, SD = 1.65) = 77.35, p <.001, =.29).
A mediation analysis ([29], PROCESS Model 4 with 5,000 bootstrap samples) on log-transformed donation amounts revealed that perceived resources significantly mediated the effect of the price promotion (1 = promotion, 0 = no promotion) on participants' donation behavior (indirect effect =.61, SE =.15, 95% CI = [.32,.93]). After controlling for the indirect effect, the direct effect of the price promotion on donation behavior was no longer significant (direct effect =.01, SE =.26, 95% CI = [−.51,.53]). A mediation analysis with price promotion as a multicategorical variable provided consistent support for the mediating role of perceived resources (for detailed results, see Table 1).
Graph
Table 1. Mediation Effect of Perceived Resources (Study 2).
| Omnibus Test of Direct Effect | F(3, 193) =.66, p =.57 |
|---|
| Condition(Baseline: No Promotion) | Direct Effect[95% CI] | Indirect Effect[95% CI] |
|---|
| 10% off | [−.62,.55] | [.20,.74] |
| 50% off | [−.71,.51] | [.28,.92] |
| 50% off with double spending | [−.38,.91] | [.33, 1.11] |
In Study 2, participants in the price promotion conditions (relative to the no-promotion condition) made a greater average donation to a charitable organization, and this positive effect increased with the magnitude of the monetary savings. Interestingly, participants in the 50%-off-with-double-spending condition reported the highest level of perceived resources—perhaps driven by both the perception that they had more money to spend and their greater monetary savings—and these participants also indicated the greatest average donation amount. Furthermore, we showed that perceived resources mediated the effect of the price promotion on donation behavior. In this study (as well as in Study 5), we also included some affective measures to examine the role of feelings such as happiness and guilt. Unlike perceived resources, these affective measures did not directly mediate the effect of price promotions on donation behavior (for supplementary analyses on these measures, see Web Appendix 5).
Study 3 used a field experiment to examine the causal impact of price promotions on donation behavior in a more realistic setting. We conducted the field experiment in collaboration with a café, where we randomly distributed discount coupons to customers. We predicted that customers who received (vs. did not receive) a discount coupon would make more donation behavior to a charitable cause.
We conducted the field experiment in a café in a large city in China in June 2020. Because there were no queues at the cashier counter of this café, we could use random assignment without the customers noticing. The field experiment followed a two-cell (promotion vs. no promotion) between-subjects design, and we introduced the promotion manipulation after customers gave their orders to the cashier. To avoid potential data contamination, we did not include subsequent orders made by repeat customers. In the price promotion condition, a 10 RMB discount coupon was applied to the customer's order. In the no-promotion condition, customers received no discount. Customers were randomly assigned to one of the two conditions. In one day (between 9 a.m. and 9 p.m.), we observed 121 orders (37.2% female customers).
After each customer paid the bill, the research assistant, dressed as a staff member, handed the customer their receipt with an attached donation appeal and collected the café's copy of the receipt, which recorded the order details and the manipulation information. Meanwhile, another research assistant, disguised as a customer, recorded each customer's gender, estimated age, and number of companions.
The donation appeal flyer solicited donations to help pay the medical expenses of a 28-year-old woman diagnosed with leukemia. To reduce a potential norm of reciprocity, we chose this real donation appeal, which was posted by the recipient herself on a large online charity fundraising platform. Posters with the same information were also displayed at the cashier counter (for the field experiment setting, see Web Appendix 2). The research assistant who was disguised as a staff member was blind to our hypothesis and was instructed not to interact with customers regarding the donation appeal. We attached the QR code to the donation appeal flyer, so customers could make their donation decision at their own tables privately instead of deciding at the cashier counter. Customers who intended to donate could scan the QR code printed on the donation appeal flyer to view detailed information about the donation cause and make an actual donation. To match the donation data with the experimental conditions, we asked participants to enter the last four digits of their order number after they made their donation. At the end of the day, research assistants transferred all donations to the woman on behalf of the customers.
Of the 121 orders observed, 20 orders were excluded because the customer did not take the receipt and attached donation flyer. To cleanly manipulate price promotion, we also removed eight orders made by customers using other discounts and one order for which the customer refused to accept the coupon. As a result, 92 orders qualified for our analysis.
First, we examined the percentage of customers who donated in each condition. A binary logistic regression revealed that customers in the promotion condition were more likely to donate (17.50%) than those in the no-promotion condition (1.92%; Wald χ2( 1) = 4.75, p =.03). Next, we examined whether having a companion affected the customer's donation decision. When included as a covariate in our analysis, having a companion did not have a significant effect (Wald χ2 ( 1) = 2.45, p =.12), and the effect of the price promotion on the donation rate remained significant (Wald χ2 ( 1) = 4.55, p =.03).
The donation amount was positively skewed (skewness = 4.23, SE =.25). Thus, we log-transformed the donation amount, as in Study 2. The pattern of results remained the same, so we report the untransformed means for ease of interpretation. A one-way ANOVA revealed a positive effect of the price promotion on the donation amount (Mpromo = ¥1.00, SD = ¥2.52 vs. Mno promo = ¥.08, SD = ¥.56; F( 1, 90) = 7.29, p =.008, =.07). In addition, having a companion did not significantly affect the donation amount (F( 1, 89) = 2.01, p =.16, =.02), and the effect of the price promotion remained significant after controlling for this covariate (F( 1, 89) = 6.78, p =.01, =.07).
The results of this field experiment thus provided strong causal evidence for the positive effect of price promotions on donation behavior. Specifically, we found that a discount at a café significantly increased actual donation behavior among café customers.
In the next few studies (Studies 4–6), we aimed to test the underlying mechanism of perceived resources by examining whether the positive effect of price promotions on consumers' donation behavior is attenuated if we disrupt the impact of price promotions on consumers' perceived resources. In Study 4, we examined whether directing consumers' focus to the amount of money they just spent would reduce their perceived resources and hence attenuate the positive effect of price promotions on donation behavior.
Three hundred thirty-five students from a large university in Singapore (65.1% female; Mage = 22.4 years) completed an online experiment in exchange for a chance to win SGD $20. The experiment followed a 2 (promotion vs. no promotion) × 2 (control vs. focus on money spent) between-subjects design, and participants were randomly assigned to one of the four conditions.
As a cover story, we told participants that we were interested in their preferences for food delivery services. Participants were asked to imagine that they had been endowed with $20 and were instructed to use some of it to purchase a voucher; participants then had a choice to donate some of the leftover endowment to a charity. To make the experiment incentive-compatible, we followed recent research ([ 2]) and informed participants that we would randomly select ten participants and actually fulfill their decisions (i.e., they would receive their chosen voucher and any remaining endowment that did not go to the charity). To avoid a selection issue (i.e., that participants in the promotion condition might be more likely to make a purchase), we required participants to make a purchase choice between two product options.
In the no-promotion condition, participants were told that they could purchase a $10 voucher from either Grabfood or Foodpanda, two popular food delivery companies in Singapore. In the promotion condition, participants were told that they could purchase a $20 voucher for $10 (50% off) from either of these two companies. We controlled the final purchase price ($10 in both conditions) to eliminate the potential concern that participants in the promotion condition would spend less money on the required task and thus have more disposable money for a donation. After making the purchase decision, participants in the focus-on-money-spent condition were asked to recall how much money they spent on the voucher, whereas those in the control condition did not receive this instruction.
Subsequently, all participants took a survey in which they were asked to indicate how much they were willing to donate (between $0 and $10) to the Children's Charities Association to help physically, mentally, and socially disadvantaged children in Singapore. Participants were told that they would be expected to send their indicated donation amount if they won the lottery. Finally, they completed the same perceived resources scale (α =.89) used in Study 2.
After completing the study, all participants provided their email addresses for the lottery draw. Ten randomly chosen participants received the cash prize and were given the donation method to fulfill their promise to the charity.
A binary logistic regression analysis on the donation rate revealed a significant interaction effect between the price promotion and focus manipulation (Wald χ2( 1) = 6.86, p <.01). There was also a significant main effect of the focus manipulation (Wald χ2( 1) = 12.35, p <.001), whereas the main effect of the price promotion was not significant (Wald χ2( 1) = 1.48, p =.22). Within the control condition, participants in the promotion condition were more likely to donate to the charity (95.35%) than those in the no-promotion condition (84.34%; Wald χ2( 1) = 5.06, p =.02). This effect was eliminated in the focus-on-money-spent condition (promotion: 70.73% vs. control: 79.76%; Wald χ2( 1) = 1.80, p =.18).
No outliers were identified for donation amount, and there was no skewness problem (skewness = −.13, SE =.13). An ANOVA with the price promotion and focus manipulation as independent variables and donation amount as the dependent variable revealed a significant interaction effect (F( 1, 331) = 5.54, p =.02, =.02). There was also a significant main effect of the focus manipulation (F( 1, 331) = 17.42, p <.001, =.05), whereas the main effect of the price promotion was not significant (F( 1, 331) =.75, p =.39, =.002). Within the control condition, the average donation amount was greater in the promotion condition (Mpromo = $7.12, SD = $3.31) than in the no-promotion condition (Mno promo = $5.80, SD = $3.83; F( 1, 331) = 5.22, p =.02, =.02). Within the focus-on-money-spent condition, however, the effect of the price promotion on the donation amount was eliminated (Mpromo = $4.44, SD = $3.98 vs. Mno promo = $5.05, SD = $3.88; F( 1, 331) = 1.10, p =.30, =.003).
An ANOVA on the perceived resources index revealed a significant main effect of the price promotion (Mpromo = 4.86, SD = 1.27 vs. Mno promo = 4.05, SD = 1.46; F( 1, 331) = 29.21, p <.001, =.08) and a significant main effect of the focus manipulation (Mcontrol = 4.62, SD = 1.52 vs. Mfocus = 4.29, SD = 1.43; F( 1, 331) = 4.56, p =.03, =.01). However, the interaction term was not significant (F( 1, 331) = 1.05, p =.31, =.003).
The results from this study replicated the previous findings using an incentive-compatible online experiment. Furthermore, we found that the positive effect of the price promotion on donation behavior was eliminated when consumers were guided to think about the money they spent in the promotion. However, we did not observe a significant interaction effect on the perceived resources index. We suspect that although the focus manipulation significantly reduced participants' subjective experience of saving money, participants in the promotion condition had more actual monetary resources (i.e., a voucher worth $20) than those in the no-promotion condition (in which the voucher was worth only $10). In the next study, we address this limitation by using a cleaner manipulation.
Study 5 aimed to test budget overspending as a moderator of the effect of price promotions on donation behavior. Specifically, we predicted that the positive effect of price promotions on perceived resources (and, subsequently, on donation behavior) would be attenuated when the purchase exceeded consumers' mental budget.
Five hundred fourteen undergraduate students in China (43.8% female, all ≥18 years old) from an online subject pool participated in this study for monetary compensation. The study followed a 2 (promotion vs. no promotion) × 2 (over budget vs. within budget) between-subjects design.
All participants imagined that they were browsing one of their favorite online shops. In the promotion conditions, participants were told that the online shop was having a 50% off sale, and they spent ¥500 on products that originally cost ¥1,000. In the no-promotion conditions, participants were told that they spent ¥500 in the online shop. In the over-budget (within-budget) conditions, participants were told that their spending exceeded (was within) their budget. Subsequently, participants imagined that when they paid the bill through Alipay, they noticed that the platform was raising money to plant trees in a remote area in China. The cost of planting each tree was ¥3, and each participant could donate a maximum of ten trees. Participants indicated the number of trees they were willing to donate. Finally, they completed the same perceived resources scale (α =.79) used in Study 2.
Across conditions, a large majority of participants were willing to donate at least one tree (within-budget/promotion: 98.44%, within-budget/no-promotion: 96.12%, over-budget/promotion: 93.80%, over-budget/no-promotion: 93.75%). A binary logistic regression analysis on the donation rate revealed only a significant main effect of budget overspending (Wald χ2 ( 1) = 3.79, p =.05); neither the main effect of the price promotion nor the interaction effect was significant (ps >.34).
Given that the dependent variable was count data, we conducted a Poisson regression (e.g., [16]; [38]) on the number of trees donated, with budget overspending and the price promotion as independent factors. The analysis revealed a significant main effect of the price promotion (Wald χ2( 1) = 3.74, p =.05), no significant main effect of budget overspending (Wald χ2( 1) =.25, p =.62), and most importantly, a significant interaction effect (Wald χ2( 1) = 3.71, p =.05). Specifically, within the within-budget condition, the average donation amount was greater in the promotion condition (Mpromo = 6.07, SD = 3.20) than in the no-promotion condition (Mno promo = 5.26, SD = 3.37; Wald χ2( 1) = 7.38, p =.007). This effect did not occur within the over-budget condition (Mpromo = 5.76, SD = 3.32 vs. Mno promo = 5.76, SD = 3.41; Wald χ2 ( 1) <.001, p =.99).
An ANOVA on perceived resources revealed a significant main effect of the price promotion (F( 1, 510) = 44.30, p <.001), a significant main effect of budget overspending (F( 1, 510) = 26.21, p <.001), and most importantly, a significant interaction effect (F( 1, 510) = 7.27, p =.007). Specifically, among the within-budget participants, those in the promotion condition reported greater perceived resources (Mpromo = 5.27, SD = 1.11) than those in the no-promotion condition (Mno promo = 4.20, SD = 1.30; F( 1, 510) = 43.73, p <.001). This effect was attenuated among the over-budget participants (Mpromo = 4.38, SD = 1.31 vs. Mno promo = 3.93, SD = 1.44; F( 1, 510) = 7.84, p =.005).
Next, we tested whether perceived resources mediated the effect of the price promotion on the donation amount and whether budget overspending moderated the path from the price promotion to perceived resources. A moderated mediation analysis ([29], PROCESS Model 7 with 5,000 bootstrap samples) with perceived resources as the mediator and budget overspending as the moderator revealed a significant index of moderated mediation (b = −.57, SE =.22, 95% CI = [−1.01, −.15]). Specifically, in the within-budget conditions, the indirect effect of the price promotion on donation via perceived resources was positive and significant (b =.98, SE =.16, 95% CI = [.68, 1.32]). However, in the over-budget conditions, this indirect effect was attenuated (b =.42, SE =.16, 95% CI = [.11,.75]).
Results from Study 5 provided further support for the underlying mechanism related to perceived resources. Specifically, the study showed that budget overspending attenuates the positive effect of price promotions on consumers' donation behavior via its influence on perceived resources. To further test the moderating role of budget overspending, we conducted an additional study in which we manipulated the product type. Prior research suggests that consumers usually have mental budgets for necessity purchases, whereas indulgence purchases often entail budget overspending ([34]; [55]). This additional study followed a 2 (promotion vs. no promotion) × 2 (necessity purchase vs. indulgence purchase) between-subjects design, and it is reported in Web Appendix 3.
The objective of Study 6 was twofold. First, we aimed to provide further field evidence for the effect of price promotions on consumers' donation behavior. To this end, we collaborated with a café to run price promotions, and we examined customers' actual donation behavior. Second, we wanted to examine the perceived resources mechanism by comparing the effects of an instant discount coupon versus a rebate coupon. Specifically, consumers who received a rebate coupon could use the money to offset their spending in the next (but not the current) purchase. In other words, the rebate coupon did not offer immediate monetary savings, so the rebate coupon (unlike the discount coupon) should not increase consumers' current perceived resources. Therefore, we predicted that consumers who received the instant discount coupon would exhibit greater donation behavior than those who received the rebate coupon or no coupon.
We conducted this field experiment in a café on the campus of a large university in China in October 2019. The field experiment followed a three-cell between-subjects design (no promotion vs. instant discount coupon vs. rebate coupon). To minimize the risk that customers would notice the existence of other conditions, which could potentially cause data contamination, we did not randomize the conditions within a day. Instead, we conducted the study over the course of six days (Tuesday, Wednesday, and Thursday for two consecutive weeks) and randomly assigned one experimental condition to each day (yielding two days per condition). In total, we observed 399 customers (48.6% female), with 107, 127, and 165 customers in the no-promotion, instant-discount-coupon, and rebate-coupon conditions, respectively.
The field experiment commenced when customers came to order food/drinks from the cashier. In the no-promotion condition, customers did not receive any kind of coupon. In the instant-discount-coupon condition, the cashier gave each customer a ¥5 coupon (valid until November 30) and told them that they could use the coupon on their current order. In the rebate-coupon condition, the cashier gave each customer a ¥5 coupon (valid until November 30) and told them that the coupon was valid only on future purchases. Customers paid and then waited at the pick-up counter for their orders; as they waited, a research assistant approached them with a donation appeal poster for House of Kindness, a charity run by the university's student union to help replace broken free-sharing umbrellas provided for students and teachers on the campus. The same poster was also displayed at the café's pick-up counter (for the field experiment setting, see Web Appendix 4). The research assistant asked each customer to consider donating, and customers who agreed proceeded to donate any amount by scanning the QR code on the poster. At the same time, another research assistant, who pretended to be a server in the café, observed and recorded each customer's gender, estimated age, number of companions, and donation behavior (yes/no), and she collected each customer's receipt when they picked up their order.
A binary logistic regression on the donation rate revealed a marginally significant effect of the price promotion (Wald χ2( 2) = 4.52, p =.10). Specifically, the donation rate in the instant-discount-coupon condition (52.76%) was higher than in both the rebate-coupon condition (40.61%; Wald χ2( 1) = 4.24, p =.04) and the no-promotion condition (42.99%; Wald χ2( 1) = 2.21, p =.13). Importantly, there was a significant difference between the instant-discount-coupon condition (52.76%) and the other two conditions combined (41.54%; Wald χ2( 1) = 4.11, p =.04). The results remained robust after controlling for the presence of a companion (Wald χ2( 2) = 4.86, p =.09), which had an insignificant effect (Wald χ2( 1) =.54, p =.46).
The donation amount varied from ¥0 to ¥100. Due to the high variance, we identified and removed six outliers that were three standard deviations above or below the mean. The donation amount was positively skewed (skewness = 1.81, SE =.12), so we log-transformed the donation amount; the pattern of results remained the same regardless of the log-transformation, and we report the untransformed means for ease of interpretation. An ANOVA revealed an insignificant effect of the price promotion on the donation amount (F( 2, 390) = 2.03, p =.13, =.01). Specifically, the average donation amount was directionally higher in the instant-discount-coupon condition (Minstant = ¥3.45, SD = ¥4.48) than in both the no-promotion condition (Mno promo = ¥2.87, SD = ¥4.50; F( 1, 390) = 2.12, p =.15, =.005) and the rebate-coupon condition (Mrebate = ¥2.86, SD = ¥4.93; F( 1, 390) = 3.74, p =.05, =.01). Importantly, there was a significant difference between the instant-discount-coupon condition and the other two conditions combined (F( 1, 390) = 3.76, p =.05, =.01). The effect remained robust after controlling for the presence of a companion (F( 2, 389) = 2.14, p =.12, =.01), which had an insignificant effect (F( 1, 389) =.30, p =.59, =.001).
The results of this study provided further field evidence in a real shopping environment for the positive effect of price promotions on consumers' donation behavior. Furthermore, we found that customers who received immediate monetary savings (i.e., the instant discount coupon) were more likely to donate than customers whose monetary savings were not immediate (i.e., the rebate coupon).
Study 7 investigates another boundary condition—the time interval between the price promotion and the donation solicitation—that we hypothesized would moderate the positive effect of price promotions on donation behavior. As time passes after a price promotion, consumers may spend their savings on another purchase or may gradually adapt to the savings; both outcomes should eliminate any increase in perceived resources that the promotion originally conferred. Thus, we expected that the positive effect of price promotions on donation behavior would be stronger if the donation was solicited immediately after the price promotion and would be attenuated if there was a delay between the price promotion and donation solicitation.
Two hundred ninety-four shoppers who made purchases on December 12 in the 2019 Alibaba's Double 12 sales event (63.4% female, all ≥18 years old) were recruited from an online panel in China on either December 12–13 (124 shoppers) or December 20–21 (170 shoppers) to complete this study in exchange for monetary compensation. As a cover story, we told all shoppers that we were interested in understanding their shopping experience. As in Study 1, shoppers completed a few ostensibly unrelated surveys. In the first survey, they were asked to indicate how much money (in RMB) they actually spent on their purchases on the same 22-point scale used in Study 1. We did not measure perceived savings to rule out demand effects as a possible explanation for the results of Study 1.
In the second survey, shoppers saw a donation appeal from the charity Wardrobe of Love, which raises funds to purchase new winter clothes for children in need. Shoppers indicated their intention to donate to this charity on a seven-point scale (1 = "definitely will not donate," and 7 = "definitely will donate").
Consistent with the findings of Study 1, a linear regression analysis revealed that the amount spent during the Double 12 sales event was significantly positively correlated with shoppers' intention to donate to the charity (b =.09, t(292) = 3.59, p <.001). More importantly, a linear regression with donation intention as the dependent variable and the amount spent and time interval (−.5 = immediately after the promotion,.5 = one week after the promotion) as independent variables revealed a significant main effect of the amount spent (b =.09, t(290) = 3.65, p <.001) and a significant main effect of the time interval (b =.76, t(290) = 3.49, p <.001). These main effects were qualified by a marginally significant interaction effect between the time interval and amount spent (b = −.08, t(290) = −1.66, p <.10).
Specifically, among shoppers who were surveyed immediately after the Double 12 promotion event, the donation intention significantly increased with the amount spent in the promotion (b =.13, t(290) = 3.36, p <.001), but this effect did not occur among shoppers who were surveyed one week after the promotion event (b =.05, t(290) = 1.62, p >.10). Study 7 thus showed that the positive effect of price promotions on donation behavior is stronger when the donation is solicited immediately after the price promotion, and it disappears over time.
Across a set of seven studies, using both field and experimental data, we provide robust evidence that price promotions can increase consumers' donation behavior (for a summary of the results, see Table 2). Study 1 provided correlational field evidence for the proposed effect. Study 2 manipulated the presence as well as the magnitude of the price promotion and provided causal evidence for the effect. Study 2 also showed that perceived resources mediated the effect of price promotions on donation behavior. Study 3 provided further causal evidence by implementing a price promotion in a field experiment. Next, we provided further support for the perceived resources mechanism by showing that the positive effect of price promotions on consumers' donation behavior was attenuated when consumers focused on how much money they spent (rather than saved) in the promotion (Study 4), when the purchase involved budget overspending (Study 5), and when the monetary savings could not be realized immediately (Study 6). Finally, we showed that the effect was attenuated by a longer delay (one week vs. one day) between the price promotion and donation solicitation (Study 7). Together, our findings converged to establish the robustness of the positive effect of price promotions on consumers' donation behavior.
Graph
Table 2. Summary of the Effect of Price Promotions on Donation Behavior in Each Study.
| Study | Sample Size | Data Collection Method | Focal Effects |
|---|
| Coefficient Between Money Spent in the Price Promotion and Charitable Behavior |
|---|
| 1 | 135 | Online survey (China) | .26a |
| 7 | 294 | Online survey (China) | Immediately after promotion:.13a |
| | | One week after promotion:.05c |
| Donation Amount | Donation Rate |
| | | Promotion | No Promotion/Other | Promotion | No Promotion/Other |
| 2 | 200 | Online experiment (USA) | $16.74d | $7.54a | 74.17%d | 65.31%c |
| 3 | 92 | Field experiment (China) | ¥1.00 | ¥.08a | 17.50% | 1.92%a |
| 4 | 335 | Online experiment (Singapore) | Control:$7.12 | Control:$5.80a | Control:95.35% | Control:84.34%a |
| Focus on money spent:$4.44 | Focus on money spent:$5.05c | Focus on money spent:70.73% | Focus on money spent:79.76%c |
| 5 | 514 | Online experiment (China) | Within budget:6.07 | Within budget:5.26a | Within budget:98.44% | Within budget:96.12%c |
| Over budget:5.76 | Over budget:5.76c | Over budget:93.80% | Over budget:93.75%c |
| 6 | 399 | Field experiment (China) | ¥3.45 | ¥2.87b,e | 52.76% | 41.54%a,e |
| Web Appendix 3 | 320 | Online experiment (USA) | Necessity:$12.32 | Necessity:$5.14a | Necessity:81.25% | Necessity:70.00%b |
| Indulgence:$6.15 | Indulgence:$5.77c | Indulgence:70.00% | Indulgence:81.25%b |
- 40022242920988260 a The difference between the promotion condition and the no-promotion/other condition was significant (p <.05).
- 50022242920988260 b The difference between the promotion condition and the no-promotion/other condition was marginally significant (p <.10).
- 60022242920988260 c The difference between the promotion condition and the no-promotion/other condition was insignificant (p >.10).
- 70022242920988260 d The promotion conditions were pooled to compare with the no-promotion condition.
- 80022242920988250 e The no-promotion condition and rebate-coupon condition were pooled to compare with the instant-discount-coupon condition.
It is worth noting that although we found a consistent effect on the donation amount, we found mixed results on the donation rate. This might have been partially caused by a procedural difference in the donation rate measurement: in our field experiments (Studies 2 and 6), we first asked customers whether they would like to donate; only customers who decided to donate then moved on to make a donation. In other words, it was a two-step donation decision as we explicitly asked customers to make two sequential decisions: ( 1) whether to donate and ( 2) how much money to donate. In the other studies, however, we asked all participants to indicate the donation amount, and we computed the donation rate by coding those who indicated a donation amount of "0" as "didn't donate" and those who indicated a nonzero donation amount as "donated." Nevertheless, these results are consistent with the prior findings that different factors may affect donation choice and donation amount (e.g., [19]; [23]).
Furthermore, we should acknowledge that the effect of price promotions on donation behavior may be multiply determined. For example, prior research has suggested that affective responses such as happiness ([32]) and guilt ([ 7]) may influence charitable behavior. We measured some affective responses in Studies 2 and 5 (see Web Appendix 5). In Study 5, while we found some evidence that greater perceived resources could also increase happiness, happiness did not directly mediate the effect of price promotions on donation behavior. Furthermore, guilt did not mediate the observed effect in our studies, and a guilt-based account would predict a stronger effect for purchases associated with guilt (e.g., budget overspending in Study 5; an indulgence purchase in the additional study)—but we found an attenuated effect in these scenarios.
This research makes several important theoretical contributions. First, while existing research has examined the effect of price promotions on firms' performances and consumers' purchasing behaviors, there is a gap in the literature regarding whether and how price promotions can have important social consequences. This research fills that gap. Specifically, this research shows that price promotions can boost consumers' perceived resources and thereby increase their donation behavior—a positive social consequence. Second, while promotions and donation behavior have been studied together in the cause-related marketing literature, our research is unique in its focus on donation behavior as a consequence of a promotion rather than as the promotion itself ([56]; [59]). Third, our research adds to the donation behavior literature by identifying price promotions as a novel situational factor that drives consumers' donation behavior. More broadly, this research answers the call for a greater understanding of when and why marketing activities can contribute to a better world by improving consumer and societal welfare.
This research offers pertinent and actionable implications for charitable organizations. Specifically, our findings may help charitable organizations make three important decisions:
- Whom to target: Consumers who have participated in price promotions. Our research indicates that these consumers have a greater charitable tendency, and it should be easier to identify and target them than to reach out to potential donors on the basis of individual characteristics (e.g., sympathy, donation history).
- When to solicit donations: Immediately after consumers make purchases. A few years ago, a global movement named Giving Tuesday was initiated by New York City's 92nd Street Y and the United Nations Foundation in the post-Thanksgiving season. Our findings not only help explain the success of this Giving Tuesday phenomenon but also provide insights about the timing for government or international organizations to initiate charitable campaigns.
- How to increase the effectiveness of donation appeals: Our research indicates that charitable organizations should pair their donation appeals with promotions for necessities (vs. indulgences) that offer immediate discounts (vs. future rebates). Furthermore, the donation appeals should direct consumers' focus toward the money they saved (vs. spent) in the promotion. These are ecologically valid factors in the marketplace, and charitable organizations can take advantage of them to optimize their donation appeals.
Furthermore, this research suggests that firms can use price promotions as great opportunities to collaborate with charitable organizations. For example, the outdoor brand Patagonia has made a commitment since 2016 to donate 100% of its profits from Black Friday to charities ([39]). Unfortunately, in traditional cause-related marketing practices, consumers might doubt a firm's prosocial motivation because the benefits for the charity are contingent on consumers' purchases from the firm ([18]; [24]). Our findings suggest that by soliciting donations after consumers complete their purchases, firms can cultivate a purer image of corporate social responsibility. This strategy was exemplified recently by Ralph Lauren, which partnered with the World Health Organization to fight the COVID-19 pandemic by soliciting donations from customers immediately after they submitted their orders on the store's official online shop. This collaborative strategy between firms and nonprofit organizations represents a win-win situation that can benefit both stakeholders and contribute to a better world.
This research also raises interesting directions for future studies. First, while this research focused on the positive impact of price promotions on monetary donation behaviors, it would be interesting to explore the generalizability of this effect to nonmonetary donation behaviors (e.g., volunteering). We found preliminary evidence for this in Study 1, but more studies are needed to establish a robust effect. Second, in addition to exploring different types of promotions, future research could examine whether different elements of a promotion also moderate the effect of price promotions on donation behavior. For example, it would be interesting to examine whether the percentage or the absolute size of the discount has a greater effect on consumers' perceived resources. Third, future research could investigate other boundary conditions, such as consumers' chronic resource availability or price consciousness. These research directions would further enrich our understanding of the social implications of price promotions and could offer relevant insights for firms and consumer welfare.
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242920988253 - Do Promotions Make Consumers More Generous? The Impact of Price Promotions on Consumers' Donation Behavior
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242920988253 for Do Promotions Make Consumers More Generous? The Impact of Price Promotions on Consumers' Donation Behavior by Kuangjie Zhang, Fengyan Cai and Zhengyu Shi in Journal of Marketing
The original stimuli appeared in Chinese.
- 1. Give directions to a stranger who loses his/her way.
- 2. Give money to beggars or strangers who need it.
- 3. Give food to beggars or homeless people who need it.
- 4. Donate money to charitable organizations.
- 5. Donate goods or clothing to charitable organizations.
- 6. Volunteer for charitable organizations.
- 7. Help strangers carry things (e.g., luggage, bags).
- 8. Let someone in need queue in front of you at a supermarket, fast-food restaurant, etc.
- 9. Offer your own seat to a standing stranger in need on a bus or metro.
- 10. Participate in community service activities.
- 11. Raise funds for charitable organizations.
- 12. Help people in other ways.
- 13. Donate money via online charity platforms (e.g., Alibaba charity platform, Tencent charity platform, JD Foundation charity platform).
Footnotes 1 Kuangjie Zhang and Fengyan Cai contributed equally to this research.
2 Vikas Mittal
3 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
4 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The authors gratefully acknowledge the financial support provided by Grant NSFC71922017, 71672110, 71832008, 91646205 and 91746206 from China, and the financial support from Nanyang Technological University and Shanghai Jiao Tong University.
5 Kuangjie Zhang https://orcid.org/0000-0002-8760-2316
6 Online supplement: https://doi.org/10.1177/0022242920988253
References Ailawadi Kusum L., Gedenk Karen, Lutzky Christian, Neslin Scott A. (2007), "Decomposition of the Sales Impact of Promotion-Induced Stockpiling," Journal of Marketing Research, 44 (3), 450–67.
Allard Thomas, Hardisty David J., Griffin Dale. (2019), "When 'More' Seems Like Less: Differential Price Framing Increases the Choice Share of Higher-Priced Options," Journal of Marketing Research, 56 (5), 826–41.
Andreoni James, Nikiforakis Nikos, Stoop Jan. (2017), "Are the Rich More Selfish Than the Poor, or Do They Just Have More Money? A Natural Field Experiment," National Bureau of Economic Research, https://www.nber.org/papers/w23229.
Arkes Hal R., Joyner Cynthia A., Pezzo Mark V., Nash Jane Gradwohl, Siegel-Jacobs Karen, Stone Eric. (1994), "The Psychology of Windfall Gains," Organizational Behavior and Human Decision Processes, 59 (3), 331–47.
Ashworth Laurence, Darke Peter R., Schaller Mark. (2005), "No One Wants to Look Cheap: Trade-Offs Between Social Disincentives and the Economic and Psychological Incentives to Redeem Coupons," Journal of Consumer Psychology, 15 (4), 295–306.
Ashworth Laurence, McShane Lindsay. (2012), "Why Do We Care What Others Pay? The Effect of Other Consumers' Prices on Inferences of Seller (Dis)Respect and Perceptions of Deservingness Violation," Journal of Retailing, 88 (1), 145–55.
7 Basil Debra Z., Ridgway Nancy M., Basil Michael D. (2008), "Guilt and Giving: A Process Model of Empathy and Efficacy," Psychology & Marketing, 25 (1), 1–23.
8 Bell David R., Chiang Jeongwen, Padmanabhan V. (1999), "The Decomposition of Promotional Response: An Empirical Generalization," Marketing Science, 18 (4), 504–26.
9 Berman Jonathan Z., Bhattacharjee Amit, Small Deborah A., Zauberman Gal. (2020), "Passing the Buck to the Wealthier: Reference-Dependent Standards of Generosity," Organizational Behavior and Human Decision Processes, 157, 46–56.
Blair Edward A., Landon E. LairdJr. (1981), "The Effects of Reference Prices in Retail Advertisements," Journal of Marketing, 45 (2), 61–9.
Blattberg Robert C., Neslin Scott A. (1990), Sales Promotions: Concepts, Methods and Strategies. Englewood Cliffs, NJ: Prentice Hall.
Cai Fengyan, Bagchi Rajesh, Gauri Dinesh K. (2016), "Boomerang Effects of Low Price Discounts: How Low Price Discounts Affect Purchase Propensity," Journal of Consumer Research, 42 (5), 804–16.
Cai Fengyan, Wyer Robert S.Jr. (2015), "The Impact of Mortality Salience on the Relative Effectiveness of Donation Appeals," Journal of Consumer Psychology, 25 (1), 101–12.
Cavanaugh Lisa A., Bettman James R., Luce Mary Frances. (2015), "Feeling Love and Doing More for Distant Others: Specific Positive Emotions Differentially Affect Prosocial Consumption," Journal of Marketing Research, 52 (5), 657–73.
Chandon Pierre, Wansink Brian, Laurent Gilles. (2000), "A Benefit Congruency Framework of Sales Promotion Effectiveness," Journal of Marketing, 64 (4), 65–81.
Coxe Stefany, West Stephen G., Aiken Leona S. (2009), "The Analysis of Count Data: A Gentle Introduction to Poisson Regression and Its Alternatives," Journal of Personality Assessment, 91 (2), 121–36.
Darke Peter R., Chung Cindy M.Y. (2005), "Effects of Pricing and Promotion on Consumer Perceptions: It Depends on How You Frame It," Journal of Retailing, 81 (1), 35–47.
Dean Dwane Hal. (2003), "Consumer Perception of Corporate Donations Effects of Company Reputation for Social Responsibility and Type of Donation," Journal of Advertising, 32 (4), 91–102.
Dickert Stephan, Sagara Namika, Slovic Paul. (2011), "Affective Motivations to Help Others: A Two-Stage Model of Donation Decisions," Journal of Behavioral Decision Making, 24 (4), 361–76.
Dovidio John F., Piliavin Jane Allyn, Schroeder David A., Penner Louis A. (2006), The Social Psychology of Prosocial Behavior. Mahwah, NJ: Lawrence Erlbaum Associates.
Duclos Rod, Barasch Alixandra. (2014), "Prosocial Behavior in Intergroup Relations: How Donor Self-Construal and Recipient Group-Membership Shape Generosity," Journal of Consumer Research, 41 (1), 93–108.
Ethical Consumers (2018), "Boycott Black Friday," (November 20), https://www.ethicalconsumer.org/retailers/boycott-black-friday.
Fajardo Tatiana M., Townsend Claudia, Bolander Willy. (2018), "Toward an Optimal Donation Solicitation: Evidence from the Field of the Differential Influence of Donor-Related and Organization-Related Information on Donation Choice and Amount," Journal of Marketing, 82 (2), 142–52.
Foreh Mark R., Grier Sonya. (2003), "When Is Honesty the Best Policy? The Effect of Stated Company Intent on Consumer Skepticism," Journal of Consumer Psychology, 13 (3), 349–56.
Gertz Marisa, Porter GeraldJr, Roeder Jonathan. (2018), "How Black Friday Became America's Greediest Holiday," Bloomberg(November 18), https://www.bloomberg.com/news/photo-essays/2018-11-18/how-black-friday-became-america-s-greediest-holiday.
Gourville John T., Soman Dilip. (1998), "Payment Depreciation: The Behavioral Effects of Temporally Separating Payments from Consumption," Journal of Consumer Research, 25 (2), 160–74.
Grewal Dhruv, Monroe Kent B., Krishnan R. (1998), "The Effects of Price-Comparison Advertising on Buyers' Perceptions of Acquisition Value, Transaction Value, and Behavioral Intentions," Journal of Marketing, 62 (2), 46–59.
Havens John J., O'Herlihy Mary A., Schervish Paul G. (2007), "Charitable Giving: How Much, by Whom, to What, and How?" in The Nonprofit Sector: A Research Handbook, Powell Walter W., Steinberg Richard, eds. London: Yale University Press, 542–67.
Hayes Andrew F. (2017), Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach, 2nd ed. New York: Guilford Publications.
Heath Chip, Soll Jack B. (1996), "Mental Budgeting and Consumer Decisions," Journal of Consumer Research, 23 (1), 40–52.
Heilman Carrie M., Nakamoto Kent, Rao Ambar G. (2002), "Pleasant Surprises: Consumer Response to Unexpected In-Store Coupons," Journal of Marketing Research, 39 (2), 242–52.
Isen Alice M., Levin Paula F. (1972), "Effect of Feeling Good on Helping: Cookies and Kindness," Journal of Personality and Social Psychology, 21 (3), 384–88.
Kharpal Arjun. (2019), "Alibaba Breaks Singles Day Record with More than $38 Billion in Sales," CNBC(November 11), https://www.cnbc.com/2019/11/11/alibaba-singles-day-2019-record-sales-on-biggest-shopping-day.html.
Kivetz Ran, Simonson Itamar. (2002), "Self-Control for the Righteous: Toward a Theory of Precommitment to Indulgence," Journal of Consumer Research, 29 (2), 199–217.
Klebnikov Sergei. (2019), "Cyber Monday 2019 by the Numbers: A Record $9.4 Billion Haul," Forbes(December 3), https://www.forbes.com/sites/sergeiklebnikov/2019/12/03/cyber-monday-2019-by-the-numbers-a-record-94-billion-haul/#b3a8e232ef0b.
Korndörfer Martin, Egloff Boris, Schmukle Stefan C. (2015), "A Large Scale Test of the Effect of Social Class on Prosocial Behavior," PLoS One, 10 (7), e0133193.
Krishna Aradhna, Briesch Richard, Lehmann Donald R., Yuan Hong. (2002), "A Meta-Analysis of the Impact of Price Presentation on Perceived Savings," Journal of Retailing, 78 (2), 101–18.
Kupor Daniella, Tormala Zakary. (2018), "When Moderation Fosters Persuasion: The Persuasive Power of Deviatory Reviews," Journal of Consumer Research, 45 (3), 490–510.
Lauletta Tyler. (2016), "Patagonia Is Donating 100% of Its Black Friday Sales to Charity This Year," Business Insider(November 23), https://www.businessinsider.com/patagonia-black-friday-donate-100-percent-of-sales-to-charity-2016-11/.
Lee Leonard, Tsai Claire I. (2014), "How Price Promotions Influence Postpurchase Consumption Experience over Time," Journal of Consumer Research, 40 (5), 943–59.
Lee Saerom, Winterich Karen Page, Ross William T.Jr. (2014), "I'm Moral, but I Won't Help You: The Distinct Roles of Empathy and Justice in Donations," Journal of Consumer Research, 41 (3), 678–96.
Levontin Liat, Ein-Gar Danit, Lee Angela Y. (2015), "Acts of Emptying Promote Self-Focus: A Perceived Resource Deficiency Perspective," Journal of Consumer Psychology, 25 (2), 257–67.
Lichtenstein Donald R., Netemeyer Richard G., Burton Scot. (1990), "Distinguishing Coupon Proneness from Value Consciousness: An Acquisition-Transaction Utility Theory Perspective," Journal of Marketing, 54 (3), 54–67.
Lyubomirsky Sonja, King Laura, Diener Ed. (2016), "The Benefits of Frequent Positive Affect: Does Happiness Lead to Success?" Psychological Bulletin, 131 (6), 803–55.
Mela Carl F., Gupta Sunil, Lehmann Donald R. (1997), "The Long-Term Impact of Promotion and Advertising on Consumer Brand Choice," Journal of Marketing Research, 34 (2), 248–61.
Mukhopadhyay Anirban, Johar Gita Venkataramani. (2007), "Tempted or Not? The Effect of Recent Purchase History on Responses to Affective Advertising," Journal of Consumer Research, 33 (4), 445–53.
Piff Paul K., Kraus Michael W., Côté Stéphane, Cheng Bonnie Hayden, Keltner Dacher. (2010), "Having Less, Giving More: The Influence of Social Class on Prosocial Behavior," Journal of Personality and Social Psychology, 99 (5), 771–84.
Raghubir Priya, Corfman Kim. (1999), "When Do Price Promotions Affect Pretrial Brand Evaluations?" Journal of Marketing Research, 36 (2), 211–22.
Reed AmericusII, Aquino Karl, Levy Eric. (2007), "Moral Identity and Judgments of Charitable Behaviors," Journal of Marketing, 71 (1), 178–93.
Roux Caroline, Goldsmith Kelly, Bonezzi Andrea. (2015), "On the Psychology of Scarcity: When Reminders of Resource Scarcity Promote Selfish (and Generous) Behavior," Journal of Consumer Research, 42 (4), 615–31.
Schindler Robert M. (1998), "Consequences of Perceiving Oneself as Responsible for Obtaining a Discount: Evidence for Smart-Shopper Feelings," Journal of Consumer Psychology, 7 (4), 371–92.
Shady Franklin, Lee Leonard. (2020), "Price Promotions Cause Impatience," Journal of Marketing Research, 57 (1), 118–33.
Shiv Baba, Carmon Ziv, Ariely Dan. (2005), "Placebo Effects of Marketing Actions: Consumers Get What They Pay For," Journal of Marketing Research, 42 (4), 383–93.
Sober Elliott, Wilson David Sloan. (1999), Unto Others: The Evolution and Psychology of Unselfish Behavior. Cambridge, MA: Harvard University Press.
Stilley Karen M., Inman J. Jeffrey, Wakefield Kirk L. (2010), "Spending on the Fly: Mental Budgets, Promotions, and Spending Behavior," Journal of Marketing, 74 (3), 34–47.
Strahilevitz Michal, Myers John G. (1998), "Donations to Charity as Purchase Incentives: How Well They Work May Depend on What You Are Trying to Sell," Journal of Consumer Research, 24 (4), 434–46.
Thaler Richard. (1985), "Mental Accounting and Consumer Choice," Marketing Science, 4 (3), 177–266.
Thaler Richard. (1999), "Mental Accounting Matters," Journal of Behavioral Decision Making, 12 (3), 183–206.
Varadarajan P. Rajan, Menon Anil. (1988), "Cause-Related Marketing: A Coalignment of Marketing Strategy and Corporate Philanthropy," Journal of Marketing, 52 (3), 58–74.
White Katherine, Peloza John. (2009), "Self-Benefit Versus Other-Benefit Marketing Appeals: Their Effectiveness in Generating Charitable Support," Journal of Marketing, 73 (4), 109–24.
Wiepking Pamala, Bekkers René. (2012), "Who Gives? A Literature Review of Predictors of Charitable Giving II: Gender, Family Composition and Income," Voluntary Sector Review, 3 (2), 217–45.
Wiepking Pamala, Breeze Beth. (2012), "Feeling Poor, Acting Stingy: The Effect of Money Perceptions on Charitable Giving," International Journal of Nonprofit and Voluntary Sector Marketing, 17 (1), 13–24.
Winterich Karen Page, Mittal Vikas, Aquino Karl. (2013), "When Does Recognition Increase Charitable Behavior? Toward a Moral Identity-Based Model," Journal of Marketing, 77 (3), 121–34.
Winterich Karen Page, Mittal Vikas, Ross William T.Jr. (2009), "Donation Behavior Toward In-Groups and Out-Groups: The Role of Gender and Moral Identity," Journal of Consumer Research, 36 (2), 199–214.
Yoo Boonghee, Donthu Naveen, Lee Sungho. (2000), "An Examination of Selected Marketing Mix Elements and Brand Equity," Journal of the Academy of Marketing Science, 28 (2), 195–211.
~~~~~~~~
By Kuangjie Zhang; Fengyan Cai and Zhengyu Shi
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 47- Do Spoilers Really Spoil? Using Topic Modeling to Measure the Effect of Spoiler Reviews on Box Office Revenue. By: Jun Hyun (Joseph) Ryoo; Xin (Shane) Wang; Shijie Lu. Journal of Marketing. Mar2021, Vol. 85 Issue 2, p70-88. 19p. 1 Color Photograph, 2 Diagrams, 8 Charts, 1 Graph. DOI: 10.1177/0022242920937703.
- Database:
- Business Source Complete
Record: 48- Emotional Calibration and Salesperson Performance. By: Kidwell, Blair; Hasford, Jonathan; Turner, Broderick; Hardesty, David M.; Zablah, Alex Ricardo. Journal of Marketing. Nov2021, Vol. 85 Issue 6, p141-161. 21p. 2 Diagrams, 5 Charts, 4 Graphs. DOI: 10.1177/0022242921999603.
- Database:
- Business Source Complete
Record: 49- Erratum to "Do Nudges Reduce Disparities? Choice Architecture Compensates for Low Consumer Knowledge". Journal of Marketing. Sep2021, Vol. 85 Issue 5, p161-161. 1p. DOI: 10.1177/00222429211031013.
- Database:
- Business Source Complete
Erratum to "Do Nudges Reduce Disparities? Choice Architecture Compensates for Low Consumer Knowledge"
Mrkva, Kellen, Nathaniel A. Posner, Crystal Reeck, and Eric J. Johnson (2021), "Do Nudges Reduce Disparities? Choice Architecture Compensates for Low Consumer Knowledge," Journal of Marketing, 85 (4), 67–84. (Original DOI: 10.1177/0022242921993186)
In this article, an author affiliation was omitted. The corrected author information of Kellen Mrkva is presented here and the online version has been updated to reflect the correct affiliation.
Kellen Mrkva is Lecturer in Marketing, University of Bath School of Management, UK, and was Postdoctoral Scholar, Columbia Business School, Columbia University, USA (email: km3386@columbia.edu).
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 50- Evolution of Consumption: A Psychological Ownership Framework. By: Morewedge, Carey K.; Monga, Ashwani; Palmatier, Robert W.; Shu, Suzanne B.; Small, Deborah A. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p196-218. 23p. 1 Black and White Photograph, 5 Charts. DOI: 10.1177/0022242920957007.
- Database:
- Business Source Complete
Evolution of Consumption: A Psychological Ownership Framework
Technological innovations are creating new products, services, and markets that satisfy enduring consumer needs. These technological innovations create value for consumers and firms in many ways, but they also disrupt psychological ownership––the feeling that a thing is "MINE." The authors describe two key dimensions of this technology-driven evolution of consumption pertaining to psychological ownership: ( 1) replacing legal ownership of private goods with legal access rights to goods and services owned and used by others and ( 2) replacing "solid" material goods with "liquid" experiential goods. They propose that these consumption changes can have three effects on psychological ownership: they can threaten it, cause it to transfer to other targets, and create new opportunities to preserve it. These changes and their effects are organized in a framework and examined across three macro trends in marketing: ( 1) growth of the sharing economy, ( 2) digitization of goods and services, and ( 3) expansion of personal data. This psychological ownership framework generates future research opportunities and actionable marketing strategies for firms aiming to preserve the positive consequences of psychological ownership and navigate cases for which it is a liability.
Keywords: access-based consumption; big data; digitization; privacy; psychological ownership; sharing economy
Technological innovations are rapidly changing the consumption of goods and services. In modern capitalist societies, consumption is evolving from a model in which people legally own private material goods to access-based models in which people purchase temporary rights to use shared, experiential goods ([ 7]; [30]; [101]). Many urban consumers have replaced car ownership, once a symbol of independence and status, with car- and ride-sharing services that provide access to a vehicle or transportation when needed. Physical pictures occupying frames, wallets, and albums have been replaced with digital photographs; moreover, songs, books, movies, or magazines can be pulled down from the cloud at any time to suit a consumer's mood. Half the world population now buys, sells, generates, and consumes goods and information online through connected devices ([48]), generating vast quantities of personal data about their consumption patterns and private lives. The many benefits that these technological innovations and new business models offer to consumers––from convenience to lower economic cost to greater sustainability––makes legal ownership of many physical private goods undesirable and unnecessary ([79]). Consumers are not, however, simply exchanging the consumption of solid goods (i.e., enduring, ownership-based, and material) for liquid goods and services (i.e., ephemeral, access-based and dematerialized; [ 8]; [11]). We argue that relationships between consumers and their goods are changing.
Aligned with a Marketing Science Institute priority (2018–2020) to examine how economic macro trends are influencing consumers, we examine how this technology-driven evolution in consumption affects consumer behavior. We focus on ways in which changing consumption patterns are threatening, transferring, and creating new opportunities to cultivate psychological ownership—the feeling that something is MINE ([43]). It is a psychological state that is distinct from legal ownership. In contrast to the benefits accrued through consumers' reduced legal ownership of goods (for reviews, see [ 8]; [30]; [69]; [101]), a commensurate reduction in psychological ownership should typically be detrimental to both consumers and firms.
Psychological ownership is, in many ways, a valuable asset. It satisfies important consumer motives and has value-enhancing consequences. Within consumers, psychological ownership satisfies an effectance motive––a basic and chronic motive to have control and mastery over their environment, and motives to express their identity to others and themselves ([15]). Moreover, the feeling that a good is "MINE" enhances attitudes toward the good, strengthens attachments to the good, and increases its perceived economic value (for reviews, see [31]; [86]; [95]; [96]). Downstream consequences of value to firms include increased consumer demand for goods and services offered by the firm, willingness to pay for goods, word of mouth, and loyalty ([ 4]; [41]; [42]; [119]). Given these important consequences, we argue that preserving psychological ownership in the technology-driven evolution of consumption underway should be a priority for marketers and firm strategy.
Our article starts with the proposal that technological innovations are changing consumption along two dimensions: ( 1) replacing legal ownership of private goods with legal access to goods and services owned and used by others and ( 2) replacing "solid" material goods with "liquid" experiential goods (for examples, see Figure 1). We theorize that important consequences for consumer behavior are determined by the way these changes affect psychological ownership for goods and services—that is, by threatening, transferring, or creating new opportunities to preserve it. We identify underlying mechanisms of each effect on psychological ownership as well as relevant concepts to guide thinking and responses. To illustrate the value of our framework, we discuss these ideas in the context of three relevant macro trends in marketing: ( 1) growth in the sharing economy, ( 2) digitization of goods and services, and the ( 3) expansion of personal data. For each trend, our framework offers new predictions, opportunities for future research, and recommended marketing actions. We then note important caveats—cases in which psychological ownership could be undesirable or a liability to consumers and firms. We conclude by outlining next steps for consumer and strategy research within the three trends that we discuss in depth, and beyond, to other areas and broader questions.
Graph: Figure 1. Evolution of consumption: dimensions of change and examples.Notes: Consumption is evolving along two dimensions of change. Consumers are replacing legal ownership of goods with legal access to goods and replacing "solid" material goods with "liquid" experiential goods. Examples are sorted into quadrants; their location within a quadrant does not imply different values relative to others listed in that quadrant.
Psychological ownership occurs when one feels, subjectively speaking, that a thing is "MINE." It can be considered a form of emotional attachment between consumers and the goods and services they use ([107]). Antecedents of psychological ownership––perceived control, self-investment, and knowledge––do overlap with many of the property rights typically included in the "bundle of rights" provided by legal ownership of private goods ([85]). However, even though legal ownership may often precede psychological ownership, legal ownership of a good is not a requirement to feel psychological ownership for it ([100]). Consumers feel psychological ownership for ideas and goods to which they have no legal claim, such as theories and neighborhoods ([106]; [120]). At the same time, consumers feel little ownership for organizations and goods to which they do have legal claim, such as companies in which they hold stock and sports memorabilia they plan to sell ([72]; [98]). The Web Appendix provides a review of psychological ownership, including ( 1) motives and antecedents, ( 2) processes linking antecedents to outcomes, ( 3) consequences of psychological ownership, and ( 4) moderators and boundary conditions of these relationships.
Psychological ownership has value-enhancing consequences, which stem from an association of a good with the self and/or categorization of the good as "MINE." Due to psychological ownership, traits associated with the self and positive self-associations are transferred to the good, increasing emotional attachment to the good and enhancing its perception and value ([14]; [45]; [125]). Explicit categorization of the good as "MINE" appears to reframe the reference point from which it is viewed, changing the evaluation of the good from something that could be gained to something that could be lost. Loss aversion and the heightened attention to positive features of the goods that accompany this reframing increase its value, making people more reluctant to exchange it for money or other goods (for reviews, see [31]; [85]; [86]). Even goods that have more negative than positive features, if consumers actively choose to acquire them, benefit from the value-enhancing effects of psychological ownership ([126]).
Attachment between the self and good for which psychological ownership is felt parallels attachment between consumer and brand ([94]; [115]). As with an attachment between consumer and brand, psychological ownership for a good is positively associated with consumer demand, willingness to pay, customer satisfaction, relationships, word of mouth, and competitive resistance, as noted previously. Psychological ownership is thus a valuable asset for firms to preserve, capture, and redirect.
In short, documented effects of psychological ownership show it to be generally value-enhancing for consumers and firms ([31]; [86]; [95]). Our perspective is consistent with this evidence. Our focus is thus on how to preserve the value inherent in psychological ownership for goods, services, and brands in the face of technological change. Of course, there are exceptional cases in which consumers and firms find psychological ownership undesirable. To date, demonstrations of its liabilities have been limited to extreme cases, as when a good is associated with a personal failure or a disgusting stimulus ([70]; [73]). Subsequently, we identify more common instances in which consumers and firms may benefit from a decline in psychological ownership, an area ripe for future research to explore.
We propose that technological innovations are driving an evolution in consumption along two major dimensions. The first dimension of change is from a model of legal ownership, in which consumers purchase and consume their own private goods, to a model of legal access, in which consumers purchase temporary access rights to goods and services owned and used by others. The second dimension of change is from consuming solid material goods to liquid experiential goods. In this section, we unpack each change and how it affects psychological ownership. In general, we argue that the changes reduce psychological ownership and the value that accompanies it, but their effects are not uniformly negative. Table 1 identifies cases in which each change threatens psychological ownership; cases in which it transfers psychological ownership to other goods, groups, and brands; and cases in which changes in consumption patterns create new opportunities to preserve psychological ownership at prechange levels. Table 1 also includes recommended marketing actions to leverage each effect on psychological ownership, which are described in greater detail in the sections discussing the macro trends of the sharing economy, digitization, and personal data.
Graph
Table 1. Evolution of Consumption: A Psychological Ownership Framework.
| Dimension of Change | Threats to Psychological Ownership | Transfers of Psychological Ownership | Opportunities to Preserve Psychological Ownership |
|---|
| Legal ownership to legal access. Personal ownership of private goods is replaced with temporary access rights to use collectively consumed goods and services. | Fractional ownership. Bundle of rights associated with good divided among agents holding property rights to use, profit, change, or transfer ownership.♦ Emphasize liquidity and economic value. Impermanence. Consumers no longer expect to keep goods—they assume goods will be returned, impairing reference-point shift to owner ("My...").♦ Extend/guarantee duration and consistency of consumption experience.
| Collective consumption. Ownership felt for private goods transfers to goods used collectively ("MINE" to "OURS"). Reduced importance of individual goods, potential contaminated by dissociative group associations. Psychological ownership transfers to consumer communities.♦ Develop object history/intimate knowledge, encourage self-investment, deploy counterconditioning, and develop consumer communities.
| More consumer choice. Improved preference-matching due to more (often immediately) available options, increases perceived control.♦ Provide larger assortments, increase mass customization. New channels for self-expression. Social media and reputation systems integral to access-based consumption platforms provide new outlets for social signaling.♦ Develop social media applications and marketing strategy, encourage microblogging, offer access to aspirational brands/goods with positive signal value.
|
| Material to experiential. Material goods are replaced with physical or digital experiential goods. | Intangibility. Consumers are less able to touch, hold, and physically manipulate experiential goods than physical goods.♦ Develop haptic interfaces; interactive content; offer control over rate and timing of consumption; emphasize sensory features. Reduced evaluability. Ownership status is harder to determine (e.g., ownership of a vacation less clear than ownership of a vacation home).♦ Make goods indexical connections–cues for personally meaningful events (e.g., cross sell physical goods, usage history reminders); gamification.
| Higher categorization level. Category for which psychological ownership is experienced rises from individual goods to intermediary devices, platforms, and brands.♦ Vertical integration, brand alliances, servitization, relationship marketing, intermediary device personalization.
| Greater self-identification. Experiences are easier to integrate with self-concept than material goods (e.g., experiential purchases may generate more positive self-signals).♦ Leverage identity marketing (e.g., "I am a skier" > "I own skis").
|
1 Notes: ♦ = recommended marketing actions to manage psychological ownership threats, transfers, and opportunities.
In traditional capitalist markets, consumption of a private good was typically bound to sole, legal ownership of it. New access-based business models, made possible by technology-mediated platforms, fracture this model. Whereas property rights are typically bundled in private ownership (e.g., use, modify, profit from, or transfer rights; [58]), fractional ownership models unbundle property rights, allowing consumers to acquire a right to temporarily use goods and services that are often shared with tens, hundreds, or thousands of consumers (e.g., by paying for or sharing personal data; [30]; [123]). These models are distinct from previous models of collective consumption within families and communities ([38]). They relinquish ownership rights to firms and strangers and shift the goal of collaborative consumption. In collectives and families, the goal is to help others and facilitate relationship building. In access-based models, the goal is typically to provide financial or efficiency gains for consumers and firms ([68]).
Access-based models facilitate the creation of new products (e.g., social media platforms, video conferencing), and provide considerable benefits by changing the way existing products are consumed. By relinquishing private legal ownership of goods, access-based consumption offers consumers greater economic value, better preference matching, convenience gains from avoiding the entanglements of ownership (e.g., maintaining a car or vacation home), more sustainable means of consumption (e.g., digital books), and the use of both scarce and new goods that would otherwise be unaffordable or infeasible (e.g., luxury goods and social media platforms, respectively). The economic, temporal, and social benefits derived from the absence of legal ownership have been well documented (e.g., [ 8]; [57]; [69]; [101]). We argue that when access-based models induce a commensurate reduction in psychological ownership, however, there are negative downstream effects for consumers and firms. We briefly introduce how access-based consumption affects psychological ownership by threatening it, by causing it to be transferred, and by creating opportunities to preserve it.
Access-based consumption models threaten psychological ownership in two ways (see Table 1). First, fractional ownership models of access-based consumption divide property rights across agents, who may each possess one or more of the legal rights to ( 1) use a good; ( 2) profit from its use or sale; ( 3) modify the form, substance, or location of the good; or ( 4) transfer possession of some or all of these rights between agents ([52]). This change impinges on perceived control over access-based goods, a critical antecedent of psychological ownership ([ 5]). Second, the impermanence associated with access-based goods also threatens psychological ownership ([ 8]). Psychological ownership often entails the expectation that one will possess a good in the future. This expectation shifts the reference point from which the good is evaluated, as something that is to be lost, rather than as a potential gain. When consumers expect goods to be returned or relinquished, however, they do not shift the reference point from which they evaluate the good. They are users who perceive the good like a "buyer" would, not as an "owner" would. Users view its consumption as a temporary gain in their happiness or utility, not as part of a new status quo that will be lost when they give back the good ([86]).
Access-based models may also effectively transfer psychological ownership away from individual goods and toward consumer communities. Collective consumption of access-based goods may threaten psychological ownership for individual goods because they are used ([64]). They circulate among many consumers synchronously or asynchronously ([37]). Their circulation makes them interchangeable means to fulfill a goal. Therefore, consumers may use a good but not view it as "MINE" or unique or special ([80]). Their circulation also makes the symbolic meaning of access-based goods particularly vulnerable to contamination by dissociative social groups, persons, or acts ([60]). When consuming these used, circulating, or fungible goods, psychological ownership that would normally be directed toward an individual good ("It's MINE") may be replaced by psychological ownership of the group of consumers who use it ([41]; [97]). Collective psychological ownership is a feeling that all consumers of a good or service share ownership of it ("It's OURS") and gives each consumer a claim to membership, belonging, and ownership of the community formed ([97]).
Finally, we see two opportunities for access-based consumption models to preserve psychological ownership at levels commensurate with the level observed for private goods. First, access-based consumption offers large assortments to consumers. More consumer choice could increase feelings of psychological ownership for goods and services through the greater control it provides to consumers ([59]; [87]). A second opportunity stems from the new channels for self-expression that access-based models provide. Self-expression is a fundamental motive driving the desire to own and consume ([15]), and access-based consumption facilitates this identity signaling ([17]). Access to more choices within and across product categories, and to new channels such as social media platforms, provides consumers means to more precisely signal authentic and desired identities as well as to accumulate social capital, attention, and future economic gain ([ 6]; [41]; [67]).
New technologies are replacing "solid" material goods (i.e., tangible objects that are acquired and owned by consumers) with "liquid" experiential substitutes (i.e., events or experiences that one encounters and lives through) to fulfill a variety of hedonic and utilitarian wants and needs ([ 7]; [11]; [17]; [47]). This mirrors a shift in consumer demand, driven by millennials but also applicable to other generations, whereby consumers now prefer to spend money on experiences rather than things and have increased the share of their income spent on experiences ([ 9]). Beyond the multitude of new experiential offerings made possible through the expansion of the sharing economy, digitization, and an information economy driven by personal data (discussed subsequently in detail), firms are making significant investments in servitization and experiential offerings. Firms now offer a variety of product-focused services and experiences to consumers postpurchase. In many cases, even the acquisition of material goods is becoming refocused on its experiential components. Brick-and-mortar retailers, seeking differentiation from more convenient online platforms, for instance, have embraced "experiential shopping" (or "shoppertainment") with pop-up shops, live events, interactive displays, activities, product lessons, and interactions with experts ([44]).
Many goods could be classified as material or experiential (e.g., a DVD is a tangible material object, but the film it plays is an intangible experience). Our classification scheme sorts goods according to the focal acquisition goal—to have a thing or an experience. A consumer could acquire an album with the goal to expand her record collection, or to listen to the music pressed into its vinyl form ([25]). Even traditional solid goods (e.g., cars, computers, phones, watches) are often now also sold with accompanying experiential features (e.g., applications such as GPS, music streaming, and games). We predict that eventually the material versus experiential distinction will be blurred to the extent that consumers will view most goods as experiential by default. Next, we briefly introduce how the change from material to experiential consumption affects psychological ownership by threatening it and causing it to be transferred, as well as how this change creates opportunities to preserve it.
Two threats to psychological ownership arise from the substitution of material goods with experiential goods. The first is the intangibility of experiential goods. Psychological ownership is typically imbued through physical cues such as holding, touching, and manipulating a material object, which instantiate perceived control over it ([95]; [100]). This lack of physical interaction should consequently reduce psychological ownership for experiential goods—and, thus, their value—to consumers ([ 4]).
A second threat to psychological ownership is the reduced evaluability of ownership––the difficulty evaluating who owns experiential goods, such as determining which property rights belong to consumers, owners, and intermediaries ([11]; [25]). When a consumer buys a concert ticket to a live event, what rights does that afford her other than access to the show? Can she be denied admission if she fails to comply with security and health protocols? Can she film it for personal consumption or share her recording on social media? Whether a consumer, intermediary, or firm "owns" an experience is often ambiguous, even when firms strive to make legal ownership transparent (e.g., who holds which property rights), and is muddled further when firms make legal ownership strategically opaque. Consumers who buy digital books, for instance, often mistakenly believe they have purchased more than the right to permanently view them ([56]).
If consumers think of experiential goods at a higher categorization level than similar material goods (i.e., at a more abstract level), psychological ownership may transfer from individual goods (e.g., a book) to branded services, platforms (e.g., Audible), or technological devices used to consume them (e.g., a tablet). Vertical transfers may direct psychological ownership for material goods to brands of experiential goods or the platform through which experiential goods are accessed. Self–brand attachments may strengthen, and possession–self attachments may weaken, as experiential goods replace material goods ([32]; [39]). If psychological ownership manifests at the brand level, it can have positive downstream effects on consumer demand. Germans who felt more psychological ownership for a car-sharing service more frequently booked cars from that service, and students who felt more psychological ownership for a music streaming platform reported using it more often each week ([41]). Horizontal transfers may direct psychological ownership from material goods to the intermediary devices used to access experiential goods. Phones, computers, smart panels, watches, and other technological devices may accrue greater psychological ownership, value, and significance in the eyes of consumers (e.g., [81]).
One opportunity to preserve psychological ownership at levels commensurate with feelings for material goods comes from consumer's greater self-identification with experiential than with material goods (e.g., a trip to Italy vs. an Italian jacket; [25]; [46]). We posit that the more positive social signal provided by experiential than by material purchases ([10]) may undergird their potent value as self-signals. Consumers may forge stronger attachments to experiential than material purchases, because they are more socially appropriate means with which to define the self.
As evidence of the value of our psychological ownership framework, we present three macro trends in marketing disrupting existing business models, whose effects on consumer behavior are mediated by changes in psychological ownership: ( 1) growth in the sharing economy, ( 2) digitization of goods and services, and ( 3) expansion of personal data. We selected these trends because they are disrupting the marketplace and are active foci of interdisciplinary research. For each trend, following our framework, we identify specific threats to psychological ownership, transfers of psychological ownership to other stimuli, and opportunities to preserve psychological ownership at prechange levels. Marketing actions are then recommended to counter the threats and leverage transfers and opportunities. Exemplary case studies appear in Table 2 (ride sharing), Table 3 (digital music), and Table 4 (health and wellness), which concretely illustrate the explanatory power of our psychological ownership framework for scholars and practitioners.
Graph
Table 2. Case Study #1: Ride Sharing.
| Dimension of Change | Threats to Psychological Ownership | Transfers of Psychological Ownership | Opportunities to Preserve Psychological Ownership |
|---|
| Legal ownership to legal access. Private ownership of a car replaced with temporary access rights to use a collectively consumed car. | Fractional ownership. The right to drive, sell, and control use of a car reduced to access to specific rides.♦ Emphasize cost savings and convenience of not owning a car. Impermanence. Each ride is with a different car and driver, impairing development of psychological ownership.♦ Repeat service delivery with favorite vehicle types, makes, models, and drivers.
| Collective consumption. Private use of a car is replaced by use of cars in a fleet that circulates among a group of consumers, some potentially diseased (e.g., "covidiots").♦ Provide car features, driver history, celebrity brand ambassadors, and high sanitary standards; ask users to help keep cars clean; develop consumer communities (e.g., Uber Pool).
| More consumer choice. Improved preference-matching between car type, user, and occasion increases perceived control.♦ Optimize assortment of transportation options for specific uses (e.g., airport trips, commuting, dining out, groceries). New channels for self-expression. Positive feedback and displaying aspirational brand use on social media facilitate social signaling♦ Two-sided reputation systems, aspirational offerings (e.g., rider ratings, luxury/exotic vehicles).
|
| Material to experiential. Ownership of a material car is replaced with access to the experience of a car ride. | Intangibility. Consumers are less free to touch and manipulate ride experience than their own physical cars.♦ Provide choice of routes, sensory settings (e.g., temperature, conversation, music). Reduced evaluability. Ownership status is harder to determine; ownership of a ride is less clear than ownership of a car.♦ Provide consumers with record of trips, cars, drivers, and history with platform; gamify travel (e.g., pin map with landmarks visited).
| Higher categorization level. Psychological ownership shifts from a specific car to smartphone, platform, or brand.♦ Marketing emphasis on relationship with platform (e.g., Uber), optimizing customer satisfaction (mobile applications, experience).
| Greater self-identification. Goal of ride easier to integrate with self-concept than physical stimuli (e.g., road trip versus type of car driven).♦ Identity marketing (e.g., minimal, sustainable lifestyle—use car only when necessary).
|
2 Notes: ♦ = recommended marketing actions to manage psychological ownership threats, transfers, and opportunities.
Graph
Table 3. Case Study #2: Digital Music.
| Dimension of Change | Threats to Psychological Ownership | Transfers of Psychological Ownership | Opportunities to Preserve Psychological Ownership |
|---|
| Legal ownership to legal access. Privately owned albums replaced with temporary access rights to use collectively consumed albums, songs, and videos. | Fractional ownership. Rights to use, sell, share, or gift an album are replaced with access rights to album, song, or platform catalog.♦ Emphasize cost savings, convenience. Impermanence. Permanent ownership is replaced with access rights contingent on composition of platform catalog or longevity of software or firm.♦ Maintain consistency in offerings (e.g., recordings), guarantee long-term access to purchases.
| Collective consumption. Listening to a private library of music replaced with consumption of a catalog available to all platform users; ownership transfers from album to consumer group.♦ Provide information about recordings and artists; feature artist/influencers in marketing communications; make opportunities for cocreation (e.g., playlist, remixes); cultivate consumer groups (e.g., events, social media marketing).
| More consumer choice. Access to larger libraries increase match between state-dependent preferences and music available.♦ Provide omnichannel (mobile, desktop, offline) access to more songs, artists, and recordings in platform catalogs. New channels for self-expression. Consumers comment, review, discuss music (e.g., Twitter, YouTube, Reddit); create and share new music (e.g., SoundCloud).♦ Encourage microblogging, reviews, editing and publishing tools, increase access to new and rare recordings.
|
| Material to experiential. Physical records, tapes, and CDs are replaced by songs, downloaded to or streamed on personal device. | Intangibility. Consumers are less able to touch, hold, and manipulate digital music than physical records, CDs, tapes.♦ Use touchscreen and gesture-based menus and controls; skeuomorphic controls (e.g., virtual turntables); include album covers, videos, and samples in music. Reduced evaluability. Ownership of downloaded and purchased digital album is more ambiguous than ownership of a physical album.♦ Visual ownership and usage cues (pictorial menus, playlists), cross-sell physical merchandise (branded apparel, posters, household goods), gamification (top songs, percent of favorite artist's library heard).
| Higher categorization level. Psychological ownership transfers from album to smartphone, headphones, or platform.♦ Emphasis on relational marketing, develop mobile applications, personalization of intermediary devices (e.g., customizable headphones).
| Greater self-identification. Consumers more readily identify with artist or song than physical album/CD/tape.♦ Provide history of songs, artists, albums (e.g., lyrics, biographies, discographies), connect artists with salient social identities and causes.
|
3 Notes: ♦ = recommended marketing actions to manage psychological ownership threats, transfers, and opportunities.
Graph
Table 4. Case Study #3: Health and Wellness Data.
| Dimension of Change | Threats to Psychological Ownership | Transfers of Psychological Ownership | Opportunities to Preserve Psychological Ownership |
|---|
| Legal ownership to legal access. Private paper office records are now accessed and shared through platforms, applications, and intermediary firms. | Fractional ownership. Private records controlled by consumer are replaced with electronic data shared without knowledge by firms and third parties through data exchange (e.g., Cures Act).♦ Emphasize benefits of accurate and accessible health and medication history. Impermanence. Permanent paper office records are replaced with electronic records contingent on platform longevity (e.g., MyChart).♦ Standardize records platform across providers; guarantee access to records.
| Collective consumption. Private health and fitness data are replaced by data that are collectively consumed (e.g., heart rate displays in fitness classes). Health status ownership/identity transfer from individual to social group (e.g., "My diabetes." to "Our diabetes.").♦ Develop patient communities (e.g., collective goals), solicit self-investment.
| More consumer choice. Consumers gain new opportunities to select and manage data inputs, outputs, and visualizations from tests and medical devices, review records and results online (e.g., 23andMe, Apple Health)♦ Increase data integration and personalization across devices.New channels for self-expression. Consumers can disclose health and wellness data to social media or applications (e.g., Fitbit; Nike+; Peloton).♦ Encourage microblogging, offer social media applications.
|
| Material to experiential. Physical medical records indicative of health status replaced with in vivo electronic dashboards plotting health over time (or in real time). | Intangibility. Physical records of and interactions with patient at doctor's office, replaced with cloud-based electronic records and communications.♦ Increase consumer control over how and when they consume their data. Reduced evaluability. Greater ambiguity for ownership of continuous heart rate data than static report (e.g., app display vs. report from doctor).♦ Increase access to longitudinal data and account personalization (e.g., trends in health states, photos and avatars); gamification of goals, states, activity (e.g., miles run, REM sleep).
| Higher categorization level. Psychological ownership transfers from private records to intermediary devices and platforms used to record or display data (e.g., wearables, MyChart).♦ Relational marketing, personalize intermediary devices (e.g., smartwatches).
| Greater self-identification. Increase in data provides deeper portrait of health status and history, increasing identification with it (e.g., light/deep sleeper, low heart rate).♦ Health status treated as social identity in positioning and marketing communications.
|
4 Notes: ♦ = recommended marketing actions to manage psychological ownership threats, transfers, and opportunities.
Sharing has traditionally been restricted to familiar others, such as family members and homogeneous collaborative or cooperative social groups ([68]). The new sharing economy is comprised of strangers, who together participate in "a scalable socio-economic system that employs technology-enabled platforms that provide users with temporary access to tangible and intangible resources that may be crowdsourced" ([30], p. 7). Its many forms of collaborative consumption include renting, reselling, lending, simultaneous consumption, and resource pooling ([20]). Sellers provide temporary usage rights for unused goods in exchange for profit. Buyers acquire access rights to those goods without worrying about outright purchase or upkeep. Thus, value is created for both parties ([35]; [68]). Sharing platforms lower matching costs between sellers and buyers, and secure the exchange of money, by strengthening trust through reputation systems ([ 7]; [29]; [112]).
The staggering growth of products available and platforms for sharing, including bicycles, boats, cars, clothes, homes, offices, rides, and scooters (e.g., Airbnb, Bird, Blue Bikes, Lyft, Poshmark, Rent the Runway, Turo, Uber, WeWork) may threaten the long-term viability of private ownership. For instance, personal car ownership declines when sharing is a viable option ([83]), perhaps most for those who do not see car ownership as central to their identity ([18]). As an example, Table 2 illustrates how ride sharing threatens, transfers and creates opportunities to preserve psychological ownership.
Fractional ownership models prevalent in the sharing economy threaten psychological ownership, whether access-based goods are rented in exchange for payment or borrowed for free. Consumers report feeling less psychological ownership for rented goods than goods they privately own. This gulf is widened when goods are free. Consumers feel less psychological ownership for borrowed than rented goods. Indeed, they feel no more psychological ownership for borrowed goods than goods they merely evaluate ([ 5]). Marketing actions can be taken to counter threats posed by fractional ownership. First, marketers could emphasize the benefits of reduced costs and dependencies when forgoing legal ownership (e.g., avoiding car payments, gasoline, parking, cleaning, insurance, and general maintenance; [57]). Second, firms can recruit consumers as both users and suppliers, or "prosumers" ([30]; [102]). Seeing the transaction from the role of supplier should increase value by increasing consumers' attention to what is gained through fractional ownership ([86]).
A second threat to psychological ownership from sharing markets is that consumers rightly expect their ownership rights and possession of goods to be temporary. Marketers could counter this threat by extending access to goods and services consumed in the present, or promising future access to those particular goods and services ([31]; [100]). A dress could be lent for longer, a ride-share platform could provide consumers with frequent access to their highest-rated vehicles and drivers, or a home rental service could give a consumer first claim to her favorite past rental on the same set of dates each year.
In the sharing economy, consumers interact with individual goods, but those goods are not the goal of consumption. The goods are fungible means to an end. Most consumers use a ride-share platform for transportation, for example, not to have the experience of riding in a particular car. The ensuing transfer of psychological ownership from individual goods to user communities can create a "tragedy of the commons" ([54]), whereby individual users take less care and responsibility for a shared good than they would if it were theirs alone. [ 7] note such negative reciprocity for car sharing. Contamination concerns may also loom large in the sharing economy. Consumers may be disgusted by sleeping in a bed in a rental property that has been slept in by many others, or worried about riding in a car previously used by a sick passenger.
Multiple marketing actions can be implemented to preserve psychological ownership with such transfers. One marketing action to counter the lack of a unique relationship with any particular good may be to emphasize what is unique about the goods, such as their features, history, or owner ([49]; [71]). Second, beyond maintaining and advertising high standards for sanitation, background checks, and screening for irresponsible users, firms may use counterconditioning ([78]). Attractive, trustworthy brand ambassadors and clean and modern goods may counter the negative associations from dissociative groups and contamination concerns ([ 2]). Third, marketers could also try to retain psychological ownership at the group level, developing consumer communities around common geographic regions, interests, or goals (e.g., Uber Brooklyn; Uber Coachella; Uber Pool for work). Membership in such groups could reduce behaviors associated with reduced personal responsibility, such as obstructing sidewalks with electric scooters, and increase the attractiveness of sharing goods as a substitute for private goods ([41]).
A shift from legal ownership to legal access also offers opportunities to preserve psychological ownership. More ride-sharing options enable users to better satisfy unique needs than car-buying consumers with one vehicle for all purposes (e.g., commuting, grocery shopping, travel). Decision aids may facilitate such preference matching. Soliciting the purpose of a trip or inferring it from locations (e.g., restaurants, airports), may allow a ride-sharing service to recommend suitable transportation options (e.g., a large SUV to carry luggage). Platform design can incorporate customization opportunities, such as choosing the brand of car or music in a ride share, the color of an outfit, or the towels and bath products in a home rental. Firms can also coordinate matches between customers and goods, such as when hotels configure mutable features of rooms to loyalty program member preferences (e.g., minibar, pillows). Psychographics should enable firms to target promotion-focused consumers willing to take risks with novel experiences and product categories, particularly as product trials are freed from the costs of long-term ownership.
Another opportunity to preserve psychological ownership is via self-expression, expressing preferences and identities with goods that would otherwise be unaffordable or untenable to consumers. A student might rent a designer gown through a platform for a special occasion or social media post. A couple on a date night might treat themselves to a ride in a limousine, a car that would be impractical and onerous for them to privately own. Being able to use and broadcast use of aspirational and luxury goods through sharing platforms may produce greater identification with, psychological ownership for, and loyalty to brands accessible through the platform, which consumers may not normally buy. This includes goods used infrequently (e.g., formal attire, party supplies), that are costly to maintain (e.g., boats, vacation homes), or that are expensive to buy (e.g., handbags, yard equipment). Firms may further benefit from facilitating user posting of experiences on social media for social signaling and from soliciting user feedback. Vacationers may feel greater attachment to a rental after sharing pictures of it, or after expressing their values by writing a review of the home ([55]).
In the sharing economy, consumers may remain in physical contact with "solid" material goods, but the focal goal is not to own material goods. It is to consume goods in "liquid" experiential forms ([ 7]; [30]; [101]). A ride-share user purchases a ride, not a car. A vacationer purchases access to a home, not the home itself. A freelancer buys access to a workspace and its amenities, not the property on which she works. A first threat is raised by the intangibility of such experiential goods. This reduces physical control, and thus perceived control over the consumption experience. To offset this threat, marketers could use techniques that restore control through other dimensions, such as providing consumers with touchscreen interfaces (e.g., smartphones; [21]), or control over when and how goods will be consumed (e.g., scheduling rides and routes; [13]), the sensory features of the experience (e.g., temperature, music), and less tangible options (e.g., interactions with the driver or owner; [103]).
Second, the rights afforded by the purchase of a shared good (e.g., a ride, rental of a vacation home) are more subjective and less evaluable than the rights afforded by private ownership of good (e.g., a car, a home; [11]; [25]). Consumers buy a contract for a ride from point A to point B, or to use a house for several nights, but which rights are included in that contract can be ambiguous. The end result is that consumers may not be able to discern (or feel) ownership of the experiential good they have purchased. To enhance the evaluability of owning shared experiential goods, marketers could cross-sell or bundle private material goods that serve as a marker of the experiential purchase. Tangible goods can serve as reminders of personal memories and meaningful consumption episodes ([121]). The French Laundry gives diners a branded wooden clothespin, for instance, as a souvenir of their extravagant meal. Such cues create value through the indexical connections they form, tangible links between consumers and meaningful events ([50]). Platforms could provide consumers with other cues such as usage history records or gamify use, such as by pinning maps with landmarks visited.
Psychological ownership for the concrete, tangible, material goods used in the sharing economy may be transferred to the more abstract, intangible branded platforms and intermediary devices through which experiential goods are accessed. While this may reduce psychological ownership for any individual experience, positive effects of this transfer could include higher brand loyalty, competitive resistance, and word of mouth for brands and intermediary devices ([ 3]). We recommend that marketers emphasize the relationship with the platform in their strategy and actions. Consumers may care less about how the particular brands of cars available through a ride-share platform reflect on their identity, for instance, than the fairness of its prices or its treatment of drivers.
The sharing economy may afford particular opportunities to preserve psychological ownership. Consumers may more readily identify with collections of unusual experiences (e.g., renting a 1980s Mercedes convertible while vacationing in California) than with material merchandise that does not reflect their authentic selves (e.g., buying the same convertible to drive to work; [63]). A consumer can purchase experiences to signal that she is adventurous or on trend ([ 7]; [16]). Firms positioned toward identity marketing could target consumers who identify as "minimalists," who prefer to avoid entanglement in the responsibilities of ownership ([57]). The appeal of using products collectively could be highlighted to appeal to consumers who identity with sustainable consumption, and firms could address their environmental concerns with premium sustainable offerings (e.g., electric cars, passive houses).
Digitization of goods and services, wherein information is converted into a numerical format, has evolved from niche scientific and commercial applications in the 1950s and 1960s into a technology that has spread across and transformed society. Consumers exhibit strong demand for digital goods. There has been a recent rise in consumer demand for some vintage physical goods such as vinyl records ([90]), but many analog products and services have been, or are being, replaced by digital substitutes. Digital cameras outsold analog camera sales by 2003. Both were outsold by smartphones in 2006, which were used to take most of the more than 1 trillion photographs taken in 2017 ([23]). By 2018, record labels earned more through streaming services than physical CD sales. Mass digitization of millions of books is currently underway by Google, the Open Content Alliance, and Microsoft ([27]). Digital currencies, from dollars to information-based currencies such as Bitcoin and Ethereum, may eventually replace cash.
Digital goods provide similar consumption experiences as their physical counterparts, but their immateriality confers numerous advantages. A digital photograph can be shared instantly with friends and family members. It can be recovered even if the phone used to take it is lost or broken. Digital music and books can be purchased and accessed at home, on the beach, or in the air––anywhere with wireless access—from a pocket-sized device, never scratching, fading, or tearing. Digital goods have many environmental benefits, from lower carbon footprints to no waste on disposal ([82]). Effects of digitization on psychological ownership for goods, and its downstream consequences, are less clearly positive. As an example, Table 3 illustrates how digitization threatens, transfers, and creates opportunities to preserve psychological ownership of music.
Digitization is replacing permanent ownership models with access-based consumption models in many domains ([30]; [123]). In the case of music, private ownership of physical albums is being replaced with access-based consumption of digital downloads and streamed music (Table 3). Streaming is now the most popular way to consume music. Diffusion of digital access-based models is also widespread for books, email, films, magazines, maps, news, and television.
Access-based consumption of digital goods typically entails the temporary right to use a good, housed on a cloud server, which is owned and fractionated by a third-party provider. Consumers cannot sell, trade, or gift digital goods for which they purchased "permanent" access; they have only purchased a right to personally consume it. Consumers often do not even own digital consumption objects they create (e.g., annotated books, avatars in games, playlists). We suggest that this fractional model of ownership threatens the psychological ownership felt by owner-users, potentially transferring perceived ownership to the platforms and brands providing consumers access to digital goods. Indeed, consumers feel less psychological ownership and are thus less willing to pay for digital books, films, and photographs than their physical counterparts, ([ 4]; see also [108]). In addition, even though users spend more than an hour of their time each day on social media platforms each day, they are willing to forgo access to their content and online social networks for relatively small sums of money ([22]). Marketing actions for firms to address this threat could highlight the considerable economic and transactional benefits of access-based digital goods, which are often more attractive than the benefits of legally owning private goods ([109]).
Second, consumers (rationally) view their ownership of access-based digital goods as impermanent. Streamed goods are often not even rented. Consumers pay for access to a platform's catalog, and individual goods are only possessed for the duration of their consumption. The ability to consume access-based digital goods—even goods that consumers themselves created—is typically determined by the platform on which they are hosted ([84]). Consumers may thus not feel ownership even for the digital goods they can "permanently" access. Indeed, consumers are willing to pay more to purchase than rent utilitarian physical goods (e.g., a hardcover textbook), but they are not willing to pay more to purchase than rent similar digital goods ([ 4]; [ 5]). We suggest that marketers respond to impermanence threats by assuring consumers that they will have continued access to the same digital goods. Platforms could extend streaming access to favorite titles in their catalog, or guarantee access to digital goods purchased "permanently" for a specified time period. When updating platform designs and formats, we conjecture that retaining elements that instill a perception of continuity may reduce this threat.
Issues around transfer of psychological ownership due to the collective consumption of digital goods raise different concerns than those described in the sharing economy. Digitization should mitigate physical contamination of goods, but consumers may still be concerned about acquiring digital goods from dissociative groups, who may add malware or viruses. We speculate that contamination may also affect digital goods at higher construal levels. Whereas consumers may be primarily concerned with the previous owners of one copy of a physical good (e.g., "This paperback of The Fountainhead was owned by a white nationalist"), consumers may be concerned with the previous and other owners of any copy of a digital good (e.g., "The Fountainhead is popular on Facebook with white nationalists"). As contamination effects become more diffuse, however, they may also become more diluted. Contamination may be more potent when it applies to one rather than to all copies of a particular good. As digitization facilitates the coordination of social groups around collective activities and interests (e.g., games, music, news, photography, design, literature, videos), ownership for goods may be replaced with ownership for these consumer communities ([97]). Consumers may feel psychological ownership for the community itself as well as for their contributions that further the goals and formation of these groups (e.g., posts, comments, virtual objects).
Marketing actions to retain psychological ownership for an individual digital good include providing consumers with more information about its background (e.g., history; critical reviews and summaries; information about individual artists, actors, or musicians involved in its production; [71]), and counterconditioning by featuring beloved artists, awards, or celebrity users in marketing communications for the good (e.g., social media influencer endorsements). Marketers who aim to benefit from the transfer could grow consumer communities by creating officially licensed clubs, posting content in spaces where consumers interact with each other and brands or artists (e.g., Facebook fan pages, Twitter), and providing consumers ways to engage with and invest their time and energy in digital objects and these social groups (e.g., hosting forums, posting reviews and comments, creating collaborative quests and interconnected worlds; [40]). That investment is likely to foster a feeling of psychological ownership for digital consumption objects (e.g., avatars, posts, virtual cities; [61]; [91]), which have considerable value for firms as means to lock in consumers to their platforms ([84]).
Digitization provides opportunities to preserve psychological ownership through the panoply of options and channels for the self-expression it affords consumers. Digital goods enhance control and provide consumers with large assortments of content to match their preferences. Consumers typically can choose which digital media to consume anytime, anywhere, with even more choice on the go than when choosing similar kinds of physical goods at brick and mortar retailers (e.g., books, games, movies, music). Digital goods can also enhance control by facilitating the personalization of consumption experiences. The increased control imbued by enhanced consideration sets and customization may create a greater level of psychological ownership than is experienced for comparable physical goods ([59]; [87]). Marketing actions that can leverage these benefits include maintaining large choice sets, even as recommendation systems improve ([62]), offering consumers ways to customize their consumption experiences, and direct control over those experiences or the content offered (e.g., in games or media feeds). Low marginal costs and image filters for digital photographs, for instance, allow consumers to capture many images of the same subject and edit the photograph that best realizes their vision ([118]). As illustrated by the consumer backlash against Apple for adding U2's Songs of Innocence album to user libraries in 2014 ([12]), firms should avoid curating consumer content without their explicit consent.
A second opportunity to preserve psychological ownership stems from the many new ways digital goods allow consumers to create and signal their identity to others through the cocreation of public digital consumption objects. Indeed, consumers invest considerable labor in creating and curating their image, content, and contacts on social media, in games, and in online virtual worlds ([84]). Marketing actions that facilitate these forms of self-branding and identity signaling would provide consumers with ways to share their preferences for and consumption of digital goods through social media and recommendation systems, and by including aspirational digital goods in their catalog of offerings (e.g., Pinterest walls, upvotes and downvotes, digital artifacts, new or exclusive content).
Digitization, by definition, translates analog material media to an immaterial digital format that can be transmitted and consumed experientially through a variety of devices, including computers, smartphones, tablets, headphones, radios, and wearable devices. Digitization can also facilitate new material forms of consumption and exchange. For example, 3D printing may present consumers with new ways to buy, share and create material goods, based on digital plans acquired from business-to-customer or customer-to-customer markets, exchanges, or collaborations.
One threat posed by this transformation is intangibility. The immateriality of digital goods imbues them with many remarkable benefits but prevents consumers from having physical interactions with digital goods ([21]; [95]; [100]). Consequently, consumers are less likely to establish a feeling of psychological ownership for digital goods, which leads them to value digital goods less than similar physical goods ([ 4]). Marketing actions to directly address this threat include interfaces that restore physical cues signaling control ([21]), allowing consumers to control the rate, time, and place at which digital goods are consumed ([13]) and positioning digital goods along sensory dimensions where they outshine physical analogues (e.g., [103]). Digital games allow consumers to navigate virtual worlds with joysticks, touchscreens, or their bodies (e.g., Xbox Kinect), for instance, to play at any time with people around the world and explore complex novel worlds. Online courses might benefit from haptic annotation tools, the ability to watch lectures at accelerated rates or asynchronously, the opportunity to save screenshots of slides and whiteboards, and novel animations that would be infeasible to incorporate in offline courses.
A second threat to psychological ownership is reduced evaluability. It is often difficult to determine who owns experiential, digital goods ([92]). Consumers may incorrectly identify who owns the rights to share and transmit the goods, particularly in contexts where they are allowed to share physical goods. A consumer might see that it is illegal to sell a stranger access to her streaming account but will freely share access with roommates or family members. Beyond cross-selling and bundling physical goods with digital goods to create physical reminders of ownership (e.g., toys, clothing), digital goods may be able to serve as indexical reminders of meaningful memories by incorporating usage history features that identify when, where, and with whom they were consumed. Digital photographs, for instance, already include information about their date, location, and the people included in the photograph. Digital goods are ripe for gamification, whereby levels of ownership may be indicated by completion of real or arbitrary goals and status levels (e.g., pages read each week).
Digital goods may lead consumers to transfer psychological ownership from the particular good being consumed (e.g., "My LP") to higher levels of categorization or abstract properties of the consumption experience, such as the genre, artist, recording, brand, or platform (e.g., "I'm listening right now to Kind of Blue by Miles Davis on my Spotify playlist"). This could also lead consumers to feel greater ownership for the services and intermediary devices they use to consume digital goods, such as platforms and smartphones ([41]), as those touch points will be the primary means by which consumers control experiential goods ([13]). We suggest that digital goods are likely to be perceived more as services than goods. Consumers expect interactions with firms to entail the delivery of a consumption experience or experiences over time and to be an enduring relationship, rather than a fleeting transactional exchange (e.g., buying access to stream an evolving catalog of music vs. buying a vinyl album, respectively). Firms need to adapt their marketing strategy toward this service orientation in the minds of their consumers. Problems with digital goods, for example, are thus likely to be perceived as service failures, and strategies to maintain customer satisfaction may need to change. On the upside, servitization may then become a potential route through which to preserve psychological ownership at the brand level. Depending on the level at which psychological ownership manifests, brands may need to retain and develop consumer brand attachment through vertical integration or brand alliances that allow them to sell intermediary devices, which may become important means of self-expression (e.g., recognizable designs for smartphones, headphones, laptops).
One opportunity to preserve psychological ownership is that the experiential nature of digital goods may increase consumer identification. Identity marketing strategies, such as emphasizing associations or the fit between digital goods and salient consumer identities (e.g., trendiness or sustainability) may be particularly effective ([19]). Given their flexible categorization, if digital goods are marketed as experiences rather than as digital substitutes for material goods (e.g., as readings of books by their authors vs. as audio books), consumers may more strongly identify with their consumption and feel levels of psychological ownership comparable to that felt for their material substitutes.
The expansion in the recording of and analytics to manage and use personal data, defined as "any information that relates to an identified or identifiable living individual" ([33]), is fundamentally changing life and business, particularly how marketing is done for firms and experienced by consumers ([93]). Technological advances in collection, storage, and analysis as well as the transformative shift to online search, shopping, and fulfillment has both enabled and enhanced the value of firms using consumer data to power their marketing decisions. Consumers are realizing that their personal data have significant value ([76]). They want a share of that value as well as protection of their privacy ([99]). Regulatory bodies are dramatically increasing the legal ownership rights of consumers to their personal data by requiring consumers to "opt in" to permit firms to use/sell the data (e.g., General Data Protection Regulation, California Consumer Privacy Act; [28]). In early 2020, two U.S. states have passed and nine other states are in final stages of passing new consumer data regulations, where "we're witnessing the beginning of a massive shift toward protection for consumer data and accountability for businesses that control and process it" ([104], p. 1).
The changing regulatory policies illuminate a tension between firms and consumers with regard to who owns the incredible breadth and depth of personal data. Firms try to capture as much data as possible on potential and existing customers to target the "best" consumers with the right products at the right time, increasing sales and profits. This data, once constrained to the history of a consumer at a single business, is increasingly associated with identity-relevant information about all facets of their lives (e.g., locations visited, photographs and videos, search history, medical and genetic information). In this context, firms would like to reduce consumers' psychological ownership of their personal data because this would promote consumer sharing their data with fewer restrictions or needs for compensation. As emerging firms (e.g., Datawallet, Midata) offer consumers opportunities to regain control of their personal data and sell it to firms, consumers may become more concerned with retaining ownership rights ([ 1]). Understanding these changes and identifying heterogeneous segments will be key to effective marketing strategies related to personal data and consumer privacy. As an example, Table 4 illustrates how the expansion of personal data threatens, transfers, and creates opportunities to preserve psychological ownership of health and wellness data.
In the past, consumers received and saved paper copies of their financial transactions, providing them physical ownership of these data. Now, consumers receive online access to platforms of financial intuitions providing cloud-based digital records of their personal financial data on as-needed basis. In government and business sectors, digitization is rapidly replacing physical documents with digital files from taxes to driving and medical records (e.g., [24]). Housing consumer data and giving consumer online access can result in switching barriers and consumer loyalty ([26]), but we argue that this model is changing consumer psychological ownership of their personal data.
First, access-based models are fractionalizing data ownership. Data is becoming more distributed, which could threaten consumers' psychological ownership of their data. Once private to consumers, data is now gathered and sold (or shared) by companies to third-party vendors (e.g., advertisers). The results of genetic testing were once accessible only to the consumer and her doctor. Firms such as 23andMe now offer consumers access rights to their genetic records, which are also shared (anonymously) with the parent company, other firms, and researchers. Tax records were once physical documents consumers prepared (perhaps with an accountant) and submitted to the government, keeping private physical copies stored in their files. Now taxes are prepared through intermediary platforms that keep a digital record, which the platforms use to market credit cards and loans back to their consumers. Even private copies of records stored by consumers in an electronic form may be accessible to cloud server hosts (e.g., Dropbox, Google). Location data, once exclusive to consumers, is now tracked by phone companies, government, GPS, and sold for profit (e.g., for mobile advertising).
Initial technological and purchase trends associated with fractional ownership reduced consumer data privacy (social media, peer-to-peer payments, online shopping), but this is being offset by new technologies (blockchain, two-factor authentication) and regulations addressing data privacy concerns. Privacy and anonymity can be provided in exchanges by the use of cryptocurrency (e.g., Bitcoin), blockchain open source commuting platforms (e.g., Ethereum), or emerging decentralized autonomous organization, a complex form of smart contracts using token governance rules ([128]), which offer multiple research opportunities. Marketers may find that these technologies give consumers real and perceived control over their data, reducing threats to psychological ownership posed by fractional models of legal ownership.
Second, the perceived impermanence of personal data threatens psychological ownership in situations where electronic access replaces permanent storage of a "hard copy" (e.g., lab reports, tax returns). As with digital goods, access to these data depends on the longevity and security of the hosting platform. When platforms hosting data close, or organizations change where their data is housed, data not transferred to new platforms may be lost. The frequency and scope of data breaches and ransomware attacks are additional salient reminders of the impermanence of personal data, even when firms prioritize privacy ([77]). Marketing actions providing consumers with the permanence necessary to preserve psychological ownership for their data may include long-term file storage, and continuity in file structures and platform interfaces. Providing real safeguards and privacy protections should be an effective marketing strategy to attract consumers with security-based psychological ownership concerns (e.g., Datawallet, DuckDuckGo, Midata).
A change in the consumption of personal data and experiences may transfer psychological ownership from the individual to the collective space ([61]). Most consumer data were formerly consumed individually or among family members. Now, with the increased availability and consumption of metadata, social media, community forums, and other network-based apps, those data are now often consumed jointly or collectively. Power companies present the energy consumption of individual households and their neighbors side by side ([105]). Patients share information in online health forms about their health conditions with strangers ([113]), which may provide them with a feeling of membership in and ownership of a patient community. Workout classes at Orangetheory Fitness publicly display identifiable consumer heart rate data, in real time, on the same monitor with others in their class. The normative influence of social comparison and the emotional relief of sharing experiences can be powerfully motivating, but may replace psychological ownership of personal data with membership in the groups with which it is shared.
Firms may increase collective psychological ownership for this data by soliciting consumer investment in its inputs; facilitating prosharing norms by asking consumers to share experiences, strategies, and ideas (e.g., medical symptoms and treatments; [111]); having consumers vote on goals for the community to pursue (e.g., how to reduce energy consumption); and helping consumers further the goals shared by the group (e.g., fundraising for members struggling to make their health care payments). Firms can present group-level data as a benchmark of progress toward collective goals, or to differentiate rival groups (e.g., competitions between neighborhoods in average household energy consumption). Platforms dependent on user-generated content may be particularly invested in such forms of community building, which are known to increase member contributions and usage ([111]).
Access-based models also afford potential opportunities to preserve psychological ownership. Consumers have more choice as they select and manage data inputs, outputs, and visualizations from medical tests and devices. These choices can be facilitated by increased data integration and personalization. Regulatory changes are also helpful in offering more choice in privacy options, such as via the "right to be forgotten." Customizable disclosure settings give consumers the ability to selectively remove their data from the collective space and increase their individual privacy ([34]). Fine-tuning desired disclosure levels across multiple platforms and audiences could increase perceived control of the data. To foster psychological ownership, developing and communicating policies that give the customer greater control and choice over which data is harvested or shared will be important, such as by providing consumers with an opt-out default as they trade access for personal data ([ 1]). Other means to preserve perceived control include enhancing consumer control over shared data with analysis tools for evaluating and displaying personal data shared with a firm.
A second way to preserve psychological ownership of personal data is through the considerable opportunities for self-expression and social group membership afforded by publishing personal data. While the majority of users do not post personal information on social media ([111]), many consumers do divulge a variety of personal data online, such as their location on Foursquare or Instagram, their employment on Twitter or LinkedIn, their family on Facebook, and their spending on Yelp, Amazon, or Mint. Firms can facilitate new channels for positive social signaling—such as ways to express desirable knowledge, experience, or status—to increase data disclosure and consumer ownership. This strategy may work best with digital natives, extraverts, and narcissists, who are particularly likely to disclose personal information on social media platforms ([111]).
The expansion of the collection and use of personal data in business is recategorizing data that was once associated with material or physical records as experiential. Data that was "static" in the past, such as a physical report of heart rate and blood pressure measured once during an annual physical, are often now continuously collected and displayed in real time on wearable devices or through application dashboards with animation, audio, and gamification ([66]; [74]; see Table 5). Another emerging and potentially sensitive source of experiential personal data comes from the Internet of Things, as many home appliances (refrigerators, washers) and systems (electrical, HVAC, water) are continuously monitored and their output harvested, capturing activity about consumers' daily lives ([124]).
Graph
Table 5. Evolution of Consumption and Psychological Ownership: Open Questions.
| Dimensions of Change | Research Questions |
|---|
| Legal Ownership to Legal Access |
| Consumer issues | When does access-based consumption increase and decrease demand for future private ownership of goods? How do risks of future loss (e.g., discontinued access) affect PO? Are antecedents and consequences of individual and collective PO different? Do larger consideration sets and more customization increase PO? Are access-based goods weaker influences on, and expressions of, self-identity? Does social signaling increase or crowd out PO? Is PO developed for aspirational goods and brands through access-based use? Does selling access to goods reduce PO for owners/prosumers? Do consumers feel reduced PO for goods chosen with recommendation systems? Are threats and opportunities to PO culturally specific (e.g., individualistic cultures)? Is consumer well-being improved, in the aggregate, with the substitution of access-based models for legally owned goods?
|
| Firm issues | What access-based models best preserve PO (e.g., rent-to-own, rent, streaming)? Can impermanence threats be mitigated in access-based models (e.g., guarantees)? Which marketing strategies help increase PO for brands? How should choice be balanced with choice overload (assortment sizes vs. recommendation system)? Are access-based goods downward stretches for luxury/status brands? When should PO be reduced for personal data versus adopting proprivacy positioning? What are the net effects of threats and opportunities on PO by technology/context?
|
| Material to Experiential |
| Consumer issues | What material goods cannot be fully replaced by experiential goods? What interface designs/application features preserve PO (e.g., haptic, rate control)? Do different sensory features instantiate PO for material and experiential goods? What determines PO of an experience (e.g., indexicality, goal achieved)? When are associated material and experiential goods PO complements or substitutes (e.g., movie and smartphone, song and band T-shirt, trip and souvenir)? What determines categorization level of PO (e.g., good, device, platform, brand)? Why is there greater self-identification for experiential goods than material goods? Are threats and opportunities to PO generationally specific (e.g., digital natives)?
|
| Firm issues | Is adoption of experiential goods impaired/facilitated by owning material substitutes? When should firms implement fully experiential vs. hybrid offerings (e.g., music, courses)? When will demand for material complements justify cross selling (e.g., books)? How should indexical connections and gamification for experiential goods be implemented? Does vertical integration of brands with platforms capture transfer of PO? When should experiential goods be marketed as services? How do experiential versus material purchases affect PO for brands and intermediaries? When are firm versus consumer values more important for identity marketing?
|
5 Notes: PO = psychological ownership.
These more experiential forms of data may threaten psychological ownership due to intangibility, more ambiguous evaluations of ownership, and the higher categorization level at which experiential data are construed. Consumers may feel less control over disclosure of intangible cloud-based continuous data than static physical records. Perceived control may be particularly impaired if firms remove actual user control by fixing the manner in which data is collected, accessed, and presented. A shift to experiential consumption of data, however, could increase psychological ownership of that data if firms give consumers more control of its disclosure, display, and delivery, facilitating identification with the data and its consumption (e.g., see their health data as an indicator of "me" rather than "it"; [40]; [125]). Internet-enabled devices and wearables could give consumers the ability to "mute" data reporting. Platforms can facilitate the accessibility of data when consumers desire it. At any time of day or night, a patient may receive test results and request referrals from her primary physician on MyChart or initiate a prescription refill via SMS or IVR communication with her pharmacy. Psychological ownership could also be enhanced through haptic (e.g., touchscreen) interfaces and dashboards that control privacy settings (e.g., [21]).
A second threat to psychological ownership that arises from the immateriality of data is reduced evaluability, meaning that it is difficult to determine who owns the data. A consumer might feel less psychological ownership for a dynamic heart rate report during a fitness class than for a printout reporting her static heart rate during a physical because ownership of the dynamic data is more ambiguous. It may belong to the consumer, the firm that manufactured the device on which it is recorded, the firm supporting the application on which it is displayed, or the firm running the cloud server where it is stored. In other cases, consumers may claim ownership for data that are not "theirs." When consumers use the internet to answer questions, for instance, they misattribute possession of that knowledge to themselves ([122]). Indexing or gamifying data to form a record of meaningful personal events (e.g., exercise classes, family birthdays, graduation), or making it a meaningful story in itself, such as achieving a health or wellness goal, may bolster consumer psychological ownership.
A shift from more material to experiential forms of personal data may prompt a transfer in psychological ownership between categorization levels, from the individual data (e.g., my cholesterol level) to the applications and intermediary devices and platforms that provide access to that data (e.g., iHealth, iPhone, or MyChart, respectively). Consumers may feel considerable ownership of their accounts and devices. They may also hold platforms and firms rather than themselves responsible for security. Beyond providing consumers with opportunities to personalize their accounts and intermediary devices, firms should prioritize customer satisfaction and position brands and platforms in ways that allow consumers to feel psychological ownership for them (e.g., highlight identity consistency, emphasize the unique history of the company or platform, encourage consumer self-investment).
A related opportunity to preserve psychological ownership for personal data as it shifts to more experiential forms is to capitalize on consumer identification with experiences. As data evolve from static documents to dynamic portraits of the self across time, data may provide a record of experiences that confirm important identities to consumers. A record of a run could be a social signal to potentially broadcast to others but could also reaffirm an important identity to a consumer (e.g., runner, athlete, fit). Identity marketing, whether integrated into data capture or display or positioning, could create feelings of ownership for these dynamic experiential records of consumers' lives.
We view psychological ownership as an asset that is typically valuable for consumers and firms to preserve ([41]; [86]), even in cases in which legal ownership is inconvenient or undesirable. Of course, there are caveats where consumers, firms, or both may benefit from its decline. We suggest four important cases for each.
Consumers may find psychological ownership to be undesirable ( 1) when it would amplify the pain of a sure loss, ( 2) when it would link them with identity-incongruent goods, ( 3) when it would increase the meaning of negative events or decrease the meaning of positive events, or ( 4) when a good will be shared. We discuss each of these points. First, when possession of goods is short term, consumers may wish to forgo psychological ownership to reduce the pain felt when returning goods, such as a rental car or dress, and thus avoid the strong feelings of loss felt when selling their car or donating their clothing ([116]). This avoidance is evident in the lack of psychological ownership felt by expert traders for goods they expect to sell ([72]) and by consumers of borrowed and rented goods ([ 4]; [ 5]).
Second, because psychological ownership changes how consumers perceive not only the good but also themselves ([125]), they may avoid psychological ownership for goods that are identity incongruent. A cinephile may prefer to digitally stream a film before committing to the self-signal that buying it entails, for example, and pornography consumers may prefer to not feel psychological ownership for their browsing and search history.
Third, consumers may eschew psychological ownership of goods that would increase the meaningfulness of negative events, such as a funeral or personal failure ([73]), and goods that would muddle other reminders of meaningful positive events (e.g., memorabilia from an unmemorable conference at a place where they vacationed with family; [127]).
Fourth, consumers may try to avoid high levels of psychological ownership for goods that will be shared with others. Feeling greater psychological ownership for personal data could change consumers' personal comfort equilibrium with trading their data for free access to platforms that will sell it (e.g., Facebook), and prompt them to discontinue use of those desirable and "free" goods and services. Reduced psychological ownership should help reduce jealousy or territoriality when sharing physical goods ([65]). Psychological ownership for a good, and a more general attachment to goods ([36]), should thus be key predictors of engaging in the supply side of the sharing economy. For example, firms may find that a prospective homeowner who has yet to develop psychological ownership for a home ([89]; [110]) should be more comfortable with renting her home to strangers. Having decided to rent it, she might even purposely furnish it in a style that is discordant with her personal taste to establish a boundary between the properties in which she lives and lets.
We identify four cases in which firms may benefit if consumers feel low levels of psychological ownership for goods, intermediaries, and brands: ( 1) when changes in access rights are likely, ( 2) when consumers are the product, ( 3) when it creates frictions in sharing markets, and ( 4) when service quality is inconsistent. First, like consumers, firms may prefer low levels of psychological ownership when access to goods is short-lived. When Microsoft ended sales of eBooks in April 2019, it deleted and refunded all books purchased through the platform. Consumers who felt stronger psychological ownership for the books in their digital library may have felt greater loss and anger when their access rights were revoked. More generally, for any digital goods or personal data, strong psychological ownership may breed resentment that access rights cannot be shared with or transferred to other consumers through sales, gifts, or inheritances.
Second, many firms earn considerable profit from "free" services by mining and selling consumer personal data. In such cases, it may benefit firms to enact policies, contracts, and contexts that minimize psychological ownership of personal data (e.g., [ 1]). Consumers with high psychological ownership for their data may demand a share of profits or divulge less personal information ([76]).
Third, if consumers feel high levels of psychological ownership for particular goods and brands, it may create frictions in matching consumer demand and supply, similar to market frictions in the endowment effect literature ([31]; [86]). A consumer with strong attachment to and psychological ownership for Mercedes cars, for instance, might be reluctant to book a car from a car-sharing platform if only Fords are available. Consumers who feel psychological ownership for a "third place"––a social space other than at home or work, such as a seat in a café, bar, or park––may be more likely to visit it but will linger in that space ([51]). Firms may wish to keep psychological ownership low for access-based and experiential goods so that consumers are more receptive to a variety of goods and brands, or turn over quickly.
Fourth, when dealing with consumers with high psychological ownership, firms will need to more carefully manage expectations and customer satisfaction ([117]). The value-enhancing effects of psychological ownership, if it has been transferred from the good to the brand, may heighten expectations and make firms more accountable for service failures in the eyes of consumers. If a ride-share car breaks down during a ride, for example, the consumer may hold the platform responsible rather than the driver or the automotive brand. Preserving psychological ownership may thus be a counterproductive exercise for platforms when service failures are likely.
Applying our psychological ownership framework and associated concepts to three macro trends in marketing identifies many opportunities for future research, some of which we previously outlined. Table 5 suggests additional opportunities for exploration. Psychological ownership is a central theme, but the list engages with a variety of major themes in marketing research. In consumer behavior, our framework informs research examining how technology is changing the self-concept, as well as critical relationships between consumers and technologies, goods, brands, and other consumers (e.g., [53]).
Researchers focused on firm strategy and technological innovation will find that our framework delineates important considerations, boundaries, and opportunities for the acceptance and adoption of new consumption models and technologies. Many traditional brands have stumbled when entering access-based markets (e.g., car-sharing services such as BMW's ReachNow and GM's Maven) or when launching digital products (e.g., Barnes & Noble's Nook e-reader). Marketing strategists navigating the transformation from private material goods to access-based experiential goods cannot solely focus on and tout benefits of relinquishing legal ownership. Marketers should consider trade-offs between legal and psychological ownership as well as how to maintain the attachments, value, and loyalty to goods and brands that consumers derive from psychological ownership. Behavioral researchers need to identify the brands and sectors for which those attachments, value enhancements, and loyalties are most contingent on the preservation of psychological ownership (e.g., luxury goods). Firms and strategy researchers should test when product development, branding, and repositioning strategies preserve psychological ownership (e.g., servitization, vertical integration, brand alliances), which could be a lifeline for struggling industries and firms (e.g., retail, telecommunications, financial services). We have made many such suggestions throughout this article.
The threats and opportunities to preserve psychological ownership identified by our framework generalize beyond the three macro trends in marketing we explore here to many technology-driven trends reshaping modern economies and life. Psychological ownership may affect consumer motivations for sustainable consumer behavior. It could increase preservation of shared resources, as it does for private goods. It could also be counterproductive and increase the consumption of those resources, if consumers anticipate others using them. Remote work and the move from live personal interactions toward virtual interactions is an area experiencing growth, accelerated by the COVID-19 pandemic. If remote work is the future of employment, how will virtual interactions affect psychological ownership among the parties involved? Will employees who work from home feel more or less psychological ownership for their ideas, projects, and firms, as compared to a live office environment? Will students feel less psychological ownership for online courses and degrees received for remote learning? Automation and artificial intelligence in both firm and residential applications is another such trend. Psychological ownership has numerous direct applications to its intersections with retailing and labor. Consumers may feel less psychological ownership and attachment to items chosen or purchased by or with the help of a recommendation system if using recommendation systems feels like relinquishing choice to another agent. The desirability of psychological ownership may then be an important factor in determining for which product categories recommendation systems, touchscreens, and voice interfaces should be integrated as decision aids or replace live salespeople. More generally, whether consumers feel psychological ownership for intelligent devices may depend critically on their positioning (e.g., tool vs. intelligent agent).
Although we have suggested that transfer can occur, an important question remains regarding what happens to the aggregate level of psychological ownership felt by a consumer in response to these changes. When a consumer relinquishes a traditional good, does the aggregate level of psychological ownership she experiences also decline? Psychological ownership once felt for her amassed library of books, movies, and photographs, for instance, could decrease as it is digitized or transferred to devices and streaming platforms. Indeed, if psychological ownership is bundled into devices or platforms, diminishing marginal utility suggests that it will decline in the aggregate ([114]). However, psychological ownership satisfies core motivational drivers, so consumers may instead strive to maintain a set level of aggregate psychological ownership for their various attachments. They may then transfer the psychological ownership lost for one good to other targets (e.g., goods, devices, platforms). Our article focuses on changes to psychological ownership felt for individual goods, but how technology-driven consumption changes affect the aggregate level of psychological ownership consumers experience is a question critical for understanding the ebbs and flows of psychological ownership.
Finally, we do not address heterogeneity in the experience of psychological ownership, but it is likely that features of psychological ownership are not universal or static. They are manifested differently across cultures as well as within cultures with different forms of economic transaction. Psychological ownership does not appear to generate the same degree of value enhancement for East Asians or descendants of East Asian cultures, for instance, as it does for White Americans or people descended from European cultures ([75]). Generational differences may affect how psychological ownership is affected by the macro trends we have identified. Digital natives who have grown up with music streaming and targeted mobile advertising may be less threatened. Firms need guidance to develop and deploy effective targeting and positioning strategies across cultures, generations, and other groups.
Technological innovations are changing consumption models from permanent legal ownership of private physical goods to access-based use of temporary, experiential, and collective goods. Consumers benefit from forgoing legal ownership of goods in these fractional ownership models (e.g., money, time, effort; [ 8]; [69]). However, giving up legal ownership does not imply that psychological ownership, a generally desirable source of value for both firms and consumers, must or should also be relinquished.
We illustrate the worth of a psychological ownership framework for anticipating and understanding consumer responses to this technology-driven evolution in consumption. Our framework predicts when technological innovations will threaten, transfer, and create opportunities to preserve this valuable asset, and it identifies accompanying research opportunities for marketing scholars. We have mapped our framework to three key macro trends: ( 1) growth in the sharing economy, ( 2) digitization of goods and services, and ( 3) the expansion of personal data. For each trend, we offer recommendations for how managers can counter threats to psychological ownership and leverage opportunities to preserve or enhance it through a variety of strategies. We also note cases in which consumers and firms benefit from letting psychological ownership decline. More broadly, our framework applies to many sectors where technology is changing consumption, and it is informative for managers vying to attract and retain customers within these new environments. It outlines many ways in which psychological ownership will continue to be a valuable lens through which to view, understand, forecast, and manage the consumer experience.
Supplemental Material, JM.18.0480.R5_Web_Appendix_PDF - Evolution of Consumption: A Psychological Ownership Framework
Supplemental Material, JM.18.0480.R5_Web_Appendix_PDF for Evolution of Consumption: A Psychological Ownership Framework by Carey K. Morewedge, Ashwani Monga, Robert W. Palmatier, Suzanne B. Shu and Deborah A. Small in Journal of Marketing
Footnotes 1 Online supplement: https://doi.org/10.1177/0022242920957007
2 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 ORCID iD Carey K. Morewedge https://orcid.org/0000-0001-7502-9279
References Acquisti Alessandro, John Leslie K., Loewenstein George. (2013), "What Is Privacy Worth?" Journal of Legal Studies, 42 (2), 249–74.
Argo Jennifer J., Dahl Darren W., Morales Andrea C. (2008), "Positive Consumer Contagion: Responding to Attractive Others in a Retail Context," Journal of Marketing Research, 45 (6), 690–701.
Asatryan Vahagn S., Oh Haemoon. (2008), "Psychological Ownership Theory: An Exploratory Application in the Restaurant Industry," Journal of Hospitality & Tourism Research, 32 (3), 363–86.
Atasoy Ozgun, Morewedge Carey K. (2018), "Digital Goods Are Valued Less Than Physical Goods," Journal of Consumer Research, 44 (6), 1343–57.
5 Bagga Charan K., Bendle Neil, Cotte June. (2019), "Object Valuation and Non-Ownership Possession: How Renting and Borrowing Impact Willingness-to-Pay," Journal of the Academy of Marketing Science, 47 (1), 97–117.
6 Barasch Alixandra, Berger Jonah. (2014), "Broadcasting and Narrowcasting: How Audience Size Affects What People Share," Journal of Marketing Research, 51 (3), 286–99.
7 Bardhi Fleura, Eckhardt Giana M. (2012), "Access-Based Consumption: The Case of Car Charing," Journal of Consumer Research, 39 (4), 881–98.
8 Bardhi Fleura, Eckhardt Giana M. (2017), "Liquid Consumption," Journal of Consumer Research, 44 (3), 582–97.
9 Barton Christine, Koslow Lara, Beauchamp Christine. (2014), "How Millenials Are Changing the Face of Marketing Forever," BCG (January 15), https://www.bcg.com/publications/2014/marketing-center-consumer-customer-insight-how-millennials-changing-marketing-forever.
Bastos Wilson, Brucks Merrie. (2017), "How and Why Conversational Value Leads to Happiness for Experiential and Material Purchases," Journal of Consumer Research, 44 (3), 598–612.
Bauman Zygmunt. (2000), Liquid Modernity. Cambridge, UK : Polity Press.
Baxter Weston L., Aurisicchio Marco. (2018), "Ownership by Design," in Psychological Ownership and Consumer Behavior, Peck Joann, Shu Suzanne B., eds. Cham, Switzerland : Springer, 119–35.
Baxter Weston L., Aurisicchio Marco, Childs Peter R.N. (2015), "A Psychological Ownership Approach to Designing Object Attachment," Journal of Engineering Design, 26 (4–6), 140–56.
Beggan James K. (1992), "On the Social Nature of Nonsocial Perception: The Mere Ownership Effect," Journal of Personality and Social Psychology, 62 (2), 229–37.
Belk Russell W. (1988), "Possessions and the Extended Self," Journal of Consumer Research, 15 (2), 139–68.
Belk Russell W. (2010), "Sharing," Journal of Consumer Research, 36 (5), 715–34.
Belk Russell W. (2013), "Extended Self in a Digital World," Journal of Consumer Research, 40 (3), 477–500.
Belk Russell W. (2014), "You Are What You Can Access: Sharing and Collaborative Consumption Online," Journal of Business Research, 67 (8), 1595–1600.
Bhattacharjee Amit, Berger Jonah, Menon Geeta. (2014), "When Identity Marketing Backfires: Consumer Agency in Identity Expression," Journal of Consumer Research, 41 (2), 294–309.
Botsman Rachel, Rogers Roo. (2010), "Beyond Zipcar: Collaborative Consumption," Harvard Business Review, 88 (10), 30.
Brasel S. Adam, Gips James. (2014), "Tablets, Touchscreens, and Touchpads: How Varying Touch Interfaces Trigger Psychological Ownership and Endowment," Journal of Consumer Psychology, 24 (2), 226–33.
Brynjolfsson Erik, Collis Avinash, Eggers Felix. (2019), "Using Massive Online Choice Experiments to Measure Changes in Well-Being," Proceedings of the National Academy of Sciences, 116 (15), 7250–55.
Cakebread Caroline. (2017), "People Will Take 1.2 Trillion Digital Photos this Year—Thanks to Smartphones," Business Insider (August 31), https://www.businessinsider.com/12-trillion-photos-to-be-taken-in-2017-thanks-to-smartphones-chart-2017-8.
Campbell David F.J, Hanschitz Georg. (2018), " Digitalization of Tax: Epistemic Tax Policy," in Handbook of Cyber-Development, Carayannis Elias G., Campbell David F.J., Efthymiopoulos Marios Panagiotis, eds. Cham, Switzerland : Springer International.
Carter Travis J., Gilovich Thomas. (2010), "The Relative Relativity of Material and Experiential Purchases," Journal of Personality and Social Psychology, 98 (1), 146.
Chaudhuri M., Voorhees Clay M., Beck Joshua M. (2019), "The Effects of Loyalty Program Introduction and Design on Short- and Long-Term Sales and Gross Profits," Journal of the Academy of Marketing Science, 47, 640–58.
Coyle Karen. (2006), "Mass Digitization of Books," Journal of Academic Librarianship, 32 (6), 641–45.
Downes Larry. (2018), "GDPR and the End of the Internet's Grand Bargain," Harvard Business Review (April 9), https://hbr.org/2018/04/gdpr-and-the-end-of-the-internets-grand-bargain.
Eckhardt Giana M., Bardhi Fleura. (2015), "The Sharing Economy Isn't About Sharing at All," Harvard Business Review (January 28), https://hbr.org/2015/01/the-sharing-economy-isnt-about-sharing-at-all.
Eckhardt Giana M., Houston Mark B., Jiang Baojun, Lamberton Catherine, Rindfleisch Aric, Zervas Giorgos. (2019), "Marketing in the Sharing Economy," Journal of Marketing, 83 (5), 5–27.
Ericson Keith M., Füster Andreas. (2011), "Expectations as Endowments: Evidence on Reference-Dependent Preferences from Exchange and Valuation Experiments," Quarterly Journal of Economics, 126 (4), 1879–1907.
Escalas Jennifer Edson, Bettman James R. (2005), "Self-Construal, Reference Groups, and Brand Meaning," Journal of Consumer Research, 32 (3), 378–89.
European Commission (2020), "What Is Personal Data?" (accessed September 24), https://ec.europa.eu/info/law/law-topic/data-protection/reform/what-personal-data_en.
Faitelson Yaki. (2019), "Why U.S. GDPR-Style Privacy Laws Are Good for Business," Forbes (December 19), https://www.forbes.com/sites/forbestechcouncil/2019/12/19/why-u-s-gdpr-style-privacy-laws-are-good-for-business/#2b6f9f3d8756.
Farronato Chiara, Fradkin Andrey. (2018), "The Welfare Effects of Peer Entry in the Accommodation Market: The Case of Airbnb," Working Paper No. 24361, National Bureau of Economic Research.
Ferraro Rosellina, Escalas Jennifer Edson, Bettman James R. (2011), "Our Possessions, Our Selves: Domains of Self-Worth and the Possession–Self Link," Journal of Consumer Psychology, 21 (2), 169–77.
Figueiredo Bernardo, Scaraboto Daiane. (2016), "The Systemic Creation of Value Through Circulation in Collaborative Consumer Networks," Journal of Consumer Research, 43 (4), 509–33.
Findlay Isobel M. (2018), "Precursors to the Sharing Economy: Cooperatives," in The Rise of the Sharing Economy: Exploring the Challenges and Opportunities of Collaborative Consumption, Albinsson Pia A., Yasanthi Perera B., eds. Santa Barbara, CA : Praeger, 9–28.
Fournier Susan. (1998), "Consumers and Their Brands: Developing Relationship Theory in Consumer Research," Journal of Consumer Research, 24 (4), 343–73.
Franke Nikolaus, Schreier Martin, Kaiser Ulrike. (2010), "The 'I Designed It Myself' Effect in Mass Customization," Management Science, 56 (1), 125–40.
Fritze Martin P., Marchand André, Eisingerich Andreas B., Benkenstein Martin. (2020), "Access-Based Services as Substitutes for Material Possessions: The Role of Psychological Ownership," Journal of Service Research, 23 (3), 368–85.
Fuchs Christoph, Prandelli Emanuela, Schreier Martin. (2010), "The Psychological Effects of Empowerment Strategies on Consumers' Product Demand," Journal of Marketing, 74 (1), 65–79.
Furby Lita. (1991), "Understanding the Psychology of Possession and Ownership: A Personal Memoir and an Appraisal of Our Progress," Journal of Social Behavior and Personality, 6 (6), 457–63.
Ganesan Shankar, George Morris, Jap Sandy, Palmatier Robert W., Weitz Barton. (2009), "Supply Chain Management and Retailer Performance: Emerging Trends, Issues, and Implications for Research and Practice," Journal of Retailing, 85 (1), 84–94.
Gawronski Bertram, Bodenhausen Galen V., Becker Andrew P. (2007), "I Like It, Because I Like Myself: Associative Self-Anchoring and Post-Decisional Change of Implicit Evaluations," Journal of Experimental Social Psychology, 43 (2), 221–32.
Gilovich Thomas, Kumar Amit. (2015), "We'll Always Have Paris: The Hedonic Payoff from Experiential and Material Investments," Advances in Experimental Social Psychology, 51, 147–87.
Gilovich Thomas, Kumar Amit, Jampol Lily. (2015), "A Wonderful Life: Experiential Consumption and the Pursuit of Happiness," Journal of Consumer Psychology, 25 (1), 152–65.
Goldfarb Avi, Greenstein Shane M., Tucker Catherine E. (2015), Economic Analysis of the Digital Economy. Chicago : University of Chicago Press.
Grayson Kent, Martinec Radan. (2004), "Consumer Perceptions of Iconicity and Indexicality and Their Influence on Assessments of Authentic Market Offerings," Journal of Consumer Research, 31 (2), 296–312.
Grayson Kent, Shulman David. (2000), "Indexicality and the Verification Function of Irreplaceable Possessions: A Semiotic Analysis," Journal of Consumer Research, 27 (1), 17–30.
Griffiths Merlyn A., Gilly Mary C. (2012), "Dibs! Customer Territorial Behaviors," Journal of Service Research, 15 (2), 131–49.
Haase Michaela, Kleinaltenkamp Michael. (2011), "Property Rights Design and Market Process: Implications for Market Theory, Marketing Theory, and SD Logic," Journal of Macromarketing, 31 (2), 148–59.
Hamilton Ryan, Ferraro Rosellina, Haws Kelly L., Mukhopadhyay Anirban. (2021), "Traveling with Companions: The Social Customer Journey," Journal of Marketing, 85 (1), 68–92.
Hardin Garrett. (1968), "The Tragedy of the Commons," Science, 162 (3859), 1243–48.
He Daniel, Melumad Shiri, Pham Michel Tuan. (2018), "The Pleasure of Assessing and Expressing Our Likes and Dislikes," Journal of Consumer Research, 46 (3), 545–63.
Helm Sabrina V., Ligon Victoria, Stovall Tony, Riper Silvia Van. (2018), "Consumer Interpretations of Digital Ownership in the Book Market," Electronic Markets, 28 (2), 177–89.
Hodder Ian. (2012), Entangled: An Archaeology of the Relationships Between Humans and Things. Chichester, UK : John Wiley & Sons.
Honoré Anthony M. (1961), "Ownership," in Oxford Essays in Jurisprudence. Oxford : Oxford University Press, 107–147.
Huang Yunhui, Wang Lei, Shi Junqi. (2009), "When Do Objects Become More Attractive? The Individual and Interactive Effects of Choice and Ownership on Object Evaluation," Personality and Social Psychology Bulletin, 35 (6), 713–22.
Inbar Yoel, Pizarro David A., Knobe Joshua, Bloom Paul. (2009), "Disgust Sensitivity Predicts Intuitive Disapproval of Gays," Emotion, 9 (3), 435–39.
Karahanna Elena, Xu Sean Xin, Zhang Nan. (2015), "Psychological Ownership Motivation and Use of Social Media," Journal of Marketing Theory and Practice, 23 (2), 185–207.
Karakayali Nedim, Kostem Burc, Galip Idil. (2018), "Recommendation Systems as Technologies of the Self: Algorithmic Control and the Formation of Music Taste," Theory, Culture & Society, 35 (2), 3–24.
Keinan Anat, Kivetz Ran. (2010), "Productivity Orientation and the Consumption of Collectable Experiences," Journal of Consumer Research, 37 (6), 935–50.
Kim Jungkeun. (2017), "The Ownership Distance Effect: The Impact of Traces Left by Previous Owners on the Evaluation of Used Goods," Marketing Letters, 28 (4), 591–605.
Kirk Colleen P., Peck Joann, Swain Scott D. (2018), "Property Lines in the Mind: Consumers' Psychological Ownership and Their Territorial Responses," Journal of Consumer Research, 45 (1), 148–68.
Koivisto Jonna, Hamari Juho. (2019), "The Rise of Motivational Information Systems: A Review of Gamification Research," International Journal of Information Management, 45, 191–210.
Kuehn Kathleen M. (2016), "Branding the Self on Yelp: Consumer Reviewing as Image Entrepreneurship," Social Media + Society, 2 (4), 1–9.
Lamberton Cait. (2016), "Collaborative Consumption: A Goal-Based Framework," Current Opinion in Psychology, 10, 55–9.
Lamberton Cait P., Rose Randall L. (2012), "When Is Ours Better Than Mine? A Framework for Understanding and Altering Participation in Commercial Sharing Systems," Journal of Marketing, 76 (4), 109–25.
Lerner Jennifer S., Small Deborah A., Loewenstein George. (2004), "Heart Strings and Purse Strings: Carryover Effects of Emotions on Economic Decisions," Psychological Science, 15 (5), 337–41.
Li Charis X., Lutz Richard J. (2019), " Object History Value in the Sharing Economy," in Handbook of the Sharing Economy, Belk Russell W., Eckhardt Giana M., Bardhi Fleura, eds. Cheltenham, UK : Edward Elgar Publishing.
List John A. (2003), "Does Market Experience Eliminate Market Anomalies?" Quarterly Journal of Economics, 118 (1), 41–71.
Loewenstein George, Issacharoff Samuel. (1994), "Source Dependence in the Valuation of Objects," Journal of Behavioral Decision Making, 7 (3), 157–68.
Lurie Nicholas H., Mason Charlotte H. (2007), "Visual Representation: Implications for Decision Making," Journal of Marketing, 71 (1), 160–77.
Maddux William W., Yang Haiyang, Falk Carl, Adam Hajo, Adair Wendi, Endo Yumi, et al. (2010), "For Whom Is Parting with Possessions More Painful? Cultural Differences in the Endowment Effect," Psychological Science, 21 (12), 1910–17.
Marthews Alex, Tucker Catherine E. (2017), " Government Surveillance and Internet Search Behavior," working paper, doi.org/10.2139/ssrn.2412564.
Martin Kelly D., Borah Abhishek, Palmatier Robert W. (2017), "Data Privacy: Effects on Customer and Firm Performance," Journal of Marketing, 81 (1), 36–58.
Mason Elizabeth C., Richardson Rick. (2012), "Treating Disgust in Anxiety Disorders," Clinical Psychology: Science and Practice, 19 (2), 180–94.
Matzler Kurt, Veider Viktoria, Kathan Wolfgang. (2015), "Adapting to the Sharing Economy," MIT Sloan Management Review, 56 (2), 71–77.
McEwan Stephanie, Pesowski Madison L., Friedman Ori. (2016), "Identical but Not Interchangeable: Preschoolers View Owned Objects as Non-Fungible," Cognition, 146, 16–21.
Melumad Shiri, Pham Michel Tuan. (2020), "The Smartphone as a Pacifying Technology," Journal of Consumer Research, 47 (2), 237–55.
Mi Zhifu, Coffman D'Maris. (2019), "The Sharing Economy Promotes Sustainable Societies," Nature Communications, 10 (1), 1–3.
Mishra Gouri S., Clewlow Regina R., Mokhtarian Patricia L., Widaman Keith F. (2015), "The Effect of Carsharing on Vehicle Holdings and Travel Behavior: A Propensity Score and Causal Mediation Analysis of the San Francisco Bay Area," Research in Transportation Economics, 52, 46–55.
Molesworth Mike, Watkins Rebecca, Denegri-Knott Janice. (2016), "Possession Work on Hosted Digital Consumption Objects as Consumer Ensnarement." Journal of the Association for Consumer Research, 1 (2), 246–61.
Morewedge Carey K. (2020), " Psychological Ownership: Implicit and Explicit," working paper.
Morewedge Carey K., Giblin Colleen E. (2015), "Explanations of the Endowment Effect: An Integrative Review," Trends in Cognitive Sciences, 19 (6), 339–48.
Morewedge Carey K., Gray Kurt, Wegner Daniel M. (2010), "Perish the Forethought: Premeditation Engenders Misperceptions of Personal Control," in Self-Control in Brain, Mind, and Society, Hassin Ran R., Ochsner Kevin N., Trope Yaacov, eds. Oxford, UK : Oxford University Press, 260–78.
Morewedge Carey K., Shu Lisa L., Gilbert Daniel T., Wilson Timothy D. (2009), "Bad Riddance or Good Rubbish? Ownership and Not Loss Aversion Causes the Endowment Effect," Journal of Experimental Social Psychology, 45 (4), 947–51.
Nash Jane Gradwohl, Rosenthal Robert A. (2014), "An Investigation of the Endowment Effect in the Context of a College Housing Lottery," Journal of Economic Psychology, 42, 74–82.
Nielsen (2019), "Nielsen Music Mid-Year Report," (accessed September 24), https://www.nielsen.com/wp-content/uploads/sites/3/2019/06/nielsen-us-music-mid-year-report-2019.pdf.
Norton Michael I., Mochon Daniel, Ariely Dan. (2012), "The IKEA Effect: When Labor Leads to Love," Journal of Consumer Psychology, 22 (3), 453–60.
Oram Jon H. (1997), "The Costs of Confusion in Cyberspace," Yale Law Journal, 107 (3), 869–74.
Palmatier Robert. W., Martin Kelly D. (2018), An Intelligent Marketer's Guide to Data Privacy, working draft of forthcoming book, in press, Palgrave Macmillan.
Park C. Whan, MacInnis Deborah J., Priester Joseph. (2008), Brand Attachment: Construct, Consequences and Causes. Boston : Now Publishers.
Peck Joann, Shu Suzanne B. (2009), "The Effect of Mere Touch on Perceived Ownership," Journal of Consumer Research, 36 (3), 434–47.
Peck Joann, Shu Suzanne B. (2018), Psychological Ownership and Consumer Behavior. New York : Springer Publishing.
Pierce Jon L., Jussila Iiro. (2010), "Collective Psychological Ownership Within the Work and Organizational Context: Construct Introduction and Elaboration," Journal of Organizational Behavior, 31 (6), 810–34.
Pierce Jon L., Rubenfeld Stephen A., Morgan Susan. (1991), "Employee Ownership: A Conceptual Model of Process and Effects," Academy of Management Review, 16 (1), 121–44.
Rainie Lee, Anderson Janna. (2014), "The Future of Privacy," Pew Research Center (December 18), https://www.pewresearch.org/internet/2014/12/18/future-of-privacy/.
Reb Jochen, Connolly Terry. (2007), "Possession, Feelings of Ownership, and the Endowment Effect," Judgment and Decision Making, 2 (2), 107–14.
Rifkin Jeremy. (2001), The Age of Access: How the Shift from Ownership to Access is Transforming Modern Life. New York : Penguin Business.
Ritzer George, Jurgenson Nathan. (2010), "Production, Consumption, Prosumption: The Nature of Capitalism in the Age of the Digital 'Prosumer'," Journal of Consumer Culture, 10 (1), 13–36.
Schmitt Bernd H. (2010), Customer Experience Management: A Revolutionary Approach to Connecting with your Customers. Hoboken, NJ : John Wiley & Sons.
Schryver Kyle. (2019), "The Future of Data Privacy in the United States," CPO Magazine (August 1), https://www.cpomagazine.com/data-protection/the-future-of-data-privacy-in-the-united-states/.
Schultz P. Wesley, Nolan Jessica M., Cialdini Robert B., Goldstein Noah J., Griskevicius Vladas. (2007), "The Constructive, Destructive, and Reconstructive Power of Social Norms," Psychological Science, 18 (5), 429–34.
Shaw Alex, Li Vivian, Olson Kristina R. (2012), "Children Apply Principles of Physical Ownership to Ideas," Cognitive Science, 36 (8), 1383–1403.
Shu Suzanne B., Peck Joann. (2011), "Psychological Ownership and Affective Reaction: Emotional Attachment Process Variables and the Endowment Effect," Journal of Consumer Psychology, 21 (4), 439–52.
Siddiqui Shakeel, Turley Darach. (2006), "Extending the Self in a Virtual World," in Advances in Consumer Research, Vol. 33, Pechmann Connie, Price Linda, eds. Duluth, MN : Association for Consumer Research, 647–48.
Sinclair Gary, Tinson Julie. (2017), "Psychological Ownership and Music Streaming Consumption," Journal of Business Research, 71, 1–9.
Strahilevitz Michal A., Loewenstein George. (1998), "The Effect of Ownership History on the Valuation of Objects," Journal of Consumer Research, 25 (3), 276–89.
Sun Na, Rau Patrick Pei-Luen, Ma Liang. (2014), "Understanding Lurkers in Online Communities: A Literature Review," Computers in Human Behavior, 38, 110–17.
Tadelis Steven. (2016), "Reputation and Feedback Systems in Online Platform Markets," Annual Review of Economics, 8, 321–40.
Tanis Martin. (2008), "Health-Related On-Line Forums: What's the Big Attraction?" Journal of Health Communication, 13 (7), 698–714.
Thaler Richard. (1985), "Mental Accounting and Consumer Choice," Marketing Science, 4 (3), 199–214.
Thomson Matthew, MacInnis Deborah J., Park Whan C. (2015), "The Ties That Bind: Measuring the Strength of Consumers' Emotional Attachments to Brands," Journal of Consumer Psychology, 15 (1), 77–91.
Trudel Remi, Argo Jennifer J., Meng Matthew D. (2016), "The Recycled Self: Consumers' Disposal Decisions of Identity-Linked Products," Journal of Consumer Research, 43 (2), 246–64.
Tsiros Michael, Mittal Vikas, Ross William T. Jr. (2004), "The Role of Attributions in Customer Satisfaction: A Reexamination," Journal of Consumer Research, 31 (2), 476–83.
Van Dijck José. (2008), "Digital Photography: Communication, Identity, Memory," Visual Communication, 7 (1), 57–76.
Vandewalle Don, Dyne Linn Van, Kostova Tatiana. (1995), "Psychological Ownership: An Empirical Examination of Its Consequences," Group & Organization Management, 20 (2), 210–26.
Verkuyten Maykel, Martinovic Borja. (2017), "Collective Psychological Ownership and Intergroup Relations," Perspectives on Psychological Science, 12 (6), 1021–39.
Wallendorf Melanie, Arnould Eric J. (1988), "'My Favorite Things': A Cross-Cultural Inquiry into Object Attachment, Possessiveness, and Social Linkage," Journal of Consumer Research, 14 (4), 531–47.
Ward Adrian F. (2013), "Supernormal: How the Internet Is Changing Our Memories and Our Minds," Psychological Inquiry, 24 (4), 341–48.
Watkins Rebecca D., Denegri-Knott Janice, Molesworth Mike. (2016), "The Relationship Between Ownership and Possession: Observations from the Context of Digital Virtual Goods," Journal of Marketing Management, 32 (1/2), 44–70.
Wedel Michel, Kannan P.K. (2016), "Marketing Analytics for Data-Rich Environments," Journal of Marketing, 80 (6), 97–121.
Weiss Liad, Johar Gita Venkataramani. (2016), "Products as Self-Evaluation Standards: When Owned and Unowned Products Have Opposite Effects on Self-Judgment," Journal of Consumer Research, 42 (6), 915–30.
Ye Yang, Gawronski Bertram. (2016), "When Possessions become Part of the Self: Ownership and Implicit Self-Object Linking," Journal of Experimental Social Psychology, 64, 72–87.
Zauberman Gal, Ratner Rebecca K., Kim Kyu B. (2008), "Memories as Assets: Strategic Memory Protection in Choice over Time," Journal of Consumer Research, 35 (5), 715–28.
Zyskind Guy, Nathan Oz, Pentland Alex. (2015), " Decentralizing Privacy: Using Blockchain to Protect Personal Data," 2015 IEEE Security and Privacy Workshops, 180–84.
~~~~~~~~
By Carey K. Morewedge; Ashwani Monga; Robert W. Palmatier; Suzanne B. Shu and Deborah A. Small
Reported by Author; Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 51- Examining Why and When Market Share Drives Firm Profit. By: Bhattacharya, Abhi; Morgan, Neil A.; Rego, Lopo L. Journal of Marketing. Jul2022, Vol. 86 Issue 4, p73-94. 22p. 1 Diagram, 10 Charts. DOI: 10.1177/00222429211031922.
- Database:
- Business Source Complete
Examining Why and When Market Share Drives Firm Profit
Many firms use market share to set marketing goals and monitor performance. Recent meta-analytic research reveals the average economic impact of market share performance and identifies some factors affecting its value. However, empirical understanding of why any market share–profit relationship exists and varies is limited. The authors simultaneously examine the three primary theoretical mechanisms linking firm market share with profit. On average, they find that most of the variance in market share's positive effect on firm profit is explained by market power and quality signaling, with little support for operating efficiency as a mechanism. They find a similar explanatory role of the three mechanisms in conditions where market share negatively predicts profit (for niche firms and those "buying" market share). Using these mechanism insights, the authors show that the value of market share differs in predictable ways between firms and across industries, providing new understanding of when managers may usefully set market share goals. The authors also provide new insights into how market share should be measured for goal setting and performance monitoring. They show that revenue market share is a predictor of firm profit while unit market share is not, and that relative measures of revenue market share can provide greater predictive power.
Keywords: market share; quality; efficiency; market power; niche; firm profit; revenue share; unit share
Many firms use market share to set goals and monitor marketing performance, and market share is also widely used in research examining marketing's performance impact ([24]; [40]). [20] recent meta-analytic study (hereinafter, E-H 2018) reports a significant positive relationship between a firm's market share and its economic performance and identifies contingencies affecting this relationship. However, while the literature suggests several reasons market share may drive firm performance, few empirical studies have directly examined any (and none more than one) of these mechanisms. Thus, little is known about the underlying "why" of mechanism(s) linking firms' market share and economic performance and how they may both explain previously identified moderators and facilitate identification of additional moderators of this important relationship. In addition, when understanding of the mechanisms linking market share with firm performance suggests that it is economically valuable to measure market share for goal setting and performance monitoring purposes, managers currently have no empirical insights into how to do so.
These knowledge gaps are important because understanding why market share is linked to firms' future profit can provide new insights into when and where market share is most likely to be valuable. While many firms use market share as a marketing performance metric, our research identifies new ways for managers to assess when this is most appropriate—and when it may not be. Because market share is such a common marketing goal, this is also important in delineating the role that marketing plays in determining firm performance and in understanding contingencies that may affect this role. Exploring the predictive value of alternative measures of market share, we also provide important new insights into how market share goals should be set and performance assessed via different market share measurement options in terms of unit versus revenue market share and absolute versus relative market share.
In addressing these key questions, this study offers several contributions. First, we provide the first direct empirical assessment of the three primary causal mechanisms that have been theorized to link market share with firm profit: market power, operating efficiency, and quality signaling. Using direct measures, we examine each of these three mechanisms simultaneously and show that both market power and quality signaling are key mechanisms linking market share with firm profit. On average, we find little evidence of theorized economies of scale and learning benefits of market share, but we identify conditions under which such efficiency benefits do exist. We find no support for a fourth theorized mechanism linking market share negatively with profit as a result of a strong competitor orientation. However, we do find support for the same three mechanisms in conditions under which the market share–firm profit relationship is negative—for niche firms and when a firm "buys" market share. Overall, these findings provide important new empirical insights into market share's value-creating role.
Second, using these new causal mechanism insights, we explore the consistency of the market share–profit relationship across different types of marketplaces and firms where the relative value of market share via the three mechanisms may be expected to vary. We show that the market share–profit relationship varies across industries and firms, and that the different causal mechanisms identified provide high explanatory power for such variations; thus, all three theories from which the hypothesized mechanisms arise can be "correct." In addition, this insight provides an empirically supported way for managers to identify when setting market share goals and monitoring market share performance may be more or less valuable. In contrast, we find that using indirect contingencies to try to infer the mechanisms linking market share with performance relationship often does not align with the directly observed mechanism effects, further indicating the value of direct measures in understanding the "why" mechanisms involved.
Third, we extend recent meta-analytic insights regarding the nature of the relationship between market share and firms' economic performance by using direct measures of the three most widely cited mechanisms: measures of both revenue and unit market share and different market share benchmarks, firm size controls to isolate the benefits of market share versus firm scale, and different econometric approaches to address panel data and endogeneity estimation concerns. These aspects of our study enable us to provide several new insights. For example, we show that for most firms, economies of scale arise from firm size and not firm market share. They also allow us to identify which market share metrics are most predictive of profit for different types of firms and the economic value of increasing market share on these metrics. This is useful new knowledge for managers because it provides new insights into how market share should be measured in goal setting and performance monitoring as well as the scale of profit benefits that may be expected from any gain in a firm's market share.
The article is organized as follows. First, we develop a conceptual framework and hypothesize relationships involving the three key mechanisms by which market share may be linked with firms' future profit. Next, we use the three mechanisms to identify three conditions under which the market share–profit relationship may be expected to be stronger versus weaker. We then describe the data set assembled and analysis approaches used to test the hypotheses and discuss the results. Having shown that the three mechanisms collectively mediate the market share–profit relationship, we then assess whether this remains true even under conditions when the market share–profit relationship is negative. Next, having shown that managers can use knowledge of the three mechanisms to identify when market share is likely to be economically valuable for their firm, we assess how managers may best measure market share. Finally, we assess the implications of our study for theory and practice and identify new questions for future research suggested by our findings.
Much of the theorizing regarding market share and firm performance in economics and management concerns related but distinct phenomena such as firm size and market concentration. We focus only on relationships that directly pertain to firm market share and the mechanisms underlying its economic value. As a result, we center our market share conceptualization on revenue market share—units sold × realized price (i.e., sales revenue) divided by total market sales revenue. In doing so, we conceptualize and measure the "market" as comprising firms selling similar product/service offerings. However, we also examine unit market share—units sold divided by total market unit sales—as well as several different operationalizations of revenue market share in robustness checks and post hoc analyses.
The marketing literature generally views market share as an indicator of the success of a firm's efforts to compete in a product-marketplace (e.g., [13]; [63]). From this perspective, market share is an outcome of a firm's marketing efforts including its advertising and promotion, product/service offering quality and price, channel and customer relationships, and selling activities ([24]). All of these are evaluated relative to those of other suppliers by customers (channel members and end users) when they consider and select offerings, which is what conceptually distinguishes a firm's market share (how the firm's sales compare with those of the total market) from its sales revenue (the number of units sold × price). Importantly, this means that (unlike sales revenue) market share is not a component variable in any indicators of firm economic performance,[ 6] so there is no synthetic (or "hard-wired") market share–firm economic performance relationship.
Historically, the empirical literature provided conflicting and equivocal answers concerning the "main effect" relationship between firms' market share and their economic performance (e.g., [11]; [36]; [37]). However, the recent E-H (2018) meta-analysis using more sophisticated methodological approaches has provided new insight on this question, showing a generally positive effect of market share on firm economic performance. We corroborate this in our data and focus our hypothesizing on why this relationship exists and how this "why" understanding may help explain and predict differences in the strength of the relationship across firms and industries.
While several explanations have been independently proffered for why a firm with higher market share may enjoy superior economic performance, three mechanisms are much more widely discussed than others. As Figure 1 shows, we focus our theorizing on these mechanisms and consider how each may link a firm's market share with its profit.
Graph: Figure 1. Conceptual framework.
The first proposed mechanism by which market share may be linked with firm profit is via market power (i.e., the firm's ability to influence the price of its product/service offerings by exercising control over demand, supply, or both; e.g., [10]; [59]). Industrial organization theory posits that firms enjoy superior profit when they are able to charge higher prices than rivals, which is determined by the availability of alternatives to customers and firms' ability to create and/or control resources that give them stronger market positions (e.g., [57]). Market share may be a resource that provides a firm with the opportunity for greater market power over both "upstream" suppliers and "downstream" channels and customers and thereby control prices in several ways.
For upstream suppliers, buyer firms with higher end-user market share are more attractive, which may allow them to negotiate lower prices and/or higher-quality inputs from their suppliers ([ 9]). For example, Apple's smartphone market share allows it to both charge app developers for selling their products and enforce strict quality controls on the apps it sells. It may also increase supplier willingness to cooperate with others in the buyer's supply network to further lower the buyer's input costs and improve input quality ([28]). For downstream channels, higher–market share firms are more attractive upstream partners because they generate end-user demand for more and/or higher-value products. They may also attract larger customer numbers and/or more frequent interactions for channels to engage in cross-selling. This may enable higher–market share firms to negotiate better list prices than rivals in downstream channels and to benefit from greater channel cooperation (e.g., preferred shelf-space, merchandizing support). For example, PepsiCo's snacks division leverages its leading market share position to obtain preferential shelf and display access in many U.S. retail chains. The input and go-to-market cost and quality benefits of higher–market share firms should allow them to provide better value offerings, which may thus allow them to charge higher prices to end users (as in the case with Apple) and/or enjoy higher profit margins on each unit sold (e.g., Walmart). Thus,
- H1: The positive effect of market share on firm profit is mediated by the firm's market power.
The second theorized mechanism by which a firm's market share may lead to profit is via increasing the firm's operating efficiency (e.g., [17]). Disputing market power arguments, the "Chicago school" in economics argues that market share is an outcome of firm efficiency that allows a firm to sell quality-equivalent offerings at lower prices than rivals, attracting greater demand (e.g., [14]; [50]). Following this logic, strategic management scholars propose that higher market share may also allow firms to further increase their efficiency in a recursive relationship with lowering firm costs via learning effects (e.g., [ 1]; [29]). Much of this logic is framed in terms of a firm's position on the production "experience curve" as a function of the volume of units sold, with greater experience allowing production-related learning and lower production costs (e.g., [30]). Thus, firms selling a greater number of units produce more and learn how to do so more efficiently. For example, Tesla has used its greater accumulated experience in producing electric vehicles (EVs) to lower its costs compared with rivals.
Conceptually, this may also be possible via market share impacting the number of interactions a firm has with suppliers, channels, and customers, enhancing opportunities for higher–market share firms to learn and use knowledge gained to improve their supply-and-demand chains ([55]). For example, Tesla has used its greater EV sales to learn how to drive improvements in battery designs and configurations from suppliers as well as to optimize its own software to increase EV range. More interactions also increase the likelihood that suppliers, channels, and customers will trust higher–market share firms, increasing information sharing, lowering coordination costs, and enhancing cooperation in changes designed to enhance the firm's supply-and-demand chains ([16]; [27]). This should enable higher–market share firms to lower costs and enhance supply-and-demand chain quality and reliability, allowing superior value offerings for customers and/or greater margins. Thus,
- H2: The positive effect of market share on firm profit is mediated by the firm's operating efficiency.
The third mechanism by which market share may enhance firm profit is by signaling unobserved quality. Information economics theory posits that customers' limited evaluative knowledge often makes it difficult for them to observe "true" product/service quality (e.g., [38]; [41]). Empirical studies also show that customers are often unable to accurately (or confidently) evaluate an offering's quality prior to making purchase decisions, and they frequently rely on indirect cues (e.g., [47]; [61]). Market share may signal quality by increasing the credibility of firm claims and thereby lowering customer perceived risk ([21]; [34]). Customers may also infer that "everyone can't be wrong" in choosing the offerings of a high–market share firm (e.g., [18]). For example, Toyota campaigns have touted that its products are "#1 for a Reason." Thus, to the extent that market share signals higher quality, it should increase future demand and reduce customer churn. It may also lower the firm's costs relative to rivals, because alternative ways to signal quality (e.g., advertising) may be more costly.
Market share may also signal quality to suppliers and channel members. Firms that are perceived to be producing high-quality offerings may be viewed by suppliers as not just attractive buyers, in terms of their own demand, but also as potentially providing a halo image spillover benefit. Similar to customers viewing them as having "too much to lose" to provide inferior offerings, supplier choices made by high–market share firms may be viewed as being based on ensuring high quality and reliable inputs to protect their reputation and market position. For example, Apple's suppliers are frequently identified as such in business press reports. This could also apply to channel partners where selling offerings that are perceived as higher quality can provide a halo effect making the channel member more attractive to other suppliers and end-user customers (e.g., [42]). All of these arguments suggest the following:
- H3: The positive effect of market share on firm profit is mediated by the firm's perceived quality.
Prior research suggests that the value of market share varies across industries (e.g., [ 4]), indicating that setting market share goals may be more beneficial for some firms than others. To explore this, E-H's (2018) meta-analysis examines the sample characteristics most commonly reported in prior studies and reports that market share is more valuable in business-to-customer (B2C) markets and in markets with medium market concentration, whereas it is less valuable in the banking industry. While offering initial useful insight to managers, these boundary conditions are limited in number and scope—and the "why" mechanisms involved are unobserved. Robust empirical understanding of the mechanisms using direct assessments should allow additional boundary conditions to be identified and provide empirically verified principles for managers to distinguish when they should and should not care about market share.
To provide an initial assessment of the predictive value of our mechanism results and offer new insights for managers, we next examine the extent to which the market share–profit relationship varies under conditions in which each of the three mechanism in turn may be expected a priori to be more versus less important. For each mechanism, we identify a condition expected to be particularly impactful on that particular market share–profit pathway. However, in our analyses we also allow for the possibility that each of the conditions we identify may affect the strength of all three mechanisms linking market share with profit. First, in terms of market power we examine industries characterized by higher customer switching costs, where firms are more easily able to retain customers. Firms should benefit more from the market power provided by market share when switching costs are high because they are better placed to increase prices without fear of customers switching ([23]; [58]; [60]).
Second, in terms of the value of operating efficiency in explaining the market share–profit relationship, the literature suggests that cost-reducing learning effects are more likely earlier in the life of a firm (e.g., [66]). For example, "experience effect" studies of the value of a firm's cumulative doubling of output show that this is more likely to occur early in a firm's existence (e.g., [31]). In addition, learning effects require changing and adapting firms' processes—which tend to become more rigid over time (e.g., [54]). Thus, younger firms are less knowledgeable in their operations and less "set in their ways," providing incentives to seek out the learning opportunities presented by market share and the ability to exploit the efficiency-enhancing knowledge gained via process changes.
Third, to explore conditions where the quality-signaling value of market share may be stronger, we examine differences between "service-dominant" and "goods-dominant" industries.[ 7] A key difference between these markets is the greater intangibility of service offerings, which creates more quality uncertainty for customers ([68]). Under such conditions, customers are more likely to use cues such as market share as indicators of the quality of a firm's offerings (e.g., [12]). Interestingly, this prediction is the opposite of E-H (2018), who reason that physical goods manufacturers may benefit more from efficiency, and that this may be more important in driving profit than any dampening of the quality-signaling effect of market share in physical goods-focused markets. We explore this reasoning empirically when we directly examine the three mechanisms underpinning the market share–profit relationship.
We therefore hypothesize the following:
- H4: The effect of market share on firm profit via market power is stronger in marketplaces with higher switching costs.
- H5: The effect of market share on firm profit via efficiency is stronger for younger firms.
- H6: The effect of market share on firm profit via perceived quality is stronger for firms selling service- versus product-dominant offerings.
We combine secondary data from a variety of sources. From Compustat, we obtained data to construct measures of market power and operating efficiency, firm economic performance indicators, firm-specific controls, and a set of industry and competitive context control variables. Equitrend provided data on the perceived quality of firms' offerings. To calculate measures of unit market share, we use unit sales data from the Global Market Information Database (GMID). We assembled our initial data set by merging data from Compustat and GMID. To test the mediation hypotheses, we also require data from Equitrend, for which our access covers only the years 2000–2013. Because each data source has distinct firm and year coverage, the compiled data set used to confirm the main effect of market share on firm profit and test the hypothesized mediation effects contains 3,058 firm-year observations from 244 individual firms, operating in 126 North American Industry Classification System (NAICS) four-digit industries, 2000 through 2013. The average firm in this sample has $13.81 billion in assets and has been operating for 45 years. Table 1 shows summary statistics and correlations for the main variables in our sample and additional details are contained in Web Appendix 1. To test H4–H6, we also required American Customer Satisfaction Index (ACSI) data (to measure switching costs), which reduced our sample for testing these three hypotheses to 2,629 firm-year observations from 207 firms (2000–2013).[ 8]
Graph
Table 1. Descriptive Statistics and Correlations (N = 3,058).
| Variables | Mean | SD | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 |
|---|
| 1. Firm Profit ($M) | 839.10 | 1,104.10 | 1.00 | | | | | | | | | | | | | |
| 2. Market Share (%) | 6.85 | 9.59 | .14 | 1.00 | | | | | | | | | | | | |
| 3. Sales Revenue ($M) | 3,907.08 | 5,747.18 | .77 | .14 | 1.00 | | | | | | | | | | | |
| 4. Market Power (%) | 30.79 | 10.90 | .23 | .13 | .09 | 1.00 | | | | | | | | | | |
| 5. Firm Efficiency (Index) | 50.09 | 9.06 | .10 | .08 | .38 | .23 | 1.00 | | | | | | | | | |
| 6. Perceived Quality (Index) | 65.35 | 16.95 | .18 | .07 | .06 | .35 | −.05 | 1.00 | | | | | | | | |
| 7. Firm Size ($B) | 13.81 | 62.89 | .39 | .16 | .69 | .17 | .03 | .22 | 1.00 | | | | | | | |
| 8. Market Growth ($M) | 122.23 | 523.18 | .10 | .02 | .03 | .14 | −.04 | −.10 | .30 | 1.00 | | | | | | |
| 9. Advertising ($M) | 53.92 | 255.75 | .44 | .34 | .36 | .07 | −.01 | .01 | .31 | .03 | 1.00 | | | | | |
| 10. R&D ($M) | 64.92 | 366.32 | .40 | .30 | .22 | .15 | −.02 | −.07 | .32 | .05 | .48 | 1.00 | | | | |
| 11. Service Indicator (0/1) | .19 | .39 | −.02 | −.10 | −.15 | −.03 | .18 | −.05 | .08 | −.04 | −.08 | −.05 | 1.00 | | | |
| 12. Switching Costs (Index) | −.01 | 1.11 | .33 | −.14 | .06 | .16 | .07 | .02 | .18 | .14 | .19 | .07 | .11 | 1.00 | | |
| 13. Firm Age (Years) | 45.30 | 41.30 | .21 | .05 | .09 | .14 | .10 | .21 | .18 | −.08 | .49 | .41 | −.08 | .34 | 1.00 | |
| 14. Niche Focus (Index) | 2.43 | 8.10 | .02 | −.15 | −.02 | .12 | .06 | .05 | −.07 | −.06 | .06 | .13 | .12 | .01 | −.06 | 1.00 |
1 Notes: All descriptive statistics are for the "raw" (i.e., untransformed) variables. Correlations with an absolute value larger than.046 are significant at p < .01, and those greater than.035 are significant at p < .05.
The Appendix contains definitions and operationalization details of all variables described next.
Market share is the percentage of a market's total sales garnered by a firm over a specified time period ([25]). The market may consist of all suppliers selling products/services with the same characteristics, or those that are thought of similarly by customers and are purchased for the same use. We follow [35] to compute a measure of market share using a set of competitors and market definitions derived from business descriptions in firm 10-Ks. This allows market definitions to be dynamic, where a firm may move in and out of any given market depending on whether its offerings changed over time and thus compete with a different set of firms.
To compute market share, we divide the total sales of each firm by the aggregate sales for that market for that year, where the market is dynamically defined as described previously using data from all 22,076 firms in Compustat for the 2000–2013 period. In defining markets, we note that each firm has a similarity/competition score with respect to any other firm (i.e., all possible dualities are computed) in the Compustat database. In line with [35], the number of competitors can be defined using a threshold of similarity scores and/or specified number of nearest neighbors (e.g., 50 or 20). We combine the two approaches and specify 50 as the largest number of neighbors, while also imposing a minimum threshold limit. Thus, our market definition comprises a maximum of 50 firms per industry, while allowing for fewer firms, to maintain a minimum level of similarity among competitors in the same market.[ 9]
To assess the robustness of the findings using this dynamic measure of market share, we also use a more static approach, defining markets via each firm's primary NAICS designation using the four-digit level that researchers suggest most closely represents the real "competed" market (e.g., [44]). To calculate this, we first collect the total revenue-by-industry data that comprise gross domestic product (i.e., total expenditures on products and services) for all four-digit NAICS industries from the U.S. Bureau of Economic Analysis, which allows us to account for the sales of firms that are private, small, or otherwise not available in Compustat. We then divide the total sales of each firm by the gross domestic product value for that four-digit NAICS industry for that year. Firm market shares are computed from their revenues in their primary NAICS markets.
We use net income as our primary measure of firm profit, obtained from Compustat. We use this indicator of absolute firm profit (while controlling for asset size in our model) because economic theories of the value of market share assume that maximizing the amount of profit—not the efficiency with which profit is generated, which is what "return on asset" (or investment) relative profit measures capture—is a firm's superordinate performance objective.
We use profit elasticity relative to the industry average (similar to [ 8]) to indicate firm-level market power. This is calculated by estimating regressions of firms' profit (net income) on their total variable costs for each industry as follows:
Graph
where π is firm profit and tvc is the firm's total variable cost (Cost of Goods Sold + Selling, General and Administrative Expenses) for firm i at time t. Both profit and variable costs are scaled by firm size (total assets). Because profit and costs are natural log transformed, the β from this regression captures the average profit elasticity within the industry, with less negative βs indicating the average ability of firms within the industry to mark up prices when costs rise and thus exercise market power (e.g., [39]). Firm-specific residuals measure each firm's margins relative to its industry's average, providing an indicator of firm's market power ([ 8]). Positive residuals (equivalent to less negative elasticities) indicate greater market power, and negative residuals (i.e., more negative elasticities) indicate weaker market power. Web Appendix 2a indicates favorable face validity for this measure.
From an economic theory viewpoint, this concerns producing goods and services in ways that optimize the combination of inputs to produce maximum output at the minimum cost ([ 5]). To operationalize productive (in)efficiency, we use a stochastic frontier estimation approach. Following [ 5], we use operating expense as the input and total sales as the output. In stochastic frontier estimation, the firm in the industry with the lowest input requirements to produce a given set of outputs forms the efficiency frontier and the closeness of a firm's inputs-to-outputs to this frontier determines its relative (to the industry's most efficient firm) efficiency. Web Appendix 2b provides evidence of strong face validity for this measure.
We use the perceived quality measure of brands from the Equitrend database, which comprises consumer ratings on an 11-point perceived quality scale. For multibrand firms, we take the mean perceived quality of all brands owned by the firm.[10] Face validity assessments for this measure (see Web Appendix 2c) provide strong support for the measure.
We use ACSI data and follow [53] to construct an industry-level measure of switching costs as the "excess loyalty" displayed by customers to suppliers using the residual of regressing each industry's customers' loyalty onto its customers' satisfaction, controlling for time fixed effects (FEs). This measure has been shown to have strong face validity ([53]), and we also find evidence of this (Web Appendix 3).
Service- (vs. product-) dominant industries is a dummy variable identifying firms operating in nonbanking (banks have idiosyncratic characteristics we later explore) service-focused industries using Fama–French industry definitions ([22]).
Firm age is the number of years since the firm's founding using information from annual reports and websites.
In addition to firm and year FEs used to control for unobserved heterogeneity, we employ several firm- and industry-level covariates in our analyses, including firm size, operationalized as the logarithm of each firm's total assets to account for scale economies not captured by market share, and the firm's advertising and research-and-development (R&D) expenditures to control for firm-level heterogeneity. We also control for market growth that may affect the profit outcomes of market share ([56]), captured as the year-to-year change in total market sales.
The Appendix and Web Appendix 1 summarize descriptive statistics for all variables used in our analyses. To enable log-log specification and interpretation in our analyses and reduce deviations from normality present in several of our measures (market share, firm profit, market power, firm efficiency, perceived quality, advertising expense, R&D expense, and market growth), we applied log transformations to our data.[11]
We empirically test the hypothesized relationships using a fixed-effects autoregressive (FE-AR) estimation approach ([65]) for several reasons. First, we are using panel data, and the Hausman test indicates that an FE correction is needed to address unobserved heterogeneity and separate between time-variant and -invariant firm-specific errors. Second, several of our measures are longitudinally persistent, raising concerns about serial correlation—the AR correction of the errors addresses any potential bias to the estimates. The modified Durbin–Watson and Baltagi–Wu LBI tests indicate that an AR1 correction is appropriate. In addition, we control for heteroskedasticity using cluster-adjusted robust standard errors at the firm level. Finally, we estimate our hypothesis-testing models using generalized least squares (GLS), because OLS are statistically inefficient and may result in biased inference in the presence of serially correlated residuals.
We first verify the average positive relationship between market share and profit (E-H 2018) and estimate the total effect using the following model specification:
Graph
( 1)
where i stands for firm and t for time (year), ζi is a time-invariant firm FE, and εi, t + 1 is the random error representing all unobserved influences on future profit, modeled as an AR1 process such that εi, t + 1 = ρεi, t + ηi, t + 1 and where |ρ|<1 and ηi, t + 1 is an independent and identically distributed (i.i.d) error. Market Share, Firm Size, Advertising, R&D, and Market Growth are as described previously, and Year FEs are mutually exclusive year dummies. Lagged regressors are used to alleviate concerns due to simultaneity and reverse causality (i.e., future profit should not impact past market share).
Having selected an appropriate estimation approach given the nature of our data, we next deal with potential endogeneity concerns with respect to omitted variables—of which reverse causality and simultaneity are special cases ([65]). We examine the potential for the presence and effect of such endogeneity concerns using a Gaussian copula correction to Equation 1 and assess the presence and effect of any endogeneity (including potential selection bias introduced by the various data sets on which we draw for our measures) via a likelihood ratio test of whether there is a significant difference between the uncorrected set of parameter estimates and the endogeneity-corrected set ([65]).[12] Once we show that potential endogeneity issues are not material, we empirically test H1–H3 using an identical FE-AR approach by estimating the following equations:
Graph
(2\rm a)
Graph
(2\rm b)
Graph
(2\rm c)
Graph
(2\rm d)
where Market Power, Firm Efficiency, Perceived Quality, Switching Costs, Services Dummy, and Firm Age are as described in the variable measurement section, and all other variables and subscripts follow Equation 1. Finally, we empirically test H4–H6 by estimating the moderated-mediation contingencies and include interactions between Market Sharei,t and Switching Costsi,t, Services Dummyi,t, and Firm Agei,t in Equations 2a–2d. To estimate the relative effects of the three hypothesized mediation mechanisms (market power, firm efficiency, and quality signaling) and three moderated-mediation contingencies (switching costs, firm age, and services), we follow [51] using [64] approach to augment the FE-AR estimation.
Prior to testing the hypothesized mechanisms, we first verify the main effect results indicated in the E-H (2018) meta-analysis in our sample using several variants of the model specification detailed in Equation 1. We begin by estimating a model with FEs and cluster-adjusted robust standard errors that includes only the covariates as predictors (M1), to which we then add market share (M2), allowing us to verify the main effect of market share on firm profit and reveal its incremental predictive power. We also estimate this same model using an FE-AR error correction and cluster-adjusted robust standard errors (M3) to demonstrate the stability of the estimates across the different statistical corrections proposed. In M4 we examine whether the reported estimates suffer from endogeneity bias by including a Gaussian copula for the Market Share variable as a control function to empirically correct endogeneity bias. The likelihood ratio test for joint parameter differences ([65]) indicates that the endogeneity-corrected estimates in M4 are not statistically different from those in M3.
As Table 2 shows, the estimates are consistent across all four models, demonstrating the robustness of the effect of market share on firm profit. In addition, while the Gaussian copula estimate in M4 is significant (.048, p < .05) indicating the presence of some omitted variable endogeneity, the likelihood ratio test indicates no significant difference in the market share parameter estimates between M3 (β = .137) and M4 (β = .159). This supports the use of an FE-AR( 1) (i.e., model specification M3) estimation approach and confirms that any remaining bias is modest and does not substantively impact the estimates. In a robustness check, we also replaced the dynamic market share measure with a four-digit NAICS alternative and again confirmed the main effect (Web Appendix 5). Finally, we further verified that endogeneity bias does not unduly influence our findings using a difference-in-differences version of Equation 1 comparing the market share–profit relationship for firms in industries that experience an exogenous demand shock (exit of bankrupt firms) with those that do not. The results (Web Appendix 6) again confirm the main effect findings.
Graph
Table 2. Main Effect of Market Share on Firm Profit.
| Models and Dependent Variables |
|---|
| M1 | M2 | M3 | M4 |
|---|
| Independent Variables | Profit(t + 1) | Profit(t + 1) | Profit(t + 1) | Profit(t + 1) |
|---|
| Main Effect | | | | |
| Market Share(t) | | .153** | .137** | .159** |
| | (.053) | (.038) | (.052) |
| Controls | | | | |
| Firm Size(t) | .228* | .208*** | .521*** | .291*** |
| (.113) | (.067) | (.045) | (.051) |
| Advertising(t) | .234*** | .130** | .073*** | .098* |
| (.061) | (.045) | (.023) | (.044) |
| R&D(t) | .061* | .044 | .066*** | .042*** |
| (.027) | (.025) | (.014) | (.010) |
| Market Growth(t) | .020 | .012 | .002 | .029* |
| (.017) | (.020) | (.002) | (.013) |
| Market Share(t)COPULA | | | | .048* |
| | | | (.021) |
| Specification Tests | | | | |
| Wald χ2 | 125.32 | 198.12 | 188.36 | 115.92 |
| R2 | .57 | .59 | .58 | .59 |
| Rho_AR | | | .40 | .43 |
- 2 *p < .05.
- 3 **p < .01.
- 4 ***p < .001.
- 5 Notes: All model specifications estimated using 3,058 firm-year observations. M1/M2: GLS estimation, FEs and cluster-adjusted robust standard errors. M3/M4/M5: GLS estimation, FEs with AR errors and cluster-adjusted robust standard errors. Z-test difference in share coefficients between M3 (.137) and M4 (.159) = .64 (p > .05).
Collectively, these analyses verify the main effect results in E-H (2018) that, on average, firm market share positively predicts future firm profit—and the effect sizes reported on Table 2 are both consistent and aligned with the average elasticity of.132 reported by E-H (2018), further enhancing confidence in our findings. Table 2 results also show the suitability of the FE-AR error correction and cluster-adjusted robust standard errors GLS estimation approach (model specification M3), which we employ in the hypothesis-testing analyses.
As Table 3 shows, in testing H1–H3 we find support for both market power in M1a (.230, p < .001) and quality signaling (.141, p < .05) in M1c as mechanisms linking market share with firm profit. However, while M2 confirms that firm efficiency predicts firm profit (.129, p < .001), M1b reveals that a firm's efficiency is not predicted by its market share (.024, p > .1). Thus, on average we find no evidence supporting efficiency as a mechanism linking firm market share and profit in our sample. Overall, these results provide support for H1 and H3 but not for H2. As M2 shows, all three of the mechanism variables are significant predictors of firm profit, and the main effect of market share becomes insignificant (.031, p > .10) in the presence of these three variables. To examine the relative strength of the mediator role played by the three mechanism variables in explaining the market share–profit relationship, we follow [64] approach. This reveals that the three mechanisms collectively explain 77.37% of the total effect of market share on firm profit, with 63.21% of this flowing through market power, 33.96% via perceived quality, and 2.83% through firm efficiency.
Graph
Table 3. Mechanism for Market Share Effect on Firm Profit.
| Models and Dependent Variables |
|---|
| M1a | M1b | M1c | M2 |
|---|
| Independent Variables | Power(t + 1) | Efficiency(t + 1) | Quality(t + 1) | Profit(t + 1) |
|---|
| Direct Effect | | | | |
| Market Share(t) | .230*** | .024 | .141* | .031 |
| (.081) | (.016) | (.065) | (.018) |
| Indirect Effect | | | | |
| Market Power(t) | | | | .302*** |
| | | | (.042) |
| Firm Efficiency(t) | | | | .129*** |
| | | | (.029) |
| Perceived Quality(t) | | | | .274*** |
| | | | (.061) |
| Controls | | | | |
| Firm Size(t) | .029* | .027*** | .039*** | .210*** |
| (.013) | (.006) | (.008) | (.029) |
| Advertising(t) | .020 | .021 | .022* | .090* |
| (.023) | (.020) | (.010) | (.043) |
| R&D(t) | .032** | .013*** | .028** | .023*** |
| (.011) | (.002) | (.011) | (.005) |
| Market Growth(t) | .012 | .007 | .012 | .008 |
| (.019) | (.009) | (.010) | (.007) |
| Specification Tests | | | | |
| −Log-likelihood | 2,810.17 | | | |
| R2 | .16 | .18 | .10 | .68 |
- 6 *p < .05.
- 7 **p < .01.
- 8 ***p < .001.
- 9 Notes: 3,058 firm-year observations covering 244 firms for the 2000–2013 period (Equitrend available 2000–2013). Total effect (from Table 2: M3).137 (100.00%) minus direct effect (from M1a).031 (22.63%) = indirect effect of.106 (77.37%). Indirect effect via ( 1) Power = .067 (63.21%); ( 2) Quality = .036 (33.96%); and ( 3) Efficiency = .003 (2.83%).
To check the robustness of the mechanism results, we conducted four additional analyses. First, to check for any potential scale effect of absolute sales revenue beyond firm size, we reestimated our model using market share ranks and adding firm sales revenue as a separate control. The estimates replicated the hypothesis-testing results (Web Appendix 7). Second, to check for any potential biasing effect of firm orientation to market share ([43]) we used text analysis of 10-K reports to construct an annual measure of each firm's market share focus based on the number of times "market share" is mentioned relative to the total number of words. When this is added to our model, we find that the results remain essentially unchanged (Web Appendix 8). Third, to ensure that results are robust to alternative firm performance measures, we replaced net profit in turn with return on assets and Tobin's q as dependent variables. As shown in Web Appendices 9 and 10, we replicate the hypothesis-testing results. Fourth, we also checked that a firm's competitor orientation—a potential fourth mechanism linking market share (negatively) with firm profit ([ 3])—does not explain additional variance in the market share–profit relationship. Using 10-K reports and [ 7] text-based measure, we computed the competitor orientation of each firm in our sample and included this in our model. As Web Appendix 11 shows, we find that while competitor orientation predicts firm market share, it does not materially affect the market share–profit relationship.
Having demonstrated the robustness of the hypothesized mechanism results, we next examine whether the market share–profit relationship may be stronger in industry and firm conditions in which each of the three mechanism variables in turn may be expected a priori to be more versus less important as captured in H4–H6. The results are summarized in Table 4, with M1 showing that firms in industries with higher customer switching costs are more profitable (.137, p < .05), and M2 supporting H4 by confirming that market share is more valuable in such industries (.087, p < .001) via its stronger effect on market power (.157, p < .05). In addition, M4c reveals that firms also gain stronger perceived quality benefits from market share in industries with higher switching costs (.203, p < .05), suggesting that some of the switching costs we observe are due to customers continuing to choose a provider because of positive relational bonds that may influence both customers and others' perceptions of the quality of such firms' offerings.
Graph
Table 4. Main Effect and Mechanisms for Market Share Effect on Firm Profit in Hypothesized Moderators.
| Model Specifications (M) and Dependent Variables |
|---|
| M1 | M2 | M3a | M3b | M3c | M3d | M4a | M4b | M4c | M4d |
|---|
| Independent Variables | Profit(t + 1) | Profit(t + 1) | Power(t + 1) | Efficiency(t + 1) | Quality(t + 1) | Profit(t + 1) | Power(t + 1) | Efficiency(t + 1) | Quality(t + 1) | Profit(t + 1) |
|---|
| Direct Effects | | | | | | | | | | |
| Market Share(t) | .118*** | .114*** | .105* | .031 | .278** | .017 | .136*** | .028 | .149*** | .030 |
| Indirect Effects | | | | | | | | | | |
| Market Power(t) | | | | | | .210*** | | | | .223*** |
| Firm Efficiency(t) | | | | | | .075** | | | | .083*** |
| Perceived Quality(t) | | | | | | .169*** | | | | .163*** |
| Moderators | | | | | | | | | | |
| Switching Costs(t) | .137* | .149* | .093* | .005 | .051* | .013 | .107 | .015 | .077* | .027 |
| Firm Age(t) | .178 | .208 | −.002 | .013 | .006 | .015* | .034 | −.031 | .004 | .019 |
| Services Dummy(t) | −.058* | −.059* | .017 | −.033* | .008 | −.004 | .093 | .509*** | .028 | −.006 |
| Interaction Effects | | | | | | | | | | |
| Market Share(t) × Switching Costs(t) | | .087*** | | | | | .157* | .017 | .203* | .033 |
| Market Share(t) × Firm Age(t) | | −.069*** | | | | | −.048 | −.109*** | −.092* | −.043 |
| Market Share(t) × Services Dummy(t) | | .056*** | | | | | −.006 | .148*** | .012 | .020 |
| Controls | | | | | | | | | | |
| Firm Size(t) | .514*** | .534*** | .025*** | .031*** | .046*** | .028*** | .030*** | .083* | .041*** | .046*** |
| Advertising(t) | .278*** | .281*** | .008 | .022 | .006 | .011 | .007 | .010 | .039 | .039*** |
| R&D(t) | .274*** | .272*** | .039*** | .010 | .059* | .034*** | .029*** | .012 | .062*** | .031*** |
| Market Growth(t) | .014 | .015* | .011 | .008*** | .002 | .004 | .011 | .017 | .012 | .003 |
| Specification Tests | | | | | | | | | | |
| Wald χ2 | 303.11 | 358.07 | | | | | | | | |
| −Log-likelihood | | | 2,489.31 | | | | 2,913.87 | | | |
| R2 | .50 | .52 | .24 | .21 | .22 | .69 | .25 | .29 | .26 | .70 |
- 10 *p < .05.
- 11 **p < .01.
- 12 ***p < .001.
- 13 Notes: 2,629 firm-year observations covering 207 firms for the 2000–2013 period (sample size due to ACSI data availability).
The interactions reported for M2 also show that market share is generally less valuable for older firms (−.069, p < .001), and consistent with H5, the mechanism estimates in M4b provide strong evidence supporting the expected effect of market share on firm efficiency being weaker for older firms (−.109, p < .001). This is aligned with our rationale that efficiency-enhancing learning effects associated with market share accrue mainly to firms that are earlier in their development. M4c estimates also reveal that older firms benefit less from market share via quality signaling (−.092, p < .05). We reason that older firms that have been in the marketplace for longer are likely to be better known and also that firm age may indicate a firm's stability and lower risk, which reduce the signaling value of its market share.
In terms of services-dominant firms, the significant positive estimate in M2 for the services × market share interaction (.056, p < .001) indicates that service firms benefit more from market share. However, our mechanism estimates in M4c show that this is not a result of the expected strengthening of the quality-signaling benefit of market share (.012, p > .10) posited in H6 but rather, as shown in M4b, that service firms benefit more from the efficiency-enhancing effect of market share (.148, p < .001).[13] Because controlling for scale effects via firm size isolates the efficiency-enhancing learning effects of market share, this finding suggests that market share provides a greater opportunity for service firms to learn how to operate more efficiently and to use this knowledge to change their operations to do so. We reason that this may be because the greater direct customer interactions from higher market share are more valuable in helping service firms learn how to efficiently deal with customer heterogeneity, and that applying what is learned may also be less capital-intensive for service firms (vs. manufacturers).
To provide additional insight into how the hypothesized moderators affect the profit value of market share via the three mechanisms, we examined these effects in an additional analysis (Table 5). Of the.086 total effect (elasticity) of market share on profit when the moderator variables are included in the model,.056 is indirect (65% of the total) via the three mechanisms, with 62% of this flowing through market power, 6% through firm efficiency, and 32% via perceived quality. Consistent with the H4 testing results (Table 4), the effect of market share on firm profit is strengthened by switching costs, with the total effect amplified by.287 for each unit increase in switching costs, of which.195 is indirect via market power (50.9%), firm efficiency (2.5%), and perceived quality (46.6%). These direct and indirect effects of switching costs on market share's effect on firm profit are proportionately lower (higher) at lower (higher) levels of switching costs (i.e., ± one standard deviation around average switching costs) with the indirect effects flowing through the three mechanisms in very similar percentages.
Graph
Table 5. Indirect Effects for Market Share Effect on Firm Profit in Hypothesized Moderators.
| Market Share–Profit Effects | Indirect Effect Mechanisms |
|---|
| Moderator Variable Conditions | Total Effect | Direct Effect | % of Total | Indirect Effect | % of Total | Power | Efficiency | Quality |
|---|
| Overall | .086* | .030 | 34.9% | .056* | 65.1% | 62.0% | 6.0% | 32.0% |
| Switching costs | .287*** | .092* | 32.1% | .195*** | 67.9% | 50.9% | 2.5% | 46.6% |
| +1 SD | .345*** | .111* | 32.2% | .234*** | 67.8% | 51.2% | 2.4% | 46.4% |
| −1 SD | .218*** | .073* | 33.5% | .145*** | 66.5% | 51.1% | 2.4% | 46.5% |
| Service dominant | .032* | .020 | 62.5% | .012 | 37.5% | 41.0% | 21.0% | 38.0% |
| Product dominant | −.032* | −.010 | 31.2% | −.022 | 68.8% | 54.0% | 3.0% | 43.0% |
| Firm age | −.136*** | −.014 | 10.3% | −.122*** | 89.7% | 12.1% | 45.5% | 42.4% |
| +1 SD | −.170*** | −.018 | 10.6% | −.152*** | 89.4% | 17.1% | 40.2% | 42.7% |
| −1 SD | −.081* | .011 | −13.6% | −.092* | 113.6% | 2.7% | 56.8% | 40.5% |
- 14 *p < .05.
- 15 **p < .01.
- 16 ***p < .001.
- 17 Notes: 2,629 firm-year observations covering 207 firms for the 2000–2013 period (sample size due to ACSI data availability).
Consistent with H5 testing results (Table 4), the total effect of market share on firm profit is also amplified for service-dominant firms by an extra.032, of which.012 is indirect (38% of the total) and flows through market power (41.0%), firm efficiency (21.0%), and perceived quality (38.0%). Meanwhile, for product-dominant firms, the total effect is reduced by −.032, of which −.022 is indirect, with 54.0% flowing through market power, 3.0% through firm efficiency, and the remaining 43.0% via perceived quality.
Finally, in line with H6 testing results (Table 4), Table 5 shows the effect of market share on profit is weakened by firm age with each additional year reducing the total effect of market share on profit by −.136, of which −.122 is indirect (90% of the total) and flows through market power (12.1%), firm efficiency (45.5%), and perceived quality (42.4%). As we expected, the total effect of firm age on the market share–profit relationship is more pronounced for very high (old) versus very low (young) age levels, with a marked increase in the indirect effect flowing through firm efficiency (from 40.2% to 56.8%) and decrease in that flowing through market power (17.1% to 2.7%) in the case of very young firms. This is consistent with our Table 4 hypothesis testing results revealing stronger efficiency gains with market share for younger firms.
Aligned with E-H's (2018) finding that 82% of market share–performance elasticities in prior research are positive (82% of the same elasticities in our sample are also positive), our hypotheses are framed in terms of a net positive performance effect of market share. However, conceptual arguments concerning potential negative outcomes of market share have also been proposed (e.g., E-H 2018; [34]). Drawing on our theorizing, we expect that the three mechanisms we identify should empirically capture any negative and positive effects of market share. For example, any associated diseconomies of scale will reduce a firm's efficiency while a reduction in perceived exclusivity will affect the quality-signaling value of market share. To empirically verify this expectation, we identify two conditions under which market share's positive benefits may be outweighed by negative consequences, such that larger market share might reduce firm profit and reestimate the mediation effects of the market power, firm efficiency, and quality-signaling mechanism in these conditions.
One condition in which market share may negatively predict profit concerns firms with a strategic focus on serving a smaller segment of a market, usually a group of customers with a distinct set of needs and requirements (e.g., [49]). For example, Louboutin specializes in high-fashion stiletto shoes. By serving distinctive needs, niche-focused firms make money by occupying positions in a segment of a broader market in which competition is more limited (e.g., [19]). As a result, they may not serve enough customers to gain market power benefits from market share, and their specialist positioning may diminish any quality-signaling benefit. They are also unlikely to gain from any learning effects in production. However, niche-focused firms with higher overall market shares are likely to have achieved this by selling to customers beyond their original niche ([62]). This may negatively impact the firm's profitability by reducing its original niche appeal via a negative effect on perceived quality (e.g., [34]) and also attract more competition (e.g., [32]). These downsides may outweigh any potential market power and/or firm efficiency benefits of having a larger market share.
Another circumstance when market share may negatively impact profit is when firms "buy" market share by lowering prices relative to rivals. This is analogous to findings in the sales promotion literature that price promotions often produce negative returns (e.g., [33]). In this circumstance, any market share gain via greater market power and the ability to charge higher prices is not only relinquished but reversed. In addition, because there is a price-perceived quality heuristic among customers in many markets (e.g., [52]), charging lower prices may offset any quality-signaling benefit of higher market share, and the net result on perceived quality could be negative. Our previous results suggest that in most circumstances, these negative market power and quality-signaling effects are likely to outweigh any firm efficiency gains via learning produced by increasing market share.
To assess the robustness of our mechanism results under conditions when the market share–profit relationship may be negative, we first identified firms that are likely pursuing a niche strategy by combining a new text measure indicator of the degree to which a firm has a niche strategic emphasis (for details, see Web Appendices 4a and 4b) with the number of brands they market (both firms with both a high niche-focus in their product-market coverage strategy and those that offer only a single brand are likely to be niche firms). The face validity assessments in Web Appendices 4a and 4b support this identification logic. Second, to identify firms that may be "buying" market share, we created a dummy variable indicator for firm-years in which a firm both reduced its average prices (computed using GMID data) and experienced a positive market share change.
We then reestimated our market share–profit models from Table 3 with the addition of the new niche firm measure and buying share dummy indicator, along with their respective interactions with market share. As Table 6 shows, model M1 shows that higher market share reduces profit for niche firms (−.115, p < .05). As we expected, M2c reveals that this is a result of a strong negative effect of market share via perceived quality (−.062, p < .001). M1 also shows that the effect of market share on firm profit is significantly lower for firms "buying" market share (−.036, p < .001).[14] The mechanism results indicate that this is caused by a significant reversal in both the market power (M2a: −.047, p < .001) and firm efficiency (M2b: −.033, p < .001) effects of market share and a reduction of the perceived quality mechanism to insignificance (M2c: −.022, p > .1). These findings suggest that any supplier input cost benefits of greater market power from market share are more than offset by lowering downstream prices to "buy" the market share. In addition, consistent with the well-known "bullwhip" effect, rapid increases in short-term demand resulting from lowering price seems to disrupt the efficient production and delivery of these firms' products and services. Overall, the Table 6 results provide support for the robustness of the three mechanism variables in mediating the relationship between firm market share and profit, even in the relatively rare conditions under which the relationship is negative.
Graph
Table 6. Moderating Effect and Mechanism When We Include Conditions in Which Market Share May Have a Negative Effect on Profit.
| Model Specifications and Dependent Variables |
|---|
| M1 | M2a | M2b | M2c | M2d |
|---|
| Profit(t + 1) | Power(t + 1) | Efficiency(t + 1) | Quality(t + 1) | Profit(t + 1) |
|---|
| Direct Effect | | | | | |
| Market Share(t) | .058*** | .091*** | .034 | .108*** | .033 |
| Indirect Effect | | | | | |
| Market Power(t) | | | | | .218*** |
| Firm Efficiency(t) | | | | | .095*** |
| Perceived Quality(t) | | | | | .179*** |
| Moderators | | | | | |
| Switching Costs(t) | .149* | .118 | .021 | .081*** | .041 |
| Services Dummy(t) | −.042* | .088 | .510*** | .027 | −.010 |
| Firm Age(t) | .193 | .037 | −.036 | .008 | .021 |
| Niche Focus Firms(t) | .078*** | −.025*** | .027 | .119** | .180* |
| Buying Share Dummy(t) | .016 | .024 | −.032* | −.009 | −.026 |
| Prior Moderator Effects | | | | | |
| Share(t) × Switching Costs(t) | .050* | .162* | .022 | .200* | .037 |
| Share(t) × Services Dummy(t) | .063*** | −.011 | .166*** | .018 | .019 |
| Share(t) × Firm Age(t) | −.053*** | −.055 | −.113*** | −.078 | −.009 |
| Proposed Negative Moderators | | | | | |
| Share × Niche Focus Firms(t) | −.115** | −.016 | −.001 | −.062*** | −.010 |
| Share × Buying Share Dummy(t) | −.036*** | −.047*** | −.033*** | −.022 | −.036 |
| Controls | | | | | |
| Firm Size(t) | .490*** | .035*** | .086* | .039*** | .049*** |
| ADV(t) | .233*** | .008 | .010 | .018 | .041* |
| R&D(t) | .241*** | .031*** | .015 | .055*** | .059*** |
| Market Growth(t) | .022*** | .023 | .018 | .015 | .012 |
| Specification Tests | | | | | |
| −Log-likelihood | | 3,104.92 | | | |
| R2 | .55 | .30 | .39 | .20 | .72 |
- 18 *p < .05.
- 19 **p < .01.
- 20 ***p < .001.
- 21 Notes: 2,629 firm-year observations covering 207 firms for the 2000–2013 period (sample size due to ACSI data availability). For Niche Firms, indirect effect = 58%, of which Power = 21%; Efficiency = 0%; and Quality = 79%. For Firms Buying Share, Indirect Effect = 33%, of which Power = 56%, Efficiency = 22%, and Quality = 22%.
Having provided robust evidence to support the three mechanisms, to offer additional insight on the utility of the direct measures of the three mechanisms employed, we also examined how the results compare with previous indirect inferences regarding these mechanisms drawn from observable moderators of the market share–profit relationship. To accomplish this, we first replicated E-H's (2018) measures as well as main effect and substantive moderator results (banking services, concentration, and B2C). We then examined the mechanisms explaining the effect of these moderators of the market share–profit relationship in our sample, and the results are revealing (Web Appendix 12). For example, we find that while E-H's theorizing focuses on quality signaling, the reason for the stronger market share–profit relationship in B2C industries is a significant strengthening of all three mechanisms relative to business-to-business (B2B) industries (market power:.143, p < .001; efficiency:.044, p < .05; quality:.082, p < .05). In addition, we find that while banks are in general more profitable (.426, p < .01) and have greater market power (.042, p < .05), this is in spite of—not due to—their market share (−.087, p < .05). In fact, results reveal that market share reduces banks' profitability by lowering their efficiency (−.410, p < .001). We also find a direct moderating effect for concentration (.109, p < .05), whereas E-H found a nonlinear effect, and we observe that this is via increasing the market power benefit of market share (.110, p < .01). These results show that using moderators to indirectly infer the three mechanisms underlying the market share–profit relationship often does not do a good job of isolating these mechanisms. This reinforces the value of direct empirical understanding of the mechanisms linking market share with firm profit in predicting when market share is more valuable and thus when managers should set market share goals.
The new empirical understanding of the mechanisms linking market share with firm profit revealed in our analyses can help managers evaluate when market share may be a valuable goal. When its value is indicated, a manager's next task is to decide how to measure market share for goal setting and performance monitoring. To provide insights on this question, we examined two key market share measure design choices facing managers. First, "share of what?," in terms of unit sales volume or sales revenue, should be used in computing market share ([ 6]). Managers use both types of indicators to track market share, and both rank among the most popular measures of marketing performance in practice (e.g., https://marketbusinessnews.com/financial-glossary/market-share/). The second is "relative to what?," in terms of whether and how the firm's market share is benchmarked—as an absolute value (% of total market sales) or relative to others in the market (the market share leader or the top three players).
To provide insights on the first question, we replicated model M3 in Table 2 and replaced the sales revenue market share with unit sales volume market share using the same dynamic market definition. As we show in Table 7, in contrast to revenue market share (M2:.151, p < .05), unit market share (M1:.009, p > .1) does not predict firm profit. This result is robust to all of the same checks performed on our revenue market share main effect testing analyses and also to using benchmarked (vs. absolute) values of unit market share. Post hoc analysis of the mechanisms associated with unit share (Web Appendix 13) reveal that although it has a small positive effect on both market power and firm efficiency (consistent with the learning effect logic that market share is a proxy for number of units produced), this is insufficient to overcome the significant negative relationship with quality signaling. We reason that the weaker effect of unit (vs. revenue) market share on market power is a result of unit market share ignoring prices charged to customers (a downstream indicator of market power). The negative quality-signaling effect of unit market share is consistent with both ignoring price (which is often a quality cue for customers) and the notion that ubiquity reduces perceived exclusivity (e.g., [34]). These results show that when the presence of the three mechanisms indicates market share's value, managers should set market share goals and monitor performance in terms of revenue market share.
Graph
Table 7. Market Share–Profit Relationship Using Alternative Market Share Measures and Benchmarks.
| Market Share Measure, Model, Benchmark, and Dependent Variable |
|---|
| Unit Market Share | Revenue Market Share | Revenue Market Share | Revenue Market Share |
|---|
| M1 | M2 | M3 | M4 |
|---|
| Independent Variables | Absolute | Absolute | Relative to Market Leader | Relative to Top 3 |
|---|
| Profit(t + 1) | Profit(t + 1) | Profit(t + 1) | Profit(t + 1) |
|---|
| Main Effect | | | | |
| Market Share(t) | .009 | .151* | .222*** | .392*** |
| Controls | | | | |
| Firm Size(t) | .201*** | .270*** | .213*** | .243*** |
| Advertising(t) | .081** | .121*** | .121*** | .123*** |
| R&D(t) | .033 | .024 | .033 | .030 |
| Market Growth(t) | .001 | .001 | .004 | .006 |
| Specification Tests | | | | |
| Wald χ2 | 115.23 | 188.91 | 210.81 | 167.81 |
| R2 | .18 | .59 | .52 | .52 |
- 22 *p < .05.
- 23 **p < .01.
- 24 ***p < .001.
- 25 Notes: 3,058 firm-year observations covering 244 firms for the 2000–2013 period, except for model specification M1, which is estimated using 2,214 firm-year observations covering 235 firms for the period 2004–2013 (due to GMID data availability). In a subsequent robustness check, model specifications M2 through M4 were reestimated using the same 2,214 firm-year observations, and estimates remain identical.
In terms of the "relative to what?" question, in Table 7 we compared the market share–profit estimates of the absolute value of market share used in the main effect testing (M2) and two different relative market share benchmark operationalizations: relative to the market share leader (M3) and relative to the combined market share of the top three market share firms (M4).[15] The results indicate that benchmarked measures of firm market share provide stronger predictive power (of future profit) (M3:.222, p < .001; M4:.392, p < .001, respectively) than using absolute market share (M2:.151, p < .05). Subsequent analysis of the three mechanisms show that this is a result of the relative market share measures "dialing up" the market share–market power link (Web Appendix 14). This is likely due to such "relative to others in the same industry" measures capturing some of the industry-level market concentration power that our previous analysis showed increased the market share–market power relationship in terms of both switching costs (which are higher when markets have fewer equivalent players) and average market share (as an indicator of market concentration in the E-H [2018] replication analyses).
This study offers several new insights into theories of firm behavior and performance. First, economic theory assumes that market share predicts firm profit but offers different reasons for why this relationship exists. We provide the first simultaneous test of three mechanisms proffered in competing economic theories for this relationship and show that in combination, they explain the vast majority of the variance in the market share–profit relationship. This suggests that individual single-theory lens explanations of the mechanisms linking market share with profit are incomplete, and all three mechanisms can provide higher (or lower) explanatory power under different conditions. While, on average, market power provides the highest level, and firm efficiency the lowest level, of explanatory power, we also identify conditions under which the reverse is true (e.g., for young firms). Thus, none of the three theories from which the hypothesized mechanisms arise is "correct" or "incorrect," but market power and quality signaling generally explain more of the variance in the market share–profit relationship across firms and industries.
Second, our results offer new insights into efficiency-enhancing experience-based "learning effects" identified in strategic management theorizing ([ 2]). Management scholars have used this logic to explain why market share (a proxy for the number of times a firm may have produced a value offering) may be positively related to firm profit (e.g., [29]). We find that while firm efficiency is valuable (predicts profit), on average it is explained mainly by a firm's size rather than its market share. This suggests that for most firms, scale economies are more important in driving profit than economies of learning. However, for young firms, we find that market share delivers significant efficiency benefits above and beyond those associated with size, and we also find significant efficiency benefits from market share among service businesses. This suggests that "learning by doing" effects occur where organizational routines are less set and when firms can use experience gained to update and change their processes with lower investments.
Third, we find support for information economics theorizing on the value of signals of unobservable firm quality. While prior research has explored market share's role in consumer evaluations of quality ([34]), we provide the first empirical evidence that market share generally signals firm quality and thereby increases firm profit. The negative effects on perceived quality we observe when using unit (vs. revenue) market share also suggest that price combines with market share in signaling quality to customers. In addition, we find that market share's positive quality-signal effect depends on previously unidentified industry and firm conditions (stronger for younger firms, in B2C markets, and for those with switching costs).
For researchers, our study also has broader implications. Not least, it clearly shows the effect that sampling can have on the findings and inferences drawn in firm-level empirical research. We find wide variance in both the main market share–profit relationship and in the specific mechanisms accounting for the relationship across industries. Thus, samples made up of a single industry, or an industry dominated by certain types of firms, would lead to very different results and widely varying inferences being drawn as to which theory may be supported in empirical tests. This is unlikely to be unique to the market share phenomenon we examine. In addition, our study also reveals the desirability of directly observing (or at least finding direct indicators of) mechanisms believed to underlie relationships of interest. In particular, our results highlight the need for researchers to be careful about using indirect contingencies to infer such unobserved mechanisms when there may be more than one mechanism involved.
This study also provides new insights for managers regarding how market share should be measured. Although unit (volume) market share is widely used in practice to set marketing goals and monitor performance (e.g., auto and motorcycle manufacturers, many consumer packaged goods companies), our results reveal that it is not predictive of firm profit, whereas revenue (value) market share is. We also find that in terms of predicting profit, relative (to others) measures of revenue market share can be superior to absolute measures. Post hoc analyses suggest that such relative measures can enhance the market power value of the observed market share, and that benchmarking a firm's market share relative to the top three market share firms versus the market share leader offers a stronger predictor of future profit. This is aligned with the intuition that benchmarking against others provides an indicator of both the firm's market share and the concentration present in the marketplace, which we show interact significantly in predicting firm performance.
To provide finer-grained managerial insights, we also examined ( 1) which measures of market share were the strongest predictors of future profit for different types of firms to help managers select the most appropriate market share metrics for goal setting and performance monitoring and ( 2) the average profit value of a 1% increase in the average firms' market share for different types of businesses to give managers a calibration of the dollar-value benefits that may be expected when evaluating costs associated with share building strategies. Given our sample size, we are somewhat limited in how fine-grained we can be in these analyses without running into power issues. We therefore split our sample in a managerially meaningful way by identifying firms on the basis of whether they serve primarily consumer or business customers and whether their value offerings are mainly product- versus service-based. As shown in Table 8, the results vary across the four cells, with B2C product firm and B2B service firm profit being most strongly predicted by absolute revenue market share, whereas for B2C service and B2B product firms, it is revenue share relative to the top three market share players. The one-year profit increases associated with a 1% improvement in the average firm's market share vary across the four cells from a low of just over $1 million to almost $6 million. These findings have clear and important implications for managers setting market share goals and monitoring market share performance in their firms and offer a useful dollar benefit scale calibration for managers with respect to the potential payoffs they may expect from investments in market share–building strategies.
Graph
Table 8. Managerial Matrix: Metrics.
| | Products | Services |
|---|
| B2C | Strongest market share–profit predictor | Absolute revenue share | Relative to top three revenue share |
| Mean firm market share | 6.80% | 7.19% |
| Profit value of 1% increase in mean market share | From 6.80% to 6.87%:.121% (p < .001) × $840 million = $1.02 million | From 7.19% to 7.26%:.704% (p < .001) × $840 million = $5.9 million |
| Observations | 1,910 firm/year observations (136 firms) | 484 firm/year observations (52 firms) |
| B2B | Strongest market share–profit predictor | Relative to top 3 revenue share | Absolute revenue share |
| Mean firm market share | 6.68% | 7.31% |
| Profit value of 1% increase in mean market share | From 6.68% to 6.75%:.309% (p < .001) × $840 million = $2.6 million | From 7.31% to 7.38%:.146% (p < .01) × $840 million = $1.2 million |
| Observations | 322 firm/year observations (32 firms) | 342 firm/year observations (24 firms) |
26 Notes: Unit share is not predictive of firm profit in any one of the four cells. Reported elasticities estimated via a model specification equivalent to M3 in Table 2, with the noted strongest market share predictor measure as a regressor and using the observations specific to each of the Product/Services and B2C/B2B cells. Profit increase $ values are for a 1% increase in the mean firm's market share in each cell (e.g., 7.310% to 7.383%) not an increase of 1 point of total market share (e.g., from 7.310% to 8.310%). Because we estimate log-log models, the estimated coefficients in each condition can be interpreted as market share–profit elasticities (%) which can be converted to a dollar profit value by multiplying them by the mean profit in our sample (i.e., $840 million).
In terms of where managers would be advised to pursue market share to a greater or lesser degree, our results provide several new insights (Table 9). For younger firms and for nonbanking services firms, it may make sense to set market share goals and monitor performance. It may also be more beneficial for firms operating in marketplaces with high levels of quality uncertainty and those with higher switching costs. However, it may make less sense for banks and firms in industries in which pricing power is low and/or quality is relatively certain. Older firms may also find market share to be of less value as a marketing goal and performance metric. Firms pursuing a niche strategy would be well advised to either ignore market share or ensure that they assess it only within their selected niche market definition. Finally, we show that, while relatively rare, "buying share" is not a profitable move.
Graph
Table 9. Managerial Matrix: Contingency Effects on Share-Profit Mechanisms.
| Relative Mechanism Importance |
|---|
| Contingency | Market Power | Firm Efficiency | Perceived Quality |
|---|
| Switching costs (high) | + | n.s. | + |
| Service (vs. product) | n.s. | + | n.s. |
| Firm age (older) | n.s. | − | − |
| Concentration (more) | + | n.s. | n.s. |
| B2C (vs. B2B) | + | + | + |
| Banking (vs. others) | n.s. | − | n.s. |
27 Notes: n.s. = not significant. This table summarizes analyses reported in Table 4 and Web Appendix 12, with mechanism importance indicated relative to the average displayed by all firms in our sample.
For policy makers, this study provides new insights with respect to when market share may lead to market power and potential abuse that requires regulation. Importantly, our results show that firm profits from market share result from quality signaling and learning-based efficiencies as well as market power. Thus, policy makers need to be careful not to directly equate market share and market power; we show that while they are often related, they are far from synonymous. Rather, our results suggest that regulatory authorities can be less concerned by a firm's market share in marketplaces where customer quality uncertainty is significant and where efficiency-enhancing learning benefits from market share may exist (e.g., young firms, service firms). In such conditions, market share could enhance rather than harm consumer welfare by reducing consumer–firm information asymmetry and potentially lowering costs.
This study has some limitations that should be taken into account when considering the findings. First, because we require public data to explore our research questions, our sample is naturally skewed toward larger firms. While we include small, nonpublic firms in the definition of the total market sales used in constructing the robustness check NAICS measure of market share, we are unable to include such firms' individual market shares in the hypothesis testing because these firms' sales data are private. Although we have a wide range of market shares in our sample (with a low of less than 1%, a high of 77%, and a mean of less than 7%), and no evidence of range restriction in our key variables, researchers with access to private firm data could test the generalizability of this study's findings to firms with much smaller market shares.
Second, our data are focused on firms with U.S. listings. However, including studies covering broader geographies and longer time period data, E-H (2018) suggest that the market share–profit relationship is weaker in recent times in Western Europe than the United States, so future research in other regions is required to examine how the mechanisms we identify may differ across geographies. Third, our study examines market share at a firm level. However, market shares may also be computed at other levels (e.g., brand or geographic market level). A post hoc analysis of monobrand firms in our sample suggests that the same market share–profit main effect and mechanism relationships hold (Web Appendix 15); however, research is required to confirm this.
Our study also reveals several new avenues for theoretically interesting and managerially relevant research. First, we find that the vast majority of market share's effect on profit is mediated through its effects on firm market power, perceived quality, and efficiency. This suggests that new theorizing regarding why market share is valuable may be of limited value. However, in light of our findings, new research on the details of how each of the three mechanisms works is clearly required. For example, what is the relative effect of market power on upstream versus downstream parties, and how much is contributed by cost reductions versus pricing versus coordination benefits? Similarly, what types and levels of quality uncertainty create conditions that lead to market share's value in signaling quality? How much of market share's signal value is to upstream versus downstream parties?
Second, this study reveals market power, quality signaling, and operating efficiency as the mechanisms linking market share with firm profit. Because market share is a market-based outcome of firms' marketing efforts, this raises the interesting possibility that these three mechanisms may also mediate the relationship between other marketing-related phenomena and firm performance. For example, are market-based assets such as brand equity and customer relationships also linked to firm profit via the same three mechanisms? Are there also other mechanisms that may be available to such market-based assets but not to market share?
Third, given that market share is more or less valuable under different market and firm conditions—and that buying share is both rare and ineffective—does it also matter how firms create and leverage market share? For example, are market shares more or less valuable to firms pursuing low-cost business strategies versus those pursuing differentiated advantages? Are the three mechanisms linking market share and profit the same for these different strategies, or are some mechanisms more important to one strategy than another? Addressing these questions would provide important new insights for both managers and researchers.
Graph
Appendix: Measure Details
| Variables | Measurement Details | Data Source/Literature |
|---|
| Firm Profit | Net income of the firm (Item NI). | Compustat |
| Market Share (Revenue) | Percentage of an industry or market's total sales garnered by a particular firm over a specified time period. Markets are defined through text analysis of similarity between product-market descriptions within 10-Ks. Sales for each firm obtained from Compustat. | SEC, CompustatHoberg and Phillips (2010) |
| Market Share (Units) | Units sold by each firm were obtained directly using the GMID (Euromonitor) database. Market definition for firms with unit share data calculated as for revenue share. | GMID |
| Market Power (Power) | Operationalized based on a profit elasticity measure following Boone (2008), estimated by regressing (at the industry level) firms' profit (Item NI) on their total costs (Items COGS and XSGA). Firm-specific residuals are used to calibrate each firm's margins relative to industry average, providing a firm-level indicator of market power. | CompustatBoone (2008) |
| Firm Efficiency (Efficiency) | Concerns producing goods and services with the optimal combination of inputs to produce maximum output at the minimum cost. We use a stochastic frontier estimation approach with operating expense (Item XOPR) as the input and total sales (Item SALE) as the output. | CompustatBauer, Berger, and Humphrey (1993) |
| Perceived Quality (Quality) | Measured using customer perceived quality ratings of the firm's brand(s) from Equitrend database. | EquitrendMorgan and Rego (2009) |
| Switching Costs | These are perceived costs associated by the firm's customers with moving to an alternative supplier. We calibrate these costs as the degree to which customers exhibit loyalty to a firm that cannot be explained by the level of satisfaction delivered by the firm's offerings. Using ACSI data, we estimate customer-level loyalty as a latent factor comprising variables capturing customers' repurchase intentions and price sensitivity. Satisfaction is the ACSI measure detailed previously. We estimate switching costs for each firm/year as the residual of regressing each firm's customers' loyalty onto its customers' satisfaction, controlling for industry and time.Loyalty(it) = β0 + β1 × Satisfaction(it) + ID(it) + YD(it) + ε(it), where ID(it) are industry and YD(it) year dummies. ε(it) is the residual of this regression and is used as our estimate of switching costs, which are firm- and year-specific. | ACSI (firm/year-level aggregation of individual-level respondent survey response data).Rego, Morgan, and Fornell (2013) |
| Niche-Focused Strategy (Niche) | Text analysis employing a new dictionary utilizing an inductive word search with exemplar niche firms. The analysis is then performed using a bag-of-words approach where each firm gets a score corresponding to the ratio of niche-related words and total words in each firm 10-K. To ensure that we were isolating the types of niche firms where market share was expected to be negatively associated with profit, suggested in the theorizing (i.e., those pursuing a single niche in a market vs. those targeting several different segments with different offerings), we then identified mono- versus multibrand firms by multiplying the niche-focus score for each firm by the dummy variable (1 for monobrand firms, 0 for multibrand firms). | New measure |
| Service-Dominant Markets (Services) | Dummy variable identifying service firms/ industries using Fama–French NAICS industries. | Fama and French (2008) |
| Firm Age | Number of years of operation of the firm since incorporation, obtained from the firm's annual reports and websites. | |
| Industry Concentration | Industry-level average market share. | Edeling and Himme (2018) |
| B2C versus B2B Firms | Dummy variable capturing whether the firm caters mainly to business customers. Each firm was coded manually by three coders who used information on categorization from secondary sources such as Hoover's. Reliability was >85%. | |
| Services (Banking) | Dummy variable capturing whether a firm belongs to the banking sector (SIC Code 602). | Compustat |
| Competitor Orientation | Text analysis of 10-K reports following dictionaries on competitor orientation (as a part of Market Orientation) developed in prior literature (Zachary et al. 2011). | SECZachary et al. (2011) |
| Controls | |
| Firm Size | The firm's reported total assets (Item AT). | Compustat |
| Market Growth Annual | change in cumulative industry sales (Item SALE). | Compustat |
| R&D Expense | Firm's reported expenditures on Research and Development (Item XRD). | Compustat |
| Advertising Expense | Firm's reported expenditures on Advertising (Item XAD) | Compustat |
| Robustness Check Variables | |
| ROA | The ratio of current year income before extraordinary items (Item IB) to the firm's previous year total assets (Item AT). | Compustat |
| Tobin's q | Ratio of the firm's market value to the replacement cost of physical and intangible capital of the firmWe measure the firm's market value as the market value of outstanding equity (Items PRCC_F × CSHO), plus the book value of debt (Items DLTT + DLC), minus the firm's current assets (Item ACT). The firm's replacement cost of physical capital is measured as the book value of property, plant, and equipment (Item PPEGT). Intangible capital is estimated as the sum of the firm's knowledge capital (the capitalized value of firm R&D expenditures) and organizational capital (a fraction of the capitalized value of firm SGA expenditures) following Peters and Taylor (2017). | Peters and Taylor (2017) |
| Alternate Market Power | Operationalized based on Lerner Index as profit margin relative to price. Average variable costs are used as a proxy for marginal costs, operationalized using total variable costs divided by sales (Items XOPR and SALE). Average price was estimated dividing sales revenues (Item SALE) by unit sales (obtained from GMID database). | GMID, CompustatBoone (2008) |
| Market Share Focus | Based on text analysis of 10-K reports, estimated as the ratio of the number of times "market share" is reported relative to the total number of words in the annual 10-K report. | New measure |
| Perceived Quality | Measured via average annual perceived quality ratings of the firm's brand(s) from the Brand Asset Valuator database. | Brand Asset Valuator Mizik and Jacobson (2005) |
| Perceived Quality | Measured using average annual firm quality ratings from Fortune's World's Most Admired Companies database. | AMACCretu and Brodie (2007) |
28 Notes: SEC = Securities & Exchange Commission; SGA = selling and general administrative.
Footnotes 1 Debanjan Mitra
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Abhi Bhattacharya https://orcid.org/0000-0002-6921-7430
5 Online supplement:https://doi.org/10.1177/00222429211031922
6 Sales revenue is used in calculating revenue market share but is conceptually and arithmetically distinct from it. The correlation between revenue market share and sales revenue in our data is.14. Nonetheless, we assess how this may affect our hypothesis testing results in a robustness check using revenue market share ranks and controlling for revenue.
7 Although firms may sell both types of offerings, for brevity we use the simple terms "services" and "products" to denote which type of value offering is the primary focus of the firm.
8 The main effect and mediation hypothesis testing results reported are robust to using only this smallest (n = 2,629) "core" data set used in testing H4–H6 firms (see Web Appendices 16 and 17). A list of all firms contained in the full hypothesis-testing data set is provided in Web Appendix 20.
9 All variables calculated using industry-level data in our hypothesis testing use the same industry definitions.
As we show in Web Appendices 18 and 19, our analyses are robust to using alternative firm quality indicators from Fortune's "World's Most Admired Companies" database and Young & Rubicam's Brand Asset Valuator for the sample subsets where these data were available.
We applied a log (x + 1) transformation to all variables; for variables that include negative values (e.g., profit), we transformed these via −log(|x| + 1) to preserve rank (e.g., [26]).
Even after our log transformation, the nonnormal distribution of the market share variable still meets the requirements for the use of a copula approach (Shapiro–Wilk test (Z = 7.217, V = 16.888, p > z = .00).
E-H (2018) find a marginally (p < .10) stronger effect of market share on performance in manufacturing industries, which is inconsistent with our findings. However, 92% of the service firms in their sample are banks, and using only banks and simultaneous cross-sectional analyses as they do, we reproduce E-H's results. Thus, differences in banks' accounting and financial reporting appear to affect the observed economic impact of market share in ways not true of other service firms.
We also found this to be true for contemporaneous profit in post hoc tests. Such negative effects may be well-known in practice, as buying market share does not seem to be common or a long-term strategy (we find fewer than 7% of firm-year observations where firms appear to be buying market share, and very few examples of these firms doing so in sequential periods).
Results from [62] indicate that most industries evolve to an equilibrium with three large market share firms.
References Amit Raphael. (1986), " Cost Leadership Strategy and Experience Curves ," Strategic Management Journal , 7 (3), 281 – 92.
Argote Linda. (2011), " Organizational Learning Research: Past, Present, and Future ," Management Learning , 42 (4), 439 – 46.
Armstrong J. Scott , Collopy Fred. (1996), " Competitor Orientation: Effects of Objectives and Information on Managerial Decisions and Profitability ," Journal of Marketing Research , 33 (2), 188 – 99.
Bass Frank M. , Cattin Philippe , Wittink Dick R.. (1978), " Firm Effects and Industry Effects in the Analysis of Market Structure and Profitability ," Journal of Marketing Research , 15 (1), 3 – 10.
Bauer Paul W. , Berger Allen N. , Humphrey David B.. (1993), " Efficiency and Productivity Growth in US Banking, " in The Measurement of Productive Efficiency: Techniques and Applications , Fried H.O. , Lovell C.A.K. , Schmidt S.S. , eds. Oxford, UK : Oxford University Press.
Bendle Neil T. , Bagga Charan K.. (2016), " Metrics that Marketers Muddle ," MIT Sloan Management Review , 57 (3), 73 – 82.
Bhattacharya Abhi , Misra Shekhar , Sardashti Hanieh. (2019), " Strategic Orientation and Firm Risk ," International Journal of Research in Marketing , 36 (4), 509 – 27.
Boone Jan. (2008), " A New Way to Measure Competition ," Economic Journal , 118 (531), 1245 – 61.
Boulding William , Staelin Richard. (1990), " Environment, Market Share, and Market Power ," Management Science , 36 (10), 1160 – 77.
Bresnahan Timothy F. (1989), " Empirical Studies of Industries with Market Power ," Handbook of Industrial Organization , 2 , 1011 – 57.
Buzzell Robert D. , Gale Bradley T. , Sultan Ralph G.M.. (1975), " Market Share: A Key to Profitability ," Harvard Business Review , 53 (1), 97 – 106.
Carman James M. (1990), " Consumer Perceptions of Service Quality: An Assessment of the SERVQUAL Dimensions ," Journal of Retailing , 66 (1), 33 – 55.
Chaudhuri Arjun , Holbrook Morris B.. (2001), " The Chain of Effects from Brand Trust and Brand Affect to Brand Performance: The Role of Brand Loyalty ," Journal of Marketing , 65 (2), 81 – 93.
Conner Katherine R. (1991), " A Historical Comparison of Resource-Based Theory and Five Schools of Thought Within Industrial Organization Economics: Do We Have a New Theory of the Firm? " Journal of Management , 17 (1), 121 – 54.
Cretu Anca E. , Brodie Roderick J.. (2007), " The Influence of Brand Image and Company Reputation Where Manufacturers Market to Small Firms: A Customer Value Perspective ," Industrial Marketing Management , 36 (2), 230 – 40.
Dabholkar Pratibha A. , Johnston Wesley J. , Cathey Amy S.. (1994), " The Dynamics of Long-Term Business-to-Business Exchange Relationships ," Journal of the Academy of Marketing Science , 22 (2), 130 – 45.
Demsetz Harold. (1974), Toward a Theory of Property Rights. Basingstoke, UK : Palgrave Macmillan.
DiMaggio Paul , Louch Hugh. (1998), " Socially Embedded Consumer Transactions: For What Kinds of Purchases Do People Most Often Use Networks? " American Sociological Review , 63 (5), 619 – 37.
Echols Ann , Tsai Wenpin. (2005), " Niche and Performance: The Moderating Role of Network Embeddedness ," Strategic Management Journal , 26 (3), 219 – 38.
Edeling Alexander , Himme Alexander. (2018), " When Does Market Share Matter? New Empirical Generalizations from a Meta-Analysis of the Market Share–Performance Relationship ," Journal of Marketing , 82 (3), 1 – 24.
Erdem Tülin , Swait Joffre. (2004), " Brand Credibility, Brand Consideration, and Choice ," Journal of Consumer Research , 31 (1), 191 – 8.
Fama Eugene F. , French Kenneth R.. (2008), " Dissecting Anomalies ," Journal of Finance , 63 (4), 1653 – 78.
Farrell Joseph , Klemperer Paul. (2007), " Coordination and Lock-In: Competition with Switching Costs and Network Effects, " in Handbook of Industrial Organization, Vol. 3 , Armstrong M. , Porter R. , eds. Amsterdam : North Holland Publishing.
Farris Paul W. , Bendle Neil T. , Pfeifer Phillip E. , Reibstein David J.. (2010), Marketing Metrics: The Definitive Guide to Measuring Marketing Performance. Upper Saddle River, NJ : Pearson Education.
Ferrier Walter J. , Smith Ken G. , Grimm Curtis M.. (1999), " The Role of Competitive Action in Market Share Erosion and Industry Dethronement: A Study of Industry Leaders and Challengers ," Academy of Management Journal , 42 (4), 372 – 88.
Galizzi Monica , Zagorsky Jay L.. (2009), " How Do On-the-Job Injuries and Illnesses Impact Wealth? " Labour Economics , 16 (1), 26 – 36.
Glazer Rashi. (1991), " Marketing in an Information-Intensive Environment: Strategic Implications of Knowledge as an Asset ," Journal of Marketing , 55 (4), 1 – 19.
Gooner Richard A. , Morgan Neil A. , Perreault William D.. (2011), " Is Retail Category Management Worth the Effort (and Does a Category Captain Help or Hinder)? " Journal of Marketing , 75 (5), 18 – 33.
Haleblian Jerayr , Kim Ji-Yub , Rajagopalan Nandini. (2006), " The Influence of Acquisition Experience and Performance on Acquisition Behavior: Evidence from the U.S. Commercial Banking Industry ," Academy of Management Journal , 49 (2), 357 – 70.
Hall Graham , Howell Sydney. (1985), " The Experience Curve from the Economist's Perspective ," Strategic Management Journal , 6 (3), 197 – 221.
Hambrick Donald C. (1983), " High Profit Strategies in Mature Capital Goods Industries: A Contingency Approach ," Academy of Management Journal , 26 (4), 687 – 707.
Hamlin Robert , Henry James , Cuthbert Ron. (2012), " Acquiring Market Flexibility via Niche Portfolios: The Case of Fisher & Paykel Appliance Holdings Ltd.," European Journal of Marketing , 46 (10), 1302 – 19.
Hanssens Dominique M., (2015) Empirical Generalizations About Marketing Impact. Cambridge, MA : Marketing Science Institute.
Hellofs Linda L. , Jacobson Robert. (1999), " Market Share and Customers' Perceptions of Quality: When Can Firms Grow Their Way to Higher Versus Lower Quality? " Journal of Marketing , 63 (1), 16 – 25.
Hoberg Gerard , Phillips Gordon. (2010), " Product Market Synergies and Competition in Mergers and Acquisitions: A Text-Based Analysis ," Review of Financial Studies , 23 (10), 3773 – 3811.
Jacobson Robert. (1988), " Distinguishing Among Competing Theories of the Market Share Effect ," Journal of Marketing , 52 (4), 68 – 80.
Jacobson Robert , Aaker David A.. (1985), " Is Market Share All That It's Cracked Up to Be? " Journal of Marketing , 49 (3), 11 – 22.
Jin Ginger Zhe , Leslie Phillip. (2003), " The Effect of Information on Product Quality: Evidence from Restaurant Hygiene Grade Cards ," Quarterly Journal of Economics , 118 (2), 409 – 51.
Kasman Adnan , Oscar Carvallo , (2014), " Financial Stability, Competition and Efficiency in Latin American and Caribbean Banking ," Journal of Applied Economics , 17 (2), 301 – 24.
Katsikeas Constantine S. , Morgan Neil A. , Leonidou Leonidas C. , Tomas M G.. Hult (2016), " Assessing Performance Outcomes in Marketing ," Journal of Marketing , 80 (2), 1 – 20.
Kirmani Amna , Rao Akshay R.. (2000), " No Pain, No Gain: A Critical Review of the Literature on Signaling Unobservable Product Quality ," Journal of Marketing , 64 (2), 66 – 79.
Knight John G. , Holdsworth David K. , Mather Damien W.. (2007), " Country-of-Origin and Choice of Food Imports: An In-Depth Study of European Distribution Channel Gatekeepers ," Journal of International Business Studies , 38 (1), 107 – 25.
Maciel Andre F. , Fischer Eileen. (2020), " Collaborative Market Driving: How Peer Firms Can Develop Markets Through Collective Action ," Journal of Marketing , 84 (5), 41 – 59.
Massey Patrick. (2000), " Market Definition and Market Power in Competition Analysis: Some Practical Issues ," Economic and Social Review , 31 (4), 309 – 28.
Mizik Natalie , Jacobson Robert. (2005), " How Brand Attributes Drive Financial Performance ," MSI Reports , 3 , 21 – 39.
Morgan Neil A. , Rego Lopo L.. (2009), " Brand Portfolio Strategy and Firm Performance ," Journal of Marketing , 73 (1), 59 – 74.
Parker Jeffrey R. , Lehmann Donald R. , Xie Yi. (2016), " Decision Comfort ," Journal of Consumer Research , 43 (1), 113 – 33.
Peters Ryan H. , Taylor Lucian A.. (2017), " Intangible Capital and the Investment- q Relation ," Journal of Financial Economics , 123 (2), 251 – 72.
Porter Michael E. (1996), "What Is Strategy?" Harvard Business Review (November/December), 2–19.
Posner Richard A. (1979), " The Chicago School of Antitrust Analysis ," University of Pennsylvania Law Review , 127 (4), 925 – 48.
Preacher Kristopher J. , Rucker Derek D. , Hayes Andrew F.. (2007), " Addressing Moderated Mediation Hypotheses: Theory, Methods, and Prescriptions ," Multivariate Behavioral Research , 42 (1), 185 – 227.
Rao Akshay R. , Monroe Kent B.. (1989), " The Effect of Price, Brand Name, and Store Name on Buyers' Perceptions of Product Quality: An Integrative Review ," Journal of Marketing Research , 26 (3), 351 – 57.
Rego Lopo L. , Morgan Neil A. , Fornell Claes. (2013), " Reexamining the Market Share–Customer Satisfaction Relationship ," Journal of Marketing , 77 (5), 1 – 20.
Repenning Nelson P. , Sterman John D.. (2002), " Capability Traps and Self-Confirming Attribution Errors in the Dynamics of Process Improvement ," Administrative Science Quarterly , 47 (2), 265 – 95.
Richardson James. (1993), " Parallel Sourcing and Supplier Performance in the Japanese Automobile Industry ," Strategic Management Journal , 14 (5), 339 – 50.
Romanelli Elaine. (1989), " Environments and Strategies of Organization Start-Up: Effects on Early Survival ," Administrative Science Quarterly , 34 (3), 369 – 87.
Scherer Frederic M. , Ross David. (1990), Industrial Market Structure and Economic Performance , 3rd ed. Boston : Houghton-Mifflin.
Shi Mengtze. (2013), " A Theoretical Analysis of Endogenous and Exogenous Switching Costs ," Quantitative Marketing and Economics , 11 (2), 205 – 30.
Shy Oz. (1995). Industrial Organization: Theory and Applications. Boston : MIT Press.
Snyder Scott Andrew. (2008), " System and Method for Assisting Customers in Choosing a Bundled Set of Commodities using Customer Preferences," U.S. Patent 7,430,531.
Teas R. Kenneth , Agarwal Sanjeev. (2000), " The Effects of Extrinsic Product Cues on Consumers' Perceptions of Quality, Sacrifice, and Value ," Journal of the Academy of Marketing Science , 28 (2), 278 – 90.
Uslay Can , Ayca Altintig Z. , Winsor Robert D.. (2010), " An Empirical Examination of the 'Rule of Three': Strategy Implications for Top Management, Marketers, and Investors ," Journal of Marketing , 74 (2), 20 – 39.
Varadarajan Rajan. (2020), " Customer Information Resources Advantage, Marketing Strategy and Business Performance: A Market Resources Based View ," Industrial Marketing Management , 89 , 89 – 97.
Wetzel Hauke , Hattula Stefan , Hammerschmidt Maik , van Heerde Harald J.. (2018), " Building and Leveraging Brand Equity: Evidence from 50 Years of German Professional Soccer ," Journal of the Academy of Marketing Science , 46 (4), 591 – 61.
Wooldridge Jeffrey. (2015), Introductory Econometrics: A Modern Approach. Scarborough, ON : Nelson Education.
Yli-Renko Helena , Autio Erkko , Sapienza Harry J.. (2001), " Social Capital, Knowledge Acquisition, and Knowledge Exploitation in Young Technology-Based Firms ," Strategic Management Journal , 22 (6/7), 587 – 613.
Zachary Miles A. , McKenny Aaron F. , Short Jeremy C. , Davis Kelly M. , Wu Di. (2011), " Franchise Branding: An Organizational Identity Perspective ," Journal of the Academy of Marketing Science , 39 (4), 629 – 45.
Zeithaml Valerie A. , Bitner Mary J. , Gremler Dwayne D.. (1996), Services Marketing. New York : McGraw Hill.
~~~~~~~~
By Abhi Bhattacharya; Neil A. Morgan and Lopo L. Rego
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 52- Expression of Concern. Journal of Marketing. Sep2021, Vol. 85 Issue 5, p162-162. 1p. DOI: 10.1177/00222429211026311.
- Database:
- Business Source Complete
Expression of Concern
The American Marketing Association (AMA) issues an expression of concern related to Jayati Sinha and Rajesh Bagchi (2019), "Role of Ambient Temperature in Influencing Willingness to Pay in Auctions and Negotiations," Journal of Marketing, 83 (4), 121–138 at the suggestion of Editor in Chief Christine Moorman.
A third party contacted the journal with concerns that the data in this article may have been fabricated, based on the identification of unusual patterns in three studies reported in the article. After receipt of the concerns, the journal arranged for a review by a panel of independent scholars. The authors fully cooperated with the independent review by providing copies of the original completed questionnaires and associated spreadsheet files used in the analysis for studies 2a, 2b, and 2c – the three studies under question. The independent reviewers were unable to confirm the third-party's concerns about data fabrication.
However, in comparing the original questionnaires and Excel files, four anomalies were identified: (1) data appearing in the spreadsheet files when no values were present in the original surveys; (2) data in the spreadsheet files having different values from those that appear in the original surveys; (3) large outliers (outside the valid response range) whose values were changed in the entered data; and (4) entries in the original survey being scratched over with new results circled. Anomalies were found in all three studies, with Study 2b having the largest number.
The anomalies described in points (2) and (4) were deemed less serious as the anomalies described in point (2) could be attributed to random data entry errors, and the changed entries described in point (4) may happen in surveys and experiments.
Point (1), the entry of data in the spreadsheet files that was not present in the raw files, and point (3), the failure to identify the occurrence of outliers and the action of changing the outlier data to results more consistent with other reported data without disclosing these changes in the article, were deemed more serious infractions of data management.
The number of instances of (1) and (3) are very small relative to the number of data points checked in Studies 2a–2c. Upon independently reanalyzing the data, it was found that the results do not change when correcting for these anomalies. However, given the nature of the errors, an Expression of Concern was deemed prudent.
The first author, Jayati Sinha, had informed the journal that the data were entered exclusively by the first author.
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 53- Faculty Research Incentives and Business School Health: A New Perspective from and for Marketing. By: Stremersch, Stefan; Winer, Russell S.; Camacho, Nuno. Journal of Marketing. Sep2021, Vol. 85 Issue 5, p1-21. 21p. 3 Diagrams, 5 Charts, 1 Graph. DOI: 10.1177/00222429211001050.
- Database:
- Business Source Complete
Faculty Research Incentives and Business School Health: A New Perspective from and for Marketing
Grounded in sociological agency theory, the authors study the role of the faculty research incentive system in the academic research conducted at business schools and business school health. The authors surveyed 234 marketing professors and completed 22 interviews with 14 (associate) deans and 8 external institution stakeholders. They find that research quantity contributes to the research health of the school, but not to other aspects of business school health. The r-quality of research (i.e., rigor) contributes more strongly to the research health of the school than research quantity. The q-quality (i.e., practical importance) of research does not contribute to the research health of the school but does contribute positively to teaching health and several other dimensions of business school health. The authors conclude that faculty research incentives are misaligned: ( 1) when monitoring research faculty, the number of publications receives too much weight, while creativity, literacy, relevance, and awards receive too little weight; and ( 2) faculty feel that they are insufficiently compensated for their research, while (associate) deans feel they are compensated too much for their research. These incentive misalignments are largest in schools that perform the worst on research (r- and q-) quality. The authors explore how business schools and faculty can remedy these misalignments.
Keywords: business school; incentives; organizational health; practical importance; research; relevance; rigor; scientometrics
Business schools consider academic research by their faculty as one of the main pillars in their business model and allocate a large part of their resources to it (e.g., faculty time, labs, research budgets). At the same time, prior research across fields, including marketing, has heavily debated whether the academic research that business school professors conduct adds value to the business schools that employ them (see Table 1[ 5]).
Graph
Table 1. Selected Papers on the Role of Academic Research in Business Schools.
| Articlea | Field | Journal | No. of Citesb (Google) | Focusc | Conceptual (C) or Empirical (E) | Summary |
|---|
| Bettis (2012) | Strategy | Strategic Management Journal | 133 | Rigor of research | C | Data snooping may be prevalent among strategy researchers, even though evidence is lacking. This raises serious issues for the future of strategic management research. |
| Jaworski (2011) | Marketing | Journal of Marketing | 199 | Practical importance of research | C | Research faculty in marketing need a better understanding of managerial practice to conduct research that is relevant to marketing managers' needs. To do so, marketing scholars must spend more time with managers. |
| Kaplan (2011) | Accounting | Accounting Review | 294 | Practical importance of research | C | Accounting scholars have distanced themselves from the professional practice of accounting. Accounting scholars should devote more resources to fundamentally understand contemporary and future practice. |
| Lehmann, McAlister, and Staelin (2011) | Marketing | Journal of Marketing | 151 | Practical importance of research | E | The level of analytical rigor has risen steadily in academic marketing journals. While, ceteris paribus, rigor is desirable, other desirable characteristics, such as practical importance, communicability, and simplicity, have been downplayed. |
| Lilien (2011) | Marketing | Journal of Marketing | 167 | Adoption of research by practice | C | There is a large and widening academic–practitioner gap in the research published by marketing scholars. There are many successful marketing model developments, but their level of usage in practice is low. |
| Reibstein, Day, and Wind (2009) | Marketing | Journal of Marketing | 424 | Practical importance of research | C | The widening divergence between marketing academia and practice has become detrimental to the long-term health of the marketing field. |
| Mitra and Golder (2008) | Marketing | Journal of Marketing | 55 | Consequences of research for business schools | E | Academic research has positive long-term effects on the perceptions of academics, recruiters, and program applicants, and on education performance. A persistent increase of three single-author A-level articles per year is associated with an improved MBA ranking by one place. |
| Rosemann and Vessey (2008) | Information systems | MIS Quarterly | 445 | Practical importance of research | C | To improve practical importance, information systems researchers should combine rigor and relevance by conducting applicability checks with practitioners. |
| Bartunek (2007) | Management | Academy of Management Journal | 529 | Practical importance of research andadoption of research by practice | C | Management researchers need to develop a "relational scholarship of integration" with practitioners (i.e., develop fuller and emotionally engaging relationships in which each has an equal role, rather than being restricted to distant and one-sided "translation" efforts). |
| Gulati (2007) | Management | Academy of Management Journal | 499 | Practical importance of research and rigor of research | C | Many "serious scholars" presume that colleagues writing for practitioners ("management types") lack rigor, and those writing primarily for other scholars lack practical importance. It is important for management scholars to seek out room for reconciliation between rigor and practical importance. |
| Rynes, Giluk, and Brown (2007) | Management | Academy of Management Journal | 608 | Adoption of research by practice | E | Practitioner and bridge journals provide little coverage of some of the research findings deemed most important by scholars. When they do offer coverage, this coverage is often inconsistent with research evidence. |
| Shapiro, Kirkman, and Courtney (2007) | Management | Academy of Management Journal | 450 | Practical importance of research | E | The theory–practice gap stems not only from a "translation" problem (i.e., translating research for a practice audience) but also from a "production" problem (i.e., producing research that is relevant for a practitioner audience). |
| Tushman and O'Reilly III (2007) | Management | Academy of Management Journal | 280 | Practical importance of research | C | Executive education is a fertile and underleveraged setting to shape research that is both rigorous and relevant. |
| Bennis and O'Toole (2005)d | Management | Harvard Business Review | 2,984 | Practical importance of research andconsequences of research for business schools | C | Rigor crowded out most of the practical importance of the research conducted at business schools. The science model may not be applicable to business schools. Business is "essentially a human activity in which judgements are made with messy, incomplete and incoherent data" (Bennis and O'Toole 2005, p. 99). |
| Vermeulen (2005) | Management | Academy of Management Journal | 266 | Practical importance of research | C | Research that lacks rigor cannot be relevant. Business school scholars should strive to conduct research that "makes a difference." |
| Pfeffer and Fong (2002)d | Management | Academy of Management Learning & Education | 2,369 | Practical importance of research andconsequences of research for business schools | E | Business school research is making a modest contribution to management practice at best, especially when compared with research and ideas from consulting firms, journalists, and companies. |
| Rynes, Bartunek, and Daft (2001) | Management | Academy of Management Journal | 1,369 | Practical importance of research | C | The diffusion of academic knowledge to practitioners is slow. Practical knowledge gathered from practitioners can enhance scientific progress. |
| Trieschmann et al. (2000) | Management | Academy of Management Journal | 380 | Consequences of research for business schools | E | Research performance (e.g., number of first-tier publications) and MBA performance (e.g., business press rankings) have different determinants. |
| Benbasat and Zmud (1999) | Information systems | MIS Quarterly | 1,550 | Practical importance of research | C | Information Systems academic research lacks practical importance because it emulated the rigor of other academic fields. |
| AMA Taskforce (1988) | Marketing | Journal of Marketing | 138 | Practical importance of research | C | Marketing suffers from several structural impediments to the development and dissemination of knowledge with long-term impact. |
1 a We constrained the selection of articles in Table 1 to those published in journals on the UTD list.
- 2 b We collected the number of Google Scholar citations for the listed papers on February 1, 2021.
- 3 c While many papers cover multiple dimensions, we attempted to define the focus of the respective papers rather narrowly. The four dimensions we categorize papers on are ( 1) "consequences of research for business schools," which includes papers that explicitly take a business school perspective (as contrasted to a field perspective); ( 2) "practical importance of research," which includes papers that address threats to q-quality (e.g., the gap between academia and practice) from the perspective of academics; ( 3) "adoption of research by practice," which includes papers that address the limited application of academic research, from the perspective of practitioners; and ( 4) "rigor of research," which includes papers that address threats to r-quality (e.g., low replicability of studies, low rigor and scientific integrity of research).
- 4 d We included these articles published in journals outside the UTD list as an exception to the rule, because of their strong impact.
On the positive side, academic research may enhance a professor's relevant knowledge base, which can be transferred to students and motivate them to study the subject ([34]). Academic research may also signal teaching quality to high-quality prospective students ([12]). Business school faculty or deans may also advocate certain schools on the basis of their academic research performance, thus affecting school choices and driving high-quality students and faculty to research-intensive schools ([34]). On the negative side, scholars have lamented the lack of practical importance of business school research (e.g., [26]; [31]; [43]; [52]). In addition, science fraud cases in business schools have called into question the integrity and rigor of academic research in management ([13]).
Prior literature has hinted that the faculty research incentive system of business schools, composed of monitoring and compensation instruments, may be responsible for the main concerns on rigor (formally, r-quality) and practical importance (formally, q-quality) that are voiced about business school research ([29]; [31]; [41]; [60]). The purpose of this article is to examine the effects of the faculty research incentive system on the execution of the research task by faculty and, thereby, on a holistic set of business school outcomes, which, following prior work in the educational literature (e.g., [25]), we conceptualize as "business school health." Business school health is the extent to which a business school performs well ( 1) at the technical level (i.e., research and teaching), ( 2) at the institutional level (i.e., external support and institutional integrity), and ( 3) at the managerial level (i.e., leadership support, administrative support, and resource support). We define all key terms in Table 2.
Graph
Table 2. Key Construct Definitions and Representative Papers.
| Construct | Definition | Representative Works |
|---|
| Business School Health | The extent to which a business school performs well (1) at the technical level, (2) at the institutional level, and (3) at the managerial level. | Hoy, Tarter, and Kottkamp (1991) |
| Performance at the technical level | The extent to which the business school has high research health (i.e., research faculty are viewed as leading in their respective fields, publish regularly in leading journals, and assume academic leadership positions), and high teaching health (i.e., the school offers an excellent learning environment with high standards for teaching). |
| Performance at the institutional level | The extent to which the business school has high external support (i.e., very good relationships with alumni and donors, who commit substantial resources to the school), and high institutional integrity (i.e., faculty and students uphold the highest standards of integrity). |
| Performance at the managerial level | The extent to which the business school has high leadership support (i.e., a high-quality leadership team and clear faculty performance standards), high administrative support (i.e., professional administrative staff that is supportive to faculty, students, and visitors), and high resource support (i.e., adequate facilities and resources to help faculty effectively perform their work). |
| Research Task of the Faculty | The research task of business school faculty is to produce research of sufficient quantity and quality. | Gomez-Mejia and Balkin (1992) |
| Research quantity | The total volume of academic research produced by a scholar, or a group of scholars. | Lightfield (1971) |
| r-quality | Academic research that adheres to "objective, scientific standards" (Bennis and O'Toole 2005, p. 99). Often equated to rigor. | Bennis and O'Toole (2005),Ellison (2002)Lehmann, McAlister and Staelin (2011),Vermeulen (2007) |
| q-quality | Academic research that provides insights that "practitioners find useful for understanding their own organizations and situations better than before" (Vermeulen 2007, p. 755). Often equated to practical importance. |
| Faculty Research Incentive System | The set of monitoring and compensation instruments that a business school puts in place to steer the research of its faculty and minimize agency problems such as the faculty not doing enough research or doing research that is not good enough. | Shapiro (2005) |
| Faculty research monitoring instruments | The set of devices that business schools use to measure research faculty's effort or outcomes. | Joseph and Thevaranjan (1998) |
| Faculty research compensation instruments | The set of rewards that business schools use to align the actions of research faculty with the objectives of the business school. | Ahuja and Yayavaram (2011) |
5 Notes: For a complete set of construct definitions and corresponding operationalizations, see Table W1 in the Web Appendix, section W2.
This research offers two main contributions. First, many articles take a scholarly field perspective rather than a business school perspective. Exceptions ([11]; [34]; [39]; [56]) focus on specific business school outcomes (e.g., master of business administration [MBA] ranking) or specific research metrics (e.g., number of publications) and often contradict each other, with some being very negative and others being more positive. This article also takes a business school perspective, but it offers more elaboration on faculty research incentives, faculty research task, and business school outcomes (i.e., business school health) than prior research. Second, prior work suggesting that the faculty research incentive system is one of the main culprits for today's state of affairs (see, e.g., [31]; [41]; [60]) did not theoretically conceptualize this faculty research incentive system or offer empirical evidence of its misalignment. This article does both.
We theoretically ground our hypotheses in sociological agency theory ([49]). We provide empirical evidence from ( 1) a survey of 234 marketing professors in business schools across 20 countries (response rate of 62.6%), ( 2) qualitative interviews with 14 (associate) deans of 13 business schools in the United States and Europe, and ( 3) qualitative interviews with 8 external stakeholders representing external institutions of marketing scholarship (e.g., the American Marketing Association) and marketing practice at large multinational firms.
Our main conclusions are as follows. Research task incentives are badly designed, on average. Among monitoring instruments, we find that business schools, on average, overweight number of publications in faculty evaluations while creativity, literacy, relevance to nonacademics, and awards (in order of importance) receive too little weight. Among compensation instruments, we find, on average, that faculty feel they are insufficiently compensated, whereas (associate) deans feel that faculty are compensated too much for their research. We find that badly designed incentive systems are more prevalent in schools that perform below the median on research quality—that is, r-quality (i.e., rigor) and q-quality (i.e., practical importance). We do not find such a relationship between badly designed research incentives and research quantity.
Regarding the research task of the faculty, we find that research quantity contributes to business school research health but not to other aspects of business school health. The r-quality of research contributes more strongly to business school research health than research quantity and q-quality of research. The q-quality of research does not contribute to business school research health but does contribute positively to business school teaching health as well as several other dimensions of business school health, such as external support (by alumni and donors) and institutional integrity.
Our findings have important implications for business schools and research faculty. First, business schools need to develop better research metrics to monitor the academic research of their faculty. Second, business schools need to improve alignment with their faculty on compensation. Third, business schools need to improve the quality (especially q-quality) of their faculty's research. We provide specific suggestions how business schools can follow up on each of these three main implications.
We develop a sociological agency framework on business school research (see Figure 1), in which we distinguish four elements: ( 1) constituents (e.g., principal, agents, institutions), ( 2) incentive instruments[ 6] the principal uses to motivate the agent (e.g., publication metrics), ( 3) the task of the agent (e.g., research), and ( 4) desirable outcomes for the principal (e.g., business school health).
Graph: Figure 1. A sociological agency theory perspective on academic research in business schools.
The business school is a "collective principal," comprising a chain of delegation in a system of peers, akin to complex administrative structures often found in international organizations (e.g., [35]). Business schools typically operate within a university, which oversees the school's incentive system (exceptions exist, e.g., INSEAD) and are divided into disciplinary units or departments, each of which influences the school's incentive system (see top of Figure 1). The agent in our framework is a research or tenure-track faculty member. The business school incentivizes the research of agents by monitoring and compensating the faculty member's research task.
External institutions are organizations outside the governance of the business school that play an essential role in social monitoring because principal–agent relationships are "enacted in a broader social context and buffeted by outside forces" ([49], p. 269).[ 7] Building on [ 2], we discern two external institutions of special relevance[ 8]: ( 1) endorsement institutions and ( 2) cohesion institutions (see the bottom of Figure 1; for a primer and nonexhaustive list of these institutions in the marketing field, see section W1 in the Web Appendix).
Endorsement institutions verify information about agents, conduct analyses to compare or rank agents, and endorse agents. Examples of such institutions in marketing that endorse faculty are premier journals that publish their research (e.g., the Journal of Marketing) or associations (e.g., the American Marketing Association [AMA]) that have a variety of awards for research. Cohesion institutions ensure collective action by enabling the provision of collective goods. Collaborative research platforms, such as the Marketing Science Institute (MSI) or Institute for the Study of Business Markets (ISBM), are good examples of such cohesion institutions (note that institutions can provide endorsement as well as cohesion, as is the case for the AMA).
Principals use monitoring instruments to measure an agent's effort or outcomes ([27]), of which the following are relevant for business school research (e.g., [30]): ( 1) number of publications, ( 2) number of citations, ( 3) peer recognition, ( 4) awards, ( 5) relevance to nonacademics, ( 6) literacy,[ 9] and ( 7) creativity. Compensation instruments are the rewards, pecuniary and nonpecuniary, that principals use to align the actions of agents with their own objectives, of which the following are relevant for business school research (e.g., [21]): ( 1) salary, ( 2) performance-based salary increases, ( 3) publication bonuses paid as salary supplements,[10] ( 4) research budgets, ( 5) publication bonuses paid as supplementary research budget,[11] ( 6) academic freedom, and ( 7) reduced teaching loads.
The faculty research task is to produce research of sufficient quantity ("doing enough research") and quality ("doing research that is good enough"). Research quantity relates to the total volume of research produced by a scholar (e.g., [30]). For research quality, we distinguish "r-quality" from "q-quality" ([17]). Academic research is of high r-quality (i.e., rigorous) if it adheres to "objective, scientific standards" ([11], p. 99), which means that "the various elements of a theory are consistent, that potential propositions or hypotheses are logically derived, that data collection is unbiased, measures are representative and reliable, and so on" ([61], p. 755). Academic research is of high q-quality (i.e., practically important) if it provides insights that "practitioners find useful for understanding their own organizations and situations better than before" ([61], p. 755).
Building on the classic work of [38] and [25], we define a healthy business school as a business school that performs well at three levels: ( 1) the technical level, ( 2) the institutional level, and ( 3) the managerial level. At the technical level, a healthy business school has high research health (i.e., research faculty are seen as leading in their respective fields, publish regularly in leading journals, and assume academic leadership positions) and high teaching health (i.e., the school offers an excellent learning environment with high standards for teaching). At the institutional level, a healthy business school has high external support (i.e., very good relationships with alumni and donors, who commit substantial resources to the school) and high institutional integrity (i.e., faculty and students uphold the highest standards of integrity). At the managerial level, a healthy business school has strong leadership support (i.e., a high-quality leadership team and clear faculty performance standards), strong administrative support (i.e., professional administrative staff that is supportive to faculty, students, and visitors), and strong resource support (i.e., adequate facilities and resources to help faculty effectively perform their work).
Next, we develop our hypotheses, starting with the effects of incentive instruments on the research task of the faculty,[12] after which we turn to the effects of the research task on business school health (for a graphical overview, see Figure 2).
Graph: Figure 2. The effect of the faculty research incentive system on the research task of the faculty and business school health.
According to agency theory (e.g., [23]), incentive instruments increase an agent's motivation by raising the marginal cost of bad performance (through monitoring) and/or the marginal reward of good performance (through compensation). Higher motivation, in turn, leads the agent to work harder and to perform better on their task. However, there are multiple reasons to expect the effect of incentive instruments on the research task of professors in business schools to be more nuanced.
Incentive instruments may be improperly weighted and deviate from what both agents and principals see as the optimal incentive system, because optimal incentives are typically costly to design and implement ([27]). For instance, often, quality is more expensive to monitor than quantity ([23]). In the context of business schools, an increasing number of automated scientometric tools make the monitoring of research quantity inexpensive, while the monitoring of research quality remains expensive for multiple reasons: ( 1) it is more difficult to objectify quality than to objectify quantity, ( 2) it is more difficult to compare research quality across domains than to compare research quantity across domains, and ( 3) senior business school administrators may have been detached from high-quality research activities themselves for a long time. Consequently, business schools may design incentive systems that overweight research quantity, possibly at the expense of research quality.
Incentive systems that overweight quantity may lead faculty to become extrinsically motivated to publish as many papers as possible, possibly leading them to ignore quality ([23]) or to engage in undesirable practices to game the metrics rather than optimize the task itself. An example is "salami publishing" (i.e., trying to squeeze as many papers as possible out of a research project). Therefore, we expect improperly weighted incentive instruments to increase the quantity of faculty's research.
However, such an increase in quantity may come at the expense of a decrease in (r- and q-) quality of the faculty's research. Badly designed incentive systems reduce the intrinsic motivation of the agent because agents in badly designed incentive systems may feel underappreciated, which impairs self-esteem, or externally pressured, which impairs self-determination ([19]). Impaired self-esteem reduces agents' persistence in difficult tasks ([33]), which is critical to improve or sustain r-quality ([ 3]; [17]). Impaired self-determination reduces creativity ([ 5]), which is an important precursor to q-quality ([51]). For instance, [14], p. 5) argues that "home run" papers "pose new questions that we had never thought to ask" or "allow us to see existing problems and solutions from a new perspective." Therefore, we hypothesize:
- H1: In business schools with improperly weighted incentive instruments, research faculty (a) produce a higher quantity of research, (b) produce research of lower r-quality, and (c) produce research of lower q-quality compared with business schools with properly weighted incentive instruments.
Next, we postulate the effects of the research task of the faculty on research health and teaching health as well as on external support and institutional integrity. We do not develop ex ante expectations for the managerial level of business school health.[13]
Scholars who publish a high research quantity (controlling for quality) have higher visibility than scholars who publish a low research quantity ([54]). Scholars who frequently "survive" peer review also demonstrate to others they know "what is needed, correct, and valued" by the research system ([29], p. 156) and typically attract more collaborations, increasing their belongingness to the academic community. Higher visibility and belongingness increase the extent to which a scholar attains academic leadership. Therefore, we hypothesize:
- H2: The research health of a business school increases with the production of a higher quantity of research by its research faculty.
For research quality, the effect on research health may be more nuanced; we expect increases in r-quality of faculty's research to contribute more strongly to research health of a business school than increases in q-quality of faculty's research. [ 1] show that scholars acclaim stronger reputational rewards to basic than to applied science because basic research requires a higher level of scientific ability than applied research. Basic research is typically higher in rigor than applied research, which, in turn, is typically higher in practical importance ([58]). [ 3] calls this the "hardness bias," which he also attributes to the greater agreement among scholars on r-quality than on q-quality. In turn, the greater reputational rewards faculty may derive from increments in r-quality, as compared with increments in q-quality, fuel opportunities to take up leadership roles in journals and in the academic research community ([17]). Therefore:
- H3: The research health of a business school increases more as research faculty produce research of higher r-quality than as research faculty produce research of higher q-quality.
Research quantity may have two opposite effects on teaching health. On the one hand, a high volume of research may give faculty members a broader knowledge base in their teaching subjects, increasing their ability to set high teaching standards and to motivate students' interest in the subject ([34]). On the other hand, research and teaching activities compete for faculty time. Assuming a time constraint, the more research faculty allocates time to writing papers, the less they allocate time to preparing classes, creating teaching materials, and meeting with students. [12] analytically show that increasing research output may deteriorate teaching quality. Therefore, we formulate two alternative hypotheses:
- H4a: The teaching health of a business school increases as research faculty produce a higher quantity of research.
- H4b: The teaching health of a business school decreases as research faculty produce a higher quantity of research.
Faculty members who produce research high in q-quality typically immerse themselves in real-world managerial practice through consulting, case writing, or executive education ([61]). Such immersion, in turn, increases a faculty member's usage of concrete concepts, which are easier to understand than abstract concepts ([57]). In contrast, high r-quality faculty tends to abstract from contextual details to focus on the key underlying properties of a situation or problem ([29]). Moreover, the strong theoretical and methodological grounding of high r-quality faculty may lead them to underestimate that abstract concepts may not be obvious to less informed audiences. Therefore, we expect faculty who produce research high in q-quality (high in r-quality) to use more concrete (more abstract) concepts when teaching students. Teaching in concrete rather than abstract language is more effective because it enhances student comprehension and memory retention ([47]), which, in turn, may ensure high teaching standards. Therefore, we hypothesize:
- H5: The teaching health of a business school increases more as research faculty produce research of higher q-quality than as research faculty produce research of higher r-quality.
We expect increases in research quantity to contribute less to external sponsors' (i.e., alumni and donors) willingness to donate their time or money to the school than increases in research (r- and q-) quality. Using self-reported data from alumni, [32] show that donors' self-esteem increases more when they donate to a high-prestige than to a low-prestige school. The production of high-quality research is a more important driver of the prestige of an academic institution than the production of a high quantity of research ([15]), for two main reasons.
First, a rare favorable outcome (e.g., publishing a "home run" paper) conveys more information about an individual's ability than being able to achieve several less favorable outcomes ([50]). Thus, research quality is more significant than research quantity in eliciting recognition through awards, appointments to prestigious academic departments, and overall prestige among national and international peers ([15]).
Second, the awards and accolades bestowed to high-quality scholars serve as signals of appreciation and recognition by external experts. [32] argue that academic institutions can symbolically manage such quality signals as "identity anchors" that increase the salience of the institution among alumni and donors and, ultimately, their willingness to support the institution. Accordingly, we hypothesize:
- H6: The external support for a business school increases more as research faculty produce research of higher quality (in both r- and q-quality) than as research faculty produce a higher quantity of research.
Faculty who conduct research high in r-quality are more likely to adopt and disseminate the latest scientific guidelines that heavily endorse research integrity ([36]) and, in turn, may develop a stronger overall "moral muscle" that transcends domains ([ 9]) than faculty who conduct research low in r-quality. Faculty with a stronger moral muscle may more effectively disseminate ethical values to students. Therefore, we hypothesize:
- H7: The institutional integrity of a business school increases as research faculty produce research of higher r-quality.
We do not have theoretical expectations regarding the effects of research quantity and q-quality on a business school's institutional integrity. We explore such effects empirically.
We control for several other effects in our empirical tests. First, we empirically explore the effects of the research task of the faculty on the managerial level of business school health, for which we did not posit ex ante expectations. Second, we allow for correlated error terms when we estimate the effects of the research task of the faculty on the different dimensions of business school health. In this manner, we accommodate for the existence of feedback loops that we conceive in two main ways (see right-to-left arrows at the top of Figure 2): ( 1) business school health may influence the faculty in the execution of their research task and ( 2) the faculty's execution of the research task may lead to adjustments in monitoring and compensation.
In this section, we provide empirical evidence from surveying marketing faculty members and interviewing (associate) deans of business schools and external stakeholders.
We invited 374 marketing academics across 168 business schools to respond to our survey; 234 responded (62.6%). Of these, 182 (77.8%) respondents work at research-intensive schools (i.e., schools where tenure criteria are mainly research focused) and 149 of the respondents (63.7%) work at business schools that are ranked in the Top 100 Financial Times (FT) Global MBA ranking. For further details on survey sampling, questionnaire structure, analysis, and results, see section W2 in the Web Appendix and visit www.frisbuss.com.[14]
Regarding the faculty research incentive system, we asked respondents if, at their school, each of the seven monitoring instruments we study receives far too little weight (−2), too little weight (−1), just the right weight (0), too much weight (+1), or far too much weight (+2). Regarding compensation, we asked respondents whether they felt research faculty at their school receive far too little (−2), too little (−1), just the right level (0), too much (+1), or far too much (+2) of each of the seven compensation instruments we study.
Regarding the research task of the faculty, we asked respondents whether the performance of research faculty at their business school in each of the three dimensions of the research task (i.e., research quantity, r-quality, and q-quality of research) was "very low," "low," "moderate," "high," or "very high."
To measure business school health, we created a 21-item scale (see Table 3) by adapting earlier measures of [25] to the business school context. We conducted a principal component analysis with varimax rotation on this scale. The scree plot suggested a seven-component structure with all items loading on their expected theoretical dimensions.[15] The seven components accounted for 82.6% of the total variance, with the largest component accounting for 12.7% of the total variance. All loadings were greater than the recommended threshold of.60, with the lowest being.73 (see Table 3). Next, we conducted a seven-factor confirmatory factor analysis (CFA; χ2 = 251.08, p <.01, d.f. = 168). The fit indices for this model meet the recommended standards (comparative fit index [CFI] =.98, nonnormed fit index [NNFI] =.97, root mean square error of approximation [RMSEA] =.05, square root mean residual [SRMR] =.04).[16]
Graph
Table 3. Business School Health Scale Items and Factor Loadings.
| Items | Factor Loading(PCA)a | Factor Loading(CFA)b |
|---|
| Research Health (M = 3.46; SD =.98; CR =.92; AVE =.78) | | |
| Our faculty is seen as leading in research by peers internationally. | .90 | .94 |
| Our faculty publishes regularly in the best journals in their respective fields. | .89 | .91 |
| Our faculty takes up leadership positions in the academic research community. | .86 | .80 |
| Teaching Health (M = 3.71; SD =.78; CR =.82; AVE =.60) | | |
| The school sets high standards for teaching. | .85 | .86 |
| Faculty accepts their responsibility toward providing students with an excellent learning environment. | .80 | .86 |
| Faculty that do well in the classroom are well respected in the school. | .84 | .71 |
| External Support (M = 3.34; SD = 1.05; CR =.93; AVE =.81) | | |
| Our school has the support of external stakeholders (alumni, donors) who are willing and able to commit substantial resources (e.g., time, money) to the school. | .88 | .91 |
| Our school has a very good relationship with external stakeholders (alumni, donors). | .87 | .92 |
| It is easy for our school to call on external stakeholders (alumni, donors) when times get tough. | .85 | .88 |
| Institutional Integrity (M = 3.58; SD =.92; CR =.81; AVE =.59) | | |
| Our school is able to maintain high integrity despite possible pressure from external influencers. | .84 | .85 |
| Our school and faculty commit to the highest standards of integrity on a daily basis, even if this comes at a short-term cost. | .80 | .89 |
| We communicate stronger ethical values to our student and faculty body than most of our peers. | .75 | .75 |
| Leadership Support (M = 3.35; SD = 1.06; CR =.89; AVE =.74) | | |
| The school's leadership maintains clear standards for faculty performance. | .86 | .90 |
| The school's leadership lets faculty know what is expected of them. | .86 | .86 |
| Our leadership team is of high quality. | .73 | .88 |
| Administrative Support (M = 3.51; SD =.91; CR =.84; AVE =.63) | | |
| Our administrative staff (i.e., PA's and secretaries, program support staff, business development staff, people division, etc.) is very supportive to faculty such that faculty can focus on their primary responsibilities. | .77 | .85 |
| Our administrative staff is greatly appreciated by our students and by visitors to our school. | .81 | .72 |
| Our administrative staff is very professional, and their competences are well developed. | .79 | .90 |
| Resource Support (M = 3.65; SD =.91; CR =.80; AVE =.57) | | |
| Our school has great facilities in which to perform our work. | .74 | .69 |
| We have adequate resources for all tasks assigned to us. | .86 | .89 |
| We have access to resources and materials when we need them to perform our work effectively. | .85 | .95 |
- 6 a Loadings from a principal component analysis with varimax rotation on the full set of 21 items without predetermined factors. Each item had its highest loading in its theorized factor, which we report here.
- 7 b Standardized loadings obtained from a confirmatory factor analysis with items preloaded on the seven business school health dimensions.
- 8 Notes: CR = composite reliability ([ 7]); AVE = average variance extracted ([18]).
Overall, our business school health scale exhibits good psychometric properties. All seven dimensions show composite reliabilities above the recommended threshold of.70 ([ 7]), the smallest being.80 (see Table 3). All factor loadings were positive, highly significant (minimum z-value was 18.95; all p-values below.01), and at least ten times as large as the standard errors establishing convergent validity ([20]). For all pairs of business school health dimensions, the square root of the average variance extracted for both dimensions was greater than their correlation, which demonstrates acceptable discriminant validity ([18]). We averaged respondents' answers across each set of three items for each business school health dimension to produce seven summated scales.
We addressed CMV ex ante by ( 1) promising confidentiality to respondents ([40]), ( 2) using well-defined response labels that varied across questions ([42]), and ( 3) asking respondents to evaluate their business school's performance rather than their own performance, triggering high involvement and informant reliability ([24]). Ex post, we found that ( 1) the largest factor in our principal component analysis accounted for only 12.7% of the variance explained, and ( 2) a single-factor CFA model fits the data worse than our hypothesized model (CFI =.49, NNFI =.43, RMSEA =.20, SRMR =.12). Both findings are inconsistent with severe CMV.
Figure 3 shows the average value (μ) for each incentive instrument. The asterisks depict whether this value is significantly different from 0; 0 indicates that the weight given to that instrument is "just right." On average, Figure 3 shows that business schools' research incentive systems are badly designed.
Graph: Figure 3. Misalignment of incentive instruments.* p <.10.** p <.05.*** p <.01.aThe question asked for each monitoring instrument was "At your school, do you feel that the following metrics on research faculty receive too much or too little weight?" (−2 = "Far too little weight," −1 = "Too little weight," 0 = "The weight is just right," +1 = "Too much weight," and +2 = "Far too much weight").bThe question asked for each compensation instrument was "At your school, do you feel that research faculty receive too little or too much of each of the following as rewards for their research?" (−2 = "Far too little," −1 = "Too little," 0 = "Just right," +1 = "Too much," and +2 = "Far too much").Notes: The asterisks represent the p-values for t-tests comparing the mean score for the perceived appropriateness of the weight given to each instrument to 0 (which means the weight is "just right"). All p-values are two-sided. In the case of compensation questions, respondents could answer "not applicable"; thus, we indicate the sample used to compute mean responses next to each compensation instrument's label in the right panel.
Of the monitoring instruments (Figure 3, Panel A), we find that the "number of publications" receives too much weight (μ =.39; t = 6.88, p <.01). All other monitoring instruments receive too little weight, especially so for (in order) ( 1) creativity (μ = −.65; t = −12.95, p <.01), ( 2) literacy (μ = −.49; t = −10.05, p <.01), and ( 3) relevance to nonacademics (μ = −.44; t = −8.58, p <.01).
Of the compensation instruments (Figure 3, Panel B), we find that respondents consider research faculty at their school to be insufficiently compensated, except for the academic freedom they get, especially so for (in order) ( 1) bonuses paid as research budget (μ = −.84; t = −10.34, p <.01), ( 2) bonuses paid as salary (μ = −.77; t = −9.47, p <.01), and ( 3) reduced teaching loads (μ = −.67; t = −10.36, p <.01).
To test H1, we first generated a 2 × 2 matrix according to a median split of respondents as below median or above median in terms of the performance on research quantity and r-quality of their business school (Figure 4, Panel A). Then, for each respondent, we computed the mean absolute deviation (MAD) from 0, aggregated across all seven monitoring and compensation instruments.[17] We then averaged these individual scores to obtain MADM and MADC for each of the cells in the 2 × 2 matrix.
Graph: Figure 4. Misalignment of incentive instruments: variation according to research quantity and r-quality and q-quality.Notes: To measure whether faculty research incentive instruments are improperly weighted (i.e., misaligned) we computed mean absolute deviations (MAD). Specifically, we first computed individual MAD scores, which are the averages of the absolute deviations between a respondent's scores in all items of a given scale (say, all seven monitoring instruments) and the central point of the scale (which indicates that the weight given to a given instrument is "just right"). The values reported in this figure are the averages, across respondents in a given cell, of these individual MAD scores for monitoring instruments (MADM) and for compensation instruments (MADC). To avoid a skewed split, we randomly classified respondents in the "median category" (e.g., those with a score of 4 for research quantity) as either "below median" or "above median" using a proportion that ensures that approximately half of the respondents are classified as "below median" and the other half as "above median" in each dimension.
We ran two one-way analyses of variance of the MADM and MADC by respondents across the four cells in Figure 4, Panel A. Fisher–Hayter post hoc tests[18] show that there are no significant differences in the extent to which incentive instruments are properly weighted (i.e., MADM and MADC) in schools with above-median versus below-median research quantity (see Web Appendix, section W2). Thus, we are not able to confirm H1a.
Consistent with H1b, Fisher–Hayter post hoc analyses show that in schools with above-median r-quality (i.e., upper cells in Figure 4, Panel A), monitoring instruments are more properly weighted (i.e., lower MADM) than in schools with below-median r-quality (i.e., lower cells), an effect that is significant both at low levels of research quantity (p <.05) and at high levels of research quantity (p <.01). Compensation instruments are more properly weighted (i.e., lower MADC) in schools with above-median r-quality (i.e., upper cells) than in schools with below-median r-quality (i.e., lower cells), an effect that is significant at the 10% level at low levels of research quantity (p <.10) and approaches significance at high levels of research quantity (p =.14).
We used the same approach to generate a 2 × 2 matrix according to a median split on research quantity and q-quality (Figure 4, Panel B). We then ran two one-way analyses of variance of the MADM and MADC by respondents across the four cells in Figure 4, Panel B. Consistent with H1c, Fisher–Hayter post hoc analyses show that in schools with above-median q-quality (i.e., upper cells in Figure 4, Panel B), monitoring instruments are more properly weighted (i.e., lower MADM) than in schools with below-median q-quality (i.e., lower cells), an effect that is significant at low levels of research quantity (p <.05) but not at high levels of research quantity (p =.18). We do not find such a contrast for compensation instruments (MADC).
To test H2–H7, we estimated a multivariate regression system of the seven dimensions of business school health on the three dimensions of the research task (research quantity, r-quality, and q-quality), with correlated error terms across the seven equations (see the Web Appendix, section W2). The Lagrange multiplier test proposed by Breusch and Pagan confirms that the covariance matrix between error terms is not diagonal (χ2(21) = 573.8, p <.01). The fit of the model is satisfactory. The R2-statistic is highest for research health (.46), which befits the primary focus of our investigation.
We depict our results in Table 4. The first four rows show the parameter estimates of research task on business school health, whereas the subsequent seven rows show the residual correlations among the different business school health dimensions. Confirming H2, we find that higher research quantity is associated with higher research health (β =.28; p <.01). We also find that the higher the r-quality of faculty research, the higher the research health of a business school (β =.52, p <.01). In contrast, q-quality has no significant effect on research health (β =.05; p =.32). A Wald test rejected the null hypothesis that the parameters for r- and q-quality are equal (F = 10.11, p <.01), thereby confirming H3.
Graph
Table 4. Impact of Faculty Research on Business School Health.
| Research Health | Teaching Health | External Support | Institutional Integrity | Leadership Support | Admin. Support | Resource Support |
|---|
| Regression Estimates | | | | | | | |
| Constant | .57** | 3.09*** | 2.34*** | 2.06*** | 1.30*** | 2.20*** | 2.56*** |
| Research quantity | .28*** | –.03 | –.17* | .06 | .12 | –.04 | .01 |
| r-quality ("rigor") | .52*** | .02 | .21*** | .17** | .36*** | .26*** | .22*** |
| q-quality ("practical importance") | .05 | .21*** | .27*** | .25*** | .14** | .18*** | .10 |
| Residual Correlations | | | | | | | |
| Research health | | .22 | .32 | .07 | .16 | .15 | .23 |
| Teaching health | .22 | | .33 | .44 | .29 | .31 | .28 |
| External support | .32 | .33 | | .34 | .36 | .39 | .38 |
| Institutional integrity | .07 | .44 | .34 | | .47 | .38 | .38 |
| Leadership support | .16 | .29 | .36 | .47 | | .46 | .39 |
| Administrative support | .15 | .31 | .39 | .38 | .46 | | .47 |
| Resource support | .23 | .28 | .38 | .38 | .39 | .47 | |
| N = | 234 | 234 | 234 | 234 | 234 | 234 | 234 |
| R2 = | .46 | .08 | .12 | .16 | .20 | .14 | .09 |
- 9 *p <.10.
- 10 **p <.05.
- 11 *** p <.01.
- 12 Notes: All p-values are two-sided. The first four rows depict the parameter estimates from our multivariate regression. The subsequent seven rows depict the correlations obtained from the residual correlation matrix. We rely on a multivariate regression because it allows us to jointly estimate the seven models as one regression system while accounting for error correlations. Multivariate regression is a special case of Zellner's seemingly unrelated regression with identical regressors across equations, in which case the seemingly unrelated regression estimator simplifies to ordinary least squares in each equation. Yet, because it is a joint estimator, the multivariate regression also estimates between-equation error correlations, allowing us to efficiently test coefficients across equations.
Confirming neither H4a nor H4b, we find that a higher research quantity does not have a significant effect on teaching health (β = −.03, p =.67). We also find that a higher q-quality of faculty research is associated with higher teaching health of a business school (β =.21, p <.01), whereas higher r-quality is not (β =.02, p =.69). A Wald test rejected the null hypothesis that the parameters for r- and q-quality are equal (F = 4.53, p <.05), thereby confirming H5.
Confirming H6, we find that research quantity may negatively affect external support (β = −.17, p <.10), while higher levels of r-quality (β =.21, p <.01) and of q-quality (β =.27, p <.01) positively affect external support. A Wald test showed that the coefficients for r-quality and q-quality are not significantly different from one another (F =.28, p =.59).
Confirming H7, we find a positive effect of r-quality (β =.17, p <.05) on institutional integrity. We find no significant effect of research quantity on institutional integrity (β =.06, p =.44) and a positive and significant effect of q-quality on institutional integrity (β =.25, p <.01).
As to other effects, we observe that schools with high r-quality research have strong leadership support (β =.36, p <.01), strong administrative support (β =.26, p <.01), and strong resource support (β =.22, p <.01). Schools with high q-quality research have strong leadership support (β =.14, p <.05), administrative support (β =.18, p <.01), and resource support (β =.10, p =.12). We do not find any association between research quantity and leadership support (β =.12, p =.18), administrative support (β = −.04, p =.64), or resource support (β =.01, p =.91).
Next, we report on the interviews we conducted with (associate) deans and with representatives of external institutions. These interviews took 35 minutes on average and yielded a total of 164 pages of single-spaced transcripts.
We conducted phone interviews with seven deans (four former and three current) and seven associate deans (two former and five current) at 13 business schools in the United States and Europe (for more information, see section W3 in the Web Appendix), who are good informants ([24]). We opted for a "phenomenological" approach that is in-depth but nondirective in nature ([55]). We audio-recorded the interviews (except for two who did not give permission), which were subsequently transcribed by a research assistant and double-checked by one of the authors for accuracy. Our interviews led to the following insights.
First, virtually all (associate) deans we interviewed expressed that there is an overreliance on effortless metrics (especially counting number of publications, but also number of citations) often at the expense of more effortful metrics such as creativity, literacy, and relevance to nonacademics. Of the 14 (associate) deans we interviewed, 11 recognized this overreliance on effortless metrics, and 9 explicitly mentioned they saw this trend as problematic for business schools, as highlighted by the following quotes:
I definitely have seen just what I feel is an overreliance on the cohort table and the numbers. And I feel that that was something that I have kind of raised but I do not feel that I necessarily had any impact in terms of trying to say this is just one piece of information. (Former vice-dean for faculty at a U.S. FT Top 25 school)
When I started in 2000–2001, it was about the quality of the journals and what the outside reviewers said. So initially, there was very light weight put on citation counts, and then over time, it started to increase a bit and then we got a couple of people elected to the promotion and tenure committee who were like, "We don't even have to look at quality, we can tell from the citation counts whether these things are any good or not." (Former dean at a U.S. FT Top 30 school)
[Awards] should weigh a lot even when compared with contemporary productivity metrics, but in all honesty, contemporary productivity metrics are some of the most overused metrics to gauge academics. (Current dean of research at a non-U.S. FT Top 75 school)
My frustration is, when I'm drawing on a department chair for information, I get counts such as they had 27 publications, 4 in premier outlets, and this was the citation count. (Current dean at a large U.S. public school)
I remember when Google Scholar first came out, there was a lot of skepticism about it...but that has definitely been adopted as the norm. And I think the ease of checking it and following it has caused a drift toward weighing it more heavily. (Former dean at a U.S. FT Top 15 school)
Are we just giving up on our ability to be doing all the heavy work? I think we are relying too much on the ease of numbers. (Current dean at a U.S. FT Top 75 school)
I personally view it [a growing reliance on counting] as a very negative trend because people start gaming the citation count. (Current dean at a U.S. FT top 100 school)
Now that we have metrics and now that people are scored on those metrics, I think that the system does—it shouldn't, but it does—put a greater emphasis on those numbers and less on, for example, creativity. (Current vice-dean at a U.S. FT Top 10 school)
Second, 9 out of the 14 (associate) deans we interviewed found business school professors overpaid for the research they do, in contrast with the views of research faculty in our survey. The following three quotes illustrate their views:
People come with their hands out all the time. I do not get it. It is just wrong. And I think we get paid really well. We have been historically. And we get things that other university faculty just do not get, like guaranteed summers. I mean, talk to someone in public health, right? It has become an absurdity to me, and it's very unsustainable. (Current dean at an FT Top 75 school)
The financial incentives that exist right now in the field are, to a certain extent, disturbing the market. I think the financing model of the top 100 business schools in the U.S. sooner or later will explode....It is a crisis waiting to happen. (Current dean of research at a non-U.S. FT Top 75 school)
Nowadays, it is too hard to get faculty to do things, so you start compensating, paying for everything. (Current dean at a large U.S. public school)
Nearly all the (associate) deans we interviewed also expressed a negative opinion on publication bonuses, again in contrast with research faculty in our survey. The following two quotes are representative of this generalized negative feeling:
We do not have bonuses for publications, and I do not find those a good idea; they may trigger perverse behaviors. (Current vice-dean at a public non-U.S. business school)
I think that, at least among our faculty, if a bonus were paid directly for a paper, it would make faculty feel like coin operated. And I think that would lead to a culture impact that would not serve us. (Former dean at a U.S. FT Top 15 school)
Third, the interviews largely confirmed that research quantity and research quality (both in r-quality and q-quality) are important for a business school's research health. Nine of our interviewees expressed a more positive view on the extent to which their school's faculty was achieving this on r-quality than on q-quality:
Basic science tries to understand how the world works, applied science tries to develop applications. I believe that management research is now 99% "basic" and only 1% "applied." (Current vice-dean at a public non-U.S. business school)
We like to see people who hit a home run, like, "this is a really good paper."...There's a lot of acceptance of low productivity rates if the quality of the home runs is there. (Former deputy dean at a U.S. FT Top 30 school)
I feel increasingly frustrated by the extent to which we talk to other academics and we do work that is not addressing the issues and questions that are really most pressing in the world of business or the world more broadly, and that we could be a lot more relevant and we could be speaking to practice a lot more. (Former vice-dean for faculty at a U.S. FT Top 25 school)
At some level, most of the work that I see that goes on doesn't connect to management....Sometimes the research is so technical that it's not acceptable to a broader audience. (Former deputy dean at a U.S. FT top 30 school)
When I look at what's in the journals, it strikes me that most of it is pretty irrelevant to what's going on in the world. So, I think that's a huge issue. (Former dean at a U.S. FT Top 30 school)
Fourth, while basically all the (associate) deans we interviewed viewed teaching health as fundamental, four of our interviewees expressed concerns with the impact of the research task of the faculty on teaching health, as illustrated by the following two quotes:
We have a management department...and I think at this point, there's maybe two people in there who could be teaching exec ed. And that is where your leadership people should be...and they just can't do it. At some level, we may kick ourselves out of business. (Current dean at a U.S. FT Top 75 school)
It seems every marketer wants to be a social scientist and wants to stop selling cookies. I mean, there are a lot of marketing scholars that fundamentally do not study marketing topics anymore and just look at topics that are generic social science research topics. (Current dean of research at a non-U.S. FT Top 75 school)
We conducted phone interviews with eight external stakeholders including ( 1) current or past leaders at five external institutions of marketing scholarship (e.g., MSI); and ( 2) senior marketing practitioners at three large multinational firms (the former global chief marketing officer of a large multinational technology corporation, the current chief executive officer [CEO], and an executive vice president [EVP] at two of the world's largest market research firms), who are or have been involved with these external institutions.
The interviews with external stakeholders yielded the following key insights. First, consistent with our theorizing, interviewees expressed that business schools track endorsement institutions' monitoring of their faculty. As a former chair-elect of the AMA Board of Directors pointed out:
[The] AMA aims to promote the creation of cutting-edge marketing content both through the journals and through the awards inside the Foundation. Faculty go back and list those awards on their annual reviews and use that as part of their argument for where they should stand inside their institution.
Second, again consistent with our theorizing, interviewees confirmed that cohesion institutions enable the provision of a common base of knowledge, sharing of such knowledge, data access, or connections to practice, all of which supports research faculty in their research agenda.[19] For instance,
MSI's Young Scholars program helps juniors develop a strong cohort. They get more invited to talks, it gets them the opportunities to be recruited, and it starts research collaborations. (Former executive director of MSI)
I think institutions such as MSI or the ISBM can facilitate research that has both academic rigor and has got practical merit. (Former director of a cohesion institution that bridges academia and practice)
At every conference, we have a panel of practitioners and a practitioner speaker. And people like that. (Cofounder of a cohesion institution that bridges academia and practice)
What I found interesting about the "7 Big Problems in Marketing" work at the AMA is that we were really trying to get at which problems practitioners today are facing, or that we see coming,...and those were defined kind of collectively between academics and practitioners. (Former global chief marketing officer of a large multinational technology corporation)
In my last trip [to an MSI meeting], in San Francisco right before COVID-19, there was a cocktail [hour] where we had various academics explaining their research with whiteboards; we could walk out and talk about their research, et cetera....It was really interesting. (Current EVP at one of the world's largest market research firms)
Third, nearly all external stakeholders expressed a more negative view on the extent to which business school faculty is achieving q-quality in research (vs. the extent to which it is achieving r-quality in research), consistent with the insights we obtained from interviewing the (associate) deans:
[Academic research in business schools] feels like a small set of people speaking to each other about something that nobody cares about. I may be a little harsh here, but it is often not applicable to the kind of problems I see. (Current senior executive at a cohesion institution that bridges academia and practice)
I think there is a stereotype we have on our side is that academic research is "out of touch" with reality. (Current EVP at one of the world's largest market research firms)
Academic research is highly differentiating but not necessarily as relevant....And obviously, the two things are easily at odds....If you are highly relevant, you are not "different." And I think that's the challenge. Does academic research want to be more relevant? Or does it want to maintain its differentiation? Because while it clearly is rigorous, it is largely unassailable, I would say, to the business community. (Current CEO at a leading market research firm in the United States)
Most of my peers in the business functions [marketing, strategy, and corporate reputation] would only look at academic research if it was sort of quoted in the context of another business story. Our analytics folks will definitely go deeper into academic papers, specifically if it's helping them. (Current CEO at a leading market research firm in the United States)
Our results have three main implications for business schools and the research faculty they employ. In these implications, we embed several conjectures that can provide fertile ground for future research to provide empirical testing.
Research monitoring instruments in business schools are, on average, badly designed. Low-effort metrics, such as the number of publications, receive too much weight in faculty evaluation, whereas effortful metrics such as creativity, literacy, relevance to nonacademic audiences, and awards receive too little weight. Business schools with badly designed monitoring instruments perform worse on (r- and q-) quality of research than business schools with well-designed monitoring instruments. Business schools need to develop better research metrics. Business schools that take this message to heart could consider multiple pathways.
First, business schools could devote more effort to otherwise low-effort metrics to make them more informative. For instance, schools can correct aggregate publication counts for journal status. Journal of Consumer Research, Journal of Marketing, Journal of Marketing Research, and Marketing Science are journals that publish, on average, higher-quality articles than other journals in marketing (according to the UTD list, which is the most stringent list on quality). Alternatively, schools could correct aggregate citation counts for ( 1) whether a scholar's highly cited papers were original contributions in premier journals or review articles in secondary journals, ( 2) whether a scholar's articles are consistently in the top 20% cited papers or bottom 20% cited papers of a journal, ( 3) whether a scholar's top five or top ten cited articles were published in premier or secondary journals, and ( 4) whether a scholar's work is mainly cited by papers in premier or secondary journals.
Second, business schools could consider low-effort metrics such as the number of publications or citations only as a starting point for faculty evaluation rather than an end point. For instance, for citations, it would be meaningful to rank a professor's work according to Web of Science citations, after which the five highest-ranked articles are assigned for reading to a committee, which assesses the r- and q-quality of the respective five papers after reading them. Ideally, these committees would provide thorough evaluations of the work, rather than a mere summary. One (associate) dean also told us about the practice of assigning discussants on specific papers of a candidate up for a promotion and tenure (P&T) evaluation to stimulate reading and evaluation. Instead of scientometrically picking the best three to five papers for reading, schools could also ask the candidate to pick three to five of their best papers and ensure that evaluators read and discuss those papers.
Third, business schools could add creativity and literacy of scholarly work to the evaluation process, piggybacking on recent work enabling their reliable and valid measurement.[20] Business schools could also improve creativity training and coaching of doctoral students and young faculty ([51]). Innovation management as a field has shown that creativity, ideation, idea development, are all processes that can be trained with tools; doctoral students and young faculty could be trained on such tools (for examples of such tools, see www.frisbuss.com).
Fourth, business schools could make the system of reference letters used for P&T decisions more effective by ( 1) providing a cohort list to which the candidate should be compared, ( 2) making evaluation criteria such as creativity and literacy explicit, and ( 3) involving a more heterogeneous set of letter writers. To prevent gaming of cohort lists, schools could decide on a universal set of reference schools, such as the 10–20 schools that perform similarly or a little better on the FT overall or UTD research rankings. The cohort for a specific candidate in a P&T process could consist of two types of faculty members of the reference schools: ( 1) all research faculty with a similar "time since doctoral degree" (e.g., ±1–2 years) and ( 2) all faculty of the same rank for which the candidate is considered who received their doctorate no more than five years prior to the candidate. To source letter writers, business schools could ( 1) source academic experts from the entire discipline across silos, instead of purely from the silo to which the candidate belongs and ( 2) allow nonacademics (e.g., alumni, students, professionals) to write letters, as we observed in one school we studied where a typical P&T package could have up to 50 letters.
Faculty members feel undercompensated, whereas several (associate) deans feel they are overcompensated for the research they do. Business schools where faculty feel more appropriately compensated perform better on r-quality of research than business schools where faculty feel less appropriately compensated. Business schools that aim to improve the alignment with their faculty on compensation can do so in multiple ways.
First, business schools could give faculty a better understanding of the entire organization, its operations, and its finances. Some schools have a well-developed habit of organizing faculty meetings where they transparently cover all aspects of the school's business. In one of the business schools we studied, faculty meetings periodically cover the school's income statement, sales forecasts, and balance sheet to increase faculty's understanding of the economics of the school. Other schools do not share—or purposefully hide—financials, which prohibits the faculty from seeing their salary and contribution in the context of the bigger picture.
Second, business schools could showcase what staff, administrators, (associate) deans, and other senior faculty do on a day-to-day basis to improve the school's health. We have seen "a day in the life of..." presentations by deans to give faculty a better idea of what kinds of internal and external pressures they are facing. Transparency on such direct contributions to the health of the school may put the research accomplishments of a research faculty member (such as another Journal of Marketing or Journal of Marketing Research publication being freshly accepted) into perspective.
Third, business schools could promote teamwork and collaboration among faculty within the same school, fostering a high-commitment environment. Such collaborations may stimulate the faculty's emotional identification with the school. While considering such promotion, schools also need to put checks in place against undesirable practices, such as forcing people into collaborations, free-riding in collaborations, or junior faculty trading in coauthorships for political or teaching support, often from senior faculty, among others.
Fourth, business schools could increase the leverage over faculty to ensure that their research faculty meet the outside world also from a compensation perspective. Specifically, we believe that business school professors would benefit from practicing in their professional area just as medical school professors benefit from seeing patients or law professors benefit from assisting in writing and enforcing legislation, practicing law, or performing expert witness services. Outside activity by professors would also give them an outside valuation on their time. Such external valuation could ( 1) bring the compensation demanded from the school more in line with actual valuation by external stakeholders and ( 2) complement the pecuniary reward from the school, lowering the faculty's dependency on the school's paycheck.
Research r-quality is a stronger driver of business school research health than research quantity. Compared with research quantity, research r- and q-quality are stronger drivers of business school health dimensions other than business school research health. Research quantity can even negatively affect external support. The (associate) deans report that the business schools they lead have made more progress on r-quality than on q-quality and that they are concerned about a further decline in q-quality in recent years. This viewpoint is shared by the external stakeholders we interviewed. Business schools that want to improve the r-quality and/or q-quality of their faculty's research can do so in multiple ways.
First, business schools could focus audits of their research activities more on quality than on quantity. Business schools that want to increase r-quality could investigate whether their metrics sufficiently reward quality, whether they allocate research money sufficiently based on quality, and whether its faculty is sufficiently represented on the Editorial Review Boards of the best journals in the field. Business schools that want to increase q-quality could investigate whether the school sufficiently stimulates consulting by faculty high in r-quality (as recommended in [43]] and [52]]), whether research centers fundamentally engage with practice or are mostly "lipstick on a pig" (as one our interviewed associate deans put it), whether research faculty high in r-quality teach in executive MBA or open and custom programs (which provide more socialization with practice than undergraduate or daytime MBA programs), and whether the portfolio of research faculty profiles is balanced sufficiently both on r-quality and q-quality.[21]
Second, business schools could consider complementing internal audits (e.g., of a multidepartment committee chaired by the research dean) with external audits by a panel of outside faculty with outstanding research records, preferably on both r- and q-quality, and with a good understanding of business school health. For schools that have not done a research audit for a while, these findings and suggestions could stimulate them to organize such audits. For schools that already perform such audits regularly, our findings indicate that the aforementioned topics should make such audits more impactful and focused on today's major challenges of business schools.
Third, business schools could benchmark their experiences with those of successful business schools, or role models, which can serve as yardsticks for improving their research faculty incentive systems. Role models help clarify an "aspiration gap" (i.e., the difference between a level of performance that one aspires to achieve and the level of performance that one already has). Moreover, different business schools have different aspiration levels and thus place different weights across different dimensions of the research task they want to optimize. Thus, each business school should benchmark its faculty research incentive system with that of a weighted combination of other business schools chosen to generate a "synthetic role model" that closely resembles the performance that the school aspires on research quantity, r-quality, and q-quality. As an illustration of the usage of these "synthetic role models," we present, in Table 5, three stylized synthetic role models that may serve as inspiration for schools aiming to increase their performance in research quantity (SRM-Qty), r-quality (SRM-R), or q-quality (SRM-Q).
Graph
Table 5. Synthetic Role Models According to a School's Leading Research Task Optimization Goal.
| SRM-Q:Stimulate q-quality | This school values thought leadership in a substantive area. Therefore, the relevance of faculty research to nonacademics is greatly appreciated. Its faculty publishes strong dual impact contributions. Faculty in these schools are typically leading expert witnesses, leading consultants, or (co)founders of firms that are spin-offs of their academic work, in addition to their professor duties. The faculty's work shows high creativity and is recognized by awards from academic and nonacademic endorsement institutions that recognize relevance (e.g., INFORMS Buck Weaver Award). In faculty assessment committees, committee members evaluate and appreciate "translational" publications (e.g., books or publications in practitioner-oriented outlets). P&T decisions are extremely selective. There are no bonuses because outstanding performance is expected as a regular duty. While faculty have very high academic freedom, it is bounded by very strong professional expectations. |
| SRM-R:Stimulate r-quality | This school almost exclusively values publications in journals that are recognized as leading in their respective fields. For tenure, only publications in these journals count. Highly cited papers are seen as "home runs" if they exemplify an original contribution. Best paper awards from top journals and awards from leading endorsement institutions are "hard currency." Editorship of leading journals is considered strong service to the department and to the school and often receives teaching credit. Faculty are expected to take leadership positions and participate in steering committees in academic cohesion institutions (e.g., Association of Consumer Research, INFORMS). Committees that assess faculty go beyond counting the number of top journal publications and read the work of the candidate in detail. Faculty receive salary increases and promotions as their scientific prestige increases among the leading scholarly international community. Academic freedom is a fundamental reward. Receiving tenure is the ticket for such intellectual freedom but is not given to many in a selective promotion and tenure system. Reference letters are typically asked from highly prolific scholars in top journals who are completely independent from the candidate (no coauthors, no supervisor-student relationships, etc.) |
| SRM-Qty:Stimulate research quantity | This school frequently measures the number of publications of its faculty, typically weighted according to journals' standing in the Thomson Institute for Scientific Information's quantiles, or by standards in the field (e.g., A journals). P&T decisions are not very selective, with promotion and tenure typically occurring as soon as a candidate crosses a clear quantity cutoff. The metric counting system is very well developed into a protocol and is very clearly monitored by committees, such as P&T committees. This monitoring system is externally audited by specialized institutions or external review committees often composed of leading international scholars. Faculty are expected to join academic cohesion institutions, which are valued because they enrich faculty's collaboration networks. Faculty get sufficient dedicated research time and are assigned a personal research budget that is sufficient to execute their research. Faculty are rewarded in terms of career progression according to the amount of research that they publish. Faculty receive a salary that is based on research output. The teaching load varies according to publication volume. |
Fourth, faculty could consciously strengthen the cohesion institutions that support the promotion of socialization with practitioners (e.g., AMA, MSI, Theory + Practice in Marketing) and business schools could encourage and support such efforts. Within such cohesion institutions, faculty could stimulate action that increases q-quality of research of high r-quality. For instance, institutions such as MSI could give fewer, larger grants, possibly assigning a corporate sponsor to steer such larger grants, or grant funding only to research teams that combine academics and practitioners. Under its present organizational structure (the senior leadership team being fully composed solely of practitioners), the AMA has failed to make the connection between academics and practitioners (as noted by the representative from the AMA we interviewed). Business school marketing faculty could aid in building a new model within the AMA.
Several limitations of this article may give rise to future research. First, our empirical evidence is self-reported from a survey with research faculty, interviews with (associate) deans and interviews with representatives of external institutions. While self-reports enable us to cover a broad set of topics, each of the relationships we establish could potentially fuel secondary data research. Several secondary data studies (e.g., [34]; [39]) have examined the effect of research on teaching, but none have examined the effect of research on other business school health dimensions, such as external support or institutional integrity, all of which could be gauged by secondary data also (e.g., endowment statistics, online chatter of student communities). Future research should also better examine other constituents' perceptions of business school health (e.g., students, recruiters, donors, alumni).
Second, our conceptual derivation and empirical evidence only limitedly exposes the causal mechanisms at work. In fact, we have been prudent throughout the article to clearly identify instances where our data permits us only to offer logical conjectures and to claim correlation rather than causation. Thus, future research that goes from correlation to causation would be very fruitful; it could also document more precisely the nature of the feedback mechanisms that we introduced. Future research could also more elaborately document the behavioral mechanisms in place that lead business schools to excessively monitor numbers and insufficiently monitor creativity or literacy. One can conceive behavioral experiments with academic assessors on research metrics, how people use them, and under which conditions decisions can be (de)biased.
Third, we explored the variance in incentive misalignment across schools on a limited number of school descriptors. Research could easily expand on a larger set of school descriptors. For instance, do the effects we study depend on whether the school offers executive education, where the school is located (United States vs. international), whether the school is private or public, or how high the tuition fees are that it is charging?
Fourth, we took a step beyond our empirical inquiry to conceptualize what business schools could do to positively affect the present state of affairs. Some of the recommendations we gave seem easy to implement, whereas others are more difficult and would benefit from a more elaborate conceptualization than the length and scope of this article allow. For instance, how can business schools create a stronger sense of common purpose among its faculty such that the faculty is less self-interest seeking? Alternatively, how can business schools favor more reading and less counting? How can they better monitor creativity and literacy? The latter question can also fuel scientometric research to address some of the alternative metrics we suggest.
Despite these limitations, we feel that we have made a significant contribution to understanding the role of faculty research in business school health. At the very least, we hope that we have sparked a dialogue to get more (marketing) faculty and business school administrators to rethink how academic research can make business schools healthier.
Supplemental Material, sj-docx-1-jmx-10.1177_00222429211001050 - Faculty Research Incentives and Business School Health: A New Perspective from and for Marketing
Supplemental Material, sj-docx-1-jmx-10.1177_00222429211001050 for Faculty Research Incentives and Business School Health: A New Perspective from and for Marketing by Stefan Stremersch, Russell S. Winer and Nuno Camacho in Journal of Marketing
Footnotes 1 Leigh McAlister
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The authors received financial support for this research from Erasmus School of Economics (Erasmus University Rotterdam) and Stern School of Business (New York University).
4 https://doi.org/10.1177/00222429211001050
5 Table 1 lists the most prominent articles that have appeared in the journals indexed by the University of Texas at Dallas (UTD) Research Ranking that covered the role of faculty research in business schools. It does not include articles focusing primarily on scientometric properties of research (e.g., [53]; Stremersch, Verniers, and Verhoef 2007).
6 For brevity, in our theorizing we refer to the "faculty research incentive system" as "incentive instruments." We treat both terms as synonyms. We focus solely on business schools' incentives for agents' research task. Therefore, we do not examine incentives for other tasks of these agents (e.g., the teaching task).
7 Social monitoring may occur when the business school appeals to external institutions (e.g., by considering whether the agent received or has been a finalist in external institutions' awards) to monitor agents. Social monitoring may also occur when the business school appeals to peers of the agent at other business schools (e.g., to write reference letters for faculty promotion), or to peers of senior administrators (e.g., when deans evaluate each other in business school rankings and Association to Advance Collegiate Schools of Business reviews). We represent such social monitoring with two left-right arrows in the center and two bottom-up arrows at the bottom of Figure 1.
8 [2] raise three other types of institutions that are responses to market failure issues that have not been connected to principal–agent theory and do not seem relevant in our context.
9 By "literacy," we mean how well-read a scholar is (i.e., the extent to which they have access to a large knowledge base, in line with the [6], p. 2) definition of "information literacy": a person's "ability to locate, evaluate, and use effectively the needed information." We provided this definition in our faculty survey.
Performance-based salary increases are permanent compensation increases, whereas publication bonuses are one-time compensation increases and, thus, may have different effects on agents' behavior.
The distinction between publication bonuses paid as salary supplements versus paid as supplementary research budget is important because it taps into the classic distinction between pecuniary and nonpecuniary rewards. Specifically, even though publications bonuses paid as supplementary research budget are monetary in nature, the benefits that a research faculty member derives from such bonuses are nonpecuniary (e.g., easier access to data and equipment, higher travel allowances to visit conferences)
Note that all hypotheses are formulated ceteris paribus.
We develop our theorizing for each of the dimensions of business school health, instead of merely at the overall level. We do so for two reasons: (1) we consider the seven dimensions to be noncompensatory and (2) the antecedents we study may have different effects across the different dimensions. Note that we consider business school health to be a superordinate label, instead of a formative construct, and the seven subordinate dimensions as facets that collectively define it ([16]).
"Frisbuss" stands for Faculty Research Incentive Systems in Business Schools.
The eigenvalue criterion suggested a six-component structure combining the items of institutional integrity and leadership support. However, the seventh component had an eigenvalue very close to 1 (.93) and confirmatory factor analyses (discussed subsequently) showed that the seven-component solution fits the data better.
We also ran a factor model with a latent business school health factor as second-order construct and the seven dimensions of business school health as first-order constructs. This second-order factor model had a worse fit than the first-order factor model according to all indices (CFI =.96, NNFI =.96, RMSEA =.06, SRMR =.19).
The MAD is a proxy for the extent to which a respondent perceives the mix of incentive instruments at their business school as improperly weighted, as it gives us the average absolute deviations from the mid-point of the scales indicating whether a given instrument receives the "right weight."
We use Fisher–Hayter's procedure because it has more power compared with other post hoc comparison methods such as Tukey's test. Note that Fisher's least significant difference (LSD) also has more power than Tukey's test, but it does not correct for multiple comparisons, which may inflate Type I error. Fisher–Hayter's test is a revised version of the LSD test proposed by Hayter to overcome the weaknesses of the LSD test.
Note that cohesion institutions often go beyond supporting their members' research agendas and help with agenda setting (e.g., MSI Research Priorities).
To measure literacy, business schools may, for example, evaluate the quality of a scholar's bibliographies and citation practices ([37]). To measure creativity, business schools may measure the extent to which an author's citations contain atypical combinations of prior work ([59]).
One way to screen candidates on their potential to produce high q-quality research may be to include practitioners in the search committees for new faculty.
References Agarwal Rajshree , Ohyama Atsushi. (2013), " Industry or Academia, Basic or Applied? Career Choices and Earnings Trajectories of Scientists ," Management Science , 59 (4), 950 – 70.
Ahuja Gautam , Yayavaram Sai. (2011), " Perspective—Explaining Influence Rents: The Case for an Institutions-Based View of Strategy ," Organization Science , 22 (6), 1631 – 52.
Akerlof George A.. (2020), " Sins of Omission and the Practice of Economics ," Journal of Economic Literature , 58 (2), 405 – 18.
AMA Task Force on the Development of Marketing Thought (1988), " Developing, Disseminating, and Utilizing Marketing Knowledge ," Journal of Marketing , 52 (4), 1 – 25.
Amabile Teresa M.. (1998), " How to Kill Creativity ," Harvard Business Review , 76 (5), 76 – 87.
American Library Association (2000), Information Literacy Competency Standards for Higher Education. Chicago, IL : Association of College & Research Libraries.
Bagozzi Richard P. , Yi Youjae. (1988), " On the Evaluation of Structural Equation Models ," Journal of the Academy of Marketing Science , 16 (1), 74 – 94.
Bartunek Jean M.. (2007), " Academic-Practitioner Collaboration Need Not Require Joint or Relevant Research: Toward a Relational Scholarship of Integration ," Academy of Management Journal , 50 (6), 1323 – 33.
Baumeister Roy F. , Exline Julie Juola. (1999), " Virtue, Personality, and Social Relations: Self-Control as the Moral Muscle ," Journal of Personality , 67 (6), 1165 – 94.
Benbasat Izak , Zmud Ron W.. (1999), " Empirical Research in Information Systems: The Practice of Relevance ," MIS Quarterly , 23 (1), 3 – 16.
Bennis Warren G. , O'Toole Jim. (2005), " How Business Schools Have Lost Their Way ," Harvard Business Review , 83 (5), 96 – 104.
Besancenot Damien , Faria Joao Ricardo , Vranceanu Radu. (2009), " Why Business Schools Do So Much Research: A Signaling Explanation ," Research Policy , 38 (7), 1093 – 1101.
Bettis Richard A.. (2012), " The Search for Asterisks: Compromised Statistical Tests and Flawed Theories ," Strategic Management Journal , 33 (1), 108 – 13.
Bradlow Eric T.. (2008), " Editorial: Enticing and Publishing the Home Run Paper ," Marketing Science , 27 (1), 4 – 6.
Cole Stephen , Cole Jonathan R.. (1967), " Scientific Output and Recognition: A Study in the Operation of the Reward System in Science ," American Sociological Review 32 (3) 377 – 90.
Edwards Jeffrey R.. (2011), " The Fallacy of Formative Measurement ," Organizational Research Methods , 14 (2), 370 – 88.
Ellison Glenn. (2002), " Evolving Standards for Academic Publishing: A q-r Theory ," Journal of Political Economy , 110 (5), 994 – 1034.
Fornell Claes , Larcker David F.. (1981), " Evaluating Structural Equation Models with Unobservable Variables and Measurement Error ," Journal of Marketing Research , 18 (1), 39 – 50.
Frey Bruno S. , Jegen Reto. (2001), " Motivation Crowding Theory ," Journal of Economic Surveys , 15 (5), 589 – 611.
Gerbing David W. , Anderson James C.. (1988), " An Updated Paradigm for Scale Development Incorporating Unidimensionality and Its Assessment ," Journal of Marketing Research , 25 (2), 186 – 92.
Gomez-Mejia Luis R. , Balkin David B.. (1992), " Determinants of Faculty Pay: An Agency Theory Perspective ," Academy of Management Journal , 35 (5), 921 – 55.
Gulati Ranjay. (2007), " Tent Poles, Tribalism, and Boundary Spanning: The Rigor-Relevance Debate in Management Research ," Academy of Management Journal , 50 (4), 775 – 82.
Holmstrom Bengt , Milgrom Paul. (1991), " Multitask Principal-Agent Analyses: Incentive Contracts, Asset Ownership, and Job Design ," Journal of Law, Economics & Organization , 7 , 24 – 52.
Homburg Christian , Klarmann Martin , Reimann Martin , Schilke Oliver. (2012), " What Drives Key Informant Accuracy? " Journal of Marketing Research , 49 (4), 594 – 608.
Hoy Wayne K. , Tarter C. John , Kottkamp Robert B.. (1991), Open Schools/Healthy Schools: Measuring Organizational Climate. Beverly Hills, CA : SAGE Publications.
Jaworski Bernard J.. (2011), " On Managerial Relevance ," Journal of Marketing , 75 (4), 211 – 24.
Joseph Kissan , Thevaranjan Alex. (1998), " Monitoring and Incentives in Sales Organizations: An Agency-Theoretic Perspective ," Marketing Science , 17 (2), 107 – 23.
Kaplan Robert S.. (2011), " Accounting Scholarship that Advances Professional Knowledge and Practice ," Accounting Review , 86 (2), 367 – 83.
Lehmann Don R. , McAlister Leigh , Staelin Richard. (2011), " Sophistication in Research in Marketing ," Journal of Marketing , 75 (4), 155 – 65.
Lightfield E. Timothy. (1971), " Output and Recognition of Sociologists ," American Sociologist , 6 (2), 128 – 33.
Lilien Gary L.. (2011), " Bridging the Academic-Practitioner Divide in Marketing Decision Models ," Journal of Marketing , 75 (4), 196 – 210.
Mael Fred , Ashforth Blake E.. (1992), " Alumni and Their Alma Mater: A Partial Test of the Reformulated Model of Organizational Identification ," Journal of Organizational Behavior , 13 (2), 103 – 23.
McFarlin Dean B. , Baumeister Roy F. , Blascovich Jim. (1984), " On Knowing When to Quit: Task Failure, Self-Esteem, Advice, and Nonproductive Persistence ," Journal of Personality , 52 (2), 138 – 55.
Mitra Debanjan , Golder Peter N.. (2008), " Does Academic Research Help or Hurt MBA Programs? " Journal of Marketing , 72 (5), 31 – 49.
Nielson Daniel L. , Tierney Michael J.. (2003), " Delegation to International Organizations: Agency Theory and World Bank Environmental Reform ," International Organization , 57 (2), 241 – 76.
Nosek Brian A. , Alter George , Banks George C. , Borsboom Denny , Bowman Sara D. , Breckler Steven J.. (2015), " Promoting an Open Research Culture ," Science , 348 (6242), 1422 – 25.
Oakleaf Megan. (2009), " Using Rubrics to Assess Information Literacy: An Examination of Methodology and Interrater Reliability ," Journal of the American Society for Information Science and Technology , 60 (5), 969 – 83.
Parsons Talcott. (1951). The Social System. Glencoe, IL : The Free Press.
Pfeffer Jeffrey , Fong Christina T.. (2002), " The End of Business Schools? Less Success Than Meets the Eye ," Academy of Management Learning & Education , 1 (1), 78 – 95.
Podsakoff Phillip M. , MacKenzie Scott B. , Lee Jeong-Yeon , Podsakoff Nathan P.. (2003), " Common Method Biases in Behavioral Research: A Critical Review of the Literature and Recommended Remedies ," Journal of Applied Psychology , 88 (5), 879 – 903.
Reibstein David J. , Day George , Wind Jerry. (2009), " Guest Editorial: Is Marketing Academia Losing Its Way? " Journal of Marketing , 73 (4), 1 – 3.
Rindfleisch Aric , Malter Alan J. , Ganesan Shankar , Moorman Christine. (2008), " Cross-Sectional Versus Longitudinal Survey Research: Concepts, Findings, and Guidelines ," Journal of Marketing Research , 45 (3), 261 – 79.
Roberts John H. , Kayande Ujwal , Stremersch Stefan. (2014), " From Academic Research to Marketing Practice: Exploring the Marketing Science Value Chain ," International Journal of Research in Marketing , 31 (2), 127 – 40.
Rosemann Michael , Vessey Iris. (2008), " Toward Improving the Relevance of Information Systems Research to Practice: The Role of Applicability Checks ," MIS Quarterly , 32 (1), 1 – 22.
Rynes Sara L. , Bartunek Jean M. , Daft Richard L.. (2001), " Across the Great Divide: Knowledge Creation and Transfer Between Practitioners and Academics ," Academy of Management Journal , 44 (2), 340 – 55.
Rynes Sara L. , Giluk Tamara L. , Brown Kenneth G.. (2007), " The Very Separate Worlds of Academic and Practitioner Periodicals in Human Resource Management: Implications for Evidence-Based Management ," Academy of Management Journal , 50 (5), 987 – 1008.
Sadoski Mark , Goetz Ernest T. , Fritz Joyce B.. (1993), " Impact of Concreteness on Comprehensibility, Interest, and Memory for Text: Implications for Dual Coding Theory and Text Design ," Journal of Educational Psychology , 85 (2), 291 – 304.
Shapiro Debra L. , Kirkman Bradley L. , Courtney Hugh G.. (2007), " Perceived Causes and Solutions of the Translation Problem in Management Research ," Academy of Management Journal , 50 (2), 249 – 66.
Shapiro Susan. (2005), " Agency Theory ," Annual Review of Sociology , 31 , 263 – 84.
Shugan Steven M. , Mitra Debanjan. (2009), " Metrics—When and Why Nonaveraging Statistics Work ," Management Science , 55 (1), 4 – 15.
Stewart David W.. (2020), " Creativity and Publication in Marketing ," AMS Review , 10 , 65 – 72.
Stremersch Stefan. (2021), " Commentary on Kohli & Haenlein: The Study of Important Marketing Issues: Reflections ," International Journal of Research in Marketing , 38 (1), 12 – 17.
Stremersch Stefan , Camacho Nuno , Vanneste Sofie , Verniers Isabel. (2015), " Unraveling Scientific Impact: Citation Types in Marketing Journals ," International Journal of Research in Marketing , 32 (1), 64 – 77.
Stremersch Stefan , Verniers Isabel , Verhoef Peter C.. (2007), " The Quest for Citations: Drivers of Article Impact ," Journal of Marketing , 71 (3), 171 – 93
Thompson Craig J. , Locander William B. , Pollio Howard R.. (1989), " Putting Consumer Experience Back into Consumer Research: The Philosophy and Method of Existential-Phenomenology ," Journal of Consumer Research , 16 (2), 133 – 46.
Trieschmann James S. , Dennis Alan R. , Northcraft Gregory B. , Niemi Albert W. Jr. (2000), " Serving Constituencies in Business Schools: MBA Program versus Research Performance ," Academy of Management Journal , 43 (6), 1130 – 41.
Trope Yaacov , Liberman Nira. (2010), " Construal-Level Theory of Psychological Distance ," Psychological Review , 117 (2), 440 – 63.
Tushman Michael , O'Reilly Charles A. III. (2007), " Research and Relevance: Implications of Pasteur's Quadrant for Doctoral Programs and Faculty Development ," Academy of Management Journal , 50 (4), 769 – 74.
Uzzi Brian , Mukherjee Satyam , Stringer Michael , Jones Ben. (2013), " Atypical Combinations and Scientific Impact ," Science , 342 (6157), 468 – 72.
Vermeulen Freek. (2005), " On Rigor and Relevance: Fostering Dialectic Progress in Management Research ," Academy of Management Journal , 48 (6), 978 – 82.
Vermeulen Freek. (2007), " I Shall Not Remain Insignificant: Adding a Second Loop to Matter More ," Academy of Management Journal , 50 (4), 754 – 61.
~~~~~~~~
By Stefan Stremersch; Russell S. Winer and Nuno Camacho
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 54- Fields of Gold: Scraping Web Data for Marketing Insights. By: Boegershausen, Johannes; Datta, Hannes; Borah, Abhishek; Stephen, Andrew T. Journal of Marketing. Sep2022, Vol. 86 Issue 5, p1-20. 20p. 1 Diagram, 4 Charts, 1 Graph. DOI: 10.1177/00222429221100750.
- Database:
- Business Source Complete
Fields of Gold: Scraping Web Data for Marketing Insights
Marketing scholars increasingly use web scraping and application programming interfaces (APIs) to collect data from the internet. Yet, despite the widespread use of such web data, the idiosyncratic and sometimes insidious challenges in its collection have received limited attention. How can researchers ensure that the data sets generated via web scraping and APIs are valid? While existing resources emphasize technical details of extracting web data, the authors propose a novel methodological framework focused on enhancing its validity. In particular, the framework highlights how addressing validity concerns requires the joint consideration of idiosyncratic technical and legal/ethical questions along the three stages of collecting web data: selecting data sources, designing the data collection, and extracting the data. The authors further review more than 300 articles using web data published in the top five marketing journals and offer a typology of how web data have advanced marketing thought. The article concludes with directions for future research to identify promising web data sources and embrace novel approaches for using web data to capture and describe evolving marketplace realities.
Keywords: web scraping; application programming interface; API; crawling; validity; user-generated content; social media; big data
The accelerating digitization of social and commercial life has created an unprecedented number of digital traces of consumer and firm behavior. Every minute, users worldwide conduct 5.7 million searches on Google, make 6 million commercial transactions, and share 65,000 photos on Instagram ([76]). The resulting web data—enormous in size, diverse in form, and often publicly accessible on the internet—is a potential goldmine for marketing scholars who want to quantify consumption, gain insights on firm behavior, and track social activities difficult or costly to observe otherwise. The importance of web data for marketing research is reflected in a growing number of impactful publications across all methodological traditions, including consumer culture theory, consumer psychology, empirical modeling, and marketing strategy.
Researchers can use web scraping and application programming interfaces (APIs) to efficiently collect web data at scale. Web scraping is the process of developing software to automatically collect information displayed in a web browser. For example, researchers can scrape Amazon's website to construct data sets of online consumer reviews. Because many websites and web apps are publicly accessible, data sets can be generated without involving data providers. In contrast, some data providers also offer APIs for programmatic access to their internal databases. For example, scholars can apply for academic research access to retrieve data from the Twitter API. Researchers can also access a wide range of algorithms via APIs. For instance, Google offers advanced image and video recognition through its Cloud Vision API (for additional examples and explanations, see Table W1 in Web Appendix A).
Data extracted from the internet, at first sight, might resemble other organically generated data sets that address related research questions (e.g., a firm's clickstream data). Yet, collecting web data for academic use in a highly automated manner may prompt a set of novel and sometimes insidious validity challenges. Validity concerns may arise from, among others, ( 1) failing to capture contextual information in a rapidly changing environment (e.g., updates to the website's data-generating process), ( 2) not sufficiently aligning the psychological processes of interest with the frequency of data extraction on review platforms (e.g., the collected information does not capture the time when the behavior occurred), ( 3) overlooking the influence of algorithmic interference on e-commerce websites (e.g., the effect of personalization algorithms on information display), or ( 4) failing to retain raw website or API data necessary for construct validation, sampling, and analysis.
Against this background, this article makes three interlinked contributions. First, we develop a methodological framework that highlights how addressing validity concerns arising from web scraping and APIs requires the joint consideration of idiosyncratic technical and legal/ethical concerns. Within marketing, guidance exists for collecting web data in the consumer culture theory research tradition, particularly using netnography (e.g., [46], [47]). A handful of articles address selected challenges that occur during the automatic extraction of web data (e.g., sampling; [39]). Outside of marketing, tutorials and books primarily focus on technical details for the automatic extraction of web data (see Table W2 in Web Appendix B). Yet, neither these resources nor methodological articles in other disciplines (e.g., [19]; [52]) address the broad spectrum of validity concerns arising from the automatic collection of web data for academic use. It is this void that our methodological framework fills. In discussing the methodological framework, we offer a stylized marketing example for illustration and provide recommendations for addressing challenges researchers encounter during the collection of web data via web scraping and APIs.
Second, despite the use of web data in marketing for two decades, no systematic review reflects on how it has and could advance marketing thought. Importantly, understanding the richness and versatility of web data is invaluable for scholars curious about integrating it into their research programs. To offer these insights, we have systematically reviewed more than 300 articles in the top five marketing journals across two decades that have used web data. We leverage our coding to reveal which web sources have been considered and how data have been extracted. The resulting typology of web data may spark the imagination of researchers interested in generating new marketing insights from web data.
Finally, we use our methodological framework and typology to unearth new and underexploited "fields of gold" associated with web data. We seek to demystify the use of web scraping and APIs and thereby facilitate broader adoption of web data across the marketing discipline. Our future research section highlights novel and creative avenues of using web data that include exploring underutilized sources, creating rich multisource data sets, and fully exploiting the potential of APIs beyond data extraction. We particularly highlight the value of web scraping and APIs for research streams that have not yet embraced them at scale.
In what follows, we provide an overview of the use of web data in marketing and document four pathways through which web data have advanced marketing thought. We then introduce our methodological framework to help researchers make sensible design decisions when automatically extracting web data. We conclude with directions for future research.
Across the top five marketing journals, marketing researchers increasingly use information available on the internet. For example, the share of web data–based publications has more than tripled in the last decade, from about 4% in 2010 to 15% in 2020 (see the thick line in Figure 1). The growing use of web data has been fueled by its increased accessibility and associated time and cost savings. Most of the 313 identified web data–based articles rely on web scraping (59%); APIs are used much more sparingly (12%), and some articles combine web scraping and APIs (9%). The remaining articles—especially netnographic work—use web data but tend to extract it manually (20%). The median annual citation count of articles using web data is 7.55, compared with 3.90 for publications not using web data.
Graph: Figure 1. Increased use of web data in marketing (2001–2020).
Some of the earliest uses of web data in marketing can be attributed to the development of netnography to study online communities (e.g., [46], [45]). Subsequently, the first quantitative marketing scholars extracted web data at scale (e.g., [27]). Today, all subfields—including marketing strategy and consumer behavior—have embraced web data.
Online word of mouth and social media are the most prominent domains of inquiry using web scraping (see Table W3 in Web Appendix C). The most widely used data source in academic marketing research is Amazon (38 articles). Other prevalent sources are Twitter (30), IMDb (24), Facebook, and Google Trends (both 22; see Table W4 in Web Appendix C).
Via a comprehensive literature review, we next identify the four central pathways through which web data facilitate the creation of new knowledge in marketing.
Web data can boost the field's relevance by enabling marketing scholars to study novel phenomena. For example, initial work using web data focused on novel online phenomena that emerged at the beginning of this century, such as online conversations ([27]) and the impact of consumer reviews on sales ([14]). Web data are well suited to provide fertile grounds for inductive research to develop novel theories about emerging marketing phenomena (e.g., brand public; [ 4]).
Gathering data via web scraping or APIs often decreases the time between the occurrence of a marketplace phenomenon and the availability of data for academic research. This inherent timeliness of web data continues to be an essential lever for marketing scholars to advance our understanding of emerging substantive topics such as the sharing economy (e.g., Airbnb; [93]), access-based business models (e.g., Spotify; [15]), and fake online content (e.g., [ 3]). More generally, web data enable researchers to weigh in on contemporary issues before any "conventional" data sets become available, such as measuring the effect of pandemic lockdown policies on consumption ([74]).
Web data can create knowledge by allowing researchers to move closer to marketing's "natural habitat" ([83]). Some of the most used web sources contain commercial outcome variables relevant to marketing stakeholders and are difficult or costly to collect otherwise. Examples are sales (e.g., The-Numbers.com), sales ranks (e.g., Amazon), online searches (e.g., Google Trends), and donations (e.g., contributions to a Kiva project).
As web data can be collected unobtrusively, they can effectively complement more controlled data collection methods. Using web data, researchers can demonstrate that focal psychological processes occur outside the confines of a controlled laboratory environment and stylized experimental stimuli ([64]). Consider, for instance, the controversy around the decoy effect ([37])—one of the most prominent context effects in consumer behavior. Using experiments, [23] questioned the robustness and practical relevance of the decoy effect. In response, [90] built a panel data set from an online diamond market using web data. Their work not only shows that the decoy effect emerges in a high-stakes setting but also, more importantly, reaffirms its practical significance by quantifying its profit implications for the diamond retailer.
Another benefit of using web data to boost ecological value is that they can often be collected without the data provider's direct involvement. Thereby, researchers can limit the interference of data suppliers or collaborating firms to ensure that the societal relevance of a particular research question is given precedence over business objectives (e.g., firms might be unwilling to share data about the tracking tools they use on websites; [81]). Further, using web data, researchers can ensure the publication of research findings, regardless of how palatable they are to the organizations that are being studied.
As much of the data produced by consumers and firms is inherently unstructured, extracting insights can be challenging ([88]). Thus, marketing researchers have leveraged web data for developing methods that deal with and extract insights from different types of unstructured data, such as textual, image, and video data. For instance, web data have fueled the rapid improvement of automated text analysis (see [ 6]) and the large-scale analysis of image and video content (e.g., [55]; [57]).
The availability of network data on the internet (e.g., friend or product networks), along with outcome variables (e.g., posts, likes, sales ranks), has further enabled the use and advancement of methods for analyzing networks (e.g., [67]). Given their wealth and richness, web data have also stimulated the development of novel methods that can complement or replace traditional marketing research methods (e.g., using user-generated content to construct accurate multidimensional scaling maps of brands; [65]).
Web data can advance marketing knowledge by allowing researchers to measure constructs more precisely and obtain more valid inferences. For example, the collection of adequate control variables is often difficult. To capture seasonality in purchase patterns across a wide range of geographical markets and calendar years, researchers have used APIs to construct continuous (vs. dichotomous) variables that accurately reflect national holidays (e.g., HolidayAPI; [16]). Web data also allow researchers to efficiently operationalize new measures at scale, such as weather conditions based on the location of users' IP addresses (e.g., Weather Underground; [53]).
Relative to non-web data sources, researchers can collect data on the behavior of many consumers and firms at higher frequencies ([ 1]). Such data enhance statistical power, enable identification of causal effects, and facilitate the examination of theoretically relevant variation within individuals over time (e.g., how various psychological distances shape review content for the same consumer; [35]) or how effects unfold over time (e.g., the impact of video elements on virality over time; [77]).
Table 1 presents a typology of the four central pathways through which web data have advanced marketing thought. The typology highlights web data-based studies that investigate key marketing constructs across different entities, from consumers to organizations and other marketing stakeholders. For example, [80] explored a new phenomenon (tweeting), focusing on consumers (i.e., their motivation to tweet). These pathways for knowledge creation from web data are not mutually exclusive. Combining different pathways might be particularly promising for making breakthrough contributions.
Graph
Table 1. How to Create Knowledge Using Web Data: A Typology.
| Effect on ... | Primary Pathways of Knowledge Creation Using Web Data |
|---|
| Pathway 1:Studying New Phenomena | Pathway 2:BoostingEcological Value | Pathway 3:Facilitating Methodological Advancement | Pathway 4: Improving Measurement |
|---|
| Consumers(e.g., social media use, consumer learning) | Toubia and Stephen (2013) test the motivations of users to contribute content to social media. | Sridhar and Srinivasan (2012) explore peer effects in evaluating online product reviews. | Huang (2019) studies how picture quality improves due to consumer learning. | Huang et al. (2016) exploit within-user variation to measure how psychological distances interact. |
| Organizations(e.g., sales and profits of firms, donations to nonprofits) | Chevalier and Mayzlin (2006) demonstrate the impact of online reviews on book sales. | Wu and Cosguner (2020) probe the prevalence and profit implications of decoy effects. | Netzer et al. (2012) mine user-generated content to identify market structures. | Datta et al. (2022) gather national holidays across 14 countries and 11 years to capture seasonality. |
| Other marketing stakeholders(e.g., market reaction of investors, public health outcomes) | Hermosilla, Gutiérrez-Navratil, and Prieto-Rodríguez (2018) examine how consumers' aesthetic preferences create biases in firms' hiring decisions. | Blaseg, Schulze, and Skiera (2020) examine whether consumers are protected against false price advertising claims on Kickstarter. | Tirunillai and Tellis (2012) develop novel online metrics based on user-generated content to predict stock returns. | Kim and KC (2020) explore the effect of ads for erectile dysfunction drugs on birth rates. |
1 Notes: The table highlights illustrative and diverse examples of web data–based studies and corresponding outcome variables, cross-tabulated by four pathways through which web data have advanced marketing thought (the columns) and three of the most studied actors in marketing research (the rows).
Next, we introduce our methodological framework, which outlines an approach for making design decisions that enhance the validity of web data collected via web scraping and APIs. Researchers interested in learning more about the technical details of automatically extracting web data can consult our curation of technical tutorials in Web Appendix B or the digital companion to this article (available at https://web-scraping.org), which features a searchable database of all marketing articles in the top five marketing journals using web data.
In automatically collecting web data using web scraping and APIs, researchers make seemingly innocuous design decisions. However, as we will show, these decisions often involve trade-offs about research validity, technical feasibility, and legal/ethical risks[ 5] that are not always apparent. How researchers resolve these trade-offs shapes the credibility of research findings by enhancing or undermining statistical conclusion validity, internal validity, construct validity, and external validity ([73]).
We develop a methodological framework to provide guidance for the automatic collection of web data using web scraping and APIs. Figure 2 offers a stylized view of this process involving three key stages—source selection, collection design, and data extraction. Researchers typically start with a broad set of potential data sources and eliminate some of them as a function of three key considerations—validity, technical feasibility, and legal/ethical risks. These three considerations appear in the corners of an inverted pyramid, with validity at the bottom to underscore its importance. Given the difficulty in projecting the exact characteristics of the final data set before it is collected, researchers often revisit these considerations as they design, prototype, and refine their data collection. Failure to resolve technical or legal/ethical issues might mean that web data cannot inform the research question meaningfully.
Graph: Figure 2. Methodological framework for collecting web data.
Our framework deliberately focuses on collecting web data rather than its subsequent analysis. Analyzing web data involves many familiar methodological challenges encountered with organically generated data (e.g., cleaning to remove erroneous data or create measures, selecting observations, addressing endogeneity). However, approaches for the valid collection of web data are not yet documented nor commonplace in marketing research.
The methodological framework—designed to guide the automatic extraction of web data at scale—is agnostic to research paradigms. It is applicable to both deductive (i.e., identifying compelling web data to test hypotheses) and inductive (i.e., observing interesting irregularities in web data to identify novel marketing concepts and/or novel relationships between constructs) approaches to theory building. We next highlight the idiosyncratic challenges encountered when collecting web data and summarize solutions to these challenges in Tables 2–4. For expositional clarity, we focus on web scraping in our text.[ 6] To illustrate the key challenges encountered in designing the data collection, we gradually introduce a stylized marketing example involving the collection of book reviews from Amazon.
Graph
Table 2. Challenges and Solutions in Selecting Web Data Sources.
| Challenge #1.1: Exploring the Universe of Potential Web Sources |
| Reason for importance | As web sources vastly differ in quality, stability, and retrievability, researchers might be tempted to consider dominant or familiar platforms only. A thorough exploration of the data universe permits compelling theory testing and identifying novel, emerging marketing phenomena that may be difficult to notice otherwise. |
| Solutions and best practices | Assume the perspective of different stakeholders (e.g., consumers, analysts, managers) during the search process Browse through public API directories (e.g., ProgrammableWeb, GitHub) Broaden geographic search criteria (e.g., non-Western) Identify adjacent data sources (e.g., using Google Trend's "related search queries") Expand search to nonprimary data providers (i.e., aggregators, databases) Carefully vet the provider's description of relevant metrics Determine the conditions necessary to access data (e.g., requirement to log in on a website, creating an API key, subscribing to an API, possibility to signal academic status/scientific use) Verify whether it is possible to opt out of firm-administered experiments or whether the site is accessible without cookies Use the website or make some initial API requests to assess information availability (in the case of APIs, assess which authentication procedure is necessary to obtain data)
|
| Challenge #1.2: Considering Alternatives to Web Scraping |
| Reason for importance | Because web scraping is the most popular extraction method for web data, researchers may overlook alternative ways to extract data. APIs provide a documented and authorized way to obtain web data for many sources. Some sources also provide readily available data sets. Using such alternatives leads to time savings and minimizes exposure to legal risk. |
| Solutions and best practices | Expand search by explicitly including terms such as "API" or "data set download" Explore whether the source or third parties (e.g., public data platforms, researchers) offer data sets for download and assess their terms of use If a data source provides an API and a website, understand the differences in what data could be retrieved from them (e.g., by screening the API documentation) and how well the API can be accessed (e.g., using packages) Use stable versions of an API, and subscribe to a source's API support updates
|
| Challenge #1.3: Mapping the Data Context |
| Reason for importance | Web data are usually not accompanied by extensive documentation. Identifying potentially relevant contextual information early on is essential for the relevance and validity of the research. |
| Solutions and best practices | Screen blogs, press releases, a source's software "changelogs," or use Google's reverse search to identify important (technical) developments Build an initial understanding of the presence of algorithms by visiting the source with different devices at different times or by inspecting the site's source code Understand changes to the data-generating process (e.g., by studying changes over time using archive.org) Inspect the robots.txt file and assess how the source requires users to agree to their terms of service (e.g., preferrable "browsewrap" vs. less preferable "clickwrap" agreements) Scrutinize popularity, legitimacy, and business model of data sources (e.g., by using firm reports, stock filings, news, and social media, other data providers like Statista) Explore forums where users of the source talk about the source (e.g., Reddit) Assess whether the data links to other data sets (e.g., by spotting common IDs) Map out "worst-case" scenarios for research objectives in the case that the data source changes (e.g., discontinuation of an API, removal of a website)
|
Graph
Table 3. Challenges and Solutions in Designing Web Data Collections.
| Challenge #2.1: Which Information to Extract from Which Pages? |
Validity Challenges [V] Which information is necessary to justify construct operationalization and allow analysis? Which metadata might enhance internal and external validity? Is information subject to algorithmic biases or missing data? Are there significant changes to the data-generating process?
Legal/Ethical Challenges [L] Is all of the required information publicly accessible, or is a login required? Does the data contain personal or sensitive information, and can subjects be identified? Is there a sufficient scientific justification for using the data? How large is the overlap between the research objective and the original intent of subjects disclosing the data?
Technical Challenges [T] Is all information extractable? Are there any limits to iterating through pages or endpoints? Does the extraction software obtain information reliably?
| Solutions and Best PracticesExplore different types of pages to detect unique vs. identical information [V] Explore whether alternative ways to browse/navigate the site (e.g., URLs, clicking, scrolling, logging into the site) provides different or reveals new information [T] Explore how extraction methods (e.g., "headless" HTTP requests vs. simulated browsing, different user agents, screen width, login status, use of different packages) affect information display [V, T] Assess the accuracy of timestamps (e.g., time zones) [V] Save screenshots of pages that describe the calculation of metrics [V] Explore (temporarily available) information in the source code of a website using the browser's "inspect" tools [V] Assess the presence of technical roadblocks (e.g., captchas) [T] Assess how data was generated historically at the source (e.g., via archive.org) [V] Explore limits to iterating through pages [T] Obtain information from various sources to reduce dependency on data provider [L] If possible, opt out of firm-administered experiments or block cookies; alternatively, identify relevant metadata that can be used to control for the presence of algorithms [V]
|
| Challenge #2.2: How to Sample? |
Validity Challenges [V] Is the sample size sufficient to effectively inform the research question? To which population does the sample generalize? Is the sampling frame corresponding to the research objective (e.g., randomness)? How prevalent is panel attrition?
Legal/Ethical Challenges [L] Does the data represent an excessive portion relative to all data available? Can the data be obtained in similar forms elsewhere, or is the research question only answerable with the targeted data? Are some of the sampled units (potentially) vulnerable?
Technical Challenges [T] Is the required sample size technically feasible? Can external information (e.g., IDs) be consistently matched to the data?
| Solutions and Best PracticesAssess characteristics of the population (e.g., using secondary sources) [V] Explore options to sample directly from the source (e.g., from different pages, randomization through filtering/searching, obtaining usernames from forums, see also Neuendorf 2017 and Humphreys and Wang 2018) [V] Choose lists or pages that are not affected by algorithmic influence [V] Refresh sample (or use multiple types of sampled units) to assess the stability of sample and counterbalance panel attrition [V] Discard units from the sample to prevent data collection from subjects falling under prohibitive national and supranational legislation (e.g., GDPR) [L] Explore external sources to inform the sampling frame [V], or facilitate linkage [T] Assess the efficiency of different navigation paths and their impact on sample size [T] Pseudo-anonymize or discard sensitive or personal information [L] Ensure that no excessive amount of data (e.g., data on all users) is collected (absolute volume, relative volume) [L] Reexamine alternative sources to improve justification of data extraction [L]
|
| Challenge #2.3: At Which Frequency to Extract the Data? |
Validity Challenges [V] Is the extraction frequency in sync with the studied phenomena? Is the refresh rate of the source sufficient? Is the data (thought to be archival) really archival? Is the information consistently available across all periods of interest? Does the order and frequency in which information is retrieved induce bias?
Legal/Ethical Challenges [L] Does the extraction frequency pose an excessive load on the source? Does collecting more data at higher frequencies make the data more sensitive?
Technical Challenges [T] Does the desired extraction frequency pose new technical hurdles? How can the stability of data collection be guaranteed, and different collection batches be distinguished?
| Solutions and Best PracticesExplore the gains in collecting data multiple times rather than once (e.g., in a "live" data collection) [V] Adhere to best practices in setting the extracting frequency (e.g., five requests per second for APIs, one request per two seconds for web scraping) [L, T] Experiment with technical parameters (e.g., number of computers) to balance technically feasible sample size and desired frequency of data extraction [T] Formulate, test, and refine data source theory (Landers et al. 2016) [V] Reinspect the robots.txt file to avoid exceeding retrieval limits for selected pages [T] Consider randomizing extraction order for sampled units over time [V] Consider (cost) implications for storage and computation time [T] Consider getting in touch with the data provider if the targeted data set is infeasible to extract via web scraping or APIs [T, L] Devise a schedule for the automatic extraction of the data (e.g., using Windows Task Manager or Cron) [T, V]
|
| Challenge #2.4: How to Process the Data During the Collection? |
Validity Challenges [V] Could erroneous processing lead to unexpected data loss? Could there be any significant scientific value in retaining the raw data?
Legal/Ethical Challenges [L] Is the collected data in conflict with prohibitive laws (e.g., GDPR)? Is the collected data sufficiently secured from unauthorized access? Is anonymization or pseudonymization required?
Technical Challenges [T] Which storage facilities to use to accommodate the expected data (size, location, format, encoding) Is normalization necessary?
| Solutions and Best Practices Retain raw data (e.g., HTML pages, JSON responses) whenever possible [V, T] Always parse some minimal amount of data (e.g., timestamps) to facilitate monitoring checks in real-time [V, T] Remove sensitive and personal information on the fly; if personal or sensitive information is strictly required to meet the research objective, consider pseudo-anonymizing (potentially via third parties) [L] Verify data storage during collection meets legal requirements for potentially sensitive or personal data [L] Ensure proper encoding of (non-English) characters, retain correct digit separators and correct data format
|
Graph
Table 4. Challenges and Solutions in Extracting Web Data.
| Challenge #3.1: Improving Performance |
| Reason for importance | In scaling up the data collection, researchers might encounter new technical issues. For example, the data collection could stop unexpectedly or proceed much slower than anticipated. |
| Solutions and best practices | When scraping, use stable selectors (e.g., tags, classes, attributes, styles associated with particular information) and make only selective use of error handling When using APIs, choose a stable and supported version Attempt to reparse data from stored raw data if the extraction failed Check for traces of being banned/blocked/slowed down by the website (e.g., by scanning the content that was retrieved) Notify data sources about potential bandwidth issues with APIs Update the technically feasible retrieval limit, and recalculate desired sample size, extraction frequency, etc. Verify that computing resources are appropriate and reliable (e.g., scale up or down servers, verify that database runs optimally) Move data to a remote (and more scalable) file storage or database Consider potential benefits from using cloud computing (e.g., for extended, uninterrupted data collection) vs. benefits from local setups (e.g., due to security or privacy concerns) Budget the expected costs of API subscriptions, cloud computing and data storage and transfer
|
| Challenge #3.2: Monitoring Data Quality |
| Reason for importance | Monitoring is critical to be timely alerted to data quality issues. Setting up a monitoring system allows researchers to intervene before discarding data altogether. |
| Solutions and best practices | Log each web request (i.e., URL call), along with response status codes, timestamps of when the collection was started, and when the request was made Save raw data (i.e., source code of HTML websites), along with the parsed data for triangulation Verify whether the raw data was correctly parsed (e.g., for a sample of information, compare raw data and parsed data) Check file sizes or the number of observations at regular intervals Set up a monitoring tool to timely alert you to any future issues (e.g., based on the number of files retrieved or requests made, file sizes retrieved, time the collection last ran, budget spent) Automatically generate reports on data quality (e.g., using RMarkdown) Record issue(s) in a logbook (e.g., in the documentation); especially if considered critical for data quality
|
| Challenge #3.3: Documenting Data |
| Reason for importance | Researchers are responsible for documenting the data set they produce from web data. Building documentation during the collection is important to guarantee accuracy and completeness, which facilitates use, reuse, and replicability. |
| Solutions and best practices | Maintain a logbook in which to note important events (e.g., when the collection broke down and why) Start writing the documentation in the early stages of collecting the data, and make use of templates (e.g., Datasheets for Datasets; Gebru et al. 2020) Keep and organize copies of relevant files (e.g., screenshots of the website at the time of data extraction, the API documentation, details on variable operationalization with summary statistics, information about the context) Have a plan for long-term archival storage (e.g., re3data.org, The Dataverse Project, Zenodo), anonymization (e.g., generating synthetic versions of sensitive data), and consider which license to use for the data (e.g., Creative Commons)
|
A critical first step in the use of web data is selecting the data source(s). We examine three challenges faced by researchers in this selection process. First, it is essential that researchers explore the universe of potential sources (challenge #1.1). Second, researchers need to consider the range of possible extraction methods (challenge #1.2). Third, it is crucial to map the context in which the data are generated (challenge #1.3). Table 2 summarizes our recommendations for tackling these challenges.
In the absence of conventional gatekeepers (e.g., data providers), researchers can select from countless web data sources. For example, there are 2.1 million online retailers in the United States alone ([20]). Further, websites and APIs differ greatly in scope (e.g., number of users), data quality (e.g., consistency), and retrievability (e.g., extraction limits). Even within the same product category, data sources differ vastly. For example, Amazon reports a book's sales rank (an aggregate outcome metric for product sales), whereas Goodreads reports users' reading behavior (an individual outcome metric for consumers' usage intensity).
Faced with a vast universe of potential sources, researchers may be tempted to focus on familiar platforms only ([89]). For instance, Amazon is the most used web data source in marketing (see Table W4 in Web Appendix C). Amazon might be a relevant source to extract book reviews, given its broad assortment and user base. Yet, in other cases, researchers might miss opportunities for identifying novel, emerging marketing phenomena or conduct more compelling theory testing without a thorough exploration of potential sources. Researchers can avoid the pitfalls of defaulting to dominant sources by actively considering a broad spectrum of websites and APIs, ranging from highly popular (e.g., Amazon) to less popular sources (e.g., Goodreads), from primary data providers (e.g., YouTube) to data aggregators (e.g., Social Blade), and from platforms with global reach (e.g., Twitter) to more regional ones (e.g., Taringa!). Another strategy to move beyond familiar sources is to adopt alternative perspectives. For instance, researchers can consider websites or APIs that are used by consumers, analysts, or managers. API directories at GitHub or programmableweb.com can facilitate identifying potentially relevant APIs.
A broad exploration of web data sources may lead researchers to discover sources that may be more permissive for (academic) data extraction or less likely to trigger ethical concerns. For example, websites that do not require logging into the site to reveal information are typically more scraping-friendly than sites that first require registering a user account. In the case of Amazon, researchers can obtain most information without logging in and do not have to explicitly provide their agreement to the website's Terms of Service. To reduce legal (e.g., breaches of contract, as researchers have provided explicit agreements to the terms of service) and ethical (e.g., website users may consider their data private) risks, researchers should refrain from creating fake accounts to access information requiring a login. By explicitly declaring their academic status (e.g., when registering at the site using the institutional email address), researchers might be able to diminish their exposure to legal risk.
When exploring web sources, researchers need to examine whether theoretical constructs can be operationalized in a valid manner ([91]). A healthy level of skepticism is warranted when using idiosyncratic metrics from APIs or websites. For example, researchers might be interested in scraping the price tier of restaurants from Yelp. Yet, it is not entirely clear how Yelp computes this metric from individual consumer ratings.
To determine when to stop exploring sources, researchers need to assess to what extent the selected source(s) improve(s) on alternatives. One way to justify selecting a single web source is the presence of unique features. For example, a researcher studying how observers react to humor in reviews might prefer Yelp to alternative platforms as it is the only source featuring "funny" votes (e.g., [60]). At other times, researchers may be indifferent between potential sources and can draw from multiple sources to boost the generalizability of their findings (e.g., tweets and restaurant reviews; [61]). Collecting data from multiple sources is often useful because even similar types of information (e.g., consumer comments) may affect marketing outcomes differently, depending on source characteristics (e.g., forums vs. microblogs; [72]). Data aggregators—some of which offer authorized data access via APIs—facilitate the collection of such multisource data.
The popularity of web scraping may lead to the conclusion that it should be preferred over other methods. However, some web sources offer access to data via APIs ([12]). In general, extracting data via APIs is more scalable and less likely to invoke the same level of legal risks compared with web scraping. Although some sources offer unconstrained APIs that do not require authentication, others require (paid) subscriptions and authentication procedures. Some sources, such as Twitter, have recently started offering APIs for academic research. In the case of Amazon, an API offering access to consumer reviews is currently not available.
In addition to APIs, numerous other options exist for researchers to obtain web data. For example, some data providers (e.g., Yelp, IMDb), public data platforms (e.g., Kaggle, The Dataverse Project), and researchers (e.g., [59]) provide documented web data sets that can readily be used for academic research. There are many potential use cases of such data sets, but less than 5% of all web data–based articles in marketing used such data sets. To avoid the pitfall of defaulting to web scraping for data extraction, researchers can expand their search by explicitly including terms such as "API" or "data set download" in their search queries.
Relative to other frequently used archival sources in marketing, web data entail large and often undocumented complexities. Thus, it is critical that researchers map the data context, which involves identifying relevant contextual developments that may undermine the validity of the research if gone unnoticed.
First, mapping the data context may reveal changes in the underlying data structure. For example, a major change in a platform's user interface may affect subsequent consumer behavior. Second, mapping the data context enables researchers to identify relevant pieces of information for collection together with the focal web data. For example, researchers may discover an external website (e.g., Statista) that offers information about the composition of a focal data source's user base. If stored, such data could eventually be used to detect changes in the composition of the user base or verify the representativeness of the extracted data. Third, mapping the data context may reveal unknown information, potentially allowing researchers to discover novel research opportunities. For example, researchers may use the (unexpected) launch of a new recommendation system at a music streaming service as a natural experiment to investigate the impact of recommendations on music consumption.
To understand and map the contextual complexity of web data, researchers can immerse themselves in the ecosystem surrounding the focal source by signing up and using the source, tracking press releases, social media, and scanning the competitive environment. Helpful tools include a search engine's advanced search features, newsletters, and alerts on leading business and technology magazines (e.g., TechCrunch.com, WSJ.com, FT.com). The website's source code may also hold valuable information about potentially relevant environmental changes. Sometimes, researchers may also detect the presence of algorithms on the site that may threaten the validity of the collected data. For example, Amazon's product pages personalize information based on which preceding products were viewed—even without users logging into the site.
After narrowing down potential sources, researchers decide which information to extract from them (challenge #2.1), how to sample (challenge #2.2), at which frequency to extract the information (challenge #2.3), and how to process the information during the collection (challenge #2.4). Table 3 summarizes these challenges and corresponding solutions.
In the absence of any "downloadable" data set, the first challenge lies in deciding which information to extract from a source. Researchers begin by browsing the web page to identify from which pages to extract which information. In our Amazon example, some of the most commonly used pages are product pages (e.g., [14]) and review pages (e.g., [84]). Generally, pages such as those from e-commerce platforms contain information from the company's database, offering researchers the opportunity to capture some of the information available at a company. Collecting such data involves iterating through a set of related pages (i.e., browsing through many product pages and corresponding review pages in our Amazon example) and saving the data as it becomes visible.
As the goal of websites like Amazon is rarely the provision of data sets for academic research, it is often necessary to combine information from different pages (e.g., book descriptions from the product page and ratings from the review page). It is particularly difficult to recognize the subtleties of available information, which makes the decision from which pages to extract information challenging. For example, researchers interested in building a data set of book reviews would find total ratings both on the product and review page, but only the review page reveals all product reviews.[ 7] Yet, neither the product nor the review pages contain all the biographical information available on a reviewer's profile page. Widely exploring a website or API is necessary for identifying information relevant for subsequent analysis (e.g., construct operationalization). The amount and type of information also often vary (e.g., depending on screen width or whether the user is logged in). In this phase, researchers should assess the degree to which the information could be considered personal or sensitive under different regulatory regimes (e.g., the European Union's General Data Protection Regulation [GDPR]), which may require planning measures such as pseudo-anonymization of reviewer names. Researchers may also reassess whether all information needs to be captured to meet the research objective. Suppose reviewer names are strictly necessary (e.g., because they allow for matching different sources). In that case, researchers can explore whether the targeted web data source offers ways to exclude subjects governed by prohibitive privacy regulations (e.g., by using filters).
An important threat to internal validity in any study involving web data is algorithmic interference (e.g., [91]). The (visual) design of websites that facilitates usability can undermine the validity of the collected data if gone unnoticed and unaddressed. Especially when deciding which information to extract, it is important to reexamine the website or API for the presence of algorithms. For example, the order in which the researchers in our example visited the website while designing their data extraction could affect which related books are displayed on the product pages on Amazon. Other algorithms that often affect the display of data on websites are sorting algorithms (e.g., by popularity or mixed with sponsored search results) and filtering algorithms (e.g., showing subsets of the data). Algorithmic interference is often hard to detect without being sensitive to it. To account for potential algorithmic interference, the researcher might extract variables as part of an algorithm's more extensive set of input variables, which offers opportunities to control for them in the empirical analysis (e.g., the order in which books were extracted in our Amazon example).
Researchers also need to establish the intertemporal stability of available information. Because the web is constantly evolving, the information on a page might not have been generated via the same process over time, undermining the internal validity of the data. Some changes to sources are drastic enough to alter how the data were created in the first place, introducing measurement error ([87]). For example, Amazon shifted to a positive-only evaluation of reviews by removing the "not helpful" vote button in 2018, and it no longer displays "not helpful" counts next to reviews ([28]). This change might have impacted review content (e.g., users writing shorter review texts). Yet, researchers collecting data today cannot find any traces of these "not helpful" votes. A tool for examining changes to relevant information on a website is the Wayback Machine (archive.org). Researchers can use this tool to retroactively inspect websites over time (e.g., [58]) or submit their own website links for archiving.
Finally, collecting metadata that "annotates" the data collection enhances internal and external validity (e.g., storing the timestamp of data extraction, whether an API request was completed successfully, or the IP address from which the data request was made). Such metadata can be used not only for diagnostic purposes but also to link the extracted web data to other data sets. For example, in our Amazon example, the collected data could be linked to other data using IP-based geolocations (e.g., linking geolocation and web search data; [86]) or timestamps (e.g., linking reviews to stock prices; [78]).
A second challenge in designing the data extraction lies in deciding how to sample from the data source. In particular, in the absence of access to the data source's entire database, it is difficult or impossible to draw a random sample from the population (e.g., all products) available at the data source. Instead, researchers need to devise their own sampling frame to reveal the units they want to sample from the website ([66]). For example, researchers could scan the site for an index of all products that could inform their sampling. In our example studying reviews at Amazon, multiple such indexes may be available. Should products be sampled from the bestseller page for books (so-called exposure-based populations; [66]) or instead from the category page for books (i.e., availability-based populations)? Choices like this result in different data and may even invalidate inferences, as sampling frames might inadvertently induce systematic bias ([39]).
One common validity challenge in choosing how to sample is determining how many units (e.g., books) are sufficient to inform the research question. From a validity standpoint, it would be ideal to collect information on the entire population (e.g., all books available at Amazon). However, Amazon does not have an obvious page to extract all books. Imagine that a research team wanted to collect information about all marketing books sold on Amazon. The bestseller page, for example, lists only the top 100 bestsellers. By manually changing pagination parameters in the URL, the top 400 bestsellers can be revealed. Yet, this list of 400 books neither constitutes the entire population nor represents a random sample of marketing books sold at Amazon. Alternatively, when starting from the product overview pages, these pages list an imprecise number of books (e.g., "over 60,000"), which can only be viewed up to page 50. With each result page featuring 24 organic search results, this approach would produce 1,600 books per category at best. Thus, researchers need to consider other ways to identify more books on Amazon, such as searching for books using various keywords. To expand the number of sampled units, researchers could collect data multiple times, use other keywords, or tweak search parameters to reveal more data by requesting narrower subsets from the database (e.g., only books published during a specific month).
Even if a list of the population (e.g., all books) could be retrieved, it may be infeasible to extract data within a reasonable time frame. While sample size requirements are mostly concerned with a researcher's inferential goals (e.g., [50]), few articles make the resource constraints that affect collecting web data explicit (e.g., [68]). For example, with web data, a study's sample size critically depends on technical parameters such as the number of computers used for data extraction or the number of pages that need to be visited. We illustrate how to calculate the technically feasible sample size in Web Appendix F, which may effectively complement traditional sample size calculations commonplace in marketing.
As a result of these complications, researchers often restrict their sample size. One way to motivate a compelling sampling frame is to use external sources that can be linked to the web data. For instance, the New York Times or Publishers Weekly bestseller lists might be a starting point for sampling books ([14]). An alternative approach focuses on internal data available at the source itself. Researchers may have to allocate substantial time to identify ways to sample from the focal source. Sometimes, starting the data collection from a page unrelated to the focal pages of interest might facilitate collecting a more representative sample (e.g., by reducing geographical biases; [86]). For example, on Amazon, researchers could first sample reviewers and associated demographic information (available at the user profile of reviewers) and subsequently retrieve data on all reviewed products. Similar to how researchers build network data from an initial set of products or users, the sampling units retrieved from an initial set of pages can be considered seeds. In choosing seeds, researchers should be cautious about drawing from vulnerable populations (e.g., minors) or infringing on prohibitive privacy regulations.
Web data are nonstatic, as they change often or might disappear altogether. Therefore, researchers need to consider at which frequency to extract information. This decision encompasses whether to collect data once or multiple times and when to run (and potentially schedule) the data extraction. Consideration of the frequency and schedule is challenging but required to ensure the intertemporal stability of measurement, which is critical for internal and construct validity.
From a technical and legal perspective, it is most desirable to extract data only once. Single extractions are less likely to represent a burden on the firm's servers, and the extracted data often only represent a limited snapshot of the entire database, reducing the risks of copyright infringement. Further, such data may be more likely to respect users' "right to be forgotten," which is part of the privacy laws in some jurisdictions. Yet, single data extraction might raise several validity issues that can easily go unnoticed. For instance, in our example, researchers extracting book reviews once from Amazon will not be able to identify whether any of the archival information has changed. Only when extracting data multiple times can researchers systematically notice changes on the site, which may lead to the identification of "fake" reviews that have been removed by the platform (e.g., [29]). More generally, researchers can compare information over time to detect whether data that initially appeared to be archival is truly archival (i.e., does not change over time).
Another concern is that a single extraction may not produce a data set that adequately maps onto the focal processes of interest. For example, suppose researchers in our example want to examine whether a review by a so-called "Top 1000 Reviewer" leads to more subsequent reviews from other users. However, the researcher merely observes that the reviewer is a top reviewer at the time of data extraction. This does not necessarily imply that this user had the same status when the review was first posted and thus was most likely to affect subsequent reviewing behavior of other users. Formulating and testing the essential assumptions about the data, including the relation between the time of data extraction and the focal (psychological) processes, is thus critical. The formulation of such assumptions is called a "data source theory" ([52]). Testing and refining the data source theory helps take proactive steps to enhance internal and construct validity. In the preceding example, it would thus be necessary to collect data from these review pages closer to the original posting date, ensuring that reviewers classified as "Top Reviewers" had that status when their reviews became visible.
When extracting data more than once, automatic scheduling can help ensure consistency and contribute to validity. Scheduling is beneficial if the required information is only available in real-time. For example, sales ranks at Amazon are updated hourly for popular products, and historical sales ranks cannot be retrieved. Suppose researchers in our example were interested in studying the sales performance of books over time. In that case, they could repeatedly extract the books' sales ranks from the product pages at Amazon. Sometimes fixed intervals enhance validity (e.g., every Monday, 8 a.m.). In other circumstances (e.g., when collecting data from many pages), it may be better to vary the starting time or weekday of the data extraction.
Another decision is whether to set an end date for the data extraction. Collecting data over extended periods offers the potential for researchers to build a programmatic stream of research and stumble into unexpected natural experiments (e.g., [13]). Especially for longitudinal data collections, continuing the data collection while the project is in the review process brings numerous benefits, such as the ability to update the data (e.g., a longer time frame, new measures). Yet, concerns about technical feasibility (e.g., storage requirements, continued availability of data source) grow as the data extraction horizon extends. Similarly, from an ethical perspective, the longer the data extraction, the greater the likelihood of potentially identifying individuals via triangulation. Next to ethics, long-term data collection also places a heavier load on servers, potentially increasing exposure to legal risks.
As a final step in designing the data extraction, researchers decide how to process the information while it is collected. Any kind of web data collection requires a minimal degree of processing, given that the information is available in a computer's memory (e.g., in the browser or the software processing the API output) and still needs to be stored in files or databases. Thus, this processing step occurs before data sets are cleaned or analyzed.
When deciding on how to process information during the extraction, researchers must balance potential efficiency gains from molding raw web data into readily usable data sets with the potential threats to validity due to "on-the-fly" processing. For example, in our Amazon example, researchers may be tempted to remove seemingly unnecessary information (e.g., image links in reviews), apply text processing (e.g., removing characters used as separators), or force specific information (e.g., prices) to be stored in a strictly numeric format. Such on-the-fly processing promises to produce essential efficiency gains, as the data set resulting from the extraction could directly be analyzed. However, because on-the-fly processing decisions are usually made after the inspection of only a limited number of pages in early prototypes of the data collection, it is difficult to guarantee their correctness. For example, using our example, what if the initial screening revealed only pictures posted in a review, while the extensive data collection revealed the need to capture video files? Given this and related challenges, keeping the raw data (such as the source code of websites, API output, or any media files loaded at the time of data extraction) is ideal from a validity perspective. For example, even if the data collection breaks, researchers could still process and use the information after debugging their extraction code. Retaining the raw data can also help reduce Type 1 errors by increasing transparency about researchers' degree of freedom in collecting and processing the web data. Yet, retaining the raw data prompts significant concerns about the technical feasibility and ethical risks. From a technical standpoint, storing the raw data might require databases to retain their original structure and facilitate processing, especially for projects involving many raw data files collected over extended periods. Keeping all raw data might raise questions regarding the right to store the raw data—especially if it is not (pseudo-) anonymized before storage.
Finally, retaining the raw data allows researchers to refine their extraction design at later project stages. For example, a researcher might have collected Amazon reviews in 2018—around the time of the removal of the "not helpful" voting feature. Although extracting "not helpful" votes was not part of the original extraction design, researchers would be able to use the raw web data to examine the effect of the removal of these "not helpful" votes.
After source selection and designing the data collection, researchers gradually transition to turning their small-scale prototype into stable extraction software. In so doing, researchers face three challenges. First, researchers may need to improve the performance of their extraction software when operating it automatically at scale (challenge #3.1). Second, they may need to implement monitoring checks to be alerted to any issues arising during extended data collections (challenge #3.2). Third, researchers should compile information important for documenting the final data set (challenge #3.3). Table 4 contains a summary of solutions and best practices to these challenges.
In scaling up their data extraction, researchers may notice that the extraction software frequently breaks across a larger number of pages or runs significantly slower than expected. Such technical challenges, if unaddressed, have the potential to undermine research validity (e.g., missing data, not meeting sample size requirements). A practical solution to preempt these and similar challenges involves capturing the focal information in different ways and storing raw data—especially in the early stages of data collection and for more ambitious, large-scale web data collection projects. To track whether the extraction targets are met, researchers can log the (timestamped) URLs of scraped pages and visualize the performance of the extraction software over an extended period. The resulting "effective" extraction frequency can then be used in recomputing the technically feasible sample size (see Web Appendix F). Novel web scraping services promise to handle technical difficulties efficiently (e.g., ScrapingBee, Zyte).
As a next step, researchers consider which metadata can help them diagnose issues with the data collection in real-time. Especially when websites constantly change, monitoring the health of web scrapers can be a tedious task. Researchers should consider performance at a higher level (e.g., the file sizes of extracted raw data) and lower level (i.e., the accuracy of the information in resulting data files) to assess whether the collection is proceeding as expected. When collecting over long periods, automatic reporting can greatly facilitate monitoring. Finally, alerts (e.g., via email or mobile) can help researchers detect predefined data issues quickly.
During the data extraction, researchers need to record relevant information about the data in real-time. This is an essential step in building documentation, enabling future data usage by the researcher(s) who collected the data and other scholars. Even after the data extraction has ended, researchers can continuously refine the documentation as they become familiar with the characteristics of the data (e.g., variables that were erroneously captured, missing values).
Accurate and comprehensive documentation is particularly critical given that collecting web data tends to involve repeated iterations between discovery (and often troubleshooting) and confirmation (i.e., subsequent analyses that are outside the scope of our framework). Designing web data extractions requires a different mindset compared with experiments or archival research. Unlike running experiments, the extraction design for collecting web data may be in flux, even when the collection is already running. Relative to traditional archival research in which data sets are sufficiently annotated, researchers are in charge of accurately recalling details about the data collection. Such details encompass information about the data composition (e.g., sampled units), extraction process (e.g., annotated code, detected errors during the collection), and processing details (e.g., applied cleaning steps). The template of [24] provides a useful starting point for building the documentation for a data set collected via web scraping or APIs. Given that contextual changes are inevitable (see challenge #1.3), documenting the source's institutional background (e.g., screenshots, corporate blog posts, API documentation) is crucial.
An unprecedented gold rush of web data has enriched the marketing discipline for two decades—over 300 published articles provide countless examples of impactful marketing insights using web data. With the ever-increasing digitization of social and commercial life, it is hard to imagine that the heyday of this gold rush might subside any time soon. Yet, are marketing's currently productive mines the only or the most promising sources of marketing insights in the future? Which novel approaches and technologies are necessary to capture and describe evolving marketplace realities?
To identify directions for future research, we have reviewed more than 300 articles to provide a snapshot of the current state of web data in marketing. We use these insights to inform the subsequent discussion, which we organize along the four pathways through which web data can advance marketing thought (as summarized in Table 1). We supplement our discussion with key elements from our methodological framework (see Figure 2) and inspiring use cases from other disciplines.
Next, we discuss how researchers can use source selection to branch out to new or underutilized sources for studying emerging substantive topics. We also highlight how researchers can design more complex, longitudinal, and multisource web data sets to reveal otherwise invisible phenomena.
Our review reveals that marketing research draws from a somewhat concentrated list of web sources (see Table W4 in Web Appendix C). We encourage researchers to focus on underused or niche sources that have received limited or no attention in marketing. Web data are often prized, as they allow for collecting "consequential dependent variables from the 'real world'" ([40], p. 357). Identifying new sources or novel consequential variables constitutes a promising avenue for discovering emerging phenomena.
Consider, for example, the twilight state of the nascent legal cannabis industry in the United States. While more states are legalizing cannabis for medical and recreational use, the market value of the legal U.S. cannabis industry was still less than a third of the illegal market in 2020 (i.e., $20 billion vs. $66 billion; [22]). Using surveys, media coverage, and in-depth interviews, marketing scholars have begun to explore how such legalized markets emerge and seek legitimacy ([38]). Sociologists and organizational scholars, in turn, have already used web data to compile intriguing data sets from sources such as Weedmaps. Using these data, they examine, for example, how existing medical cannabis dispensaries have repositioned themselves after the entry of recreational dispensaries ([34]) or how consumers deal with potential stigma transfer (Khessina, Reis, and Cameron Verhaal 2021). By leveraging similar web data, marketing researchers could explore intriguing marketing questions. For instance, how should brands position themselves (e.g., brand personalities, emphasis on product vs. service), depending on the strength of categorical stigma? What are the potential public health and welfare implications of the increasing competition among cannabis dispensaries or their growing social media activities?
In addition to being attuned to work in other disciplines, a low-tech route for source exploration is provided by Similarweb, which allows researchers to browse website rankings by region or category. Given the broad accessibility of web sources worldwide, the dominance of Northern American and European data sources is surprising. Not a single article focuses exclusively on African web sources, and only a handful of articles use some African data (e.g., [49]). Possible starting points for branching out into these underexplored marketplaces could be popular websites such as Nairaland.com (online community), bidorbuy.co.za (auction platform), and Jumia.com.ng (e-commerce).
Most published marketing articles use web data gathered from a single source. Only very few articles collect data from a large number of web sources (i.e., 50 or more web sources). Following the lead of these articles, we encourage marketing researchers to envision unique data sets compiled from many and diverse sources. For example, in economics, [10] collected online and offline prices for individual goods sold by 56 large multichannel retailers in ten countries (i.e., United States, United Kingdom, Argentina, Australia, Brazil, Canada, China, Germany, Japan, and South Africa) between 2014 and 2016. This "Billion Prices Project" (bpp.mit.edu; [11]) exemplifies how creative and ambitious data collection from diverse web sources can fuel entire research programs. Especially if sufficiently documented, such web data are poised to unearth new fields of gold for the marketing discipline.
As researchers decide which information to extract (see challenge #2.1), they may overlook novel information on sources they already know. Therefore, refocusing on different information may also reveal how to study novel phenomena on frequently used sources. Adopting a "discovery mode" may reveal that phenomena of high societal relevance such as gender or racial issues are occurring at frequently used sources such as TripAdvisor ([69]), Kickstarter ([92]), and DonorsChoose ([ 2]). For example, in entrepreneurship, [92] scraped Kickstarter information to examine whether male African American founders are less successful in crowdfunding. Researchers in marketing, in turn, could build on these and similar ideas to explore whether biases exist in other online market exchanges.
Another promising lever for exploring emerging phenomena is the extraction frequency (challenge #2.3). In most articles, the data were extracted once (e.g., on a single occasion). Extracting data once is sufficient for many research objectives, such as demonstrating the prevalence of a phenomenon in the marketplace (e.g., [79]). Yet, researchers can also uncover novel marketing phenomena by creatively envisioning web data sets that only reveal the phenomenon if the information is extracted multiple times. For example, [29] leverage the observation that Amazon removed certain reviews to study the market for "fake" reviews. Specifically, they combine repeatedly web-scraped data from Amazon with hand-coded data from large private groups on Facebook used to solicit fake reviews to examine the short- and long-term impact of such rating manipulations. This example illustrates that data imperfections (e.g., data modifications discovered when mapping the data context, see challenge #1.3) can be opportunities to pose novel research questions rather than merely nuisances that warrant correction.
As a second direction for knowledge discovery, web data are often used to increase the ecological value of marketing research by complementing carefully controlled experiments. Triangulating findings generated via different methods is fruitful. Yet, there are many other underutilized avenues for how researchers can select and extract web data to infuse ecological validity into experiments and other types of marketing studies.
By carefully selecting websites and APIs, researchers can enhance the ecological validity of their experiments (e.g., through more realistic or diverse stimuli and measures). This enormous potential has hardly been realized in marketing, particularly at scale (for a creative smaller-scale application, see [62]). Social psychologists demonstrate the full potential of such an approach. Consider, for example, [33], who scraped 87 real-world profiles of doctors (including their fitness habits) from the website of a health insurance provider. These profiles served as the foundation for a novel stimulus-sampling paradigm wherein participants in experiments were presented with randomly selected subsets (i.e., five fitness-focused and five non-fitness-focused profiles). In doing so, the authors first ground the phenomenon in the field (i.e., that doctors signal their fitness habits) and then use stimuli created from real profiles to demonstrate that overweight and obese individuals are less likely to choose fitness-focused doctors for their own care. Such triangulation and the creation of larger and more representative samples of naturalistic stimuli enhance the replicability and generalizability of experimental effects ([42]). The experimental paradigms in core marketing topics (e.g., branding, advertising, pricing) and methods (e.g., lab experiments, conjoint studies) could benefit from similar applications to mimic real marketplaces. For instance, branding or advertising researchers might develop stimuli based on data extracted from sources like crowdfunding platforms or Bing's Image Search API (e.g., brand logos, ads, and slogans).
While field experiments continue to be prized for their realism and high ecological value ([83]), very few published marketing articles use APIs to run field experiments (e.g., [51]; [80]). There are many untapped opportunities to run field experiments administered by researchers rather than cooperating partners (e.g., firms or charities). Using APIs to run field experiments gives researchers more control over the design and debriefing processes and allows for monitoring of granular participant behavior over longer periods. Thus, web data–based field experiments potentially produce more precise effect sizes and allow researchers to capture long-term effects ([26]). In such experiments, researchers might randomly assign users to different treatments, such as adding (vs. not adding) followers on Twitter ([80]) or assigning (vs. not assigning) Reddit's Gold Awards to user posts ([ 9]). By gathering high-frequency data via APIs, researchers can analyze how experimental treatments influence outcomes such as posting or the creativity of user-generated content. Alternatively, APIs can be leveraged to infuse realism into experiments, as embodied in [56], who developed "Hoogle," a mock search engine that relies on APIs offered by Google but only displays organic search results that are not altered based on previous user queries. We foresee many more creative future applications of web data to facilitate such field experimentation.
A core topic in marketing research is to develop marketing metrics that can guide managerial decision making. Traditionally, many metrics have been based on offline information and established data providers (e.g., [21]). Given the continued growth and diversification of web data, it is tempting for marketing managers to focus more on web data for managing firm growth and profitability. Yet, deciding which information to select and extract for marketing insight is challenging (see challenge #2.1). More research is needed to help managers avoid succumbing to the streetlight effect (i.e., an "overreliance on readily available data due to ease of measurement and application, irrespective of their growth objective"; [18], pp. 164–65). But, how can researchers get started?
Over the last decade, scholars have begun to explore which types of web data could proxy or improve on existing core marketing metrics. For example, managers may use search data extracted from Google to spot trends in the relative importance of their firm's product attributes, which is more cost effective than traditional methods ([17]). Mining Twitter data provides cheaper, real-time, and more actionable measures and insights about brand reputation than existing survey-based metrics like the Brand Asset Valuator data from the advertising agency VMLY&R ([71]). Yet, in other circumstances, readily and cheaply available web data might not be a good substitute for more expensive or established proprietary data sources to uncover market structure ([70]).
An exciting direction for future research is to explore what web data sources should be selected or combined to generate marketing insights that fuel firm growth. For instance, many novel metrics rely on textual data ([ 6]). This focus limits applications to markets using the same language employed by the original method (i.e., mostly English). Future research could explore what other types of web data might enable the creation of metrics and insights that allow real-time monitoring and managing diverse global markets. What insights can managers draw from differences and commonalities between the volume of different kinds of internet searches available via Google Trends (e.g., web search vs. image search vs. Google Shopping vs. YouTube search)? Alternatively, what insights about consumer preferences (or any other stakeholder) can be extracted from short videos posted on platforms such as TikTok?
A fascinating opportunity arises from providing microservices via APIs to marketing stakeholders. This means that researchers not only use APIs to retrieve data but can also operate their own APIs to examine real marketplaces (e.g., using rplumber.io in R). Researchers in data science, for example, offer firms a framework for testing multiarmed bandit policies via APIs while at the same time gathering field experimental data ([48]). Marketing researchers could use similar API-powered microservices to study emerging topics such as recommendation systems (and resulting biases) or tap into a firm's customer relationship management system to validate new customer churn models. At a small scale, researcher-powered APIs could lower the entry barriers for firms to experiment with novel algorithms that have not yet been implemented in major software packages.
The provision of APIs provides access to novel types of data, while also increasing the timeliness and ecological value of such data. For example, consider the differences between web data collected by a web scraper and the underlying clickstream data stored in the company's database. The website may merely show aggregate statistics about the number of reviews posted. At the same time, the underlying clickstream data also feature information on every website visit (e.g., time, IP address). As with self-administered APIs, researchers define which information a company should submit (e.g., as input to a recommendation algorithm). Thus, researchers can gain access to unique firm data that are otherwise difficult to obtain. For example, large-scale studies with image and video data are still scarce in marketing. Offering image and video analysis as microservices may generate knowledge discovery for new image sources, such as GIFs used in social media (e.g., Giphy).
Web data also have advanced marketing by improving measurement by efficiently collecting diverse variables. Therefore, as a fourth direction, we discuss how web data can improve measurement across the discipline, particularly by rejuvenating interest in core marketing topics (e.g., market orientation, advertising; for an overview of these topics, see [41]). Relatedly, researchers can also leverage APIs to effectively integrate algorithms for processing unstructured data at scale into empirical analyses ([88]).
Most marketing articles gather web data to describe and examine behavior occurring online. As documented in Table W4 in Web Appendix C, many of the used sources in marketing are focused on online consumer behaviors, such as e-commerce websites (e.g., Amazon), online reviewing platforms (e.g., Yelp), social media sites (e.g., Twitter), and search engines (e.g., Google Trends). Relatively less research has focused on firm behavior online. Yet, by doing so, researchers could explore many core marketing constructs (e.g., service orientation, sustainability). For example, researchers could systematically collect information available on the websites of many firms to analyze which organizational factors influence how firms signal their service orientation (e.g., employees' digital presence; [30]) or environmental credentials (e.g., the B Corporations certification; [25]) to customers and other stakeholders.
We encourage marketing researchers who have not yet used web data in their research to consider websites and APIs as valuable, rich, and timely sources to exploit the increased digitization of all forms of behaviors—not only online behavior. A recent example of bringing web data into an established "offline" research stream is [32], who scraped the annual reports of more than 8,000 firms from AnnualReports.com between 1998 and 2016. Web sources contain historical information about periods, even long before the web in its current form existed (e.g., 1998 in this case). The authors subsequently use these reports to develop a novel text-based measure of marketing excellence derived from firm letters to shareholders. Many other untapped online sources (e.g., job posting platforms) offer new insights into how firms communicate their marketing capabilities to external stakeholders beyond consumers, such as prospective employees, social activists, and investors.
Particularly for the marketing–finance interface, the web features many understudied forms of investor-facing communication that are ripe for collection at scale. For example, which type of marketing topics besides marketing excellence (e.g., marketing capabilities, brand positioning, pricing) should top management emphasize to investors to increase firm valuation during investor relations presentations, investor days, or earnings calls? Researchers could also examine the relative importance of the content versus the delivery (e.g., the tone of the speaker on the recording of an investor day; see [85]). Such multimodal data can also benefit the inferences made in established research streams.
APIs offer many opportunities for improving measurement—some of which are unexpected. For example, consumer researchers planning to run longitudinal studies might consider APIs for automating processes for managing participants at scale, thereby reducing the operating costs (and potentially boosting sample size). The Amazon Mechanical Turk API and the various Prolific Academic APIs (e.g., Study API) are good starting points for running multiwave studies.
APIs also enable much more than just retrieving data. For example, to reduce validity concerns in long-term data collection, researchers can use the Pushover API (https://pushover.net/api) to send monitoring alerts to their smartphones. The API of Amazon Web Services allows for the orchestration of virtual computing infrastructure (e.g., to capture data from different countries). Another fruitful avenue in which APIs are currently underused in marketing is facilitating stimuli selection. For example, a classic area of inquiry in marketing is how (background) music affects product and brand perceptions and choices (e.g., [ 8]). In 2022, background music is quite different (e.g., self-chosen, more diverse use cases). Researchers could use the Spotify Web API to select stimuli from millions of mood, sleep, or study playlists, thereby discovering perfect "lookalikes" that only differ on one focal attribute (e.g., tempo) but not on other acoustic attributes available at the API (e.g., valence, loudness). Even in this simple example, there might be a substantive interest in better understanding the effect of new background music on consumption choices, especially given the shift to working and studying from home.
Web data have unearthed many fields of gold in marketing. However, extracting data for generating relevant and valid research insights is challenging. Our article highlights validity concerns that require the joint consideration of idiosyncratic technical, legal, and ethical questions. We introduce a novel methodological framework (Figure 2), offer practical solutions (Tables 2–4), and outline directions for future research to enable researchers to create impactful and credible marketing knowledge. While our focus is primarily on authors, our work also spotlights crucial validity concerns to scholars reviewing web data–based research and practitioners interested in deriving accurate and actionable marketing insights from web data.
We hope that our work encourages marketing scholars to integrate web data into their research programs. While web data often provide compelling answers to the question, "Assuming that this hypothesis is true, in what ways does it manifest in the world?" ([ 5], p. 1455), this does not imply that web data are relevant for all research projects. Web data sources tend to feature a large N (i.e., many users) with many V (i.e., different pieces of information for potential variables) observable over a large T (i.e., many observations over extended periods of time and at a very granular level; [ 1]). Yet, collecting web data via web scraping or APIs provides limited information about the browsing behavior of individuals on the website that led to the creation of the data in the first place. Significant synergies exist by enriching clickstream stream data capturing such browsing processes with web data retrieved from web scraping and APIs (e.g., [54]).
Our work aims to bridge entrenched training silos (e.g., between quantitative marketing and consumer behavior). We encourage scholars to further integrate and leverage existing best practices with regard to the collection and analysis of web data (e.g., preregistration, addressing endogeneity). There is significant untapped potential for collaborations across methodological traditions to explore and exploit new fields of gold. Collecting valid web data can enable marketing as a discipline to enhance its relevance and assert intellectual leadership on important emerging substantive topics that are also increasingly studied in fields such as computer science, information systems, and management science ([63]).
We would be remiss not to mention the nonmonetary costs of collecting web data via web scraping and APIs. While browsing the web is (mainly) free, researchers should not assume that collecting web data is costless. The prototype of a data collection can be ready and running in a matter of hours. Yet, researchers will often find out that the data collection does not work entirely as intended or encounter some of the challenges discussed in our methodological framework. Just like with any other method, the devil is in the details.
Web data democratize data access and make our discipline more inclusive for scholars who would otherwise find it difficult to obtain access to data. To further reduce entry barriers, it would be helpful to create incentives (e.g., journal space) for rich web data sets and their documentation, like the Billion Prices Project ([11]). Similarly, authors can make their algorithms or data available for other researchers by sharing code publicly or deploying API-based microservices that can increase their methods' adoption and offer unique opportunities for field experimentation. In summary, web data present a golden opportunity to examine important marketing questions, now and in the future.
sj-pdf-1-jmx-10.1177_00222429221100750 - Supplemental material for Fields of Gold: Scraping Web Data for Marketing Insights
Supplemental material, sj-pdf-1-jmx-10.1177_00222429221100750 for Fields of Gold: Scraping Web Data for Marketing Insights by Johannes Boegershausen, Hannes Datta, Abhishek Borah and Andrew T. Stephen in Journal of Marketing
Footnotes 1 Oded Netzer
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Marketing Science Institute (grant #4000678) and the Dutch Research Council (NWO #451–17–028).
4 Johannes Boegershausen https://orcid.org/0000-0002-1429-9344 Hannes Datta https://orcid.org/0000-0002-8723-6002
5 A comprehensive discussion of the legality and ethics of the automatic collection of web data is beyond the scope of this article. For a discussion of key issues, see [52] and [82], pp. 181−86). As legal experts are unlikely to be familiar with the various decisions that researchers make during data extraction, Web Appendix D provides a list of concerns to consider when seeking legal counsel.
6 Our discussion largely extends to APIs, unless noted otherwise. For example, with APIs, researchers use "endpoints" instead of "pages" (see Web Appendix A).
7 Table W6 in Web Appendix E contains URLs that accompany the Amazon example.
References Adjerid Idris , Kelley Ken. (2018), " Big Data in Psychology: A Framework for Research Advancement ," American Psychologist , 73 (7), 899 – 917.
Agarwal Saharsh , Sen Ananya. (2022), " Antiracist Curriculum and Digital Platforms: Evidence from Black Lives Matter ," Management Science , 68 (4), 2932 – 48.
Anderson Eric T. , Simester Duncan I.. (2014), " Reviews Without a Purchase: Low Ratings, Loyal Customers, and Deception ," Journal of Marketing Research , 51 (3), 249 – 69.
Arvidsson Adam , Caliandro Alessandro. (2016), " Brand Public ," Journal of Consumer Research , 42 (5), 727 – 48.
Barnes Christopher M. , Dang Carolyn T. , Leavitt Keith , Guarana Cristiano L. , Uhlmann Eric L.. (2018), " Archival Data in Micro-Organizational Research: A Toolkit for Moving to a Broader Set of Topics ," Journal of Management , 44 (4), 1453 – 78.
Berger Jonah , Humphreys Ashlee , Ludwig Stephan , Moe Wendy W. , Netzer Oded , Schweidel David A.. (2020), " Uniting the Tribes: Using Text for Marketing Insight ," Journal of Marketing , 84 (1), 1 – 25.
Blaseg Daniel , Schulze Christian , Skiera Bernd. (2020), " Consumer Protection on Kickstarter ," Marketing Science , 39 (1), 211 – 33.
8 Bruner Gordon C. (1990), " Music, Mood, and Marketing ," Journal of Marketing , 54 (4), 94 – 104.
9 Burtch Gordon , He Qinglai , Hong Yili , Lee Dokyun. (2022), " How Do Peer Awards Motivate Creative Content? Experimental Evidence from Reddit ," Management Science , 68 (5), 3488 – 4506.
Cavallo Alberto. (2017), " Are Online and Offline Prices Similar? Evidence from Large Multi-Channel Retailers ," American Economic Review , 107 (1), 283 – 303.
Cavallo Alberto , Rigobon Roberto. (2016), " The Billion Prices Project: Using Online Prices for Measurement and Research ," Journal of Economic Perspectives , 30 (2), 151 – 78.
Chen Eric Evan , Wojcik Sean P.. (2016), " A Practical Guide to Big Data Research in Psychology ," Psychological Methods , 21 (4), 458 – 74.
Chen Yubo , Wang Qi , Xie Jinhong. (2011), " Online Social Interactions: A Natural Experiment on Word of Mouth Versus Observational Learning ," Journal of Marketing Research , 48 (2), 238 – 54.
Chevalier Judith A. , Mayzlin Dina. (2006), " The Effect of Word of Mouth on Sales: Online Book Reviews ," Journal of Marketing Research , 43 (3), 345 – 54.
Datta Hannes , Knox George , Bronnenberg Bart J.. (2018), " Changing Their Tune: How Consumers' Adoption of Online Streaming Affects Music Consumption and Discovery ," Marketing Science , 37 (1), 5 – 21.
Datta Hannes , van Heerde Harald J. , Dekimpe Marnik G. , Steenkamp Jan-Benedict E.M.. (2022), " Cross-National Differences in Market Response: Line-Length, Price, and Distribution Elasticities in Fourteen Indo-Pacific Rim Economies ," Journal of Marketing Research , 59 (2), 251 – 70.
Du Rex Yuxing , Hu Ye , Damangir Sina. (2015), " Leveraging Trends in Online Searches for Product Features in Market Response Modeling ," Journal of Marketing , 79 (1), 29 – 43.
Du Rex Yuxing , Netzer Oded , Schweidel David A. , Mitra Debanjan. (2021), " Capturing Marketing Information to Fuel Growth ," Journal of Marketing , 85 (1), 163 – 83.
Edelman Benjamin. (2012), " Using Internet Data for Economic Research ," Journal of Economic Perspectives , 26 (2), 189 – 206.
etailinsights (2021), "How Many Etailers Are in the US?" infographic (accessed December 25, 2021), https://www.etailinsights.com/online-retailer-market-size.
Farris Paul W. , Bendle Neil T. , Pfeifer Phillip E. , Reibstein David J.. (2010), Marketing Metrics: The Definitive Guide to Measuring Marketing Performance. Upper Saddle River , NJ : Pearson Education.
Franklin Joshua. (2021), " Banking on Cannabis: The New Network of Lenders for a Semi-Legal Industry ," Financial Times (August 25) , https://on.ft.com/3BeC3Lt.
Frederick Shane , Lee Leonard , Baskin Ernest. (2014), " The Limits of Attraction ," Journal of Marketing Research , 51 (4), 487 – 507.
Gebru Timnit , Morgenstern Jamie , Vecchione Briana , Vaughan Jennifer Wortman , Wallach Hanna , Daumé III Hal , et al. (2020), "Datasheets for Datasets," arXiv preprint arXiv:1803.09010.
Gehman Joel , Grimes Matthew. (2017), " Hidden Badge of Honor: How Contextual Distinctiveness Affects Category Promotion among Certified B Corporations ," Academy of Management Journal , 60 (6), 2294 – 2320.
Gneezy Ayelet. (2017), " Field Experimentation in Marketing Research ," Journal of Marketing Research , 54 (1), 140 – 43.
Godes David , Mayzlin Dina. (2004), " Using Online Conversations to Study Word-of-Mouth Communication ," Marketing Science , 23 (4), 545 – 60.
Hanna Robin. (2018), "Amazon on a Positive Note: The End of Downvoting," (accessed December 1, 2021), https://sellics.com/blog-amazon-on-a-positive-note-the-end-of-downvoting/.
He Sherry , Hollenbeck Brett , Proserpio Davide. (2022), " The Market for Fake Reviews ," Marketing Science (published online February 25) , https://doi.org/10.1287/mksc.2022.1353.
Herhausen Dennis , Emrich Oliver , Grewal Dhruv , Kipfelsberger Petra , Schoegel Marcus. (2020), " Face Forward: How Employees' Digital Presence on Service Websites Affects Customer Perceptions of Website and Employee Service Quality ," Journal of Marketing Research , 57 (5), 917 – 36.
Hermosilla Manuel , Gutiérrez-Navratil Fernanda , Prieto-Rodríguez Juan. (2018), " Can Emerging Markets Tilt Global Product Design? Impacts of Chinese Colorism on Hollywood Castings ," Marketing Science , 37 (3), 356 – 81.
Homburg Christian , Theel Marcus , Hohenberg Sebastian. (2020), " Marketing Excellence: Nature, Measurement, and Investor Valuations ," Journal of Marketing , 84 (4), 1 – 22.
Howe Lauren C. , Monin Benoît. (2017), " Healthier Than Thou? 'Practicing What You Preach' Backfires by Increasing Anticipated Devaluation ," Journal of Personality and Social Psychology , 112 (5), 718 – 35.
Hsu Greta , Kovács Balázs , Koçak Özgecan. (2019), " Experientially Diverse Customers and Organizational Adaptation in Changing Demand Landscapes: A Study of US Cannabis Markets, 2014–2016 ," Strategic Management Journal , 40 (13), 2214 – 41.
Huang Ni , Burtch Gordon , Hong Yili , Polman Evan. (2016), " Effects of Multiple Psychological Distances on Construal and Consumer Evaluation: A Field Study of Online Reviews ," Journal of Consumer Psychology , 26 (4), 474 – 82.
Huang Yufeng. (2019), " Learning by Doing and the Demand for Advanced Products ," Marketing Science , 38 (1), 107 – 28.
Huber Joel , Payne John W. , Puto Christopher. (1982), " Adding Asymmetrically Dominated Alternatives: Violations of Regularity and the Similarity Hypothesis ," Journal of Consumer Research , 9 (1), 90 – 98.
Huff Aimee Dinnin , Humphreys Ashlee , Wilner Sarah J. S.. (2021), " The Politicization of Objects: Meaning and Materiality in the U.S. Cannabis Market ," Journal of Consumer Research , 48 (1), 22 – 50.
Humphreys Ashlee , Wang Rebecca Jen-Hui. (2018), " Automated Text Analysis for Consumer Research ," Journal of Consumer Research , 44 (6), 1274 – 1306.
Inman J. Jeffrey , Campbell Margaret C. , Kirmani Amna , Price Linda L.. (2018), " Our Vision for the Journal of Consumer Research : It's All About the Consumer ," Journal of Consumer Research , 44 (5), 955 – 59.
Jedidi Kamel , Schmitt Bernd H. , Sliman Malek Ben , Li Yanyan. (2021), " R2M Index 1.0: Assessing the Practical Relevance of Academic Marketing Articles ," Journal of Marketing , 85 (5), 22 – 41.
Judd Charles M. , Westfall Jacob , Kenny David A.. (2017), " Experiments with More Than One Random Factor: Designs, Analytic Models, and Statistical Power ," Annual Review of Psychology , 68 (1), 601 – 25.
Khessina Olga M. , Reis Samira , Verhaal J. Cameron. (2021), " Stepping out of the Shadows: Identity Exposure as a Remedy for Stigma Transfer Concerns in the Medical Marijuana Market ," Administrative Science Quarterly , 66 (3), 569 – 611.
Kim Tongil "TI" , KC Diwas. (2020), " Can Viagra Advertising Make More Babies? Direct-to-Consumer Advertising on Public Health Outcomes ," Journal of Marketing Research , 57 (4), 599 – 616.
Kozinets Robert V.. (2001), " Utopian Enterprise: Articulating the Meanings of Star Trek's Culture of Consumption ," Journal of Consumer Research , 28 (1), 67 – 88.
Kozinets Robert V.. (2002), " The Field Behind the Screen: Using Netnography for Marketing Research in Online Communities ," Journal of Marketing Research , 39 (1), 61 – 72.
Kozinets Robert V.. (2020), Netnography: The Essential Guide to Qualitative Social Media Research , 3rd ed. , London : SAGE Publications.
Kruijswijk Jules , van Emden Robin , Parvinen Petri , Kaptein Maurits. (2020), " StreamingBandit: Experimenting with Bandit Policies ," Journal of Statistical Software , 94 (9), 1 – 47.
Kübler Raoul , Pauwels Koen , Yildirim Gökhan , Fandrich Thomas. (2018), " App Popularity: Where in the World Are Consumers Most Sensitive to Price and User Ratings? " Journal of Marketing , 82 (5), 20 – 44.
Lakens Daniël. (2022), "Sample Size Justification," Collabra: Psychology, 8 (1): 33267.
Lambrecht Anja , Tucker Catherine , Wiertz Caroline. (2018), " Advertising to Early Trend Propagators: Evidence from Twitter ," Marketing Science , 37 (2), 177 – 99.
Landers Richard N. , Brusso Robert C. , Cavanaugh Katelyn J. , Collmus Andrew B.. (2016), " A Primer on Theory-Driven Web Scraping: Automatic Extraction of Big Data from the Internet for Use in Psychological Research ," Psychological Methods , 21 (4), 475 – 92.
Li Chenxi , Luo Xueming , Zhang Cheng , Wang Xiaoyi. (2017), " Sunny, Rainy, and Cloudy with a Chance of Mobile Promotion Effectiveness ," Marketing Science , 36 (5), 762 – 79.
Li Jingjing , Abbasi Ahmed , Cheema Amar , Abraham Linda B.. (2020), " Path to Purpose? How Online Customer Journeys Differ for Hedonic Versus Utilitarian Purchases ," Journal of Marketing , 84 (4), 127 – 46.
Li Xi , Shi Mengze , Wang Xin. (2019), " Video Mining: Measuring Visual Information Using Automatic Methods ," International Journal of Research in Marketing , 36 (2), 216 – 31.
Liu Jia , Toubia Olivier. (2018), " A Semantic Approach for Estimating Consumer Content Preferences from Online Search Queries ," Marketing Science , 37 (6), 930 – 52.
Liu Liu , Dzyabura Daria , Mizik Natalie. (2020), " Visual Listening In: Extracting Brand Image Portrayed on Social Media ," Marketing Science , 39 (4), 669 – 86.
Martin Kelly D. , Borah Abhishek , Palmatier Robert W.. (2017), " Data Privacy: Effects on Customer and Firm Performance ," Journal of Marketing , 81 (1), 36 – 58.
McAuley Julian. (2021), " Recommender Systems Datasets," (accessed February 2, 2021), https://cseweb.ucsd.edu/∼jmcauley/datasets.html.
McGraw A. Peter , Warren Caleb , Kan Christina. (2015), " Humorous Complaining ," Journal of Consumer Research , 41 (5), 1153 – 71.
Melumad Shiri , Meyer Robert. (2020), " Full Disclosure: How Smartphones Enhance Consumer Self-Disclosure ," Journal of Marketing , 84 (3), 28 – 45.
Moore Sarah G.. (2015), " Attitude Predictability and Helpfulness in Online Reviews: The Role of Explained Actions and Reactions ," Journal of Consumer Research , 42 (1), 30 – 44.
Moorman Christine , van Heerde Harald J. , Moreau C. Page , Palmatier Robert W.. (2019), " Challenging the Boundaries of Marketing ," Journal of Marketing , 83 (5), 1 – 4.
Morales Andrea C. , Amir On , Lee Leonard. (2017), " Keeping It Real in Experimental Research—Understanding When, Where, and How to Enhance Realism and Measure Consumer Behavior ," Journal of Consumer Research , 44 (2), 465 – 76.
Netzer Oded , Feldman Ronen , Goldenberg Jacob , Fresko Moshe. (2012), " Mine Your Own Business: Market-Structure Surveillance Through Text Mining ," Marketing Science , 31 (3), 521 – 43.
Neuendorf Kimberly A.. (2017), The Content Analysis Guidebook. Thousand Oaks , CA : SAGE Publications.
Oestreicher-Singer Gal , Libai Barak , Sivan Liron , Carmi Eyal , Yassin Ohad. (2013), " The Network Value of Products ," Journal of Marketing , 77 (3), 1 – 14.
Peng Jing , Agarwal Ashish , Hosanagar Kartik , Iyengar Raghuram. (2018), " Network Overlap and Content Sharing on Social Media Platforms ," Journal of Marketing Research , 55 (4), 571 – 85.
Proserpio Davide , Troncoso Isamar , Valsesia Francesca. (2021), " Does Gender Matter? The Effect of Management Responses on Reviewing Behavior ," Marketing Science , 40 (6), 1199 – 1213.
Ringel Daniel M. , Skiera Bernd. (2016), " Visualizing Asymmetric Competition Among More Than 1,000 Products Using Big Search Data ," Marketing Science , 35 (3), 511 – 34.
Rust Roland T. , Rand William , Huang Ming-Hui , Stephen Andrew T. , Brooks Gillian , Chabuk Timur. (2021), " Real-Time Brand Reputation Tracking Using Social Media ," Journal of Marketing , 85 (4), 21 – 43.
Schweidel David A. , Moe Wendy W.. (2014), " Listening in on Social Media: A Joint Model of Sentiment and Venue Format Choice ," Journal of Marketing Research , 51 (4), 387 – 402.
Shadish William , Cook Thomas D. , Campbell Donald Thomas. (2002), Experimental and Quasi-Experimental Designs for Generalized Causal Inference. Boston : Houghton Mifflin.
Sim Jaeung , Cho Daegon , Hwang Youngdeok , Telang Rahul. (2022), " Frontiers: Virus Shook the Streaming Star: Estimating the COVID-19 Impact on Music Consumption ," Marketing Science , 41 (1), 19 – 32.
Sridhar Shrihari , Srinivasan Raji. (2012), " Social Influence Effects in Online Product Ratings ," Journal of Marketing , 76 (5), 70 – 88.
Statista (2021), " Media Usage in an Internet Minute as of August 2021," (accessed January 10, 2022), https://www.statista.com/statistics/195140/.
Tellis Gerard J. , MacInnis Deborah J. , Tirunillai Seshadri , Zhang Yanwei. (2019), " What Drives Virality (Sharing) of Online Digital Content? The Critical Role of Information, Emotion, and Brand Prominence ," Journal of Marketing , 83 (4), 1 – 20.
Tirunillai Seshadri , Tellis Gerard J.. (2012), " Does Chatter Really Matter? Dynamics of User-Generated Content and Stock Performance ," Marketing Science , 31 (2), 198 – 215.
Tonietto Gabriela N. , Barasch Alixandra. (2020), " Generating Content Increases Enjoyment by Immersing Consumers and Accelerating Perceived Time ," Journal of Marketing , 85 (6), 83 – 100.
Toubia Olivier , Stephen Andrew T.. (2013), " Intrinsic vs. Image-Related Utility in Social Media: Why Do People Contribute Content to Twitter? " Marketing Science , 32 (3), 368 – 92.
Trusov Michael , Ma Liye , Jamal Zainab. (2016), " Crumbs of the Cookie: User Profiling in Customer-Base Analysis and Behavioral Targeting ," Marketing Science , 35 (3), 405 – 26.
Vanden Broucke Seppe , Baesens Bart. (2018), Practical Web Scraping for Data Science: Best Practices and Examples with Python. Berkeley , CA : Apress.
Van Heerde Harald J. , Moorman Christine , Page Moreau C. , Palmatier Robert W.. (2021), " Reality Check: Infusing Ecological Value into Academic Marketing Research ," Journal of Marketing , 85 (2), 1 – 13.
Villarroel Ordenes Francisco , Ludwig Stephan , de Ruyter Ko , Grewal Dhruv , Wetzels Martin. (2017), " Unveiling What Is Written in the Stars: Analyzing Explicit, Implicit, and Discourse Patterns of Sentiment in Social Media ," Journal of Consumer Research , 43 (6), 875 – 94.
Wang Xin , Lu Shijie , Li X.I. , Khamitov Mansur , Bendle Neil. (2021), " Audio Mining: The Role of Vocal Tone in Persuasion ," Journal of Consumer Research , 48 (2), 189 – 211.
Wang Yang , Chaudhry Alexander. (2018), " When and How Managers' Responses to Online Reviews Affect Subsequent Reviews ," Journal of Marketing Research , 55 (2), 163 – 77.
Weber Matthew S.. (2018), " Methods and Approaches to Using Web Archives in Computational Communication Research ," Communication Methods and Measures , 12 (2/3), 200 – 215.
Wedel Michel , Kannan P.K.. (2016), " Marketing Analytics for Data-Rich Environments ," Journal of Marketing , 80 (6), 97 – 121.
Wells William D. (2001), " The Perils of N = 1 ," Journal of Consumer Research , 28 (3), 494 – 98.
Wu Chunhua , Cosguner Koray. (2020), " Profiting from the Decoy Effect: A Case Study of an Online Diamond Retailer ," Marketing Science , 39 (5), 974 – 95.
Xu Heng , Zhang Nan , Zhou Le. (2020), " Validity Concerns in Research Using Organic Data ," Journal of Management , 46 (7), 1257 – 74.
Younkin Peter , Kuppuswamy Venkat. (2018), " The Colorblind Crowd? Founder Race and Performance in Crowdfunding ," Management Science , 64 (7), 3269 – 87.
Zervas Georgios , Proserpio Davide , Byers John W.. (2017), " The Rise of the Sharing Economy: Estimating the Impact of Airbnb on the Hotel Industry ," Journal of Marketing Research , 54 (5), 687 – 705.
~~~~~~~~
By Johannes Boegershausen; Hannes Datta; Abhishek Borah and Andrew T. Stephen
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 55- From Waste to Taste: How "Ugly" Labels Can Increase Purchase of Unattractive Produce. By: Mookerjee, Siddhanth (Sid); Cornil, Yann; Hoegg, JoAndrea. Journal of Marketing. May2021, Vol. 85 Issue 3, p62-77. 16p. 1 Color Photograph, 1 Diagram, 3 Graphs. DOI: 10.1177/0022242920988656.
- Database:
- Business Source Complete
From Waste to Taste: How "Ugly" Labels Can Increase Purchase of Unattractive Produce
Food producers and retailers throw away large amounts of perfectly edible produce that fails to meet appearance standards, contributing to the environmental issue of food waste. The authors examine why consumers discard aesthetically unattractive produce, and they test a low-cost, easy-to-implement solution: emphasizing the produce's aesthetic flaw through "ugly" labeling (e.g., labeling cucumbers with cosmetic defects "Ugly Cucumbers" on store displays or advertising). Seven experiments, including two conducted in the field, demonstrate that "ugly" labeling corrects for consumers' biased expectations regarding key attributes of unattractive produce—particularly tastiness—and thus increases purchase likelihood. "Ugly" labeling is most effective when associated with moderate (rather than steep) price discounts. Against managers' intuition, it is also more effective than alternative labeling that does not exclusively point out the aesthetic flaw, such as "imperfect" labeling. This research provides clear managerial recommendations on the labeling and the pricing of unattractive produce while addressing the issue of food waste.
Keywords: aesthetics; debiasing; food waste; labeling; sustainability; ugliness penalty
Consumers today expect the fruits and vegetables they purchase to "look good" all year round ([56]), a demand that farmers and retailers meet by discarding large amounts of produce that fails to meet aesthetic standards. A lot of produce fails these standards, not because of disease or damage that may negatively affect taste or nutritional quality, but simply because of inherent variation in natural growth. In fact, U.S. retailers throw away $15.4 billion of edible produce each year ([ 7]), and farmers discard up to 30% of their crops because of cosmetic imperfections ([ 5]). Food waste also has damaging consequences for the environment: 96% of wasted food is left to decompose in landfills, resulting in the release of methane, a greenhouse gas that traps solar radiation and contributes to climate change ([14]). In addition, food waste leads to a waste of other valuable resources: 1.4 billion hectares of land and 25% of the world's fresh water are used to grow produce that will be later thrown away ([22]; [40]).
Recent research has started to identify factors that might increase consumers' acceptance of unattractive produce, including marketing message framing ([16]; [48]; [53]), reduced pricing ([ 2]), and individual differences in environmental awareness ([11]; [36]; [57]). Most relevant to the present investigation, [16] proposed that consumers devalue unattractive produce, in part because imagining eating it negatively affects how they view themselves. Thus, a marketing message boosting consumers' self-esteem, "You Are Fantastic! Pick Ugly Produce!," increases purchase compared with a message simply stating, "Pick Ugly Produce." While this research identifies a straightforward managerial intervention, both the intervention and the comparison messages labeled unattractive produce "ugly," so the effect of "ugly" labeling in isolation is unclear.
We build on this prior work by investigating the effect of labeling unattractive produce as "ugly" (what we call "ugly" labeling) and comparing it with several alternative labels. We believe that further investigation is warranted because most retailers do not label unattractive produce in any specific way, and when they do, there is great variation in how unattractive produce is labeled. Indeed, although "ugly" labeling was employed by French retailer Intermarché in 2014, subsequent campaigns by other retailers have used more understated labels to promote unattractive produce, such as "imperfect" or labels that aim to positively frame visual atypicality, such as produce "with personality." To assess managers' beliefs regarding the use of "ugly" labeling, we interviewed 52 grocery store managers across North America with an average of 12 years of experience, and asked them to indicate which of four labeling options ("ugly," "imperfect," "with personality," or no specific label) they would use to promote unattractive produce sold at a discounted price. Of the 52 respondents, 46% stated that they would not use any label and that just the discount was enough, followed by 33% preferring "imperfect" labeling, 17% preferring "with personality" labeling, and only 4% preferring "ugly" labeling. We also asked them to select the worst option, and 75% mentioned "ugly" labeling.
Although managers do not see merit in "ugly" labeling, our research proposes that labeling unattractive produce as "ugly" can increase purchase, not only compared with no specific labeling, but also compared with more understated—and more popular—labeling such as "imperfect." We demonstrate the effectiveness of "ugly" labeling through a combination of field and online experiments, and elucidate the underlying mechanism. We show that consumers saddle unattractive produce with an "ugliness penalty" ([23], p.1181) that negatively affects expectations of the produce's key attributes—particularly tastiness—and thus affects purchase intentions. "Ugly" labeling corrects for these biased, negative expectations because it directly points out the aesthetic flaw as their source, in line with research that has shown a corrective effect when drawing observers' attention toward the source of a biased judgment ([49]; [51]). Further, while price discounts can motivate consumers to purchase unattractive produce ([ 2]), we show that "ugly" labeling is most effective when associated with a moderate price discount, because large discounts in conjunction with the "ugly" label send conflicting signals regarding the quality of the produce.
Our work makes several contributions. While prior research on two-sided persuasion ([12]; [42]) has shown that weak negative information added to a positive description can improve product evaluation, our research demonstrates that emphasizing negative information can have positive effects in the absence of any accompanying positive information. We also contribute to research that has investigated how awareness of influence has a corrective effect on biased judgment ([49]), extending prior findings to a consumption context.
Our research also provides guidance to managers on how to label and price unattractive produce. While retailers believe "imperfect" labeling or no specific labeling to be more effective than "ugly" labeling, we demonstrate that the opposite is the case. Our research may therefore partly explain the unsuccessful attempts by Whole Foods and Walmart to sell unattractive produce by labeling it "imperfect" ([ 9]). Our hope is that our research can assist managers in designing campaigns that can benefit their organizations and reduce food waste.
Unattractive produce is that which has a significant natural aesthetic deviation in shape and/or color from prototypical produce, but has no damage or disease that could affect safety, taste, or nutrition ([11]; [16]). [16] suggest that consumers reject unattractive produce because imagining eating such produce makes consumers view themselves as less attractive, less moral, less healthy, and so on. We propose that produce unattractiveness also influences how consumers view the produce itself. Extant research in social and consumer psychology shows that people stereotypically attribute a "beauty premium" to attractive individuals and objects, and, conversely, they saddle unattractive individuals and objects with an "ugliness penalty" that negatively affects perceptions beyond aesthetics. Indeed, physically unattractive individuals are perceived as less intelligent and less sociable than attractive individuals ([17]), and unattractive products are perceived as lower in quality and usability ([26]; [55]).
Regarding potential "ugliness penalty" effects in the realm of produce, we consider three categories of attributes: tastiness, healthiness, and naturalness. Tastiness refers to produce's hedonic, multisensory qualities: not only its flavor, but also its juiciness or crispiness ([ 3]). Healthiness refers to nutritional value. Naturalness refers to the absence of chemicals (e.g., pesticides, preservatives), which is characteristic of organic produce ([54]). In addition to these categories, there can be additional safety concerns in the case of moldy, rotten, or damaged produce. However, our definition of unattractive produce explicitly excludes these concerns as retailers have strict regulations preventing the sale of unsafe produce.
There is clear evidence in the literature of a positive association between aesthetic appeal and tastiness—thus, consumers should expect unattractive produce to be less tasty than attractive produce. Visual appearance, including color and shape, has a strong impact on inferences about a food's sensory quality ([11]; [29]). In the domain of produce, multiple studies have shown that a wide range of fruits and vegetables with atypical (vs. typical) colors were expected to be less tasty, although the actual taste was equivalent ([34]; [47]; [50]).
Research also points to a positive association between aesthetic appeal and healthiness. In the domain of produce (as opposed to many processed foods), consumers largely expect tasty foods to be healthier ([19]), so if unattractiveness negatively affects tastiness expectations, it should also negatively affect healthiness expectations. In line with this proposition, two studies have found that carrots with atypical (vs. typical) colors and bell peppers with uneven (vs. even) shape were expected to be not only less tasty, but also less healthy ([20]; [47]).
The association between attractiveness and naturalness is less straightforward. On the one hand, classic aesthetic patterns that are considered beautiful (e.g., the golden ratio, Fibonacci proportions) stem from the natural world ([41]; [45]), hinting at a possible positive correlation between perceived attractiveness and naturalness—as [20] found in the domain of food presentation. On the other hand, within the domain of fresh produce, cosmetic imperfections generally stem from nature ([16]); thus, there may rather be a negative correlation between attractiveness and naturalness expectations. In line with this perspective, several studies suggest that consumers expect natural, organic, and/or pesticide-free produce to be less attractive ([ 6]; [15]; [52]; [58]), especially eco-conscious consumers ([35]). Importantly, the expectation that unattractive produce is more natural is not biased, but due to the fact that the absence of chemicals (pesticides, preservatives) results in cosmetic imperfections ([ 6]).
As detailed previously, prior literature suggests that consumers expect unattractive produce to be less tasty and less healthy. Expectations regarding naturalness are less clear but tend toward a reverse effect given that natural/organic produce is more likely to be visually imperfect. Note that there is no factual reason to expect unattractive produce to be less tasty or less healthy; in fact, assuming that unattractive produce is more natural/organic, it should also be more tasty and more healthy, as suggested by a meta-analysis of 343 publications that concluded that organic foods present both gustatory and nutritive benefits ([ 4]). Thus, negative expectations regarding the tastiness and healthiness of unattractive produce are biased judgments, based on stereotypes such as those uncovered in research on the "ugliness penalty."
We posit that "ugly" labeling—that is, labeling unattractive produce "ugly"—will correct for negative, biased expectations that consumers may have about the tastiness or healthiness of unattractive produce. We propose that deliberately emphasizing the unattractiveness of the produce via "ugly" labeling acts as a signal that there is nothing "wrong" with the produce other than its appearance. Further, "ugly" labeling may make consumers reevaluate the diagnosticity of visual appearance for assessing tastiness and healthiness; that is, it will make them aware of the limited nature of their spontaneous objection to unattractive produce. This proposition is in line with research that has shown that "awareness of influence" triggers validity-driven corrections of attitudes ([49]). For instance, in the domain of aesthetics, [51] found that the aesthetic design of financial documents influenced participants' investment decisions, unless their attention was drawn to the design.
In summary, our central hypotheses are that "ugly" labeling will increase purchase of unattractive produce versus when no specific label is present and that this will occur by improving attribute expectations, in particular tastiness and healthiness. We do not expect naturalness expectations to be affected by "ugly" labeling, insofar as consumer beliefs about unattractive produce being more natural are not instances of biased judgment and as such do not need correction. Formally,
- H1: "Ugly" labeling (vs. no specific label) increases the likelihood that consumers purchase unattractive produce.
- H2: The effect of "ugly" labeling on the purchase of unattractive produce is mediated by improved attribute expectations, particularly tastiness and healthiness.
It is important to consider how "ugly" labeling compares or interacts with other interventions investigated by past research and/or employed in the field. First, research has shown that price discounts can motivate consumers to purchase unattractive produce ([ 2]; [11]); indeed, it is common practice to sell unattractive produce at a discount of up to 50% ([16]). However, we propose that the depth of the discount moderates the effectiveness of "ugly" labeling. Although consumers value the economic benefit of acquiring produce for a low price, a large discount may signal low quality ([18]), thereby hindering the corrective effect of "ugly" labeling. From a managerial perspective, this suggests that "ugly" labeling along with a moderate discount may be as effective as a steeper discount in motivating purchase.
- H3: The effect of "ugly" labeling on purchase is moderated by the depth of price discount, such that "ugly" labeling is most effective when associated with a moderate (vs. steep) discount.
Second, while "ugly" labeling has generated a lot of media attention, there is great variation in the marketplace on the labeling of unattractive produce. In fact, major brick-and-mortar retailers such as Whole Foods, Loblaws (in Canada), and Tesco (in England), as well as online retailers Imperfect Foods (imperfectfoods.com) and Perfectly Imperfect Produce (perfectlyimperfectproduce.com), have preferred to use a more understated label: "imperfect." Retailers have also utilized labels that attempt to positively frame visual atypicality, such as "produce with personality" (Giant Eagle), "misfit" (Hy-Vee), or "pickuliar" (Koger). Web Appendix W1 provides a nonexhaustive list of labels used by retailers all over the world.
We have argued that "ugly" labeling unambiguously points out the aesthetic flaw in the produce, making it clear that there are no deficiencies other than unattractiveness. For this reason, alternative labels that do not point to the aesthetic flaw should not improve attribute expectations as much as "ugly" labeling does, and should therefore be less effective at motivating purchase. We compare "ugly" labeling with "imperfect" labeling (because it is the most popular label and does not point directly to aesthetics) and "with personality" labeling (as an example of a label that positively frames visual atypicality).
- H4: "Ugly" labeling is more effective than alternative labeling that does not explicitly point out the aesthetic flaw.
We first test the effectiveness of "ugly" labeling in the field at a farmers' market (Study 1) and online, with incentive-compatible choices (Study 2). We then test our proposed mechanism—an increase in tastiness and healthiness expectations—through mediation (Study 3) and moderation (Study 4). We further test whether the effectiveness of "ugly" labeling is moderated by price discounts (Study 5). Finally, we compare the effectiveness of "ugly," "imperfect," and "with personality" labeling in an online study (Study 6a) and in a field study measuring online advertising click-throughs (Study 6b).
In Study 1, we tested the effect of "ugly" labeling at a farmers' market. We ran a stand selling attractive and unattractive vegetables, and manipulated the way the unattractive produce was labeled (either "ugly" or not) by changing signage every hour. This study was preregistered (http://aspredicted.org/blind.php?x=zg7hi5).
We obtained visually attractive and unattractive carrots, potatoes, and tomatoes from a local supplier. The unattractive vegetables were crooked or oddly shaped, but were not bruised or rotten. Fifty participants recruited on Amazon Mechanical Turk (MTurk)[ 5] rated photos of these vegetables from −3 = "Much less beautiful than normal" to +3 = "Much more beautiful than normal," with a midpoint of 0 = "Normal-looking." Participants judged the unattractive vegetables as less beautiful than the attractive vegetables (carrots: M = −1.20, SD = 1.46 vs. M =.34, SD = 1.17; p <.001; tomatoes: M = −.66, SD = 2.05 vs. M = 1.34, SD = 1.21; p <.001; potatoes: M = −.36, SD =.94 vs. M =.86, SD = 1.31; p <.001).
We conducted the study at a farmers' market in a major city in Canada over four consecutive Saturdays in September 2020. We ran a stall from 10:00 a.m. to 2:00 p.m. each day, for a total of 16 hours. The stall consisted of a tent and a table, to which was attached a poster stating the name of the stand ("Sam's Produce") and, per a request by the Association of Farmers' Markets, indicating that the stand is a student project selling certified organic produce grown by local farmers (see Web Appendix W2).
On top of the table were four baskets (see Figure 1): two contained unattractive produce, and two contained attractive versions of the same produce. We used potatoes and carrots on the first day, and potatoes and tomatoes on the other three days because carrots were no longer available from our supplier. The baskets had labels attached to them. We manipulated the labels associated with the unattractive produce, such that it was explicitly called "ugly" in the "ugly" label condition ("Ugly Potatoes," "Ugly Carrots," "Ugly Tomatoes") and not in the control condition ("Potatoes," "Carrots," "Tomatoes"). Across both conditions, the attractive produce was always labeled "Potatoes," "Carrots," and "Tomatoes." We changed the labels used for the unattractive produce every hour. On the first and third days, we displayed the "ugly" label first (from 10:00 to 11:00 a.m.), while on the second and fourth days, we displayed the control label first.
Graph: Figure 1. Stimuli.Notes: For Studies 2–4, we only show stimuli in the "ugly" label condition. Stimuli in the control condition were identical, but without the label "ugly." All stimuli can be found in Web Appendix W5.
Our pricing was consistent across conditions. Following prior research ([16]) and within the range of industry practice, the unattractive produce was sold at a discount of 25%. The attractive potatoes, carrots, and tomatoes were respectively priced at CAD $2.50, $2.50, and $3.00 per pound, while the unattractive potatoes, carrots, and tomatoes were respectively priced at CAD $1.88, $1.88, and $2.25 per pound.
The stall was managed by two research assistants blind to the hypotheses. The first research assistant was in charge of switching the labels and acted as the seller, handling transactions and communicating with shoppers following a script prepared in advance and kept constant across conditions. To maximize control, the research assistant was instructed to evade the issue if shoppers asked about the labels. A second research assistant recorded the transactions and also recorded the number of individuals per hour who stopped at the stand and engaged with the seller.
Across the four days, 938 individuals (in 573 groups) stopped at the stand, and 259 individuals (in 169 groups) engaged with the seller. Two-sided binomial tests indicated no significant differences in the number of individuals stopping or engaging with the seller across labeling conditions (all ps >.21). There were 113 buyers (defined as the individuals who handled money to purchase produce), but again, there was no significant difference in the number of buyers across labeling conditions (p =.38), although labeling affected what produce was bought, as shown in the following analyses.
It was unknown whether the buyers purchased produce for themselves or also for the individuals that accompanied them, or whether the buyers used their own money or the group's pooled money. Therefore, as indicated in the preregistration, all analyses controlled for the size of the group (if a buyer was alone, group size was 1; mean group size was 1.56). The analyses also controlled for the day of the study, given that we replaced carrots with tomatoes after the first day. All effects remained significant without these covariates (see Web Appendix W3).
In the control condition, 62.5% of buyers purchased unattractive produce and 56% purchased attractive produce (these proportions do not total 100% because some buyers purchased both types of produce). In the "ugly" label condition, 81.6% bought unattractive produce and 26.5% bought attractive produce. Two logistic regressions showed that labeling unattractive produce "ugly" (vs. control) significantly increased buyers' likelihood to purchase unattractive produce (z = 2.28, p =.02) and decreased their likelihood to purchase attractive produce (z = −3.06, p =.002).
We found converging results using spending as the dependent variable (see Figure 2). On average, in the control condition, buyers purchased $2.36 (SD = 2.49) of unattractive produce and $3.35 (SD = 4.34) of attractive produce. In the "ugly" label condition they purchased $3.41 (SD = 2.83) of unattractive produce and $1.78 (SD = 3.76) of attractive produce. A mixed regression of total spending, with label ("ugly" vs. control) as a between-subjects factor and appearance (attractive vs. unattractive) as a within-subject factor, found no significant main effects (all ps >.49) but a significant interaction effect (z = 2.83, p =.005). To interpret this interaction effect, we ran a multivariate regression with spending on unattractive produce and spending on attractive produce as dependent variables. Labeling unattractive produce "ugly" (vs. control) significantly increased spending on unattractive produce (t(107) = 2.16, p =.03) and marginally decreased spending on attractive produce (t(107) = −1.79, p =.08).
Graph: Figure 2. Spending by label conditions and visual appearance of produce (Study 1).†p <.1.*p <.05Notes: Error bars: ±1 SE.
Employing a field study setting at a farmers' market, we found that buyers were more likely to purchase unattractive produce (sold at a discounted price) over attractive produce when the unattractive produce was labeled "ugly," compared with a control condition in which unattractive produce was not labeled in any specific way. "Ugly" labeling also increased average spending on unattractive produce. These results verify H1 and go against managers' intuition that merely discounting unattractive produce, without using any specific label, should be more effective than using an "ugly" label.
In absolute terms, since the "ugly" label increased purchase of cheaper (unattractive) over more expensive (attractive) produce, less total revenue was generated in the "ugly" label condition ($254.50) than in the control condition ($364.90). However, given that attractive produce is more costly (not to mention its environmental cost), after including the cost at which we purchased the produce from the suppliers, gross profit margins were higher in the "ugly" label condition ($39.30) than in the control condition ($26.00).
In Study 2, we further test the effectiveness of "ugly" labeling in the context of produce boxes purchased online. Participants decided whether to buy a box of unattractive produce or a box of attractive produce (or nothing at all), and we manipulated the label for the unattractive produce (either "ugly" or not). We used an incentive-compatible design, and this study was preregistered (https://aspredicted.org/blind.php?x=hd3iu5). All questions for this and all subsequent studies appear in Web Appendix W4.
Our stimuli consisted of a photo of attractive oranges, apples, cucumbers, and carrots, and a photo of the same items but visually unattractive (see Figure 1). Fifty MTurk participants judged the unattractive produce less beautiful than the attractive produce (M = −1.90, SD = 1.39 vs. M = 1.20, SD = 1.12; p <.001).
Because this study involved incentive-compatible choices and to increase the power of the study ([37]), we only recruited participants who would potentially be interested in purchasing produce online. We posted an ad on Facebook (shown in Web Appendix W5) targeted at people living in the United States, between 18 and 64 years of age, with an interest (determined by the Facebook pages they "like") in "Online grocer," "FreshDirect," and "AmazonFresh." The ad indicated that our research team was looking for participants, and in exchange for completing a survey, they would enter a lottery to win $30 or produce boxes. The ad never mentioned "ugly" produce to avoid recruiting participants with a specific interest in such produce. We advertised the study until 303 participants completed it (Mage = 45.20 years, SD = 12.83 years; 93% female). The high proportion of female participants is likely due to Facebook ad targeting. Participants were randomly assigned to one of two conditions: either "ugly" labeling or control.
The ad led to a study hosted on Qualtrics. In the consent form, we indicated that the chance of winning the lottery was about 15%. Then, as a cover story, participants answered 25 questions with two possible answers, reportedly designed to measure personality (e.g., "Would you rather go to a movie or to dinner alone?"). In the 25 questions, we embedded two attention checks that automatically excluded participants who failed, before they could participate in the actual study (see Web Appendix W4).
Next, participants read, "You will now enter a lottery to win $30. The prize will be paid via PayPal, Amazon eGift card or other online means of payment of your choice. If you win, you can decide to keep the $30, or to use some of this money to purchase a box of fruits & veggies delivered to your doorstep by one of our trusted partners. Produce sold by our partners meets USDA [U.S. Department of Agriculture] safety standards. We managed to get special deals on two boxes of fruits & veggies." We provided illustrations and information about these two boxes. Box 1 featured attractive oranges, apples, carrots, and cucumbers and indicated "SPECIAL PRICE: $20 (regular price: $35)," and Box 2 featured the same produce but aesthetically unattractive and indicated "SPECIAL PRICE: $15 (regular price: $25)." The label used for the attractive produce was always "Fruits and Veggies." We manipulated between subjects the label used for the unattractive produce: either "Ugly Fruits and Veggies" in the "ugly" label condition or "Fruits and Veggies" in the control condition. We show the stimulus used in the "ugly" label condition in Figure 1, and all stimuli in Web Appendix W5.
Participants were asked to indicate in advance what they would do if they won the lottery: "I want the full $30 cash prize without buying anything," or "I want Box 1 at a special price of $20 delivery included, and I get the remainder of $10 cash," or "I want Box 2 at a special price of $15 delivery included, and I get the remainder of $15 cash."
We programmed the survey such that 15% of the participants won the lottery. The winners provided their email address, and we followed up by sending them online cash payments and/or online coupons of produce box delivery companies (Farmbox Direct, Farm Fresh to You, Hungry Harvest, and Perfectly Imperfect Produce), depending on what prize they selected. If none of the companies could deliver to their address, we sent them online cash payments.
In the "ugly" label (vs. control) condition, 41.1% of participants (vs. 26.3%) decided to purchase the box of unattractive produce, 7.9% (vs. 23.0%) decided to purchase the box of attractive produce, and 51.0% (vs. 50.7%) preferred to keep the cash (see Web Appendix W6). A logistic regression showed that the likelihood of purchasing a box over keeping the cash was not different across conditions (p =.95). However, the "ugly" label (vs. control) significantly increased the likelihood of purchasing the box of unattractive produce over the box of attractive produce (z = 3.86, p <.001).
In an online study with an incentive-compatible measurement of choice and where participants had the option not to purchase any produce, we found that "ugly" labeling made consumers purchase unattractive, rather than attractive produce, in line with H1. As in Study 1, "ugly" labeling influenced produce choice, but not overall produce purchase.
In Study 3, we test our proposed mechanism: we posit that consumers have negative expectations regarding the tastiness and healthiness (but not the naturalness) of unattractive produce, and that "ugly" labeling improves these expectations. The study also addresses several alternative explanations for the positive effect of "ugly" labeling on choice. For example, it is possible that "ugly" labeling is perceived as original, surprising, or amusing ([13]). Likewise, "ugly" labeling may anthropomorphize unattractive produce, increasing sympathy ([32]; [48]). "Ugly" labeling might also enhance the perceived credibility of the seller by conveying honest information about the produce. Finally, "ugly" labeling might affect self-perceptions ([16]). We thus measure each of these constructs to test their potential role. The study was preregistered (http://aspredicted.org/blind.php?x=ah63mh).
We used photos of attractive and unattractive cucumbers. Fifty MTurk participants judged the unattractive cucumbers less beautiful than the attractive ones (M = −.84, SD = 1.54 vs. M = 1.26, SD = 1.24; p <.001).
We assigned 320 MTurk participants (Mage = 36.21 years, SD = 11.94 years; 53% female) to one of two between-subjects conditions: "ugly" label versus control. Participants were shown photos of baskets of attractive and unattractive cucumbers ostensibly sold by the same vendor and meeting USDA safety standards. Across conditions the attractive cucumbers were called "Type A" and priced at $1.26 per pound, and the unattractive cucumbers were called "Type B" and priced at $.95 per pound. We manipulated the label attached to the basket of unattractive cucumbers: "Ugly Cucumbers" in the "ugly" label condition versus "Cucumbers" in the control condition. The attractive cucumbers were always labeled "Cucumbers." The stimuli for the "ugly" label condition appear in Figure 1, and all stimuli in Web Appendix W5.
Participants indicated which produce they would purchase on a five-point scale ranging from 1 = "Definitely Cucumbers A" to 5 = "Definitely Cucumbers B," with a midpoint of 3 = "I would be indifferent."
We then measured produce attribute expectations ([28]) with a scale composed of four taste-related items (tasty, flavorful, juicy, crisp), three health-related items (healthy, nutritional, full of vitamins), four nature-related items (natural, free of pesticides, free of preservatives, organic), and three other items (ripe, fresh, clean). For each item, we asked participants to rate their expectations of Cucumbers B relative to Cucumbers A on a seven-point scale ranging from −3 = "Much more negative than Cucumbers A" to 3 = "Much more positive than Cucumbers A," with a midpoint of 0 = "Not different from Cucumbers A."
The next measurements were used to test alternative explanations. We distributed the negative self-perception scale developed by [16]: participants imagined eating Cucumbers B (i.e., the unattractive ones) and rated whether they felt 16 self-perceptions (e.g., worthless, immoral) on a seven-point scale (1 = "Not at all," and 7 = "Very much"). Credibility was assessed with four items (e.g., "I think the seller of this vegetable is trustworthy") adapted from [31] and evaluated on a seven-point scale (1 = "Strongly disagree," and 7 = "Strongly agree"). We measured anthropomorphic perceptions by asking participants to rate whether Cucumbers B reminded them of humanlike features ([32]) on a five-point scale (1 = "Not at all," and 5 = "To a great extent"). We also asked participants whether they "feel sorry," "feel compassion," and "feel sympathy" for Cucumbers B on the same five-point scale. We measured whether participants perceived the image of cucumbers B to be original, surprising, and funny (with two items: funny and amusing) on a five-point scale (1 = "Not at all," and 5 = "To a great extent"). Each construct was presented on a separate, randomized page with reminders of the stimuli.
At the end of the study, as an attention check, we asked participants to recall the prices of Cucumbers A and B. There were five possible answers and only one correct answer; those who answered incorrectly were excluded from analysis. We used the same preregistered attention check and exclusion rule across all MTurk studies (Studies 3–6a). In Web Appendix W7, we report results with and without data exclusion; the results are consistent.
Twenty-eight participants (8.8%) failed the attention check and were excluded from analysis.
An analysis of variance (ANOVA) of choice likelihood indicated that "ugly" labeling (vs. control) increased the likelihood of choosing the unattractive produce over the attractive produce (M = 3.01, SD = 1.44 vs. M = 2.54, SD = 1.42; F( 1,290) = 7.90, p =.005).
Figure 3 displays expectations about each attribute across conditions. We created indices of tastiness expectations (α =.92), healthiness expectations (α =.92), and naturalness expectations (α =.91) and performed the analyses using these indices. Although "ripe," "fresh," and "clean" contribute to taste and nutritive quality, they are conceptually distinct from the tastiness and healthiness constructs ([27]; [43]), so we did not include these items in the indices; note that "ugly" labeling did not significantly improve "ripe," "fresh," and "clean" expectations (all ps >.10).
Graph: Figure 3. Attribute expectations of visually unattractive produce by label conditions (Study 3).*p <.05.**p <.01.Notes: Error bars: ±1 SE. "Tastiness," "healthiness," and "naturalness" are indices composed of the items in the brackets.
The tastiness index was well below zero in the control condition (p <.001), indicative of an "ugliness" penalty effect on taste expectations of unattractive produce. The "ugly" label (vs. control) improved the tastiness index (M = −.08, SD = 1.06 vs. M = −.45, SD = 1.12; F( 1, 290) = 8.45, p =.004). Healthiness expectations in the control condition were not significantly different from zero (p =.93). Still, the "ugly" label (vs. control) significantly increased the healthiness index (M =.23, SD =.94 vs. M = −.01, SD =.97; F( 1, 290) = 4.59, p =.03), although to a smaller extent than the tastiness index. We found an "ugliness premium" for naturalness, with expectations above zero in the control condition (p =.01), and the "ugly" label did not further increase the naturalness index (M =.38, SD = 1.03 vs. M =.26, SD = 1.20; p =.34).
We conducted a mediation analysis ([25], Model 4) with the tastiness, healthiness, and naturalness indices as parallel mediators, choice likelihood as the dependent variable, and the label manipulation as the independent variable. As shown in Figure 4, tastiness had the strongest mediating effect (b =.17, SE =.07, 95% confidence interval [CI] = [.055,.343]), healthiness had a weaker, although significant, mediating effect (b =.08, SE =.05, 95% CI = [.008,.226]), and naturalness did not have a mediating effect (95% CI = [−.029,.137]). We conducted the same analyses for comparable conditions in Studies 4, 5, and 6a and present them in Figure 4.
Graph: Figure 4. Mediation by tastiness, healthiness, and naturalness expectations (Studies 3 to 6a).*p <.05.**p <.01.***p <.001.Notes: Parallel mediation models ([25], Model 4) were used for Studies 3, 4, 5, and 6a. We only consider comparable "ugly" label and control label conditions: for Study 4, we exclude the condition in which participants received the corrective message; for Study 5, we exclude the two larger discount (40% and 60%) conditions; for Study 6a, we exclude the "imperfect" and "with personality" label conditions. The statistics inside the figure are unstandardized regression coefficients. The 95% confidence intervals of the indirect effects below the figure are estimated with 5,000 bootstrapped samples.
We found that "ugly" labeling (vs. control) did not significantly affect self-perceptions (M = 3.17, SD =. 76 vs. M = 3.31, SD =.85; F( 1, 290) = 2.13, p =.15) or any of the measures of anthropomorphic perceptions or sympathy (all ps >.3).
"Ugly" labeling (vs. control) marginally improved credibility (α =.78; M = 5.40, SD =.93 vs. M = 5.18, SD = 1.01; F( 1, 290) = 3.61, p =.06), but credibility did not mediate the effect of labeling on choice based on a 95% confidence interval (b =.09, SE =.05, 95% CI = [−.004,.188]).
Images with "ugly" (vs. control) labels were judged funnier (r =.90; M = 2.41, SD = 1.05 vs. M = 2.02, SD =.99; F( 1, 290) = 10.54, p =.001) and more original (M = 2.65, SD = 1.05 vs. M = 2.41, SD = 1.04; F( 1, 290) = 4.25, p =.04), but not more surprising (p =.24). However, the effect of "ugly" labeling on produce choice was not mediated by humor (95% CI = [−.110,.039]) or by originality (95% CI = [−.046,.053]).
Study 3 demonstrated that "ugly" labeling increases the choice likelihood of unattractive produce (H1) and that this effect is mediated by an increase in tastiness expectations and, to a somewhat smaller extent, healthiness expectations (H2). Unattractive produce without any specific label was judged less tasty than attractive produce, in line with past research, although unattractive produce was judged just as healthy. We return to this point in the "General Discussion" section. As a preview, we find across Studies 3 through 6a that people judge unattractive produce less tasty than attractive produce, but not necessarily less healthy; thus, the effect of "ugly" labeling on choice is mediated to a larger extent by tastiness expectations than by healthiness expectations.
Naturalness expectations, credibility, self-perceptions, originality, surprise, humor, and anthropomorphic perceptions did not explain the effectiveness of "ugly" labeling.
To confirm the causality chain tested via mediation in Study 3 ("ugly" labeling → taste expectations; healthiness expectations → choice), Study 4 manipulated the mediator ([44]). We informed half the participants that aesthetic differences across produce do not pertain to differences in taste or healthiness. If the effectiveness of "ugly" labeling is due to improved taste or healthiness expectations, explicitly addressing those expectations should have the same effect as the "ugly" label. The study was preregistered (http://aspredicted.org/blind.php?x=br2xi3).
A total of 423 MTurk participants (Mage = 36.04 years, SD = 12.11 years; 54% female) were assigned to a 2 (label: ugly vs. control/no descriptor) × 2 (message: "no other difference than visual" vs. control/no message) between-subjects design.
Participants had to choose between purchasing attractive or unattractive cucumbers. The scenario, stimuli, manipulation of "ugly" labeling, prices, and measurement of choice likelihood were identical to those in Study 3. In addition to the labeling manipulation, we manipulated a message such that half the participants read the following text before seeing the stimuli: "Please be aware that although the two types of cucumbers that you will see look different, these differences in visual appearance do not pertain to any differences other than visual: for instance, they have similar gustatory or nutritive qualities."
Then, participants completed a shorter version of the attribute expectations scale: they evaluated the expected taste (tasty, flavorful, juicy, crisp) and healthiness (healthy, nutritional, full of vitamins) of the unattractive produce, relative to the attractive produce.
Twenty-two participants (5.1%) failed the attention check and were excluded.
An ANOVA of choice likelihood revealed a main effect of the message manipulation (F( 1,397) = 9.50, p =.002) and a significant message × label interaction (F( 1,397) = 4.54, p =.03). The main effect of label was not significant (p =.12). When there was no message, in line with Study 3, the "ugly" label (vs. control label) significantly increased choice likelihood of unattractive cucumbers (M = 3.29, SD = 1.42 vs. M = 2.77, SD = 1.36; t(397) = 2.64, p =.009). However, when participants were exposed to the "no other difference than visual" message, the "ugly" label (vs. control label) no longer had a significant impact (M = 3.43, SD = 1.47 vs. M = 3.51, SD = 1.40; p =.70). In addition, comparing choice likelihood across the "ugly" label/no message condition and either of the two conditions in which participants received the "no other difference than visual" message, we found no significant differences (all ps >.30). In other words, merely labeling unattractive produce "ugly" had a similar effect as informing consumers that visual differences do not pertain to other attribute differences.
We created healthiness (α =.93) and tastiness (α =.93) indices and tested a moderated mediation model ([25], Model 7) with the label manipulation as the independent variable, choice likelihood as the dependent variable, healthiness and tastiness expectations as parallel mediators, and the message moderating the link between the independent variable and the mediators. The indices of moderated mediation were significant for both tastiness (95% CI = [.10,.46]) and healthiness (95% CI = [.02,.25]). The results, reported in detail in Web Appendix W8, replicated those of Study 3: among participants who did not receive the additional message (but not among those who did), the effect of "ugly" labeling on choice was mediated by tastiness, and to a smaller extent (and marginally significantly) by healthiness, as shown in Figure 4.
Merely labeling unattractive produce "ugly" had a similar effect to informing consumers that visual differences do not pertain to healthiness or tastiness differences. This provides support for our argument that "ugly" labeling increases choice of unattractive produce because it improves expectations about tastiness and healthiness of unattractive produce (H2).
Across all studies presented herein, unattractive produce is sold at a 25% to 33% discount compared with attractive produce. Given the industrywide practice of discounting unattractive produce ([ 2]), Study 5 tests whether the depth of discount moderates the effectiveness of "ugly" labeling. We propose that "ugly" labels are more effective for moderate discounts because a large discount may signal low quality, thereby hindering the positive effect that "ugly" labels have on taste and healthiness expectations and thus on purchase (H3). The study was preregistered (https://aspredicted.org/blind.php?x=nc67z7).
A total of 709 MTurk participants (Mage = 35.38 years, SD = 11.37 years; 47% female) were assigned to a 3 (discount: 20% vs. 40% vs. 60%) × 2 (label: ugly vs. control) between-subjects design.
All participants saw an ad for two produce boxes, described as customizable boxes of fruits and vegetables that meet USDA safety standards. The ad (shown in Web Appendix W5) depicted examples of produce contained in each of the two boxes, one featuring attractive oranges, apples, carrots, and cucumbers and the other featuring the same produce but aesthetically unattractive (we used the same photos as in Study 2). The label used for the attractive produce was always "Fruits and Vegetables." We manipulated the label used for the unattractive produce: either "Ugly Fruits and Vegetables" ("ugly" label condition) or "Fruits and Vegetables" (control condition). The box with attractive produce was always priced at $20 for 5 pounds of produce. We manipulated the price of the box with unattractive produce: $16 with a "20% OFF" tag, $12 with a "40% OFF" tag, or $8 with a "60% OFF" tag. To facilitate measurement, the boxes were called "Box 1" (at the top of the ad) and "Box 2" (at the bottom); the position of the unattractive and attractive boxes was counterbalanced across participants.
Participants indicated which produce box they would rather purchase on a five-point scale ranging from 1 = "Definitely Box 1" to 5 = "Definitely Box 2," with a midpoint of 3 = "I would be indifferent." Because of the counterbalance, we reverse-coded the answers for half the participants, such that a higher number on the scale would always indicate preference for the box of unattractive produce. Then, they completed the full attribute expectations scale for the unattractive produce, relative to the attractive produce. Unlike Studies 3 and 4, this scale also included the item "sweet (fruits only)," which was also used to create the tastiness index.[ 6]
One hundred nineteen participants (16.8%) failed the attention check and were excluded.[ 7]
An ANOVA of choice likelihood with label, discount, counterbalance, and their interactions as independent variables showed that counterbalancing interacted with none of the manipulated factors (all ps >.57). We thus collapsed the results across counterbalance conditions and repeated the ANOVA, which revealed significant main effects of label (F( 1,586) = 4.24, p =.04) and price discount (F( 1,586) = 12.27, p <.001), and a significant label × discount interaction effect (F( 1,586) = 8.54, p =.004).
As shown in Figure 5, contrast analyses revealed that the "ugly" label (vs. control) significantly increased the choice likelihood of unattractive produce when the price discount was 20% (M = 2.56, SD = 1.37 vs. M = 1.94, SD = 1.21; t(584) = 3.13, p =.002). When the discount was 40%, the "ugly" label (vs. control) had a directionally positive but nonsignificant impact on choice (M = 2.66, SD = 1.42 vs. M = 2.36, SD = 1.44; t(584) = 1.49, p =.13). When the discount was 60%, the "ugly" label (vs. control) had a nonsignificant impact (p =.31). Also note that "ugly" labeling coupled with a low discount (20%) was just as effective as providing a steep price discount (60%) with or without the "ugly" label (all ps >.16).
Graph: Figure 5. Choice likelihood of unattractive produce box by label and price discount conditions (Study 5).**p <.01.Notes: Error bars: ±1 SE.
We created tastiness (α =.95), healthiness (α =.91), and naturalness (α =.92) expectations indices and tested a moderated mediation model ([25], Model 7) with the label manipulation as the independent variable; choice likelihood as the dependent variable; tastiness, healthiness, and naturalness expectations as parallel mediators; and discount moderating the link between the independent variable and the three mediators. Discount was treated as a continuous variable, given that the discounts increased linearly across conditions. There were significant main effects of "ugly" labeling on tastiness (t(586) = 3.34, p <.001) and healthiness (t(586) = 2.64, p =.009), and marginally significant label × discount interaction effects on tastiness (t(586) = −1.91, p =.056) and healthiness (t(586) = −1.95, p =.052); the other effects (including those on naturalness) were nonsignificant (all ps >.10). The indices of moderated mediation were significant for both tastiness (95% CI = [.003,.097]) and healthiness (95% CI = [.002,.080]), but not for naturalness (95% CI = [−.002,.037]). We thus do not discuss naturalness further.
When the discount was 20%, the results mirrored what we found in Studies 3 and 4: the "ugly" label (vs. control) improved tastiness expectations (M =.36, SD = 1.34 vs. M = −.36, SD = 1.77; t(584) = 3.32, p =.001) and healthiness expectations (M =.66, SD = 1.15 vs. M =.16, SD = 1.48; t(584) = 2.64, p =.008), and, as shown in Figure 4, the effect of "ugly" labeling on choice was mediated by tastiness (b =.19, SE =.10, 95% CI = [.052,.424]) and healthiness (b =.13, SE =.08, 95% CI = [.023,.344). When the discount was 40%, the effects were weaker: the "ugly" label (vs. control) marginally improved tastiness expectations (M =.10, SD = 1.43 vs. M = −.32, SD = 1.64; t(584) = 1.90, p =.06) and healthiness expectations (M =.43, SD = 1.22 vs. M =.02, SD = 1.53; t(584) = 2.11, p =.04); the effect of "ugly" labeling on choice was mediated by tastiness (b =.14, SE =.09, 95% CI = [.007,.386]) but not significantly by healthiness (b =.04, SE =.08, 95% CI = [−.071,.249]). When the discount was 60%, none of these effects were significant (all ps >.54; 95% CIs include zero).
In Study 5, "ugly" labeling was found to be most effective when associated with a moderate (vs. steeper) discount, in line with H3. Indeed, "ugly" labeling (vs. control) increased choice likelihood of unattractive produce via improved health and taste expectations when the price discount was 20%, but not when the price discount was 60%.
"Ugly" labeling allows retailers to avoid excessively discounting the price of unattractive produce: participants were just as likely to choose unattractive produce when it was labeled "ugly" and had a 20% discount as when it had a 60% discount (with or without "ugly" labeling). Indeed, while a steeper price discount naturally increases choice likelihood (as in the control condition), this was not the case in the "ugly" label condition. Although more affordable, produce with a 60% discount and an "ugly" label was expected to be less tasty and less healthy than produce with a 20% discount and an "ugly" label (tastiness: M = −.20, SD = 1.51 vs. M =.36, SD = 1.34, t(584) = 2.54, p =.01; healthiness: M =.27, SD = 1.35 vs. M =.66, SD = 1.15, t(584) = 2.06, p =.04). This is in line with our contention that steep discounts send a signal conflicting with the "ugly" label regarding produce quality.
In Study 6a we compare the effectiveness of "ugly" labeling with two other labels: "with personality" and "imperfect." "Imperfect" is used by numerous retailers and was the most popular label choice (beside no specific label) in our interview with grocery store managers. While this study has important practical implications, it also allows a further test of our theory that "ugly" labeling is most effective because it points out that the flaw in the produce is aesthetic, compared with "imperfect" and "with personality" labeling (H4). This study was preregistered (http://aspredicted.org/blind.php?x=zx2pq2).
A total of 440 MTurk participants (Mage = 34.78 years, SD = 11.73 years; 49% female) were assigned to one of four label conditions: "ugly," "imperfect," "with personality," or control.
The scenario, the stimuli (shown in Web Appendix W5), and the questions were similar to Study 5. However, unlike Study 5, the prices of the boxes were fixed at $18 for the box of attractive produce and $12 for the box of unattractive produce, and there was no discount tag. There were four labeling conditions for the box of unattractive produce: "Ugly Fruits and Vegetables," "Imperfect Fruits and Vegetables," "Fruits and Vegetables with Personality," or just "Fruits and Vegetables" (control).
Forty-nine participants (11.1%) failed the attention check and were excluded.
An ANOVA of choice likelihood revealed a significant effect of labeling (F( 3, 387) = 4.40, p =.005). As shown in Web Appendix W9, the "ugly" label increased choice of unattractive produce (M = 2.82, SD = 1.49) significantly compared with the control label (M = 2.08, SD = 1.37; F( 1, 387) = 12.98, p <.001), marginally significantly compared with the "imperfect" label (M = 2.42, SD = 1.36; F( 1, 387) = 3.62, p =.058), and directionally compared with the "with personality" label (M = 2.51, SD = 1.50; F( 1, 387) = 2.26, p =.13).
Although "imperfect" and "with personality" were less effective than "ugly," they still increased choice of unattractive produce compared with the control label. The "imperfect" versus control contrast was marginally significant (F( 1, 387) = 2.94, p =.09), and the "with personality" versus control contrast was significant (F( 1, 387) = 4.43, p =.04).
We created tastiness (α =.96), healthiness (α =.93), and naturalness (α =.93) expectation indices and tested parallel mediations ([25], Model 4). The effects of the "ugly" label (vs. control) were in line with Studies 2–4: a significant improvement in tastiness (M = −.50, SD = 1.44 vs. M = −.97, SD = 1.36; t(387) = 2.53, p =.01), a marginally significant improvement in healthiness (α =.93, M = −.09, SD = 1.26 vs. M = −.39, SD = 1.13; t(387) = 1.83, p =.07), and a nonsignificant change in naturalness (p =.23). As shown in Figure 4, tastiness mediated the effect of "ugly" labeling on choice (b =.17, SE =.09, 95% CI = [.034,.394]); however, neither healthiness nor naturalness were significant mediators (95% CI = [−.035,.136], 95% CI = [−.012,.109], respectively).
"Imperfect" labeling (vs. control) did not have any significant impact on tastiness, healthiness, and naturalness expectations (all ps >.12), and none of these categories of expectations were significant mediators (95% CIs include zero).
"With personality" labeling (vs. control) positively affected tastiness (M = −.60, SD = 1.22 vs. M = −.97, SD = 1.36; t(387) = 2.05, p =.04), and tastiness mediated the effect of "with personality" labeling on choice (b =.20, SE =.11, 95% CI = [.013,.429]). However, "with personality" labeling did not significantly influence healthiness (p =.10) or naturalness (p =.70), and these categories were not significant mediators. We discuss these effects after Study 6b.
Study 6b compares the effectiveness of the three labeling interventions in the field through ads posted on social media platforms. We used Facebook Ads Manager's Split Test (also called "A/B Test") to compare the effectiveness of different versions of an ad on click-through rates, holding all other factors constant ([ 8]; [24]; [33]).
As we were measuring click-throughs in advertising, rather than relative choice, we focused solely on ads with unattractive produce, and we only included ads with specific labels, namely "ugly," "imperfect," and "with personality" (i.e., there was no condition without a specific label). This study was preregistered (https://aspredicted.org/blind.php?x=rr88f8).
We created an ad for a "produce box" of unattractive produce using the same photos of unattractive produce as in Studies 2 and 6a. The three versions of the ad each had a different label written on the box: "Ugly Fruits and Veggies," "Imperfect Fruits and Veggies," or "Fruits and Veggies with Personality" (as shown in Figure 1). We added text at the top of the ad that reinforced the label manipulation; for instance, in the "ugly" label condition, the text was "Ugly fruits and vegetables delivered to your door, in a customizable box. Get 30% off your first order today." The call to action for the ad was a button labeled "Get Offer."
Facebook Ads Manager enabled us to determine the audience for the ad: people living in the United States, between 18 and 64 years of age, with an interest in "Online grocer," "FreshDirect," and "AmazonFresh." The ad was placed on social media platforms Facebook and Instagram, and users were randomly assigned to see one of the three versions of the ad. We programmed the campaign such that the ad would be delivered for four days, for a total cost of $600 ($200 per version). This amount was determined based on an estimated test power of 80%. Additional technical specifications appear in Web Appendix W10.
Our ads were viewed a total of 42,463 times: 14,269 in the "ugly" condition; 14,199 in the "imperfect" condition; and 13,995 in the "with personality" condition. Thus, there was no imbalance in number of views across conditions (all ps >.17).
There were 438 clicks in the "ugly" condition, 373 in the "imperfect" condition, and 404 in the "with personality" condition. We computed the click-through rate (CTR), defined as the number of clicks divided by the number of impressions ([33]), for each condition and analyzed the differences in CTR across conditions. As shown in Web Appendix W9, the "ugly" ad generated the highest CTR (3.07%) and the lowest cost per click ($.46). In line with Study 6a, the "imperfect" ad was the least effective (CTR = 2.62%; cost per click = $.54) and the "with personality" ad was in between (CTR = 2.89%; cost per click = $.50). The difference in CTR between the "ugly" ad and the "imperfect" ad was significant (χ2 = 5.04, p =.02). The differences between the "ugly" and the "with personality" ad, and between the "imperfect" and the "with personality" ad were not significant (χ2 =.82, p =.37; χ2 = 1.78, p =.18, respectively).
Studies 6a and 6b provide consistent results across very different study designs. In partial support of H4, the studies showed that "ugly" labeling was more effective than "imperfect" labeling in terms of hypothetical choice between unattractive and attractive produce (p =.058), and was also more effective at generating clicks with social media advertising in a field setting (p =.02). This is remarkable, given that the more than 50 grocery store managers that we interviewed overwhelmingly preferred "imperfect" labeling over "ugly" labeling.
The "ugly" label was directionally more effective than the "with personality" label, but the differences did not approach significance (all ps >.13), failing to support H4. In addition, "with personality" labeling (vs. control) significantly increased choice of unattractive produce, and, as for "ugly" labeling, this was mediated by tastiness expectations. In retrospect, this finding may not be inconsistent with our theorizing. The label "with personality" is a playful reference to language that suggests someone is not attractive; thus, the label may in fact point out the aesthetic flaw, albeit in a less explicit manner. To further examine this possibility, in Web Appendix W11 we report an additional study that compares the "ugly" label with yet other labels: "misshapen," "inferior," and "second-rate." We found that "ugly" was more effective than "inferior" and "second-rate," although "misshapen" was as effective as "ugly," and its effect on purchase likelihood was mediated by attribute expectations. Overall, this suggests that any label that explicitly ("ugly," "misshapen") or implicitly ("with personality") points out an aesthetic flaw may correct biased attribute expectations and increase purchase of unattractive produce.
Up to 30% of edible produce is discarded by farmers and retailers every year because of cosmetic imperfections, contributing to the environmental cost of food waste ([ 5]). Our work offers a simple marketing communications strategy that can be easily implemented to increase the appeal of unattractive produce. Specifically, across seven experiments we show that emphasizing the aesthetic flaw of unattractive produce via "ugly" labeling increases purchase, choice, and click-throughs.
Study 1 was conducted at a farmers' market and demonstrated that "ugly" labeling (vs. no specific label) increased purchase of unattractive, rather than attractive, produce. Study 2 used an incentive-compatible design and showed that "ugly" labeling significantly increased the likelihood that consumers use their lottery earnings to purchase a box of unattractive, rather than attractive, produce. Studies 3 and 4 showed through mediation and moderation that "ugly" labeling increases the choice of unattractive over attractive produce because it improves tastiness expectations and, to a smaller extent, healthiness expectations. Study 5 demonstrated that price discounts moderate the effectiveness of "ugly" labeling, and that "ugly" labeling associated with a mere 20% discount is as effective as a steep 60% discount. Studies 6a and 6b showed that "ugly" labeling is more effective than "imperfect" labeling at increasing the choice of unattractive produce and at increasing clicks on online ads. However, "ugly" labeling was not significantly more effective than "with personality" labeling (we return to this point under "Limitations").
We theorized that "ugly" labeling increases acceptance of unattractive produce because it corrects for consumers' biased, negative expectations about unattractive produce. We hypothesized that this should be the case for tastiness and healthiness expectations, but not for naturalness expectations. The results on tastiness supported our theorizing: without any specific label, unattractive produce suffered from negative tastiness expectations; "ugly" labeling systematically corrected for these negative expectations, which mediated the effect of "ugly" labeling on choice. The results on naturalness also supported our theorizing. Without any specific label, unattractive produce enjoyed positive naturalness expectations. As these positive expectations are in line with fact (the absence of pesticides, preservatives, or wax coatings necessarily yields cosmetic imperfections), they did not need to be corrected, and the mediations by naturalness were never significant. The results on healthiness were more muddled, but still consistent with our theorizing. Although healthiness expectations for unattractive produce in the absence of the "ugly" label were never significantly negative, we nonetheless found positive effects of "ugly" labeling and some mediating effects, although these effects were systematically weaker than for tastiness and not always significant (see Figure 4 for all mediation analyses; see Web Appendix W12 for all means and additional analyses).
Our research examines the effectiveness of "ugly" labeling, which was held constant in prior research examining how unattractive produce can negatively affect self-perceptions ([16]). In doing so, our research builds on this previous work by identifying another reason consumers reject unattractive produce: negative inferences about produce attributes. Our work also adds to research examining how food unattractiveness affects attribute expectations ([20]).
We also extend the literature on "awareness of influence" ([49]) to the domain of consumption. In line with this literature, we show that explicitly pointing out the source of biased attitudes—in this case, produce unattractiveness—motivates validity-driven corrections of attitudes.
Additionally, we contribute to research on persuasion. In the context that we study, simply adding one piece of negative information improves product evaluation. This contrasts with the literature on two-sided arguments ([42]) that has shown that weak negative information improves product evaluation, provided it is combined with positive information. However, the effects operate through different mechanisms. While two-sided arguments preempt counterarguments by explicitly addressing favorable and opposing views ([30]; [46]), "ugly" labeling draws consumers' attention to a nondiagnostic cue that was biasing their judgment.
While we have demonstrated the efficacy of "ugly" labeling, it is likely that any label pointing out the aesthetic flaw should increase purchase of unattractive produce. Studies 6a and 6b suggested that the "with personality" label, which hints at unattractiveness in a subtle way, was nearly as effective as the "ugly" label. Our study reported in Web Appendix W11 showed that the "misshapen" label, which clearly points out the aesthetic flaw, works as well as "ugly" to drive choice of unattractive produce, and both labels are driven by the same mechanism. Given our findings, it would be interesting to examine the extent to which other labels (e.g., "misfit," "pickuliar") are perceived as pointing to aesthetics as the source of imperfection, and whether they can also motivate purchase of unattractive produce.
Future research should also investigate heterogeneity in attractive–healthy associations and attractive–natural associations. While we found that people do not necessarily expect unattractive produce to be unhealthy, two studies found such associations ([20]; [47]). Looking at the stimuli used in these two studies, we suggest the possibility that when unattractiveness is operationalized with strong deformity or very unusual colors, it leads to unhealthiness inferences. Likewise, while we found that people expect unattractive produce to be more natural, research by [20] showed the opposite. This may be because Hagen's research focused on prepared and processed foods, for which cosmetic imperfections are unlikely to stem from nature. This discrepancy may also be related to measurement. Indeed, we measured naturalness with such items as "free of pesticides" and "free of preservatives," which may activate the knowledge that a more natural mode of production results in cosmetic imperfection, while Hagen measured naturalness with such items as "pure" and "unprocessed," which are more likely to activate notions of classic beauty.
Our work offers significant managerial contributions: it gives clear guidance to managers on whether and how to label unattractive produce, and which price discount will maximize sales. Specifically, we show that "ugly" labeling is more effective than "imperfect" labeling and works best with moderate price discounts. Importantly, these findings largely contrast with managers' beliefs. Indeed, several large brick-and-mortar and online retailers have relied on "imperfect" labeling (Web Appendix W1), and the more than 50 grocery store managers we spoke to largely preferred "imperfect" labeling, or no specific labeling, over "ugly" labeling.
"Ugly" labeling can also be a support for other better-world interventions, as shown by [16] in the case of a self-esteem boost intervention. Although this has not been tested, "ugly" labeling may also further increase the effectiveness of more labor-intensive and costly interventions that rely on educating consumers about the environmental consequences of food waste ([ 1]; [ 6]; [53]).
Online retailers who exclusively sell unattractive produce have been recently criticized for occasionally sourcing produce from industrial-scale producers, driving small-scale farmers out of business ([38]). While being cognizant of this issue, we believe that increasing consumers' interest in unattractive produce remains crucial: "ugly" labeling can be applied by smaller actors, particularly farmers, whose limited resources render them unable to meet the aesthetic demands and quotas required by retailers. "Ugly" labeling may also overcome retailers' reluctance to sell unattractive produce, whether it is because they fear a lack of consumer interest or they are concerned that steep price discounts would hurt their bottom line. Given retailers' participation in the U.S. Food Loss and Waste 2030 Champions initiative, with its objective of cutting food waste in half by 2030, our research helps reduce the uncertainty and reluctance regarding promotion of unattractive produce. In alignment with [39], which recently released a report focused on strategies to reduce food waste, our work shows how marketing can be used to shape a "better world" by providing a win-win solution to several stakeholders—from farmers and retailers to consumers and society at large.
Supplemental Material, sj-docx-1-jmx-10.1177_0022242920988656 - From Waste to Taste: How "Ugly" Labels Can Increase Purchase of Unattractive Produce
Supplemental Material, sj-docx-1-jmx-10.1177_0022242920988656 for From Waste to Taste: How "Ugly" Labels Can Increase Purchase of Unattractive Produce by Siddhanth (Sid) Mookerjee, Yann Cornil and JoAndrea Hoegg in Journal of Marketing
Footnotes 1 Cait Lamberton
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Financial support from the Social Sciences and Humanities Research Council of Canada is gratefully acknowledged.
4 Online supplement: https://doi.org/10.1177/0022242920988656
5 For all MTurk studies herein, we used the Cloud Research platform, which filters suspicious participants (e.g., bots) based on IP address and allows for the exclusion of people who participated in our other studies.
6 The results were similar when the item "sweet" was excluded from the scale.
7 Study 5 was the last MTurk study that we ran (May 2020). Because we noticed a sharp increase in low-quality responses and as indicated in the preregistration, in addition to the end-of-questionnaire attention check, we added another attention check at the beginning that automatically excluded participants who failed. We also increased the sample size in order to reach approximately 100 participants per condition, as indicated in the preregistration. As indicated earlier, we report in Web Appendix W7 all MTurk results on choice likelihood with and without data exclusion.
References Aschemann-Witzel Jessica, De Hooge Ilona, Amani Pegah, Bech-Larsen Tino, Oostindjer Marije. (2015), "Consumer-Related Food Waste: Causes and Potential for Action," Sustainability, 7 (6), 6457–77.
Aschemann-Witzel Jessica, Giménez Ana, Ares Gastón. (2018), "Consumer In-Store Choice of Suboptimal Food to Avoid Food Waste: The Role of Food Category, Communication and Perception of Quality Dimensions," Food Quality and Preference, 68, 29–39.
Auvray Malika, Spence Charles. (2008), "The Multisensory Perception of Flavor," Consciousness and Cognition, 17 (3), 1016–31.
Barański Marcin, Średnicka-Tober Dominika, Volakakis Nikolaos, Seal Chris, Sanderson Roy, Stewart Gavin B., et al. (2014), "Higher Antioxidant and Lower Cadmium Concentrations and Lower Incidence of Pesticide Residues in Organically Grown Crops: A Systematic Literature Review and Meta-Analyses," British Journal of Nutrition, 112 (5), 794–811.
Berkenkamp JoAnne, Meehan Michael. (2016), "Beyond Beauty: The Opportunities and Challenges of Cosmetically Imperfect Produce: Report No. 4: Lessons from Minnesota's Hunger Relief Community," Tomorrow's Table.
Bunn David, Feenstra Gail W., Lynch Lori, Sommer Robert. (1990), "Consumer Acceptance of Cosmetically Imperfect Produce," Journal of Consumer Affairs, 24 (2), 268–79.
Buzby Jean C., Farah-Wells Hodan, Hyman Jeffrey. (2014), "The Estimated Amount, Value, and Calories of Postharvest Food Losses at the Retail and Consumer Levels in the United States,"Economic Research Service Economic Information Bulletin No. 121, U.S. Department of Agriculture (February).
8 Castelo Noah, Bos Maarten W., Lehmann Donald R. (2019), "Task-Dependent Algorithm Aversion," Journal of Marketing Research, 56 (5), 809–25.
9 Choi Candice, McFetridge Scott. (2019), "Walmart and Whole Foods End 'Ugly Produce' Tests, Suggesting Trend May Have Limits," The Globe and Mail (February 21), https://www.theglobeandmail.com/life/article-us-grocers-end-ugly-produce-tests-suggesting-trend-may-have/.
Darley John M., Gross Paget H. (1983), "A Hypothesis-Confirming Bias in Labeling Effects," Journal of Personality and Social Psychology, 44 (1), 20.
De Hooge Ilona E., Oostindjer Marije, Aschemann-Witzel Jessica, Normann Anne, Loose Simone Mueller, Almli Valérie Lengard. (2017), "This Apple Is Too Ugly for Me! Consumer Preferences for Suboptimal Food Products in the Supermarket and at Home," Food Quality and Preference, 56, 80–92.
Ein-Gar Danit, Shiv Baba, Tormala Zakary L. (2011), "When Blemishing Leads to Blossoming: The Positive Effect of Negative Information," Journal of Consumer Research, 38 (5), 846–59.
Eisend Martin. (2009), "A Meta-Analysis of Humor in Advertising," Journal of the Academy of Marketing Science, 37 (2), 191–203.
Environmental Protection Agency (2017), "Greenhouse Gas Emissions" (accessed June 27, 2019), https://www.epa.gov/ghgemissions/overview-greenhouse-gases.
Govindasamy Ramu, Italia John, Liptak Clare. (1997), "Quality of Agricultural Produce: Consumer Preferences and Perceptions," New Jersey Agricultural Experiment Station P-02137-1-97, Rutgers University.
Grewal Lauren, Hmurovic Jillian, Lamberton Cait, Reczek Rebecca Walker. (2019), "The Self-Perception Connection: Why Consumers Devalue Unattractive Produce," Journal of Marketing, 83 (1), 89–107.
Griffin Angela M., Langlois Judith H. (2006), "Stereotype Directionality and Attractiveness Stereotyping: Is Beauty Good or Is Ugly Bad?" Social Cognition, 24 (2), 187–206.
Grunert Klaus G. (2007), "How Consumers Perceive Food Quality," in Understanding Consumers of Food Products, Frewer Lynn J., Trijp Hans Van, eds. Cambridge, UK: British Welding Research Association, 181–99.
Haasova Simona, Florack Arnd. (2019), "Practicing the (Un)Healthy = Tasty Intuition: Toward an Ecological View of the Relationship Between Health and Taste in Consumer Judgments," Food Quality and Preference, 75, 39–53.
Hagen Linda. (2021), "Pretty Healthy Food: How and When Aesthetics Enhance Perceived Healthiness," Journal of Marketing, 85 (2), 129–45.
Hahn Adam, Gawronski Bertram. (2019), "Facing One's Implicit Biases: From Awareness to Acknowledgment," Journal of Personality and Social Psychology, 116 (5), 769.
Hall Kevin D., Guo Juen, Dore Michael, Chow Carson C. (2009), "The Progressive Increase of Food Waste in America and Its Environmental Impact," PLOS ONE, 4 (11), e7940.
Hamermesh Daniel S., Biddle Jeff E. (1994), "Beauty and the Labor Market," The American Economic Review, 84 (5), 1174–94.
Hardisty David J., Weber Elke U. (2020), "Impatience and Savoring vs. Dread: Asymmetries in Anticipation Explain Consumer Time Preferences for Positive vs. Negative Events," Journal of Consumer Psychology, 30 (4), 598–613.
Hayes Andrew F. (2012), "My Macros and Code for SPSS and SAS" (accessed January 26, 2017), http://afhayes.com/spss-sas-andmplus-macros-and-code.html.
Hoegg JoAndrea, Alba Joseph W., Dahl Darren W. (2010), "The Good, the Bad, and the Ugly: Influence of Aesthetics on Product Feature Judgments," Journal of Consumer Psychology, 20 (4), 419–30.
Hornick Sharon B. (1992), "Factors Affecting the Nutritional Quality of Crops," American Journal of Alternative Agriculture, 7 (1/2), 63–68.
Hussin Siti Rahayu, Yee Wong Foong, Bojei Jamil. (2010), "Essential Quality Attributes in Fresh Produce Purchase by Malaysian Consumers," Journal of Agribusiness Marketing, 3 (December), 1–19.
Hutchings John B. (1994), "Sensory Assessment of Appearance—Methodology," in Food Colour and Appearance, Hutchings John B., ed. New York: Springer, 104–41.
Kamins Michael A., Marks Lawrence J. (1987), "Advertising Puffery: The Impact of Using Two-Sided Claims on Product Attitude and Purchase Intention," Journal of Advertising, 16 (4), 6–15.
Kirmani Amna. (1997), "Advertising Repetition as a Signal of Quality: If It's Advertised So Much, Something Must Be Wrong," Journal of Advertising, 26 (3), 77–86.
Koo Minkyung, Oh Hyewon, Patrick Vanessa M. (2019), "From Oldie to Goldie: Humanizing Old Produce Enhances Its Appeal," Journal of the Association for Consumer Research, 4 (4), 337–51.
Kupor Daniella, Laurin Kristin. (2020), "Probable Cause: The Influence of Prior Probabilities on Forecasts and Perceptions of Magnitude," Journal of Consumer Research, 46 (5), 833–52.
Leksrisompong Pattarin, Whitson Magan, Truong Van Den, Drake Maryanne. (2012), "Sensory Attributes and Consumer Acceptance of Sweet Potato Cultivars with Varying Flesh Colors," Journal of Sensory Studies, 27 (1), 59–69.
Loebnitz Natascha, Schuitema Geertje, Grunert Klaus G. (2015), "Who Buys Oddly Shaped Food and Why? Impacts of Food Shape Abnormality and Organic Labeling on Purchase Intentions," Psychology & Marketing, 32 (4), 408–21.
Makhal Annesha, Robertson Kirsten, Thyne Maree, Mirosa Miranda. (2020), "Normalising the 'Ugly' to Reduce Food Waste: Exploring the Socialisations That Form Appearance Preferences for Fresh Fruits and Vegetables," Journal of Consumer Behaviour(published online November 16), https://doi.org/10.1002/cb.1908.
Meyvis Tom, Van Osselaer Stijn M.J. (2017), "Increasing the Power of Your Study by Increasing the Effect Size," Journal of Cosumer Research, 44 (5), 1157–73.
Mull Amanda. (2019), "The Murky Ethics of the Ugly-Produce Business," The Atlantic(January 25), https://www.theatlantic.com/health/archive/2019/01/ugly-produce-startups-food-waste/581182/.
National Academies of Sciences, Engineering and Medicine (2020), A National Strategy to Reduce Food Waste at the Consumer Level, Oria Maria, Schneeman Barbara O., eds. Washington, DC: The National Academies Press.
Owen James. (2005), "Farming Claims Almost Half Earth's Land, New Maps Show," National Geographic (December 9), https://news.nationalgeographic.com/news/2005/12/agriculture-food-crops-land/.
Palmer Stephen E., Schloss Karen B., Sammartino Jonathan. (2013), "Visual Aesthetics and Human Preference," Annual Review of Psychology, 64, 77–107.
Pechmann Cornelia. (1992), "Predicting When Two-Sided Ads Will Be More Effective Than One-Sided Ads: The Role of Correlational and Correspondent Inferences," Journal of Marketing Research, 29 (4), 441–53.
Péneau Sandrine, Hoehn Ernst, Roth Hans Rudolf, Escher Felix E., Nuessli Jeannette. (2006), "Importance and Consumer Perception of Freshness of Apples," Food Quality and Preference, 17 (1/2), 9–19.
Pirlott Angela G., MacKinnon David P. (2016), "Design Approaches to Experimental Mediation," Journal of Experimental Social Psychology, 66, 29–38.
Raghubir Priya, Greenleaf Eric A. (2006), "Ratios in Proportion: What Should the Shape of the Package Be?" Journal of Marketing, 70 (2), 95–107.
Rucker Derek D., Petty Richard E., Briñol Pablo. (2008), "What's in a Frame Anyway? A Meta-Cognitive Analysis of the Impact of One Versus Two Sided Message Framing on Attitude Certainty," Journal of Consumer Psychology, 18 (2), 137–49.
Schifferstein Hendrik N.J., Wehrle Theresa, Carbon Claus-Christian. (2019), "Consumer Expectations for Vegetables with Typical and Atypical Colors: The Case of Carrots," Food Quality and Preference, 72, 98–108.
Shao Xiaolong, Jeong EunHa, Jang SooCheong Shawn, Xu Yang. (2020), "Mr. Potato Head Fights Food Waste: The Effect of Anthropomorphism in Promoting Ugly Food," International Journal of Hospitality Management, 89, 102521.
Strack Fritz, Hannover Bettina. (1996), "Awareness of Influence as a Precondition for Implementing Correctional Goals," in The Psychology of Action: Linking Cognition and Motivation to Behavior, Gollwitzer P.M., Bargh J.A., eds. New York: Guilford Press, 579–96.
Symmank Claudia, Zahn Susann, Rohm Harald. (2018), "Visually Suboptimal Bananas: How Ripeness Affects Consumer Expectation and Perception," Appetite, 120, 472–81.
Townsend Claudia, Shu Suzanne B. (2010), "When and How Aesthetics Influences Financial Decisions," Journal of Consumer Psychology, 20 (4), 452–58.
Tsakiridou Efthimia, Boutsouki Christina, Zotos Yorgos, Mattas Kostantinos. (2008), "Attitudes and Behaviour Towards Organic Products: An Exploratory Study," International Journal of Retail & Distribution Management, 36 (2), 158–75.
Van Giesen Roxanne I., Hooge Ilona E. De. (2019), "Too Ugly, But I Love Its Shape: Reducing Food Waste of Suboptimal Products with Authenticity (and Sustainability) Positioning," Food Quality and Preference, 75, 249–59.
Verhoog Henk, Bueren Edith Lammerts Van, Matze Mirjam, Baars Ton. (2007), "The Value of 'Naturalness' in Organic Agriculture," NJAS–Wageningen Journal of Life Sciences, 54 (4), 333–45.
Villegas Beatriz, Carbonell Inmaculada, Costell Elvira. (2008), "Effects of Product Information and Consumer Attitudes on Responses to Milk and Soybean Vanilla Beverages," Journal of the Science of Food and Agriculture, 88 (14), 2426–34.
Walmsley Ashley. (2017), "Produce Consumers Demand Perfection: US Rabobank Expert," North Queensland Register (August 3), https://www.northqueenslandregister.com.au/story/4831949/produce-consumers-demand-perfection/.
Xu Yang, Jeong EunHa, Jang SooCheong Shawn, Shao Xiaolong. (2021), "Would You Bring Home Ugly Produce? Motivators and Demotivators for Ugly Food Consumption," Journal of Retailing and Consumer Services, 59, 102376.
Yuan Jingxue Jessica, Yi Sungpo, Williams Helena A., Park Oak-Hee. (2019), "US Consumers' Perceptions of Imperfect 'Ugly' Produce," British Food Journal, 121 (11), 2666–82.
~~~~~~~~
By Siddhanth (Sid) Mookerjee; Yann Cornil and JoAndrea Hoegg
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 56- Generating Content Increases Enjoyment by Immersing Consumers and Accelerating Perceived Time. By: Tonietto, Gabriela N.; Barasch, Alixandra. Journal of Marketing. Nov2021, Vol. 85 Issue 6, p83-100. 18p. 1 Chart. DOI: 10.1177/0022242920944388.
- Database:
- Business Source Complete
Record: 57- Genetic Data: Potential Uses and Misuses in Marketing. By: Daviet, Remi; Nave, Gideon; Wind, Jerry. Journal of Marketing. Jan2022, Vol. 86 Issue 1, p7-26. 20p. 2 Diagrams, 2 Charts. DOI: 10.1177/0022242920980767.
- Database:
- Business Source Complete
Genetic Data: Potential Uses and Misuses in Marketing
Advances in molecular genetics have led to the exponential growth of the direct-to-consumer genetic testing industry, resulting in the assembly of massive privately owned genetic databases. This article explores the potential impact of this new data type on the field of marketing. Drawing on findings from behavioral genetic research, the authors propose a framework that incorporates genetic influences into existing consumer behavior theory and use it to survey potential marketing uses of genetic data. Applications include business strategies that rely on genetic variants as bases for segmentation and targeting, creative uses that develop consumers' sense of community and personalization, use of genetically informed study designs to test causal relations, and refinement of consumer theory by uncovering biological mechanisms underlying behavior. The authors further evaluate ethical challenges related to autonomy, privacy, misinformation, and discrimination that are unique to the use of genetic data and are not sufficiently addressed by current regulations. They conclude by proposing an agenda for future research.
Keywords: behavioral genetics; big data; consumer genomics; DNA data; ethics; segmentation; technology
In September 2018, the music streaming service Spotify announced that it would allow its 217 million users to upload their genetic data and create playlists that "match their genetic ancestry" ([61]). A few months later, Mexico's national air carrier, Aeroméxico, launched a "DNA Discounts" campaign, offering to some customers discounted flights to Mexico, with discount rates that matched the traveler's "Mexican DNA" percentage, determined by a genetic test ([144]).[ 5] These actions mark the dawn of a new age, when consumers and firms alike may access information that until recently was rarely accessible: individual-level measures of the human genome.
Such data are now available through the direct-to-consumer genetic testing (DTC-GT) market, whose total sales in 2019 exceeded all previous years combined. Most sales come from personalized DNA testing kits—plastic tubes that consumers spit into and then ship off for genomic analysis. The motives for taking a DNA test vary, ranging from the desire to uncover forgotten family histories to assessing genetic predispositions for diseases. As of 2020, more than 30 million people have already taken such personalized DNA tests ([120]). A by-product of the growing DTC-GT market is the accumulation of massive genetic data sets. Industry leaders, such as AncestryDNA and 23andMe, encourage consumers to participate in research by answering surveys about anything from dietary habits to personality, generating enormous data sets for investigating genetic associations to numerous outcomes. Because the sales growth of DTC-GT kits might be slowing down ([45]), DTC-GT companies are aiming to monetize their data to maintain growth. For example, Patrick Chung, a 23andMe board member, noted in an interview that "the long game here is not to make money selling kits, although the kits are essential to get the base level data" ([101]). In line with this notion, 23andMe has already accredited access to its data to the pharmaceutical company GlaxoSmithKline in a $300 million deal ([16]).
The abundance of privately owned DNA data is concurrent with large-scale data collection efforts of public endeavors such as the UK BioBank, which genotyped nearly half a million U.K. citizens ([20]). National genome projects have also taken off in other countries, including Sweden and Singapore ([134]). The accumulation of genetic data has already fueled the discovery of associations between genes and individual differences in many traits ([91]; [100]; [143]), from dietary habits such as coffee and tea intake ([135]), to psychological traits such as adventurousness ([68]).
The current research explores the potential impacts of the DNA revolution on the field of marketing and discusses possible uses and abuses of genetic data by marketers. It is organized as follows. First, we introduce key terms and review recent advances in the fields of behavioral genetics and genealogy. Drawing on these findings, we introduce a theoretical framework that incorporates genetic variables into existing consumer behavior theory. We rely on this framework to conceptually explore applications of genetic data for marketing strategy and research and evaluate under what circumstances genetic tools may be of value to marketers. We then raise ethical challenges that are unique to the use of genetic data in marketing, survey how current regulations address them (or not), and suggest potential solutions. Subsequently, we identify gaps in the current state of knowledge that must be filled to further advance the field and draw a research agenda to address them.
This section introduces basic concepts in human genetics and reviews related research that is relevant for the field of marketing (see Table 1). Our review is intended for readers who are not acquainted with the topic, and it focuses on research using DNA measures (the only type of genetic data currently available at scale). We admittedly abstract away from many subtleties and refer interested readers to other publications for more comprehensive reviews ([21]; [83]) and surveys of research using other genetic data modalities (for epigenetics, see [82]; for RNA sequencing, see [132]; for gene therapy, see [149]).
Graph
Table 1. Illustrative Genetics Literature of Marketing-Relevant Outcomes.
| Reference | Outcomes |
|---|
| Twin Studies of Behavioral Tendencies |
| Cesarini et al. (2009) | Altruism, risk-taking |
| Cesarini et al. (2010) | Investment portfolio risk |
| Simonson and Sela (2011) | Default bias, tendency to choose a compromise option, preference for utilitarian goods |
| Cesarini et al. (2012) | Procrastination, default bias, conjunction fallacy, cognitive reflection |
| Loewen and Dawes (2012) | Voting turnout |
| Miller et al. (2012) | Mobile phone usage (call and text frequency) |
| Cronqvist and Siegel (2014) | Stock market participation, asset allocation |
| Pedersen et al. (2015) | Blood donation |
| Hall, Loscalzo, and Kaptchuk (2015) | Susceptibility to placebo effects |
| Genetic Variants with Direct Influence on Target Outcomes |
| Health Care | |
| Giri, Zhang, and Lü (2016) | Early onset Alzheimer's disease |
| Kuchenbaecker et al. (2017) | Breast and ovarian cancer |
| Redondo, Steck, and Pugliese (2018) | Type I diabetes |
| Al-Khelaifi et al. (2019) | Muscle composition |
| Blauwendraat, Nalls, and Singleton (2020) | Parkinson's disease |
| Nutrition | |
| Eng, Luczak, and Wall (2007) | Alcohol intolerance |
| Mattar et al. (2012) | Lactose intolerance |
| Nehlig (2018) | Caffeine metabolism |
| Beauty | |
| Shimomura et al. (2008) | Wooly hair |
| Bataille et al. (2000) | Freckles |
| Pirastu et al. (2017) | Hair loss (alopecia) |
| Genome-Wide Association Studies |
| Hu et al. (2016) | Propensity to be a morning person |
| Taylor, Smith, and Munafò (2018) | Nonalcoholic beverage consumption (including coffee and tea) |
| Lee et al. (2018) | Educational attainment |
| Day, Ong, and Perry (2018) | Loneliness; attendance of gyms, clubs, or pubs; participation in religious groups |
| Sanchez-Roige et al. (2018) | Personality traits |
| Yengo et al. (2018) | Height, body mass index |
| Al-Khelaifi et al. (2019) | Athletic performance |
| Baselmans et al. (2019) | Life satisfaction, positive affect, neuroticism, depressive symptoms |
| Erzurumluoglu et al. (2020) | Smoking initiation, cigarettes/day, smoking cessation |
| Karlsson Linnér et al. (2019) | Risk tolerance, adventurousness, speeding, alcohol consumption, smoking, number lifetime sex partners |
The human genome is a sequence of about 3 billion base pairs. There are four types of bases: adenine (A), thymine (T), guanine (G), and cytosine (C). The base pairs are packaged into structures called chromosomes and are indexed based on their location on the sequence. Every human has two copies of each chromosome, one inherited from each parent. The base pairs in most genome locations are identical across all humans and are thus not informative about interindividual variability. However, there is a small number of locations (<2%) called polymorphisms where individuals commonly differ. The most common type of polymorphism is the single-nucleotide polymorphism (SNP), which denotes locations where a single base pair differs across individuals.[ 6] For most SNPs, only two possible base pair types are observed in a given species. The more frequent base pair is called the major allele, and the other is called the minor allele. As all humans inherit one chromosome from each parent, they also inherit two copies of each SNP, and thus have either zero, one, or two minor alleles in every SNP location. This property allows for the storage of an individual's genetic data in terms of numbers of minor alleles at each SNP location (0, 1, or 2). Certain SNPs are located in subsequences of base pairs called genes. Genes shape the structure and function of every cell in the human body and are involved in many biological processes, most notably the construction of proteins ([44]). The human genome includes 20,000 to 30,000 genes.
Until recently, it was extraordinarily time consuming and expensive to measure genetic variation of individuals. However, technological advances following the sequencing of the human genome by the Human Genome Project ([29]) have enabled cost-effective measurements of the genome across individuals. Common measurement techniques quantify variations in selected genome locations (typically under 1 million SNPs) where humans commonly differ. From there, around 20 million other SNPs are imputed.
Behavioral genetics is a discipline dedicated to studying the relationship between genetic code and behavioral traits (also called phenotypes). Early research in the field mainly consisted of twin studies—which rely on the fact that identical (monozygotic) twins are on average twice more genetically similar to each other than fraternal (heterozygotic) twins. Under some strong assumptions ([43]), twin studies enable us to estimate a trait's heritability—the part of its interindividual variance that can be attributed to genetics. Surprisingly, twin studies have shown that most human behavioral traits are, to some degree, heritable. This finding is commonly known as "The First Law of Behavioral Genetics" ([138]) and was illustrated for manifold phenotypes, from psychological traits such as personality to real-life outcomes such as marital status (see Table 1). Two other empirical regularities characterize findings from behavioral twin studies. The Second Law of Behavioral Genetics states that the effect of being raised in the same family is typically smaller than that of genetics. The Third Law of Behavioral Genetics denotes that substantial behavioral variations are not accounted for by either genetics or family environment. Nonetheless, the Three Laws are not without exceptions. On the one hand, many biological phenotypes that are highly relevant for marketing of health care, nutrition, and beauty products, such as lactose intolerance (for additional examples, see Table 1) are highly heritable. The downstream behavioral consequences of these traits (e.g., the tendency to buy dairy alternative products)[ 7] are expected to be more heritable. On the other hand, various culture-related characteristics, such as one's native language or nationality, are entirely driven by the environment yet can be predicted from genetic ancestry (see the "Genetic Ancestry" subsection). The Three Laws demonstrate the promises and drawbacks of using measurements of the genome in marketing. Although genomes are informative about a wide range of relevant outcomes, genetic information is usually not informative for making individual-level predictions of most behavioral traits without additional variables ([60]). A unique feature of DNA data is that they are currently immutable across one's lifespan. Thus, such measures may be informative of one's future behavior long before any other variables become informative.
Although twin studies produce heritability estimates, they remain silent about contributions of specific genetic variants to a trait's variability. The first wave of research addressing this gap consisted of candidate gene studies—theoretically motivated examinations of associations between phenotypes and SNPs located in specific genes that were a priori hypothesized to be related to them ([77]). For example, the known role of serotonin in depression motivated studies investigating the association between depression and SNPs located on serotonergic genes ([109]). Although candidate-gene studies have yielded eye-catching findings for some applications, most in the behavioral domain have failed to replicate in subsequent studies. This failure is attributed to low statistical power, a lack of appropriate correction for multiple hypotheses testing, and a lack of control for confounding factors ([25]). Development of genotyping techniques, together with massive data collection efforts, has led to a paradigm shift from candidate-gene studies to genome-wide association studies (GWAS; [143])—data-driven investigations of the relationships between phenotypes and SNPs across the entire genome. Due to the large number of associations studied, GWAS methodology emphasizes stringent correction for multiple testing, preregistration, and replication in independent samples. Over the past decade, GWAS samples have grown from thousands to millions of participants, and the increase in statistical power has allowed researchers to identify numerous replicable associations between SNPs and behavioral phenotypes (see Table 1). However, a typical behavioral trait is associated with numerous variants, each of them accounting for a very small part (R2 <.01%) of its variance, an observation known as the "Fourth Law of Behavioral Genetics" ([25]).
While the contribution of individual SNPs to the variability of most human behavioral traits is minute, one can obtain greater explanatory power by aggregating their effects to a polygenic risk score (PRS). The PRS is a linear combination of the most significant SNPs identified in its GWAS, and it becomes increasingly accurate as sample sizes increase. For example, a PRS constructed from a recent GWAS in 1 million people was able to predict 13% of the variance in the educational attainment in an independent sample ([80]). Polygenic risk scores are similarly informative on many behavioral traits, and firms that possess genetic data can construct them using GWAS summary statistics that either are publicly available ([91]) or can be obtained from other organizations. Yet a significant share of current publicly available PRSs of behavioral phenotypes are not accurate enough for making individual-level predictions. Furthermore, the predictive accuracy of PRSs typically decreases when applied to populations different from those used to estimate them (e.g., in ethnicity and socioeconomic status; [37]; [94]). Nonetheless, publicly available PRSs are typically computed from samples of only up to a million individuals, whereas DTC-GC companies have access to samples that are an order of magnitude larger. Moreover, advanced statistical techniques show a high potential for obtaining more accurate genome-based predictions (see the "Advanced Targeting and Prediction" subsection).
The possibility to quantify genetic variations in individuals has also opened up the path for studying genetic variation between populations. A common approach is to perform principal component analysis on the genetic data of a population sample in search of high-order factors that capture its variability ([ 2]). These principal components (PCs) are highly informative about one's genetic ancestry and location.[ 8] For example, a study of individuals from 51 populations worldwide found that the first PC distinguished sub-Saharan Africans from non-Africans and the second PC differentiated populations from Eastern and Western Eurasia ([84]). These findings were echoed by studies of less diverse samples that used the same method for high-resolution ancestry mapping (e.g., [106]). The PCs are also commonly used as control variables in GWAS and other population studies to account for environmental factors that vary across ethnic groups ([116]; [102]). The relevance of genetic ancestry for marketing stems from its noncausal correlation with environmental factors such as language and culture. For example, individuals of Irish ancestry are more likely to be interested in a cultural heritage trip to Ireland or celebrate Saint Patrick's Day with a pint of Guinness, and their interests may stem from cultural influences related to their ancestry. Marketers with access to genetic data may be able to infer such behavioral tendencies and the motivations underlying them and use such insight for targeting and positioning.
The Four Laws of Behavioral Genetics provide solid empirical grounds from which explorations of genetic effects on consumer behavior can embark. Translating these fundamental insights into applications, however, depends on incorporating them into consumer behavior theory and models. This section proposes such a framework, illustrated in Figure 1. Our theory extends the well-known stimulus-response model, which describes behavior as arising from the interaction between the organism (consumer) and stimulus ([10]). The stimulus is described via object variables, such as the products, prices, and brands offered, and situational variables, such as location, time of day, and context. The organism has traditionally been marked by personal variables denoting characteristics that are "stable over times and places of observation and may therefore be attributed consistently to the individual" ([10], p. 36). Typical personal variables include demographics, psychographics, and behavioral dispositions.
Graph: Figure 1. A model of genetic effects on consumer behavior.Notes: Arrows represent causal relations and interactions. Dashed lines denote important noncausal correlations.
Our framework extends the stimulus-response model by incorporating the elementary factors described in the previous section. We group these factors into three categories: ( 1) environment, which includes stable cultural, social, and geographical factors, as well as the flow of time influencing development and aging; ( 2) family factors, such as parenting style; and ( 3) the individual's genome, which depends on familial background, except for cases of adoption and recomposed families. Our framework considers familial and environmental factors as external to the organism, where the genome is within the organism and constitutes the most stable type of personal variables: it is fixed at conception and remains mostly stable throughout the lifespan. Our framework also extends the description of the organism by incorporating stable biological traits such as physiology (e.g., height), anatomy (e.g., brain structure), and typical brain function (e.g., connectivity between brain areas at rest). These biological traits are more directly influenced by genetics and typically mediate the influence of genetics on nonbiological personal traits. When such mediation occurs, the mediating biological trait is commonly referred to as an endophenotype.
As indicated by the Three Laws of Behavioral Genetics, the environment plays a major role in the development of most personal traits. Nonetheless, genetic influences affect many outcomes of interest, starting from prenatal development and early-life stages. These effects occur via interactions with familial and environmental factors and are mediated via endophenotypes that are more directly susceptible to genetic influences, such as brain anatomy ([137]). The relative impact of genetics varies by trait. In some cases, few genetic variants have strong direct effects on a biological endophenotype (e.g., lactose tolerance; for other examples, see Table 1), and genetic data will be highly informative of their downstream consequences (e.g., interest in dairy alternatives). Most personal traits, however, are only moderately heritable and are influenced by interactions between numerous genetic and environmental factors. Importantly, the genome is also informative about characteristics that are not influenced by genetics at all, due to the noncausal correlations between genetics and environmental or familial factors (dashed lines in Figure 1). If genetic data are available, such links allow for the inference of consumer characteristics such as cultural heritage and language.
The impact of genetics continues through the lifespan via two main channels. First, genetics affects later-life outcomes through its prior influence on traits that had developed earlier. For example, variants that contribute to early-life intellectual development continue to affect one's educational attainment and career in adulthood. Second, genetics continues to interact with environmental factors (e.g., time, nutrition) to influence later-life development of personal traits through biological endophenotypes such as brain anatomy and function ([131]). Although the heritability of later-life traits is typically moderate, characteristics that have a strong biological basis are well-explained by interactions between genetics and time. For example, a few SNPs explain 38% of the variance in hair loss (alopecia) in men ([111]), whose associated market size is expected to reach $3.9 billion by 2026 ([54]). These SNPs likely capture behavioral variance in this trait's downstream consequences.
The final influence of genetics on consumer behaviors, such as information search, purchase decisions, satisfaction, and word-of-mouth activity, occurs through interactions with situation and object. These effects are mediated via biological processes (e.g., changes in neural activity and hormonal levels) that regulate the consumer's emotional and physiological state, as well as cognitive processes such as attention, valuation, and memory ([112]). For instance, genetics affects one's tendency to be an early riser or a night owl ([63]), and this disposition affects arousal via interaction with the time of day (situational variable) to influence behavior. Likewise, situational stressors interact with genetics to generate a person-specific stress response, regulated by activation of the hypothalamic–pituitary–adrenal axis ([46]). This response, in turn, influences decision making (e.g., [92], [93]). Genetics also interacts with object variables, as products and marketing messages may affect genetically regulated attention, reward, and valuation processes. For example, the presence of a desirable food item (e.g., in a supermarket tasting counter) elicits an appetitive (or Pavlovian) response that may increase its subjective valuation ([19]). Animal studies suggest that individual differences in this tendency, which is biologically implemented by the dopaminergic system, is partly accounted by genetic variation ([48]). Additional interactions between genetics and object occur via indirect genetic influences on heritable traits such as personality ([96]) and behavioral dispositions such as the tendency to choose the default or compromise option in a choice set ([24]; [128]). Genetic data may allow for approximating these tendencies without having to rely on large-scale customer surveys.
Building on the framework introduced in the previous section, the following two sections discuss how the availability of genetic data may advance marketing practice and research (see Figure 2 and Table 2). We highlight that some of these applications, especially when employed by private entities in a for-profit setting, raise legal and ethical challenges (discussed in a subsequent section). It remains to be seen whether their potential benefits outweigh these concerns.
Graph: Figure 2. Genetic applications for marketing strategy.
Graph
Table 2. Using Genetics to Advance Marketing Research.
| Application | Approach | Related Literature |
|---|
| Estimating causal relations | Using genetic measures as instrumental variables via MR and other PRS-based analytical techniques | Smith and Ebrahim (2004); Zhu et al. (2018); DiPrete, Burik, and Koellinger (2018); O'Connor and Price (2018); Davies et al. (2019); Koellinger and De Vlaming (2019) |
| Accounting for otherwise unobserved heterogeneity | Including genetic PCs and PRS as control variables in observational studies and randomized-controlled trials | Price et al. (2006); Nave et al. (2018); Buniello et al. (2019); Harden and Koellinger (2020); Benjamin et al. (2012) |
| Studying person–subject and person–situation interactions | Approximating unobservable traits using their PRS and exploring their interactions with other observed variables (e.g., treatment effects) | Buniello et al. (2019); |
| Studying relationships between traits | Quantifying the variance that traits share due to genetic causes via their genetic correlations (rg), which can be estimated between any two traits for which a GWAS has ever been conducted | Lynch and Walsh (1998); Bulik-Sullivan et al. (2015) |
| Identifying biological mechanisms | Once genetic variants are linked to a behavioral trait, they can be tied to neurobiological and physiological systems via bioinformatic tools | Finucane et al. (2015); Dass et al. (2019); Aydogan et al. (2021) |
When genetic variations correspond with consumer needs, firms may rely on genetic data to divide the market into distinct, stable, and identifiable subsets to be reached with unique marketing mixes ([49]). In some cases, genetic variants are indeed directly associated with consumer needs via known mechanisms. A firm or institution could thus rely on genetic data to identify segments that benefit from its products and services. Prior research has uncovered mechanisms that link genetic variants to phenotypes that closely map onto consumer needs in various domains (see Table 1). Most current knowledge concerns outcomes related to health care, nutrition, and beauty, with applications such as promoting screening or prevention products to individuals who are at increased risk of developing pathologies such as cancer, diabetes, or Alzheimer's disease. Indeed, leading DTC-GT companies already provide information on such risks to their consumers and aspire to use their data to become the "Google of personalized health care" ([101]). As genetic databases grow in size, research for nonmedically relevant causal effects is expected to increase and yield new discoveries that are relevant for marketing strategy across domains. For example, a brand manager of a product for preventing men's hair loss could rely on a specific genetic variation linked to male pattern baldness ([111]) to identify segments that are genetically disposed to alopecia. The brand manager may even be able to identify future customers long before they show any behavioral indication that they may need the product (e.g., via web searches) and increase their awareness of the brand (e.g., by advertising to males in their late 20s who are genetically disposed to baldness in their mid-30s).
As Figure 1 illustrates, genetic variation correlates with almost every personal characteristic. As a result, genetic data can be used for reaching market segments when nongenetic managerially relevant variables cannot be easily observed at scale. In contrast to the direct use of specific genes as segmentation bases, most SNP associations to behavioral traits occur outside of genes, and marketers can leverage their cumulative information to infer other (nongenetic) segmentation bases. Once genetic data are available, a firm can construct for every individual in a target population PRSs that are predictive (to some degree) of every trait for which a GWAS has ever been performed ([18]). Similarly, a firm can rely on previous findings of genealogical research for calculating individual-level ancestry estimates to infer various culturally distinct motivations, interests, and behaviors. The usefulness of genomes as proxies for other segmentation variables crucially depends on how predictive they are of the target trait relative to other measures. Although genetic data might not be the most predictive of a target trait, it may be more convenient than other data sources such as surveys, which might be costly and subject to low response rates. Furthermore, adding genes to predictive models that use other variable types may improve their predictive accuracy at the individual level (see the "Are Genes More Predictive Than Other Measures?" subsection).
Marketers often aim to predict the probability of a single behavior (purchase, click on an ad, etc.), without necessarily understanding the underlying mechanism. As such, even a simple PRS constitutes a straightforward tool for targeting. Firms that obtain genetic data, but not samples that are large enough to estimate the coefficients used to construct a PRS, could potentially recover them from the public domain ([18]) and other organizations. More advanced statistical learning methods ([86]), including deep learning algorithms ([157]; [41]), have been adapted to genetic data to generate more accurate predictions. Furthermore, when genetic predictive estimates are available, they can be used in conjunction with other variables for early identification of consumers with high lifetime value. For instance, a coffeehouse chain may want to target consumers with a high genetic potential to enjoy espresso before they show any prior espresso-purchasing patterns in their behavioral data. While counterintuitive, such an approach would potentially allow for reaching consumers who have not yet developed an espresso consumption habit and thus are not "locked in" to a particular brand. This is in contrast to targeting based on more traditional variables (e.g., behavioral measures) that are likely to become predictive only after the person has already tried and developed the habit of consuming a competitor's brand.
A different approach using genetic data for behavioral prediction is to consider that genomes are representative of family relations and, as such, can be used to compute a comprehensive map of relatedness between individuals. Such a map can then be used for targeting in a similar manner to social network graphs ([142]; [148]). Another possibility is to compute genetic relatedness (or inversely, genetic distance) between individuals ([117]), either for the whole genome or chromosome-wise, and leverage this metric for behavioral prediction. For instance, a company could target people who are within a small genetic distance from existing clusters of loyal customers. Methods such as collaborative filtering, nearest neighbors, or more advanced machine learning algorithms could be applied to implement such strategies ([87]). Notably, geneticists are already using similar techniques that do not depend on identifying links between specific genes and a phenotype to estimate the variance in a trait that can be explained by SNP-derived genetic distance ([151]). For example, such methods have shown that 51% of the general population variance in fluid intelligence could be explained by genetic distance, quantified from SNPs, using a sample of a few thousand people ([33]).
Finally, DNA has a unique status as a "cultural icon" ([105]), which opens the door for creative uses, including new product development and repositioning of existing products and brands. Genetic data provide a new means of "knowing thyself," connecting to previously unknown genetic relatives, and building bridges between people and their ancient family histories ([139]). Leading DT-GTC companies have created several new products and positioning strategies that translate their customers' fascination with DNA into applications that promote their sense of community and personalization. Notable examples are the aforementioned partnership between Spotify and AncestryDNA and the collaboration between Airbnb and 23andMe, which developed a service that helps travelers organize cultural experiences tailored to their ancestry. Ancestry-based positioning strategies of products and services in other domains, including entertainment (e.g., period dramas such as Downton Abbey and Braveheart), food (e.g., traditional cookbooks) and tourism (e.g., museums, heritage sites) could similarly benefit from such partnerships and creative uses. Similar strategies could employ PRS or single genetic variants. For example, most elite power athletes have a specific variant of the ACTN3 gene that encodes a protein expressed in muscle fibers ([ 3]; [79]). A sporting brand may be able to develop positioning strategies that generate a sense of community among amateur athletes who carry this variant and promote their sense of identification with brand ambassadors who also carry it.
Genetic data can refine and substantiate existing theories of consumer behavior by illuminating the nature of relationships between traits and revealing the biological mechanisms underlying individual differences in behavior. Some of these applications are similar in nature to uses of genetics in other fields of the social sciences ([11]; [60]), where others are unique to marketing research.
In many domains of consumer research, it is not feasible to study causal relations between variables experimentally. For example, experimentally studying the causal relationship between one's consumption habits and long-term happiness ([52]; [124]) would require randomly assigning individuals into groups that differ in their consumption habits or in their well-being. Such assignment could be extremely difficult for some variables and even unethical (e.g., if a group is required to worsen dietary habits, creating a threat to their health). Furthermore, studying such causal relations using observational data is also not straightforward. First, many personal and environmental factors (e.g., socioeconomic status, personality) confound the relationship between the explanatory variable and outcome. Second, there exists a possibility of reverse causality.
Sometimes, it is possible to overcome the aforementioned limitations using instrumental variables (IVs; [ 4]). Instrumental variables are factors that cause changes in the explanatory variable of interest (e.g., consumption habits) and have no other independent effects on the outcome (e.g., happiness), enabling one to estimate the causal effect without bias due to confounds and reverse causality. Under some circumstances, genetic measures can be used as IVs. This is possible because the transmission of genetic variants from parents to offspring is determined via a "genetic lottery" that is independent from environmental factors (conditioned on the parents' genomes). Furthermore, because genetic variations are not influenced by one's environment or habits, reverse causality is not a concern.
The most common method that uses genetic measures as IVs is Mendelian randomization (MR; [130]), which can be thought of as a natural experiment that occurs at the time of conception. Mendelian randomization uses genetic variants that have well-established causal influences on the explanatory trait as IVs to quantify the trait's causal effect on an outcome. For example, medical researchers have been using variants that regulate alcohol metabolism as IVs for studying the long-term causal effects of alcohol consumption on outcomes such as cardiovascular disease and cognitive decline (e.g., [26]). When using genetic data to infer causal relations, it is important to keep a careful eye on the assumptions of the methods used to estimate the effects. One crucial issue is that the transmission of genes occurs at random only within a family. Therefore, MR studies should ideally rely on within-family designs that compare genetic variation between related individuals (e.g., sibling pairs, parent–offspring trios). Mendelian randomization studies that do not use such designs are susceptible to biases of various sources ([34]). A second important assumption of MR is that the genes used as IVs affect the outcome only via their effect on the explanatory trait (a criterion called "exclusion restriction"). It is therefore important that the mechanisms linking the genetic IVs and the explanatory variable are well-understood, and that the genes' prevalence in the population studied does not correlate with unobservable environmental factors that might influence the outcome ([74]).
One limitation of MR is that most genetic variants are relatively weak instruments, because their associations with personal traits of interest are small. Moreover, genetic variants typically correlate with multiple traits that could influence an outcome, a phenomenon called "pleiotropy." Several statistical techniques that rely on summary statistics from large-scale GWAS (instead of single variants) have been recently proposed to overcome these issues ([36]; [108]; [156]). Each of these methods relies on a different set of assumptions concerning the relationships between genetics and other variables that are included in (or omitted from) the model, for estimating a causal effect. To mitigate concerns that claims of causality are driven by any specific assumptions, it is crucial to verify that a study's conclusion is consistent across methods. We anticipate that continuing development of such methods, together with the growing availability of data sets that include genetic measures of related-individuals, will provide a fertile ground for investigations of causal relations for a broader range of settings in the near future.
Genetic variation between individuals is fixed across the lifespan and can be related to many outcomes of interest to consumer researchers. As such, including genetic variables (most notably PRSs and genetic PCs) in statistical models that quantify any nongenetic effects provides a means to control for unobservable factors that would otherwise be a part of the model's error. Such reduction of the model's error would increase the study's statistical power and allow estimating model parameters of interest with less uncertainty ([11]). For illustration, consider a field experiment aiming to test the efficiency of different campaigns for preventing smoking initiation among teenagers. In such settings, PRSs can explain one's genetic tendency to smoke, as well as variance related to many preexisting personal characteristics that are not contaminated by the treatment and could be related to future smoking (e.g., extraversion). Including such PRSs in the model would therefore allow for quantifying the treatment effect more accurately.
Genetic measures are also useful in studying how consumers differentially respond to marketing stimuli or situational contexts. As noted previously, generic variants per se are not of great interest to marketers, but they allow for calculation of PRSs (based on any previously published GWAS) to approximate personal characteristics that cannot be easily measured in large samples (e.g., intelligence, personality) or when the participants' tendencies are not yet expressed behaviorally. Going back to the smoking-prevention field experiment example, constructing PRSs for many unobservable traits in the sample could be used for carrying a post hoc analysis to investigate whether certain individuals more strongly respond to a certain treatment versus another.
Because genetic variation correlates with many personal characteristics, it provides a means for studying the relationships between traits and whether they arise from genetic or environmental causes. A useful method for quantifying the genetic overlap between traits is estimating their genetic correlation (rg), which measures the amount of variance they share due to genetic causes ([90]). A useful feature of genetic correlations is that they can be estimated between any two traits for which GWAS has ever been conducted—even for traits that have not been measured in the same sample ([17]). A recent example for insight obtained from genetic correlations comes from a GWAS of general risk tolerance in a sample of over 1 million people (Karlsson [68]). This study found that the genetic correlations between general risk tolerance and many domain-specific risky behaviors—such as substance use, speeding on motorways, and self-employment—were substantially larger than the correlations observed between the behavioral phenotypes. This finding indicates that common genetic causes influence all these phenotypes, where the translation of this genetic tendency to each of the domain-specific risky behaviors depends on environmental factors.
Genetic data can enrich marketing theory by illuminating biological mechanisms that underlie behavior, akin to research in the field of consumer neuroscience ([113]). Apart from straightforward genetic effects on traits like lactose intolerance, genetic analyses can provide insight into how different brain systems mediate the influence of genetics on complex behavioral traits, such as economic preferences and consumption patterns. Although brain imaging studies have long ago uncovered multiple systems that are functionally involved in emotional and cognitive processes, linking functional brain measures to differences across individuals is not straightforward, because of their low test-retest reliability ([39]) and the high cost of obtaining such measures at scale. Genetic variation, in contrast, can be measured reliably and inexpensively in large samples, and once genetic variants are linked to a behavioral trait, they can be tied to neurobiological systems via bioinformatic tools (e.g., [47]). For example, the recent GWAS of general risk tolerance pointed to multiple brain systems that are genetically associated with the trait, including the prefrontal cortex, the amygdala and mid-brain regions involved in reward processing ([68]). An alternative promising approach is to derive biologically informed PRSs, which reflect aggregate effects of variants related to known biological systems (e.g., the dopaminergic genes) on a target phenotype, and investigate their relationship with biological endophenotypes ([32]). The rapid development of bio-annotation techniques, together with the formation of data sets that include both genetic and brain-imaging measures ([ 6]), will facilitate additional discoveries of gene-brain-behavior pathways in the near future.
Similar to other data types, some marketing uses of genetic data can improve individuals' well-being and have a positive impact on society as a whole. For example, focused early interventions based on genetic data may help health care providers reach patients at high risk for conditions such as diabetes and hypertension and provide them strategies that mitigate these risks (e.g., via physical exercise and diet; [146]). However, genetic data might facilitate manipulation and exploitation of vulnerable individuals ([133]). For example, e-cigarette companies could use genetic data to target teenagers who are more genetically prone to develop nicotine addiction ([121]). Yet the use of human genetic data by marketers raises even further ethical and legal challenges. These issues are the result of several unique features of genetic data, which contain immutable and identifiable information that is predictive of future behavior and disease, both for the individual and their genetic relatives. For this reason, genetic data have been considered particularly sensitive even within the medical field, a view known as "genetic exceptionalism" ([56]). In this section, we highlight serious ethical issues that emerge from these unique properties, review the current state of legislation in this area, and propose possible solutions.
Except for monozygotic twins, genetic data can be uniquely attributed to one person: A mere 60 to 300 randomly selected SNPs are sufficient to identify an individual ([154]). Anonymizing genetic data without destroying a large share of the information is not a simple task. Some methods, for instance, try to balance anonymity and information preservation by clustering the data before analysis ([88]). Even when the data are labeled as anonymized, however, the inherent information they contain could allow for potential reidentification attacks ([150]). Due to the combination of this unique identifiability property and the rich information content of genetic data, using them for research requires obtaining informed consent from study participants ([12]). Nonetheless, even in the ethically stricter research setting, acceptable anonymization and consent practices have been subject to heterogenous standards ([38]).
Because most current human genetic research involves analysis of secondary data that have been typically collected long before hypotheses are formed, obtaining consent is challenging. A common solution has been to ask participants to consent for all future research that falls within a broadly defined scope. For example, 23andMe informs customers who volunteer to participate in research that "the topics to be studied span a wide range of traits and conditions" and that "some of these studies may be sponsored by or conducted on behalf of third parties."[ 9] Similar consent procedures are used in practice by other DTC-GT firms and biobanks. Advocates of the broad consent approach argue that it provides an ideal trade-off between participants' autonomy and the public interest to benefit from research outputs ([59]). However, it is unclear whether genetic research subjects can fully appreciate the potential benefits and risks of any future research at the time of consent. For instance, it is unlikely that 23andMe customers could foresee that access to their data would be sold to a pharmaceutical company under the broad label of "research." To overcome these issues, scholars have proposed using dynamic or hybrid consenting protocols, where individuals can opt in to studies or withdraw their consent online ([71]; [114]).
To complicate matters further, one's genetic data are informative not only about oneself but also about one's nongenotyped relatives. This issue was recently illustrated in the apprehension of Joseph James DeAngelo, the alleged Golden State Killer, who was arrested after a fraction of his genome could be matched to the DNA of distant relatives, who uploaded their genetic data to a searchable public genealogical database ([118]). Although relatives of genetic research participants are potentially identifiable, current guidelines do not require obtaining their consent yet recommend that participants consult relatives when deciding to take part in research ([98]). These guidelines may change in the future, as genetic identification technology advances.
In summary, several unique issues make it difficult to anonymize data and obtain fully informed consent from participants of genetic research. Because this is an active area of study, we recommend that researchers closely monitor the emerging literature on the topic and ensure that their studies comply with the latest ethical guidelines. It is imperative that analyses of publicly available genetic data, collected thanks to public funding, produce discoveries that benefit society as a whole. As for research using privately owned genetic data, it is crucial that informed consent is obtained and that all studies fall beyond a shadow of doubt under the scope of research to which participants had consented.
Many of the features that turn genetic data into a marketing opportunity also raise fundamental privacy and security challenges. Genetic data are identifiable, predictive of virtually every aspect of one's life, and are even informative about one's relatives—and thus, could enable firms to target individuals who never opted to share any information. Given that major companies have been known to keep "shadow profiles" of individuals who did not register for their services ([50]), this potential privacy threat is imminent. The assembly of privately owned genetic databases also gives rise to security concerns, as major data breaches become increasingly common ([27]). In these cases, third parties obtain data against the will of both the consumer and the data holder. Once leaked, data will likely be used regardless of any regulation or ethical norm.
As massive volumes of genetic data reside on the servers of private firms, the question arises as to whether legislation and practice sufficiently protect the privacy of consumers from having their data exploited against their interests. While leading DTC-GT companies argue that their research complies with ethical guidelines, and they have a clear interest to avoid public controversies, it is unclear whether they follow the same principles when using data for marketing. As of July 2020, market leader 23andMe indicates in its (unilaterally modifiable) privacy statement that it would not process genetic data for marketing purposes without explicit consent, implying that it may do so if consent is given.[10] Furthermore, ethical recommendations are likely not a priority for all entities that own genetic data. In the absence of legal regulation and transparency, it becomes difficult to know exactly how private companies use the data.
Surprisingly, current federal laws in the United States concerning the use of genetic data have little implications on the DTC-GT industry, and U.S. lawmakers have mostly remained silent (with some notable exceptions) regarding potential regulations on the use of genetic data. As a consequence, the license to use and share genetic data for marketing purposes depends on the privacy policy of each individual company. Currently, a large number of U.S.-based DTC-GT companies do not provide their customers with any privacy information (on their website or the testing-kit packages) prior to the purchase of DNA kits, and the policies of many of the remaining companies indicate that they may use genetic data for purposes other than delivering ancestry and health reports. Furthermore, companies often reserve the right to share genetic data with third parties in cases of merger, acquisition, or bankruptcy, or to modify their privacy policies without notification ([62]).
In contrast to U.S. federal law, the recent European General Data Protection Regulation, commonly known as GDPR, explicitly recognizes genetic data as "sensitive" under Article 9 ([125]) and provides unique protection against sharing of genetic data (even semianonymized). Under current European regulations, one has to provide "explicit consent to the processing of personal data for one or more specified purposes."[11] Nonetheless, consumers have been known to easily approve mining of their data without reading the legal terms and services conditions ([107]). Once such consent is provided, virtually every marketing application becomes possible, despite the strict sharing restrictions in place. Furthermore, DTC-GT companies can process genetic data and use them for running marketing campaigns on behalf of other companies, without having to directly share them. For example, DTC-GT companies can offer to forward a message to a subsample of their clients satisfying some criteria on behalf of other entities, without disclosing any data, just as Facebook allows advertisers to target its own users without sharing their data ([96]). Thus, regulatory limits to genetic data sharing may end up simply granting DTC-GT companies a monopoly over the data. Finally, it is important to recognize that the power of regulation might be limited. Industry practices typically advance faster than the policies trying to regulate them, with regulations doing too little too late after malpractice had already been exposed ([64]). Moreover, technology giants have a long history of violating data protection laws and do not appear to be deterred by financial disincentives, as indicated by numerous condemnations and legal battles between regulatory agencies and these entities.
A possible solution to the privacy and security issues—which, in our view, is crucial for the continuing growth of the DTC-GT market—is adoption of industry standards that guarantee acceptable practices. One such framework, previously proposed to address similar challenges in artificial intelligence (AI) research, can be directly applied to genetic data ([136]). This framework, namely the "Four Pillars of Perfectly Privacy-Preserving AI," articulates four principles for maintaining privacy, security, and usability of data: ( 1) training data privacy: a malicious actor will not be able to recover genetic data from other accessible information (e.g., model output); ( 2) input privacy: a user's genetic data should not be observed by other parties, including the model creator; ( 3) output privacy: the output of a model should not be visible by anyone except for the user whose data are being analyzed; and ( 4) model privacy: the model (trained or not) should be protected from being stolen by a malicious party.
A specific strength of this framework is that privacy is considered from both sides. From the consumer side, data and the inferences (e.g., genetic reports, ads selected for the consumer) are not visible to the company. From the company's side, its algorithms and parameters (e.g., GWAS weights) are not visible to the consumer. Importantly, no data have to be kept on the consumer side if a sufficiently strong encryption algorithm is applied to them before delivery to third-party servers. Thus, the consumer would only need to preserve the encryption key.
While algorithms satisfying some or all of the aforementioned criteria are still under active research, several methods to perform privacy-preserving GWAS already exist ([65]; [141]; [153]). With these methods, the GWAS's summary statistics (e.g., weights, p-values) are known to the analyst, yet the PRSs can only be computed on the consumer side. Concurrently, general methods to allow for privacy-preserving versions of machine learning algorithms are being developed and can be expected to be adapted to genetic data mining, following their nonprivacy preserving counterparts ([85]; [127]). An additional advantage of such methods is that users can withdraw their data from the pool unilaterally by deleting either the data or the encryption key. However, even though such methods are constantly being developed, it is far from clear whether companies will end up adopting them. Major industry actors might not feel compelled to change their practice without strong incentives. A possible solution would be to enforce the use of these technologies through regulations. We can, for instance, picture a legal framework wherein a DTC-GT kit cannot be sold in a country without adhering to a framework of this type.
In November 2013, the U.S. Food and Drug Administration ordered 23andMe to suspend its genetic health reporting service until the company provided sufficient evidence to support clinical claims made in its reports. The company, which at that time had already sold half a million kits, relaunched the service only two years later, with less elaborate reporting that emphasized the probabilistic nature of genetic diagnosis ([115]). Although regulation of health-related genetic applications has tightened up, this cautionary tale illustrates how companies might use the scientific image of genetics in their consumers' minds to oversell the utility of genetic information. When doing so, marketers could rely on genetic data to make pseudo-scientific claims that promote the appeal of products and services, as commonly done in the wellness industry ([ 7]).
In the United States, because nonmedical genetic applications do not pose direct health risks to consumers, the Federal Trade Commission, rather than the Food and Drug Administration, is responsible for regulating potentially deceptive marketing messages that make claims on the utility of genetic data ([70]). However, such oversight might be difficult to exercise, for three main reasons. First, although genetic data are only moderately informative of most human behavioral traits, they do indeed contain some information. As a result, it is difficult to argue that genetic-based recommendations are entirely deceitful. Second, consumers' perceptions of genetics ([155]) might make them prone to believe that genetic-based recommendations are always backed by scientific evidence, even when such claims are not made explicitly. Third, people have a poor intuitive sense of probabilities and thus might be prone to overestimate the informativeness of genetic-based recommendations even when their probabilistic nature is communicated ([140]). In our view, regulation should ensure that companies disclose the science underlying any scientific claims (and its limitations), attempt to communicate probabilistic information intuitively, and avoid the use of deterministic language when appropriate ([147]).
Similar to discrimination based on other unchangeable characteristics, negative treatment of individuals based on their actual (or assumed) genetic markup is a potential source of distress, exclusion, and loss of opportunities ([14]). Furthermore, such discrimination might deter individuals from taking genetic tests that could improve their health care or from participating in genetic research that benefits society as a whole ([67]). To date, most conversations concerning genetic discrimination among ethics and law scholars have focused on potential abuses of genetic data by insurance providers and employers ([66]; [81]). Yet marketing applications of genetic data give rise to similar concerns. Aeroméxico's aforementioned DNA-discounts campaign is a recent prominent example of what is essentially genetic-based price discrimination. While it is unclear whether customers indeed received DNA discounts, the campaign was covered by major popular media outlets and, in general, was received positively by the public.
From a legal standpoint, the 1996 Health Insurance Portability and Accountability Act and 2008 Genetic Information Nondiscrimination Act prohibit insurance companies (for specific types of policies) and employers from discriminating against people based on genetic information, but they do not protect individuals from discrimination in other circumstances. However, some state laws, most notably California's Unruh Civil Rights Act, explicitly ban businesses from discriminating against consumers based on genetic information. Similarly, Florida statutes have provisions requiring notification of an individual if genetic information was used in any decision to grant or deny any insurance, employment, mortgage, loan, credit, or educational opportunity. In our view, discrimination based on one's genetic information is a serious issue that should be addressed in the same way as other types of discrimination.
A final nontrivial concern is that marketing strategies that rely on consumers' genes for predicting their preferences and behavior might generate self-reinforcing loops ([55]) that perpetuate inequality and deprive consumer's exploration of options that do not align with their genetic markup. For example, providers of SAT preparation kits could offer promotions to high school students who are genetically disposed to higher education ([80]) and, by doing so, give preferential treatment to individuals who are already in an advantageous position.
Forthcoming discoveries in the field of behavioral genetics will undoubtedly advance our understanding of how genetics interacts with the environment to influence behavior. However, assessing the utility of genetic tools for the advancement of marketing theory and practice, and accurately evaluating the severity of ethical concerns, would require addressing several gaps of knowledge in the current literature (summarized in the Appendix).
Many genetic associations of phenotypes that are of interest to marketers have been identified over the past decade. Nonetheless, the genetic underpinnings of many traits that are more closely tied to consumer behavior and are known to be heritable (see Table 1) have remained elusive. There are two likely reasons for this gap. First, marketing scholars have largely neglected the influence of genetics on consumer behavior (with a few notable exceptions, e.g., [128]). Research in related fields, however, points to genetic effects on many traits that are central to consumer behavior theory and practice. Examples include investment decisions ([23]; [31]), altruism and trust ([22]; [110]), susceptibility to placebo effects ([58]), voting turnout ([89]), and mobile phone usage patterns ([99]). Molecular genetic studies of these traits would be a straightforward extension that can enrich marketing theory and support industry applications.
Second, genetic data sets that include fine-grained behavioral measures are scarce. Behavioral geneticists have overcome this limitation by using measures that are more readily available at scale as proxies for traits that are laborious to measure, an approach that was shown to boost statistical power of genetic discovery ([122]). Genetic research of consumer behavior can similarly benefit from such an approach. For example, a twin study found that one's disposition to display decision biases shares a common genetic variance with performance in the cognitive reflection test ([24]), suggesting that this brief measure could serve as a proxy for such behavioral dispositions. Another possible solution would be to preselect genetic loci that have already been identified as associated with related phenotypes in large-scale GWAS. This would drastically reduce the number of hypotheses to be tested and, thus, the sample size required to obtain sufficient statistical power.
On a final note, the capacity (or lack thereof) to obtain detailed phenotypic measures at scale to complement the genetic measures may be less of a concern for behavioral marketing metrics. Companies in the DTC-GT industry possess relationship management data of millions of customers and likely know whether they were early adopters, responded to email advertisements, shared coupons with their friends, and consulted health or ancestry reports and, furthermore, what device was used to access them. Thus, large-scale genetic data sets that contain high-resolution measures of consumer behavior already exist and could be employed to unveil the genetic foundations of many aspects of consumer behavior. Such explorations will generate insights that advance not only the field of marketing, but also the discipline of behavioral genetics.
Behavioral genetic research typically focuses on identifying variants that have causal effects on a target trait and quantifying the variance they account for. Many marketing applications of genetic data, however, do not depend on whether genetic variants are indeed causally related to a trait but, rather, on whether they are more informative than other readily available measures. These two questions are not interchangeable for two reasons. First, genomes correlate with many personal characteristics that have no genetic basis. In traditional genetic analysis, noncausal correlations are of no interest ([116]). For marketing applications, however, noncausal genetic associations carry information that is useful for identifying segments and reaching targets. Second, many behavioral dispositions can be accurately predicted from records of their downstream consequences (e.g., personality can be estimated from digital footprints; [75]; [102]). This empirical observation is not of particular interest to geneticists, yet it is crucial for marketers deciding on what data to base their strategy. As noted previously, the degree to which genomes are more predictive than other measures likely varies by trait. To the best of our knowledge, only one study to date (whose outcome measure was longevity) systematically compared the predictive accuracy of models that use different sets of variables ([69]). This constitutes an important gap that should be filled as genetic data sets become more available to marketing researchers.
Marketing applications, such as segmentation and targeting, often depend on identifying people at the extreme ends of a trait's distribution as opposed to explaining the variance in the general population. For example, a manager of an eco-friendly luxury car brand would be interested in reaching people who are willing to pay a lot for "green" products ([78]) rather than accounting for heterogeneity in this tendency in the general population. However, the goal of most behavioral genetic research to date has been to estimate how much of a trait's variance in the general population can be attributed to genetics (using summary statistics such as estimated heritability or R2). Future studies should shed light on the capacity to use genetic data for identifying segments at the extreme ends of the behavioral distribution, using techniques such as discriminant analysis.
As described in the "Applications for Marketing Strategy" section, there are several cases when a marketing researcher can use genetic data to accurately predict a variable of interest, denoted by y (e.g., propensity for pattern baldness). However, genetic information might not be available for the entire population of potential consumers. In such cases, it may be possible to leverage the share of the population with genetic information to predict y for the remaining nongenotyped population. To this end, researchers must first predict the variable of interest in the population using genetic information (e.g., using a genetic estimator y′, such as a PRS) and then estimate a model to capture the link between nongenetic variables (e.g., demographics) and the predicted variable of interest y′. Finally, the model can be used to predict the variable of interest in the nongenotyped population, without having to rely on genetic data. The feasibility of this approach crucially depends on the capacity to estimate y′, which is a function of the genome, from other observables. We are not aware of any research relying on this approach to date, and its potential performance remains to be studied. Answering this question is crucial for evaluating the utility of genetic variables as segmentation bases.
A final important open question concerns how consumers would feel about the use of their genetic data by marketers. On the one hand, it seems plausible that at least some people would welcome marketing applications of genetic data if it got them discounts or better recommendations, which saves search costs. On the other hand, such usage is expected to raise privacy concerns that are similar to those invoked in relation to other data types. Yet there are several additional unique matters, related to the image of genetics in the minds of consumers. One major concern relates to historical misconceptions surrounding genetics, which were used in the past to justify racist worldviews and policies responsible for some of the worst crimes against humanity ([72]). Business strategies that insensitively use consumers' genetic data might therefore invoke strong negative reactions. Furthermore, although the true causal effects of genetic factors on most human traits are moderate in size and occur via interactions with the environment, genetics is often associated with biological determinism ([30]). As such, the use of genetics for matching consumers with products, services, and ads might increase beliefs in the existence of potentially deterministic aspects of behavior ([13]; [155]) and threaten consumers' sense of autonomy ([145]). A final concern is that an individual's genome contains sensitive information that they may not be aware of, for example, about future health risks such as cancer or Parkinson's disease. Marketers should be cautious to avoid exposing consumers to information they might not want to know ([51]) or prefer to receive with the appropriate counseling in a medical setting.
The substantial size of the DTC-GT market, despite poor regulation, suggests that these issues may not be a major concern for many customers. Moreover, many individuals voluntarily share their genetic data with third-party interpretation services ([57]) and websites that use them solely for making product recommendations. Indeed, in addition to ancestry-based playlists (Spotify) and cultural experiences (Airbnb), other services have recently emerged, recommending wines, travel destinations, and even romantic matches purportedly tailored to their consumers' DNA. Yet it is possible that the market trends merely reflect consumer ignorance. A recent survey found that while many DTC-GT customers presumed that they were sufficiently informed about privacy issues, their expectations were often inconsistent with company practices ([28]). For example, consumers' most common expectation, that DTC-GT companies would not share their data with third parties, was often at odds with the firms' actual privacy policies. Thus, it remains to be seen whether consumers' attitudes toward the use of genetic data for marketing differ from how they (dis)regard the use of other types of digital records, and whether there are means to mitigate such effects (e.g., by increasing transparency; [73]).
This article is a first attempt to assess how the massive amounts of data accumulated in genetic databases will influence the field of marketing. We developed a framework that incorporates genetic variables into consumer behavior theory and used it to explore potential applications of genetic data in marketing. We further evaluated ethical and legal challenges, and we highlighted gaps of knowledge that should be addressed by future research. Despite the gaps of knowledge in the published literature, we note that DTC-GT firms and governments already have access to the genetic data of millions of individuals. Therefore, business strategies that employ genetic data are likely already implemented, to some degree, by organizations. With the fast accumulation of genetic data and the rapid advances in methodology for genetic-based inference, the use of genetic data for marketing research and practice is likely to become increasingly common in the future.
- Which specific genetic variants are linked to relevant marketing outcomes (e.g., customer relationship management measures)?
- To what extent is genetic data predictive of consumer behavior, above and beyond nongenetic variables traditionally used in marketing?
- Do genetic data allow identification of individuals at the extreme ends of distributions (e.g., heavy espresso drinkers)?
- To what degree can genetic variation be approximated from nongenetic measures traditionally used in marketing (e.g., geodemographics)?
- How will consumers react to the use and monetization of their genetic data by marketers?
Footnotes 1 Vikas Mittal
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Gideon Nave https://orcid.org/0000-0001-6251-5630
5 Aeroméxico has yet to comment on whether DNA discounts were indeed given as advertised.
6 There are at least 88 million known human genetic polymorphisms, many of them are located outside genes. Non-SNP variants include short sequence insertions or deletions (called indels) and structural variants such as inversions, deletions, duplications, and translocations ([5]).
7 The dairy alternatives market is expected to reach $41.80 billion by 2026 ([54]).
8 Although genetic ancestry correlates with race and ethnicity, the latter two are socially defined categories that vary by culture ([129]).
9 See https://www.23andme.com/about/consent/ (accessed January 25, 2021).
See https://www.23andme.com/about/privacy/ (accessed January 25, 2021).
See https://gdpr-info.eu/art-9-gdpr/ (accessed January 25, 2021).
References Adhikari Bhim M. , Jahanshad Neda , Shukla Dinesh , Glahn David C. , Blangero John , Fox Peter T. , et al. (2018), " Comparison of Heritability Estimates on Resting State fMRI Connectivity Phenotypes Using the ENIGMA Analysis Pipeline, " Human Brain Mapping , 39 (12), 4893 – 4902.
Alexander David H. , Novembre John , Lange Kenneth. (2009), " Fast Model-Based Estimation of Ancestry in Unrelated Individuals ," Genome Research , 19 (9), 1655 – 64.
Al-Khelaifi Fatima , Diboun Ilhame , Donati Francesco , Botrè Francesco , Abraham David , Hingorani Aroon , et al. (2019), " Metabolic GWAS of Elite Athletes Reveals Novel Genetically-Influenced Metabolites Associated with Athletic Performance ," Scientific Reports , 9 (1), 19889.
Angrist Joshua D. , Imbens Guido W. , Rubin Donald B.. (1996), " Identification of Causal Effects Using Instrumental Variables ," Journal of the American Statistical Association , 91 (434), 444 – 55.
Auton Adam , Brooks Lisa D. , Durbin Richard M , Garrison Erik P. , Kang Hyun Min , Korbel Jan O. , et al. (2015), " A Global Reference for Human Genetic Variation ," Nature , 526 (7571), 68 – 74.
Aydogan G. , Daviet R. , Karlsson Linnér R. , Hare T.A. , Kable J.W. , Kranzler H.R. , et al. (2021), " Genetic Underpinnings of Risky Behavior Relate to Altered Neuroanatomy ," Nature Human Behaviour , https://www.nature.com/articles/s41562-020-01027-y.
Baker Stephanie Alice , Rojek Chris. (2020), " The Online Wellness Industry: Why It's So Difficult to Regulate ," The Conversation , (February 20), https://theconversation.com/the-online-wellness-industry-why-its-so-difficult-to-regulate-131847.
Baselmans Bart M.L. , Jansen Rick , Ip Hill F. , Dongen Jenny van , Abdellaoui Abdel , Weijer Margot P. van de , et al. (2019), " Multivariate Genome-Wide Analyses of the Well-Being Spectrum ," Nature Genetics , 51 (3), 445 – 51.
Bataille Veronique , Snieder Harold , MacGregor Alex J. , Sasieni Peter , Spector Tim D.. (2000), " Genetics of Risk Factors for Melanoma: An Adult Twin Study of Nevi and Freckles ," Journal of the National Cancer Institute , 92 (6), 457 – 63.
Belk Russell W.. (1975), " Situational Variables and Consumer Behavior ," Journal of Consumer Research , 2 (3), 157 – 64.
Benjamin Daniel J. , Cesarini David , Chabris Christopher F. , Glaeser Edward L. , Laibson David I. , Guðnason Vilmundur , et al. (2012), " The Promises and Pitfalls of Genoeconomics ," Annual Review of Economics , 4 (1), 627 – 62.
Beskow L.M. , Burke W. , Merz J.F. , Barr P.A. , Terry S. , Penchaszadeh V.B. , et al. (2001), " Informed Consent for Population-Based Research Involving Genetics ," JAMA , 286 (18), 2315 – 21.
Bhattacharjee Amit , Berger Jonah , Menon Geeta. (2014), " When Identity Marketing Backfires: Consumer Agency in Identity Expression ," Journal of Consumer Research , 41 (2), 294 – 309.
Billings P.R. , Kohn M.A. , de Cuevas M. , Beckwith J. , Alper J.S. , Natowicz M.R.. (1992), " Discrimination as a Consequence of Genetic Testing ," American Journal of Human Genetics , 50 (3), 476 – 82.
Blauwendraat Cornelis , Nalls Mike A. , Singleton Andrew B.. (2020), " The Genetic Architecture of Parkinson's Disease ," The Lancet Neurology , 19 (2), 170 – 78.
Brodwin Erin. (2018), " DNA Testing Company 23andMe Has Signed a $300 Million Deal with a Drug Giant. Here's How to Delete Your Data if That Freaks You Out ," Business Insider , (July 25), https://www.businessinsider.com/dna-testing-delete-your-data-23andme-ancestry-2018-7.
Bulik-Sullivan Brendan , Finucane Hilary K. , Anttila Verneri , Gusev Alexander , Day Felix R. , Loh Po-Ru , et al. (2015), " An Atlas of Genetic Correlations Across Human Diseases and Traits ," Nature Genetics , 47 (11), 1236 – 41.
Buniello Annalisa , MacArthur Jacqueline A.L. , Cerezo Maria , Harris Laura W. , Hayhurst James , Malangone Cinzia , et al. (2019), " The NHGRI-EBI GWAS Catalog of Published Genome-Wide Association Studies, Targeted Arrays and Summary Statistics 2019 ," Nucleic Acids Research , 47 (D1), D1005 – 12.
Bushong Benjamin , King Lindsay M. , Camerer Colin F. , Rangel Antonio. (2010), " Pavlovian Processes in Consumer Choice: The Physical Presence of a Good Increases Willingness-to-Pay ," American Economic Review , 100 (4), 1556 – 71.
Bycroft Clare , Freeman Colin , Petkova Desislava , Band Gavin , Elliott Lloyd T. , Sharp Kevin , et al. (2018), " The UK Biobank Resource with Deep Phenotyping and Genomic Data ," Nature , 562 (7726), 203 – 09.
Calladine Chris R. , Drew Horace R. , Luisi Ben F. , Travers Andrew A.. (2004), Understanding DNA: The Molecule and How It Works , 3rd ed. San Diego : Elsevier Academic Press.
Cesarini David , Dawes Christopher T. , Johannesson Magnus , Lichtenstein Paul , Wallace Björn. (2009), " Genetic Variation in Preferences for Giving and Risk Taking ," Quarterly Journal of Economics , 124 (2), 809 – 42.
Cesarini David , Johannesson Magnus , Lichtenstein Paul , Sandewall Örjan , Wallace Björn. (2010), " Genetic Variation in Financial Decision-Making ," Journal of Finance , 65 (5), 1725 – 54.
Cesarini David , Johannesson Magnus , Magnusson Patrik K.E. , Wallace Björn. (2012), " The Behavioral Genetics of Behavioral Anomalies ," Management Science , 58 (1), 21 – 34.
Chabris Christopher F. , Lee James J. , Cesarini David , Benjamin Daniel J. , Laibson David I.. (2015), " The Fourth Law of Behavior Genetics ," Current Directions in Psychological Science , 24 (4), 304 – 12.
Chen Lina , Smith George Davey , Harbord Roger M. , Lewis Sarah J.. (2008), " Alcohol Intake and Blood Pressure: A Systematic Review Implementing a Mendelian Randomization Approach ," PLoS Medicine , 5 (3), e52.
Cheng Long , Liu Fang , Yao Danfeng (Daphne). (2017), " Enterprise Data Breach: Causes, Challenges, Prevention, and Future Directions ," Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery , 7 (5), e1211.
Christofides Emily , O'Doherty Kieran. (2016), " Company Disclosure and Consumer Perceptions of the Privacy Implications of Direct-to-Consumer Genetic Testing ," New Genetics and Society , 35 (2), 101 – 23.
Collins Francis S. , Morgan Michael , Patrinos Aristides. (2003), " The Human Genome Project: Lessons from Large-Scale Biology ," Science , 300 (5617), 286 – 90.
Condit Celeste M. , Ofulue Nneka , Sheedy Kristine M.. (1998), " Determinism and Mass-Media Portrayals of Genetics ," American Journal of Human Genetics , 62 (4), 979 – 84.
Cronqvist Henrik , Siegel Stephan. (2014), " The Genetics of Investment Biases ," Journal of Financial Economics , 113 (2), 215 – 34.
Dass Shantala A. Hari , McCracken Kathryn , Pokhvisneva Irina , Chen Lawrence M. , Garg Elika , Nguyen Thao T.T. , et al. (2019), " A Biologically-Informed Polygenic Score Identifies Endophenotypes and Clinical Conditions Associated with the Insulin Receptor Function on Specific Brain Regions ," EBioMedicine , 42 , 188 – 202.
Davies G. , Tenesa A. , Payton A. , Yang J. , Harris S.E. , Liewald D. , et al. (2011), " Genome-Wide Association Studies Establish That Human Intelligence Is Highly Heritable and Polygenic ," Molecular Psychiatry , 16 (10), 996 – 1005.
Davies Neil M. , Howe Laurence J. , Brumpton Ben , Havdahl Alexandra , Evans David M. , Smith George Davey. (2019), " Within Family Mendelian Randomization Studies ," Human Molecular Genetics , 28 (R2), R170 – 79.
Day Felix R. , Ong Ken K. , Perry John R.B.. (2018), " Elucidating the Genetic Basis of Social Interaction and Isolation ," Nature Communications , 9 (1), 2457.
DiPrete Thomas A. , Burik Casper A. P. , Koellinger Philipp D.. (2018), " Genetic Instrumental Variable Regression: Explaining Socioeconomic and Health Outcomes in Nonexperimental Data ," Proceedings of the National Academy of Sciences of the United States of America , 115 (22), E4970 – 79.
Duncan L. , Shen H. , Gelaye B. , Meijsen J. , Ressler K. , Feldman M. , Peterson R. , Domingue B.. (2019), " Analysis of Polygenic Risk Score Usage and Performance in Diverse Human Populations ," Nature Communications , 10 (1), 3328.
Elger Bernice S. , Caplan Arthur L.. (2006), " Consent and Anonymization in Research Involving Biobanks: Differing Terms and Norms Present Serious Barriers to an International Framework ," EMBO Reports , 7 (7), 661 – 66.
Elliott Maxwell L. , Knodt Annchen R. , Ireland David , Morris Meriweather L. , Poulton Richie , Ramrakha Sandhya , et al. (2020), " What Is the Test-Retest Reliability of Common Task-Functional MRI Measures? New Empirical Evidence and a Meta-Analysis ," Psychological Science , 31 (7), 792 – 806.
Eng Mimy Y. , Luczak Susan E. , Wall Tamara L.. (2007), " ALDH2, ADH1B, and ADH1C Genotypes in Asians: A Literature Review ," Alcohol Research & Health: The Journal of the National Institute on Alcohol Abuse and Alcoholism , 30 (1), 22 – 7.
Eraslan Gökcen , Avsec Žiga , Gagneur Julien , Theis Fabian J.. (2019), " Deep Learning: New Computational Modelling Techniques for Genomics ," Nature Reviews Genetics , 20 (7), 389 – 403.
Erzurumluoglu A. Mesut , Liu Mengzhen , Jackson Victoria E. , Barnes Daniel R. , Datta Gargi , Melbourne Carl A. , et al. (2020), " Meta-Analysis of up to 622,409 Individuals Identifies 40 Novel Smoking Behaviour Associated Genetic Loci ," Molecular Psychiatry , 25 (10), 2392 – 409.
Evans David M. , Martin Nicholas G.. (2000), " The Validity of Twin Studies ," GeneScreen , 1 (2), 77 – 79.
Ezkurdia Iakes , Juan David , Rodriguez Jose Manuel , Frankish Adam , Diekhans Mark , Harrow Jennifer , et al. (2014), " Multiple Evidence Strands Suggest That There May Be as Few as 19,000 Human Protein-Coding Genes ," Human Molecular Genetics , 23 (22), 5866 – 78.
Farr Christina. (2020), " 23andMe Lays off 100 People as DNA Test Sales Decline, CEO Says She Was 'Surprised' to See Market Turn ," CNBC (January 23), https://www.cnbc.com/2020/01/23/23andme-lays-off-100-people-ceo-anne-wojcicki-explains-why.html.
Federenko Ilona S. , Nagamine Mitsue , Hellhammer Dirk H. , Wadhwa Pathik D. , Wüst Stefan. (2004), " The Heritability of Hypothalamus Pituitary Adrenal Axis Responses to Psychosocial Stress Is Context Dependent ," Journal of Clinical Endocrinology and Metabolism , 89 (12), 6244 – 50.
Finucane Hilary K. , Bulik-Sullivan Brendan , Gusev Alexander , Trynka Gosia , Reshef Yakir , Loh Po-Ru , et al. (2015), " Partitioning Heritability by Functional Annotation Using Genome-Wide Association Summary Statistics ," Nature Genetics , 47 (11), 1228 – 35.
Flagel Shelly B. , Watson Stanley J. , Robinson Terry E. , Akil Huda. (2007), " Individual Differences in the Propensity to Approach Signals vs Goals Promote Different Adaptations in the Dopamine System of Rats ," Psychopharmacology , 191 (3), 599 – 607.
Frank Ronald Edward , Massey William F. , Wind Yoram. (1972), Market Segmentation. New York : Prentice Hall.
Garcia David. (2017), " Leaking Privacy and Shadow Profiles in Online Social Networks ," Science Advances , 3 (8), e1701172.
Gigerenzer Gerd , Garcia-Retamero Rocio. (2017), " Cassandra's Regret: The Psychology of Not Wanting to Know ," Psychological Review , 124 (2), 179 – 96.
Gilovich Thomas , Kumar Amit , Jampol Lily. (2015), " A Wonderful Life: Experiential Consumption and the Pursuit of Happiness ," Journal of Consumer Psychology , 25 (1), 152 – 65.
Giri Mohan , Zhang Man , Lü Yang. (2016), " Genes Associated with Alzheimer's Disease: An Overview and Current Status ," Clinical Interventions in Aging , 11 , 665 – 81.
Globe Newswire (2020), " Alopecia Treatment Global Industry to 2026 - Growth Opportunities in Emerging Markets ," press release (June 17) , https://www.globenewswire.com/news-release/2020/06/17/2049337/0/en/Alopecia-Treatment-Global-Industry-to-2026-Growth-Opportunities-in-Emerging-Markets.html.
Grafanaki Sofia. (2017), " Drowning in Big Data: Abundance of Choice, Scarcity of Attention and the Personalization Trap: A Case for Regulation ," Richmond Journal of Law & Technology , 24 (1), https://jolt.richmond.edu/volume24_issue1_grafanaki/.
Green Michael J. , Botkin Jeffrey R.. (2003), " Genetic Exceptionalism in Medicine: Clarifying the Differences between Genetic and Nongenetic Tests ," Annals of Internal Medicine , 138 (7), 571 – 75.
Guerrini Christi J. , Wagner Jennifer K. , Nelson Sarah C. , Javitt Gail H. , McGuire Amy L.. (2020), " Who's on Third? Regulation of Third-Party Genetic Interpretation Services ," Genetics in Medicine , 22 , 4 – 11.
Hall Kathryn T. , Loscalzo Joseph , Kaptchuk Ted J.. (2015), " Genetics and the Placebo Effect: The Placebome ," Trends in Molecular Medicine , 21 (5), 285 – 94.
Hansson Mats G. , Dillner Joakim , Bartram Claus R. , Carlson Joyce A. , Helgesson Gert. (2006), " Should Donors Be Allowed to Give Broad Consent to Future Biobank Research? " The Lancet Oncology , 7 (3), 266 – 69.
Harden K. Paige , Koellinger Philipp D.. (2020), " Using Genetics for Social Science ," Nature Human Behaviour , 4 (6), 567 – 76.
Hassan Aisha. (2018), " Spotify and Ancestry Can Use Your Real DNA to Tell Your 'Musical DNA' ," Quartz (September 22), https://qz.com/quartzy/1399279/spotify-can-use-your-ancestry-dna-test-to-tell-your-musical-dna/.
Hazel James W. , Slobogin Christopher. (2018), " Who Knows What, and When: A Survey of the Privacy Policies Proffered by US Direct-to-Consumer Genetic Testing Companies ," Cornell Journal of Law and Public Policy , 28 (1), 35 – 66.
Hu Youna , Shmygelska Alena , Tran David , Eriksson Nicholas , Tung Joyce Y. , Hinds David A.. (2016), " GWAS of 89,283 Individuals Identifies Genetic Variants Associated with Self-Reporting of Being a Morning Person ," Nature Communications , 7 , 10448.
Isaak Jim , Hanna Mina J.. (2018), " User Data Privacy: Facebook, Cambridge Analytica, and Privacy Protection ," Computer , 51 (8), 56 – 59.
Johnson Aaron , Shmatikov Vitaly. (2013), " Privacy-Preserving Data Exploration in Genome-Wide Association Studies ," in Proceedings of the 19th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York : Association for Computing Machinery , 1079 – 87.
Joly Yann , Braker Maria , Huynh Michael Le. (2010), " Genetic Discrimination in Private Insurance: Global Perspectives ," New Genetics and Society , 29 (4), 351 – 68.
Joly Yann , Feze Ida Ngueng , Song Lingqiao , Knoppers Bartha M.. (2017), " Comparative Approaches to Genetic Discrimination: Chasing Shadows? " Trends in Genetics , 33 (5), 299 – 302.
Karlsson Linnér , Pietro Biroli Richard , Kong Edward , Meddens S. Fleur W. , Wedow Robbee , Fontana Mark Alan , et al. (2019), " Genome-Wide Association Analyses of Risk Tolerance and Risky Behaviors in over 1 Million Individuals Identify Hundreds of Loci and Shared Genetic Influences ," Nature Genetics , 51 (2), 245 – 57.
Karlsson Linnér Richard , Koellinger Philipp D.. (2020), " Genetic Risk Scores of Disease and Mortality Capture Differences in Longevity, Economic Behavior, and Insurance Outcomes ," medRxiv (April 2) , https://www.medrxiv.org/content/10.1101/2020.03.30.20047290v1.
Kasperbauer T.J. , Wright David E.. (2020), " Expanded FDA Regulation of Health and Wellness Apps ," Bioethics , 34 (3), 235 – 41.
Kaye Jane , Whitley Edgar A. , Lund David , Morrison Michael , Teare Harriet , Melham Karen. (2015), " Dynamic Consent: A Patient Interface for Twenty-First Century Research Networks ," European Journal of Human Genetics , 23 (2), 141 – 46.
Kevles Daniel J.. (1985), In the Name of Eugenics: Genetics and the Uses of Human Heredity. Cambridge, MA : Harvard University Press.
Kim Tami , Barasz Kate , John Leslie K.. (2019), " Why Am I Seeing This Ad? The Effect of Ad Transparency on Ad Effectiveness ," Journal of Consumer Research , 45 (5), 906 – 32.
Koellinger Philipp D. , Vlaming Ronald de. (2019), " Mendelian Randomization: The Challenge of Unobserved Environmental Confounds ," International Journal of Epidemiology , 48 (3), 665 – 71.
Kosinski Michal , Stillwell David , Graepel Thore. (2013), " Private Traits and Attributes Are Predictable from Digital Records of Human Behavior ," Proceedings of the National Academy of Sciences of the United States of America , 110 (15), 5802 – 5.
Kuchenbaecker Karoline B. , Hopper John L. , Barnes Daniel R. , Phillips Kelly-Anne , Mooij Thea M. , Roos-Blom Marie-José , et al. (2017), " Risks of Breast, Ovarian, and Contralateral Breast Cancer for BRCA1 and BRCA2 Mutation Carriers ," JAMA , 317 (23), 2402 – 16.
Kwon Jennifer M. , Goate Alison M.. (2000), " The Candidate Gene Approach ," Alcohol Research & Health , 24 (3), 164 – 68.
Laroche Michel , Bergeron Jasmin , Barbaro-Forleo Guido. (2001), " Targeting Consumers Who Are Willing to Pay More for Environmentally Friendly Products ," Journal of Consumer Marketing , 18 (6), 503 – 20.
Lee Fiona X.Z. , Houweling Peter J. , North Kathryn N. , Quinlan Kate G.R.. (2016), " How Does α-Actinin-3 Deficiency Alter Muscle Function? Mechanistic Insights into ACTN3, the 'Gene for Speed' ," Biochimica et Biophysica Acta , 1863 (4), 686 – 93.
Lee James J. , Wedow Robbee , Okbay Aysu , Kong Edward , Maghzian Omeed , Zacher Meghan , et al. (2018), " Gene Discovery and Polygenic Prediction from a 1.1-Million-Person GWAS of Educational Attainment ," Nature Genetics , 50 , 1112 – 21.
Lemmens Trudo. (2004), " Genetics and Insurance Discrimination: Comparative Legislative, Regulatory and Policy Developments and Canadian Options ," SSRN (February 24) , https://papers.ssrn.com/sol3/papers.cfm?abstract_id=495404.
Lester Barry M. , Conradt Elisabeth , Marsit Carmen. (2016), " Introduction to the Special Section on Epigenetics ," Child Development , 87 (1), 29 – 37.
Lewis Ricki. (2017), Human Genetics: The Basics , 2nd ed. New York : Routledge.
Li Jun Z. , Absher Devin M. , Tang Hua , Southwick Audrey M. , Casto Amanda M. , Ramachandran Sohini , et al. (2008), " Worldwide Human Relationships Inferred from Genome-Wide Patterns of Variation ," Science , 319 (5866), 1100 – 1104.
Li Ping , Li Jin , Huang Zhengan , Li Tong , Gao Chong-Zhi , Yiu Siu-Ming , et al. (2017), " Multi-Key Privacy-Preserving Deep Learning in Cloud Computing ," Future Generation Computer Systems , 74 , 76 – 85.
Libbrecht Maxwell W. , Noble William Stafford. (2015), " Machine Learning Applications in Genetics and Genomics ," Nature Reviews Genetics , 16 (6), 321 – 32.
Lin Eugene , Lane Hsien-Yuan. (2017), " Machine Learning and Systems Genomics Approaches for Multi-Omics Data ," Biomarker Research , 5 (January), 2.
Lin Jun-Lin , Wei Meng-Cheng. (2009), " Genetic Algorithm-Based Clustering Approach for K-Anonymization ," Expert Systems with Applications , 36 (6), 9784 – 92.
Loewen Peter John , Dawes Christopher T.. (2012), " The Heritability of Duty and Voter Turnout ," Political Psychology , 33 (3), 363 – 73.
Lynch Michael , Walsh Bruce. (1998), Genetics and Analysis of Quantitative Traits. Sunderland, UK : Sinauer.
MacArthur Jacqueline , Bowler Emily , Cerezo Maria , Gil Laurent , Hall Peggy , Hastings Emma , et al. (2017), " The New NHGRI-EBI Catalog of Published Genome-Wide Association Studies (GWAS Catalog) ," Nucleic Acids Research , 45 (Database Issue), D896 – 901.
Margittai Zsofia , Nave Gideon , Strombach Tina , van Marijn , Wingerden , Schwabe Lars , Kalenscher Tobias. (2016), " Exogenous Cortisol Causes a Shift from Deliberative to Intuitive Thinking ," Psychoneuroendocrinology , 64 , 131 – 35.
Margittai Zsofia , Nave Gideon , Wingerden Marijn Van , Schnitzler Alfons , Schwabe Lars , Kalenscher Tobias. (2018), " Combined Effects of Glucocorticoid and Noradrenergic Activity on Loss Aversion ," Neuropsychopharmacology , 43 (2) 334 – 41.
Martin Alicia R. , Kanai Masahiro , Kamatani Yoichiro , Okada Yukinori , Neale Benjamin M. , Daly Mark J.. (2019), " Clinical Use of Current Polygenic Risk Scores May Exacerbate Health Disparities ," Nature Genetics , 51 (4), 584 – 91.
Mattar Rejane , Mazo Daniel Ferraz de Campos , Carrilho Flair José. (2012), " Lactose Intolerance: Diagnosis, Genetic, and Clinical Factors ," Clinical and Experimental Gastroenterology , 5 (July), 113 – 21.
Matz S. , Kosinski M. , Nave G. , Stillwell D.. (2017), " Psychographic Persuasion as an Effective Approach to Digital Mass Communication ," Proceedings of the National Academy of Sciences of the United States of America , 114 , (48), 12714 – 19.
McGue Matt , Lykken David T.. (1992), " Genetic Influence on Risk of Divorce ," Psychological Science , 3 (6), 368 – 73.
McGuire Amy L. , Caulfield Timothy , Cho Mildred K.. (2008), " Research Ethics and the Challenge of Whole-Genome Sequencing ," Nature Reviews Genetics , 9 (2), 152 – 56.
Miller Geoffrey , Zhu Gu , Wright Margaret J. , Hansell Narelle K. , Martin Nicholas G.. (2012), " The Heritability and Genetic Correlates of Mobile Phone Use: A Twin Study of Consumer Behavior ," Twin Research and Human Genetics , 15 (1), 97 – 106.
Mills Melinda C. , Rahal Charles. (2019), " A Scientometric Review of Genome-Wide Association Studies ," Communications Biology , 2 (January), 9.
Murphy Elizabeth. (2013), " Inside 23andMe Founder Anne Wojcicki's $99 DNA Revolution ," Fast Company (October 14), https://www.fastcompany.com/3018598/for-99-this-ceo-can-tell-you-what-might-kill-you-inside-23andme-founder-anne-wojcickis-dna-r.
Nave Gideon , Minxha Juri , Greenberg David M. , Kosinski Michal , Stillwell David , Rentfrow Jason. (2018), " Musical Preferences Predict Personality: Evidence from Active Listening and Facebook Likes ," Psychological Science , 29 (7), 1145 – 58.
Nave Gideon , Jung Wi Hoon , Karlsson Linnér Richard , Kable Joseph W. , Koellinger Philipp D.. (2018), " Are Bigger Brains Smarter? Evidence from a Large-Scale Preregistered Study ," Psychological Science , 30 (1), 43 – 54.
Nehlig Astrid. (2018), " Interindividual Differences in Caffeine Metabolism and Factors Driving Caffeine Consumption ," Pharmacological Reviews , 70 (2), 384 – 411.
Nelkin Dorothy , Lindee M. Susan. (1995), The DNA Mystique: The Gene as a Cultural Icon. Ann Arbor : University of Michigan Press.
Novembre John , Johnson Toby , Bryc Katarzyna , Kutalik Zoltán , Boyko Adam R. , Auton Adam , et al. (2008), " Genes Mirror Geography within Europe ," Nature , 456 (7218), 98 – 101.
Obar Jonathan A. , Oeldorf-Hirsch Anne. (2020), " The Biggest Lie on the Internet: Ignoring the Privacy Policies and Terms of Service Policies of Social Networking Services ," Information, Communication & Society , 23 (1), 128 – 47.
O'Connor Luke J. , Price Alkes L.. (2018), " Distinguishing Genetic Correlation from Causation Across 52 Diseases and Complex Traits ," Nature Genetics , 50 (12), 1728 – 34.
Ogilvie A.D. , Battersby S. , Fink G. , Harmar A.J. , Goodwin G.M. , Bubb V.J. , Smith C.D.. (1996), " Polymorphism in Serotonin Transporter Gene Associated with Susceptibility to Major Depression ," The Lancet , 347 (9003), 731 – 33.
Pedersen Ole Birger , Axel Skytthe , Rostgaard Klaus , Erikstrup Christian , Edgren Gustaf , Nielsen Kaspar R. , et al. (2015), " The Heritability of Blood Donation: A Population-Based Nationwide Twin Study ," Transfusion , 55 (9), 2169 – 74.
Pirastu Nicola , Joshi Peter K. , Vries Paul S. De , Cornelis Marilyn C. , McKeigue Paul M. , Keum NaNa , et al. (2017), " GWAS for Male-Pattern Baldness Identifies 71 Susceptibility Loci Explaining 38% of the Risk ," Nature Communications , 8 (1), 1584.
Plassmann Hilke , Ramsøy Thomas Zoëga , Milosavljevic Milica. (2012), " Branding the Brain: A Critical Review and Outlook ," Journal of Consumer Psychology , 22 (1), 18 – 36.
Plassmann Hilke , Venkatraman Vinod , Huettel Scott , Yoon Carolyn. (2015), " Consumer Neuroscience: Applications, Challenges, and Possible Solutions ," Journal of Marketing Research , 52 (4), 427 – 35.
Ploug Thomas , Holm Søren. (2015), " Meta Consent: A Flexible and Autonomous Way of Obtaining Informed Consent for Secondary Research ," BMJ , 350 , h2146.
Pollack Andrew. (2015), " 23andMe Will Resume Giving Users Health Data ," The New York Times , (October 21) , https://www.nytimes.com/2015/10/21/business/23andme-will-resume-giving-users-health-data.html.
Price Alkes L. , Patterson Nick J. , Plenge Robert M. , Weinblatt Michael E. , Shadick Nancy A. , Reich David. (2006), " Principal Components Analysis Corrects for Stratification in Genome-Wide Association Studies ," Nature Genetics , 38 (8), 904 – 09.
Queller David C. , Goodnight Keith F.. (1989), " Estimating Relatedness Using Genetic Markers ," Evolution , 43 (2), 258 – 75.
Ram Natalie , Guerrini Christi J. , McGuire Amy L.. (2018), " Genealogy Databases and the Future of Criminal Investigation ," Science , 360 (6393), 1078 – 79.
Redondo Maria J. , Steck Andrea K. , Pugliese Alberto. (2018), " Genetics of Type 1 Diabetes ," Pediatric Diabetes , 19 (3), 346 – 53.
Regalado Antonio. (2019), " More Than 26 Million People Have Taken an At-Home Ancestry Test ," MIT Technology Review (February 11), https://www.technologyreview.com/2019/02/11/103446/more-than-26-million-people-have-taken-an-at-home-ancestry-test/.
Richtel Matt , Kaplan Sheila. (2018), " Did Juul Lure Teenagers and 'Get Customers for Life'? " The New York Times (August 27) , https://www.nytimes.com/2018/08/27/science/juul-vaping-teen-marketing.html.
Rietveld Cornelius A. , Esko Tõnu , Davies Gail , Pers Tune H. , Turley Patrick , Benyamin Beben , et al. (2014), " Common Genetic Variants Associated with Cognitive Performance Identified Using the Proxy-Phenotype Method ," Proceedings of the National Academy of Sciences of the United States of America , 111 (38), 13790 – 94.
Sanchez-Roige Sandra , Gray Joshua C. , MacKillop James , Chen C.-H. , Palmer Abraham A.. (2018), " The Genetics of Human Personality ," Genes, Brain, and Behavior , 17 (3), e12439.
Schmitt Bernd , Brakus J. Joško , Zarantonello Lia. (2015), " From Experiential Psychology to Consumer Experience ," Journal of Consumer Psychology , 25 (1), 166 – 71.
Shabani Mahsa , Borry Pascal. (2018), " Rules for Processing Genetic Data for Research Purposes in View of the New EU General Data Protection Regulation ," European Journal of Human Genetics , 26 (2), 149 – 56.
Shimomura Yutaka , Wajid Muhammad , Ishii Yoshiyuki , Shapiro Lawrence , Petukhova Lynn , Gordon Derek , et al. (2008), " Disruption of P2RY5, an Orphan G Protein–Coupled Receptor, Underlies Autosomal Recessive Woolly Hair ," Nature Genetics , 40 (3), 335 – 39.
Shokri R. , Shmatikov V.. (2015), " Privacy-Preserving Deep Learning ," in Proceedings of the 22nd ACM SIGSAC Conference on Computer and Communications Security. New York : Association for Computing Machinery , 1310 – 21.
Simonson Itamar , Sela Aner. (2011), " On the Heritability of Consumer Decision Making: An Exploratory Approach for Studying Genetic Effects on Judgment and Choice ," Journal of Consumer Research , 37 (6), 951 – 66.
Smedley Audrey , Smedley Brian D.. (2005), " Race as Biology Is Fiction, Racism as a Social Problem Is Real: Anthropological and Historical Perspectives on the Social Construction of Race ," American Psychologist , 60 (1), 16 – 26.
Smith George Davey , Ebrahim Shah. (2004), " Mendelian Randomization: Prospects, Potentials, and Limitations ," International Journal of Epidemiology , 33 (1), 30 – 42.
Smith Stephen M. , Elliott Lloyd T. , Alfaro-Almagro Fidel , McCarthy Paul , Nichols Thomas E. , Douaud Gwenaëlle , et al. (2020), " Brain Aging Comprises Many Modes of Structural and Functional Change with Distinct Genetic and Biophysical Associations ," eLife , 9 , e52677.
Stark Rory , Grzelak Marta , Hadfield James. (2019), " RNA Sequencing: The Teenage Years ," Nature Reviews Genetics , 20 (11), 631 – 56.
Susser Daniel , Roessler Beate , Nissenbaum Helen. (2019), " Online Manipulation: Hidden Influences in a Digital World ," Georgetown Law Technology Review , 4 (1), https://georgetownlawtechreview.org/online-manipulation-hidden-influences-in-a-digital-world/GLTR-01-2020/.
Swede Helen , Stone Carol L. , Norwood Alyssa R.. (2007), " National Population-Based Biobanks for Genetic Research ," Genetics in Medicine , 9 (3), 141 – 49.
Taylor Amy E. , Smith George Davey , Munafò Marcus R.. (2018), " Associations of Coffee Genetic Risk Scores with Consumption of Coffee, Tea and Other Beverages in the UK Biobank ," Addiction , 113 (1), 148 – 57.
Thaine Patricia , Penn Gerald. (2020), " Perfectly Privacy-Preserving AI: What Is It and How Do We Achieve It? " WSDM 2020 Workshop on Privacy and Natural Language Processing , (February 7) , Houston, TX , 2573.
Thompson P.M. , Cannon Tyrone D. , Narr Katherine L. , Erp Theo van , Poutanen Veli-Pekka , Huttunen Matti , et al. (2001), " Genetic Influences on Brain Structure ," Nature Neuroscience , 4 (12), 1253 – 58.
Turkheimer Eric. (2000), " Three Laws of Behavior Genetics and What They Mean ," Current Directions in Psychological Science , 9 (5), 160 – 64.
Tutton Richard. (2004) " 'They Want to Know Where They Came From': Population Genetics, Identity, and Family Genealogy ," New Genetics and Society , 23 (1), 105 – 20.
Tversky Amos , Kahneman Daniel. (1983), " Extensional Versus Intuitive Reasoning: The Conjunction Fallacy in Probability Judgment ," Psychological Review , 90 (4), 293 – 315.
Uhlerop Caroline , Slavkovic Aleksandra , Fienberg Stephen E.. (2013), " Privacy-Preserving Data Sharing for Genome-Wide Association Studies ," Journal of Privacy and Confidentiality , 5 (1), 137 – 66.
Van den Bulte Christophe , Wuyts Stefan. (2007), Social Networks and Marketing. Cambridge, MA : Marketing Science Institute.
Visscher Peter M. , Wray Naomi R. , Zhang Qian , Sklar Pamela , McCarthy Mark I. , Brown Matthew A. , et al. (2017), " 10 Years of GWAS Discovery: Biology, Function, and Translation ," American Journal of Human Genetics , 101 (1), 5 – 22.
Vora Shivani. (2019), " Take a DNA Test, Then Buy an Airplane Ticket ," The New York Times , (January 22) , https://www.nytimes.com/2019/01/22/travel/ancestry-dna-test-travel.html.
Wertenbroch Klaus , Schrift Rom Y. , Alba Joseph W. , Barasch Alixandra , Bhattacharjee Amit , Giesler Markus , et al. (2020), " Autonomy in Consumer Choice ," Marketing Letters , 1 – 11.
Whelton Seamus P. , Chin Ashley , Xin Xue , He Jiang. (2002), " Effect of Aerobic Exercise on Blood Pressure: A Meta-Analysis of Randomized, Controlled Trials ," Annals of Internal Medicine , 136 (7), 493 – 503.
Williams-Jones Bryn , Ozdemir Vural. (2008), " Challenges for Corporate Ethics in Marketing Genetic Tests ," Journal of Business Ethics , 77 (1), 33 – 44.
Wind Yoram. (1994), " Marketing and Social Networks ," Advances in Social Network Analysis: Research in the Social and Behavioral Sciences , 171 , 254.
Wirth Thomas , Parker Nigel , Ylä-Herttuala Seppo. (2013), " History of Gene Therapy ," Gene , 525 (2), 162 – 69.
Wjst Matthias. (2010), " Caught You: Threats to Confidentiality Due to the Public Release of Large-Scale Genetic Data Sets ," BMC Medical Ethics , 11 , 21.
Yang Jian , Benyamin Beben , McEvoy Brian P. , Gordon Scott , Henders Anjali K. , Nyholt Dale R. , et al. (2010), " Common SNPs Explain a Large Proportion of the Heritability for Human Height ," Nature Genetics , 42 (7), 565 – 69.
Yengo Loic , Sidorenko Julia , Kemper Kathryn E. , Zheng Zhili , Wood Andrew R. , Weedon Michael N. , et al. (2018), " Meta-Analysis of Genome-Wide Association Studies for Height and Body Mass Index in ∼700,000 Individuals of European Ancestry ," Human Molecular Genetics , 27 (20), 3641 – 49.
Yu Fei , Ji Zhanglong. (2014), " Scalable Privacy-Preserving Data Sharing Methodology for Genome-Wide Association Studies: An Application to iDASH Healthcare Privacy Protection Challenge ," BMC Medical Informatics and Decision Making , 14 (Suppl 1), S3.
Zaaijer Sophie , Gordon Assaf , Speyer Daniel , Piccone Robert , Groen Simon Cornelis , Erlich Yaniv. (2017), " Rapid Re-Identification of Human Samples Using Portable DNA Sequencing ," eLife , 6 , e27798.
Zheng Yanmei , Alba Joseph W.. (2021), " Consumer Self-Control and the Biological Sciences: Implications for Marketing Stakeholders ," Journal of Marketing , https://doi.org/10.1177%2F0022242920983271.
Zhu Zhihong , Zheng Zhili , Zhang Futao , Wu Yang , Trzaskowski Maciej , Maier Robert , et al. (2018), " Causal Associations between Risk Factors and Common Diseases Inferred from GWAS Summary Data ," Nature Communications , 9 (1), 224.
Zou James , Huss Mikael , Abid Abubakar , Mohammadi Pejman , Torkamani Ali , Telenti Amalio. (2019), " A Primer on Deep Learning in Genomics ," Nature Genetics , 51 (1), 12 – 18.
~~~~~~~~
By Remi Daviet; Gideon Nave and Jerry Wind
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 58- Getting a Handle on Sales: Shopping Carts Affect Purchasing by Activating Arm Muscles. By: Estes, Zachary; Streicher, Mathias C. Journal of Marketing. Feb2022, p1. DOI: 10.1177/00222429211061367.
Ahead of Print- Database:
- Business Source Complete
Record: 59- Gift or Donation? Increase the Effectiveness of Charitable Solicitation Through Framing Charitable Giving as a Gift. By: Wang, Phyllis Xue; Wang, Yijie; Jiang, Yuwei. Journal of Marketing. Sep2022, p1. DOI: 10.1177/00222429221081506.
Ahead of Print- Database:
- Business Source Complete
Record: 60- Gimmicky or Effective? The Effects of Imaginative Displays on Customers' Purchase Behavior. By: Keh, Hean Tat; Wang, Di; Yan, Li. Journal of Marketing. Sep2021, Vol. 85 Issue 5, p109-127. 19p. 6 Color Photographs, 1 Diagram, 2 Charts, 1 Graph. DOI: 10.1177/0022242921997359.
- Database:
- Business Source Complete
Gimmicky or Effective? The Effects of Imaginative Displays on Customers' Purchase Behavior
Prior research indicates the strategic importance of the store environment in enhancing customers' shopping experience and their purchase decisions. This article examines the effects of imaginative displays on customers' purchase behavior. An imaginative display is constructed using multiple units of the same product in a novel or innovative yet aesthetically appealing form, which could be themed (i.e., having a particular shape mimicking an object) or unthemed. Six studies in both lab and field settings show that, relative to standard displays (i.e., non-novel and neutral aesthetics), imaginative displays can increase customers' purchase behavior and intentions. Importantly, for themed imaginative displays, these effects work through the dual mechanisms of affect-based arousal and cognition-based inferred benefits, which are contingent on congruence between display form and perceived product benefit. Findings from this research not only contribute to the literature on in-store display and store atmospherics but also have significant practical implications for retailers. Specifically, while imaginative displays may appear gimmicky, they can favorably influence customers' purchase behavior and increase product sales at relatively low costs.
Keywords: aesthetics; arousal; in-store display; inferred product benefits; novelty
Retail atmosphere, whether established in-store or through various retail touch-points, can influence customers' product choices and perceptions of the retailer ([59]). For instance, strategically altering the store environment can enhance customers' shopping experience and their purchase behavior ([ 1]; [33]). In particular, judicious use of in-store displays can influence customers' behavior and increase retailers' sales (e.g., [14]; [60]). Companies such as Coca-Cola often create novel and aesthetically appealing displays in stores, which have been shown to increase store sales (see the Web Appendix).
A survey of 2,400 supermarket shoppers indicates that half of them recalled seeing at least one display during their shopping trips, with endcap and freestanding displays being the most prominent; importantly, this survey also shows that one in six purchases were made when a display was present in store ([54]). Moreover, displays are particularly useful in generating unplanned purchases of regularly purchased product categories, with an increase of almost 40% from the baseline level ([32]), and are even more effective than temporary price reductions ([49]).
In this vein, prior research has examined two important facets of in-store display: display form ([12]) and display context ([72]). "Display form" refers to how the products are displayed (e.g., [17]; [50]), while "display context" refers to contextual cues surrounding the display (e.g., [22]; [72]). In extending the literature on display form and display context, the present research examines the effects of imaginative displays that are both novel (i.e., innovative) and aesthetically appealing on customers' purchase behavior. We consider design novelty and aesthetics two critical elements of imaginative displays, as customers tend to be attracted to new things and their response to the new display form is influenced by visual impression of the display ([55]). The visual design of a novel stimulus enables a retailer to gain customers' attention and stand out from its competitors ([ 7]; [56]).
Specifically, we investigate how imaginative displays that are novel and aesthetically appealing increase customers' purchase behavior compared with standard displays (i.e., non-novel and neutral aesthetics). To this end, we tap into the literatures on in-store display (e.g., [ 1]), innovativeness (e.g., [55]), and aesthetic design (e.g., [30]). We reveal two mechanisms underlying the effects of imaginative displays: affective response of arousal and cognitive response of inferred product benefits. Finally, we identify (in)congruence between inferred benefit from the display form and perceived product benefit as a contextual cue that moderates these core effects.
We conduct six empirical studies. Studies 1 and 2 are field experiments that show the effects of imaginative displays on actual purchases. We then test the underlying mechanism of arousal using two complementary designs, by measuring arousal (Study 3a) and by manipulating arousal (Study 3b). Study 4 tests the dual mediating effects of arousal and inferred product benefits from themed imaginative displays (i.e., with particular shapes mimicking actual objects). Finally, Study 5 tests the moderating role of congruence between display form and perceived product benefit on the core effects. Figure 1 presents the conceptual model.
Graph: Figure 1. Conceptual framework.
In the following sections, we first review the relevant literature and develop the hypotheses shown in the framework. Next, we report empirical results from six studies designed to test the hypotheses. Finally, we discuss the theoretical and practical contributions of our research, as well as its limitations and future research directions.
Considerable research examines the effects of in-store displays on customer behavior ([12]; [14]; [32]). A key stream of research examines the effects of display form on customer responses, including display direction ([17]), display formation ([50]), space-to-product ratio of the display ([63]), display organization ([12]), shelf scarcity ([52]), digital display ([60]), display completeness ([57]), and online virtual display ([10]).
Another research stream suggests that customers' evaluations of the product on display are also influenced by the display context, such as scent in the shopping environment ([22]), coordination of product grouping ([36]), surface material of the tablecloth ([72]), spatial layout of the display stand ([13]), and assortment organization ([61]). The present research extends both streams by examining the dual effects of novelty and aesthetics of imaginative displays on customers' purchase behavior. Table 1 summarizes the positioning of the present research vis-à-vis prior research on in-store displays.
Graph
Table 1. Positioning of Present Research in the In-Store Product Display Literature.
| Streams | Study | Display Variables | Relevant Findings |
|---|
| Display Context | Fiore, Yah, and Yoh (2000) | Display versus no display fragrance versus no fragrance | Appropriate fragrance enhances the effect of display on product evaluation and purchase intention. |
| Lam and Mukherjee (2005) | Coordination and juxtaposition of display | Coordination has an impact on consumers' evaluation and purchase intention toward the target item only when the item is juxtaposed with a complementary item. |
| Zhu and Meyers-Levy (2009) | Material of display table (modern vs. natural) | Surface material of display fixture has an assimilation effect for interdependent self-viewer and a contrast effect for independent self-viewer on consumer evaluation toward a neutral product. |
| Cavazza and Gabrielli (2015) | Shelf versus display stand with different shapes | Display stands have a positive impact on purchase intention only for little-known brands. |
| Sarantopoulos et al. (2019) | Assortment organization (complement based vs. substitute based) | Complement-based (vs. substitute-based) assortment organization leads to increased purchases and expenditures. |
| Display Form | Razzouk, Seitz, and Kumar (2002) | Completeness versus incompleteness of a stack | Products displayed in an incomplete stack are preferred to those in a visually complete stack. |
| Parker and Lehmann (2011) | Shelf-based scarcity (scarce vs. full stock) | Scarce products are preferred due to popularity inference than those in full stock. |
| Castro, Morales, and Nowlis (2013) | Display organization versus disorganization; product quantity | Disorganized and partially stocked shelves result in lower purchase intention of ingestible products but higher purchase intention for noningestible products. |
| Nordfält et al. (2014) | Display formation (waterfall only vs. waterfall + bin) | Customers were more likely to stop and examine products from the waterfall display if it emptied into a bin. |
| Deng et al. (2016) | Direction of display (horizontal vs. vertical) | Horizontal display increases perceived assortment variety and leads to more variety being chosen than vertical display. |
| Sevilla and Townsend (2016) | Space-to-product ratio (high vs. low) | More interstitial space improves perceptions of both product aesthetics and store prestige, which increases product valuation and purchase intention. |
| Roggeveen, Nordfält, and Grewal (2016) | Digital display (with vs. without) | Digital display increases sales in hypermarket, not in supermarket or convenience stores. |
| The present research | Product display (imaginative display vs. standard display) | Imaginative display (i.e., novel and aesthetically appealing display) increases customers' actual purchase and purchase intentions. Themed imaginative display increases affective response of arousal and cognitive response of inferred product benefits from the concept theme. |
The product display form consists of a number of elements that retailers choose and blend into a whole to achieve a particular sensory effect ([ 7]; [31]). Prior research suggests that an effective product display design should be perceived as innovative and visually appealing ([55]). We conceptualize an imaginative display in terms of the degree of deviation from the prototypical category exemplar of physical appearance and visual attractiveness of the design. Specifically, we define an imaginative display as a product display constructed using multiple units of the same product in a novel yet aesthetically appealing form. Thus, an imaginative display combines the two critical elements of novelty and aesthetics to achieve optimal visual effectiveness. The novel element is operationalized via the innovative and unusual appearance of the display compared with the prototype ([48]), while the aesthetic element reflects its ability to please the visual senses ([ 8]). Note that our definition of an imaginative display excludes other forms of product displays not constructed using multiple units of the same product (e.g., inflatable displays).
In marketing, "novelty" usually refers to a stimulus that is unfamiliar to the consumer ([29]) and reflects a comparison of the object with previous versions in the same or proximal categories ([55]). According to categorization theory, consumers' repeated exposure to seeing different products in the same category would lead them to develop a prototype consisting of the average value of the design features of that category ("central tendency"; [ 2]), which becomes representative of the product category ([69]). Product appearances that significantly deviate from (vs. resemble) the prototype are more novel or innovative ([48]). Moreover, perceived novelty of a stimulus not only can be due to its inherent characteristics but also could be contextual, such as when contrasted against other nearby elements ([34]).
Prior research has examined consumers' responses to various forms of aesthetics, such as product aesthetics ([ 9]; [11]), graphic aesthetics ([44]), web aesthetics (i.e., aesthetic formality and aesthetic appeal; [70]), and solicitation aesthetics ([68]). In extending the literature, we contrast a novel and aesthetically appealing imaginative display against a non-novel and neutral aesthetic standard display. Retailers use imaginative displays to attract and generate positive customer responses; therefore, we exclude aesthetically unappealing product displays, as merely imagining the unattractive display could negatively affect customers' self-perception and lower their willingness to pay for the product ([24]).
Our conceptualization of an imaginative display as being novel and aesthetically appealing highlights the combined effects of novelty and aesthetics. In this vein, novelty in the retail context can provide customers with a memorable consumption experience ([58]). A novel product appearance makes the product visually prominent compared with other products, which in turn affects customers' purchase decision ([15]). For instance, [60] show that the novelty of new digital displays in hypermarkets led to a sales lift of 17%, and even when the novelty wore off after five months, the sales lift remained at 3%.
Meanwhile, high aesthetic appeal can be pleasurable and has positive effects on customer responses to products and brands ([30]; [63]). For example, the aesthetic appeal of web design can increase online purchases through increased satisfaction ([70]). Similarly, aesthetically appealing packaged goods generate higher purchase intentions and lead to greater market share compared with aesthetically unappealing competitors ([56]). Thus, we hypothesize that an imaginative display heightens visual salience and attractiveness of the product ([15]; [35]), which increases purchase behavior. More formally:
- H1: An imaginative display, compared with a standard display, increases customers' purchases and purchase intentions.
Early research in psychobiology has established that the collative properties of a stimulus (e.g., novelty, complexity, incongruity) can influence pleasure through the mediating effect of arousal ([ 5]). For example, a work of art that is novel can influence arousal level and, subsequently, pleasure and interest ([ 6]). A recent study of the arrangement of a salad dish on a plate (arranged to look like a painting by Kandinsky vs. a regular dish vs. another organized in a neat but nonartistic way) reveals that the aesthetically pleasing dish enhanced diners' rating of the dish ([45]).
Similarly, environmental stimuli can influence individuals' emotional states (i.e., arousal and pleasure dimensions), which in turn influence their behavior ([43]). The arousal dimension reflects customers' feelings related to excitement and stimulation, while the pleasure dimension reflects their positive emotions such as happiness and satisfaction. [20] show that novelty as a measure of information rate is positively related to customer arousal. Other stimuli such as music, scent, and color can also influence arousal ([25]; [47]). For example, familiar music that is played at a novel pitch increases stimulation ([65]). Furthermore, aesthetic formality has a negative influence on arousal, whereas aesthetic appeal has a positive influence on arousal ([70]).
Increased arousal in turn positively influences customers' willingness to buy ([ 1]) and product preferences ([18]) due to arousal misattribution ([62]). That is, customers often misattribute the positive feelings from one stimulus to the target product they are evaluating ([18]). Accordingly, we propose that imaginative displays would lead to arousal misattribution, which in turn positively influences customers' purchase of the product. More formally:
- H2: An imaginative display, compared with a standard display, increases arousal, which in turn increases customers' purchases and purchase intentions.
In addition to influencing the affective response of arousal, an imaginative display may also influence customers' cognitive responses simultaneously ([ 7]). For instance, an irregularly sliced graphic design (i.e., novel and aesthetic) in the background of an ad can elicit greater arousal and also impart a more favorable embodied meaning than an intact curved design (i.e., less novel and less aesthetic; [44]). That is, consumers infer meaning and make judgments about the target object when exposed to visual aesthetics ([ 9]). In this vein, in-store display as a form of product design can communicate values, beliefs, and benefits to customers ([ 8]). For example, a novel (i.e., metallic fabric vs. burlap) tablecloth can have a contextual effect on consumers' inferred evaluation of the product placed on it (i.e., trendy vs. natural) ([72]).
Moreover, customer perception of an aesthetically appealing ensemble of products can transfer to the evaluation of the individual products ([36]). In particular, in-store displays designed with concept themes can potentially increase perceived value and brand equity of the display products ([42]). For example, a themed in-store display in the form of birds' wings can be used to promote products such as superfoods to evoke their symbolic meanings of lightness and spirituality ([42]). Accordingly, we further propose that a themed imaginative display can communicate embodied meanings that will transfer to the product constituting the display (i.e., inferred product benefits). A favorable inference will lead to greater cognitive elaboration about the product attributes ([40]), which enhances customers' purchase decision. Prior research shows that generating more product attribute–related thoughts in turn enhances product evaluation and purchase decision ([37]). More formally:
- H3: A themed imaginative display, compared with a standard display, increases the inference of product benefits, which in turn increases customers' purchases and purchase intentions.
The assumptions underlying H2 and H3 are that arousal elicited by an imaginative display would be misattributed and meaning inferred from the imaginative display would transfer to the product itself, which increase customers' purchases. However, the arousal-based and inference-based effects are context dependent and can have either a positive or negative outcome depending on the context ([11]; [68]; [70]). Thus, we further propose that (in)congruence between display form and perceived product benefit will moderate the effects of the dual processes on purchase intention.
Specifically, we draw on research showing that products are evaluated more favorably when they are congruent with similar cues in the environment ([ 4]). For instance, [22] show that congruence between a garment (sleepwear) on display and the appropriate environmental fragrance enhances customers' approach responses, and this effect is mediated by their pleasurable experience. Moreover, conceptual congruence between the thematic display context and the product can improve product evaluation by generating positive feelings and more product attribute-related thoughts ([37]). Thus, congruence between inferred benefits from the display form and perceived product benefit would facilitate arousal misattribution and meaning transfer to increase purchase intention.
By contrast, incongruence between inferred benefits from the display form and perceived product benefit would prevent arousal misattribution and dampen meaning transfer to the product, thus lowering purchase intention. For instance, aesthetic enhancement of donation solicitation that is incongruent with cost implications (e.g., using gold ink) can backfire and lower donations ([68]). Moreover, the affective experience of arousal could be positive or negative ([18]; Noseworthy, Di [51]). Incongruity could evoke additional arousal, and extremely high arousal is aversive and leads to negative feelings such as irritation and anxiety, which lower preference for the product ([11]; [51]). To illustrate, an imaginative display in the form of a battle tank could lead customers to feel positive arousal (i.e., pleasant feeling) and infer "strength and power" for a product positioned on a congruent benefit (e.g., energy drink), which increases purchase intention. In contrast, although the same battle tank display for a product with an incongruent benefit (e.g., relaxation drink) may also lead customers to feel aroused and infer energy from the display form, they are unrelated to the product itself, which disrupt arousal misattribution and meaning transfer, and subsequently lower purchase intention. Taken together, congruence between display form and perceived product benefit moderates the effects of an imaginative display on purchase behavior. More formally:
- H4: The effects of an imaginative display (H1) and the underlying processes (H2 and H3) on customers' purchases and purchase intentions are moderated by congruence between display form and perceived product benefit, such that the effects hold when they are congruent but are mitigated when they are incongruent.
Study 1 tests the main effect of an imaginative display on sales revenue in a grocery store in a major Australian city using a one-factor two-level (product display: imaginative vs. standard) between-subjects design. The imaginative display was designed in the form of a 17-story quasi-circular cone on a cuboid base constructed using boxes of facial tissues (retail price $1.99), while the standard display consisted of only the cuboid base (Appendix A). A pretest (between-subjects design, N = 115) on the display showed that the imaginative display was perceived to be more novel (Mimaginative = 5.70, SD =.98 vs. Mstandard = 2.30, SD = 1.57, F( 1, 133) = 195.18, p <.001, ηp2 =.63) and aesthetically appealing (Mimaginative = 5.95, SD =.83 vs. Mstandard = 3.72, SD = 1.18, F( 1, 133) = 136.53, p <.001, ηp2 =.547) than the standard display.[ 7]
The display was located near the checkout counter. Store employees restocked the facial tissues after each purchase. Customers were not aware of the research being conducted. The store manager provided information on the daily facial tissue unit sales, the store's daily revenue, and the relevant cost information. We conducted this study over a two-week period (Week 1 = standard display, Week 2 = imaginative display).
We first calculated the daily facial tissue sales revenue (M = 24.31, SD = 12.52) by multiplying the daily quantity sold (M = 12.21, SD = 6.29) with the unit price ($1.99). A simple regression revealed a significant effect of product display (1 = imaginative, 0 = standard) on daily facial tissue sales revenue (β =.58, t(12) = 2.45, p =.031), which supports H1.
In addition, we conducted a stepwise regression analysis on daily facial tissue sales revenue to capture the unique variance explained by the imaginative display by considering the potential covarying effect of daily store revenue. We first entered daily store revenue in the baseline model, and the overall model was nonsignificant (R2 =.15, F( 1, 12) = 2.11, p =.17). Thus, daily facial tissue sales revenue did not covary with daily store revenue (β =.38, t(12) = 1.45, p =.17). Next, we added product display into the model, and the overall model became significant (R2 =.56, F( 2, 11) = 6.91, p =.011). We observed a significant effect of the imaginative display on daily facial tissue sales revenue (β =.64, t(11) = 3.18, p =.009), in addition to a significant effect of daily store revenue (β =.48, t(11) = 2.35, p =.038). Thus, the explanatory power of the model was improved due to the effect of the imaginative display (ΔR2 =.41, F-change ( 1, 11) = 10.09, p =.009).
We also calculated return on investment (ROI) of the imaginative display using cost and sales information from the retailer. The unit sales lift, or difference in aggregate daily facial tissue unit sales between Week 2 (imaginative display) and Week 1 (standard display), was 49 units. The net profit per box of tissue was $.89 (i.e., total net profit = $43.61), while the extra labor cost to set up and maintain the imaginative display relative to the standard display was 1.5 hours, equivalent to $28.50. Thus, ROI for the imaginative versus standard display was (43.61 – 28.50) / 28.50 = 53.02%.
Study 1 provides field evidence supporting the positive effect of an imaginative display on sales revenue and ROI (H1). However, we acknowledge that the imaginative display in this study was taller than the standard display; thus, it is plausible that the effect could be due to the height difference. To eliminate this potential confounding effect, we conducted Study 2.
We designed Study 2 to replicate the main effect of an imaginative display on sales in a confectionery store in a large Australian city. To minimize biases from customers' preexisting brand preferences, we chose a little-known chocolate brand, Duc d'O, as the product stimulus.
Different from Study 1, Study 2 uses a one-factor three-level (product display: imaginative vs. standard–high vs. standard–low) between-subjects design. All displays were constructed using boxes of Duc d'O chocolates (Appendix A). The imaginative display was designed in the form of a quasi-cylindrical form on a cuboid base of chocolates. The standard–high display was a cuboid-shaped display on the same base, while the standard–low display had only an elevated cuboid base. The imaginative display was the same height as the standard–high display. We placed the product display near the entrance/exit of the store. This study took place over four days. We used a different display on each of the first three days and rotated the order of the three displays on the last day. Next to the display were two signs showing the price ($5) and inviting customers to take part in the survey in exchange for a $10 store voucher.
A research assistant stood near the entrance and invited each approaching customer to participate in the survey. Customers who agreed to take part received a paper questionnaire on a clipboard. Participants were first asked some questions about their store perceptions. Then they were asked to look at the product display and evaluate it in terms of novelty (M = 4.19, SD = 1.78, r =.91) and aesthetics (M = 4.99, SD = 1.33, α =.96). Participants also rated the extent to which they liked eating chocolates (M = 6.16, SD = 1.02, α =.93) and their familiarity with the Duc d'O brand (M = 2.05, SD = 1.57, α =.95). All measures were rated on seven-point scales (see Appendix B). Finally, participants indicated their gender and age.
After completing the questionnaire, participants were thanked and given a $10 store voucher that was valid only on that day and could be used toward any purchase, with no minimum spending. When they finished shopping, they redeemed the voucher at the counter. The cashier retained the voucher, stapled a duplicate receipt to it, and recorded the number of boxes of Duc d'O chocolates purchased using the voucher at the end of each session.
Across the four-day experiment, 1,416 customers purchased products in the store (386, 356, 302, and 372 for each day, respectively), among whom 250 customers (66.80% female, Mage = 39.79 years; N = 84, 83, and 83 for the imaginative, standard–high, and standard–low displays, respectively) participated in the study (17.66% participation rate).
Analysis of variance (ANOVA) results show that the three product displays differed significantly in novelty (F( 2, 247) = 35.06, p <.001, ηp2 =.22). Participants perceived the imaginative display to be more novel (M = 5.35, SD = 1.30) than the standard–high (M = 3.77, SD = 1.73, t(247) = 7.84, p <.001) and standard–low (M = 3.43, SD = 1.68, t(247) = 6.46, p <.001) displays, with no significant difference between the standard displays (p =.171). Similarly, ANOVA results showed that the three product displays differed significantly in aesthetics (F( 2, 247) = 26.85, p <.001, ηp2 =.18). The imaginative display was perceived to be more aesthetically appealing (M = 5.76, SD =.89) than the standard–high (M = 4.72, SD = 1.28, t(247) = 6.91, p <.001) and standard–low (M = 4.47, SD = 1.39, t(247) = 5.55, p <.001) displays, with no significant difference between the standard displays (t(247) = 1.36, p =.175). Thus, the three product displays were manipulated successfully. Moreover, participants across the three display conditions did not differ in their liking for chocolates (F( 2, 247) =.61, p =.55, ηp2 =.005) or brand familiarity (F( 2, 247) =.95, p =.39, ηp2 =.008).
We first coded purchases of Duc d'O chocolates using the voucher (1 = purchase, 0 = no purchase). A binary logistic regression revealed that customers in the imaginative display condition (48.81%) purchased significantly more chocolates than those in the standard–high (16.87%; B = −1.55, χ2( 1) = 17.93, p <.001) and standard–low (19.28%; B = −1.38, χ2( 1) = 15.33, p <.001) display conditions, with no significant difference between the standard display conditions (B =.16, χ2( 1) =.16, p =.69). Including liking for chocolates and brand familiarity as covariates did not change the conclusion. Liking for chocolates and brand familiarity had no significant effects on actual purchase (ps >.11). These results support H1.
Study 2 further supports the positive effect of the imaginative display on actual purchase relative to the standard displays, regardless of height. While Studies 1 and 2 provide external validity for the effect of the imaginative display (H1), the field settings precluded a controlled environment to test the proposed underlying mechanisms, for which we turned to laboratory experiments.
Studies 3a and 3b test the mediating role of arousal underlying the effect of an imaginative display (H2) using two complementary designs: Study 3a uses a measured-mediation design while Study 3b uses an experimental causal-chain design ([66]). Moreover, as visual salience of a novel display could draw customer attention, which would increase their purchase intention ([35]), and visual complexity and perceived difficulty in constructing the novel display could also heighten visual salience ([21]), we measured attention drawing, visual complexity, and perceived difficulty to elicit the unique effect of arousal in Study 3a.
While Study 2 eliminated display height as an alternative explanation, the quantity of items in the product display could also potentially influence customer perceptions and their purchase decision. Therefore, to eliminate this alternative explanation, Study 3a uses a one-factor three-level (product display: imaginative vs. standard–large quantity vs. standard–small quantity) between-subjects design. We recruited 261 participants on Amazon Mechanical Turk (MTurk) who received financial compensation. We excluded six participants who failed the attention check questions, leaving 255 responses for the analyses (47.84% female; Mage = 37.11 years, SD = 12.28).
The imaginative display was in the form of a quasi-spiral-staircase structure above a cuboid base constructed using boxed tubes of toothpaste. The standard–small and standard–large quantity displays had only the cuboid base, with more items in the standard–large display (see Appendix A). The imaginative and standard–large displays had the same number of items.
Participants were randomly assigned to one of the three product display conditions. They read a scenario about a trip to the grocery store to buy toothpaste in which they encountered a product display of a new, unspecified brand of toothpaste. They saw an image of the product display and indicated their purchase intention for the toothpaste (M = 4.13, SD = 1.61, α =.94) and level of arousal (M = 4.62, SD = 1.41, α =.92). They also rated the product display in terms of attention drawing (M = 5.86, SD = 1.21, α =.90), visual complexity (M = 3.52, SD = 1.99, r =.98), power (1 = "not at all powerful," and 7 = "extremely powerful," M = 3.97, SD = 1.44), and perceived difficulty in construction (M = 4.70, SD = 1.76, α =.94). Furthermore, they evaluated novelty (M = 4.05, SD = 2.05, r =.92), aesthetics (M = 4.87, SD = 1.43, α =.97), and perceived quantity (1 = "very scarce," and 7 = "very abundant," M = 6.47, SD =.87) of the product display. Unless otherwise specified, we used the same measurement items across all studies (Appendix B).
ANOVA results showed that the three product displays differed significantly in novelty (F( 2, 252) = 97.12. p <.001, ηp2 =.44). The imaginative display was perceived to be more novel (M = 5.96, SD = 1.09) than the standard–large quantity (M = 3.27, SD = 1.79, t(252) = 11.34, p <.001) and standard–small quantity (M = 2.95, SD = 1.65, t(252) = 12.71, p <.001) displays, with no significant difference between the standard displays (t(252) = 1.34, p =.18). Similarly, ANOVA results showed that the three product displays differed significantly in aesthetics (F( 2, 252) = 30.36, p <.001, ηp2 =.19). The imaginative display was perceived to be more aesthetically appealing (M = 5.76, SD = 1.17) than the standard–large quantity (M = 4.40, SD = 1.23, t(252) = 6.86, p <.001) and standard–small quantity (M = 4.45, SD = 1.45, t(252) = 6.65, p <.001) displays, with no significant difference between the standard displays (t(252) =.23, p =.82). Thus, we conclude that the three product displays were manipulated successfully.
ANOVA results revealed that product display manipulation did not significantly affect perceived quantity (F( 2, 252) = 1.71, p =.18, ηp2 =.013). In addition, we observed a significant main effect of product display on power (F( 2, 252) = 11.84, p <.001, ηp2 =.086). Participants perceived the imaginative display (M = 4.56, SD = 1.32) to embody more power than the standard–large (M = 3.80, SD = 1.34, t(252) = 3.57, p <.001) and standard–small (M = 3.57, SD = 1.49, t(252) = 4.66, p <.001) displays, with no significant difference between the standard displays (t(252) = 1.09, p =.278). Thus, we controlled for power in the subsequent analyses.
Analysis of covariance (ANCOVA) results revealed a significant main effect of product display on purchase intention (F( 2, 251) = 3.97, p =.02, ηp2 =.031). Planned contrasts showed that the imaginative display led to higher purchase intention (M = 4.79, SD = 1.53) compared with the standard–large quantity (M = 3.93, SD = 1.46; t(252) = 3.61, p <.001) and standard–small quantity (M = 3.69, SD = 1.66; t(252) = 4.59, p <.001) displays, with no significant difference between the standard displays (t(252) =.98, p =.33). These results further support H1. ANOVA results without power as a covariate did not change the significant effect of product display (F( 2, 252) = 11.64, p <.001, ηp2 =.084).
ANCOVA results revealed a significant main effect of product display on arousal (F( 2, 251) = 18.00, p <.001, ηp2 =.125). Specifically, the imaginative display (M = 5.43, SD = 1.17) led to greater feelings of arousal than the standard–large quantity (M = 4.07, SD = 1.20, t(252) = 6.88, p <.001) and standard–small quantity (M = 4.36, SD = 1.47, t(252) = 5.46, p <.001) displays, with no significant difference between the standard displays (t(252) = −1.39, p =.15). ANOVA results without power as the covariate did not change the significant effect of display on arousal (F( 2, 252) = 26.33, p <.001, ηp2 =.173).
Similarly, ANCOVA results showed the significant main effects of product display on attention drawing (F( 2, 251) = 7.71, p <.01, ηp2 =.058), visual complexity (F( 2, 251) = 66.76, p <.001, ηp2 =.347), and perceived difficulty (F( 2, 251) = 51.98, p <.001, ηp2 =.293). Controlling for these variables and power, ANCOVA results did not change the significant main effects of product display on purchase intention (F( 2, 248) = 3.88, p =.022, ηp2 =.03) and arousal (F( 2, 248) = 7.02, p <.001, ηp2 =.054).
Next, we conducted multicategorical mediation analyses using [28] PROCESS macro (Model 4). We constructed the bootstrapping at a 95% confidence interval (CI) with 10,000 samples with the imaginative display as the reference group, such that we had two dummy variables: D1, which compared the imaginative display with the standard–small display (imaginative display = 0, standard–small display = 1, standard–large display = 0), and D2, which compared the imaginative display with the standard–large display (imaginative display = 0, standard–small display = 0, standard–large display = 1). The model included display as the independent variable, arousal as the focal mediator, and attention drawing, visual complexity, and perceived difficulty as parallel mediators. In so doing, we sought to elicit the unique effect of arousal above and beyond these other potential effects.
Results showed only the mediation effects of arousal for the imaginative display compared with the standard–small display (D1: b = –.18, SE =.08, 95% CI = [–.355, –.050]), and for the comparison between the imaginative display and the standard–large display (D2: b = –.26, SE =.09, 95% CI = [–.468, –.085]). We observed no direct mediation effects of attention drawing (D1: b = −.005, SE =.05, 95% CI = [–.114,.101]; D2: b = –.004, SE =.04, 95% CI = [−.194,.081]), visual complexity (D1: b = −.006, SE =.18, 95% CI = [–.391,.353]; D2: b = −.006, SE =.17, 95% CI = [−.357,.323]), and perceived difficulty (D1: b =.28, SE =.16, 95% CI = [−.052,.589]; D2: b =.21, SE =.13, 95% CI = [–.042,.450]).
Moreover, we tested attention drawing and arousal as serial mediators between display and purchase intention using [28] PROCESS macro (Model 6; constructed at 95% CI with 10,000 bootstrapped samples). Results showed the significant single mediation of arousal (path 1) for the imaginative display versus standard–small display comparison (D1: b = –.31, SE =.11, 95% CI = [–.552, –.125]), and for the imaginative display versus standard–large display comparison (D2: b = –.42, SE =.12, 95% CI = [–.678, –.201]). Moreover, there were significant serial mediation effects (path 2) for D1 (b = –.06, SE =.03, 95% CI = [–.127, –.018]) and D2 (b = –.05, SE =.02, 95% CI = [–.097, –.014]). However, further analyses contrasting the two indirect effects (path 1 minus path 2) revealed that the single mediation effects of arousal were significantly stronger than the serial mediation effects in both comparisons (contrast for D1: b = –.25, SE =.11, 95% CI = [–.480, –.061]; contrast for D2: b = –.37, SE =.12, 95% CI = [−.633, −.163]). These results suggested that attention drawing could influence arousal, consistent with the literature ([ 6]). However, the effect of attention drawing on purchase behavior was fully mediated by arousal. That is, arousal alone was sufficient to explain the proposed effect after accounting for the effect of attention drawing.
Study 3a supports the mediation effect of arousal underlying the effect of the imaginative display on purchase intention, above and beyond the effects of visual complexity and perceived difficulty in constructing the display.
We conducted Study 3b to replicate the mediating effect of measured arousal found in Study 3a using a causal-chain design ([66]). If arousal is indeed the underlying mechanism, it should generate an effect similar to that of the imaginative display on purchase intention. Thus, if arousal was induced prior to showing participants the product displays, the relative advantage of the imaginative display on purchase intention against the standard display would diminish. Furthermore, we conjectured that for the imaginative display to be effective, it should be perceived to be both novel and aesthetically appealing. Thus, if one element (e.g., aesthetic appeal) was missing, then the efficacy of the imaginative display would be diminished. To test this assumption, Study 3b incorporates a new display condition that was more novel but not different in aesthetics compared with the standard display. We expected that the effect of this novel but nonaesthetic display would not differ from that of the standard display.
Study 3b uses a 3 (product display: imaginative vs. novel–nonaesthetic vs. standard) × 2 (arousal: high vs. low) between-subjects design. We recruited 279 participants (43.37% female; Mage = 36.87 years, SD = 10.73) on MTurk, who received financial compensation.
We first primed arousal following Noseworthy, Di Muro, and Murray's procedure ([51]; Study 1). Participants were randomly assigned to one of two groups of images, all drawn from the International Affective Picture System ([38]). Each group contained 18 images, and each image was displayed for six seconds. Images in the two groups were similar in pleasantness but varied in arousal. Participants indicated progressive change in their arousal level on an adjustable semantic differential scale (−50 = very relaxed, +50 = very excited). When they felt no further change in their arousal level, they clicked on the "No change" button and transitioned to a shopping scenario for toothpaste. Participants saw one of the three displays. The imaginative display had a quasi-spiral-staircase form as in Study 3a while the novel–nonaesthetic display had a pillar form with the same height as the imaginative display. The standard display was the standard-large quantity display used in Study 3a. All three displays had the same number of items in them (Appendix A). In addition, we removed the background of all images for a cleaner test. Following that, participants indicated their purchase intention for the toothpaste (M = 4.19, SD = 1.38, α =.93).
We conducted a pretest to verify the manipulations of the three displays (between-subjects design, N = 96). ANOVA results showed that the three displays differed significantly in novelty (F( 2, 93) = 15.85, p <.001, ηp2 =.254). The imaginative display was perceived to be more novel (M = 5.97, SD =.80) than the novel–nonaesthetic (M = 4.65, SD = 1.62, t(93) = 3.40, p =.001) and standard (M = 3.84, SD = 1.91, t(93) = 5.96, p <.001) displays, and the novel–nonaesthetic display was more novel than the standard display (t(93) = 2.12, p =.037). Similarly, ANOVA results showed that the three displays differed significantly in aesthetics (F( 2, 93) = 5.61, p =.005, ηp2 =.108). The imaginative display was perceived to be more aesthetically appealing (M = 5.78, SD =.94) than the novel–nonaesthetic (M = 5.01, SD = 1.40, t(93) = 2.32, p =.022) and standard (M = 4.72, SD = 1.51, t(93) = 3.26, p =.002) displays, with no significant difference between the novel–nonaesthetic and standard displays (t(93) =.89, p =.38). Thus, the three product displays were manipulated successfully.
ANOVA results showed that participants reported being more excited in the arousal condition (M = 18.80, SD = 23.42) and more relaxed in the nonarousal condition (M = −15.93, SD = 21.56; F( 1, 277) = 165.97, p <.001, ηp2 =.375). Thus, our manipulation of arousal was successful.
A 3 (display) × 2 (arousal) ANOVA on purchase intention revealed the significant main effects of product display (F( 2, 273) = 4.48, p =.012, ηp2 =.032) and arousal (F( 1, 273) = 8.38, p =.004, ηp2 =.030). Importantly, we observed a significant interaction effect of product display × arousal (F( 2, 273) = 3.09, p =.047, ηp2 =.022). Decomposing the interaction, the simple effect of display was significant in the low-arousal condition (F( 2, 273) = 7.45, p =.001, ηp2 =.052) but not in the high-arousal condition (F( 2, 273) =.08, p =.924, ηp2 =.001). Specifically, in the low-arousal condition, the imaginative display led to significantly higher purchase intention (M = 4.57, SD = 1.21) compared with the novel–nonaesthetic (M = 3.63, SD = 1.20, t(136) = 3.69, p <.001) and standard (M = 3.66, SD = 1.31, t(136) = 3.49, p =.001) displays, with no difference between the novel–nonaesthetic and standard displays (t(136) = −.10, p =.92). However, in the high-arousal condition, purchase intention did not significantly differ across the three displays (Mimaginative = 4.48, SD = 1.33 vs. Mnovel–nonaesthetic = 4.41, SD = 1.38 vs. Mstandard = 4.38, SD = 1.53, all ps >.70). These results support H2.
Viewed another way, priming high arousal increased purchase intention for the novel–nonaesthetic (Mlow-arousal = 3.63 vs. Mhigh-arousal = 4.41, F( 1, 273) = 7.87, p =.005, ηp2 =.028) and standard (Mlow-arousal = 3.66 vs. Mhigh-arousal = 4.38, F( 1, 273) = 6.55, p =.011, ηp2 =.023) displays, but not for the imaginative display (F( 1, 273) =.124, p =.725, ηp2 =.00).
Studies 3a and 3b affirm the positive effect of the imaginative displays on customers' purchase intentions (H1), as mediated by arousal (H2). While Study 3a measured arousal to establish causality between display form and arousal, Study 3b manipulated arousal to establish causality between arousal and purchase intention. Importantly, Study 3b results support our conjecture about the two critical elements of novelty and aesthetic appeal embedded in the imaginative display driving arousal. Specifically, the imaginative display led to higher purchase intention compared with the novel–nonaesthetic and standard displays, with no difference between the latter two displays. Moreover, inducing arousal enhanced purchase intentions for the novel–nonaesthetic and standard displays, similar to the effect of the imaginative display.
Having shown the affect-based arousal process underlying the effects of imaginative displays in Studies 3a and 3b (H2), we next turn to examining the cognition-based inference process (i.e., inferred product benefits; H3). We proposed that customers would infer certain meanings from themed imaginative displays, which would transfer to the product constituting the display, thus influencing customers' purchase decision. We tested the dual mechanisms of arousal (H2) and inferred product benefits (H3) simultaneously in Study 4.
While Studies 3a and 3b used unbranded products to increase generalizability and practical implications of the hypothesized effects (H1–H3), Study 4 used two actual brands, Charmin and Sorbent. A pretest (within-subject design, N = 102) showed that U.S. participants were significantly more familiar with the Charmin brand (M = 6.36, SD = 1.17) than with the Sorbent brand (M = 1.49, SD = 1.13, t(101) = 27.17, p <.001). We expected that the effect of the imaginative display on purchase intention would apply to both familiar and less familiar brands.
Study 4 used a 2 (product display: imaginative vs. standard) × 2 (brand: Charmin vs. Sorbent) between-subjects design. We recruited 256 participants on MTurk (44.92% female, Mage = 37.30 years, SD = 9.99), who received financial compensation.
The imaginative display was in the form of a bear stacked above a cuboid base constructed using individually wrapped rolls of bathroom tissue, while the standard display consisted of the elevated cuboid base of the display. Both displays had the same number of items in them (Appendix A). Participants read a scenario about a recent trip to the grocery store to buy bathroom tissue and were randomly assigned to see one of the two product displays with either Charmin or Sorbent brand. Following that, participants indicated their purchase intention (M = 4.41, SD = 1.66, α =.95), feeling of arousal (M = 4.04, SD = 1.69, α =.93), and inferred strength of the bathroom tissue (1 = "not at all strong," and 7 = "very strong;" M = 4.93, SD = 1.47).
We conducted a pretest to verify the manipulations of the two product displays (between-subjects design, N = 101). ANOVA results showed that the imaginative display was perceived to be more novel (M = 6.35, SD =.73) than the standard display (M = 2.81, SD = 1.94, F( 1, 99) = 145.31, p <.001, ηp2 =.595). Similarly, the imaginative display was perceived to be more aesthetically appealing (M = 5.88, SD = 1.05) than the standard display (M = 4.39, SD = 1.30, F( 1, 99) = 39.66, p <.001, ηp2 =.286). Thus, the two product displays were manipulated successfully.
A 2 (display) × 2 (brand) ANOVA showed a significant main effect of display (F( 1, 252) = 10.24, p =.002, ηp2 =.039), such that the imaginative display led to higher purchase intention (M = 4.73, SD = 1.58) than the standard display (vs. M = 4.09, SD = 1.68). We also observed a significant main effect of brand (F( 1, 252) = 20.08, p <.001, ηp2 =.074), such that participants had higher purchase intention for the more familiar Charmin brand (M = 4.87, SD = 1.59) than for the less familiar Sorbent brand (M = 3.98, SD = 1.62). However, the interaction effect of display × brand was nonsignificant (F( 1, 252) =.08, p =.778, ηp2 =.00).
Planned contrasts showed that the imaginative display increased purchase intention compared with the standard display for both Charmin (Mimaginative = 4.33, SD = 1.60 vs. Mstandard = 3.64, SD = 1.57; F( 1, 252) = 4.13, p =.043, ηp2 =.016) and Sorbent brands (Mimaginative = 5.16, SD = 1.45 vs. Mstandard = 4.58, SD = 1.67; F( 1, 252) = 6.26, p =.013, ηp2 =.024). Thus, the imaginative display increased purchase intention, regardless of brand familiarity.
Similarly, a 2 × 2 ANOVA revealed a significant main effect of product display (F( 1, 252) = 83.56, p <.001, ηp2 =.249), such that the imaginative display led to greater arousal (M = 4.89, SD = 1.56) than the standard display (M = 3.21, SD = 1.38). However, the interaction effect of display × brand was nonsignificant (F( 1, 252) =.83, p =.364, ηp2 =.003). Planned contrasts showed that the imaginative display led to higher arousal than the standard display for both Charmin (Mimaginative = 5.00, SD = 1.51 vs. Mstandard = 3.49, SD = 1.49; F( 1, 252) = 32.86, p <.001, ηp2 =.115) and Sorbent (Mimaginative = 4.78, SD = 1.63 vs. Mstandard = 2.95, SD = 1.22; F( 1, 252) = 52.12, p <.001, ηp2 =.171). Thus, the imaginative display led to greater arousal, regardless of brand familiarity.
A 2 × 2 ANOVA on inferred strength of the bathroom tissue showed a significant main effect of display (F( 1, 252) = 35.12, p <.001, ηp2 =.122), such that the bathroom tissue was perceived to be stronger for the imaginative display (M = 5.44, SD = 1.36) than for the standard display (M = 4.42, SD = 1.41). We observed a marginally significant interaction effect of display × brand (F( 1, 252) = 3.75, p =.053, ηp2 =.015). Planned contrasts showed that participants inferred greater strength for the bathroom tissue for the imaginative display than for the standard display for both Charmin (Mimaginative = 5.50, SD = 1.38 vs. Mstandard = 4.79, SD = 1.47; F( 1, 252) = 8.41, p =.004, ηp2 =.032) and Sorbent (Mimaginative = 5.38, SD = 1.34 vs. Mstandard = 4.07, SD = 1.26; F( 1, 252) = 30.50, p <.001, ηp2 =.108) brands. Thus, the imaginative display led to a meaning transfer of inferred product benefit (i.e., strength) from the imaginative display to the displayed product, regardless of brand familiarity.
We tested for dual mediation using PROCESS Model 4 with arousal and inferred product benefit as parallel mediators. Results showed significant dual mediation effects of arousal (b =.73, SE =.13, 95% CI = [.456, 1.028]) and inferred product benefit (b =.36, SE =.09, 95% CI = [.185,.565]) between product display and purchase intention. Including the brand as a covariate did not change the dual mediation effects of arousal (b =.70, SE =.14, 95% CI = [.453,.985]) and inferred product benefit (b =.34, SE =.09, 95% CI = [.173,.531]). These results support H2 and H3.
Study 4 supports the dual mediation effects of arousal and inferred product benefits underlying the effects of the imaginative display for two actual brands, one familiar and the other less familiar. That is, affectively, consumers felt greater arousal; cognitively, consumers inferred greater product strength from the themed imaginative display in the form of a bear; and both mechanisms increased their purchase intention.
We designed Study 5 to examine the moderating effect of congruence between display form and perceived product benefit on the core effects (H4). We hypothesized that the core effects are moderated by congruence between display form and perceived product benefit, such that congruence would enhance the effects of the imaginative display on customers' purchase intention, while incongruence would attenuate or even reverse the effects.
Study 5 used a 2 (product display: imaginative vs. standard) × 3 (product benefit: energized vs. relaxed vs. control) between-subjects design. We recruited 480 participants on Prolific (65.00% female, Mage = 33.65 years, SD = 8.06) in exchange for financial compensation.
Participants read a scenario about having a very busy period at work and going to the grocery store to buy ( 1) energy drinks to boost attention span and energy levels (energized condition), ( 2) relaxation drinks to reduce stress and calm down (relaxed condition), or ( 3) natural mineral water (control condition). In the store, they came across either an imaginative display or a standard display of a new beverage. The imaginative display was in the form of a battle tank stacked above a cuboid base constructed using cans of a beverage, while the standard display consisted of the elevated cuboid base. Both displays had the same number of items in them (Appendix A). Following that, participants indicated their purchase intention for the beverage (M = 3.32, SD = 1.64, α =.94), feeling of arousal from the display (M = 3.89, SD = 1.46, α =.90), inference of energy from the display (M = 3.86, SD = 1.37, α =.91), and inference of relaxation from the display (M = 2.93, SD = 1.40, α =.96) (Appendix B).
We conducted a pretest to verify the manipulations of the two display stimuli (between-subjects design, N = 99). ANOVA results showed that the imaginative display was perceived to be more novel (M = 6.12, SD = 1.24) compared with the standard display (M = 3.16, SD = 1.93; F( 1, 97) = 82.30; p <.001, ηp2 =.46). Similarly, the imaginative display was perceived to be more aesthetically appealing (M = 5.45, SD = 1.50) than the standard display (M = 4.74, SD = 1.17; F( 1, 97) = 6.74, p =.011; ηp2 =.065).
Moreover, participants inferred the imaginative display to have greater energy benefit (Mimaginative = 5.61, SD = 1.35 vs. Mstandard = 4.01, SD = 1.74; F( 1, 97) = 26.01, p <.001, ηp2 =.212), but lower relaxation benefit compared with the standard display (Mimaginative = 2.39, SD = 1.63 vs. Mstandard = 3.23, SD = 1.62; F( 1, 97) = 6.52, p =.012, ηp2 =.063). Thus, the product displays and inferred benefits were manipulated successfully.
We also conducted a pretest to verify the perceived product benefits (between-subjects design, N = 111). Participants were asked to think about an energy drink (vs. relaxation drink) and indicate the extent to which the drink would make them energized (1 = "not at all energized," and 7 = "extremely energized;" M = 4.24, SD = 1.67) and relaxed (1 = "not at all relaxed," and 7 = "extremely relaxed;" M = 4.52, SD = 1.99) after consuming the beverage.
ANOVA results showed that the energy drink led participants to perceive feeling more energized (M = 4.98, SD = 1.16) compared with the relaxation drink (M = 3.43, SD = 1.78; F( 1, 109) = 29.92, p <.001, ηp2 =.215). Conversely, the relaxation drink led participants to perceive feeling more relaxed (M = 5.81, SD = 1.27) than they did with the energy drink (M = 3.34, SD = 1.80; F( 1, 109) = 68.21, p <.001, ηp2 =.385). Thus, product benefits were manipulated successfully.
A 2 × 3 ANOVA on purchase intention revealed a significant interaction effect of display × product benefit (F( 2, 474) = 16.44, p <.001, ηp2 =.065). We observed a main effect of product benefit (F( 2, 474) = 19.44, p <.001, ηp2 =.076). The effect of product display was nonsignificant (F( 1, 474) =.53, p =.47, ηp2 =.001). Decomposing the interaction (Figure 2), planned contrasts showed that the imaginative display increased purchase intention for the energy drink compared with the standard display (Mimaginative = 4.33, SD = 1.75 vs. Mstandard = 3.39, SD = 1.58; F( 1, 474) = 14.38, p <.001, ηp2 =.029). In contrast, the imaginative display lowered purchase intention for the relaxation drink compared with the standard display (Mimaginative = 2.26, SD = 1.29 vs. Mstandard = 3.31, SD = 1.60; F( 1, 474) = 18.48, p <.001, ηp2 =.038). The effect of display was nonsignificant for the natural mineral water (p =.43).
Graph: Figure 2. Interaction effects of display form and perceived product benefit (Study 5).Notes: Error bars = ±1 SE. ***p <.001.
Viewed another way, relative to the control condition (i.e., mineral water), the imaginative display (F( 2, 474) = 35.80, p <.001, ηp2 =.131) increased purchase intention for the energy drink (Menergy = 4.33 vs. Mmineral = 3.26, t(237) = 4.22, p <.001) but lowered purchase intention for the relaxation drink (Mrelaxation = 2.26 vs. Mmineral = 3.26, t(237) = −4.44, p <.001). For the standard display, participants' purchase intentions did not significantly differ across the three product conditions (all ps >.56). Thus, compared with the standard display, the imaginative display in the form of a battle tank increased purchase intention when it was congruent with the product benefit (i.e., energy) but decreased purchase intention when it was incongruent with the product benefit (i.e., relaxation), in support of H4.
A 2 × 3 ANOVA on arousal revealed a significant effect of product display (F( 1, 474) = 107.99, p <.001, ηp2 =.186), such that the imaginative display evoked greater arousal (M = 4.51, SD = 1.32) compared with the standard display (M = 3.26, SD = 1.30). We observed a significant effect of product benefit (F( 2, 474) = 4.44, p =.012, ηp2 =.018) but a nonsignificant interaction of display × product benefit (F( 2, 474) =.68, p =.51, ηp2 =.003).
A 2 × 3 ANOVA on inferred energy revealed a significant effect of product display (F( 1, 474) = 18.71, p <.001, ηp2 =.038), such that participants inferred greater energy benefit from the imaginative display (M = 4.12, SD = 1.34) than from the standard display (M = 3.59, SD = 1.35). We observed a significant effect of product benefit (F( 2, 474) = 11.04, p <.001, ηp2 =.045) but a nonsignificant interaction effect of display × product benefit (F( 2, 474) =.33, p =.72, ηp2 =.001).
To test the moderating effects of congruence between the dual mechanisms and purchase intention (H4), we conducted a moderated mediation analysis using PROCESS Model 15 ([28]). We specified product benefit (i.e., the moderator) as a multicategorical moderator with the mineral water as the reference group, which resulted in two dummy variables: D1 compared the relaxation drink with the control condition (mineral water = 0, relaxation drink = 1, energy drink = 0), while D2 compared the energy drink with the control condition (mineral water = 0, relaxation drink = 0, energy drink = 1). Moreover, we included inferred relaxation benefit as an alternative explanation for the reversed effect of the display on the lower purchase intention for the relaxation drink.
As expected, the moderating effects of perceived product benefit on purchase intention were qualified by the three significant interaction effects of display × product benefit (F( 2, 465) = 4.79, p =.009), arousal × product benefit (F( 2, 465) = 3.44, p =.033), and inferred energy × product benefit (F( 2, 465) = 3.08, p =.046) but a nonsignificant interaction of inferred relaxation × product benefit (F( 2, 465) =.87, p =.419). Specifically, we observed a significant interaction of display × D2 (b =.69, SE =.31, t = 2.18, p =.03) but a nonsignificant interaction of display × D1 (b = –.28, SE =.32, t = –.87, p =.38) on purchase intention, suggesting that product benefit moderated the effect of display form on purchase intention. Moreover, we observed a significant interaction of inferred energy × D2 (b =.33, SE =.13, t = 2.46, p =.014) but a nonsignificant interaction of inferred energy × D1 (b =.19, SE =.11, t = 1.62, p =.11) on purchase intention, suggesting that product benefit moderated the effect of inferred energy from the display on purchase intention. Conversely, we observed a significant interaction of arousal × D1 (b = −.29, SE =.11, t = −2.47, p =.014) but a nonsignificant interaction of arousal × D2 (b = −.08, SE =.12, t = −.63, p =.52) on purchase intention, suggesting that product benefit moderated the effect of arousal from the display on purchase intention.
Importantly, bootstrapping results showed a significant moderated mediation effect via inference of energy for the energy drink (D2: index =.17, SE =.09, 95% CI = [.018,.366]) but not for the relaxation drink (D1: index =.10, SE =.08, 95% CI = [–.038,.271]). Conversely, the data showed a significant moderated mediation effect via arousal for the relaxation drink (D1: index = −.36, SE =.17, 95% CI = [−.701, −.050]) but not for the energy drink (D2: index = −.10, SE =.18, 95% CI = [−.465,.260]),[ 8] suggesting an aversive effect of arousal for the incongruent product. We observed nonsignificant moderated mediation effects through inference of relaxation for both D1 (95% CI = [−.197,.113]) and D2 (95% CI = [–.116,.253]). These results supported the second-stage moderated mediation effect as proposed (H4).
Study 5 supports the moderating effects of congruence between display form and perceived product benefit on the main effect of the imaginative display on purchase intention, and on the effects of arousal and inferred product benefits on purchase intention (H4). Compared with a standard display, the themed imaginative display (i.e., battle tank) increased purchase intention for a congruent product (i.e., energy drink) due to the positive effect of arousal and inferred product benefit. Conversely, the same imaginative display lowered purchase intention for an incongruent product (i.e., relaxation drink) due to the aversive effect of arousal.
The present research extends the literature on in-store display form and context by revealing the favorable effects of imaginative displays on customers' purchase behavior (field experiments in Studies 1 and 2). Importantly, we show that this effect can be explained by the dual mechanisms of affect-based arousal (Studies 3a−5) and cognition-based inferred benefits from imaginative displays (Studies 4 and 5). Moreover, we identify congruence between display form and perceived product benefit as a moderator on the main and mediating (i.e., arousal and inferred benefits) effects (Study 5). These findings were obtained using varying forms of imaginative display modeled after actual imaginative displays (Appendix A). The product categories encompassed both utilitarian (i.e., facial tissue, toothpaste, bathroom tissue, and beverage) and hedonic (i.e., chocolates) products for familiar, less familiar, and unspecified brands. In addition, the samples included both actual shoppers (Studies 1 and 2) and online participants (Studies 3a–5) from Australia, the United States, and the United Kingdom, which attested to the robustness of our findings. Taken together, these findings contribute to the in-store display and store atmospherics literature; in addition, they have important managerial implications.
In extending the in-store display and store atmospherics literature, the present research examines the effects of imaginative displays on customers' purchase behavior. Specifically, such imaginative displays pertain to the domain of display form, while inferred benefits embodied by imaginative displays pertain to the domain of display context. In particular, we reveal that imaginative displays must be both novel and aesthetically appealing. While the use of imaginative displays may appear gimmicky, they can positively influence customers' purchase behavior, product sales, and ROI at relatively low costs.
Our findings also extend research on the ensemble effect, which suggests that customers' attitudes toward an ensemble of complementary products can influence their evaluation of the individual product ([36]). We reveal that an imaginative display consisting of multiple units of one product can also increase customers' purchase behavior. By examining the joint effects of novelty and aesthetics, we extend the current literature, which tends to focus on novelty alone, contributing new insights to the product display literature.
Prior research on store atmospherics has examined the effects of environmental factors such as music, scent, and color on customer arousal, which in turn enhances their purchase decision ([22]; [25]; [47]). We show that customer arousal can also stem from viewing novel and aesthetically appealing imaginative displays. Besides arousal, we reveal a cognition-based process, whereby themed imaginative displays (i.e., with particular shapes mimicking actual objects such as a bear and a battle tank) convey embodied meanings (e.g., strength and energy) that transfer to the products constituting the display, which increase customers' purchase intention. The dual mechanisms of arousal and inferred product benefits underlying imaginative displays empirically support [ 7] conceptual model of product form. Moreover, we identify the moderating factor of congruence between display form and perceived product benefit, such that congruence will increase customers' purchase intention, while incongruence will lower their purchase intention. We show that arousal has a positive (negative) effect on purchase intention when display form and product benefit are congruent (incongruent), yielding new insights on the polarizing effects of arousal.
In general, retailers benefit from having in-store displays, which can generate unplanned purchases for frequently purchased product categories ([32]). Although retailers are increasingly using imaginative displays in their stores (see examples in the Web Appendix), to our knowledge, prior research has not systematically examined their effects on customers' purchase behavior and store sales. To this end, our two field experiments show that imaginative displays increase product sales while also providing positive ROI for the display. This is borne out by industry practice; for instance, Coke Zero's novel inverted pyramid display increased sales by 13% at select supermarkets implementing the display (see the Web Appendix).
Importantly, we reveal that efficacy of the imaginative display is determined jointly by its novelty and aesthetic elements, rather than by its height (Study 2) or the quantity of products in the display (Study 3a). Managerially, an imaginative display offers a cost-effective way to increase sales and ROI compared with a standard display (Study 1). Our conversations with several retailers revealed that some ideas for their imaginative displays came from employees. Thus, it would be beneficial to solicit ideas for imaginative display from employees, who in turn might feel pride when their creations are on display. Obviously, this does not preclude retailers from engaging the services of design professionals. For example, the nonprofit organization Canstruction (canstruction.org) regularly holds exhibitions and competitions of canned food imaginative displays, whereby teams of volunteers, youth groups, and/or Canstruction contractors compete and construct some rather amazing structures, at the end of which all food is donated to local food banks ([26]).
We find two mechanisms underlying the effect of the imaginative display: arousal and inferred product benefits. Potentially, the effect of the imaginative display on arousal can be complemented by other contextual stimuli such as congruent music, color, and scent in the store ([22]; [25]; [47]). Moreover, retailers should ensure congruence between inferred benefits from the imaginative display form and perceived product benefit. For example, our imaginative display of a battle tank that embodies strength leads to greater purchase intention for an energy drink that has a congruent benefit but lowers purchase intention for a relaxation drink that has an incongruent benefit.
Notwithstanding the new insights from the current research findings, we acknowledge several limitations that provide opportunities for future research. First, we note that the product stimuli used in all six studies represent low-involvement packaged goods (i.e., chocolates, bathroom tissue, and toothpaste). It is possible that product involvement could moderate the effect of the imaginative display. For example, would an expensive wine or perfume gain more sales if an imaginative display was utilized? Presumably, a customer seeking to buy a specific fine wine may be less influenced by the imaginative display, but another customer who is uncertain of which wine to purchase may well be persuaded by the display. Second, Study 3b used an incidental technique to manipulate arousal (i.e., external stimulation) rather than task-related arousal (i.e., due to the imaginative display). Prior research suggests that incidental affect and task-related affect could have differential effects ([23]). Thus, future research could determine whether nuanced differences exist between task-related arousal and incidental arousal for imaginative displays.
Moreover, it is not clear if our findings would apply to fresh or perishable items such as seafood and vegetables. The product contamination literature ([46]) suggests that some customers may not take well to fresh food items that have been handled by others, particularly if the imaginative display is intricate and takes considerable time to construct. Perceptions of contamination and concerns about product hygiene may lead to undesirable effects ([12]). This conjecture awaits further research.
Finally, while we examined imaginative displays that are novel and aesthetically appealing, we recognize that a novel stimulus may also be aesthetically unappealing ([44]; [69]). While [24] suggest that consumers tend to avoid unattractive produce due to altered self-perceptions, they do make some exceptions; for example, ironically, some consumers are embracing "ugly" Crocs footwear ([71]). This effect may be moderated by customers' need for uniqueness ([64]). This possibility merits further examination.
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921997359 - Gimmicky or Effective? The Effects of Imaginative Displays on Customers' Purchase Behavior
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921997359 for Gimmicky or Effective? The Effects of Imaginative Displays on Customers' Purchase Behavior by Hean Tat Keh, Di Wang and Li Yan in Journal of Marketing
Images for Studies 1 and 2 (Figures A1 and A2) were untouched photographs of stimuli used in the field experiments. Images for Studies 3a–5 (Figures A3–A6) were rendered using Autodesk 3ds MAX, a computer graphics program. All novel product displays were modeled after actual displays used by retailers.
Graph: Figure A1. Study 1.
Graph: Figure A2. Study 2.
Graph: Figure A3. Study 3a.
Graph: Figure A4. Study 3b.
Graph: Figure A5. Study 4.
Graph: Figure A6. Study 5.
Graph
| Construct | Measurement Items (Seven-point scales) | Reliability (α or r) |
|---|
| Purchase intention (MacKenzie, Lutz, and Belch 1986) | How likely are you to buy the product on display?Not at all likely/very likely | .94 (S3a),.93 (S3b),.95 (S4),.94 (S5) |
| Not at all probable/very probable |
| Not at all possible/very possible |
| Arousal (Baker, Levy, and Grewal 1992)a | How did you feel when looking at the display?Alive, inactive, drowsy, idle, lazy, slow | .92 (S3a) |
| Arousal (Kim and Lakshmanan 2015) | How did you feel when looking at the display?Mellow/fired upLow energy/high energyPassive/active | .93 (S4),.90 (S5) |
| Novelty (Dahl and Moreau 2002) | How would you evaluate the display you saw?Not at all innovative/extremely innovative | .88 (S1),.91 (S2),.92 (S3a),.81 (S3b),.87 (S4),.88 (S5) |
| Not at all original/extremely original |
| Aesthetics (Lam and Mukherjee 2005) | How would you evaluate the display you saw?Very offensive/very enjoyable | .97 (S1),.96 (S2),.97 (S3a),.96 (S3b),.96 (S4),.98 (S5) |
| Very poor looking/very nice looking |
| Very displeasing/very pleasing |
| Very unattractive/very attractive |
| Very bad appearance/very good appearance |
| Very ugly/very beautiful |
| Brand familiarity (Batra et al. 2000) | How familiar are you with the brand?Not at all familiar/extremely familiar | .95 (S2) |
| Never heard of it before/hear about it very often |
| Never bought it before/buy it very often |
| Never used it before/use it very often |
| Liking of chocolates (Lee, Keller, Sternthal 2010) | How much do you like eating chocolates?Do not like at all/like very muchNot at all enjoyable/extremely enjoyableNot at all pleasurable/extremely pleasurable | .93 (S2) |
| Attention drawing (Lam and Mukherjee 2005) | How would you evaluate the display you saw?Very inconspicuous/very conspicuousNot at all eye catching/extremely eye catchingNot at all noticeable/extremely noticeableNot at all attention drawing/extremely attention drawing | .90 (S3a) |
| Visual complexity (Pieters, Wedel, and Batra 2010) | How would you evaluate the display you saw?Not at all complex/extremely complexNot at all complicated/extremely complicated | .98 (S3a) |
| Perceived difficulty (Diehl, van Herpen, and Lamberton 2015) | How would you evaluate setting up the display you saw?Not at all difficult/extremely difficultVery little time/a lot of timeVery little effort/a lot of effortVery little planning/a lot of planning | .94 (S3a) |
| Inference of Energy (adapted from Sundar and Noseworthy 2010)a | To what extent do you think the drink on display can help you:Improve power; increase endurance; stay active | .91 (S5) |
| Inference of Relaxation (adapted from Havlena and Holbrook 1986)a | To what extent do you think the drink on display can help you:Relax; de-stress; calm down | .96 (S5) |
1 a1 = not at all, 7 = very much.
Footnotes 1 Hean Tat Keh and Di Wang contributed equally.
2 Dhruv Grewal
3 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
4 The author(s) received no financial support for the research, authorship, and/or publication of this article.
5 Hean Tat Keh https://orcid.org/0000-0002-3328-5669 Di Wang https://orcid.org/0000-0002-2116-4018 Li Yan https://orcid.org/0000-0002-8392-7009
6 Online supplement: https://doi.org/10.1177/0022242921997359
7 In all studies, novelty and aesthetics were highly correlated:.82 (Study 1),.73 (Study 2),.67 (Study 3a),.69 (Study 3b),.63 (Study 4), and.52 (Study 5). In addition, factor analysis showed that the two constructs loaded onto the same factor. Manipulation checks on the composite index of the two constructs did not significantly differ from using them separately. We report the reliabilities of each construct in Appendix B.
8 The conditional effect of arousal on purchase intention was significant for the energy drink (b =.19, p =.024).
References Baker Julie , Levy Michael , Grewal Dhruv. (1992), " An Experimental Approach to Making Retail Store Environmental Decisions ," Journal of Retailing , 68 (4), 445 – 60.
Barsalou Lawrence W.. (1985), " Ideals, Central Tendency, and Frequency of Instantiation as Determinants of Graded Structure in Categories ," Journal of Experimental Psychology: Learning, Memory, and Cognition , 11 (4), 629 – 54.
Batra Rajeev , Ramaswamy Venkatram , Alden Dana L. , Steenkamp Jan-Benedict E. M. , Ramachander S.. (2000), " Effects of Brand Local and Nonlocal Origin on Consumer Attitudes in Developing Countries ," Journal of Consumer Psychology , 9 (2), 83 – 95.
Berger Jonah , Fitzsimons Gráinne. (2008), " Dogs on the Street, Pumas on Your Feet: How Cues in the Environment Influence Product Evaluation and Choice ," Journal of Marketing Research , 45 (1), 1 – 14.
Berlyne Daniel E.. (1950), " Novelty and Curiosity as Determinants of Exploratory Behavior ," British Journal of Psychology , 41 (1–2), 68 – 80.
Berlyne Daniel E.. (1974), Studies in the New Experimental Aesthetics. Washington, DC : Hemisphere Publishing.
Bloch Peter H.. (1995), " Seeking the Ideal Form: Product Design and Consumer Response ," Journal of Marketing , 59 (3), 16 – 29.
Bloch Peter H.. (2011), " Product Design and Marketing: Reflections after Fifteen Years ," Journal of Product Innovation Management , 28 (3), 378 – 80.
9 Bloch Peter H. , Brunel Frederic F. , Arnold Todd J.. (2003), " Individual Differences in the Centrality of Visual Product Aesthetics: Concept and Measurement ," Journal of Consumer Research , 29 (4), 551 – 65.
Breugelmans Els , Campo Katia. (2011), " Effectiveness of In-Store Displays in a Virtual Store Environment ," Journal of Retailing , 87 (1), 75 – 89.
Buechel Eva C. , Townsend Claudia. (2018), " Buying Beauty for the Long Run: (Mis)predicting Liking of Product Aesthetics ," Journal of Consumer Research , 45 (2), 275 – 97.
Castro Iana A. , Morales Andrea C. , Nowlis Stephen M.. (2013), " The Influence of Disorganized Shelf Displays and Limited Product Quantity on Consumer Purchase ," Journal of Marketing , 77 (4), 118 – 33.
Cavazza Nicoletta , Gabrielli Veronica. (2015), " Affordant Shapes of Product Holder Influence Product Evaluation and Purchase Intention ," Current Psychology , 34 (2), 447 – 65.
Chandon Pierre , Wesley Hutchinson J. , Bradlow Eric T. , Young Scott H.. (2009), " Does In-Store Marketing Work? Effects of the Number and Position of Shelf Facings on Brand Attention and Evaluation at the Point of Purchase ," Journal of Marketing , 73 (6), 1 – 17.
Creusen Mariëlle E. H. , Schoormans Jan P. L.. (2005), " The Different Roles of Product Appearance in Consumer Choice ," Journal of Product Innovation Management , 22 (1), 63 – 81.
Dahl Darren W. , Moreau Page. (2002), " The Influence and Value of Analogical Thinking During New Product Ideation ," Journal of Marketing Research , 39 (1), 47 – 60.
Deng Xiaoyan , Kahn Barbara E. , Rao Unnava H. , Lee Hyojin. (2016), " A 'Wide' Variety: Effects of Horizontal Versus Vertical Display on Assortment Processing, Perceived Variety, and Choice ," Journal of Marketing Research , 53 (5), 682 – 98.
Di Muro Fabrizio , Murray Kyle B.. (2012), " An Arousal Regulation Explanation of Mood Effects on Consumer Choice ," Journal of Consumer Research , 39 (3), 574 – 84.
Diehl Kristin , Herpen Erica van , Lamberton Cait. (2015), " Organizing Products with Complements versus Substitutes: Effects on Store Preferences as a Function of Effort and Assortment Perceptions ," Journal of Retailing , 91 (1), 1 – 18.
Donovan Robert J. , Rossiter John R.. (1982), " Store Atmosphere: An Environmental Psychology Approach ," Journal of Retailing , 58 (1), 34 – 57.
Elazary Lior , Itti Laurent. (2008), " Interesting Objects Are Visually Salient ," Journal of Vision , 8 (3), 1 – 15.
Fiore Ann Marie , Yah Xinlu , Yoh Eunah. (2000), " Effects of a Product Display and Environmental Fragrancing on Approach Responses and Pleasurable Experiences ," Psychology & Marketing , 17 (1), 27 – 54.
Garg Nitika , Jeffrey Inman J. , Mittal Vikas. (2005), " Incidental and Task-Related Affect: A Re-Inquiry and Extension of the Influence of Affect on Choice ," Journal of Consumer Research , 32 (1), 154 – 59.
Grewal Lauren , Hmurovic Jillian , Lamberton Cait , Reczek Rebecca Walker. (2019), " The Self-Perception Connection: Why Consumers Devalue Unattractive Produce ," Journal of Marketing , 83 (1), 89 – 107.
Hagtvedt Henrik , Adam Brasel S.. (2017), " Color Saturation Increases Perceived Product Size ," Journal of Consumer Research , 44 (2), 396 – 413.
Hammer Monika. (2016), " 'Canstruction' on Display at Oakdale Mall ," (accessed May 9, 2017), [ http://www.wbng.com/news/local/Canstruction-on-display-at-Oakdale-Mall-374534331.html ].
Havlena William J. , Holbrook Morris B.. (1986), " The Varieties of Consumption Experience: Comparing Two Typologies of Emotion in Consumer Behavior ," Journal of Consumer Research , 13 (3), 394 – 404.
Hayes Andrew F.. (2018), Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach , 2nd ed. New York : The Guilford Press.
Hirschman Elizabeth C.. (1980), " Innovativeness, Novelty Seeking, and Consumer Creativity ," Journal of Consumer Research , 7 (3), 283 – 95.
Hoegg JoAndrea , Alba Joseph W. , Dahl Darren W.. (2010), " The Good, the Bad, and the Ugly: Influence of Aesthetics on Product Feature Judgments ," Journal of Consumer Psychology , 20 (4), 419 – 30.
Hollins Bill , Pugh Stuart. (1990), Successful Product Design: What to Do and When. London : Butterworth-Heinemann.
Inman J. Jeffrey , Winer Russell S. , Ferraro Rosellina. (2009), " The Interplay Among Category Characteristics, Customer Characteristics, and Customer Activities on In-Store Decision Making ," Journal of Marketing , 73 (5), 19 – 29.
Kaltcheva Velitchka D. , Weitz Barton A.. (2006), " When Should a Retailer Create an Exciting Store Environment? " Journal of Marketing , 70 (1), 107 – 18.
Kim Junghan , Lakshmanan Arun. (2015), " How Kinetic Property Shapes Novelty Perceptions ," Journal of Marketing , 79 (6), 94 – 111.
Krishna Aradhna , Cian Luca , Aydınoğlu Nilüfer Z.. (2017), " Sensory Aspects of Package Design ," Journal of Retailing , 93 (1), 43 – 54.
Lam Shun Yin , Mukherjee Avinandan. (2005), " The Effects of Merchandise Coordination and Juxtaposition on Consumers' Product Evaluation and Purchase Intention in Store-Based Retailing ," Journal of Retailing , 81 (3), 231 – 50.
Lam Shun Yin , Ho-ying Fu Jeanne , Li Dongmei. (2017), " The Influence of Thematic Product Displays on Consumers: An Elaboration-Based Account ," Psychology & Marketing , 34 (9), 868 – 83.
Lang Peter J. , Bradley Margaret M. , Cuthbert Bruce N.. (2008), " International Affective Picture System (IAPS): Affective Ratings of Pictures and Instruction Manual ," Technical Report A-8 , University of Florida.
Lee Angela Y. , Keller Punam Anand , Sternthal Brian. (2010), " Value from Regulatory Construal Fit: The Persuasive Impact of Fit between Consumer Goals and Message Concreteness ," Journal of Consumer Research , 36 (5), 735 – 47.
Lundh Lars-Gunnar. (1995), " Meaning Structures and Mental Representations ," Scandinavian Journal of Psychology , 36 (4), 363 – 85.
MacKenzie Scott B. , Lutz Richard J. , Belch George E.. (1986), " The Role of Attitude Toward the Ad as a Mediator of Advertising Effectiveness: A Test of Competing Explanations ," Journal of Marketing Research , 23 (2), 130 – 43.
Manavis Athanasios , Sourris Theocharis , Dimou Eva , Efkolidis Nikolaos , Kyratsis Panagiotis. (2019), " An Inspiration from Nature Design Methodology for In-Store Displays ," Journal of Packaging Technology and Research , 3 (2), 141 – 48.
Mehrabian Albert , Russell James A.. (1974), An Approach to Environmental Psychology. Cambridge, MA : MIT Press.
Meyers-Levy Joan , Zhu Rui (Juliet). (2010), " Gender Differences in the Meanings Consumers Infer from Music and Other Aesthetic Stimuli ," Journal of Consumer Psychology , 20 (4), 495 – 507.
Michel Charles , Velasco Carlos , Gatti Elia , Spence Charles. (2014), " A Taste of Kandinsky: Assessing the Influence of the Artistic Visual Presentation of Food on the Dining Experience ," Flavour , 3 (1), 7 – 17.
Morales Andrea C. , Fitzsimons Gavan J.. (2007), " Product Contagion: Changing Consumer Evaluations Through Physical Contact with "Disgusting" Products ," Journal of Marketing Research , 44 (2), 272 – 83.
Morrison Michael , Gan Sarah , Dubelaar Chris , Oppewal Harmen. (2011), " In-Store Music and Aroma Influences on Shopper Behavior and Satisfaction ," Journal of Business Research , 64 (6), 558 – 64.
Mugge Ruth , Schoormans Jan P. L.. (2012), " Product Design and Apparent Usability: The Influence of Novelty in Product Appearance ," Applied Ergonomics , 43 (6), 1081 – 88.
Neff Jack. (2008), " In-store Displays Are More Effective than Price Cuts ," (accessed August 20, 2015), [ http://adage.com/article/news/store-displays-effective-price-cuts/132767/ ].
Nordfält Jens , Grewal Dhruv , Roggeveen Anne L. , Hill Krista M.. (2014), " Insights from In-Store Marketing Experiments ," in Review of Marketing Research: Shopper Marketing and the Role of In-Store Marketing , Vol. 11 , Grewal Dhruv , Roggeveen Anne L. , Nordfält Jens , eds. Bingley, UK : Emerald , 127 – 46.
Noseworthy Theodore J. , Muro Fabrizio Di , Murray Kyle B.. (2014), " The Role of Arousal in Congruity-Based Product Evaluation ," Journal of Consumer Research , 41 (4), 1108 – 26.
Parker Jeffrey R. , Lehmann Donald R.. (2011), " When Shelf-Based Scarcity Impacts Consumer Preferences ," Journal of Retailing , 87 (2), 142 – 55.
Pieters Rik , Wedel Michel , Batra Rajeev. (2010), " The Stopping Power of Advertising: Measures and Effects of Visual Complexity ," Journal of Marketing , 74 (5), 48 – 60.
POPAI (2012), " 2012 Shopper Engagement Study ," (accessed May 2, 2014), [ http://www.advancingretail.org/sites/default/files/resources/POPAI%202012%20Shopper%20Engagement%20Study.pdf ].
Radford Scott K. , Bloch Peter H.. (2011), " Linking Innovation to Design: Consumer Responses to Visual Product Newness ," Journal of Product Innovation Management , 28 (s1), 208 – 20.
Raghubir Priya , Greenleaf Eric A.. (2006), " Ratios in Proportion: What Should the Shape of the Package Be? " Journal of Marketing , 70 (2), 95 – 107.
Razzouk Nabil Y. , Seitz Victoria , Kumar Vijay. (2002), " The Impact of Perceived Display Completeness/Incompleteness on Shoppers' In-Store Selection of Merchandise: An Empirical Study ," Journal of Retailing and Consumer Services , 9 (1), 31 – 35.
Rego Arménio , Júnior Dálcio Reis , Cunha Miguel Pina e , Stallbaum Gabriel , Neves Pedro. (2014), " Store Creativity Mediating the Relationship Between Affective Tone and Performance ," Managing Service Quality , 24 (1), 63 – 85.
Roggeveen Anne L. , Grewal Dhruv , Schweiger Elisa B.. (2020), " The DAST Framework for Retail Atmospherics: The Impact of In- and Out-of-Store Retail Journey Touchpoints on the Customer Experience ," Journal of Retailing , 96 (1), 128 – 37.
Roggeveen Anne L. , Nordfält Jens , Grewal Dhruv. (2016), " Do Digital Displays Enhance Sales? Role of Retail Format and Message Content ," Journal of Retailing , 92 (1), 122 – 31.
Sarantopoulos Panagiotis , Theotokis Aristeidis , Pramatari Katerina , Roggeveen Anne L.. (2019), " The Impact of a Complement-Based Assortment Organization on Purchases ," Journal of Marketing Research , 56 (3), 459 – 78.
Schachter Stanley , Singer Jerome. (1962), " Cognitive, Social, and Physiological Determinants of Emotional State ," Psychological Review , 69 (5), 379 – 99.
Sevilla Julio , Townsend Claudia. (2016), " The Space-to-Product Ratio Effect: How Interstitial Space Influences Product Aesthetic Appeal, Store Perceptions and Product Preference ," Journal of Marketing Research , 53 (5), 665 – 81.
Simonson Itamar , Nowlis Stephen M.. (2000), " The Role of Explanations and Need for Uniqueness in Consumer Decision Making: Unconventional Choices Based on Reasons ," Journal of Consumer Research , 27 (1), 49 – 68.
Simonton Dean Keith. (2010), " Emotion and Composition in Classical Music: Historiometric Perspectives ," in Handbook of Music and Emotion: Theory, Research, Applications , Juslin Patrik N. , Sloboda John A. , eds. Oxford, UK : Oxford University Press , 205 – 22.
Spencer Steven J. , Zanna Mark P. , Fong Geoffrey T.. (2005), " Establishing a Causal Chain: Why Experiments Are Often More Effective than Mediational Analyses in Examining Psychological Processes ," Journal of Personality and Social Psychology , 89 (6), 845 – 51.
Sundar Aparna , Noseworthy Theodore J.. (2014), " Place the Logo High or Low? Using Conceptual Metaphors of Power in Packaging Design ," Journal of Marketing , 78 (5), 138 – 51.
Townsend Claudia. (2017), " The Price of Beauty: Differential Effects of Design Elements with and without Cost Implications in Nonprofit Donor Solicitations ," Journal of Consumer Research , 44 (4), 794 – 815.
Veryzer Robert W. Jr , Wesley Hutchinson J.. (1998), " The Influence of Unity and Prototypicality on Aesthetic Responses to New Product Designs ," Journal of Consumer Research , 24 (4), 374 – 85.
Wang Yong Jian , Minor Michael S. , Wei Jie. (2011), " Aesthetics and the Online Shopping Environment: Understanding Consumer Responses ," Journal of Retailing , 87 (1), 46 – 58.
Wu Jasmine. (2019), " Ugly Is In: How Crocs have Taken over Teen Footwear, and Sent the Stock Soaring ," (accessed January 8, 2021), [ https://www.cnbc.com/2019/07/15/ugly-is-in-crocs-have-taken-over-teen-footwear-and-boosted-its-stock.html ].
Zhu Rui , Meyers-Levy Joan. (2009), " The Influence of Self-View on Context Effects: How Display Fixtures Can Affect Product Evaluations ," Journal of Marketing Research , 46 (1), 37 – 45.
~~~~~~~~
By Hean Tat Keh; Di Wang and Li Yan
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 61- GMO Labeling Policy and Consumer Choice. By: Kim, Youngju; Kim, SunAh; Arora, Neeraj. Journal of Marketing. May2022, Vol. 86 Issue 3, p21-39. 19p. 6 Color Photographs, 1 Diagram, 3 Charts, 3 Graphs. DOI: 10.1177/00222429211064901.
- Database:
- Business Source Complete
GMO Labeling Policy and Consumer Choice
Most scientists claim that genetically modified organisms (GMOs) in foods are safe for human consumption and offer societal benefits such as better nutritional content. However, many consumers remain skeptical about their safety. Against this backdrop of diverging views, the authors investigate the impact of different GMO labeling policy regimes on the products consumers choose. Guided by the literature on negativity bias, structural alignment theory, and message presentation, and based on findings from four experiments, the authors show that consumer demand for GM foods depends on the labeling regime policy makers adopt. Both absence-focused ("non-GMO") and presence-focused ("contains GMO") labeling regimes reduce the market share of GM foods, with the reduction being greater in the latter case. GMO labels reduce the importance consumers place on price and enhance their willingness to pay for non-GM products. Results indicate that specific label design choices policy makers implement (in the form of color and style) also affect consumer responses to GM labeling. Consumer attitudes toward GMOs moderate this effect—consumers with neutral attitudes toward GMOs are influenced most significantly by the label design.
Keywords: GMO; food claim; voluntary labeling; mandatory labeling; public policy and marketing
Although the use of genetically modified (GM) foods has become widespread across the world, scientific and public opinions diverge about their safety. Most scientists agree that GM foods are invaluable because they offer increased nutritional content, a higher yield per acre, and a better shelf life ([51]). They also agree that GM foods are as safe for humans and the environment as non-GM foods ([18]). Yet some scientists disagree ([29]), citing concerns about possible long-term effects of GM foods on human health and the environment ([10]). On the demand side, many consumers question the safety of GM foods and their scientific promise ([30]). Indeed, a 2013 New York Times poll showed that 75% of Americans expressed concern about genetically modified organisms (GMOs) in their food, and most worried about their potential health effects. It is important to note here that the baseline consumer knowledge on this issue is low ([64]). The most extreme opponents of GM foods think they have the most knowledge about the issue, but research shows that their scientific literacy is low ([21]).
The opposing views that firms and consumers have about GM foods create a fundamental tension in how such foods should be labeled, which is the central focus of this research. On the one hand, consumers and advocacy groups believe that GM foods are potentially risky; therefore, policy makers must mandate GMO labeling. In a mandatory labeling regime, food manufacturers are required to include labels such as "contains GMO" when their foods are GM. The most commonly used argument in support of such labeling is consumers' right to know. On the other hand, food manufacturers rely on scientific evidence to claim that GM foods are as safe as conventionally grown foods. As a result, they argue that mandatory labeling arbitrarily singles out GMO technology for specific attention and misleads consumers into thinking that they should be concerned about consuming GM foods ([60]). Therefore, food manufacturers support a voluntary GMO labeling policy, where firms have the freedom to use a "non-GMO" label when appropriate.
The discordant views about the safety of GM foods between firms and consumers, as well as the demands for GMO labeling by consumer advocacy groups (e.g., the Non-GMO Project) create a substantial challenge for policy makers in their efforts to develop a GMO labeling policy. As a result, GMO labels vary a great deal around the world (see Figure 1). For example, the United States allows firms to display non-GMO labels on their products if they wish. Brazil, the world's second-largest GM producer after the United States, adopted a mandatory GM label that features a black T inside a yellow triangle. The letter T stands for the Portuguese word transgenicos ("transgenics"), and the symbol resembles a caution sign indicating an upcoming T-junction ([ 6]). Similar logos have been adopted by South American countries such as Bolivia and Uruguay.
Graph: Figure 1. GMO labels.
In light of the diverse GMO policy regimes that currently exist, an important prerequisite for carefully constructing a GMO labeling policy is a theory-based understanding of whether and how consumers shift their choices under the different GMO labeling regimes. The intention behind a labeling policy that requires the disclosure of a GMO ingredient as a horizontally differentiated attribute is that it simply allows consumers to make choices that reflect their taste differences. However, an externality of such a policy may be that it leads consumers to treat GMO ingredients as a vertically differentiated attribute, signaling that non-GM foods are of a higher quality than GM foods.
To investigate the product quality–related implications empirically, we examine how the different GMO labeling policy regimes impact consumers' choice and their willingness to pay for non-GM products as well as the market shares of GM and non-GM products. Guided by the policy question of mandatory versus voluntary labeling for GM foods, we investigate the substantial impact of different GMO policy regimes on choices consumers make. More specifically, the purpose of our research is to answer the following research questions that have significant GMO labeling policy implications. Using the theory-driven terminology adopted by [ 1], in the remainder of the article we refer to the mandatory labeling regime as "presence-focused" (contains GMO) and the voluntary labeling regime as "absence-focused" (non-GMO).
- Does the labeling policy (absence-focused vs. presence-focused) affect a consumer's choice of GM products?
- Does the labeling policy (absence-focused vs. presence-focused) affect other aspects of the consumer's choice process, such as their price sensitivity and willingness to shop in a category?
- Is the consumer's choice impacted according to whether the GMO related information disclosure is complete (presence-focused and absence-focused) or partial (presence-focused or absence-focused)? Complete GMO information disclosure occurs when policy makers mandate presence-focused labeling and firms that produce non-GM products display an absence-focused label. Partial disclosure occurs when either presence or absence labels are present.
- Do consumers behave differently depending on the GM label's presentation format (e.g., color, theme)? Which consumers are most likely to alter choices because of the label format?
To answer these key policy-related questions, we combine insights from the social psychology literature with rigorous consumer choice models to make novel predictions about the effect of GMO labeling changes on consumers' demand for GM products. We develop our theory based on the literature on negativity bias (e.g., [31]), structural alignment theory (e.g., [25]), and message presentation (e.g., [33]). We use choice experiments grounded in microeconomic theory ([40]) and a hierarchical Bayes model to test our hypotheses.
Across four studies involving 3,913 respondents, we study the impact of different GMO labeling regimes on demand for GM products. Our first between-subjects experiment (Study 1) examines whether consumer choice depends on the GMO labeling regime (i.e., no labeling, absence-focused, presence-focused, or both labeling conditions). Study 2 investigates how the GMO labeling regime may impact the importance consumers place on product price and product category. Study 3 shows how the alignability of GMO information, whether partial information or full information is disclosed, affects consumer choice. Finally, Study 4 investigates how the signal used in a GMO label (e.g., color) can impact consumer demand for GM foods and reveals which market segment is most likely to be affected by the signal used.
Our findings have substantive implications for two key stakeholder groups: policy makers and food manufacturers. By quantifying the effects of various GMO labeling regimes, we offer policy makers guidance on the impact of each labeling system on consumer demand. Absence-focused policies result in the smallest change in demand for GM products compared with a regime with no GM labels. Presence-focused labeling policies can substantially alter demand for GM products, and the signal contained in the GMO logo (e.g., color) also plays a critical role in consumers' perceptions of GM products. Both policy regimes create incentives for firms to expand their offerings to include more non-GM products for the market segment that prefers such products and is willing to pay more for them. The critical question for policy makers here is whether they wish to incentivize such firm behavior.
For food manufacturers, our research reveals that GM labels add an important product feature for consumers to evaluate. Such labels create vertical differentiation for many consumers by signaling that non-GM products are better than GM products. They draw attention away from factors such as price—making it less important—and allow firms to charge a premium for non-GM products. The GM label can also drive some consumers away from a category (e.g., from crackers to another non-GM snack). All of the aforementioned effects are amplified when moving from an absence-focused to presence-focused regime.
The [63], p. 1) defines GMOs as "organisms (i.e., plants, animals, or microorganisms) in which the genetic material (DNA) has been altered in a way that does not occur naturally by mating and/or natural recombination." Proponents of GM crops argue that they increase yield, lower food prices, reduce damage to crops after the harvest, make crops tolerant of stresses such as cold and heat, help fight malnourishment, and reduce reliance on chemical pesticides ([51]). Most scientists claim that there is no substantiated evidence that genetic crop modification makes foods less safe. For example, the National Academies of Sciences and Medicine ([46]) reported that food from GM crops is no more dangerous than food produced by conventional agriculture. More than 150 Nobel laureates in areas such as chemistry, physics, and medicine signed an open letter in 2016 to endorse the safety of GM foods, noting that "opposition based on emotion and dogma contradicted by data must be stopped" (Support Precision Agriculture 2016).
Although the dominant view among scientific organizations is that GMOs do not harm human health, this view is not ubiquitous. In one review article, [36] noted that a group of scientists believe that each GM product should be tested over long periods for possible side effects. The author reviewed 26 animal feeding studies that identified adverse effects or animal health uncertainties, leading him to conclude that "putative consensus about the inherent safety of transgenic crops is premature" (p. 909). A joint statement by a group of researchers ([29]) claimed that no consensus on GM food safety exists. They indicated that a conflict of interest exists in many reported studies supporting GM food because biotechnology companies often fund this research ([15]). They further noted that no epidemiological studies have examined the effects of GM food consumption on humans. They concluded that it is necessary to test the effect of GM foods on humans and over longer periods.
Several studies have documented consumers' lack of knowledge about GMOs, as noted by [64] in their review. These studies also document an overall negative attitude toward GMOs among consumers. Such negative attitudes could be driven by negative press associated with occurrences such as GM crops causing a decline in monarch butterflies, which a recent article refutes ([ 7]). The primary concerns are that growing and consuming GM crops may cause health problems and allergic reactions.
Research has shown that the most extreme opponents of GM foods know the least about GMOs but think that they know the most ([21]). People's misplaced confidence stemming from the mismatch between what they think they know about science and what they actually know ([45]) polarizes attitudes even more ([22]).
The controversy around GM foods also relates to the growing literature on science denial ([54]) that identifies social mechanisms as the basis for extreme confidence in beliefs that contradict scientific consensus ([33]). Specifically, many people have insufficient information to establish their own opinions on new technologies and scientific developments ([20]) and instead accept the opinions of people they trust ([54]). Well-known examples of science denial include vaccine safety, global warming and climate change, the rise in antibiotic resistance, and the safety of GM foods.
GMO labeling policy in the United States was absence-focused when GM foods were first released in 1994. Some food companies use third-party verification, such as the Non-GMO Project (https://www.nongmoproject.org), to highlight the non-GMO aspect of their products. However, various consumer groups and nongovernmental organizations have argued for presence-focused labeling based on consumers' right to know what is in their food. They contend that the potential harm of GM foods needs to be made explicit.
Over the years, political pressure to introduce presence-focused GMO labeling in the United States has grown. In July 2016, U.S. Congress passed a bill requiring the U.S. Department of Agriculture to establish a national disclosure standard for GMOs. The new policy has a two-year phase-in period that began in January 2020. The proposed label under this policy has a nature-friendly theme on a green or black-and-white background and uses the term "bioengineered (BE)." Dozens of nations around the world have enacted presence-focused GMO labels based on the percentage of GMOs in ingredients or how the seed was developed. The GMO percentage thresholds vary among countries that have regulations. For example, the European Union (EU) and United Kingdom set this limit at.9%, whereas Australia set it at 1% ([31]).
Policy makers in many countries are uncertain whether GM foods are safe, and their labeling rules are based on such a perspective. For example, the EU has adopted the precautionary principle (European Commission [19]) for GMO labeling. This principle is often cited in cases of scientific uncertainty and the possibility of irreversible damage. It states that "where there are threats of serious or irreversible damage, lack of full scientific certainty shall not be used as a reason for postponing cost-effective measures to prevent environmental degradation" ([49], Principle 15).
Not surprisingly, GMO labeling policies are controversial. In a compelling counterargument to EU policies, [59] acknowledged the rationality behind the precautionary principle but cautioned that rigid regulatory controls based on the idea of "possible risk" can paralyze progress. He explained that, for GMOs, the precautionary principle results in substitute risks because it interferes with the promise of mitigating hunger and disease in developing countries by using foods such as golden rice, which is bioengineered to be rich in vitamin A.
In the United States, debate over the recently adopted logo is intense because it appears to signal the government's positive attitude toward GM foods. In Brazil, opponents of the current mandatory presence-focused GM logos accuse the country's policy makers of scaremongering. In Canada, which currently has no GMO labeling legislation, petitions have been circulated supporting mandatory presence-focused GMO labeling; in a survey of adults living in Canada, 90% of the respondents expressed support for mandatory GM labeling ([56]).
In this section, we develop predictions on how GMO labeling affects different aspects of consumer choice, including preference for GM foods, price sensitivity, and product category purchase. We also outline how these predicted effects are moderated by the GMO label format. Figure 2 summarizes our main hypotheses.
Graph: Figure 2. Overview of predictions.
In an absence-focused labeling policy, manufacturers may use a "non-GMO" label when appropriate. In a world in which many consumers have negative attitudes toward GM products, non-GMO producers often use absence-focused claims (e.g., a TV ad for Triscuit[ 5]). Such claims are in line with those that highlight the absence of negatives—namely, no preservatives, no artificial colors, no chemicals, and so on. These nature-based claims that remove a negative strongly affect consumers' inferences about product's taste and healthfulness ([ 1]), even when they are irrelevant to the actual quality. In contrast, in a presence-focused GMO labeling policy, regulators mandate that GM-food products display a "contains GMO" label. For many consumers who have negative attitudes toward GM products, this information signals potential risk.
The absence- and presence-focused labeling policies outlined thus far may impact consumers' evaluations of GM foods via two separate mechanisms—information valence and information source. With regard to information valence, it is well known that people place greater weight on negative information than positive information. This negativity bias ([50]) is at the core of how consumers may evaluate a GMO label. From an evolutionary perspective, this bias occurs because we have a greater chance of surviving and thriving if we pay greater attention to negative information; negative events are more consequential than positive ones. Some argue that negative information is more informative because it is rarer ([24]), attracting more attention and thus being more "diagnostic or informative" ([53]). Previous research has documented negativity bias in a variety of contexts. For example, negative attributes are more diagnostic of product quality than positive ones ([28]), and negative reviews have a stronger effect on purchase decisions ([ 9]).
In addition, absence- and presence-focused labeling differs by their information source. In presence-focused GMO labeling, a regulatory body mandates the display of GMO labels, whereas in absence-focused labeling, this decision is voluntary and made by the firm. The perceived credibility of a message's source can affect the recipient's cognitive response ([58]). For trusted information sources, consumers accept a message without undertaking an extensive assessment of its content ([23]). Evidence suggests that consumers trust public sources (e.g., a government) more than private ones (e.g., a firm). For example, [17] found that advertising from a government source (the Federal Trade Commission) is more credible than that from a firm. Similar results have been noted for safety hazard information ([38]), environmental information ([11]), and forest-product certification seals ([48]). Given the differences in information valence and source between the two labeling regimes discussed thus far, we propose the following hypothesis:
- H1: Presence-focused GMO labeling makes consumers more sensitive to the GMO attribute when choosing a product than absence-focused GMO labeling.
Previous studies have shown that negative information is more diagnostic and, as a result, attracts more attention ([53]). In line with the argument that attention is a scarce resource, [12] demonstrated that greater attention to a previously unconsidered attribute reduces the relative importance of other attributes. Extending these theoretical findings to our research context, greater attention to GMO ingredients likely diverts consumer attention away from other product-related information, such as price. Negative presence-focused GMO labeling should reduce consumers' focus on price information more compared with absence-focused GMO labeling. Thus,
- H2: Presence-focused GMO labeling makes consumers less price sensitive than absence-focused labeling.
Choice deferral is a means of mitigating the negativity generated in uncertain or difficult choice contexts. Previous research shows that this negative feeling in such contexts increases the likelihood that consumers will defer their decision ([39]). Deferral occurs when no single option dominates a choice set or when consumers face difficult trade-offs between product attributes ([13]).
In our context, consider a Brand A, which contains GMOs, and a Brand B, which does not. Assume that a consumer prefers Brand A but prefers non-GMO ingredients. In the absence-focused condition, this consumer will choose between Brand A, not knowing whether it is a GM product, and Brand B, fully aware that it is a non-GM product because of the "non-GMO" label. Conversely, in the presence-focused condition, the same consumer will choose between Brand A, fully aware that it is a GM product because of the "contains GMO" label and Brand B, with no GMO-related knowledge. The trade-off between brand name and GM ingredients may be more difficult in the presence-focused condition because the consumer has to analyze the costs and benefits of a brand they prefer (Brand A) and an attribute they do not (GMO). This increased task difficulty may enhance the likelihood of choice deferral.
[42] show that consumers tend to view a government's default option as an implicitly recommended course of action. In their studies, when the government uses the "organ donor" default, most participants inferred that ( 1) the policy makers were willing to be donors and ( 2) people ought to be donors. In the GMO labeling context, a GMO label mandated by a regulatory body may send a negative signal that consumers should avoid a product with GMO ingredients. Growing concerns and perceived risks associated with GMOs could increase customers' uncertainty about brand quality, thus leading to choice deferral. Therefore,
- H3: Presence-focused GMO labeling makes consumers less likely to purchase the relevant product category than absence-focused GMO labeling.
The first three hypotheses focus on scenarios where the product alternatives available apply either an absence-focused label or a presence-focused label. However, when a government mandates presence-focused labeling, firms that produce non-GM products may be free to use absence-focused labeling, as is the case in the United States today. Because many consumers view non-GM products favorably, firms offering non-GM products have a strong incentive to include such information on their product packaging to differentiate themselves from firms offering GM products. Therefore, when a mandatory GMO labeling policy is implemented in the marketplace, it is plausible that most—if not all—products will display either GMO or non-GMO labels. We use structural alignment theory ([35]; [55]; [65]) to discuss the impact of partial or complete GMO-related information on consumer choice and how they drive our predictions.
Consider the following example involving two marinara sauce brands, A and B. Brand A is sold at $2.00, without providing any information on GMO attributes; Brand B is sold at $2.50 and includes a non-GMO label. In this example, the price is alignable information because the attribute is present in both options. In contrast, under either the absence-focused or presence-focused labeling, GMO-related information is only available for Brand B, making it nonalignable. The structural alignment literature suggests that consumers pay more attention to alignable attributes ([25]) and put greater weight on them ([55]).
Consistent with this discussion, when both types of GMO labels are included (i.e., "non-GMO" and "contains GMO"), consumers will give greater weight to the GMO attribute. According to the arguments used previously for H2, giving greater weight to the GMO attribute would ( 1) further reduce the weight consumers give to price information and ( 2) make them more reluctant to purchase a product in the category. Formally,
- H4: Compared with a situation where only presence-focused ("contains GMO") labels are displayed, when both absence-focused ("non-GMO") and presence-focused ("contains GMO") labels are displayed, consumers become (a) more sensitive to GMOs, (b) less sensitive to price, and (c) less likely to purchase in the product category.
A regulatory body's choice of GMO label reveals its beliefs or attitudes about GMOs and is, therefore, a critical policy decision—consumers tend to view a government's default option as an implicit recommended course of action ([42]). Moreover, [52] showed that a speaker's description signals their attitude toward an object. For example, if someone likes a team, they describe its successes, and if they do not, they note the team's failures. The descriptions a speaker chooses, even of seemingly equivalent objects, are important for listeners ([41]).
As a concrete example involving the color of a GMO label, consumers tend to infer that a product has positive, nature-related attributes when it prominently displays the color green ([61]). Similarly, the color blue signals openness, peace, and tranquility ([43]), whereas yellow signals caution. Such color choices and their associated signals are highly relevant for GMO labeling. We hypothesize that policy makers' choice of a GMO label (e.g., the color green, blue, or yellow) is important as it delivers an implicit recommendation that may influence consumers' choices.
- H5: The graphical format of the label determines how much impact the GMO attribute has on consumer choice, including (a) how sensitive consumers are to the GMO attribute, (b) how important price is to consumers, and (c) how likely consumers are to purchase in a given product category.
Consumers' prior beliefs about GMOs could also play a role in how much a labeling policy impacts them. Previous research showed that most individuals do not know enough details to establish their own perspectives on new technologies and scientific developments ([20]) and accept the position of others they trust ([54]). As a result, we anticipate that consumers in the middle, who neither like nor dislike GM products, are affected the most by the label format policy makers select.
We include four empirical studies. The first study uses a simple between-subjects design to examine the effect of different GMO labeling policies (i.e., absence, presence, or both) on consumer choice. We subsequently conduct three choice-based conjoint studies to test H1–H5. Study 2 focuses on H1, H2, and H3 by disentangling the impact of GMO labeling on different aspects of consumer choice. Study 3 tests H4, focusing on how the findings pertaining to H1–H3 are affected by GMO information disclosure (partial vs. full). Lastly, Study 4 tests H5, focusing on how the different graphical formats of GMO labeling impact our previous findings.
The goal of Study 1 was to demonstrate that different GMO labeling regimes (i.e., no GMO labeling, absence-focused labeling, presence-focused labeling, and both labeling conditions) can lead to systematic differences in demand for GM foods.
Using Amazon Mechanical Turk (MTurk), we recruited respondents in exchange for monetary compensation. To begin, we asked the respondents questions to assess whether they shopped in the focal categories (marinara sauce and pickle) and whether they were paying attention to the study instructions. Of the 2,110 respondents who completed this first step of the study, 767 (36.4%) respondents did not qualify to continue because they did not shop in the two focal categories (N = 644; 30.5%) or failed to correctly answer the attention check questions (N = 123; 5.8%). A total of 1,343 respondents (Mage = 41.0 years; female = 62%) qualified to participate in the main study and completed it. We randomly assigned these respondents to one of the four study conditions in a between-subjects design (Ncontrol = 340, Nabsence = 331, Npresence = 335, Nboth = 337).
We presented respondents with choice sets in two different product categories (marinara sauce and pickles) and asked them to select their preferred brand. We selected these two product categories because ( 1) they are frequently purchased and ( 2) they complement the less healthy product category (potato chips) that we use in our subsequent studies.
In the marinara sauce category, the first two (Prego and Newman's Own) of the three brands included in the study were GM products while the third (Barilla) was a non-GM product. Similarly, in the pickles category, two brands (Claussen and Vlasic) were GM products and the third (Mt. Olive) was a non-GM product. In both categories, we gave respondents in the control condition no information on GMOs. In the absence labeling condition, only the non-GMO label was displayed. In the presence labeling condition, only the GMO label was displayed. Finally, in the both labeling condition, both non-GMO and GMO labels were displayed. Figure 3 shows the four GMO labeling conditions for the pickles category in this study. The stimuli for the marinara sauce category are included in the Web Appendix A (Figure W1).
Graph: Figure 3. Stimuli for four labeling conditions (pickles).
Figure 4 shows the share of non-GM and GM products in each category across the four labeling conditions. We found that the GMO labeling regime significantly impacts consumer choice of the brands included. In all three GMO labeling conditions (absence, presence, and both), more participants preferred the non-GM product than in the control condition, where no GMO-related information is displayed. The magnitude of these shifts in demand toward the non-GM product (Barilla or Mt. Olive) depended on the GMO labeling condition.
Graph: Figure 4. Non-GMO vs GMO choice share across different labeling conditions.
In the marinara sauce category (Figure 4, Panel A), the choice share of non-GM product was lower in the control condition ( = .171) than in the absence ( = .236; z = −2.094, p = .046), presence ( = .370; z = −5.825, p < .001), and both labeling conditions ( = .407; z = −6.780, p < .001). In addition, the non-GM product's choice share was lower in the absence labeling condition ( = .236) than in the presence ( = .370; z = −3.761, p < .001) and the both ( = .407; z = −4.730, p = ) labeling conditions. These results imply that consumer preference for non-GM products increases progressively from control to absence to presence labeling conditions. The difference between the presence and both labeling conditions was not statistically significant ( = .370 vs. = .407; z = −.984, p = ).
Graph: Figure 5. Example choice tasks in study 2.
The findings for the pickles category (Figure 4, Panel B) are largely consistent with those for marinara sauce. The choice share of the non-GM product was lower in the control condition ( = .229) than in the absence ( = .293; z = −1.888, p = ), presence ( = .421; z = −5.327, p < .001), and both ( = .469; z = −6.552, p < .001) labeling conditions. As in the marinara sauce category, the non-GM product's choice share was lower in the absence labeling condition ( = .293) than in the presence ( = .421; z = −3.446, p < .001) and both ( = .469; z = −4.685, p < .001) labeling conditions. Once again, the difference between the presence and both labeling conditions was not statistically significant ( = .421 vs. = .469; z = −1.252, p = ).
Study 1 supports our central thesis—namely, that the way the GMO message is conveyed to consumers affects their choices. In all GMO labeling regimes, the share of non-GM products is greater than when no GMO-related information is revealed. The share of non-GM products increases progressively from control to absence to presence labeling conditions. Although these findings are interesting on their own, Study 1 raised important follow-up research questions. For example, should we expect GMO labeling to affect the importance of price to consumers and their willingness to pay (WTP) for a non-GM product? Does the manner in which the GMO message is delivered affect the product choice consumers make? Furthermore, Study 1 confounds brand and GM ingredients, as only the Barilla brand carried the GMO-free label and only Prego and Newman's Own displayed the "contains GMO" label in the marinara sauce category. To answer our next set of research questions and remove the confounding between brand and GMO ingredients, we conducted three choice-based conjoint experiments, which we report on next.
The purpose of Study 2 was twofold. First, we test H1, H2, and H3, which examine the impact of GMO labeling (absence vs. presence) on consumers' sensitivity to the GMO attribute, price, and category purchase, respectively. Second, we explore the behavioral processes associated with consumer choice under the different GMO labeling regimes. Unlike Study 1, we do not have a confound between brand and GMO ingredients in this study.
To introduce Study 2, we used incentive alignment instructions similar to those employed by [27], informing respondents that 25 of them would receive a total value of $5 based on their answers to the survey: either a product of their choice and a Walmart e-gift card for the remaining value, or a $5 Walmart e-gift card. Existing literature shows that incentive-alignment techniques make consumers likely to provide more realistic responses. When a study offers as an incentive a version of the product that is predicted to give consumers the highest utility, respondents put greater effort into the choice task, and their responses are more likely to reflect their actual preferences ([14]; [27]).
We informed respondents that the study's purpose was to understand how they evaluated potato chips. We asked the respondents to complete 14 choice tasks. In each choice task, we showed the respondents four brands of potato chips: Lay's, Herr's, Ruffles, and a private label. The price varied by brand. As in the marketplace, we set the prices of the national brands ($2.79, $3.29, $3.79) and the private brand ($1.99, $2.39, $2.79) at different levels. The two GMO label conditions were included or not included on the package. We displayed all four brands in each choice set, varying the GMO ingredients and prices from one scenario to the next in the 14 choice tasks.
We used a statistically efficient choice design based on the D-optimality criterion ([37]) for the main-effects model that included three attributes varied orthogonally across choice tasks, which prevented us from testing higher-order interactions. For example, Lay's chips could have a GMO label in one task but not in the next; by design, the brand and GMO label attribute are unconfounded, enabling us to assess consumer preference for each. For more information on conjoint analysis and how it relates to our study, please see Web Appendix B.
We adopted a dual-response method, asking respondents to indicate ( 1) which of the product alternatives they prefer and ( 2) whether they would actually buy the product they had just selected ([ 4]; [27]). The dual-response method has the advantage of encouraging respondents to slow down to think through the purchase task, making the no-choice option more likely, which is similar to a real market situation. Next, respondents answered questions about their attitudes toward GMOs, which enabled us to investigate how consumer attitudes vary across the different labeling regimes. We used a nine-point "disagree/agree" scale to ask questions regarding attention to the GMO ingredient, risk perception of the GMO ingredient, and decision uncertainty.
We randomly assigned respondents to one of the two conditions (absence- vs. presence-focused labeling) in a between-subjects design. These two between-subjects conditions differed by how the GMO label was displayed. In the absence-focused condition, only the non-GMO label was displayed, whereas in the presence-focused condition, only the GMO label was displayed. Figure 5 presents a screenshot of one of the 14 choice tasks for each condition in Study 2.
We recruited students from a Midwestern university in the United States, and they participated in the study in exchange for course credit. To begin, we asked the students questions to assess whether they shopped in the focal category (potato chips) and whether they had a dietary restriction (gluten intolerance). Of the 665 students who completed this first step of the study, 55 (8.3%) respondents did not qualify to continue because they did not shop in the focal category (N = 49; 7.4%) or were gluten-intolerant (N = 6;.9%).[ 6] A total of 610 students (Mage = 19.7 years; female = 44.4%) qualified for and completed the main study. We randomly assigned them to one of the two study conditions (Nabsence = 303, Npresence = 307).
The proportion of respondents purchasing a non-GM product was higher in the presence condition (56.38%) than in the absence condition (48.84%). The proportion of respondents who decided not to make a purchase was also higher in the presence condition (37.06%) than in the absence condition (32.77%). The average purchase price was similar in the two conditions ($3.07 in absence vs. $3.05 in presence).
This study focused on the impact of GMO labeling on brand and category choice. Given this, we modeled two decisions—whether to buy and which brand to choose—using a nested framework to model brand choice and purchase incidence ([ 2]; [26]), where the joint probability of a given consumer choosing brand j and buying it (B) is given by
Graph
( 1)
We conceptualized our model at the brand level and assumed that individual h evaluating j (j = 1, ..., J) brands chooses the brand j. Each brand j has a design vector xj that contains discrete variables to indicate the different attribute levels (e.g., GMO) and a continuous variable (e.g., price). The deterministic part of individual h's utility for brand j is linear in the predictor variables (xjβh) and, with a Type I extreme value error structure, yields a multinomial logit model [40]. The probability of individual h choosing brand j is given by
Graph
( 2)
where βh is an individual-level parameter vector that includes brand preference and sensitivity to attributes such as GMO and price. To model the buy/no-buy decision, we specified a threshold utility (γh) parameter in the model. The utility of the most attractive alternative (j) in the choice set must exceed γh for the individual to buy it (B = 1). A larger estimated threshold parameter implies a lower probability of buying in the category.
We introduced heterogeneity across individuals hierarchically with a random effects specification ([ 4]) as , where and . In our empirical context, the hyperparameter includes the average brand preference parameters, GMO sensitivity ( ), and price sensitivity ( ), while is the average threshold for category purchase. For the remaining technical details of the model, refer to Web Appendix C.
We employed the Markov chain Monte Carlo (MCMC) method to estimate the hierarchical Bayes model. Similar to others in the literature ([ 2]; [ 3]; [ 5]; [16]), we estimate the model for each experimental condition separately (i.e., and ) and use Bayesian inference to test for differences in estimates between the experimental conditions. To obtain a one-sided p-value, we calculated the fraction of the empirical posterior distribution that is inconsistent with the formulated hypothesis. For example, to test H1, we calculated the proportion of the GMO sensitivity distribution that is inconsistent with H1 (i.e., ). To claim statistical significance, we report a two-sided Bayesian p-value, which equals two times the one-sided p-value ([62]). We used similar analyses to test the remaining hypotheses.
Table 1 shows the findings of Study 2 based on the model in Equations 1 and 2. This table contains the posterior means and standard deviations of the hyperparameter estimates ( , ) for each experimental condition. The first three rows correspond to the , , and , which are the posterior means of preference for each brand (relative to the private label). The next two rows report and , the posterior means of sensitivities associated with the GMO and price attribute, respectively. The last row reports the average threshold parameter estimate.
Graph
Table 1. Study 2 Posterior Estimates of and .
| Parameters | Absence | Presence |
|---|
| Lay's | 6.11 | 4.19 |
| (.45) | (.33) |
| Herr's | −1.95 | −1.49 |
| (.94) | (.54) |
| Ruffles | 6.52 | 3.55 |
| (.53) | (.34) |
| GMO | .00 | −1.12 |
| (.11) | (.12) |
| Price | −4.31 | −2.25 |
| (.27) | (.14) |
| Category threshold | −6.49 | −2.15 |
| (.89) | (.50) |
1 Notes: Boldfaced parameters indicate that the 95% highest posterior density interval of the estimate did not include zero. The numbers in parentheses below the posterior mean are the standard deviations.
A negative parameter estimate means that, on average, consumers like this product characteristic less than the baseline. For example, a negative estimate for GMO ( ) implies that, on average, customers prefer a non-GM product over a GM product. In our studies, we can compare this parameter estimate across the GMO labeling conditions. For example, if the parameter estimate of the GMO attribute in the presence-focused condition ( ) is more negative than in the absence-focused condition ( ), we interpret this as showing that presence-focused GMO labeling amplifies consumer preference for non-GM over GM products.
The impact of GMO ingredients on product choice was not statistically significant in the absence-focused condition, but was negative and statistically significant in the presence-focused condition ( = .00 vs. = −1.12; p < .001). This finding supports H1. The parameter shows how sensitive consumers are to price changes. The more negative it is, the more consumers are sensitive to price changes. We find that the price coefficient is more negative in the absence condition than in the presence condition ( = vs. = ; p < .001); this implies that participants were less sensitive to price changes under the presence-focused labeling, thereby supporting H2.
To assess consumers' decision to buy in the product category or not, captures the average threshold value for product category purchase. The product category is purchased when the utility of at least one product exceeds this threshold; therefore, the higher the threshold parameter , the lower the probability of the category purchase. The category purchase threshold was higher in the presence condition than in the absence condition ( = −6.49 vs. = −2.15; p < .001), indicating that consumers were less likely to purchase in the potato chips category in the presence-focused condition. This result supports H3.
In summary, Study 2's findings support H1, H2, and H3 in an incentive-aligned conjoint experiment. The main takeaways thus far are that the GMO labeling regime (absence vs. presence) affects how sensitive consumers are to the GMO attribute. The labeling also impacts consumers' price sensitivity and their likelihood of making a purchase in a product category.
The second aim of Study 2 was to understand why consumers' choices differ between the two GMO labeling conditions. We examined whether GMO risk perception differs by labeling format and, if so, whether it affects the extent to which consumers pay attention to the GMO information. In each condition, we asked the respondents a series of questions pertaining to GMOs (e.g., concerns, perceived risk, attention paid). We then asked questions relevant to choice deferral. To measure choice task difficulty, we asked, "Was it difficult to decide which product to pick?" ([32]). To measure choice task uncertainty, we asked, "How certain were you that the product would be the best in each choice task?" and "How much did you regret choosing the product you picked in each choice task?" ([34]). We used a nine-point scale (1 = "not at all," and 9 = "extremely") for all of the questions asked.
Figure 6 shows that consumers have greater concern about GMO products (Mabsence = 1.92 vs. Mpresence = 3.36; p < .001) and a greater perceived risk from them (Mabsence = 2.08 vs. Mpresence = 3.65; p < .001) in the presence condition than in the absence condition. Consumers paid greater attention to GMO products (Mabsence = 2.67 vs. Mpresence = 4.01; p < .001). These results confirm two things. First, presence labeling makes consumers experience more negative feelings (greater concern and risk perception) about GMOs. Second, as documented in the negativity bias literature, these negative perceptions result in consumers paying greater attention to the GMO attribute.
Graph: Figure 6. GMO labeling: perception, attention, and choice difficulty.
Regarding the questions pertaining to choice deferral, consumers in the presence labeling condition found the choice tasks to be more difficult (Mabsence = 2.52 vs. Mpresence = 3.20; p < .001). Consumers were less certain about their choice (Mabsence = 6.12 vs. Mpresence = 5.70; p = .009), which is consistent with the choice deferral findings we reported previously. Consumers also experienced greater regret about the choices they made (Mabsence = 2.32 vs. Mpresence = 2.91; p < .001).
A useful way to quantify the impact of the two labeling regimes is to examine consumers' WTP for the non-GMO attribute by condition. The basic assumption in a choice model is that respondents consider trade-offs between attributes when they select an alternative. Therefore, one can estimate each attribute's marginal utility from the parameter estimates and obtain the WTP for an attribute as a ratio of its marginal utility and the price coefficient ([47]; [57]). For the GMO attribute, WTP is measured by evaluating the quantity . We determined that the WTP for the non-GMO attribute was much higher (p < .001) in the presence condition ($.50, SD = .07) than in the absence condition ($.00, SD = .03).
In summary, Study 2 shows that presence-focused labeling makes consumers ( 1) more sensitive toward the GMO attribute, ( 2) less sensitive toward price information, and ( 3) more reluctant to make a purchase in a category. Thus, Study 2 supports H1, H2, and H3. Presence-focused labeling enhances consumers' concerns about GMOs, encourages them to pay greater attention to GMO information, and makes their choice decision more difficult. In a presence-focused labeling regime, firms producing non-GM products could benefit from increased WTP for the non-GMO attribute, providing them with an incentive to offer more non-GM products and to charge more for them.
The objective of Study 3 was to investigate the effect of aligned GMO information on consumer choice, as predicted by H4. To this end, we added a third condition (both) whereby all product alternatives display either the absence or the presence of GMO information.
The structure of Study 3 was similar to that of Study 2, except for the following changes. First, we used different absence- and presence-labeling stimuli from those in the previous two studies (see Figure 7). Second, we did not use incentive alignment. Third, as previously noted, we added a labeling condition (both) in which the "non-GMO" label appeared on products that did not contain GMO ingredients and the "GMO" label appeared on products that did.
Graph: Figure 7. Visual descriptions of the stimuli in study 3.
Using MTurk, we recruited respondents for this study in exchange for monetary compensation. To begin, we asked the respondents questions to assess whether they shopped in the focal category for this study (potato chips) and whether they were paying attention to the study instructions. Of the 925 respondents who completed this first step of the study, 43 (4.6%) respondents did not qualify to continue because they did not shop in the focal category (N = 17; 1.8%) or failed to correctly answer the attention check questions (N = 26; 2.8%). A total of 882 respondents (Mage = 38.3 years; female = 51.5%) qualified for the main study and completed it. We randomly assigned them to one of the three study conditions (Nabsence = 290, Npresence = 299, Nboth = 293).
As in Study 2, the proportion of participants choosing the non-GM products was higher in the presence (vs. absence) labeling condition. In addition, more participants chose the non-GM product in the both labeling condition than in the other two conditions (51.56% in absence, 54.98% in presence, 58.78% in both). The proportion of participants deciding not to purchase the product category was lowest in the absence labeling condition (10.05% in absence, 13.93% in presence, 13.24% in both). The average purchase price was slightly higher in the both (vs. absence) labeling condition ($2.93 in absence, $2.95 in presence, $2.98 in both).
Table 2 summarizes the findings from Study 3 based on the model previously outlined in Equations 1 and 2. We began by investigating whether the findings in Study 2 also held for Study 3. As before, we found that the impact of GMO ingredients was stronger in the presence labeling condition ( = −.41 vs. = −1.21; p < .001). The price coefficient was more negative in the absence labeling condition ( = −4.47 vs. = −3.54; p = .032), implying that consumers were less price sensitive in the presence (vs. absence) labeling condition. The threshold parameter was higher in the presence labeling condition ( = −12.39 vs. = −9.04; p = .018), meaning that consumers were less likely to purchase the product category in the presence condition. Overall, Study 3's findings also support H1, H2, and H3.
Graph
Table 2. Study 3 Posterior Estimates of and .
| Parameters | Absence | Presence | Both |
|---|
| Lay's | 3.99 | 4.09 | 3.90 |
| (.44) | (.37) | (.43) |
| Herr's | 1.94 | 1.47 | .30 |
| (.60) | (.43) | (.54) |
| Ruffles | 4.78 | 3.57 | 3.59 |
| (.48) | (.38) | (.45) |
| GMO | −.41 | −1.21 | −1.93 |
| (.13) | (.18) | (.23) |
| Price | −4.47 | −3.54 | −2.85 |
| (.33) | (.26) | (.21) |
| Category threshold | −12.39 | −9.04 | −7.28 |
| (1.21) | (.90) | (.80) |
2 Notes: Boldfaced parameters indicate that the 95% highest posterior density interval of the estimate did not include zero. The numbers in parentheses below the posterior mean are the standard deviations.
Returning to the primary purpose of Study 3—to investigate how GMO-related information alignment affects consumers' choice behaviors—the comparison of parameter estimates between presence and both labeling conditions revealed an interesting pattern. We found that the parameter estimates for GMO, price, and category incidence were higher in the both (vs. presence) labeling condition. Respondents became more sensitive to GMO ingredients when both non-GMO and GMO labels were displayed ( = −1.21 vs. = −1.93; p = .010); they also became less price sensitive in the both labeling condition ( = −3.54 vs. = −2.85; p = .036). The difference in the threshold parameters between the both and the presence labeling conditions is not statistically significant ( = −9.04 vs. = −7.28; p = .140). Thus, Study 3 supports H4a and H4b—namely, structurally aligned information makes the GMO attribute more salient and even more important for brand choice. The evidence pertaining to category choice in Study 3 does not support H4c.
All our findings thus far have focused on the average market response. The hierarchical Bayes model also captured individual-level responses within each labeling condition. We found that some respondents had a positive GMO coefficient, meaning they preferred products with GMOs. The segment size of this "prefer GMO" segment was 39.31%, 29.43%, and 17.06% in the absence, presence, and both labeling conditions, respectively. The important takeaway is that the proportion of consumers who prefer GMO-based products decreases as we move from absence-focused to presence-focused to both label conditions. Therefore, the labeling policy regime showed a substantial impact on the size of the "prefer GMO" segment.
The goal of Study 4 was to test H5—to determine whether the graphical format of the GMO label affects the impact of the GMO attribute on consumer choice. More specifically, we aimed to understand whether the graphical format affects ( 1) how sensitive consumers are to the GMO attribute, ( 2) how important price is to consumers, and ( 3) how likely consumers are to make a purchase in a product category. To accomplish this goal, we varied the appeal of the GMO label using different colors and themes, creating a "positive" green label that casts GM products in a favorable light and a more moderate "neutral" blue label (see Figure 8). The positive-looking green label bears some resemblance to the one recently adopted in the United States,[ 7] while the neutral blue label has a similar design to the one adopted in Uruguay.[ 8]
Graph: Figure 8. Logos in three labeling conditions.
The purpose of the pretest was to show that the two GMO logos used in Study 4 (Figure 8) convey different signals to respondents. More specifically, we tested whether the green GMO label delivered a more positive signal to respondents (i.e., GMOs are beneficial and less harmful) than the neutral blue label.
Using MTurk, we recruited respondents for this pretest in exchange for monetary compensation. To begin, we asked the respondents questions to assess whether they were paying attention to the pretest instructions. Of the 208 respondents who completed this first step of the study, 6 (2.9%) did not qualify to continue because they failed to correctly answer the attention check question. The remaining 202 respondents (Mage = 41.3 years; female = 56.4%) qualified for and completed the main study. We randomly assigned them to one of the two study conditions (Nboth-positive = 101, Nboth-neutral = 101).
We asked respondents to indicate the extent to which they agree or disagree with the following statements on a nine-point scale (1 = "very unlikely," and 9 = "very likely"): ( 1) "A product with the GMO label above is likely beneficial for me," and ( 2) "A product with the GMO label above is likely harmful for me." The second statement was reverse-coded, and we reported the average for the two questions for each label. Results indicate that the natural green label led to GMOs being perceived as more beneficial than the blue label (Mgreen = 5.76, SD = 2.64 vs. Mblue = 4.82, SD = 2.32; t(200) = −2.69, p = ).
The structure of Study 4 was similar to that of Study 3. We created three between-subjects conditions, one absence labeling condition and two both labeling conditions. The absence condition used the same non-GMO label as Studies 1 and 2. In the both labeling conditions, we used the two different GMO label formats (see Figure 8): the "positive" green label and the "neutral" blue label. Our primary interest was to compare the two both conditions with the absence condition.
Figure 9 shows the stimuli for the three labeling conditions that we used in the choice-based conjoint experiment. As in Studies 2 and 3, we focused on three attributes: brand, price, and GMO; the choice design for the conjoint part was the same as before. The rest of the procedure was similar to Study 3, except that we added questions to measure consumers' prior attitude toward GM products. As we discuss subsequently, this approach enabled us to assess the interplay among the GMO labeling regimes, consumer attitudes, and demand for GM products.
Graph: Figure 9. Example of stimuli in choice tasks in Study 4.
Using MTurk, we recruited respondents for this study in exchange for monetary compensation. To begin, we asked the respondents questions to assess whether they shopped in the focal category (potato chips) and whether they were paying attention to the study instructions. Of the 1,089 respondents who completed this first step of the study, 213 (19.6%) respondents did not qualify to continue because they did not shop in the focal category (N = 109; 10.0%) or failed to correctly answer the attention check questions (N = 104; 9.6%). A total of 876 respondents (Mage = 37.5 years; female = 49.5%) qualified for and completed the main study. We randomly assigned them to one of the three conditions (Nabsence = 291, Nboth-positive = 294, Nboth-neutral = 291).
The descriptive statistics show a similar pattern to Study 3. A larger proportion of respondents chose the non-GM products in the both labeling conditions (50.15% in absence, 51.34% in both-positive, 53.91% in both-neutral).
Table 3, Panel A, reports the parameter estimates for Study 4. Comparing the absence and both-neutral (blue) labeling conditions, produced results similar to those from Study 3. Participants in the both-neutral condition were more sensitive to the GMO ingredients ( = −.39 vs. = −.79; p = .034) and less price sensitive ( = −5.60 vs. = −4.67; p = .018). The difference in the likelihood to purchase in the product category is not significant ( = −13.04 vs. = −12.18; p = .616). However, a comparison between the absence and both-positive (green) GMO labeling conditions revealed a different pattern. There was no significant difference in consumers' GMO sensitivity ( = −.39 vs. = −.43; p = ) or category purchase probability ( = −13.04 vs. = −12.00; p = ). The only difference occurred in price sensitivity—consumers were less price sensitive in the both-positive labeling condition than in the absence condition, and it is marginally significant ( = 5.60 vs. = −4.67; p = .06). The primary takeaway from the results is that the GMO labeling format determined the extent to which the GMO attribute affected consumer preference for GM foods. Stated differently, a comparison between the both-positive and both-neutral conditions revealed that the difference in the GMO sensitivity is marginally significant ( = −.43 vs. = −.79; ). This finding marginally supports H5a and has substantial policy implications, which we discuss subsequently. Neither the price sensitivity ( = −4.67 vs. = −4.39; p = .54) nor the category purchase probability ( = −12.00 vs. =−12.18; p = .912) is significantly different between the two conditions. These findings in Study 4 do not support H5b and H5c.
Graph
Table 3. Results from Study 4.
| A: Posterior Estimates of and |
|---|
| | Both |
|---|
| Parameters | Absence | Both-Positive | Both-Neutral |
|---|
| Lay's | 6.14 | 5.35 | 4.46 |
| (.52) | (.47) | (.42) |
| Herr's | 3.28 | 2.01 | 1.32 |
| (.63) | (.70) | (.51) |
| Ruffles | 6.31 | 5.33 | 3.99 |
| (.59) | (.54) | (.44) |
| GMO | −.39 | −.43 | −.79 |
| (.12) | (.14) | (.14) |
| Price | −5.6 | −4.67 | −4.39 |
| (.41) | (.34) | (.30) |
| Category threshold | −13.04 | −12.00 | −12.18 |
| (1.35) | (1.33) | (1.06) |
| B: GMO Sensitivity for Each Subgroup Across Conditions |
| Absence | Both-Positive | Both-Neutral |
| Beneficial | −.21 | −.2 | −.41 |
| (N = 238; 27.2%) | (.16) | (.15) | (.16) |
| No strong opinion | −.20 | −.18 | −.66 |
| (N = 389; 44.4%) | (.15) | (.14) | (.16) |
| Harmful | −.41 | −.92 | −1.21 |
| (N = 249; 28.4%) | (.16) | (.22) | (.22) |
3 Notes: Boldfaced parameters indicate that the 95% highest posterior density interval of the estimate did not include zero. The numbers in parentheses below the posterior mean are the standard deviations.
In our final analyses, we examined whether a respondent's prior attitude toward GMOs affected their reactions to the signals in the GMO logos. To test this idea, we asked the respondents whether they think GMOs are beneficial or harmful; we then divided them into groups based on their prior attitude toward GMOs (1–3: "Beneficial," 4–6: "No strong opinion," 7–9: "Harmful"). Of the 876 respondents, 27.2% (N = 238) answered that GMOs are beneficial, 44.4% (N = 389) responded that they have no strong opinion about GMOs, and 28.4% (N = 249) reported that they think GMOs are harmful. Using individual-level parameters, we report the average GMO sensitivity by attitude group and label condition in Table 3, Panel B.[ 9]
Drawing on the findings in the top row of Table 3, Panel B, for the group that considered GMOs beneficial (27.2% of the sample), the both-neutral GMO labeling format had a significant impact (p = .025) on GMO sensitivity. For the group with no strong opinion about GMOs (44.4% of the sample), the both-neutral GMO labeling format had a significant impact (p = .002) on GMO sensitivity (see middle row, third column of Table 3, Panel B). The other two GMO label formats (absence and both-positive) did not have any impact on GMO sensitivity for this group. This is the most important finding in Table 3, Panel B. From these results, we concluded that, for a very large segment of the sample, the GMO label format had a decisive impact on how consumers view GMOs in the product choices they make. Lastly, among participants who viewed GMOs as harmful (28.4% of the population; bottom row of Table 3, Panel B), compared with the absence condition, GMO labeling in the both-neutral condition had a larger (p = .004) impact on GMO sensitivity.
Study 4 examined the impact of GMO label format on consumer choice. Our results showed that a GMO logo can systematically influence consumer choices. The signal contained in the label format can cause large shifts in consumer preference for GM foods. Importantly, we found that the GMO label format had a greater impact on consumers who had no strong opinions about GMOs, suggesting that preference for GM foods is highly pliable for a large segment of consumers. This effect occurs when using a neutral label format in this study, suggesting that a label that signals caution (e.g., Brazil's yellow transgenico logo) is likely to have an even more pronounced effect.
Most scientists claim that GM foods are safe for human consumption and that they offer substantial advantages; meanwhile, many consumers who lack scientific knowledge are skeptical about GM foods. Fueling these conflicting views, consumer advocacy groups have asserted that consumers need to know what they consume. Against this backdrop of diverging stakeholder views, we investigated the impact of different GMO labeling regimes. In particular we studied the impacts of voluntary, absence-focused labeling ("non-GMO") versus mandatory, presence-focused labeling ("contains GMO"). A natural extension of the latter is a third regime in which both label types are displayed on products in the marketplace. We showed that each GMO policy has a substantial impact on consumer choices and creates incentives for firms that are important for policy makers to consider.
Guided by the literature on negativity bias, structural alignment theory, and message presentation, and based on the findings of our four studies, we show that each of the three labeling regimes (absence, presence, and both) greatly affects consumers' demand for GM foods. Labels such as "non-GMO" and "contains GMO" serve as negative signals for GM foods and tend to shrink their market share. The market share shrinkage effect is stronger under the mandatory policy than the voluntary policy. GMO labeling reduces the importance consumers place on product price and impacts the consumers' WTP for non-GM products. The finding pertaining to increased preference for the non-GM products is amplified when both non-GMO and GMO labels are displayed on the products.
Finally, we found that the signal policy makers decide to send via the GM label (e.g., a green logo may be viewed as an endorsement, a yellow logo as a cautionary signal) significantly affects consumer choice. Consumers' prior attitudes toward GMOs moderate this finding; consumers who are neutral toward GMOs are impacted most by the signal contained in the label.
In line with their relative impact on demand for GM foods, label regulations could be viewed at three levels: low, medium, and high impact. They correspond respectively to the three policy regimes: absence, presence, and both. Compared with a situation where no GMO labels exist, the voluntary GMO labeling policy ("non-GMO") affects the demand for GM products the least. Mandatory labels ("contains GMO") substantially affect the demand for GM products, and when both types of labels are present ("non-GMO" and "contains GMO"), the demand shifts are the highest.
Figure 10 shows that GMO labeling has an economically significant impact on consumer WTP for non-GM products and reveals several important insights. First, voluntary, absence-focused labeling results in the lowest increase in WTP for non-GM products. In two of the latter three studies (Studies 3 and 4), this increase in WTP was nonetheless statistically and economically significant (9 cents and 7 cents, respectively). In comparison, in Studies 2 and 3, the increase in WTP in the mandatory, presence-focused labeling was substantially higher, at 50 cents and 34 cents, respectively. In Study 3, when both labels were present, WTP was the highest (68 cents). These higher WTP measures in Studies 2 and 3 should be viewed in light of the GMO label used: a yellow GMO logo similar to Brazil's in Study 2, and a red GMO label in Study 3. In Study 4, in which we tested positive (green) and neutral (blue) "contains GMO" logos, the WTP was still positive and economically significant (9 and 18 cents, respectively). The upshot here is that consumer WTP critically depends on the label policy makers adopt. Across studies, findings indicate that both the voluntary and mandatory labeling regimes create incentives for firms to add premium-priced, non-GM products to their portfolio of offerings. These incentives are substantially greater in the mandatory labeling regime than in the voluntary regime.
Graph: Figure 10. Impact of GMO labeling on Willingness to Pay (WTP $).
Contrary to the opinion that mandatory GMO labeling will merely satisfy consumers' right to know and give them complete information, our findings show that any form of GMO labeling has significant externalities that policy makers must consider carefully. GMO labeling reduces the demand for GM foods and creates incentives for firms to offer higher-priced non-GM foods, which raises the question of whether policy makers intend to promote this effect.
In addition to deciding which labeling regime to implement, policy makers also have to wrestle with another critical decision related to the signal contained in the GMO label. For example, consumers may view a green label as an endorsement of GM foods and a yellow label as a signal to exercise caution. The United States has decided to implement the use of a mandatory GMO labeling system by 2022; the proposed label has a nature-friendly theme on a green or black-and-white background, along with the term "bioengineered (BE)." In contrast, the GMO label in Brazil is a yellow triangle resembling a caution sign. Our findings indicate that even a neutral GMO label may lead consumers to focus on the negative aspects of GMOs, pay less attention to price information, and become more reluctant to make a purchase in the product category. In Brazil, all these effects are likely amplified. The two labeling regimes at each end of the spectrum in Brazil and the United States, along with the insights offered by this article, may offer guidance to other countries about which labeling regime they should adopt.
Our research reveals that GM labels add an important product feature for consumers to evaluate. It draws attention away from factors such as price, allowing firms to charge a premium for non-GM products. The GM label can also drive some consumers away from a category (e.g., from crackers to another non-GM snacking category). All of these effects are amplified as we move from absence-focused (voluntary) to presence-focused (mandatory) policies. Both regimes create incentives for firms to expand their offerings to include more non-GM products for a segment of the market that prefers such products and is willing to pay more for them. As Figure 10 illustrates, consumer WTP for a non-GM product is higher in a presence-focused labeling regime. GM manufacturers inevitably lose market shares when presence-focused labeling is enforced. They face both reduced brand share and reduced category demand. Because presence-focused labeling makes consumers less price sensitive, GM food manufacturers may attempt to compensate for their sales loss by considering promotions other than price cuts.
Unlike the debate on mandatory labeling, the discussion of regulatory aspects of voluntary labeling is limited. Regulators intervene when products make unsubstantiated health benefit claims ([ 8]). Along the same lines, future research could investigate the need for policies to regulate voluntary non-GMO disclosures. To complement our findings based on choice experiments, it would be instructive to rely on purchase data over time (e.g., [44]) from sources such as Nielsen and IRI to investigate how GMO labeling affects consumer behavior. Our choice experiments offer limited evidence for category shrinkage effects because of GMO labeling; a rigorous test for such effects would be in a multicategory context using household-level panel data.
GMO labels create vertical differentiation for many consumers by signaling that non-GM products are better than GM products. They draw attention away from factors such as price—making it less important—and allow firms to charge a premium for non-GM products. Even a voluntary GMO labeling policy deserves regulatory scrutiny because it causes a decrease in demand for GM products. In comparison, a mandatory GMO labeling policy shrinks the demand for GM products even more, and the signal contained in the GMO logo (e.g., color) plays a critical role in consumers' perceptions of GM products. Both voluntary and mandatory policy regimes create incentives for firms to expand their offerings to include more non-GM products for the market segment that prefers such products and is willing to pay more for them. The critical question for policy makers here is whether they wish to promote such consumer and firm behaviors.
sj-pdf-1-jmx-10.1177_00222429211064901 - Supplemental material for GMO Labeling Policy and Consumer Choice
Supplemental material, sj-pdf-1-jmx-10.1177_00222429211064901 for GMO Labeling Policy and Consumer Choice by Youngju Kim, SunAh Kim and Neeraj Arora in Journal of Marketing
Footnotes 1 Ralf Van der Lans
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship and/or publication of this article.
4 Youngju Kim https://orcid.org/0000-0003-4413-1224 SunAh Kim https://orcid.org/0000-0001-9763-5042
5 See https://www.ispot.tv/ad/dRmG/triscuit-non-gmo-project-verified-featuring-cecily-strong.
6 This study is incentive aligned, and we shipped potato chips to a fraction of the respondents. For safety concerns, we excluded respondents who indicated that they were gluten intolerant because some potato chips may contain gluten or may be produced at a facility that makes snacks with gluten (Roehmholdt n.d.).
7 For more information, see https://www.ams.usda.gov/rules-regulations/be/symbols.
8 For more information, see https://montevideo.gub.uy/areas-tematicas/salud/regulacion-alimentaria/reglamentacion-para-alimentos-transgenicos.
9 The GMO sensitivity for attitude group g in label condition l in Table 3, Panel B, is given by posterior mean of GMO estimates for each attitudinal group (i.e., {Beneficial, No strong opinion, Harmful}, and {Absence, Both-Positive, Both-Neutral}). Here, r indexes a draw from the posterior distribution, R is the total number of draws, and Ng,l is the number of respondents in the attitude group g for label condition l. The parameter is the respondent h's GMO sensitivity for the rth draw.
References André Quentin , Chandon Pierre , Haws Kelly. (2019), " Healthy Through Presence or Absence, Nature or Science? A Framework for Understanding Front-of-Package Food Claims ," Journal of Public Policy & Marketing , 38 (2), 172 – 91.
Aribarg Anocha , Arora Neeraj , Henderson Ty , Kim Youngju. (2014), " Private Label Imitation of a National Brand: Implications for Consumer Choice and law ," Journal of Marketing Research , 51 (6), 657 – 75.
Aribarg Anocha , Burson Katherine A. , Larrick Richard P.. (2017), " Tipping the Scale: The Role of Discriminability in Conjoint Analysis ," Journal of Marketing Research , 54 (2), 279 – 92.
Arora Neeraj , Allenby Greg M. , Ginter James L.. (1998), " A Hierarchical Bayes Model of Primary and Secondary Demand ," Marketing Science , 17 (1), 29 – 44.
Arora Neeraj , Henderson Ty. (2007), " Embedded Premium Promotion: Why It Works and How to Make It More Effective ," Marketing Science , 26 (4), 514 – 31.
Borges Bárbara J.P. , Arantes Olivia M.N. , Fernandes Antonio A.R. , Broach James R. , Fernandes Patricia M.B.. (2018), " Genetically Modified Labeling Policies: Moving Forward or Backward? " Frontiers in Bioengineering and Biotechnology , 6 (181), https://www.frontiersin.org/articles/10.3389/fbioe.2018.00181/full.
Boyle John H. , Dalgleish Harmony J. , Puzey Joshua R.. (2019), " Monarch Butterfly and Milkweed Declines Substantially Predate the Use of Genetically Modified Crops ," Proceedings of the National Academy of Sciences , 116 (8), 3006 – 11.
Calfee John E. , Pappalardo Janis K.. (1991), " Public Policy Issues in Health Claims for Foods ," Journal of Public Policy & Marketing , 10 (1), 33 – 53.
Chevalier Judith A. , Mayzlin Dina. (2006), " The Effect of Word of Mouth on Sales: Online Book Reviews ," Journal of Marketing Research , 43 (3), 345 – 54.
Dannenberg Astrid. (2009), " The Dispersion and Development of Consumer Preferences for Genetically Modified Food—A Meta-Analysis ," Ecological Economics , 68 (8/9), 2182 – 92.
Darnall Nicole , Ponting Cerys , Vazquez-Brust Diego A.. (2012), " Why Consumers Buy Green ," in Green Growth: Managing the Transition to a Sustainable Economy , Diego A. Vazquez-Brust and Joseph Sarkis, eds. New York: Springer , 287 – 308.
DellaVigna Stefano. (2009), " Psychology and Economics: Evidence from the Field ," Journal of Economic Literature , 47 (2), 315 – 72.
Dhar Ravi. (1997), " Consumer Preference for a No-Choice Option ," Journal of Consumer Research , 24 (2), 215 – 31.
Ding Min. (2007), " An Incentive-Aligned Mechanism for Conjoint Analysis ," Journal of Marketing Research , 44 (2), 214 – 23.
Domingo José L. , Bordonaba Jordi G.. (2011), " A Literature Review on the Safety Assessment of Genetically Modified Plants ," Environment International , 37 (4), 734 – 42.
Dong Songting , Ding Min , Huber Joel. (2010), " A Simple Mechanism to Incentive-Align Conjoint Experiments ," International Journal of Research in Marketing , 27 (1), 25 – 32.
Dyer Robert F. , Kuehl Philip G.. (1974), " The 'Corrective Advertising' Remedy of the FTC: An Experimental Evaluation: An Empirical Study of the Effectiveness of Corrective Advertising ," Journal of Marketing , 38 (1), 48 – 54.
Egeberg Morten. (2010), " The European Commission ," in European Union Politics , 3rd ed. , Michelle Cini and Nieves Perez-Solorzano Borragan, eds. Oxford, UK: Oxford University Press, 125 – 40.
European Commission (2017), The Precautionary Principle: Decision-Making Under Uncertainty. Luxembourg: Publications Office of the European Union.
Fernbach Philip M. , Light Nicholas. (2020), " Knowledge Is Shared ," Psychological Inquiry , 31 (1), 26 – 8.
Fernbach Philip M. , Light Nicholas , Scott Sydney E. , Inbar Yoel , Rozin Paul. (2019), " Extreme Opponents of Genetically Modified Foods Know the Least but Think They Know the Most ," Nature Human Behaviour , 3 (3), 251 –5 6.
Fernbach Philip M. , Rogers Todd , Fox Craig R. , Sloman Steven A.. (2013), " Political Extremism Is Supported by an Illusion of Understanding ," Psychological Science , 24 (6), 939 – 46.
Finch David , Deephouse David , Varella Paul. (2015), " Examining an Individual's Legitimacy Judgment Using the Value–Attitude System: The Role of Environmental and Economic Values and Source Credibility ," Journal of Business Ethics , 127 (2), 265 – 81.
Fiske Susan T.. (1980), " Attention and Weight in Person Perception: The Impact of Negative and Extreme Behavior ," Journal of Personality and Social Psychology , 38 (6), 889.
Gentner Dedre , Markman Arthur B.. (1997), " Structure Mapping in Analogy and Similarity ," American Psychologist , 52 (1), 45 – 56.
Hanemann W. Michael. (1984), " Discrete/Continuous Models of Consumer Demand ," Econometrica , 52 (3), 541 – 61.
Hauser John R. , Eggers Felix , Selove Matthew. (2019), " The Strategic Implications of Scale in Choice-Based Conjoint Analysis ," Marketing Science , 38 (6), 1059 – 81.
Herr Paul M. , Kardes Frank R. , Kim John. (1991), " Effects of Word-of-Mouth and Product-Attribute Information on Persuasion: An Accessibility-Diagnosticity Perspective ," Journal of Consumer Research , 17 (4), 454 – 62.
Hilbeck Angelika , Binimelis Rosa , Defarge Nicolas , Steinbrecher Ricarda , Székács András , Wickson Fern , et al. (2015), " No Scientific Consensus on GMO Safety ," Environmental Sciences Europe , 27 (1), 1 – 6.
Hingston Sean T. , Noseworthy Theodore J.. (2018), " Why Consumers Don't See the Benefits of Genetically Modified Foods, and What Marketers Can Do About It ," Journal of Marketing , 82 (5), 125 – 40.
International Service for the Acquisition of Agri-Biotech Applications. (2004), "Pocket K No. 7: Labeling GM Foods," (August), https://www.isaaa.org/resources/publications/pocketk/7/default.asp.
Ito Tiffany A. , Larsen Jeff T. , Smith Kyle N. , Cacioppo John T.. (1998), " Negative Information Weighs More Heavily on the Brain: The Negativity Bias in Evaluative Categorizations ," Journal of Personality and Social Psychology , 75 (4), 887 – 900.
Janiszewski Chris , Wyer Robert S. Jr.. (2014), " Content and Process Priming: A Review ," Journal of Consumer Psychology , 24 (1), 96 – 118.
Jiang Yuwei , Gorn Gerald J. , Galli Maria , Chattopadhyay Amitava. (2016), " Does Your Company Have the Right Logo? How and Why Circular- and Angular-Logo Shapes Influence Brand Attribute Judgments ," Journal of Consumer Research , 42 (5), 709 – 26.
Kahan Dan M. , Jenkins-Smith Hank , Braman Donald. (2011), " Cultural Cognition of Scientific Consensus ," Journal of Risk Research , 14 (2), 147 – 74.
Kahn Barbara E.. (2017), " Using Visual Design to Improve Customer Perceptions of Online Assortments ," Journal of Retailing , 93 (1), 29 – 42.
Kivetz Ran , Simonson Itamar. (2000), " The Effects of Incomplete Information on Consumer Choice ," Journal of Marketing Research , 37 (4), 427 – 48.
Krimsky Sheldon. (2015), " An Illusory Consensus Behind GMO Health Assessment ," Science, Technology, and Human Values , 40 (6), 883 – 914.
Kuhfeld Warren F.. (2010), " Conjoint Analysis ," SAS Tech. Pap., MR H , 2010 , 681 – 801.
Lirtzman Sidney I. , Shuv-Ami Avichai. (1986), " Credibility of Sources of Communication on Products' Safety Hazards ," Psychological Reports , 58 (3), 707 – 18.
Luce Mary F.. (1998), " Choosing to Avoid: Coping with Negatively Emotion-Laden Consumer Decisions ," Journal of Consumer Research , 24 (4), 409 – 33.
McFadden Daniel. (1973), " Conditional Logit Analysis of Qualitative Choice Behavior ," in Frontiers of Econometrics , Zarembka P. , ed. New York : Academic Press.
McKenzie Craig R.M. (2004), " Framing Effects in Inference Tasks—and Why They Are Normatively Defensible ," Memory and Cognition , 32 (6), 874 – 85.
McKenzie Craig R.M. , Liersch Michael J. , Finkelstein Stacey R.. (2006), " Recommendations Implicit in Policy Defaults ," Psychological Science , 17 (5), 414 – 20.
Mehta Ravi , Zhu Rui Juliet. (2009), " Blue or Red? Exploring the Effect of Color on Cognitive Task Performances ," Science , 323 (5918), 1226 –2 9.
Moorman Christine. (1996), " A Quasi Experiment to Assess the Consumer and Informational Determinants of Nutrition Information Processing Activities: The Case of the Nutrition Labeling and Education Act ," Journal of Public Policy & Marketing , 15 (1), 28 – 44.
Motta Matthew , Callaghan Timothy , Sylvester Steven. (2018), " Knowing Less but Presuming More: Dunning-Kruger Effects and the Endorsement of Anti-Vaccine Policy Attitudes ," Social Science and Medicine , 211 , 274 – 81.
National Academies of Sciences, Engineering and Medicine (2016), Genetically Engineered Crops: Experiences and Prospects. Washington, DC: National Academies Press.
Ofek Elie , Srinivasan Venkataraman. (2002), " How Much Does the Market Value an Improvement in a Product Attribute? " Marketing Science , 21 (4), 398 – 411.
Ozanne Lucie K. , Vlosky Richard P.. (1997), " Willingness to Pay for Environmentally Certified Wood Products: A Consumer Perspective ," Forest Products Journal , 47 (6), 39 – 48.
Roehmholdt, Rachael (n.d.), "Gluten-Free Chips," blog entry (accessed January 6, 2022), https://www.rachaelroehmholdt.com/gluten-free-chips.
Rozin Paul , Royzman Edward B.. (2001), " Negativity Bias, Negativity Dominance, and Contagion ," Personality and Social Psychology Review , 5 (4), 296 – 320.
Sharma Samriti , Kaur Rajinder , Singh Anupama. (2017), " Recent Advances in CRISPR/Cas Mediated Genome Editing for Crop Improvement ," Plant Biotechnology Reports , 11 (4), 193 – 207.
Sher Shlomi , McKenzie Craig R.M.. (2006), " Information Leakage from Logically Equivalent Frames ," Cognition , 101 (3), 467 – 94.
Skowronski John J. , Carlston Donal E.. (1989), " Negativity and Extremity Biases in Impression Formation: A Review of Explanations ," Psychological Bulletin , 105 (1), 131 – 42.
Sloman Steven , Fernbach Philip. (2018), The Knowledge Illusion: Why We Never Think Alone. New York: Penguin.
Slovic Paul , MacPhillamy Douglas. (1974), " Dimensional Commensurability and Cue Utilization in Comparative Judgment ," Organizational Behavior and Human Performance , 11 (2), 172 – 94.
Somogyi Simon , Music Janet , Cunningham Cunningham C.. (2019), " Biotechnology in Food: Canadian Attitudes Towards Genetic Engineering in Both Plant- and Animal-Based Foods ," British Food Journal , 121 (12), 3181 – 92.
Sonnier Garrett , Ainslie Andrew , Otter Thomas. (2007), " Heterogeneity Distributions of Willingness-to-Pay in Choice Models ," Quantitative Marketing and Economics , 5 (3), 313 – 31.
Sternthal Brian , Dholakia Ruby , Leavitt Clark. (1978), " The Persuasive Effect of Source Credibility: Tests of Cognitive Response ," Journal of Consumer Research , 4 (4), 252 – 60.
Sunstein Cass R.. (2005), Laws of Fear: Beyond the Precautionary Principle, Vol. 6. Cambridge, UK: Cambridge University Press.
Sunstein Cass R.. (2016), " On Mandatory Labeling, with Special Reference to Genetically Modified Foods ," University of Pennsylvania Law Review , 165 (5), 1043 – 95.
Support Precision Agriculture (2016), "Laureate Letter Supporting Precision Agriculture (GMOs)," (June 29), https://supportprecisionagriculture.org/nobel-laureate-gmo-letter_rjr.html.
Underwood Robert L. , Klein Noreen M.. (2002), " Packaging as Brand Communication: Effects of Product Pictures on Consumer Responses to the Package and Brand ," Journal of Marketing Theory and Practice , 10 (4), 58 – 68.
United Nations General Assembly (1992), " Rio Declaration on Environment and Development ," (August 12), https://www.un.org/en/development/desa/population/migration/generalassembly/docs/globalcompact/A%5fCONF.151%5f26%5fVol.I%5fDeclaration.pdf.
Wedel Michel , Dong Chen. (2020), " BANOVA: Bayesian Analysis of Experiments in Consumer Psychology ," Journal of Consumer Psychology , 30 (1), 3 – 23.
World Health Organization (2014), "Frequently Asked Questions on Genetically Modified Foods," (May), https://www.who.int/foodsafety/areas%5fwork/food-technology/Frequently%5fasked%5fquestions%5fon%5fgm%5ffoods.pdf.
Wunderlich Shahla , Gatto Kelsey A.. (2015), " Consumer Perception of Genetically Modified Organisms and Sources of Information ," Advances in Nutrition , 6 (6), 842 – 51.
Zhang Shi , Markman Arthur B.. (2001), " Processing Product Unique Features: Alignability and Involvement in Preference Construction ," Journal of Consumer Psychology , 11 (1), 13 – 27.
~~~~~~~~
By Youngju Kim; SunAh Kim and Neeraj Arora
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 62- Halo or Cannibalization? How New Software Entrants Impact Sales of Incumbent Software in Platform Markets. By: Allen, B.J.; Gretz, Richard T.; Houston, Mark B.; Basuroy, Suman. Journal of Marketing. May2022, Vol. 86 Issue 3, p59-78. 20p. 1 Diagram, 8 Charts. DOI: 10.1177/00222429211017827.
- Database:
- Business Source Complete
Halo or Cannibalization? How New Software Entrants Impact Sales of Incumbent Software in Platform Markets
Platform markets involve indirect network effects as two or more sides of a market interact through an intermediary platform. Many platform markets consist of both a platform device and corresponding software. In such markets, new software introductions influence incumbent software sales, and new entrants may directly cannibalize incumbents. However, entrants may also create an indirect halo impact by attracting new platform adopters, who then purchase incumbent software. To measure performance holistically, this article introduces a method to quantify both indirect and direct paths and determine which effect dominates and when. The authors identify relevant moderators from the sensations–familiarity framework and conduct empirical tests with data from the video game industry (1995–2019). Results show that the direct impact often results in cannibalization, which generally increases when the entrant is a superstar or part of a franchise. For the indirect halo impact, superstar entrants significantly increase platform adoption, which can help all incumbents. Combining the direct and indirect impacts, the authors find that only new software that is both a superstar and part of a franchise increases platform adoption sufficiently to overcome direct cannibalization and achieve a net positive effect on incumbent software; all other types of entrants have a neutral or negative overall effect.
Keywords: cannibalization; entertainment franchise; halo effect; indirect network effects; platform markets; superstars; two-sided markets; video games
A platform market involves two or more user groups (i.e., sides of a market) whose interactions are mediated through a platform. These markets are typically characterized by indirect network effects, as the actions of agents on one side of the market affect the outcomes of agents on another side ([44]). Many of these markets consist of both platform devices and corresponding software ([24]; [42]). When a new software entrant launches into such a platform market, its ultimate success might depend on a variety of factors or be measured in a variety of ways—sales of the entrant, impacts on sales of incumbent software already on the market, or platform sales. Yet with few exceptions (e.g., [24]; [30]), extant research has focused on sales of the core platform, rather than specifying how an entrant might alter the sales of the existing (i.e., incumbent) software portfolio. In particular, new software might cannibalize incumbent sales ([46]), but it also arguably can create a halo effect that increases the sales of competitive, incumbent software ([32]). These outcomes are critical across the platform ecosystem; as [42], p. 1232) explain, "Members of a platform ecosystem often have strong vested interest in each other's fates. Because it is the overall appeal of the ecosystem that attracts end users to the platform, the success of individual members depends, at least in part, on the success of other members of the ecosystem—even those with which they may be simultaneously competing."
Consider the paths in Figure 1. A new software entrant can influence incumbent software sales directly (Path A) and positively if the entrant stimulates usage among existing platform owners. However, a negative competitive effect is more likely, as consumers choose to purchase the entrant instead of the incumbent. In addition, the entrant might exert an indirect, positive impact on the incumbent by stimulating sales of the platform (i.e., new platform adoptions) (Path B). Consumers who newly adopt the platform likely backfill their collection of platform-compatible software by purchasing incumbent software (Path C).
Graph: Figure 1. Paths by which new software entrants impact sales of incumbent software.
But how much of the entrant's impact on incumbent software sales is through direct Path A versus indirect Paths B and C? This central question has not been addressed, and answering it will provide a more holistic view of entrants and their sales implications. While variations of Path B (software → platform sales) have been studied (for a review, see [48]), scholars have "focused almost exclusively on quantity" ([29], p. 39) or size of the network. Scholars rarely differentiate the impact of entrants (instead of the quantity of software stock) or new platform adoptions (instead of the quantity of the platform's installed base).
In addition, Paths A and C have yet to be broadly addressed. A few studies examine individual software launches but do not quantify the impact on incumbent software. For example, [ 8] study how software entrants impact hardware sales but do not measure whether incumbent software sales are affected; [30] investigates both platform and software sales but assumes no competition between available software. This gap is important because research outside of platforms shows that new products can impact incumbents both positively (e.g., via spillover; [ 1]) and negatively (e.g., by increasing competition; [46]). Given this evidence, we argue that it is vital to understand the holistic impact of new software launches on incumbent software sales.
Further, it is important that new product research accounts for the complexities of platform markets, in which entrants influence incumbents both directly (as substitutes or complements [Path A]) and indirectly (through platform demand [Paths B and C]). Such insight is useful, as many new products release into platform markets (e.g., apps, video games); our approach distinguishes the impact of an entrant versus classic software supply. Rather than measure the impact of an increase in overall software stock, we examine the attributes of new software that differentially affect incumbent software (and platform) sales. These findings are not a simple extension of extant knowledge (i.e., new software affects platform sales) into a similar setting. Whereas software entrants are complements to platforms, they are primarily substitutes to incumbent software. Further, in these complex markets, purchases can put customers into active states that prompt them to purchase additional (potentially incumbent) software ([23]). We thus rely on established evidence that platform sales influence software demand, but we go further to detail the specific effects on incumbent software driven by new platform adoptions. In Figure 1, our contributions refer mainly to the understudied Paths A and C (see also Table 1).
Graph
Table 1. Studies of Indirect Network Effects in Platform Markets.
| | Traditional Network Logic | Paths Presented in Current Study |
|---|
| Authors | Relevant Research Question | Impact of Software Stock on Platform Adoption | Impact of Platform Installed Base on Software Stock or Sales | Path A: Direct Impact of New Software on Incumbent Software | Path B: Direct Impact of New Software on Platform Adoption | Path C: Indirect Impact (Through Platform) of New Software on Incumbent Software | Moderators: The Impact of New Software on Incumbents Differs Based on Software Characteristics |
|---|
| Binken and Stremersch (2009) | What is the differential impact of superstar versus nonsuperstar software on hardware adoption? | Yes | No | No | Yes (including impact of software characteristics) | No | No |
| Cox (2014) | Do software characteristics determine if software will be a blockbuster? | No | Yes | No | No | No | No |
| Gretz et al. (2019) | How is the impact of superstar software on hardware moderated by hardware's product life cycle? | Yes (software characteristics) | Yes | No | No | No | No |
| Haviv, Huang, and Li (2020) | Does software stock in a given period affect software demand in subsequent periods? | Yes | Yes | No, but measures how increases in stock affect later periods' demand | No | No | No |
| Healey and Moe (2016) | Does platform recency and innovativeness affect software adoption? | No | Yes | No | No | No | No |
| Kim, Prince, and Qiu (2014) | Does hardware quality, beyond the installed base, affect the supply of software products? | Yes | Yes | No | No | No | No |
| Landsman and Stremersch (2011) | How does multihoming impact platform sales? | Yes | Yes | No | No | No | No |
| Lee (2013) | Does vertical integration and exclusive software affect hardware and software? | Yes | Yes | Yes, robustness check of how superstars affect other superstars | No | No | No |
| Stremersch et al. (2007) | Does software lead hardware or does hardware lead software? | Yes | Yes | No | No | No | No |
| Current study | How do new software entrants affect the sales of incumbent software, directly and indirectly? | Yes | Yes | Yes | Yes | Yes | Yes |
To determine these effects, we use a dynamic model to measure the impact of software entrants on incumbent software sales and platform sales, while also integrating the contextual factors that determine heterogeneous effects. We build on [ 4] model and incorporate heterogeneity through random coefficients, along with fixed effects in a system-of-equations estimation. We use a feasible generalized least squares (FGLS) approach to test how incumbent software–specific factors influence the new product effects. Following [29], we apply these methods to a video game data set, comprising monthly sales of 13,064 games (software) and 19 consoles (platform) over 1995–2019, as well as the dates of software introductions and advertising expenditures. These recent data enable us to report current effect sizes and establish timely evidence for this rapidly evolving industry.
Overall, we contribute to research in platform markets by studying the relative impact of new software products on incumbent software sales. We report empirical generalizations for effect sizes for Paths A, B, and C, distinguishing direct from indirect impacts. With our holistic approach, we are then able to quantify the total (net) impact of an entrant by aggregating the effects of each path. In determining effects, we add context by identifying characteristics of entrants and incumbents that moderate each path. Drawing from the sensations–familiarity theoretical framework ([25]), which fits well for hedonic software (e.g., apps, video games), we operationalize software characteristics by capturing whether the software is a superstar or member of a franchise.
We find that the direct impact (Path A) of an entrant usually, but not always, results in cannibalization, depending on its characteristics. When entrants are superstars and/or members of a franchise (vs. not a superstar or franchise member—"standard software" hereinafter), they directly cannibalize incumbents more. The incumbents' characteristics offer little protection against such cannibalization. We find that new superstars significantly increase new platform adoption through the indirect Path B, which benefits incumbent superstar software and franchise members more than standard software (Path C). By combining both the indirect and direct impacts, we determine the net overall impact: Standard entrants hurt all incumbents, but a new superstar can produce a net positive halo, depending on the context. For example, a 1% increase in entrants with both superstar and franchise status drives a.0207% net increase (direct + indirect) in the sales of incumbent franchise software; the direct cannibalization of incumbent sales (via Path A) of −.0130% is overcome by a halo effect from new platform adopters (via Paths B and C) that indirectly increases incumbent sales by.0337% (.0207 = −.0130 + .0337). We show managers how to estimate both direct and indirect impacts of different types of new software based on the specific makeup of a firm's portfolio. We also show how these estimates translate to financial outcomes. We provide new insights for platform markets and new product research, while extending the sensations–familiarity framework.
To connect our findings to extant literature, we ground our conceptualization in network effects research. We then outline relevant theory from which we identify key variables.
Indirect network effects in systems markets are of enduring interest to scholars (e.g., [12]; [45]; [48]). Systems or platform markets often comprise hardware, such as a computer, game console, or smartphone, and related hardware-compatible software, such as computer programs, video games, or apps. A systems logic applies to digital platforms too (e.g., Netflix); so, to ensure the broad applicability of our study logic, we use the terms "platform" (vs. hardware; [47]) and "software."
Platform markets create both direct and indirect network effects. Direct network effects arise when the value to a user increases with the number of other platform users; for example, a multiplayer online game with many fellow players is more desirable ([33]). Indirect network effects exist if the number of platform users entices software producers to create new offerings for the platform. As the software portfolio improves (e.g., quality, variety), so does the attractiveness of the platform to users ([12]). Indirect effects apply to all members of the system: platforms need an attractive software portfolio to attract customers, and software providers prefer to develop software for platforms with large installed bases. In the vibrant research dedicated to these markets, briefly summarized in Table 1 (for a more expansive list, see Table W1 in the Web Appendix), we know of no efforts to calculate the extent of the total impact (direct and indirect) of new software entrants on sales of incumbent software.
Many platform markets exist for entertainment products (e.g., video games, movie streaming services, apps), so we rely on a theory that predicts consumer responses to hedonic products, namely, the sensations–familiarity framework ([25]). However, a similar logic can apply in nonentertainment settings (e.g., [38]). According to this theory, a person consumes hedonic products for pleasure (vs. functional benefits) that can be derived from the very act of consumption. This consumption act (e.g., playing a game) generates emotional responses (e.g., happiness, melancholy) and cognitive responses, which define people's mental representations of the experience. [ 6]) argue that consumers combine these responses to form holistic product judgments (vs. evaluating the attributes piecemeal).
The sensations–familiarity framework indicates that both sensations and a sense of familiarity can each drive emotional and cognitive responses that underlie judgment. Sensations are physiological (vs. purposive) responses, felt as a sense of arousal when nerves are stimulated and hormones (e.g., dopamine) released. Humans become satiated easily and prefer novel, multidimensional sensations, which leads to an innate desire to seek out rich, new sensations instead of repeating the same one ([35]). Familiarity refers to a consumer's sense of connectedness to a product (or its elements). Built through prior exposure, when a new product includes elements that are familiar, the sense of familiarity triggers memories and emotions that transfer to the new product ([ 9]). Familiarity also enhances processing fluency by helping consumers quickly make sense of new products ([39]).
Sensations and familiarity thus form a delicate balance. A new product that is too familiar (few new sensations) seems stale and unappealing. A new product that is too novel (stark new sensations with little link to the familiar) can overwhelm consumers by the sheer intensity of sensations (e.g., never-ending explosions in a Michael Bay movie; [25]). Thus, new product managers introducing hedonic products aim to provide the "right" levels of sensations and familiarity to maximize the products' attractiveness.
The variables we derive from this framework in turn reflect familiar elements of prior products and/or promise rich new sensations. For video game software, the task is to offer enough familiarity to connect with consumers (e.g., familiar worlds, product designs, characters) while also providing new sensations to arouse consumers (e.g., exciting new patterns of play, beloved characters in new worlds, novel designs) ([25]). Thus, we focus on two variables with distinct influences in extant studies. First, software products (e.g., video games, apps, movies) of extraordinarily high quality offer rich, desirable sensations. We thus operationalize the sensations factor as "superstar software," defined as "software titles of exceptional high quality" that often "possess unique and superior attributes" ([ 8], pp. 88–89). Superstars may achieve high payoffs, but this term refers explicitly to product quality, unlike the terms "blockbusters" or "hits," which reflect sales volume known only after product release ([ 8]). Second, we operationalize familiarity as software that is part of a franchise (e.g., Super Mario Bros., Star Wars). Extensions of a franchise leverage the influence of familiarity, but they also offer some new sensations and contexts (e.g., Super Mario Bros. 2, The Empire Strikes Back). By integrating elements of an existing brand into additional products, "the resulting set of products, in its entirety, then constitutes the 'franchise'" ([25], p. 429). Accordingly, software that is part of an established franchise should be desirable to consumers ([40]), and real-world evidence consistently shows that franchise products (e.g., sequels, prequels, reboots/remakes) often outsell similar, nonfranchise products ([33]). Both superstar and franchise variables have appeared in prior studies, but we do not know of their use to determine how new product entrants affect incumbent sales. Next, we explicate the roles these variables play in the three paths of Figure 1.
In Path A—the direct impact of new software entrants on sales of incumbent software—we are conceptualizing the behaviors of existing platform owners. In this case, a new software entrant is a potential substitute for incumbent software ([14]); they all compete for the consumer's (i.e., platform owner's) limited budget ([18]). Thus, to accurately quantify Path A, we should consider not only average effects but also how effects differ according to the characteristics of both the new entrant and the incumbent.
The characteristics of the new entrant should be particularly powerful, as the effects suggested by the sensations–familiarity framework may be most influential when consumers encounter new stimuli. For example, when they first play a new video game, the sensations offered are, to some degree, new to the world; if these novel sensations also are rich, as in the case of superstar software, they likely exert the strongest impacts on consumer preference. However, product newness also evokes uncertainty, so familiarity likely shapes consumers' responses as well. Being part of a franchise offers a comforting connection to known objects, such that relevant memories and emotions can transfer to the new entrant. Thus, we expect characteristics of both superstar and franchise status to increase the direct cannibalization caused by a new entrant.
In Path A, which reflects the consumer's choices between the new entrant and the incumbent, the characteristics of the incumbent software might matter too, but their power is less clear. At first glance, the rich sensations offered by a superstar incumbent might enhance its attractiveness and provide some level of protection against cannibalization by standard new entrants. Likewise, the familiarity benefits of being part of a franchise should leave the incumbent from a franchise (vs. nonfranchise) less susceptible to cannibalization by new entrants. However, prior applications of the sensations–familiarity framework have focused almost exclusively on new products ([ 5]), so it is not entirely clear how the findings generalize to incumbents. Being a superstar or part of a franchise may not provide incumbents with protection from competition. First, the specific sensations offered by this incumbent would be new to consumers who had yet to purchase it, but they are not new to the world. Even new purchasers likely have gained some predictive information about them, such as through word of mouth from other consumers with experience. Second, word of mouth about these experiences reduces not only novelty but also the level of uncertainty for new consumers, which might attenuate the familiarity benefit of being a franchise incumbent.
It is well known that new software influences sales of platform devices (for a review, see [41]]). We test this path to ensure a complete model and to isolate the indirect impact of new software entrants on sales of incumbent software by stimulating platform sales (Paths B and C). However, we differ from extant literature by focusing on the impact of new software entrants (vs. software stock) on new platform adopters (vs. installed base). That is, we expect software to increase platform adoption, consistent with traditional network logic, which may benefit incumbents, but we also test for distinct effects of new entrants. For example, because new superstars create excitement, they might attract increased attention and more new platform adopters ([ 8]; [21]). Franchise video games also are disproportionately attractive to consumers, so they prompt higher sales ([16]; [33]), and new franchise software might draw more new adopters to the platform.
Whereas Path A taps how a new software entrant competes directly with incumbent software among existing platform owners, Path C represents sales of incumbent software generated by new platform adoptions and thus reflects consumers' incumbent-versus-incumbent choices upon adopting the platform—specifically, whether and which incumbent offerings to buy. We thus calculate how the influx of new adopters affects demand for incumbent software (Path C), independent of the direct competitive effect of new software entrants.
Traditional cannibalization viewpoints frame a market as a zero-sum game with a finite number of customers. But in platform markets, a new product can help other products by driving increased traffic ([18]; i.e., product spillover effects; [ 1]). If a new software entrant induces new consumers to adopt the platform, they might buy other software from the existing portfolio. For example, if PlayStation loyalists wanted to play Super Smash Bros. Ultimate, released on December 7, 2018, and therefore bought a Nintendo Switch console to gain access to it, the consumers likely bought other Switch games to make full use of the console's capabilities (e.g., on-the-go gaming). The players also might try to "backfill" a collection of software favorites, played previously on the PlayStation; such cross-platform adoption is common for entertainment products ([30]).
The notion that platform sales lead to software sales is foundational to network effects theory, but we extend this traditional view. That is, our argument is not about new platform adopters in general, but on the purchases of new adopters attracted by the release of new software. We study Path C as part of our novel effort to understand the totality of Paths B and C. Further, we know of no other study that measures whether new platform adopters influence incumbent software sales immediately, that is, in the same month as platform adoption, when new adopters' budgets are likely depleted from buying the platform and new software entrant. Studies that span the platform's entire installed base cannot specify how platform adoption affects software sales immediately, because they lump existing platform owners in with new adopters. Assessments of software stock overall, rather than individual incumbent software, also cannot determine which type of software (e.g., superstar, franchise) benefits the most from greater platform adoption.
For Path C, we again apply the sensations–familiarity framework: for choices between two or more incumbent products, superstars and franchise software are more attractive than standard incumbents, so their sales may benefit most from new platform adopters.
Finally, to quantify the total net impact of a new software entrant on sales of incumbent software, we combine the direct (Path A) and indirect (Paths B and C) impacts. We want to understand in what circumstances a positive halo effect (i.e., increasing the number of platform adopters who subsequently buy incumbent software) is enough to overcome the direct cannibalization effect caused by increased competition.
On the one hand, a new product could be so keenly attractive that it cannibalizes sales of the incumbent ([46]), despite the overall increase in platform customers ([14]). With limited budgets, consumers cannot purchase an infinite amount of software; in turn, the cannibalization argument suggests that new software entrants prompt customers to choose them instead of an incumbent, resulting in a negative net impact on sales of incumbent products. On the other hand, the new entrant might attract enough new adopters, who subsequently purchase incumbent software, that it overcomes the direct competitive effects, resulting in a positive net impact. We note that [24] find that platforms with newer customers sell more content.
To reconcile these competing ideas, we use a decision rule suggested by [11]: to increase net software demand, the network effect of new platform users must be greater than the competitive effects among software products. With empirical tests, we can determine the sizes of both the direct cannibalization impact (Path A) and the indirect impact (Paths B and C) from new platform adopters, then calculate the overall net impact. But we also consider the characteristics of the new software entrant that might determine the strength of the direct and indirect impacts. Any new entrant increases the size of the software portfolio and thus should incrementally increase platform attractiveness and sales ([48]); however, to overcome cannibalization, the new software entrant needs to generate so many new platform adopters, who then purchase so much incumbent software that they create an overall net halo effect ([11]). It is likely that only software that is highly valued will attract a sufficient number of new platform adopters to overcome cannibalization. For example, prior research has demonstrated that superstar software produces a large surge in platform sales ([ 8]). Therefore, we measure if Paths B and C produce enough incremental sales to overcome Path A, and if this total effect differs based on superstar and franchise status.
Product-level competition ([18]) implies cannibalization due to substitution ([14]). Well-established factors that affect substitutability include sequel status, genre, price, exclusivity, and advertising ([25]; [33]; [42]). We control for these either directly or through use of fixed effects.
Our empirical context is the console-based video game industry, a large and significant part of the economy. Games for video game consoles generate more than $34 billion in annual sales. The entire industry is larger, including over $32 billion earned annually by computer-based and online games; mobile-device games earn over $70 billion annually ([51]).
Our setting is generalizable to other networked markets and is ideal for exploring our research questions for several reasons. First, studies in marketing (e.g., [ 8]; [29]) and related fields (e.g., [13]; [23]) use video game data to test platform market theories because video games are "a classic example of a high-tech networked market" ([20], p. 284). Second, managing product introductions is an important decision in networked industries ([31]), especially for video game firms ([ 8]). These markets observe frequent new product introductions and offer the ability to observe many products from inception to decline ([10]). Third, games are not natural complements; owning one game does not increase the utility of owning another, reducing confounds when examining cannibalization and halo effects. Fourth, characteristics in gaming (e.g., superstar status, franchise membership) map directly to the variables suggested by the sensations–familiarity framework ([ 8]; [25]; [30]).
We obtain monthly quantity and revenue data for 8,470 unique games and 19 consoles on the U.S. video game market from January 1995 through June 2019 from the market research firm The NPD Group. Many games are released on multiple consoles, creating 13,064 software/platform combinations. We structure the data as an unbalanced panel with 698,703 software/platform/month observations. The average console is on the market 100.37 months, for 1,907 platform/month observations. We match NPD data to advertising data from Kantar Media. These data include U.S. radio, television, cinema, online, outdoor, and print advertising spend for each game and each console in each month. We aggregate and obtain advertising expenditure at the software/platform/month level for games and at the platform/month level for consoles.
We classify software as new in the first month we observe sales on the platform. Software is an incumbent any time after its first month on the platform.
We operationalize familiarity with game franchise status. NPD gives a game's franchise affiliation (e.g., Call of Duty) when applicable. This identifies the initial entry in the franchise along with any extensions. We use this to classify whether incumbent or new software belongs to a franchise. Note that the first entry in a franchise is not a franchise when it is introduced; rather, it becomes a franchise when the first extension is introduced. Overall, 62.5% of software/platform observations are connected to a franchise in some way: 17.0% are first entries that become part of a franchise when an extension is introduced, and 45.5% are franchise extensions.
We use superstar status to identify software that provides rich sensations. We assess superstar status using critic evaluation, similar to other game industry studies ([ 8]; [33]). Importantly, critics typically review and provide assessments before a game's release ([53]). This means that the quality measure is determined independently from sales, which lessens endogeneity concerns. We obtain data from Moby Games (https://www.mobygames.com/), which has track record in video game research (e.g., [15]). Moby Games provides a Mobyrank, or average critic rating, for each software/platform observation. Mobyrank ranges between 0 (low) to 100 (high); we classify superstars as games with Mobyrank ≥ 90, similar to other studies ([ 8]; [21]; [26]). This operationalization has external validity; it is a popular industry cutoff as game-developer bonuses are often tied to achieving average critic scores of 90+ ([49]). We use superstar-franchise to label software that are both superstars and part of a franchise. In total, 2.427% of software/platform observations involve superstars, similar to other video game studies ([ 8]; [21]); 1.85% are superstar-franchise; and.57% are superstars that are not a part of a franchise.
For Path B, our goal is to estimate the effect of new software entrants on new platform adopters. We log-transform the continuous variables when estimating our model because we compare new platform adopters (and software sales) among a disparate set of platforms (and software) that have differing sales volumes ([ 8]). We operationalize new platform adopters as the natural log of unit sales of platform j at time t ( ) using each platform's monthly sales.
, , , and are the natural logs of the count of all, superstar, franchise, and superstar-franchise software new to platform j at time t, respectively. Note that assessing the impact of superstar-franchise software with a classic interaction effect using and is incorrect because these are binary, software-level characteristics. Rather, superstar-franchise classification occurs at the software level before creating the aggregate variables at the platform level. In addition, we use mutually exclusive classifications of superstar, franchise, and superstar-franchise to aid in interpretation.
We include five controls that influence platform demand. First, the natural log of price of platform j in time t, , adjusted for inflation using the Consumer Price Index for All Urban Consumers (CPI-U) with January 2007 = 100. Second, we control for the stock of available software ([17]) for each group ( , , , ). Stock is measured using the natural log of the count of the active software catalog (i.e., software for platform j introduced prior to t with positive sales in the period). This enables us to distinguish the effect of the software entrant variables from software stock. Third, we use natural log of advertising for platform j in t, , adjusted for inflation using the CPI-U ([45]). Fourth, we include platform age ( ) as months since introduction, and its square, because evidence suggests that the effect of age on platform demand is nonlinear ([26]; [30]). Fifth, month and year fixed effects address month-specific (e.g., holidays) and year-specific (e.g., economic shocks) trends.
Finally, we add 1 before logging the software entrant, software stock, and advertising variables to ensure that the log-transformation is defined when the level is 0.
Previous research finds that platform sales are influenced by lagged realizations of independent variables ([ 8]), so we use a dynamic model where depends on previous values ([52]):
Graph
( 1)
with
Graph
and represent coefficients with their respective dummies for the month and year in time t. We find the appropriate lag length (P = 1 in our case) by experimenting with longer lags and eliminating those that are insignificant ([19]). Including lags allows changes in variables to affect future periods; the coefficients are current-period effects; multiplying by gives total long-term effects ([19]).
Our log-log specification means that the coefficients on the software entrant variables are elasticity estimates. Specifically, captures the elasticity of platform adoption with respect to entry of standard software. The mutually exclusive classification means that , , and are the moderating effects for superstar entrants that are not part of a franchise, franchise entrants that are not superstars, and entrants that are both superstars and part of a franchise, respectively.
We also include platform fixed effects, , to control for time-invariant, platform-specific heterogeneity (i.e., individual console characteristics) that impacts adoption rates. We identify coefficients from within variance given the fixed-effects approach. We display panel-level descriptive statistics, which show measures of within variance for each variable, in Web Appendix Table W2. Table 2, Panel A, summarizes variables and definitions for the platform equation; it also includes variables that we define when discussing endogeneity.
Graph
Table 2. Select Variable Names and Definitions.
| Variable Name | Definition |
|---|
| A: Platform Adoption Estimation |
| Natural log of unit sales of platform j at time t. |
| , , , | Natural log of the counts of all, superstar, franchise, and superstar-franchise (i.e., both) software entry on platform j in time t; classifications of superstar, franchise, and superstar-franchise are mutually exclusive. |
| a | Natural log of average price of platform j in time t. |
| , , , | Natural log of the counts of all, superstar, franchise, and superstar-franchise software for platform j introduced prior to time t that are still active (i.e., positive sales in t). |
| a | Natural log of total platform level advertising expenditure for platform j in time t. |
| Platform age at time t measured in months since platform j was first on the market. |
| , | Month fixed effects, year fixed effects. |
| Installed base of platform j at time t (i.e., cumulative sales prior to time t). |
| Average software unit advertising on competing platforms other than j in time t. |
| Average age of active nonsuperstar software for platform j introduced prior to time t. |
| , | The producer price indexes for (1) electronic computer manufacturing in time t and (2) magnetic and optical recording media in time t, respectively. |
| B: Addition Variables in Incumbent Software Sales Estimation |
| Natural log of unit sales of incumbent software i corresponding to platform j in time t. |
| , , , | Software entry as defined for the Platform Adoption Estimation but only counting software in a different genre than incumbent software i. |
| a | Natural log of average price of software i on platform j in time t. |
| Indicator: a franchise extension to software i is introduced on platform j at time t. |
| , , , | Previous software introduction as in the Platform Adoption Estimation but recalculated to exclude software i. |
| , , , | Previous software introduction as defined for the Platform Adoption Estimation but recalculated for software in a different genre than incumbent software i. |
| Natural log of installed base of software i on platform j at time t. |
| a | Natural log of total software-level advertising expenditure for software i in time t. |
| Software i's age on platform j in time t in months since software entry on the platform. |
| Period fixed effects (i.e., month fixed effects interacted with year fixed effects). |
| , , | Indicators: software i on platform j is superstar, franchise, or superstar-franchise; classifications are mutually exclusive. |
| Genre fixed effects. |
| Avg. ln of software price for software released on platform j prior to software i when they were the same age as software i in time t (e.g., if software is six months old, then the variable is calculated from the prices of previously released software when these products were six months old). |
| Avg. ln of software advertising expenditure on software released on platform j prior to software i when they were the same age as software i in time t (e.g. if software is six months old then the variable is calculated from advertising expenditure of previously released software when these products were six months old). We set = 0 when = 0. |
| Percentage of franchise software introduced on platform j in the same genre as software i prior to time t out of all software introduced in the same genre as software i prior to time t. |
| Natural log of the number of franchise extensions introduced in the same genre as incumbent software i on competing platforms other than platform j in time t. |
1 aCorrected for inflation using the CPI-U, with January 2007 = 100.
For Paths A and C, our goal is to estimate the effect of new software entrants and new platform adopters on sales of incumbent software. We operationalize sales of incumbent software as the natural log of sales of software i, corresponding to platform j, at time t ( ). We measure all incumbent software on the market at time t that meet two criteria: ( 1) the software has observed sales in t and ( 2) the platform on which the software is available also has observed sales in t. The first condition ensures that we measure only active software. The second ensures that we observe the indirect (through platform sales; i.e., Path C) and direct (through competition; i.e., Path A) effects of new software on the sales of each incumbent.
To measure the direct impact of a new entrant via Path A, we include the natural log of the count of new entrants ( , , , and ) on console j at time t, operationalized as in the platform equation. However, as discussed in the conceptualization, a new entrant affects sales differently when the incumbent is in the same genre ([50]); competition will be less salient for incumbents in different genres. Thus, as controls we include the natural logs of the count of the entry variables in a different genre than incumbent software i (e.g., ).[ 5]
To assess the indirect impact of an entrant via Path C, we include the natural log of platform sales in time t, , the dependent variable from the platform equation.
We separately include the platform's installed base, operationalized as the natural log of cumulative sales of platform j prior to t ( ), to allow for differential effects of new platform adopters and previous platform adopters on incumbent software sales. We also include eight additional groups of control variables. First, we include the natural log of average price of software i on platform j in t adjusted for inflation using the CPI-U, ([30]). Second, a dummy variable ( ) takes the value 1 if any software entrant on platform j in t is a franchise extension to incumbent software i ([33]) to control for any differential effect a new entrant may have on incumbents from the same franchise (e.g., the introduction of Tomb Raider 2 may be more likely to cannibalize incumbent Tomb Raider compared with other entrants). Third, we use the natural log of the active catalog of software for platform j similar to the platform equation, but we exclude incumbent software i from the count (e.g., ). We add the natural logs of the active catalogs in a different genre than incumbent software i (e.g., ) to control for the effect of competition within and outside the genre. Fourth, we include software installed base, operationalized as the natural log of sales of software i on platform j prior to t, ([53]). Fifth, we include the natural log of advertising expenditure for software i in time t, adjusted for inflation using the CPI-U, . We also include to address any platform advertising spillover to software sales. Sixth, we utilize software/platform specific effects ( ) to account for software-/platform-specific variation. Seventh, period fixed effects address shocks common across all software/platform observations specific to any month from January 1995 to June 2019. Lastly, we measure the age of software i on platform j ( ) in months since introduction on the platform. Note that period fixed effects are perfectly collinear with the trend captured by software age. However, we include software age squared ( ) because game sales show an exponential decline with age ([10]). Period fixed effects are not perfectly collinear with because software enters at different periods throughout our time frame.
Similar to the platform equation, we add 1 before logging the software entrant, stock, and advertising variables to ensure that the log-transformation is defined when the level is 0.
We are interested in the effect of incumbent superstar, franchise, and superstar-franchise status on the impact of new software entry and new platform adopters. We let = 1 if software i on platform j is a superstar but not part of a franchise, 0 otherwise; = 1 if it is part of a franchise but not a superstar; = 1 if it is both a superstar and part of a franchise.
The functional form for the software equation is
Graph
( 2)
where
Graph
( 3)
for = , , , , and and
Graph
represents coefficients with their respective dummies for the period in time t. We include lags of the DV for a dynamic specification and follow [19] for the appropriate lag length (S = 1 in our case). As with the platform equation, the mutually exclusive classification of software along with the log-log specification means the coefficient on is the elasticity of incumbent software sales with respect to software entry; the coefficients on the other software entry variables are moderation effects. Similarly, the coefficient on is the elasticity with respect to new platform adopters.
Importantly, including the new entrant variables and new platform adopters in Equation 2 means we can interpret their impacts holding the other constant. The effect of the new entrant variables via Path A is their impact on incumbent software sales independent of new platform adoption (i.e., it represents sales to prior platform adopters). The impact of new platform adopters via Path C represents incumbent software purchases by those who just adopted the platform. Connecting Path C with Path B means that Path B–C represents the impact on incumbent software sales from new adopters spurred by new software entrants.
Note the relationship between Equations 2 and 3. The coefficients , , , , and from Equation 2 are random and allow for heterogeneous effects of the entry variables and platform adopters on each incumbent. Equation 3 specifies that the random coefficients are determined by the incumbent characteristics ( , , ), software genre fixed effects, and platform fixed effects at the software level ( ).[ 6]
Our mutually exclusive classification for entrant and incumbent software means the constant in the estimation of using Equation 3, , is the base impact that standard software entry has on incumbent software sales via Path A. The constants in the estimates for the other entry variables ( , , ) show how the base impact is moderated when the new entrant is a superstar, franchise, or superstar-franchise software. The coefficients on , , and capture the moderating effect of the incumbent characteristics.
Similarly, the constant in the estimation of is the base impact of new platform adopters on standard incumbent sales via Path C; the coefficients on incumbent characteristics are moderating effects. Subsequently, we discuss how we combine estimates from all equations to obtain the overall net (direct + indirect) impact. We display panel-level descriptive statistics in Web Appendix Table W3. Table 2, Panel B, presents variable names and definitions.
Our estimation framework addresses several issues. First, platform and software sales form a system of equations. New platform adoption and incumbent software sales are determined simultaneously; platform sales also affect the market potential of incumbent software. We jointly estimate Equations 1 and 2 to address this issue and improve efficiency by taking into account the likely correlation of error terms.
Second, we allow for parameter heterogeneity in Equation 2 with random coefficients at the software/platform level. Unfortunately, time-invariant software-/platform-specific characteristics (e.g., software exclusivity, quality, release date) likely correlate with time-variant variables (e.g., software price, advertising) so a traditional hierarchical linear approach, where is modeled as random, will produce biased estimates. To address this, we use fixed effects for along with the method from [ 4] to incorporate random coefficients. However, we expand their method to consider a system of equations with endogenous variables. The Web Appendix includes the derivation along with a detailed step-by-step guide to implementation.
We recover unbiased estimates of , , , , and for each software/platform panel as part of the procedure. We then run auxiliary regressions of Equation 3 to find the impact of incumbent software characteristics on the random coefficients. [52] notes that a more efficient FGLS estimator is available when regressing heterogeneous coefficients in random parameter models like ours. We derive and use the FGLS estimator for Equation 3 (for details, see the Web Appendix).
There are several endogeneity concerns in the platform and software equations. We use instruments that have a track record in the literature (e.g., [17]; [27]) or are derived from best practice ([22]; [36]). We give a detailed description of endogeneity concerns, instruments, and first-stage estimations in the Web Appendix (Tables W4 and W5). We emphasize that the usual diagnostics are always satisfied (e.g., first-stage F-stats > 10; insignificant Hansen's J). We briefly discuss the validity of key instruments used to identify our endogenous variables.
First, the software entry variables are endogenous in the platform equation. A platform becomes more attractive to software providers when more consumers adopt it; the platform becomes more attractive to consumers as it introduces more software ([48]). The theory of indirect network effects suggests platform installed base as a valid and relevant instrument. Platform installed base satisfies the exclusion restriction because its impact on new platform adoption is indirect, through its influence on software provision ([12]; [30]).
Second, platform price and advertising are likely endogenous as managers consider unobservables when choosing strategies. We include producer price indexes (PPIs) from related industries to capture cost shocks that impact pricing decisions ([17]; [21]). PPIs are valid because they are not set considering unobservables that impact platform demand ([36]). We interact the PPIs with platform characteristics and other instruments to aid in identification and create platform-specific instruments ([43]).
In addition, we use the average age of nonsuperstars in the active catalog to instrument for price and software entry, similar to [27]. Firms likely compensate for old catalogs with lower prices and new software. This instrument is valid if consumers are unaware of all nonsuperstar release dates when considering platform purchase, which is likely because consumers are typically unfamiliar with nonsuperstars ([ 8]).
We use the average advertising expenditure of each software available on competing platforms as an instrument for platform advertising. This is valid because individual software on competing platforms is unlikely to take into account unobservables that impact focal platform demand in a coordinated and systematic fashion when choosing advertising strategy. It is relevant because it is related to software-level advertising on the focal platform ([36]) and captures relative investments in software versus platform advertising.
Time-invariant software-/platform-specific characteristics (e.g., software exclusivity, quality, entry timing) likely correlate with other variables in Equation 2 (e.g., price). Fixed effects address this, as these characteristics do not change over the software's life. In contrast to the new platform adopters estimation, traditional fixed effects produce Nickell bias ([37]) in dynamic models when N ( = 13,064) is large relative to T ( = 53.48, on average).[ 7] Instead, we first-difference Equation 2 and instrument with lagged levels of the dependent variable ([ 3]).
We do not consider new platform adopters or software entry to be endogenous. It is unlikely that all platform adopters in a particular period act in a coordinated fashion and account for unobservables of a particular incumbent software. Similarly, it is implausible that managers coordinate software entry systematically and consider all the individual software/platform/month unobservables that affect the performance of all software already on the market.
However, software price, software advertising, and likely are endogenous. Price and advertising strategies may account for unobservables that impact software sales; the launch of a franchise sequel may relate to unobservables of franchise members on the market.
We exploit the panel nature of the data to construct instruments for price and advertising using values from other software on the platform when they were the same age as the focal software. For example, for advertising, if software is six months old, we calculate the variable from the advertising expenditure of previously released software when it was six months old. These instruments are relevant because they capture pricing and advertising trends based on software age. Similar instruments have been used in studies of movies ([28]) and games ([30]); they are valid, because it is unlikely that managers choose their software pricing and advertising strategies considering future unobservables of yet-to-be-released software.
We use two instruments for that capture trends in franchise extension introduction for similar software. First, we use the percentage of franchise software introduced out of all software introduced on the platform in the focal software's genre prior to the current period. This is relevant as it captures past trends in franchise entrants for similar software; it is valid as it is unlikely that managers coordinate franchise software entry in a systematic fashion accounting for future unobservables for all other software in the genre. Second, we use the natural log of the number of franchise extension introductions in the current period on different platforms but in the same genre as the focal software. This is valid, as it is unlikely that software on different platforms takes into account unobservables that influence focal software performance in a systematic and coordinated fashion; it is relevant because it captures franchise introduction intensity of similar software in the current period.
We assess Paths A, B, and C using the joint generalized method of moments estimation of the platform-adoption and incumbent-software sales equations, along with the heterogeneous-coefficients equation. (See estimated variance/covariance of the heterogeneous coefficients in Web Appendix Table W6.)
We first turn to the estimation of Path A, which captures the direct competitive impact new entrants have on incumbents. The full results for estimating incumbent software sales (Equation 2) are shown in Table A1 in the Appendix. While Table A1 is useful for understanding the impact of the control variables and model diagnostics, the coefficients for the new entry variables are not as meaningful because they are simply the average of the heterogeneous effects and do not address our research questions. The full heterogeneous effects, showing the results of every combination of incumbent/new entry characteristic (i.e., higher-order interactions), are provided in Table A2 in the Appendix. To simplify our presentation, Table 3 shows key results from Table A2 relating to the base impact that standard new entrants have on standard incumbents. Panel A shows how entrant type moderates the base effect. Entrants with superstar status and/or franchise affiliation increase their competitive edge over incumbents. While entry by standard software negatively impacts incumbent sales (β = −.0110, p < .01), the effect is larger when the new software is a superstar (β = −.0074, p < .10), part of a franchise (β = −.0128, p < .01), or both (β = −.0058, p < .05). New entrants that exhibit rich sensations and/or leverage familiarity hurt standard incumbents' sales more. However, familiarity in an entrant does not augment the impact of rich sensations; the effect of a superstar entrant is not significantly different from a superstar-franchise entrant (z = .3127, p > .10).
Graph
Table 3. Path A: Direct Impact of Software Entry on Incumbent Software Sales.
| A: Moderation of Software Entry on Standard Incumbent Software Sales by New Software Characteristics |
| % Change in Standard IncumbentSoftware Sales Due to a 1% Increase in New Software Entry |
| Entry by standard software | −.0110*** (.0020) |
| Moderation if entry by superstar software | −.0074* (.0043) |
| Moderation if entry by franchise software | −.0128*** (.0012) |
| Moderation if entry by superstar-franchise software | −.0058** (.0029) |
| B: Moderation of Standard Software Entry on Incumbent Software Sales by Incumbent Software Characteristics |
| % Change in Incumbent SoftwareSales Due to a 1% Increase inStandard Software Entry |
| Impact on standard incumbent | −.0110*** (.0020) |
| Moderation if incumbent is superstar software | −.0208 (.0154) |
| Moderation if incumbent is franchise software | .0013 (.0029) |
| Moderation if incumbent is superstar-franchise software | .00043 (.0064) |
- 2 *p < .1.
- 3 **p < .05.
- 4 ***p < .01.
- 5 Notes: SEs robust to platform clustering are in parentheses.
Panel B shows how incumbent characteristics moderate the impact of a standard entrant. The negative impact of a standard entrant is not attenuated if the incumbent is a superstar (β = −.0208, p > .10), part of a franchise (β = .0013, p > .10), or both (β = .00043, p > .10). Incumbents that possess rich sensations and/or familiarity do not have additional protection from direct cannibalization caused by new entrants. In addition, the combination of rich sensations and familiarity in an incumbent does not offer extra protection compared with rich sensations alone; the impact on superstar incumbents is not significantly different from the impact on superstar-franchise incumbents (z = 1.1209, p > .10).
The combination of Paths B and C is the indirect impact of software entry on incumbent software sales via new platform adoption. Considering Path B first, we show the platform adoption estimation (Equation 1) in Table 4. Consistent with prior research (e.g., [ 8]), we find that only new entrants with superstar status affect platform adoption. While entry by standard software does not have a significant impact (β = −.3935, p > .10), the effect is positively moderated when entrants are either superstar (β = 2.7717, p < .01) or superstar-franchise (β = 5.3839, p < .01). Combining the base and moderating effects, we find that new superstars (β = −.3935 + 2.7717 = 2.3782, p < .01) and new superstar-franchise (β = −.3935 + 5.3839 = 4.9904, p < .01) entrants positively influence platform adoption. The coefficient on superstar-franchise entrants is greater than the coefficient on superstar entrants (z = 1.7831, p < .10), which suggests that franchise status augments the effect of new superstars. However, franchise status alone does not increase platform adoption (β = −.2700, p > .10). Our results suggest that only new entrants with superstar status will impact new adopters enough to spur an indirect halo effect on incumbents.
Graph
Table 4. Path B: Impact of Software Entry on Platform Adoption.
| Coefficient | SE |
|---|
| −.3935 | .3308 |
| 2.7717*** | .9727 |
| −.2700 | .4524 |
| 5.3839*** | 1.3533 |
| −.3569* | .2078 |
| .0420* | .0224 |
| .9166*** | .0195 |
| .7066*** | .2006 |
| .2441*** | .0456 |
| −.7584*** | .2156 |
| .3879*** | .0947 |
| Platform, month, year, platform age, and age2 effects | Included |
| Hansen's J | 21.0896(p = .3319) |
| N | 1,828 |
- 6 *p < .1.
- 7 **p < .05.
- 8 ***p < .01.
- 9 Notes: SEs are robust to platform clustering; DV = ; first-stage F-statistics for endogenous variables , , , , , and are all >10. First-stage estimates are shown in Web Appendix Table W4.
Path C considers the impact of new platform adopters on incumbent software sales. Table 5 displays the results of the heterogeneous coefficient estimation for new platform adopters (from Table A2). Standard incumbents benefit from new platform adopters (β = .0024, p < .10). However, incumbents who are superstars (β = .0340, p < .01), part of a franchise (β = .0043, p < .05), or both (β = .0153, p < .01) benefit more. Furthermore, superstar status matters more than franchise status: the impact of new platform adopters on superstar and superstar-franchise incumbents is larger than franchise incumbents ( = 6.0213, p < .05). However, franchise status does not augment the benefit to incumbents with superstar status. Our results suggest that incumbent superstars benefit more from new platform adopters when they are not part of a franchise (z = 1.7414, p < .10). While sensation-rich and/or familiar incumbents are not protected more from the direct competitive effects (Path A), these same incumbents benefit indirectly from software entry as they attract more interest from new platform adopters.
Graph
Table 5. Path C: Impact of New Platform Adopters on Incumbent Software Sales.
| % Change in Incumbent Software Sales Due to a 1% Increase in New Platform Adopters |
|---|
| Impact on standard incumbent | .0024* (.0013) |
| Moderation if incumbent is superstar software | .0340*** (.0116) |
| Moderation if incumbent is franchise software | .0043** (.0022) |
| Moderation if incumbent is superstar-franchise software | .0153*** (.0049) |
- 10 *p < .1.
- 11 **p < .05.
- 12 ***p < .01.
- 13 Notes: SEs robust to platform clustering are in parentheses.
We present the calculations for the overall net impact for every incumbent/new entrant combination in Table 6, Panel A, with a summary in Table 6 Panel B. Next, we walk through an example of how Table 6, Panel A, is calculated.
Graph
Table 6. Combining Paths A, B, and C.
| A: Net Overall Impact (Direct + Indirect) of New Software Entry on Incumbent Software |
|---|
| Entry By ... |
|---|
| Impact On ... | Standard | Superstar | Franchise | Superstar-Franchise |
|---|
| Path A: Direct Impact from Increased Competition |
| Standard incumbent | −.0110***(.0020) | −.0184***(.0048) | −.0238***(.0024) | −.0168***(.0036) |
| Superstar incumbent | −.0318**(.0151) | −.1482***(.0359) | −.0275*(.0159) | −.0254(.0391) |
| Franchise incumbent | −.0097***(.0010) | −.0093***(.0014) | −.0076***(.0013) | −.0130***(.0020) |
| Superstar-franchise incumbent | −.0106**(.0052) | −.0699***(.0115) | −.0238***(.0061) | .0040(.0120) |
| Paths B and C: Indirect Impact from New Platform Adopters |
| Standard incumbent | −.0010(.0009) | .0058(.0036) | −.0016*(.0010) | .0121*(.0072) |
| Superstar incumbent | −.0143(.0130) | .0866**(.0416) | −.0242**(.0105) | .1817**(.0800) |
| Franchise incumbent | −.0027(.0023) | .0161***(.0059) | −.0045***(.0014) | .0337***(.0106) |
| Superstar-franchise incumbent | −.0070(.0062) | .0422**(.0188) | −.0118**(.0047) | .0885**(.0347) |
| Net Overall Impact = Direct + Indirect |
| Standard incumbent | −.0120***(.0022) | −.0127**(.0060) | −.0254***(.0025) | −.0047(.0081) |
| Superstar incumbent | −.0462**(.0199) | −.0616(.0549) | −.0516***(.0191) | .1563*(.0890) |
| Franchise incumbent | −.0123***(.0025) | .0068(.0061) | −.0121***(.0019) | .0207*(.0108) |
| Superstar-franchise incumbent | −.0176**(.0081) | −.0277(.0220) | −.0356***(.0077) | .0926**(.0375) |
| B: Summary of Net Overall Impact (Direct + Indirect) |
| Entry By ... |
| Impact On ... | Standard | Superstar | Franchise | Superstar-Franchise |
| Standard incumbent | Cannibalization | Cannibalization | Cannibalization | Net neutral |
| Superstar incumbent | Cannibalization | Net neutral | Cannibalization | Halo |
| Franchise incumbent | Cannibalization | Net neutral | Cannibalization | Halo |
| Superstar-franchise incumbent | Cannibalization | Net neutral | Cannibalization | Halo |
14 Notes: SEs in parentheses given by the linear and nonlinear restrictions embodied in the coefficient calculation ([19]). Cannibalization = negative and significant net impact; Net neutral = insignificant net impact; Halo = positive and significant net impact.
First, we calculate the total indirect impact (Paths B and C) by multiplying the elasticity of platform adoption to software entry (Path B in Table 4) with the relevant elasticity of incumbent sales to new platform adopters (Path C in Table 5). Consider the indirect impact of a superstar-franchise entrant on a franchise incumbent. For Path B, a 1% increase in superstar-franchise entry increases platform adoption by 4.9904% (see previous explanation); for Path C, a 1% increase in new platform adopters increases incumbent franchise software sales by.00676% (=.00242 + .00434 from Table 5, p < .01). Combining both, the full indirect impact of a 1% increase in superstar-franchise entry on incumbent franchise software sales is.0337% ( = 4.9904 ×.00676, p < .01).
Second, the direct impact is calculated by combining the relevant coefficients from the heterogeneous coefficient estimations. For example, the direct impact of standard software entry on franchise incumbents is −.00969% (= −.01103 + .00134 from Table A2, column 1, p < .01); this is moderated by −.00335% if the entrants are superstar-franchise software (= −.00575 + .00240 from Table A2, column 4, p < .01). Totaling, the direct impact of a 1% increase in superstar-franchise entry on incumbent franchise software is −.0130% (≈ −.00969 + −.00335, p < .05).
Third, we obtain the net overall impact by combining the direct and indirect impacts, which, for our example, equates to.0207% (=.0337 + −.0130, p < .10), a significant overall halo effect. For incumbent franchise software facing new superstar-franchise entrants, the indirect impact of greater sales from increased platform adoption outweighs the direct competitive effect.
Table 6 performs these calculations and displays results for each type of entrant on each type of incumbent, including higher-order interactions (e.g., superstar-franchise new entrant on a franchise incumbent). As firms' incumbent-software portfolios differ, Table 6 enables managers to estimate the direct, indirect, and total impact new entrants have on their own portfolio.
Importantly, we find significant net cannibalization of incumbents by new entrants without superstar status. The negative direct competitive effect of standard and franchise entrants (via Path A) outweighs any indirect effect (via Paths B and C). Incumbent franchise or superstar status does not offer protection from cannibalization due to competition with new nonsuperstars.
In contrast, the impact of superstar and superstar-franchise entrants on new platform adopters is large enough to meaningfully counteract their direct competitive effects. The impact of new superstar-franchise software results in a net halo effect for superstar, franchise, and superstar-franchise incumbents and a net neutral effect for standard incumbents. Further, while new superstars do not produce any instances of a net halo effect, the indirect impact completely offsets losses due to direct competition for all but standard incumbents.
So far, we analyze effects of entry in the current period. An advantage of dynamic models is they allow us to see how a current change in a variable impacts long-term results by working through the lags on the right-hand side. We show the long-term net overall impact of all entrant types on all incumbents in the Web Appendix, Table W7. Results are similar to those in Table 6; however, effect sizes are larger. The impact of new software entry on incumbents persists long after the introduction date, and the cumulative effect is large.
We run robustness checks for the Equation 3 results by including an indicator variable for exclusive software, dropping genre fixed effects, and dropping platform fixed effects. We also drop software/platform observations from the counts of available software if the observation is more than three years old ([26]) and estimate Equations 1–3. We detail these checks in the Web Appendix; results are similar to those presented previously. We also check single equation estimations to ensure that results are not driven by improved efficiency of joint estimation. The sign and significance level of all coefficients are the same.
By observing how thousands of new software entrants affect the sales of incumbent software, we find evidence of both cannibalization and halo effects, depending on the software attributes. We also establish specific effect sizes for the direct and indirect impacts using large-scale data.
Research in the domains of new product development and platform markets can use our findings. First, with regard to Path A, prior network market studies rarely consider the direct relationship of new software entrants with incumbent software. A common assumption is that more software stock increases the platform's installed base, which increases software sales. However, we highlight the need to incorporate competitive cannibalization effects; otherwise, the results likely overestimate the impact of new software. We also reveal contingencies that alter whether new entrants help or hurt incumbent software sales, drawing from the sensations–familiarity framework. Software characteristics moderate the effects; by applying the superstar and franchise variables in novel ways, we show that they help quantify a new entrant's impact.
Second, pertaining to Path B, which has been studied in various ways in the network effects literature, our findings move beyond investigating the impact of overall software stock on platform installed base, because we treat new software entrants as new products with unique impacts on market dynamics, not simply as increases in software stock. Furthermore, by accounting for new platform adopters separately from the past installed base of consumers, we show that new software entrants have different roles for facilitating new platform adoption.
Third, pertaining to Path C, we quantify the impact of new platform adopters on sales of incumbent software, above and beyond purchases by the platform's existing users. Table A1 shows that the impact of platform installed base on incumbent software sales is positive (β = .4359, p < .01). However, it is not comparable to the impact of new platform adopters on incumbent sales shown in Table 5, as a 1% increase in new adoptions represents a very different number of consumers than a 1% increase in platform installed base. We observe 168,758 new adopters on average in a month; the average platform installed base in a month is 18,261,152. A 1% increase in platform adoption is equivalent to a.009% (≈ 1,688/18,261,152 × 100) increase in platform installed base, resulting in a.0039% (≈.009 ×.4359) increase in incumbent software sales. Compared with results in Table 5, incumbent superstar (β = .0352, p < .01), franchise (β = .0028, p < .01) and superstar-franchise (β = .0138, p < .01) software benefits more from new adopters than an equivalent increase in platform installed base; standard incumbents (β = −.0015, p > .10) benefit less, though the difference is not statistically significant.[ 8]
Further, by accounting for new adopters attracted by new software entrants, we can test the totality of the indirect impact through platform adoption (Paths B and C). To the best of our knowledge, no study has examined and quantified this entire indirect path. Conventional wisdom suggests that new entrants help incumbents by increasing the platform's installed base, but we clarify that only new entrants with superstar status can drive platform sales so much that they result in a net positive impact (after accounting for direct cannibalization) on incumbent sales.
Fourth, we offer an extension to the sensations–familiarity framework. This theory highlights the need for a balance between sensations and familiarity, without specifying their interaction. By showing that a measure of sensations (i.e., superstar) interacts with a familiarity variable (i.e., franchise), we establish that researchers should account for the additional effects that can be created by combinations of sensations and familiarity, beyond their independent direct effects.
Fifth, we offer a new approach that accounts for the autoregressive process for platform and software demand. Using [ 4] random coefficients model and fixed effects to address heterogeneity in our panel, we extend that approach to incorporate a system-of-equations estimation while also dealing with endogeneity. We thus allow for heterogeneity in the impacts of new software entrants on incumbent software sales. Finally, by applying FGLS, rather than the less efficient ordinary least squares, we discern software-specific factors associated with halo and cannibalization effects. The Web Appendix has a step-by-step guide to implement this approach.
Our study shows managers how to measure new product performance holistically. Among existing approaches, there have been "few attempts to provide measures to quantify the effects of new products on the current product portfolio" ([46], p. 359). Practitioners often adopt ad hoc estimates of cannibalization (e.g., [ 7]; [34]). Without historical data, managers might turn to A/B testing, but that approach is expensive and also requires clean test manipulation. With our method, leveraging the elasticities in Table 6, managers can estimate the net revenue of a new entrant by including both halo and cannibalization effects, along with the new entrant's own revenue.
Here is a hypothetical, numerical example in which we calculate the impact of a new superstar-franchise entrant in the current month. The median number of superstar-franchise entrants in a month is zero; we consider an increase to one. Recall that we add 1 to the entry variables before taking the natural log; thus, one superstar-franchise entrant is a 100% increase in median superstar-franchise entry given the variable transformations (i.e., going from one to two instead of zero to one). Because a 1% increase in superstar-franchise entrants leads to a.0207% increase in franchise incumbent sales (Table 6), a 100% increase in superstar-franchise entrants leads to a 2.07% ( = 100 ×.0207) increase. Rounding from the descriptives in Tables W2 and W3, an incumbent franchise game earns ∼$179,000 per month, on average, and there are ∼235 franchise games on the market per month. Thus, the superstar-franchise entrant causes an additional $870,000 (≈ $179,000 × 2.07% × 235) in revenue for incumbent franchise games. We repeat this exercise for superstar and superstar-franchise incumbents because they also experience a significant impact from superstar-franchise entry (Table 6). The total financial impact on superstar, franchise, and superstar-franchise incumbents is ∼$1.3 million additional revenue in the month of introduction. Compared with the average of $5.9 million that new superstar-franchise software earns in its first month, the halo effect of a superstar-franchise entrant is 22% (≈ 1.3/5.9) of its own revenue. Thus, measuring only the superstar-franchise entrant's earnings underestimates the total impact.
Consider a similar exercise with a standard entrant. The median number of standard entrants is two. Going from two to three is a 33% increase given the variable transformations. Table 6 shows standard entrants have significant cannibalization effects; one standard entrant equates to a loss of roughly $224,000 in revenue from all incumbents. The average revenue earned by a standard entrant in its first month is roughly $525,000; this suggests that 43% (≈ 224/525) of the revenue earned is simply from cannibalizing incumbents. A manager of a standard entrant might overestimate profitability if considering only revenue from the entrant's own sales.
Moreover, because halo and cannibalizations effects are contextual, we show how to assess new entrants on a case-by-case basis, while accounting for specific characteristics of both the new product and the incumbent product portfolio. Managers of platforms or software firms can leverage our methods to assess the impact of a new product introduction; for managers of firms that provide platforms and software, we show how to estimate the holistic effect on revenues earned from sales of both platforms and incumbent software. Software-only firms can predict the impacts of their own new entrants and competitor entrants on their existing portfolio sales too.
Finally, platform growth strategists can look to the introduction of the right type of new software products to spur demand. Thus, it is not sufficient to recommend only the traditional viewpoints of increasing the installed base (direct network effects) or software stock (indirect network effects); instead, managers should encourage strategic introductions of the right types of software. To design such strategies, managers can apply our method to gain a holistic view of the likely performance effects of new software introductions, based on their portfolio makeup.
Limitations of our study suggest further research. First, we study one industry; studies should build on and confirm our findings' applicability in other contexts. The video game industry is similar to other platform industries, but our theory and findings most likely generalize to other settings in which customers ( 1) might favor one platform but use several platforms, ( 2) view specific software products as imperfect substitutes, ( 3) consume multiple software products that address similar needs (often hedonic vs. functional), ( 4) repeatedly consume favorite software products, and ( 5) eventually become satiated and seek variety. Accordingly, we anticipate that our findings might apply to markets for streaming platforms, apply somewhat to markets for ride-sharing platforms/drivers, but apply less so to markets such as health insurance and hospital networks.
Second, our econometric approach does not account for psychological mechanisms that underlie consumer behaviors in response to new entrants. The lack of process measures also is a common limitation of studies that use secondary data ([ 2]). Our theory development aligns with extant literature regarding why the observed effects occur, but additional research could test the psychological underpinnings of the behaviors we observe.
Third, we evaluate some important moderators in platform markets; further research could evaluate others. Studies might address other operationalizations of superstar software. Our method dichotomizes quality, but examining quality on a continuum could yield new insights.
sj-pdf-1-jmx-10.1177_00222429211017827 - Supplemental material for Halo or Cannibalization? How New Software Entrants Impact Sales of Incumbent Software in Platform Markets
Supplemental material, sj-pdf-1-jmx-10.1177_00222429211017827 for Halo or Cannibalization? How New Software Entrants Impact Sales of Incumbent Software in Platform Markets by B.J. Allen, Richard T. Gretz, Mark B. Houston and Suman Basuroy in Journal of Marketing
Graph
Table A1. Impact of New Software Entrants on Incumbent Software Sales.
| Coefficient | SE |
|---|
| a | −.0117*** | .0041 |
| a | .0007 | .0052 |
| a | −.0034 | .0037 |
| a | −.0040 | .0055 |
| a | .0062 | .0038 |
| .4105*** | .0058 |
| .0155*** | .0038 |
| −.0010 | .0066 |
| −.0023 | .0038 |
| .0051 | .0062 |
| –1.4301*** | .0803 |
| .2963 | .2435 |
| .1016 | .1536 |
| .0377** | .0176 |
| .0964 | .1337 |
| −.0135 | .0219 |
| .0020 | .1415 |
| −.0009 | .0170 |
| −.1920 | .1183 |
| .0192 | .0204 |
| −.4632*** | .0060 |
| .4359*** | .0285 |
| .0066*** | .0007 |
| .0016*** | .0002 |
| .0002*** | .00002 |
| Included |
| Arellano–Bond Tests: | AR(1) test | −40.6581***(p = .0000) |
| AR(2) test | −.5896(p = .5554) |
| Hansen's J | 21.0896(p = .3319) |
| N | 619,886 |
- 15 *p < .1.
- 16 **p < .05.
- 17 ***p < .01.
- 18 aEstimated mean (SE of estimate) of heterogeneous effects presented.
- 19 Notes: SEs are robust to software/platform clustering; DV = ; = first difference. First-stage F-statistics for endogenous variables , , , and are all >10. First-stage estimates are shown in Web Appendix Table W5.
Graph
Table A2. Incumbent Software Sales Estimation: Results of Heterogenous Coefficient Estimations.
| Path A | Path C |
|---|
| DV: | | | | | |
|---|
| DV meaning: | Impact of Standard New Entrant on... | Moderation If Entry by Superstar Software | Moderation If Entry by Franchise Software | Moderation If Entry by Superstar-Franchise Software | Impact of New Platform Adopters on... |
|---|
| Standard incumbent | −.0110***(.0020) | −.0074*(.0043) | −.0128***(.0012) | −.0058**(.0029) | .0024*(.0013) |
| Moderation if incumbent is superstar software | −.0208(.0154) | −.1090***(.0346) | .0172***(.0053) | .0122(.0376) | .0340***(.0116) |
| Moderation if incumbent is franchise software | .0013(.0029) | .0078(.0051) | .0149***(.0019) | .0024(.0044) | .0043**(.0022) |
| Moderation if incumbent is superstar-franchise software | .00043(.0064) | −.0519***(.0118) | −.0005(.0040) | .0204*(.0106) | .0153***(.0049) |
| N | 12,374 | 9,307 | 12,113 | 10,895 | 12,423 |
- 20 *p < .1.
- 21 **p < .05.
- 22 ***p < .01.
- 23 Notes: SEs robust to platform clustering are in parentheses. We include platform and genre fixed effects in each estimation. All results are FGLS with separate error covariance estimates for superstar, franchise, and exclusive incumbent software by each platform/year of software entry cluster (for details, see the Web Appendix). N reflects the number of software/platform coefficients; coefficients are only identified if there is variation in the variable within the panel.
Footnotes 1 Stefan Stremersch
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Online supplement:https://doi.org/10.1177/00222429211017827
5 NPD provides genre classification for each game in the data set, with 15 genres in total (Action/Platformer, Application, Arcade, Bundle with Multiple Genres, Fighting, Music and Rhythm, Narrative, Puzzle, Racing/Driving, Role Playing Games, Shooter, Simulation, Skill and Chance, Sports, and Strategy).
6 We recover the constant via the normal equations ([19]) after imposing the restriction that each set of fixed effects sums to zero.
7 The correlation between the lagged dependent variable and the average error in each panel does not go to zero as N goes to infinity ([37]). The average error in the panel becomes part of the transformed error term when fixed effects are eliminated through traditional demeaning. Thus, the use of fixed effects introduces endogeneity through the nonzero correlation of the lagged dependent variable and the transformed error term.
8 Coefficients are the difference between the impact of new adopters and platform installed base on incumbent sales found by summing the relevant coefficients in Table 5 and subtracting.009 × the coefficient on platform installed base in Table A1.
References Agarwal Rajshree , Bayus Barry L.. (2002), " The Market Evolution and Sales Takeoff of Product Innovations ," Management Science , 48 (8), 1024 – 41.
Allen B.J. , Dholakia Utpal , Basuroy Suman. (2016), " The Economic Benefits to Retailers from Customer Participation in Proprietary Web Panels ," Journal of Retailing , 92 (2), 147 – 61.
Arellano Manuel , Bond Stephen. (1991), " Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations ," Review of Economic Studies , 58 (2), 277 – 97.
Arellano Manuel , Bonhomme Stephane. (2012), " Identifying Distributional Characteristics in Random Coefficients Panel Data Models ," Review of Economic Studies , 79 (3), 987 – 1020.
Askin Noah , Mauskapf Michael. (2017), " What Makes Popular Culture Popular? Product Features and Optimal Differentiation in Music ," American Sociological Review , 82 (5), 910 – 44.
Belk Russell W. , Ger Guliz , Askegaard Soren. (2002), " The Missing Streetcar Named Desire ," in The Why of Consumption: Contemporary Perspectives on Consumer Motives, Goals, and Desires , Ratneshwar S. , Mick D.G. , Huffman C. , eds. London : Routledge : 98 – 199.
Bennett Dave. (2017), "3 Ways to Measure Cannibalization," Fast Casual (August 14) , https://www.fastcasual.com/blogs/3-ways-to-measure-cannibalization.
Binken Jeroen L.G. , Stremersch Stefan. (2009), " The Effect of Superstar Software on Platform Sales in System Markets ," Journal of Marketing , 73 (2), 88 – 104.
9 Bohnenkamp Björn , Knapp Ann-Kristin , Hennig-Thurau Thorsten , Schauerte Ricarda. (2015), " When Does It Make Sense to Do It Again? An Empirical Investigation of Contingency Factors of Movie Remakes ," Journal of Cultural Economics , 39 (1), 15 – 41.
Burmester Alexa B. , Becker Jan U. , van Heerde Harald J. , Clement Michel. (2015), " The Impact of Pre- and Post-Launch Publicity and Advertising on New Product Sales ," International Journal of Research in Marketing , 32 (4), 408 – 17.
Church Jeffrey , Gandal Neil. (1992), " Network Effects, Software Provision, and Standardization ," Journal of Industrial Economics , 40 (1), 85 – 103.
Church Jeffrey , Gandal Neil. (1993), " Complementary Network Externalities and Technological Adoption ," International Journal of Industrial Organization , 11 (2), 239 – 60.
Clements Matthew T. , Ohashi Hiroshi. (2005), " Indirect Network Effects and the Product Cycle: Video Games in the U.S.," Journal of Industrial Economics , 53 (4), 515 – 42.
Cohen Morris A. , Eliashberg Jehoshua , Ho Teck H.. (2000), " An Analysis of Several New Product Performance Metrics ," Manufacturing and Service Operations Management , 2 (4), 337 – 49.
Corts Kenneth S. , Lederman Mara. (2009), " Software Exclusivity and the Scope of Indirect Network Effects in the U.S. Home Video Game Market ," International Journal of Industrial Organization , 27 (2), 121 – 36.
Cox Joe. (2014), " What Makes a Blockbuster Video Game? An Empirical Analysis of US Sales Data ," Managerial and Decision Economics , 35 (3), 189 – 98.
Dubé Jean-Pierre H. , Hitsch Günter J. , Chintagunta Pradeep K.. (2010), " Tipping and Concentration in Markets with Indirect Network Effects ," Marketing Science , 29 (2), 216 – 49.
Ghose Anindya , Smith Michael D. , Telang Rahul. (2006), " Internet Exchanges for Used Books: An Empirical Analysis of Product Cannibalization and Welfare Impact ," Information Systems Research , 17 (1), 3 – 19.
Greene William H. (2011), Econometric Analysis , 7th ed. Upper Saddle River, NJ : Prentice Hall.
Gretz Richard T. , Basuroy Suman. (2013), " Why Quality May Not Always Win: The Impact of Product Generation Life Cycles on Quality and Network Effects in High-Tech Markets ," Journal of Retailing , 89 (3), 281 – 300.
Gretz Richard , Malshe Ashwin , Bauer Carlos , Basuroy Suman. (2019), " The Impact of Superstar and Non-Superstar Software on Hardware Sales: The Moderating Role of Hardware Lifecycle ," Journal of the Academy of Marketing Science , 47 (3), 394 – 416.
Grewal Rajdeep , Dharwadkar Ravi. (2002), " The Role of the Institutional Environment in Marketing Channels ," Journal of Marketing , 66 (3), 82 – 97.
Haviv Avery , Huang Yufeng , Li Nan. (2020), " Intertemporal Demand Spillover Effects on Video Game Platforms ," Management Science , 66 (10), 4788 – 4807.
Healey John , Moe Wendy M.. (2016), " The Effects of Installed Base Innovativeness and Recency on Content Sales in a Platform-Mediated Market ," International Journal of Research in Marketing , 33 (2), 246 – 60.
Hennig-Thurau Thorsten , Houston Mark B.. (2019), Entertainment Science: Data Analytics and Practical Theory for Movies, Games, Books, and Music. New York : Springer Nature.
Kim Jin-Hyuk , Prince Jeffrey , Qiu Calvin. (2014), " Indirect Network Effects and the Quality Dimension: A Look at the Gaming Industry ," International Journal of Industrial Organization , 37 , 99 – 108.
Kretschmer Tobias , Claussen Jörg. (2016), " Generational Transitions in Platform Markets—The Role of Backward Compatibility ," Strategy Science , 1 (2), 90 – 104.
Kupfer Ann-Kristin , vor der Holte Nora Pähler , Kübler Raoul V. , Hennig-Thurau Thorsten. (2018), " The Role of the Partner Brand's Social Media Power in Brand Alliances, " Journal of Marketing , 82 (3), 25 – 44.
Landsman Vardit , Stremersch Stefan. (2011), " Multihoming in Two-Sided Markets: An Empirical Inquiry in the Video Game Console Industry ," Journal of Marketing , 75 (6), 39 – 54.
Lee Robin. (2013), " Vertical Integration and Exclusivity in Platform and Two-Sided Markets ," American Economic Review , 103 (7), 2960 – 3000.
Lee Yikuan , O'Connor Gina Colarelli. (2003), " New Product Launch Strategy for Network Effects Products ," Journal of the Academy of Marketing Science , 31 (3) 241 – 55.
Liu Hongju. (2010), " Dynamics of Pricing in the Video Game Console Market: Skimming or Penetration? " Journal of Marketing Research , 47 (3), 428 – 43.
Marchand Andre. (2016), " The Power of an Installed Base to Combat Lifecycle Decline: The Case of Video Games ," International Journal of Research in Marketing , 33 (1), 140 – 54.
Mattheessen Kristin. (2019), "The Importance of Measuring Cannibalization to Achieve Product Innovation Success," Gut Check (May 2) , https://www.gutcheckit.com/blog/the-importance-of-measuring-cannibalization-to-achieve-product-innovation-success.
McAlister Leigh , Pessemier Edgar. (1982), " Variety Seeking Behavior: An Interdisciplinary Review ," Journal of Consumer Research , 9 (3), 311 – 22.
Nevo Aviv. (2000), " A Practitioner's Guide to Estimation of Random-Coefficients Logit Models of Demand ," Journal of Economics and Management Strategy , 9 (4), 513 – 48.
Nickell Stephen. (1981), " Biases in Dynamic Models with Fixed Effects ," Econometrica , 49 (6), 1417 – 26.
Noble Charles H. , Kumar Minu. (2010), " Exploring the Appeal of Product Design: A Grounded, Value-Based Model of Key Design Elements and Relationships ," Journal of Product Innovation Management , 27 (5), 640 – 57.
Reber Rolf , Schwarz Norbert , Winkielman Piotr. (2004), " Processing Fluency and Aesthetic Pleasure: Is Beauty in the Perceiver's Processing Experience ?" Personality and Social Psychology Review , 8 (4), 364 – 82.
Reddy Srinivas K. , Holak Susan L. , Bhat Subodh. (1994), " To Extend or Not to Extend: Success Determinants of Line Extensions ," Journal of Marketing Research , 31 (2), 243 – 62.
Rietveld Joost , Schilling Melissa A.. (2021), " Platform Competition: A Systematic and Interdisciplinary Review of the Literature ," Journal of Management , 47 (6), 1528 – 63.
Rietveld Joost , Schilling Melissa A. , Bellavitis Cristiano. (2019), " Platform Strategy: Managing Ecosystem Value Through Selective Promotion of Complements ," Organization Science , 30 (6), 1232 – 51.
Rossi Peter E. (2014), " Invited Paper—Even the Rich Can Make Themselves Poor: A Critical Examination of IV Methods in Marketing Applications ," Marketing Science , 33 (5), 655 – 72.
Rysman Mark. (2009), " The Economics of Two-Sided Markets ," Journal of Economic Perspectives , 23 (3), 125 – 43.
Shankar Venkatesh , Bayus Barry L.. (2003), " Network Effects and Competition: An Empirical Analysis of the Home Video Game Industry ," Strategic Management Journal , 24 (4), 375 – 84.
Srinivasan Sundara Raghavan , Ramakrishnan Sreeram , Grasman Scott E.. (2005), " Identifying the Effects of Cannibalization on the Product Portfolio ," Marketing Intelligence and Planning , 23 (4), 359 – 71.
Sriram S. , Manchanda Puneet , Bravo Mercedes Esteban , Chu Junhong , Ma Liye , Song Minjae , et al. (2015), " Platforms: A Multiplicity of Research Opportunities ," Marketing Letters , 26 (2), 141 – 52.
Stremersch Stefan , Tellis Gerard J. , Franses Philip Hans , Binken Jeroen L.G.. (2007), " Indirect Network Effects in New Product Growth ," Journal of Marketing , 71 (3), 52 – 74.
Usher William. (2015), "Destiny's Poor Metacritic Score Could Cost Bungie Millions," Cinema Blend (September 16) , https://www.cinemablend.com/games/Destiny-Poor-Metacritic-Score-Could-Cost-Bungie-Millions-67350.html.
Vulcano Gustavo , Van Ryzin Garrett , Ratliff Richard. (2012), " Estimating Primary Demand for Substitutable Products from Sales Transaction Data ," Operations Research , 60 (2), 313 – 34.
Wijman Tom. (2018), "Mobile Revenues Account for More than 50% of the Global Games Market as It Reaches $137.9.9 Billion in 2018," Newzoo (April 30) , https://newzoo.com/insights/articles/global-games-market-reaches-137-9-billion-in-2018-mobile-games-take-half.
Wooldridge Jeffrey M. (2010), Econometric Analysis of Cross-Section and Panel Data , 2nd ed. Cambridge, MA : MIT Press.
Zhu Feng , Zhang Xiaoquan. (2010), " Impact of Online Consumer Reviews on Sales: The Moderating Role of Product and Consumer Characteristics ," Journal of Marketing , 74 (2), 133 – 48.
~~~~~~~~
By B.J. Allen; Richard T. Gretz; Mark B. Houston and Suman Basuroy
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 63- Households Under Economic Change: How Micro- and Macroeconomic Conditions Shape Grocery Shopping Behavior. By: Scholdra, Thomas P.; Wichmann, Julian R.K.; Eisenbeiss, Maik; Reinartz, Werner J. Journal of Marketing. Jul2022, Vol. 86 Issue 4, p95-117. 23p. 1 Diagram, 9 Charts, 1 Graph. DOI: 10.1177/00222429211036882.
- Database:
- Business Source Complete
Households Under Economic Change: How Micro- and Macroeconomic Conditions Shape Grocery Shopping Behavior
Economic conditions may significantly affect households' shopping behavior and, by extension, retailers' and manufacturers' firm performance. By explicitly distinguishing between two basic types of economic conditions—micro conditions, in terms of households' personal income, and macro conditions, in terms of the business cycle—this study analyzes how households adjust their grocery shopping behavior. The authors observe more than 5,000 households over eight years and analyze shopping outcomes in terms of what, where, and how much they shop and spend. Results show that micro and macro conditions substantially influence shopping outcomes, but in very different ways. Microeconomic changes lead households to adjust primarily their overall purchase volume—that is, after losing income, households buy fewer products and spend less in total. In contrast, macroeconomic changes cause pronounced structural shifts in households' shopping basket allocation and spending behavior. Specifically, during contractions, households shift purchases toward private labels while also buying and consequently spending more than during expansions. During expansions, however, households increasingly purchase national brands but keep their total spending constant. The authors discuss psychological and sociological mechanisms that can explain the differential effects of micro and macro conditions on shopping behavior and develop important diagnostic and normative implications for retailers and manufacturers.
Keywords: business cycle; income shocks; consumer packaged goods; private label; national brand; discounter; supermarket
Households are subjected to constantly changing economic conditions. These changes may take place at a personal, microeconomic level, such as if the main breadwinner receives a pay raise or a household member loses a job (micro conditions). Alternatively, changes may manifest at a macroeconomic level, in terms of the business cycle, with its recurring expansions and contractions or in response to global events such as the Great Recession or the COVID-19 pandemic (macro conditions). These changing micro and macro conditions substantially affect household spending and, in turn, companies' profits. By one estimate, the Great Recession led to an average 8%, or $4,000, decrease in real annual spending among U.S. households, which amounts to $500 billion in forgone revenues ([20]).
While households tend to simply postpone purchases of durable goods to times of economic prosperity ([16]; [18]), they engage in a variety of adjustments when shopping consumer packaged goods (CPGs): switching from national brands (NBs) to cheaper brands or private labels (PLs), from supermarkets to discounters, from regular to promotional prices, or decreasing the amounts purchased altogether (e.g., [17]; [43]; [46]).
While research to date has focused intensively on how households adjust individual CPG shopping outcomes in response to changing macro conditions (e.g., [17]; [41]; [43]), this work takes a holistic view on households' CPG shopping behavior by uncovering how it is differentially affected by micro and macro conditions. This explicit distinction is important because changes in macro and micro conditions are not necessarily aligned. In fact, even the Great Recession, during which unemployment rates skyrocketed and housing prices and stock portfolios plummeted, did not equally affect the personal income and wealth of all demographic subgroups of the population ([37]) or all geographical regions ([17]). Similarly, the economic downturn caused by the COVID-19 pandemic implies particularly severe microeconomic consequences for industry sectors that depend on tourism, events, or gastronomy, with less effect on banking or the public sector ([48]). Of course, an income loss, for example, as result of sudden unemployment, may as well occur during prosperous economic times and be no lesser of an individual hardship.
Furthermore, the consequences of changing micro and macro conditions differ considerably. Whereas changing micro conditions directly affect households' ability to purchase, changing macro conditions, all else being equal, affect only households' willingness to purchase ([39]). Accordingly, households' response to changing conditions depends on whether they are affected at a micro or macro level (or both) and may manifest in very different shopping outcomes. For example, households may alter what they purchase (e.g., NBs or PLs) and where they shop (e.g., in discounters or supermarkets), as well as how much they spend and purchase. Thus, to properly disentangle the distinct effects of micro and macro conditions and to provide differentiated implications for retailers and manufacturers, holistic observations of households' shopping behavior are crucial.
We analyze a total of seven measurable and managerially relevant shopping outcomes. These outcomes reflect how households allocate their budget across brand types and store formats—their shopping basket allocation (in terms of PL and NB spending in discounters and nondiscounters)—as well as how much they spend and purchase—their shopping basket value (in terms of total spending, purchase volume, and an index of prices paid). Through the analysis, we uncover and characterize the differential effects of micro and macro conditions on households' shopping behavior by addressing the following research questions:
- To what extent do micro (i.e., income) and macro (i.e., the business cycle) conditions affect households' CPG shopping behavior?
- How do micro and macro conditions differ in terms of their effects on households' shopping basket allocation and shopping basket value?
- Do asymmetries exist between negative (i.e., income losses/economic contractions) and positive (i.e., income gains/economic expansions) conditions, and if so, do these asymmetries differ between micro and macro conditions?
We use a unique, comprehensive data set tailored to the research objectives. Drawing on the GfK Germany ConsumerScan panel, we obtain detailed information about daily CPG transactions for more than 5,000 households in Germany over a period of eight years including the Great Recession. Drawing on this, we identify what and where households shop, how much they purchase, what prices they pay, and how much they spend. Annual surveys administered to the panel provide us with longitudinal data on households' demographics and psychographics, including micro conditions in terms of household income. In addition, the panel data enable us to control for important marketing-mix elements concerning prices, assortments, and promotional activities. We further enrich the data set with macroeconomic data from the German Federal Statistical Office and advertising data from the Nielsen Company on advertising spending by all manufacturers and retailers in the sample.
The analyses show that micro and macro conditions both have a substantial impact on households' shopping behavior. Importantly, households adjust their shopping behavior without a concrete change in their budget constraints. In addition, micro and macro conditions differ substantially in their effects on households' shopping behavior. Whereas micro conditions primarily have an impact on households' basket value, macro conditions not only affect households' basket value but also cause shifts in households' basket allocation. During adverse micro conditions, households buy lower volumes and spend substantially less in total but do not shift spending to other brands or store formats. In contrast, as macro conditions change, households shift spending to PLs (from both discounters and nondiscounters) during contractions and to NBs during expansions. In addition, they increase their total spending and purchase volume during contractions. We argue that the shifts during macro conditions are driven by a greater society-wide acceptance of frugal consumption that does not emerge during changing micro conditions. These discrete effects of micro and macro conditions and the proposed underlying mechanisms have distinct managerial implications. The results also address some of the counterintuitive findings of prior studies, such as increasing total spending and purchase volumes ([46]) as well as higher prices paid ([ 8]) during the Great Recession.
Our study relates to business cycle research in marketing as summarized in Table 1.
Graph
Table 1. Literature Overview.
| Authors | Macro Conditions | Micro Conditions | Shopping Behavior(s) | Data Basis |
|---|
| Gicheva, Hastings, and Villas-Boas (2007) | Gasoline prices | | Spending share of income, out-of-home consumption, promotion shares (individually) | Weekly household-level consumption surveys, repeated cross-section, two U.S. regions from 2000 to 2004 |
| Lamey et al. (2007) | Business cycle (asymmetries) | | PL share | Annual, country-level longitudinal data, four countries spanning multiple decades |
| Lamey et al. (2012) | Business cycle | | PL share | Annual, category-level longitudinal data, U.S. from 1985 to 2005 |
| Ma et al. (2011) | Gasoline prices, GDP growth rate | | Shopping trips, total spending, purchase volume, store format, brand type, price tier, and promotion shares (individually) | Monthly, household-level longitudinal panel data, U.S. metropolitan area from 2006 to 2008 |
| Kamakura and Du (2012) | GDP growth | Household budget | Spending share of budget | Annual, household-level consumption surveys, repeated cross-section, U.S. from 1989 to 2003 |
| Lamey (2014) | Business cycle (asymmetries) | | Discounter share | Annual, country-level longitudinal data, 15 countries, spanning 17 years |
| Cha, Chintagunta, and Dhar (2016) | Regional unemployment level | | Total spending, purchase volume, prices paid, store format, brand type, price tier, and promotion shares (individually) | Annual, household-level panel data, repeated cross-section, U.S. from 2006 to 2011 |
| Dubé, Hitsch, and Rossi (2018) | (Post)recession phase | Income, wealth | PL share | Monthly, longitudinal household-level panel data, U.S. from 2004 to 2012 |
| This article | Business cycle (asymmetries) | Income (asymmetries) | Total spending, purchase volume, price index, brand type and store format shares (simultaneously) | Quarterly, longitudinal household-level panel data, Germany from 2007 to 2013 |
1 Notes:[26] and [46] argue that changes in gasoline prices reflect changes in household budgets. We regard gasoline prices as macro effects because they are experienced simultaneously but not necessarily equally by all households, as some households may rely on their cars more than others. As such, they are more similar to macro rather than micro events.
Pioneering studies in this stream show that during recessions, PL market shares ([43]) and discounter market shares ([41]) increase, and some of these effects carry over into subsequent expansion periods. [17] generally confirm these findings by analyzing PL demand at a household level, accounting for heterogeneous income and wealth effects caused by the Great Recession. They find significant short- and long-term effects on PL demand, albeit with notably smaller elasticities. [ 8] further extend the number of shopping behaviors observed. They find that unemployment caused by the Great Recession has led households to increasingly purchase products on price promotion, cheaper brands, and in cheaper store formats. Instead of traditional macroeconomic indicators, [26] and [46] use gasoline prices to operationalize changing economic conditions. They show that gasoline prices relate to a multitude of shopping behaviors such as spending, prices paid, and store format and brand type shares.
In addition to macro conditions, some of the studies in the field observe households' micro conditions. However, they are either used as time-invariant demographic control variables ([ 8]; [26]; [46]) or conceptualized as direct consequences and part of macro conditions rather than distinct conditions with idiosyncratic effects ([17]). Our study thus contributes to this literature stream by delineating the distinct effects of changing micro as well as macro conditions on households' shopping behavior. Importantly, we also account for different magnitudes and asymmetries between adverse and beneficial micro and macro conditions.
First insights into the differences between micro and macro conditions show that overall household spending on food products and alcoholic beverages increases during adverse macro conditions but decreases when micro conditions worsen ([38]). We complement these findings by analyzing a variety of shopping outcomes beyond overall spending, using actual purchase data (thus increasing external validity), and controlling for a large variety of confounding factors such as changes in the marketing mix that are associated with changes in macro conditions ([58]).
Notably, studies to date either focus on individual shopping outcomes (e.g., [17]; [43]) or model several shopping outcomes independently from each other ([ 8]; [26]; [46]). However, households have a variety of means to adjust their shopping behavior that are also highly interdependent—for example, discounters carry substantially more PLs and fewer NBs and usually feature fewer promotions in favor of an everyday low-price strategy. As such, when households switch store formats, it almost automatically also affects their brand type and promotion shares ([11]). Failing to account for these interdependencies can overestimate the effect of changing conditions on individual shopping outcomes. Therefore, we analyze multiple shopping outcomes simultaneously, controlling for their interdependencies, and thus contribute to the literature by offering a holistic picture of micro and macro conditions' effects on households' shopping behavior.
The conceptual framework (Figure 1) depicts the two main components of our study: micro and macro conditions and their effect on households' shopping behavior. We observe these behaviors through concrete and measurable shopping outcomes that, in essence, boil down to households' shopping basket value (i.e., how much households purchase and at what price) and their shopping basket allocation (i.e., how households allocate their expenditures across brand types and store formats). To get a holistic picture of micro and macro conditions' effects on households' shopping behavior, we consider the various shopping outcomes simultaneously. We also control for household demographics and psychographics as well as manufacturer and retailer adjustments to the marketing mix.
Graph: Figure 1. Conceptual framework.
We analyze changing macro conditions in terms of the business cycle on the basis of gross domestic product (GDP) (e.g., [43]; [58]) and derive micro conditions in terms of households' income. Although changing macro conditions are experienced by an entire region, by a nation, or even globally, they do not necessarily affect all households at a micro level. For example, not all households may experience income reductions, job loss, or shrinking wealth during a recession ([17]). Thus, by differentiating between micro and macro conditions, we isolate the distinct effects on shopping outcomes of changes in households' ability to purchase (micro level) and their willingness to purchase (macro level) ([39]). A negative micro shock, for example, restricts some households' shopping budgets, while households that face only adverse macro conditions lack this budget constraint. Importantly, whereas changing micro conditions are usually a personal matter, changing macro conditions affect a society at large. Thus, shifts in macro conditions can alter what type of shopping behavior is considered the norm. During recessions, for example, frugal consumption such as buying PLs or visiting discounters may become socially acceptable and even fashionable ([22]; [38]).
In addition, beneficial and adverse economic conditions exercise asymmetric effects on consumers' shopping behavior for several possible reasons, such as general pessimism following a recession, inertia in maintaining newly adopted habits, or the need to pay off debts that have accrued during a period of lower income ([11]; [43]). Thus, we investigate asymmetric effects by splitting micro and macro conditions into both adverse and beneficial changes.
We distinguish between a household's shopping basket value and shopping basket allocation. We examine shopping basket value outcomes in terms of a household's total budget spent, total volume purchased, and an index of prices paid that indicates whether a household purchases products below average market prices of these products, for example, through temporary price promotions. In this way, we can differentiate the degree to which households adjust how much they purchase and how much they spend. We discern shopping basket allocation outcomes by considering brand types and store formats jointly and differentiating between households' spending on ( 1) PLs in discounters, ( 2) PLs in nondiscounters (e.g., supermarkets, hypermarkets), ( 3) NBs in discounters, and ( 4) NBs in nondiscounters. Prior research has taken a similar approach to households' budget allocation, with studies distinguishing between PLs and NBs as different brand types (e.g., [ 3]; [63]; [56]) or discounters and nondiscounters as different store formats (e.g., [10]; [40]; [41]). This approach has the following conceptual merits.
Regarding brand types, PLs, NBs, and their competition have received ample attention from both academics and practitioners ([40]). PLs have evolved from pure economic options to covering all price tiers and even special segments such as organic foods ([27]; [35]). They have thus developed into major competitors for NBs; for example, in Germany they have gained a market share of 41%, with 95% of consumers buying PLs ([25]; [36]). The competition between NBs and PLs is distinct in that PLs are managed by retailers and, thus, they introduce an aspect of competition into their otherwise collaborative relationship with manufacturers through downward price pressure. However, at the same time, NBs and PLs benefit each other by increasing store traffic and reinforcing quality disparities ([24]; [49]). From a consumer perspective, NBs and PLs differ substantially. First, consumers perceive PLs as inexpensive and as a good value for money. Further, while NBs are generally still better known and are perceived as being of higher quality, PLs are catching up in terms of quality perception ([36]). These differences in terms of price and quality perceptions generally suggest that households will switch between these two brand types in response to changing micro or macro conditions. Thus, the explicit distinction between NBs and PLs is relevant for our research.
In terms of store formats, previous research has contrasted discounters with "traditional retailers" ([40]; [41]), supermarkets ([10]), and large retail formats ([28]; [29]). In contrast to other formats, discounters are highly optimized for cost efficiency, resulting in a substantially different retail marketing mix: store design and product presentation are austere, consumer services are reduced to a minimum, and serviced fresh foods and baked goods counters are lacking. The assortment is typically limited, especially in terms of produce; shallow, with few alternatives in each product category; and dominated by PLs, featuring relatively few NBs. As such, discounters are able to offer substantially lower prices than other store formats at the cost of service quality ([41]; [64]).
In contrast, the major nondiscount store formats, such as supermarkets, superstores, and hypermarkets, vary in floor size and assortments offered beyond CPGs (e.g., clothing, home decor, hardware) but are similar to each other in terms of prices, service quality, and CPG assortments ([40]; [41]; [64]). This is also evident from Table 2, in which we contrast market data from discount and nondiscount store formats in Germany. Therefore, distinguishing between discounters and nondiscounters is most obvious from both retailer and consumer perspectives. Despite their distinct characteristics, however, discounters and nondiscounters do not merely address different target groups but also compete directly with each other for the same consumers, as consistently argued and shown in previous research (e.g., [10]; [33]).
Graph
Table 2. Store Format Characteristics.
| Store Format | # of Storesa | Sales Area (m2/store)a | Revenues(€ mil.)a | Market Sharea | Space Prod. (€/m2) | # of SKUsa | SKU Prod.(€ mil./SKU) | PL Shareb | Service Scorec | Price Scorec |
|---|
| 1. Discounters | 16,054 | 779 | 69,800 | 45.44% | 5,584 | 2,295 | 30.4 | 65.6% | 67.1 | 82.9 |
| 2. Small retailers | 8,750 | 297 | 4,800 | 3.13% | 1,846 | — | — | — | — | — |
| 3. Supermarkets | 10,900 | 982 | 44,900 | 29.23% | 4,196 | 11,830 | 3.8 | 21.6% | 82.0 | 73.6 |
| 4. Superstores | 1,127 | 3,461 | 15,200 | 9.90% | 3,897 | 25,005 | .6 | 84.5 | 74.0 |
| 6. Hypermarkets | 851 | 7,051 | 18,900 | 12.30% | 3,150 | 48,870 | .4 | 19.6% | 79.1 | 77.9 |
| Discounters (1) | 16,054 | 779 | 69,800 | 45.44% | 5,584 | 2,295 | 30.4 | 65.6% | 67.1 | 82.7 |
| Nondiscounters (2–5) | 21,628 | 1,073 | 83,800 | 54.56% | 3,612 | 23,226 | 3.2 | 21.2% | 82.5 | 74.7 |
- 2 aSource:[21], based on 2016 data.
- 3 bSource:[25], based on 2018 data.
- 4 cSource:[16], based on 2018 data.
- 5 Notes: Data are based on the German market. Aggregated values for nondiscounters based on sums or averages weighted by market shares. Service and price scores are indexes (0–100), scores for store formats are aggregates from the 12 major retail brands that were tested. We assigned retail brands to their primary store format based on industry convention and average store size: small retailers <400 m2, supermarkets 400–2,500 m2, superstores 2,500–5,000 m2, hypermarkets >5,000 m2 average sales area.
Importantly, we do not consider the defined brand types (NBs and PLs) and store formats (discounters and nondiscounters) in isolation but in combination. This combined view is important because the brand choice cannot be viewed independently of the underlying store format. For example, because discounters carry a larger PL share than nondiscounters, PLs are more visible to households at discounters and also compete with fewer NBs. At the same time, nondiscount formats usually offer more price tiers (e.g., economy, standard, and premium) and variants (e.g., organic, locally produced, or diet) for NBs as well as PLs within a product category than discounters ([27]; [35]). As such, PL and NB assortments differ structurally between discounters and nondiscounters, and we account for these differences by the combined consideration of these brand types (PLs and NBs) and store formats (discounter and nondiscounters). Thus, by crossing the two brand types and store formats, we obtain a parsimonious, mutually exclusive, collectively exhaustive, and meaningful conceptualization of households' shopping basket allocation. Altogether, the three shopping basket value outcomes and the four shopping basket allocation outcomes holistically cover the essence of households' CPG shopping behavior.
We control for household demographics, which play an important role in explaining differences in shopping baskets (e.g., [46]). In addition, we control for a set of household psychographics: price and quality consciousness, deal proneness, and out-of-home consumption preference. Psychographics control for household heterogeneity that is not necessarily captured by demographics because, for example, even households with high income may be deal-savvy or highly price-conscious ([ 2]). Such psychographics strongly resemble consumer traits that are largely stable in short-term environmental changes but also reflect long-term societal trends, cultural developments, and the process of consumer aging ([54]).
As prior research has shown, retailers and manufacturers also react to macro conditions by adapting their marketing mix (e.g., [13]; [42]). We are less concerned with this relationship per se but control for adjustments in the marketing mix owing to their substantial influence on households' shopping behavior.
As presented in Table 2, the German CPG retail market is split rather evenly between discounters and nondiscounters, with discounters accounting for 45% of revenues and 43% of stores.[ 6] Discounters in Germany are usually located in easily accessible and densely populated areas ([64]) and have an average sales area of 779 m2, which is slightly smaller than a typical supermarket (982 m2) and substantially smaller than superstores (3,461 m2) and hypermarkets ( 7,051 m_SP_2_sp_) ([21]). However, they carry far fewer stockkeeping units (SKUs) and offer a much larger PL share (65.6%) that typically outweighs NBs ([25]). Discounters' PL shares may vary by retailer (e.g., Aldi: 96%, Lidl: 61%), but even discounters with a relatively strong focus on NBs have a substantially larger PL share than nondiscounters (e.g., Penny: 42%, Netto Marken-Discount: 40% vs. nondiscounters: 21.2%). Discounters offer substantially lower prices but also limited service, as is evident from a study by the German Institute for Service Quality ([16]), which scores stores on the basis of their prices and service (higher scores mean better prices/service). The tested discounters received substantially higher (lower) price (service) scores than their nondiscounter counterparts. Discounters' focus on functionality rather than service is also reflected in their high space productivity (i.e., revenues per store space). Similarly, annual revenues per SKU are considerably higher in discounters (€30.4 million) than in nondiscounters (€3.2 million) ([21]).
These data underline the similarity of the nondiscount store formats and their dissimilarity to discounters for the German market from both retailer and consumer perspectives, thus corroborating the previously introduced conceptual distinction between these two groups. Interestingly, this distinction is also reflected in the branding of different retail store formats in the German CPG market. For example, two major German retail companies—the REWE Group and the EDEKA Group—operate both regular supermarkets and superstores under their REWE and EDEKA umbrella brands. Their hypermarkets (REWE Center and E-Center) also incorporate many of the same brand cues. In contrast, their discounters—Penny and Netto Marken-Discount—carry retail brands that are completely distinct from their respective umbrella brand.
To reflect the particularities of the German CPG market, the data set draws on several sources and combines information across distinct aggregation levels. The primary data source is the ConsumerScan panel provided by GfK Germany, which includes transaction and survey data for panelists at the individual household level. As a major advantage, this panel covers private consumption comprehensively and representatively, spanning all German CPG retailers, including discounters that typically do not offer data for market research purposes through retail panels.[ 7] This data availability is particularly crucial, considering the substantial market share of discount stores in Germany (see Table 2). The panel also contains survey data for all panelists, based on self-reported annual demographic information (age, household size, and income) and psychographic measures (e.g., price and quality consciousness). In addition, we obtain data on weekly advertising spending that covers all major channels as well as all manufacturers and retailers from the Nielsen Company. Finally, we add publicly available GDP data from the Federal Statistical Office that indicate the aggregate economic condition. We thus build a unique, encompassing data set that combines behavioral measures with survey-based household demographics and psychographics, macroeconomic measures, and brand- and store-level advertising spending.
The initial raw data set from the ConsumerScan panel is composed of household characteristics and purchase decisions by 85,428 unique households—with 24,000 to 37,000 in any given year—that made more than 13 million shopping trips and 48 million purchases between 2006 and 2013. Purchase information is available at the SKU level for 39 product categories from 467 retailers, most of which maintain multiple stores. These products include alcoholic and nonalcoholic beverages (e.g., beer, fruit juice) and food (e.g., cereals, pasta, ice cream) as well as nonfood items (e.g., deodorants, detergents, toilet paper). For each purchased item, we have access to the unique product code, date and place of purchase, price paid, identifiers for store format, brand type, and temporary price reductions as well as specific product characteristics such as brand and manufacturer name and package size. In preparing these data, we took several cleaning and filtering steps at the purchase record and household levels. In particular, we eliminated inconsistent transaction records and households that did not remain in the panel for the entire period. This procedure is conservative and in line with prior literature (e.g., [17]). Data cleaning involved the following steps: ( 1) Removal of all cases with missing values, ( 2) removal of all cases with unusually high (more than four times the median) or unusually low (less than one-fourth the median) prices at the SKU level, ( 3) removal of all cases with SKUs purchased fewer than 25 times in the entire period.
These data-cleaning steps preserved 97.4% of all observations and 96.1% of all expenditures. To exploit the analytical potential of panelists with long purchase histories and extensive survey information, we retain only households with at least one transaction per quarter (7,441 households) and full survey information from 2006 to 2013, leaving 5,101 unique households.
To avoid structural differences between samples, we compared the filtered households with the remaining households in terms of shopping outcomes and demographics. Overall, we find only marginal deviations in purchase behaviors and demographic composition. Thus, we assume that the selected households with complete purchase histories are not structurally different from households with shorter or incomplete purchase histories. We also compare the filtered sample with information from the 2006 Microcensus ([15]). As in other studies using this type of data (e.g., [17]), our sample is only slightly older and has higher income, fewer single households and more two-person households, and fewer children. However, we find a sizable overlap in the distributions of the demographic variables, and we control for these demographics at the individual household level throughout the empirical analyses. Therefore, a lack of sample representativeness is not an issue. Detailed comparisons of the household samples are available in Web Appendix A.
In line with the conceptual framework, we consider multiple dependent variables to capture the two domains of shopping outcomes as exhaustively as possible. The first domain relates to a household's shopping basket value—that is, how much is spent by the focal household, as represented by three dependent variables. TotalSpendinght relates to the total CPG spending of household h at time t, measured in euros. PurchaseVolht refers to the total CPG purchase volume of household h at time t, again measured in euros. Note that a household's shopping basket typically contains products with different volume units (e.g., liters, grams, pieces) that cannot directly be combined into a total volume measure. Therefore, we follow [46] and use an average category price per volume unit from a one-year (here: 2006) initialization period and multiply it by the total equivalent volume units purchased in each category. This enables us to aggregate the purchase volume across categories. Accordingly, the resulting variable is expressed in euros. We note that any variations in this variable are caused by changes in volume and not changes in prices being paid that may result from switching between brand types and store formats. Therefore, we are able to clearly disentangle households' consumption (volume) from households' spending (value) of CPG purchases. Finally, PriceIndexht is constructed as an index ([ 1]) and compares, for household h at time t, the costs of the shopping basket at average market prices with the actual costs incurred by the household. These price differentials are considered for identical goods identified at the SKU level. As such, they do not reflect differences in the quality of goods purchased but whether specific SKUs in the basket were purchased at cheaper prices (e.g., through temporary price promotions). An index greater than 1 implies that a household paid more than average for the specific goods in its basket, and a value less than 1 implies that the household paid less than average. This variable, therefore, reflects households' cherry-picking behavior ([23]) and is not related to households' switching behavior between different brand types or price tiers. We provide further details on the construction of purchase volume and the price index in Web Appendix B.
The second domain of shopping outcomes relates to a household's shopping basket allocation between combinations of brand types and store formats—that is, it captures how the household is allocating its budget. We measure this allocation with the dependent variable Spendingbht in terms of household h's total spending (in euros) at time t on the respective brand type–store format combination b: (b = 1) PLs in discounters (PLDisc), (b = 2) NBs in discounters (NBDisc), (b = 3) PLs in nondiscounters (PLNonDisc), and (b= 4) NBs in nondiscounters (NBNonDisc). Altogether, these four spending variables encompass each household's total spending.[ 8]
The focal explanatory variables represent a household's individual micro conditions and the overall macro conditions. At the macro level, we first apply the Christiano–Fitzgerald random-walk filter ([ 9]) to the log-transformed quarterly GDP data to assess the general state of the economy itself. The extracted cyclical component of the GDP series constitutes the deviation from the economy's underlying long-term growth trend. Thus, periods with increases in the cyclical component indicate economic expansions, whereas periods with decreases indicate economic contractions. However, it is important to account for not only different phases of the business cycle but also the severity that comes with the depth of up- and downturns (e.g., [55]). To do so, we follow prior research ([43]; [58]) and define the magnitude of an expansion (contraction) period relative to the prior trough (peak) of the cyclical series, or the point in the cyclical component at which the quarter-on-quarter growth turns from negative to positive (from positive to negative). Therefore, we operationalize the symmetric measure of the business cycle (BCyclet) as changes in the cyclical component of GDP at time t relative to the prior peak or trough. In addition, to study potential asymmetries of macro conditions, we use the same operationalization to construct two semidummy variables that separately capture periods with an increase in the cyclical component relative to the prior trough as expansions (Expansiont) and periods with a decrease relative to the prior peak as contractions (Contractiont) of the economy. That is, Expansiont (Contractiont) takes values increasing (decreasing) with economic expansion (contraction) and 0 values during contractions (expansions).[ 9]
At the individual level, micro conditions reflect a household's financial situation, captured by the household's monthly net income. The original income data included in the ConsumerScan panel are at a yearly aggregation level and are measured in 16 income brackets.[10] We construct a continuous income variable by taking midpoint values of these brackets in euros and transform the resulting series to a quarterly sequence (the aggregation level of the shopping outcome variables) by applying linear interpolation for each household.[11] We adjust income for inflation using the consumer price index. In line with the operationalization of macro conditions, we define micro conditions as a household's income change (IncomeChangeht) relative to its previous income peak or trough. This step enables us not only to capture income changes from one period to another but also to take the higher magnitude into account, which results from income changes along consecutive periods. Furthermore, we construct semidummy variables for positive (IncomeGainht) and negative (IncomeLossht) income changes that are equivalent to the operationalization of asymmetric measures at the macro level. Thus, IncomeGainht (IncomeLossht) is defined as the difference of the log-transformed net income at time t and the prior log-transformed income trough (peak), allowing us to account for the accumulated magnitude of income gains and losses over time. IncomeLossht and Contractiont are converted to positive values for ease of interpretation.
As control variables, we include a household's value of the dependent variable from a one-year (here: 2006) initialization period t0 (TotalSpendinght0, PurchaseVolht0, PriceIndexht0, and Spendingbht0). In addition, we include demographics to control for household heterogeneity regarding household size (HhSizeht), age of the household head (Ageht), presence of children (Kidsht), and employment status (Unemployedht). We also include psychographic variables to control for heterogeneity in shopping-related traits and preferences in terms of quality (QualConsht) and price consciousness (PriceConsht), deal proneness (DealProneht), and preferences for eating out (EatOutht). While QualConsht and PriceConsht are based on fixed constructs provided by the GfK, we construct DealProneht and EatOutht from several survey questions. The associated items, factor loadings, and Cronbach's alphas appear in Web Appendix B, Table WB1. Demographic and psychographic controls are measured at an annual level, and we transform the psychographics to a quarterly series using linear interpolation.
Finally, we include controls for the marketing mix. We compute this group of variables at different levels of aggregation as appropriate for each set of models and use household-specific product category weights to incorporate household heterogeneity ([46]). Except for the advertising measures, marketing-mix controls are based on transaction information from the ConsumerScan panel. Because we construct the marketing-mix controls on the basis of observed household transactions, we use only transaction information (e.g., prices, SKUs, price-promoted SKUs) of households that are not part of the analysis sample. Thus, we avoid potential biases resulting from nesting the transactions of these focal households into the marketing-mix controls. For the basket value models, we construct absolute measures for price (Priceht), assortment size (Assortht), price promotions (Promoht), PL share in assortments (PctPLht), and advertising spending of NBs (AdvNBt) and of store format j (with j = 1 for discounters and j = 2 for nondiscounters) (AdvStorejt), which includes advertising spending on retailer brands as well as their PLs. For the basket allocation models, the marketing-mix variables for each brand type–store format combination are computed relative to the average across all brand type–store format alternatives. Thereby, we parsimoniously account for potential cross-effects. In particular, we construct relative measures for price (RelPricebht), assortment size (RelAssortbht), price promotions (RelPromobht), PL share in assortments (RelPctPLjht), and advertising spending at the store level (RelAdvStorejt). Because advertising spending at the brand level refers to NBs only, we use it as an absolute measure.
We adjust all spending and price variables for inflation using the consumer price index and advertising spending using the GDP deflator. Table 3 presents an overview of all variables and their operationalization, while Web Appendix B shows the detailed construction of the marketing-mix variables. Tables 4 and 5 provide the descriptives and correlations for variables in the shopping basket value models and shopping basket allocation models, respectively. Note the small correlations between micro and macro conditions, in support of the conceptualization of differential effects.
Graph
Table 3. Variable Operationalization.
| Variable Group | Variable | Operationalization |
|---|
| Shopping outcomes | TotalSpendinght | Total spending (in euros) by household h at time t. |
| PurchaseVolht | Total purchase volume by household h at time t measured in constant euros. |
| PriceIndexht | Index of prices paid by household h at time t. |
| Spendingbht | Spending (in euros) by household h at time t for brand type–store format combination b. |
| Micro- and macro conditions | BCyclet | Difference between the cyclical GDP component at time t and the prior trough/peak. |
| Expansiont | Difference between the cyclical GDP component at time t and the prior trough. |
| Contractiont | Difference between the cyclical GDP component at time t and the prior peak. |
| IncomeChangeht | Difference between the log-transformed monthly net income (in euros) of household h at time t and the prior income trough/peak. |
| IncomeGainht | Difference between the log-transformed monthly net income (in euros) of household h at time t and the prior income trough. |
| IncomeLossht | Difference between the log-transformed monthly net income (in euros) of household h at time t and the prior income peak. |
| Marketing-mix controls | Priceht | Net price facing household h at time t. |
| RelPricebht | Relative net price of brand type–store format combination b facing household h at time t. |
| Assortht | Number of unique SKUs facing household h at time t. |
| RelAssortbht | Relative number of unique SKUs of brand type–store format combination b facing household h at time t. |
| Promoht | Number of price-promoted SKUs facing household h at time t. |
| RelPromobht | Relative number of price-promoted SKUs of brand type–store format combination b facing household h at time t. |
| PctPLht | Percentage share of PL SKUs in the assortment facing household h at time t. |
| RelPctPLjht | Relative share of PL SKUs in assortment of store format j facing household h at time t. |
| AdvStoret | Store-level advertising spending (in million euros) at time t. |
| RelAdvStorejt | Relative store-level advertising spending of store format j at time t. |
| AdvNBt | Advertising spending (in million euros) of NBs at time t. |
| Demographic controls | HhSizeht | Number of persons in household h at time t. |
| Ageht | Age of the leading person in household h at time t. |
| Kidsht | Dummy variable, 1 if children are present in household h at time t, 0 otherwise. |
| Unemployedht | Dummy variable, 1 if the principal earner of household h is unemployed at time t, 0 otherwise. |
| Psychographic controls | QualConsht | Scale indicating quality consciousness of household h at time t; provided by GfK. |
| PriceConsht | Scale indicating price consciousness of household h at time t; provided by GfK. |
| DealProneht | Five-item scale indicating deal proneness of household h at time t. |
| EatOutht | Three-item scale indicating preference for eating out of household h at time t. |
| Time controls | Timet | Continuous variable for time t. |
| Quarterqt | Indicator variable for quarter q of the year at time t. |
| Other controls | Copulakht | Gaussian copula for marketing-mix variable k to account for potential endogeneity. |
| InvMillsbht | Inverse Mills ratio to account for potential selection effects. |
6 Notes: Items, factor loadings, and Cronbach's alphas for DealProne and EatOut are presented in Table WB1 of Web Appendix B.
Graph
Table 4. Descriptive Statistics and Correlation Matrix for Variables in the Shopping Basket Value Models.
| M | SD | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 |
|---|
| 1. TotalSpending | 184.44 | 114.40 | 1 | | | | | | | | | | | | | | | | | | | | | | | |
| 2. PurchaseVol | 183.06 | 115.62 | .94 | 1 | | | | | | | | | | | | | | | | | | | | | | |
| 3. PriceIndex | .99 | .05 | −.02 | −.15 | 1 | | | | | | | | | | | | | | | | | | | | | |
| 4. BCycle | 1.05 | 4.50 | .00 | .01 | −.01 | 1 | | | | | | | | | | | | | | | | | | | | |
| 5. Expansion | 2.42 | 2.69 | .01 | .02 | .00 | .87 | 1 | | | | | | | | | | | | | | | | | | | |
| 6. Contraction | 1.37 | 2.53 | .01 | .01 | .01 | −.85 | −.49 | 1 | | | | | | | | | | | | | | | | | | |
| 7. IncomeChange | 51.93 | 510.04 | .03 | .03 | −.01 | −.01 | −.02 | −.01 | 1 | | | | | | | | | | | | | | | | | |
| 8. IncomeGain | 176.70 | 341.42 | .00 | .01 | −.01 | −.01 | −.05 | −.04 | .80 | 1 | | | | | | | | | | | | | | | | |
| 9. IncomeLoss | 124.77 | 315.41 | −.05 | −.04 | .00 | .01 | −.02 | −.03 | −.74 | −.19 | 1 | | | | | | | | | | | | | | | |
| 10. Price | .99 | .06 | −.11 | −.16 | .02 | −.13 | −.10 | .11 | .02 | .02 | −.01 | 1 | | | | | | | | | | | | | | |
| 11. Assort | 444.89 | 113.02 | .15 | .13 | −.01 | .03 | −.03 | −.09 | .02 | .06 | .03 | .08 | 1 | | | | | | | | | | | | | |
| 12. Promo | 217.65 | 55.96 | .04 | .02 | −.01 | .05 | −.08 | −.17 | .04 | .14 | .08 | .07 | .86 | 1 | | | | | | | | | | | | |
| 13. PctPL | .33 | .04 | −.17 | −.24 | .05 | −.03 | −.06 | .00 | .01 | .02 | .00 | .72 | .17 | .18 | 1 | | | | | | | | | | | |
| 14. AdvStore | 254.98 | 27.22 | .03 | .02 | .00 | .02 | .17 | .15 | .00 | −.02 | −.01 | −.02 | −.03 | −.03 | −.04 | 1 | | | | | | | | | | |
| 15. AdvNB | 294.83 | 46.12 | .00 | .00 | .00 | .13 | .08 | −.15 | .01 | .04 | .04 | −.02 | .05 | .10 | −.01 | .25 | 1 | | | | | | | | | |
| 16. HhSize | 2.33 | 1.13 | .49 | .52 | −.14 | .00 | .01 | .01 | .07 | .05 | −.07 | .18 | .06 | −.01 | .03 | .01 | .00 | 1 | | | | | | | | |
| 17. Age | 54.36 | 12.07 | −.12 | −.13 | .07 | −.01 | −.04 | −.02 | −.10 | −.14 | .01 | −.18 | .02 | .10 | −.08 | −.02 | .01 | −.37 | 1 | | | | | | | |
| 18. Kids | .18 | .38 | .19 | .21 | −.08 | .00 | .02 | .01 | .07 | .06 | −.04 | .25 | .03 | −.03 | .12 | .01 | .00 | .56 | −.54 | 1 | | | | | | |
| 19. Unemployed | .06 | .24 | −.06 | −.01 | −.01 | .01 | .01 | .00 | −.05 | .01 | .09 | −.01 | .00 | −.03 | −.02 | −.01 | −.01 | −.04 | −.07 | .02 | 1 | | | | | |
| 20. QualCons | 2.94 | .86 | .03 | −.07 | .12 | .00 | .00 | −.01 | .01 | −.01 | −.03 | .03 | −.01 | .03 | .05 | −.01 | .01 | −.04 | .11 | −.07 | −.10 | 1 | | | | |
| 21. PriceCons | 3.14 | .93 | .00 | .13 | −.28 | .00 | .00 | .01 | −.01 | .02 | .03 | .02 | .02 | −.01 | −.03 | .01 | .00 | .14 | −.11 | .11 | .08 | −.39 | 1 | | | |
| 22. DealProne | 11.26 | 2.47 | .10 | .18 | −.28 | .00 | .01 | .01 | .02 | .02 | −.01 | −.02 | .00 | −.04 | −.04 | .01 | .00 | .18 | −.10 | .11 | .03 | −.10 | .38 | 1 | | |
| 23. EatOut | 5.41 | 2.38 | −.07 | −.10 | .05 | .01 | .01 | −.01 | .05 | .07 | .00 | −.02 | −.01 | −.02 | −.01 | −.01 | .00 | −.07 | −.32 | .09 | .03 | .03 | −.05 | −.01 | 1 | |
| 24. Time | | | −.05 | −.05 | −.01 | −.16 | −.27 | .00 | .06 | .20 | .12 | .05 | .25 | .61 | .05 | .08 | .19 | −.04 | .13 | −.05 | −.04 | .04 | −.02 | −.03 | −.01 | 1 |
7 Notes: Means and standard deviations are based on untransformed values, correlations are based on log-transformed variables except dummy variables. BCycle, Expansion, and Contraction are multiplied by 100 to be expressed in percentage deviations.
Graph
Table 5. Descriptive Statistics for Variables in the Shopping Basket Allocation Models.
| PLDisc | NBDisc | PLNonDisc | NBNonDisc |
|---|
| M | SD | M | SD | M | SD | M | SD |
|---|
| Spending | 45.47 | 46.10 | 23.64 | 33.29 | 12.98 | 16.91 | 102.35 | 93.18 |
| Price | .76 | .03 | 1.17 | .04 | .73 | .04 | 1.34 | .04 |
| Assort | .74 | .12 | .66 | .05 | .46 | .08 | 2.15 | .17 |
| Promo | .65 | .13 | .72 | .05 | .43 | .08 | 2.20 | .18 |
| PctPL | 1.46 | .04 | 1.46 | .04 | .54 | .04 | .54 | .04 |
| AdvStore | 1.09 | .07 | 1.09 | .07 | .91 | .07 | .91 | .07 |
| AdvNB | 294.83 | 46.12 | 294.83 | 46.12 | 294.83 | 46.12 | 294.83 | 46.12 |
8 Notes: PLDisc = private labels in discounters; NBDisc = national brands in nondiscounters; PLNonDisc = private labels in nondiscounters; NBNonDisc = national brands in nondiscounters.
We define regression models for the individual shopping outcomes and estimate them jointly in a system of seemingly unrelated regressions. To control for unobserved household heterogeneity, we use a random intercept specification. The three shopping basket value equations for total spending, purchase volume, and price index, as well as the four basket allocation equations for spending across four brand type–store format combinations, are specified in log-log form (excluding the dummy variables Kidsht, Unemployedht, and Quarterqt). This approach allows for an interpretation of coefficients as elasticities and accounts for the fact that households vary substantially in magnitudes of the dependent variables ([46]).[12] We first assume symmetry in each model with regard to the focal micro- and macroeconomic measures, where MacroEcont = δ1BCyclet and MicroEconht = δ2IncomeChangeht. Subsequently, we introduce asymmetric effects, where MacroEcont = γ1Expansiont + γ2Contractiont and MicroEconht = γ3IncomeGainht + γ4IncomeLossht.
We provide the specifications for the shopping basket value and shopping basket allocation models subsequently.
The three shopping basket value models are defined as follows:
Graph
( 1)
where BasketValueaht is (a = 1) TotalSpendinght, (a = 2) PurchaseVolht, (a = 3) PriceIndexht, , k is marketing mix variable k (k = 1, ..., K), q is quarter q in a given year (q = 1, ..., 4), and t is time period t at a quarterly level (t = 1, ..., T).
We control for potential endogeneity in the marketing-mix variables resulting from unobserved shocks by including Gaussian copulas ([47]), which directly model the joint distribution of the potentially endogenous regressor and the error term through control function terms. An advantage of this method is that it does not require instrumental variables that may, as in our case given the number of marketing-mix variables across brand type–store format combinations, be difficult to find ([50]). A requirement is that the endogenous regressor is not normally distributed. Anderson–Darling tests and Kolmogorov–Smirnov tests confirm this nonnormality for all marketing-mix variables at p < .001. Given the large size of the sample, we also visually inspect quantile–quantile plots, which confirm nonnormality for all marketing-mix variables. The Gaussian copula for each marketing mix variable Xht for household h at time t is Copulaht = Φ-1[H(Xht)], where Φ-1 is the inverse distribution function of the standard normal and H(·) is the empirical cumulative distribution function of Xht.
We define the four models as follows:
Graph
( 2)
where , and the subscripts are as defined before.
One issue with Equation 2 is that expenditures are zero where a household does not patronize a specific brand type–store format combination during a period. Considering only those observations with existing expenditures or adding a small constant may lead to biased estimates ([45]). This bias may be quite substantial in our case, where zero expenditures make up between 2.6% for NBs in nondiscounters and 20.8% for NBs in discounters of all the observations. To solve this issue appropriately, we follow the procedure for Type II Tobit models ([63], pp. 560–66). In a first step, we apply a probit model with a random intercept specification and pooled coefficients for brand type–store format choice. This approach allows for the fact that households may patronize multiple brand type–store format combinations. We use the same set of independent variables as in the basket allocation models and additional instrumental variables (average number of shopping trips and unique retailers visited, share of income spent on CPGs, and per capita CPG spending) for identification purposes. In a second step, we compute the inverse Mills ratio, InvMillsbht, based on the probit model results for each brand type-store format combination as InvMillsbht = φ(Xbht′η/Φ(Xbht′η), where φ is the standard normal density function, Φ is the standard normal cumulative distribution function, and η is the vector of parameters from the probit model. The inverse Mills ratio is then added for each brand type–store format combination as an additional independent variable in the basket allocation model to correct for interrelations between brand type–store format choice and spending. As before, we also add Gaussian copulas for all brand type–store format combination specific marketing-mix variables to account for potential endogeneity issues.
We use Latent GOLD 5.1 ([59]) to estimate the seemingly unrelated regression system consisting of seven equations with a maximum likelihood approach. All the models converged before reaching the maximum number of iterations. Because we use data from 2006 for parts of the variable operationalization, we run the model on data from 2007–2013. For holdout validation, we randomly sample 500 households from the filtered data set and run the final estimations on the remaining 4,601 households. Starting with an intercept, time, and sample selection control model (Model 1), we sequentially add the dependent variable from the initialization period (Model 2); marketing-mix variables and endogeneity controls (Model 3); and demographic (Model 4), psychographic (Model 5), and symmetric micro and macro variables (Model 6). Finally, we replace the symmetric with the asymmetric micro and macro variables (Model 7). Table 6 provides an overview of the model-building process and fit statistics. Relying on the Akaike and Bayesian information criteria, Model 7 offers the best fit. We further scrutinize Model 7 for overfitting. We compare its mean squared errors and mean absolute errors between the estimation and holdout sample and find that they are very similar, showing no sign of potential overfitting.
Graph
Table 6. Model Building and Fit Statistics.
| Model | Components | Estimation Sample | Parameters |
|---|
| LL | BIC | AIC |
|---|
| M1 | Intercept + Time + Sample Selection Controls | −395,450 | 791,524 | 791,048 | 74 |
| M2 | M1 + Dependent Variable from Initialization Period | −287,496 | 575,674 | 575,153 | 81 |
| M3 | M2 + Marketing Mix + Copulas | −281,705 | 564,801 | 563,739 | 165 |
| M4 | M3 + Demographics | −273,889 | 549,407 | 548,165 | 193 |
| M5 | M4 + Psychographics | −270,205 | 542,273 | 540,852 | 221 |
| M6 | M5 + Symmetric Economic Conditions | −270,034 | 542,051 | 540,539 | 235 |
| M7 | M5 + Asymmetric Economic Conditions | −269,968 | 542,036 | 540,434 | 249 |
9 Notes: LL = log-likelihood; BIC = Bayesian information criterion; AIC = Akaike information criterion. Note that only models M3–M7 can be compared to one an other as they incorporate the same set of instruments and vary only by their exogenous variables ([19]).
Although the asymmetric model (Model 7) shows the best fit, we briefly present the results from the symmetric model specification (Model 6) to check for internal consistency across the two models. Table 7 provides an overview of all significant elasticities of micro and macro conditions on basket value and basket allocation measures. The complete results of the symmetric model are available in Web Appendix C, Table WC1. Overall, we find significant influences on household shopping behavior for changes in households' micro and macro conditions. However, the nature of these influences clearly varies.
Graph
Table 7. Overview of Significant Elasticities.
| | Basket Allocation | Basket Value |
|---|
| Variable | PLDisc Spending | NBDisc Spending | PLNonDisc Spending | NBNonDisc Spending | Total Spending | Purchase Volume | Price Index |
|---|
| Symmetric model (M6) | BCycle | −.70*** | | −.63*** | .27*** | −.06* | −.06* | −.01* |
| IncomeChange | | | | .08*** | .07*** | .06*** | |
| Asymmetric model (M7) | Expansion | −.94*** | | −.71*** | .52*** | | | −.01* |
| Contraction | .36** | −.32* | .51*** | | .14** | .11* | |
| IncomeGain | | | | | | | |
| IncomeLoss | −.10** | | | −.16*** | −.12*** | −.11*** | |
- 10 *p < .1.
- 11 **p < .05.
- 12 ***p < .01.
- 13 Notes: The table illustrates only significant elasticities. PLDisc = private labels in discounters; NBDisc = national brands in nondiscounters; PLNonDisc = private labels in nondiscounters; NBNonDisc = national brands in nondiscounters. Complete results of the asymmetric Model 7 are provided in Table 8. Complete results of the symmetric Model 6 are provided in Table WC1 of Web Appendix C.
In line with economic theory, we find significant positive elasticities of income change on shopping basket value in terms of total spending (δ = .07, p < .01) and purchase volume (δ = .06, p < .01). Given that these elasticities are very similar in size and both variables are representations of a household's shopping basket in euros featuring comparable means, we can deduce that the majority of the expenditure effect is merely driven by volume adjustments. In fact, these volume adjustments are mainly attributable to purchases of NBs in nondiscounters, as indicated by the significant positive elasticity of income change on NB spending in nondiscounters (δ = .08, p < .01). Importantly, we do not find any structural shifts in households' basket allocation in that households increase (decrease) spending for a specific brand type–store format combination and simultaneously decrease (increase) spending for another.
Under changing macro conditions, the results are different. We find marginally significant negative elasticities of the business cycle on shopping basket value dimensions (i.e., total spending [δ = −.06, p < .1], purchase volume [δ = −.06, p < .1], and price index [δ = −.01, p < .1]). Though intuitively surprising, the results confirm previous studies showing countercyclical CPG spending behavior of households (in value and volume) along the business cycle (e.g., [46]). In addition, we also find several significant elasticities of the business cycle on households' shopping basket allocation. In particular, the elasticity of the business cycle on PL spending in discounters (δ = −.70, p < .01) and nondiscounters (δ = −.63, p < .01) is significantly negative, respectively, whereas it is significantly positive on NB spending in nondiscounters (δ = .27, p < .01). This finding indicates that, to some degree, households shift from PLs in discounters and nondiscounters to NBs in nondiscounters—and vice versa—when macro conditions change. Moreover, when shifting their basket allocation across brand types–store format combinations, households also tend to purchase items at lower prices, for example, through temporary price promotions, as indicated by the negative effect of macro conditions on the price index.
Table 8 shows the estimation results of the asymmetric model. For better comparability of the impact of micro and macro conditions, Figure 2 provides an overview of the asymmetric effects of micro and macro conditions on basket allocation and basket value at their respective mean values—specifically, 2.42 (1.37) for Expansiont (Contractiont) and €176.70 (€124.77) for IncomeGainht (IncomeLossht), which translates to 7.8% (5.5%) of mean income. The findings from the symmetric model are confirmed by the asymmetric model, although the asymmetric estimation results show that the underlying effects are not symmetric but differ strongly in terms of size as well as significance between beneficial and adverse conditions.
Graph: Figure 2. Asymmetric elasticities at mean values for micro and macro conditions.
Graph
Table 8. Results of Asymmetric Model 7.
| Variable | Basket Allocation | Basket Value |
|---|
| PLDisc Spending | NBDisc Spending | PLNonDisc Spending | NBNonDisc Spending | TotalSpending | PurchaseVolume | PriceIndex |
|---|
| Intercept | 2.6310** | (1.3047) | −4.4137* | (2.3163) | −2.0639 | (1.3452) | 4.6453*** | (1.0345) | −1.0209 | (2.6274) | .4872 | (2.6826) | .0460 | (.1136) |
| Random intercept | −.4634*** | (.0131) | −.1435*** | (.0308) | .1604*** | (.0241) | .2319*** | (.0260) | .0279*** | (.0107) | −.0194* | (.0103) | −.0006 | (.0006) |
| Micro and Macro Conditions |
| Expansion | −.9387*** | (.1925) | −.0063 | (.2001) | −.7139*** | (.1816) | .5186*** | (.1132) | .0198 | (.0546) | .0024 | (.0548) | −.0083* | (.0048) |
| Contraction | .3578** | (.1496) | −.3242* | (.1968) | .5068*** | (.1602) | .0485 | (.1198) | .1357** | (.0605) | .1086* | (.0600) | .0015 | (.0050) |
| IncomeGain | −.0341 | (.0413) | .0479 | (.0471) | .0284 | (.0428) | .0112 | (.0352) | .0183 | (.0206) | .0153 | (.0205) | .0011 | (.0014) |
| IncomeLoss | −.0977** | (.0447) | .0124 | (.0532) | −.0508 | (.0463) | −.1567*** | (.0380) | −.1208*** | (.0224) | −.1053*** | (.0219) | −.0005 | (.0017) |
| Controls |
| DV(t = 0) | .3535*** | (.0245) | .2117*** | (.0116) | .3187*** | (.0135) | .5997*** | (.0207) | .5910 | (.0148) | .5719*** | (.0152) | .7263 | (.0162) |
| (Rel)Price | −.6183 | (1.2878) | 1.8388 | (2.1456) | −1.5194 | (1.7101) | −.2376 | (1.5296) | .3850 | (.7840) | .0194 | (.8050) | −.0384 | (.0863) |
| (Rel)Assort | −.4142** | (.1789) | .4310 | (1.1226) | −.2307 | (.2178) | 1.5306*** | (.3764) | .4395* | (.2495) | .1980 | (.2603) | .0239 | (.0169) |
| (Rel)Promo | .1519** | (.0725) | 1.2602 | (3.2248) | .4613* | (.2525) | .8275* | (.4941) | −.4199 | (.2621) | −.2884 | (.2691) | −.0300** | (.0141) |
| (Rel)PctPL | −7.0982*** | (2.0868) | 5.3710* | (2.9790) | −1.4206** | (.6413) | −.1210 | (.2981) | −.5991*** | (.0864) | −.5763*** | (.0930) | −.0025 | (.0096) |
| (Rel)AdvStore | −.2452** | (.1031) | .0052 | (.1281) | .5303*** | (.1342) | −.2811*** | (.0778) | .1275*** | (.0173) | .1022*** | (.0194) | .0005 | (.0022) |
| AdvNB | .1265*** | (.0453) | .2301*** | (.0763) | .1619*** | (.0594) | −.2903*** | (.0418) | −.0396* | (.0223) | −.0488** | (.0230) | −.0008 | (.0022) |
| HhSize | .3706*** | (.0479) | .4305*** | (.0297) | .3406*** | (.0272) | .3722*** | (.0276) | .3077*** | (.0154) | .3250*** | (.0164) | −.0018** | (.0008) |
| Age | −.1958*** | (.0689) | −.1135* | (.0632) | −.1397** | (.0543) | .1270*** | (.0491) | .0078 | (.0228) | −.0085 | (.0224) | .0039** | (.0019) |
| Kids | −.0057 | (.0308) | −.0594* | (.0334) | −.0104 | (.0298) | −.0593** | (.0234) | −.0148 | (.0133) | −.0086 | (.0135) | −.0002 | (.0010) |
| Unemployed | −.0617** | (.0274) | −.0092 | (.0364) | .0563 | (.0342) | −.1157*** | (.0278) | −.0533*** | (.0167) | −.0169 | (.0165) | −.0010 | (.0012) |
| QualCons | −.0224 | (.0262) | .0286 | (.0278) | −.1167*** | (.0253) | .0860*** | (.0200) | .0291*** | (.0109) | −.0175 | (.0108) | .0012 | (.0008) |
| PriceCons | .0654*** | (.0216) | −.0814*** | (.0275) | .0516** | (.0231) | −.1366*** | (.0177) | −.0583*** | (.0109) | .0018 | (.0110) | −.0098*** | (.0009) |
| DealProne | .0196 | (.0349) | .2549*** | (.0402) | −.1277*** | (.0349) | .0485* | (.0268) | .0499*** | (.0153) | .0830*** | (.0152) | −.0158*** | (.0012) |
| EatOut | −.0581** | (.0243) | −.0373 | (.0258) | −.0100 | (.0232) | −.0282 | (.0186) | −.0273** | (.0107) | −.0360*** | (.0105) | .0017** | (.0008) |
| Time | −.0197 | (.0121) | .0781*** | (.0112) | .0382*** | (.0123) | −.0425*** | (.0090) | −.0233*** | (.0051) | −.0270*** | (.0056) | −.0001 | (.0006) |
| Quarter 2 | .0327* | (.0188) | −.0842*** | (.0310) | −.0355 | (.0220) | .1078*** | (.0174) | .0379*** | (.0098) | .0440*** | (.0102) | .0005 | (.0010) |
| Quarter 3 | .0154 | (.0134) | −.0695*** | (.0219) | −.0559*** | (.0157) | .0260** | (.0102) | .0002 | (.0066) | .0089 | (.0070) | .0005 | (.0007) |
| Quarter 4 | −.0026 | (.0136) | −.0076 | (.0231) | −.0193 | (.0168) | .1187*** | (.0113) | .0258*** | (.0063) | .0228*** | (.0065) | −.0001 | (.0006) |
| Copula (Rel)Price | .0483 | (.0587) | −.0626 | (.0717) | .0594 | (.0937) | .0506 | (.0458) | −.0243 | (.0446) | −.0255 | (.0459) | .0019 | (.0049) |
| Copula (Rel)Assort | .0010 | (.0284) | .0143 | (.0742) | .0191 | (.0317) | −.1207*** | (.0297) | −.0719 | (.0603) | −.0222 | (.0629) | −.0041 | (.0041) |
| Copula (Rel)Promo | .0785*** | (.0136) | −.0515 | (.2441) | −.0289 | (.0420) | −.0212 | (.0414) | .0829 | (.0659) | .0607 | (.0676) | .0051 | (.0036) |
| Copula (Rel)PctPL | .2990*** | (.0623) | −.0955 | (.0824) | .1306*** | (.0470) | .0149 | (.0228) | .0431*** | (.0103) | .0201* | (.0109) | .0015 | (.0012) |
| Copula (Rel)AdvStore | .0063** | (.0028) | .0001 | (.0048) | −.0330*** | (.0061) | .0074** | (.0033) | −.0017 | (.0012) | −.0023* | (.0014) | .0003** | (.0001) |
| Copula AdvNB | −.0057* | (.0031) | −.0018 | (.0047) | −.0124*** | (.0045) | .0027 | (.0028) | −.0042*** | (.0016) | −.0031** | (.0015) | .0001 | (.0001) |
| InvMills | .1907 | (.1656) | −.2506*** | (.0703) | −.0832 | (.0734) | −.0260 | (.2121) | | | | | | |
| N | 131,566 | | 113,092 | | 121,787 | | 139,163 | | 142,828 | | 142,828 | | 142,828 | |
| Pseudo-R2 | .59 | | | | | | | | | | | | | |
- 14 *p < .1.
- 15 **p < .05.
- 16 ***p < .01.
- 17 Notes: PLDisc = private labels in discounters; NBDisc = national brands in nondiscounters; PLNonDisc = private labels in nondiscounters; NBNonDisc = national brands in nondiscounters. Standard errors are in parentheses.
Regarding micro conditions, we again find that micro conditions primarily have an impact on households' shopping basket value but do not cause shifts in households' shopping basket allocation. However, the results reveal substantial asymmetries between beneficial and adverse micro conditions. Most notably, income gains have no effect on households' basket value or basket allocation; only income losses show significant effects. More precisely, a 1% loss in income decreases total spending and purchase volume by.12% (p < .01) and.11% (p < .01), respectively. Owing to the similar size of the elasticities, we can again assume that expenditure reductions are largely driven by volume reductions.[13] Given that income losses show no effect on households' price index, we can rule out the notion that expenditure reductions stem from households' shopping for lower prices.
Importantly in the context of income losses, we also see no evidence that households shift their basket allocation to less expensive brand type–store format combinations. Rather, we find significant negative elasticities of income losses only on NB spending in nondiscounters (γ = −.16, p < .01) and PL spending in discounters (γ = −.10, p < .05), respectively. Thereby, we can conclude that the adjustments in purchase volume—and subsequently total spending—predominantly stem from abandoning NBs in nondiscounters and PLs in discounters when income losses occur. Instead of shifting to cheaper store formats, brand types, or both, households give up the relatively more expensive NBs in nondiscounters without substituting them with cheaper alternatives such as NBs in discounters or PLs in general. This lack of substitution is also true for PLs in discounters, but in this case options for shifting to even cheaper alternatives to reduce spending are limited, and therefore, volume adjustments are households' last resort. That is, households' primary means of coping with adverse micro conditions is to reduce expenditures on specific brand types and store formats and thereby reduce shopping basket value (i.e., spending less by purchasing lower volumes) rather than adjusting basket allocation by shifting to cheaper brand types or store formats.
In contrast to adverse micro conditions (i.e., income losses), economic contractions not only have an impact on households' shopping basket value but also cause shifts in basket allocation. With regard to basket value, we find a significant increase in total spending and a marginally significant increase in purchase volume when the economy contracts: a 1% decrease in GDP, compared with its prior peak, increases total spending by.14% (p < .05) and purchase volume by.11% (p < .1). As already indicated for the symmetric model, previous studies also find countercyclical buying behavior of households during adverse macro conditions ([46]).[14] The results confirm and extend these findings by showing that increased total spending and purchase volume are not the only effects during economic downturns, as contractions also cause shifts of households' shopping basket allocation. In particular, we find significantly positive elasticities of contractions on PL spending in discounters (γ = .36, p < .05) and nondiscounters (γ = .51, p < .01), respectively; as well as a marginally significant negative elasticity of contractions on NB spending in discounters (γ = −.32, p < .1). These findings suggest that households shift from NBs to PLs during unfavorable macro conditions. Although previous studies find comparable changes (e.g., [17]; [43]), the combined results further illustrate one important phenomenon: even though households purchase PLs to a greater extent, they increase total spending and purchase volume. Moreover, the results suggest that by switching from NBs to PLs, NBs are not affected by economic downturns per se, but only in the context of discounters. That is, we only find the contraction elasticity of NB spending in discounters to be marginally significant and negative.
The estimated elasticities during economic expansions further substantiate that changing macro conditions cause shifts in households' shopping basket allocation. Inversely to contractions, we find significant negative elasticities of expansions on PL spending in discounters (γ = −.94, p < .01) and nondiscounters (γ = −.71, p < .01), respectively. At the same time, we find a significant positive effect on NB spending in nondiscounters when the economy expands (γ=.52, p < .01). In addition, the results show a marginally significant and negative elasticity of an expansion on the price index (γ = −.01, p < .1). This result complements the findings on households' shifts from PLs in discounters and nondiscounters to NBs in nondiscounters during favorable economic times. In fact, to keep their purchase volume and total expenditures steady while shifting to more expensive NBs, households seem to actively seek price-promoted items to keep the prices they pay low.
Overall, the results show major differences in the effects of micro and macro conditions on households' shopping behavior. While favorable micro conditions show no effect at all, adverse micro conditions lead households to reduce expenditures for specific brand types and store formats, resulting in lower total spending and purchase volumes. In contrast, favorable and unfavorable macro conditions primarily result in shifts of shopping basket allocation. These results highlight the importance of separating micro from macro conditions to identify their unique properties, effects, and implications.
Although the control variables included in the asymmetric Model 7 are not of primary interest, they are important to rule out rival explanations and thus to support the causal interpretability of the main results. Therefore, we briefly summarize them here; a more detailed discussion can be found in Web Appendix C. For the most part, when significant, the effects of the included control variables are intuitive and in line with prior research.
As expected, we find a marginally significant positive effect of assortment size (in terms of unique SKUs) on total spending and a significant positive effect on NB expenditures in nondiscounters. We also find several effects of promotion activity (in terms of unique SKUs sold on promotion): a negative effect on the price index, a marginally significant positive effect on NB spending in nondiscounters, a positive effect on PL spending in discounters, and a marginally significant positive effect on PL spending in nondiscounters. It is noteworthy that the effects for PLs are of smaller magnitude and confirm prior research showing that retail promotions are less positive for PLs than for NBs ([57]). We also find that the share of unique PL SKUs in the total SKU assortment has a negative effect on total spending and purchase volume, suggesting that focusing too strongly on PLs can have unfavorable consequences for retailers (e.g., [ 2]). Finally, advertising at the store level has the expected positive effect on total spending, purchase volume, and PL spending in nondiscounters, while NB advertising has an expected positive effect on NB spending in discounters.
However, we also note that some of the effects are counterintuitive. This is particularly true for the negative effects of assortment size and PL share in assortments, negative own-advertising effects, and positive cross-advertising effects as well as the absence of significant price effects. Varying perceptions of PLs and NBs in assortments (e.g., [ 5]; [14]; [34]), underlying advertising spillover effects ([ 4]), or potential difficulties when measuring advertising effects ([51]; [52]) may provide reasonable explanations for these findings. Counterintuitive marketing-mix coefficients may, however, also be caused by the aggregation level of the data (quarterly, national-level aggregation across many individual brands, retailers, and product categories).
As expected, we find that larger households tend to spend more across all four brand type–store format combinations, spend more in total, purchase larger volumes, and maintain a lower price index. Older households typically spend less on PLs in general as well as spend marginally significantly less on NBs in discounters, but more on NBs in nondiscounters while exhibiting a higher price index. Furthermore, the results suggest that households with children spend less on NBs in nondiscounters and marginally significantly less on NBs in discounters, respectively. Households that suffer from unemployment of the main breadwinner tend to spend less in total, corresponding to fewer expenditures on both NBs in nondiscounters and PLs in discounters.
In terms of psychographics, the analyses reveal many significant effects, generally underscoring the importance of accounting for such types of consumer characteristics ([ 2]). In particular, we find that quality-conscious households tend to spend more in total, more on NBs in nondiscounters, and less on PLs in nondiscounters. In comparison, price-conscious households typically spend more on PLs and less on NBs in general, spend less overall, and exhibit a lower price index. Deal-prone households, furthermore, spend more in total, purchase larger volumes, exhibit a lower price index, and spend less on PLs in nondiscounters, but significantly more on NBs in discounters and marginally significantly more in nondiscounters. Finally, households with preferences for eating out tend to spend less overall, purchase lower volumes, but exhibit a higher price index and typically show lower spending for PLs in discounters.
We perform several robustness checks to confirm the validity of the findings by applying alternative measures and indicators for micro and macro conditions. First, we use the growth rate of real GDP (e.g., [38]; [46]) and an index of consumer confidence (e.g., [ 3]) to assess the general state of the economy. To a large extent, the results are consistent in significance, direction, and magnitude with the main symmetric model (Model 6). Second, we use first-difference specifications of micro conditions rather than differences relative to prior income peaks and troughs as in the main asymmetric model (Model 7). All effects are consistent in significance and direction, even though the elasticities are of a higher order of magnitude. Third, we introduce an individual-level measure of a household's perceived financial situation into both main models. This measure captures changing perceptions of micro conditions that are not reflected in household income (e.g., wealth). Controlling for individual financial perceptions does not alter the findings regarding income, and we can confirm all effects to be consistent in terms of significance, direction, and the order of magnitude. All significant effects of the financial perception measure itself are in line with economic theory. We present and discuss these results in greater detail in Web Appendix C.
Micro and macro conditions have significant effects on households' shopping behavior and outcomes that, by extension, may affect firm performance of retailers and manufacturers. By observing shopping basket allocation across brand types and store formats as well as shopping basket value in terms of total spending, purchase volume, and an index of prices paid, this research provides an extensive analysis of how (through shopping basket allocation) and how much (through shopping basket value) households adjust the various facets of their CPG shopping behavior. Thus, we distinguish the effects caused by micro conditions in terms of income and macro conditions in terms of the business cycle. In addition, we account for possible asymmetries between adverse and beneficial conditions. These findings, based on a rigorous modeling approach and longitudinal field data, have important diagnostic and normative value for managers and contribute to previous research on business cycle effects. We provide an overview of the results and associated implications in Table 9.
Graph
Table 9. Overview of Results and Implications.
| Outcomes | Main Findings | Interpretation and Implications |
|---|
| Shopping basket allocation | PL discounter spending | Moves countercyclically with macro conditions, decreasing in expansions and increasing in contractions. Decreases with adverse micro conditions. | As social acceptance of and demand for PLs increase during contractions, discounters can narrow their price gap to NBs. This allows for more profitable price reductions that discounters should deploy cyclically to counteract shifts to nondiscounters and NBs during expansions and adverse micro conditions. Soft discounters should extend their PL portfolio during contractions. |
| NB discounter spending | Moves cyclically with macro conditions, decreasing substantially in contractions. | Buffer discounters' and manufacturers' revenue losses during adverse micro conditions. Brand managers should extend their portfolio to discounters in these conditions to counteract losses from NBs sold in nondiscounters. Especially hard discounters may profit from a larger NB portfolio. |
| PL nondiscounter spending | Moves countercyclically with macro conditions, decreasing in expansions and increasing in contractions. | Allow nondiscounters to grow revenues even during contractions. Nondiscounters can use this opportunity to extend their PL portfolios to new product categories and price-tiers and strengthen their branding to counteract shifts back to NBs during expansions. As they are unaffected by increasing budget constraints, nondiscounters may adjust prices countercyclically to reap additional revenues during contractions and defend against NBs by deploying price reductions during expansions. |
| NB nondiscounter spending | Moves cyclically, increasing during expansions. Decreases with adverse micro conditions. | Are affected the strongest by adverse micro conditions. Manufacturers and nondiscounters can react to this through status appeals in their communication. As households do not switch due to budget constraints, marketers should not waste budgets on price promotions but provide "cheap" mechanisms that provide consumers with a sense of control and frugality such as loyalty and reward programs or (digital) store fliers. |
| Shopping basket value | Total spending | Grows with adverse macro conditions. Shrinks with adverse micro conditions. | As long as households are not affected at a micro level, they increase their purchased volumes and total spending during contractions. Managers can leverage households' increased consumption and cognitive load from shifts in spending through larger package sizes and in-store promotions. Measures that provide a sense of control and frugality such as loyalty programs or quality and status appeals may further increase compensatory consumption. During expansions, retailers and manufacturers should utilize the increased deal proneness and price savviness through price promotions and couponing. |
| Purchase volume | Grows with adverse macro conditions. Shrinks with adverse micro conditions. |
| Price index | Grows with adverse macro conditions. |
The results uncover and juxtapose the specific effects of micro and macro conditions on shopping behavior. We find that both micro and macro conditions have pronounced effects on households' shopping behavior that are distinct from one another and asymmetric for positive versus negative conditions. Some findings are especially intriguing: micro conditions affect only households' overall consumption levels, whereas macro conditions also lead to structural shifts in households' budget allocation across brand types and store formats. In addition, during changing macro conditions, household adjust their shopping behavior even if they are not affected financially (as we control for income). In this section, we first summarize the results and subsequently discuss potential underlying psychological and sociological mechanisms before addressing interaction effects and asymmetries.
Although no significant adjustments in shopping basket allocation or value emerge for income gains, income losses lead to a general decline in CPG expenditures. This drop is largely driven by households purchasing less and thus spending less. The overall decrease in consumption specifically affects PLs purchased in discounters and NBs purchased in nondiscounters. These findings show that, rather intuitively, budgetary constraints lead to decreased consumption, adding to extant research that has mostly taken a spending perspective (e.g., [38]). However, the absence of structural shifts in households' budget allocation is noteworthy. Theoretically, households could also reduce spending by switching to a cheaper store format or brand type, but instead they generate savings primarily through volume reductions.
In contrast, changing macro conditions evoke structural shifts in households' basket allocation. During contractions, we see expenditures for NBs purchased in discounters being reallocated to PLs purchased in discounters and nondiscounters. While this seems intuitive, it is interesting to note that this shift is accompanied by a general increase in total spending driven by households buying more. In other words, even though households switch to PLs during contractions, they end up spending more in total.
During expansions, households reallocate their purchases from PLs (purchased in nondiscounters as well as discounters) to NBs purchased in nondiscounters. Interestingly, we also find that total spending and volumes purchased remain unaffected at the same time, because households focus more on getting deals, as indicated by a decline of the index for prices paid. As such, households switch to a more expensive brand type during expansions although their budget remains constant (as we control for income), which seems to be feasible as they increasingly purchase products on price promotion.
Several theoretical mechanisms can explain our findings. First, the findings suggest that adverse macro conditions may have a societal impact that trickles down to individual households even if they are not affected at a financial level. In trying times, frugal consumption, such as buying PLs or shopping at discounters, seems to become more socially acceptable and even fashionable ([22]; [38]), which is in line with the shifts of budgets toward PLs in (non)discounters that we observe during contractions. Just as much as frugal consumption may become increasingly commonplace during contractions, purchasing NBs may become a societal norm and is required if households want to maintain their social standing during expansions ([38]). In accordance with that norm, households seem to drop PLs in favor of NBs in nondiscounters even though they have no increase in budgets, as we see in the results. They seem to accommodate this shopping behavior by being price-savvy, shopping products on price promotion. Price promotions may also offer a welcome justification for households to abandon the PLs they have adopted during prior contractions in favor of NBs.
This reasoning is also consistent with the lack of shifts in the face of adverse micro conditions, as described previously. An income loss, independent of macro conditions, is first a personal hardship rather than one shared by society. Therefore, there is not a general move to and acceptance of PLs and discounters, as in the case of adverse macro conditions ([22]; [38])—households do not switch to these cheaper brand types or store formats but instead reduce their overall consumption. In addition, income losses may weaken self-confidence and, thus, awaken a desire to bolster one's social status ([31]; [53]), which may lead households to continue buying NBs while economizing on volume to accommodate their lower income.
Another explanation for these findings may lie in households' perception of the nature of micro and macro conditions. While a nationwide or global contraction is beyond households' direct control, personal income can be influenced through concrete actions. This discrepancy in the "mutability" of the conditions leads to different reactions in households: whereas high-mutability conditions (here: micro conditions) result in high self-regulation, planning, and prioritizing, low-mutability conditions (here: macro conditions) elicit a desire for restoration of control ([ 7]; [31]). Adverse micro conditions lead households to self-regulate by reducing their overall consumption, whereas adverse macro conditions result in a desire to restore control through actions that are perceived as more frugal (i.e., purchasing PLs). Control-restoration behaviors are also associated with compensatory consumption, such as in the form of overspending and higher food intake ([ 7]; [44]), which may explain the overall increase in household spending and which is potentially aggravated by the lack of a budgetary constraint that would limit this behavior ([62]).
Other explanations of the increased consumption may lie in households' shift to PLs, which usually are associated with larger package sizes and lower product prices and which have been shown to increase consumption ([ 6]; [61]). Similarly, these factors contribute to households' purchase of increased quantities when shopping in warehouse club stores ([ 4]). In addition, adding discounter visits to a shopping trip may increase households' spending owing to self-licensing and self-control depletion ([31]).
Like previous studies in the field, we find asymmetries between adverse and beneficial conditions for both micro and macro conditions. In the case of micro conditions, we find that income gains generally have no significant effects on shopping outcomes, whereas income losses do. This finding suggests that households are quick to decrease spending when income decreases but are slow to respond when income increases, potentially because they need to compensate for postponed purchases of durables or paying off debts ([11]). While contractions affect households' shopping basket value more extensively than expansions, the expansion elasticities for shopping basket allocation are mostly larger than during contractions. This response seems reasonable, as failing to keep up with one's surroundings during an expansion would translate into a loss of status, whereas not adopting a more frugal shopping behavior during a contraction implies an increase in status ([38]). In addition, we find more pronounced asymmetries between adverse and beneficial conditions at the macro level than at the micro level. Thus, adjustments in shopping behaviors may reverse more quickly when they are caused by changing micro conditions compared with macro conditions. Given that adverse macro conditions shift the societal acceptance of certain brand types and store formats, households' attitudes may change ([32]). This reasoning implies that macro conditions' effects on shopping outcomes linger longer than micro conditions, during which households engage in status-maintaining shopping behaviors. Therefore, these status-maintaining shopping behaviors may be a means to an end rather than an attitudinal shift and households would quickly discard them once conditions improve.
Finally, we investigate whether micro and macro conditions and the underlying mechanisms that affect households' shopping behavior moderate each other. Thus, we perform a post hoc analysis to test for possible interaction effects for which we present complete results in Web Appendix C, Table WC3.[15] Interestingly, the main effects remain unchanged while all interaction effects are insignificant, which suggests that micro and macro conditions do not moderate each other. Thus, the results indicate that the effects and mechanisms that micro and macro conditions elicit occur independently from each other. That is, if both conditions change simultaneously, their individual effects on households' shopping outcomes work in parallel.
Changing micro conditions affect shopping outcomes only when households suffer income losses rather than gains, leading to a decrease in PLs purchased in discounters and NBs purchased in nondiscounters. To buffer the negative effects of when and where they expect wages to decrease, manufacturers as well as discounters can profit from listing NBs in discounters. In particular, hard discounters such as Aldi and Lidl, whose overwhelming majority of revenues stem from their own PLs, may profit from this strategy. Thus, we provide an additional perspective to the literature investigating the role of NBs in discounters (e.g., [12]). If households indeed suffer from weakened self-confidence and want to bolster their social status as a result of adverse micro conditions ([31]; [53]), NB manufacturers and nondiscounters may leverage this reaction by using status appeals in their advertising. Because adverse micro conditions lead to a general decline in consumption, retailers and manufacturers could target product categories that are affected the most with marketing-mix actions. Changing micro conditions may be especially hard for manufacturers and retailers to identify, but with increasing availability of data through loyalty cards and online shopping, managers could detect the specific shopping outcomes associated with these changes and address those households through personalized coupons and deals.
Changing macro conditions substantially affect households' shopping basket allocation as well as value. Given the increased acceptance of PLs during contractions, retailers can use the opportunity to extend their PL portfolio into higher price tiers and product categories with high involvement and complexity ([56]). In addition, they may narrow their price gap to NBs and strengthen their branding to preemptively counteract households' shifts back to NBs during subsequent expansions. During expansions, they could then offer more attractive and profitable price promotions. In particular, nondiscounter PLs may get away with raising prices because they are unaffected by increasing budgetary constraints. Given the countercyclical susceptibility of PLs, retailers should adjust their assortment accordingly, reducing their PL share in expansions and increasing it in contractions. While hard discounters are especially susceptible to adverse micro conditions, soft discounters (i.e., discounters with a relatively low PL assortment share) should be aware of contractions owing to the substantial negative effect of NBs purchased in discounters and their comparatively low share of PLs that may compensate the losses.
Because we control for micro conditions, the reallocation of budgets to PLs that we observe during adverse macro conditions is apparently not driven by monetary factors but instead may result from changing attitudes toward frugal consumption across society ([38]) and a desire to restore control ([ 7]). If this reasoning holds, it has important implications for managers. NBs and retailers can avoid costly price reductions that are ineffective given the lack of a more constrained budget and instead use measures that provide a perception of frugality.[16] These measures may allow households to engage in behaviors that they associate with economizing but, at the same time, are economical for the retailer or manufacturer. For example, loyalty programs can offer low price discounts and small rewards, giving households the perception that they engage in frugal consumption ([45]). Distribution of (digital) store fliers may create a sense of greater control over the planned shopping trip. In addition, communication may highlight the quality and reliability of products to reduce uncertainty and increase compensatory consumption. NB managers might also consider increasing package size, as larger package size is often associated with a lower per unit price ([ 6]). Finally, NB managers and retailers can leverage the higher cognitive load and depletion of self-control resulting from switching stores and/or brands ([60]), rendering shoppers more susceptible to in-store promotions ([31]).
When individual income is controlled for, changes in observed shopping behaviors resulting from macro conditions are clearly linked to households' willingness, rather than ability, to purchase. Potential underlying changes in attitudes and societal acceptance of certain shopping behaviors provide a conclusive basis for our argumentation. However, we do not observe these changes of attitudes in the data directly. Therefore, we encourage field experiments and laboratory studies to dive deeper into the underlying psychological and sociological mechanisms that might drive these findings. These insights can be crucial in predicting how households will change their CPG shopping in reaction to other types of macro conditions, such as a worldwide pandemic.
Including demographics and psychographics, we control for household characteristics but do not account for heterogeneity in households' reaction to changing conditions, which should be addressed by future research. Heterogeneity may originate, for example, from households' differing preferences for high-quality products, with those preferring high quality potentially opting for adjustments in the volume purchased and the price paid for a good over switches to low-tier NBs and PLs. Alternatively, heterogeneity might stem from households' usual "baseline" shopping behavior because it influences whether and how they are able to economize during adverse conditions.
Future analyses could also differentiate among different product categories, especially relating to the reduction in consumption levels caused by adverse micro conditions. Some product categories may be more essential than others and, thus, consumption may not simply be reduced ([38]). Some product categories may even experience increasing consumption—for example, as households shift from soft drinks and juices to plain water.
Finally, previous research has shown that macro conditions affect marketing-mix decisions ([58]). Thus, future research could take a corporate rather than household perspective, investigating how managers detect and react to changes in micro conditions.
sj-pdf-1-jmx-10.1177_00222429211036882 - Supplemental material for Households Under Economic Change: How Micro- and Macroeconomic Conditions Shape Grocery Shopping Behavior
Supplemental material, sj-pdf-1-jmx-10.1177_00222429211036882 for Households Under Economic Change: How Micro- and Macroeconomic Conditions Shape Grocery Shopping Behavior by Thomas P. Scholdra, Julian R.K. Wichmann, Maik Eisenbeiss and Werner J. Reinartz in Journal of Marketing
Footnotes 1 The first two authors contributed equally to this work.
2 Jan-Benedict Steenkamp
3 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
4 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was partly funded by the German Research Foundation (DFG). Project ID: 244.422.352
5 Thomas P. Scholdra https://orcid.org/0000-0002-0231-0841 Julian R.K. Wichmann https://orcid.org/0000-0003-0205-5243 Maik Eisenbeiss https://orcid.org/0000-0002-0859-3512 Werner J. Reinartz https://orcid.org/0000-0002-2440-3117
6 We consider nondiscounters as comprising small retailers (<400 m2 sales area), supermarkets (400–2,500 m2), superstores (2,500–5,000 m2), and hypermarkets (>5,000 m2) ([21]). The classification of discounters is based on retail brands and is consistent within the industry; we follow this convention.
7 The data set covers the discounters Penny, Norma, Lidl, Plus, Aldi (Nord and Süd), Netto (Nord), Netto Marken-Discount (Süd), and several smaller, more regional discounters. In terms of nondiscounters, the data set covers Edeka, Rewe, Rewe Center, E-Center, Edeka Neukauf, Nah und Gut, Ihr Platz, Markant, Marktkauf, Kaufland, Hit, Real, Globus, Tengelmann Kaisers, Metro, DM, Rossmann, Müller, Budnikowski, Schlecker, and multiple smaller, more regional nondiscount retailers.
8 An alternative approach is to use a market-share model with shares-of-wallets (SOW) as dependent variables. We refrain from this approach because elasticities are not easily comparable across models, because the elasticities in the SOW models depend on total spending (e.g., [57]). Moreover, the elasticities may be misleading as they may show, for example, positive effects on SOW while the total market size is shrinking and spending levels are decreasing. We thank the review team for pointing us in this direction.
9 Note that the business cycle variable BCycle does not solely represent the cyclical component extracted from a GDP series as used, for example, by [16]. Rather, it is a combination of expansion and contraction variables as used, for example, by [43] and [58]. We use this operationalization to achieve comparability between the symmetric and asymmetric models.
The income brackets are (1) <€500, (2) €500–€749, (3) €750–€999, (4) €1,000–€1,249, (5) €1,250–€1,499, (6) €1,500–€1,749, (7) €1,750–€1,999, (8) €2,000–€2,249, (9) €2,250–€2,499, (10) €2,500–€2,749, (11) €2,750–€2,999, (12) €3,000–€3,249, (13) €3,250–€3,499, (14) €3,500–€3,749, (15) €3,750–€3,999, and (16) ≥€4,000.
We use €499 as a proxy for the lowest income bracket and €4,000 for the highest income bracket. Potential biases inferred by this approach should be minimal as the relative number of households falling into these two brackets over the observation period combined is only 7%.
The cyclical component is extracted from a log-transformed GDP series and thus expresses percentage deviations ([43]). Therefore, the coefficients associated with macro variables are elasticities, too.
We also test this more formally through an alternative model specification that uses average price paid by households (AvgPriceht) in lieu of households' total spending. AvgPriceht is defined as TotalSpendinght divided by PurchaseVolht; in contrast to PriceIndexht, it captures whether households switch to different cheaper or more expensive products. All coefficients of the focal independent variables in the AvgPriceht model are insignificant, supporting the interpretation that total spending indeed increases due to volume adjustments rather than due to switches to differently priced products. We thank the associate editor for pointing us in this direction.
In addition, the total spending elasticities for micro and macro conditions match the direction in [38] but are substantially lower, which may stem from the use of field data alongside a variety of control variables rather than survey data. For food at home, [38] estimate an elasticity of −1.0 for income and.9 for a reduction in GDP, whereas we find elasticities of −.121 for income losses and.235 for contractions.
Because interactions in the asymmetric model would necessitate four interaction effects, leading to a complex interpretation and a high potential for multicollinearity we used the symmetric model to test for interaction effects.
Although this may seem to contradict prior findings on stronger price elasticities during adverse macro conditions ([30]; [58]), these studies have not controlled for income but have assumed that an associated decrease in income would lead to greater price sensitivity (see, e.g., [58], p. 179). In this way, our study nicely supports, complements, and concretizes these previous findings.
References Aguiar Mark , Hurst Erik. (2007), " Life-Cycle Prices and Production ," American Economic Review , 97 (5), 1533 – 59.
Ailawadi Kusum L. , Ma Yu , Grewal Dhruv. (2018), " The Club Store Effect: Impact of Shopping in Warehouse Club Stores on Consumers' Packaged Food Purchases ," Journal of Marketing Research , 55 (2), 193 – 207.
Ailawadi Kusum L. , Neslin Scott A. , Gedenk Karen. (2001), " Pursuing the Value-Conscious Consumer: Store Brands Versus National Brand Promotions ," Journal of Marketing , 65 (1), 71 – 89.
Ailawadi Kusum L. , Pauwels Koen , Steenkamp Jan-Benedict E.M.. (2008), " Private-Label Use and Store Loyalty ," Journal of Marketing , 72 (6), 19 – 30.
Allenby Greg M. , Jen Lichung , Leone Robert P.. (1996), " Economic Trends and Being Trendy: The Influence of Consumer Confidence on Retail Fashion Sales ," Journal of Business & Economic Statistics , 14 (1), 103 – 11.
Anderson Eric T. , Simester Duncan. (2013), " Advertising in a Competitive Market: The Role of Product Standards, Customer Learning, and Switching Costs ," Journal of Marketing Research , 50 (4), 489 – 504.
Briesch Richard A. , Chintagunta Pradeep K. , Fox Edward J.. (2009), " How Does Assortment Affect Grocery Store Choice? " Journal of Marketing Research , 46 (2), 176 – 89.
Cakir Metin , Balagtas Joseph V. , Okrent Abigail M. , Urbina-Ramirez Mariana. (2019), " Effects of Package Size on Household Food Purchases ," Applied Economic Perspectives and Policy , 43 (2), 781–801.
Cannon Christopher , Goldsmith Kelly , Roux Caroline. (2019), " A Self-Regulatory Model of Resource Scarcity ," Journal of Consumer Psychology , 29 (1), 104 – 27.
Cha William Minseuk , Chintagunta Pradeep K. , Dhar Sanjay K.. (2016), " Food Purchases During the Great Recession ," Kilts Center for Marketing, Chicago Booth – Nielsen Dataset Paper Series 1-008, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2548758.
Christiano Lawrence J. , Fitzgerald Terry J.. (2003), " The Band Pass Filter ," International Economic Review , 44 (2), 435 – 65.
Cleeren Kathleen , Verboven Frank , Dekimpe Marnik G. , Gielens Katrijn. (2010), " Intra-and Interformat Competition Among Discounters and Supermarkets ," Marketing Science , 29 (3), 456 – 73.
Dekimpe Marnik G. , Deleersnyder Barbara. (2018), " Business Cycle Research in Marketing: A Review and Research Agenda ," Journal of the Academy of Marketing Science , 46 (1), 31 – 58.
Deleersnyder Barbara , Dekimpe Marnik G. , Sarvary Miklos , Parker Philip M.. (2004), " Weathering Tight Economic Times: The Sales Evolution of Consumer Durables over the Business Cycle ," Quantitative Marketing and Economics , 2 (4), 347 – 83.
Deleersnyder Barbara , Dekimpe Marnik G. , Steenkamp Jan-Benedict E.M. , Koll Oliver. (2007), " Win–Win Strategies at Discount Stores ," Journal of Retailing and Consumer Services , 14 (5), 309 – 18.
Deleersnyder Barbara , Dekimpe Marnik G. , Steenkamp Jan-Benedict E.M. , Leeflang Peter S.H.. (2009), " The Role of National Culture in Advertising's Sensitivity to Business Cycles: An Investigation Across Continents ," Journal of Marketing Research , 46 (5), 623 – 36.
Deleersnyder Barbara , Koll Oliver. (2012), " Destination Discount: A Sensible Road for National Brands? " European Journal of Marketing , 46 (9), 1150 – 70.
Destatis (2008), Bevölkerung und Erwerbstätigkeit - Haushalt und Familien - Ergebnisse des Mikrozensus 2006 , Fachserie 1. Wiesbaden : Statistisches Bundesamt.
DISQ (2018), " Studie Lebensmittelmärkte 2018 ," white paper, Deutsches Institut für Service-Qualität.
Dubé Jean-Pierre , Hitsch Günter J. , Rossi Peter E.. (2018), " Income and Wealth Effects on Private-Label Demand: Evidence From the Great Recession ," Marketing Science , 37 (1), 22 – 53.
Dutt Pushan , Padmanabhan Venkata. (2011), " Crisis and Consumption Smoothing ," Marketing Science , 30 (3), 491 – 512.
Ebbes Peter , Papies Dominik , van Heerde Harald J.. (2011), " The Sense and Non-Sense of Holdout Sample Validation in the Presence of Endogeneity ," Marketing Science , 30 (6), 1115 – 22.
The Economist (2011), " Hard Times: How the Economic Slowdown has Changed Consumer Spending in America ," (October 25), https://www.economist.com/graphic-detail/2011/10/25/hard-times.
EHI (2017), " Handelsdaten aktuell 2017 ," Industry Report , EHI Retail Institute.
Flatters Paul , Willmott Michael. (2009), " Understanding the Post-Recession Consumer ," Harvard Business Review , 87 (7/8), 106 – 12.
Fox Edward J. , Hoch Stephan J.. (2005), " Cherry-Picking ," Journal of Marketing , 69 (1), 46 – 62.
Geyskens Inge , Gielens Katrijn , Gijsbrechts Els. (2010), " Proliferating Private-Label Portfolios: How Introducing Economy and Premium Private Labels Influences Brand Choice ," Journal of Marketing Research , 47 (5), 791 – 807.
GfK (2019), " Raus aus dem engen Korsett ," Consumer Index, industry report , Gesellschaft für Konsumforschung , 1 – 8.
Gicheva Dora , Hastings Justine , Villas-Boas Sofia. (2007), "Revisiting the Income Effect: Gasoline Prices and Grocery Purchases," Working Paper 13614, National Bureau of Economic Research , https://www.nber.org/papers/w13614.
Gielens Katrijn , Ma Yu , Namin Aidin , Sethuraman Raj , Smith Ronn J. , Bachtel Robert C. , et al. (2021), " The Future of Private Labels: Towards a Smart Private Label Strategy ," Journal of Retailing , 97 (1), 99 – 115.
Gijsbrechts Els , Campo Katia , Nisol Patricia. (2008), " Beyond Promotion-Based Store Switching: Antecedents and Patterns of Systematic Multiple-Store Shopping ," International Journal of Research in Marketing , 25 (1), 5 – 21.
Gijsbrechts Els , Campo Katia , Vroegrijk Mark. (2018), " Save or (Over-)Spend? The Impact of Hard-Discounter Shopping on Consumers' Grocery Outlay ," International Journal of Research in Marketing , 35 (2), 270 – 88.
González-Benito Óscar , Muñoz-Gallego Pablo A. , Kopalle Praveen K.. (2005), " Asymmetric Competition in Retail Store Formats: Evaluating Inter- and Intra-Format Spatial Effects ," Journal of Retailing , 81 (1), 59 – 73.
Gordon Brett R. , Goldfarb Avi , Li Yang. (2013), " Does Price Elasticity Vary with Economic Growth? A Cross-Category Analysis ," Journal of Marketing Research , 50 (1), 4 – 23.
Hamilton Rebecca W. , Mittal Chiraag , Shah Anuj , Thompson Debora V. , Griskevicius Vladas. (2019), " How Financial Constraints Influence Consumer Behavior: An Integrative Framework ," Journal of Consumer Psychology , 29 (2), 285 – 305.
Hampson Daniel P. , McGoldrick Peter J.. (2013), " A Typology of Adaptive Shopping Patterns in Recession ," Journal of Business Research , 66 (7), 831 – 38.
Haucap Justus , Heimeshoff Ulrich , Klein Gordon J. , Rickert Dennis , Wey Christian. (2013), " Inter-Format Competition among Retailers: The Role of Private Label Products in Market Delineation ," DICE Discussion Paper, No. 101, Düsseldorf : Düsseldorf Institute for Competition Economics (DICE).
Hoch Stephen J., Eric T. Bradlow , Wansink Brian. (1999), " The Variety of an Assortment ," Marketing Science , 18 (4), 527 – 46.
Hökelekli Gizem , Lamey Lien , Verboven Frank (2017a), " The Battle of Traditional Retailers Versus Discounters: The Role of PL Tiers ," Journal of Retailing and Consumer Services , 39 (November) , 11 – 22.
Hökelekli Gizem , Lamey Lien , Verboven Frank (2017b), " Private Label Line Proliferation and Private Label Tier Pricing: A New Dimension of Competition Between Private Labels and National Brands ," Journal of Retailing and Consumer Services , 36 (May) , 39 – 52.
Ipsos (2016), " Handelsmarkenstudie 2016 ," white paper, Ipsos.
Kalleberg Arne L. , Von Wachter Tim L.. (2017), " The US Labor Market During and After the Great Recession: Continuities and Transformations ," The Russell Sage Foundation Journal of the Social Sciences , 3 (3), 1 – 19.
Kamakura Wagner A. , Du Rex Yuxing. (2012), " How Economic Contractions and Expansions Affect Expenditure Patterns ," Journal of Consumer Research , 39 (2), 229 – 47.
Katona George. (1979), " Toward a Macropsychology ," American Psychologist , 34 (2), 118 – 26.
Kumar Nirmalya , Steenkamp Jan-Benedict E.M.. (2007), Private Label Strategy: How to Meet the Store Brand Challenge. Boston : Harvard Business School Press.
Lamey Lien. (2014), " Hard Economic Times: A Dream for Discounters ," European Journal of Marketing , 48 (3/4), 641 – 56.
Lamey Lien , Deleersnyder Barbara , Dekimpe Marnik G. , Steenkamp Jan-Benedict E.M.. (2007), " How Business Cycles Contribute to Private-Label Success: Evidence from the United States and Europe ," Journal of Marketing , 71 (1), 1 – 15.
Lamey Lien , Deleersnyder Barbara , Steenkamp Jan-Benedict E.M. , Dekimpe Marnik G.. (2012), " The Effect of Business-Cycle Fluctuations on Private-Label Share: What Has Marketing Conduct Got to Do with It? " Journal of Marketing , 76 (1), 1 – 19.
Laran Juliano , Salerno Anthony. (2013), " Life-History Strategy, Food Choice, and Caloric Consumption ," Psychological Science , 24 (2), 167 – 73.
Leenheer Jorna , van Heerde Harald J. , Bijmolt Tammo H.A. , Smidts Ale. (2007), " Do Loyalty Programs Really Enhance Behavioral Loyalty? An Empirical Analysis Accounting for Self-Selecting Members ," International Journal of Research in Marketing , 24 (1), 31 – 47.
Ma Yu , Ailawadi Kusum L. , Gauri Dinesh K. , Grewal Dhruv. (2011), " An Empirical Investigation of the Impact of Gasoline Prices on Grocery Shopping Behavior ," Journal of Marketing , 75 (2), 18 – 35.
Organisation for Economic Co-operation and Development (2020), " The Territorial Impact of COVID-19: Managing the Crisis Across Levels of Government ," research report (November 10), https://www.oecd.org/coronavirus/policy-responses/the-territorial-impact-of-covid-19-managing-the-crisis-across-levels-of-government-d3e314e1/.
Park Sungho , Gupta Sachin. (2012), " Handling Endogenous Regressors by Joint Estimation Using Copulas ," Marketing Science , 31 (4), 567 – 86.
Pauwels Koen , Srinivasan Shuba. (2004), " Who Benefits from Store Brand Entry? " Marketing Science , 23 (3), 364 – 90.
Rossi Peter E.. (2014), " Even the Rich Can Make Themselves Poor: A Critical Examination of IV Methods in Marketing Applications ," Marketing Science , 33 (5), 655 – 72.
Sethuraman Raj , Gielens Katrijn. (2014), " Determinants of Store Brand Share ," Journal of Retailing , 90 (2), 141 – 53.
Sethuraman Raj , Tellis Gerard J. , Briesch Richard A.. (2011), " How Well Does Advertising Work? Generalizations from Meta-Analysis of Brand Advertising Elasticities ," Journal of Marketing Research , 48 (3), 457 – 71.
Shapiro Bradley T. , Hitsch Guenter J. , Tuchman Anna E.. (2021), " TV Advertising Effectiveness and Profitability: Generalizable Results from 288 Brands ," Econometrica , 89 (4), 1855–79.
Sivanathan Niro , Pettit Nathan C.. (2010), " Protecting the Self Through Consumption: Status Goods as Affirmational Commodities ," Journal of Experimental Social Psychology , 46 (3), 564 – 70.
Steenkamp Jan-Benedict E.M. , Fang Eric (Er). (2011), " The Impact of Economic Contractions on the Effectiveness of R&D and Advertising: Evidence from U.S. Companies Spanning Three Decades ," Marketing Science , 30 (4), 628 – 45.
Steenkamp Jan-Benedict E.M. , Geyskens Inge. (2014), " Manufacturer and Retailer Strategies to Impact Store Brand Share: Global Integration, Local Adaptation, and Worldwide Learning ," Marketing Science , 33 (1), 6 –26.
Steenkamp Jan-Benedict E.M. , Maydeu-Olivares Alberto. (2015), " Stability and Change in Consumer Traits: Evidence From a 12-Year Longitudinal Study, 2002–2013 ," Journal of Marketing Research , 52 (3), 287 – 308.
Steenkamp Jan-Benedict E.M. , Sloot Laurens. (2018), Retail Disruptors: The Spectacular Rise and Impact of the Hard Discounters. New York : Kogan Page Ltd.
Steenkamp Jan-Benedict E.M. , Van Heerde Harald J. , Geyskens Inge. (2010), " What Makes Consumers Willing to Pay a Price Premium for National Brands over Private Labels? " Journal of Marketing Research , 47 (6), 1011 – 24.
Van Heerde Harald J.. (2005), " The Proper Interpretation of Sales Promotion Effects: Supplement Elasticities with Absolute Sales Effects ," Applied Stochastic Models in Business and Industry , 21 (4/5), 397 – 402.
Van Heerde Harald J. , Gijsenberg Maarten J. , Dekimpe Marnik G. , Steenkamp Jan-Benedict E.M.. (2013), " Price and Advertising Effectiveness over the Business Cycle ," Journal of Marketing Research , 50 (2), 177 – 93.
Vermunt Jeroen K. , Magidson Jay. (2016), Technical Guide for Latent GOLD 5.1: Basic, Advanced, and Syntax. Melmont, MA : Statistical Innovations Inc.
Vohs Kathleen D. , Faber Ronald J.. (2007), " Spent Resources: Self-Regulatory Resource Availability Affects Impulse Buying ," Journal of Consumer Research , 33 (4), 537 – 47.
Wansink Brian. (1996), " Can Package Size Accelerate Usage Volume? " Journal of Marketing , 60 (3), 1 – 14.
Watson Barry , Daley Angela , Rohde Nicholas , Osberg Lars. (2020), " Blown Off-Course? Weight Gain Among the Economically Insecure During the Great Recession ," Journal of Economic Psychology , 80 (October) , 102289.
Wooldridge Jeffrey M.. (2002), Econometric Analysis of Cross Section and Panel Data. Cambridge, MA : MIT Press.
~~~~~~~~
By Thomas P. Scholdra; Julian R.K. Wichmann; Maik Eisenbeiss and Werner J. Reinartz
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 64- How Consumer Orchestration Work Creates Value in the Sharing Economy. By: Scaraboto, Daiane; Figueiredo, Bernardo. Journal of Marketing. Mar2022, Vol. 86 Issue 2, p29-47. 19p. 1 Diagram, 3 Charts. DOI: 10.1177/00222429211027777.
- Database:
- Business Source Complete
How Consumer Orchestration Work Creates Value in the Sharing Economy
Sharing economy platforms have become increasingly popular, but many platforms do not create all the value that is possible because consumers face challenges while cocreating their experiences. The authors situate the origin of these challenges in the sharing economy's hybrid cocreation logics, which combine competing communal and transactional logics. Using a qualitative study of Couchsurfing, a platform for sharing free accommodation, the authors find that consumers engage in orchestration work to overcome cocreation roadblocks and extract greater benefits from sharing economy platforms. This orchestration work consists of many actions reflected in four overarching mechanisms: consumer-to-consumer alignment, rewiring relations, trust investment, and network experimentation. The authors connect these mechanisms to known sources of value for firms (i.e., complementarities, efficiency, lock-in, and novelty) to make recommendations for how platform firms can foster consumer orchestration work and unlock the full value of consumer cocreation in the sharing economy.
Keywords: experimentation; platform firms; sharing economy; trust; value cocreation
Sharing economy platforms are common and growing quickly across industries ([45]). These platforms work as digital marketplaces in which consumers (peer service providers and service users) act as cocreation partners to one another, and platforms facilitate this cocreation. While increasingly popular, many platforms may not be creating all the value that is possible because their consumers face competing institutional logics (i.e., guiding principles) that make cocreation challenging. For example, platform consumers need to cocreate experiences despite differing in terms of their goals and values (e.g., Airbnb homeowners and guests may have different understandings of what "comfortable," "clean," or "convenient" means). Likewise, platform consumers need to reconcile their desire for impersonal transactions and meaningful social interactions when cocreating experiences (e.g., an Uber driver and rider may differ in whether they prefer a quiet ride or a pleasant conversation). Platform consumers need to manage the risk of cocreating with strangers (e.g., Couchsurfing hosts and guests must assess whether to sleep next to a stranger). Further, platform consumers need to find ways to create personalized experiences while collaborating with others who may also want to create their own personalized experiences (e.g., TaskRabbit "taskers" need to figure out how to offer their skills while attending to the specific needs of those who ask for help). We argue that consumers try to solve these challenges by engaging in orchestration work, which we define as the set of actions that consumers engage in to overcome cocreation challenges.
This article explains what platform firms can do to help consumers resolve these challenges and unlock the full value of the orchestration work that consumers are willing to undertake to acquire benefits from cocreating in the sharing economy. Whereas the platform firm could also be considered a cocreation partner, here we focus on how consumers cocreate among themselves, using platform affordances, which are the opportunities for action shaped by a platform's design features. To develop these insights, we conducted a qualitative study of Couchsurfing, a sharing economy platform launched in 2004, in which consumers, through sharing free accommodation, cocreate cultural experiences of hospitality. In some platforms, the roles of peer service provider and user are clearly separate (e.g., Uber's drivers and riders); in others, platform consumers perform both roles at the same time (e.g., Tinder's consumers). We use the term "consumer" to capture both roles given that both parties are actively participating in the platform; if only the provider or user role is creating value, we note this.
This article makes four contributions to value cocreation research and the understanding of marketing in the sharing economy. First, it introduces the hybrid cocreation logics of the sharing economy. Considering these logics, we clearly identify and outline the challenges that platform consumers face when cocreating: the challenge of heterogeneity in cocreation, the challenge of networked sociality for cocreation, the challenge of interpersonal trust, and the challenge of personalizing cocreation.
Second, this article introduces consumer orchestration work, offering a novel understanding of how platform consumers overcome cocreation roadblocks as they navigate the identified challenges. It also maps orchestration actions to associated mechanisms that constitute orchestration work: consumer-to-consumer (C2C) alignment, rewiring relations, trust investment, and network experimentation. In doing so, this article highlights the role of cocreation partners and platform affordances helpful in shaping these actions and mechanisms.
Third, this article addresses a missing link between value cocreation benefits to consumers and platform firms by explaining how consumer orchestration work can lead to complementarities, efficiency, lock-in, and novelty ([56])—known sources of value creation for platform firms. Drawing on these findings, we offer actionable and practical recommendations to sharing economy platforms for leveraging consumer orchestration work in support of value creation.
Finally, this article provides a series of research questions to spur more work on value cocreation in digital platforms, consumer experience in the sharing economy, and sharing economy governance and policy. Overall, we offer marketers a novel framework for understanding and managing value cocreation by consumers in the sharing economy.
Sharing economy platforms draw on multiple, and often competing, institutional logics that prescribe goals, norms, and identities and shape how actors feel, think, and act in that context ([47]). In particular, the contrast between communal and transactional logics ([45]) and the hybrid resulting from their interplay ([52]) creates specific conditions that guide cocreation in this context. Aligned with the understanding that institutions shape value cocreation in service ecosystems ([51]), we refer to the normative cocreation principles established by institutional logics as "cocreation logics."
Communal cocreation logics refer to the principles that guide cocreation when community-oriented partners, who are motivated by shared values and goals, interact to perform shared social practices. With roots in the personal social interactions of Gemeinschaft ([48] [1887]), communal cocreation logics usually entail relationships based on mutuality ([ 3]) and high consociality, that is, high physical and/or virtual copresence of social actors in a network ([34]). In contrast, transactional cocreation logics refer to the principles that guide cocreation when self-interested actors, who have diverse values and goals, interact through formal, contractual, and socially distanced relations ([21]). With roots in the impersonal roles of Gesellschaft ([48] [1887]), transactional cocreation logics predominantly entail one-off quid pro quo exchanges that require money or another token ([39]).
The interplay of these two logics produces the hybrid cocreation logic of the sharing economy, defined as the competing set of principles that combines communal and transactional logics to guide cocreation among consumers in the sharing economy. Mutuality may be present, but most interactions are one-off, quid pro quo exchanges between strangers guided by heterogeneous values and self-interested motivations ([16]). As such, shared practices that could work as normative structures ([51]) for organizing cocreation tend to not develop. Whereas platform firms may regulate the exchange of money and services through informal contracts, these are insufficient to organize the vastly varied cocreation interactions that happen in the sharing economy ([12]). As a result, consumers face specific challenges when cocreating experiences in the sharing economy.
A key challenge is that platform consumers are often very heterogeneous in terms of their resources, goals, and values. As a result, consumers may end up cocreating with partners who differ largely from them, which research has found can be destabilizing. Under a communal cocreation logic, consumers deal with the challenge of heterogeneity through frame alignment practices that "facilitate the accommodation of differences, legitimize heterogeneity, and protect community continuity" ([46], p. 1011). However, the communal approach implies that consumers are trying to find a common frame for cocreation, which is often not the case in the sharing economy ([12]).
Online platforms afford intense, highly consocial connections among consumers ([34]; [54]). However, because a hybrid logic is operating, these relations are more self-interested and transactional than those of communal sociality ([22]). As a result, consumers need to navigate hybrid relationships with cocreation partners who are situated between the close, identity-shaping relations of communities and the often impersonal, transactional relations of commercial settings. Although this is a ubiquitous challenge in relations mediated by digital platforms ([22]), it becomes much more crucial in hybrid cocreation logics in the sharing economy. Prior research shows that digital platform consumers navigate tensions that emerge from contrasting logics of sociality in the sharing economy ([52]). However, it remains unclear how consumers use platform affordances to establish relationships without relying on communal logics.
Sharing economy platforms are environments of distributed trust that are underscored by digital anonymity and/or ambiguity ([ 5]). Guided by hybrid logics, consumers cocreate experiences with strangers without assurance that informal regulation mechanisms, such as reputation ratings, are effective in reducing opportunism ([12]; [45]). This requires that consumers engage in "trust leaps" ([ 6]) as they cocreate with others on the platform. It is challenging, then, for consumers to assess the integrity and reliability of potential cocreation partners in ways that reduce the risk of cocreating experiences with strangers in one-off exchanges. One such risk is potential exploitation by cocreation partners, who may act on the basis of divergent values and norms. This challenge becomes even more pressing when cocreation involves intimate experiences that demand higher levels of interpersonal trust, such as sharing a home or going on a date. Prior research on the sharing economy (e.g., [29]) has highlighted that trust develops as a combination of relational and calculative aspects, and that cocreation partners should work to cultivate trust. However, it is unclear how such work unfolds.
Guided by hybrid cocreation logics, sharing economy consumers pursue personalization and continuously reconfigure resources to accommodate their shifting individual desires ([36]; [37]). Whereas cocreation partners often expect interactions to lead to personalized experiences, the cocreated nature of these experiences demands that personalization be sought through collaboration with others. This leads to personalization roadblocks. Prior research indicates that individuals strive to tailor their cocreated experiences to their specific interests through self-directed customization ([28]). However, it is unclear how consumers navigate personalization roadblocks during cocreation when their partners are guided by different values or have conflicting goals.
These challenges become more pressing when direct marketing controls (e.g., service standards and quality checks) by the platform provider are difficult to implement (e.g., when experiences happen mostly offline or outside of the platform's control), or are not desirable (e.g., when nonstandardized experiences are part of the value proposition). Given these challenges to value cocreation in the sharing economy, we develop an empirically grounded framework of consumer orchestration work aimed at addressing these challenges. Our study is motivated by the following question: How do consumers navigate the challenges of cocreating experiences in the sharing economy, especially in the absence of direct platform firm control?
Couchsurfing is often considered "the original sharing economy platform" ([44], p. 38). Like other hospitality platforms (e.g., Airbnb, HomeAway), Couchsurfing connects hosts (i.e., platform consumers who are offering accommodation) with guests (i.e., platform consumers searching for a place to stay), and the term "couchsurfer" is often used to define all platform consumers independently of the role they take at each interaction. Like many consumer collectives in the sharing economy, exchanges and interactions in the Couchsurfing network are facilitated by an online platform that also keeps track of couchsurfers' reputations ([12]). Couchsurfing is nevertheless considered a purer, more authentic collective than other platforms, such as Uber and Airbnb, because this platform firm only minimally intervenes in how its 12 million consumers cocreate experiences ([41]). At its inception, Couchsurfing was largely guided by the principles of mutuality and generalized exchange ([ 3]), with consumers sharing free accommodation and collaborating to design and maintain this nonprofit platform. Since 2011, however, the platform has been managed by Couchsurfing International Inc., a for-profit organization that has actively pursued options, including a monthly/yearly charge, to capture the many forms of value that its consumers cocreate. Changes in the platform's configuration and its recruiting of consumers have led to the proliferation of quid pro quo exchanges on it ([12]) and the implementation of advertisements and freemium features, characterizing it as a hybrid economy ([39]). Similar to the way that some Airbnb consumers are guided by the logic of hospitality, treating their guests as friends ([52]), some couchsurfers subscribe to both communal and transactional logics, seeking friendships while keeping tabs on what they give and receive when cocreating through the platform.
For the platform firm, the challenge in capturing value is that "the actual time that the two (or more) Couchsurfing partners spend together [and which] constitutes the most important part of the Couchsurfing experience" ([20], p. 510) happens offline, in intimate home settings, and is largely outside the platform firm's direct control. Thus, in contrast to other platforms in the sharing economy, where the platform firm exerts some control over the setting of the experience (e.g., Uber determines the characteristics of the cars in its fleet, Airbnb checks the features of the properties it lists), Couchsurfing presents an extreme case of how the orchestration of experiences unfolds when it is led by consumers. If seen from [34] framing, Couchsurfing offers consumers low platform intermediation and high consociality. Couchsurfing relies little on the standardization afforded by commercial mechanisms (e.g., prices are not available as a parameter for comparing potential hosts, the absence of set rules for check-in/out times requires negotiation between cocreation partners), and the platform's technical features are limited (e.g., guests cannot browse a map of hosts). Thus, experiences cocreated by couchsurfers are extremely heterogeneous, making consumer orchestration actions more necessary, frequent, and salient.
Participants consider their Couchsurfing experience in terms of a stay[ 7] (including pretrip, on-trip, and posttrip expectations, interactions, and responses). Before traveling, a couchsurfer searches the platform for hosts to find accommodation or other couchsurfers to socialize with at the travel destination. Couchsurfers then select potential hosts and exchange messages on the platform or elsewhere (e.g., via social media or SMS) to get to know each other better and plan the stay. When a stay is confirmed, the host and guest continue to interact online before their first face-to-face meeting, which usually takes place in a public place. At home, hosts may choose to share their house keys with guests, cook for them, or include them in their household routines. Often, hosts act as local guides and give insider tips to guests, enabling them to experience a place from the perspective of a local. After a stay, the Couchsurfing platform invites guests and hosts to log their experiences on the platform by writing and posting references within 14 days. References are posted on a couchsurfer's profile and classified as would stay again or would not stay again. If a participant does not leave a reference, the stay is not registered on the platform.
Consistent with consumers' behavior on other sharing economy platforms, couchsurfers may assume different roles, depending on their goals, interests, and their available resources. Couchsurfers may exclusively host, exclusively stay at other members' houses when traveling, both host and stay, or neither host nor stay, instead simply interacting with other couchsurfers in hangouts (i.e., a feature that allows members to easily find nearby couchsurfers to meet up with) or at events (i.e., gatherings organized regularly by couchsurfers).
To investigate how platform consumers orchestrate their experiences and cocreate value in the sharing economy, we used a multimethod approach that combines netnography, participant ethnographic observations, and interviews. Netnography is a qualitative method for investigating online groups, communities, and cultures for marketing purposes ([22]). It requires participant observation in existing online environments and allows the researcher to unobtrusively collect data on consumers' culturally embedded experiences.
By observing and participating through Couchsurfing and related websites, as well as social media platforms, we developed familiarity with the Couchsurfing experience, as lived by consumers. We created profiles on Couchsurfing.com, downloaded the Couchsurfing application to our smartphones, logged in regularly to check couchsurfers' profiles, read discussions in Couchsurfing groups, and followed links offered by couchsurfers while writing field notes and systematically collecting data about experience cocreation. In addition to observing and interacting with couchsurfers, we created an interactive research web page that described this research project and invited consumer participation. We also searched for other content created by couchsurfers, such as posts on Reddit, YouTube, and blogs. The netnographic research lasted 12 months (July 2016 to July 2017), but we continued accessing Couchsurfing-related websites sporadically until December 2019.
The netnographic participant observation seamlessly became ethnographic when one of this study's authors organized two Couchsurfing events to meet local hosts and travelers. Through the platform, this researcher interacted with several people who planned to attend the events and then met some of them face-to-face during one event. We also posted upcoming trips on our Couchsurfing profiles and interacted with potential hosts in the cities where we were planning to travel. We met some of these couchsurfers in hangouts during their travels, stayed at others' homes, and hosted travelers who were visiting our local areas.
Complementing these participant ethnographic observations, we conducted interviews with 40 couchsurfers (see Table 1). We recruited participants through this research project's web page, recommendations provided by members of our personal networks (who are couchsurfers), and snowball sampling. We developed an interview guide that enabled us to capture couchsurfers' perceptions of value cocreation through their experiences ([18]). With the help of two trained assistants, we conducted interviews in person and on Skype and voice recorded and transcribed them. We performed preliminary analyses following each interview and conducted additional interviews until new data became redundant.
Graph
Table 1. Description of Informants.
| Pseudonym | Gender | Age (Years) | Occupation | Couchsurfing Role(s) | Couchsurfer Since … |
|---|
| Adam | Male | 29 | Marketing manager | Guest and host | 2010 |
| Alejandro | Male | 25 | Student | Mostly host | 2015 |
| Anna | Female | 25 | Events producer | Guest | 2012 |
| Arthur | Male | 30 | IT developer | Mostly host | 2013 |
| Brima | Female | 65 | Social worker | Guest and host | 2005 |
| Callum | Male | 26 | Student | Guest | 2015 |
| Carolina | Female | 20 | Artisan | Guest | 2016 |
| Catalina | Female | 29 | Lawyer | Host | 2013 |
| Daniel | Male | 32 | Sales | Host | 2016 |
| David | Male | 30 | Cook | Guest | 2011 |
| Debora | Female | 25 | Support worker | Mostly guest | 2012 |
| Dennis | Male | 55 | English teacher | Mostly host | 2010 |
| Didi | Female | 25 | Child protection | Guest and host | 2015 |
| Elizabeth | Female | 29 | Freelancer | Host and guest | 2016 |
| Eric | Male | 25 | Plant pathologist | Mostly guest | 2015 |
| Fabio | Male | 30 | Teacher | Mostly host | 2013 |
| Fernando | Male | Late 20s | IT technician | Guest | 2017 |
| Frederico | Male | 32 | Geologist/street musician | Mostly guest | 2014 |
| Gastón | Male | N.D. | IT developer | Guest and host | 2016 |
| Gina | Female | 30 | Translator | Guest and host | 2012 |
| Jenna | Female | 56 | Housewife | Mostly host | 2012 |
| Jennifer | Female | 23 | Personal trainer | Mostly guest | 2013 |
| John | Male | 50 | Engineer | Mostly host | 2014 |
| José | Male | 25 | Graphic designer | Guest and host | 2013 |
| Lena | Female | 32 | Waitress | Mostly host | 2011 |
| Linus | Male | 33 | Business analyst | Guest and host | 2008 |
| Lucas | Male | 38 | IT developer | Host | 2010 |
| Marc | Male | 30 | Psychologist | Guest and host | 2009 |
| Maria | Female | 32 | N.D. | Guest | 2015 |
| Mark | Male | 37 | Financial adviser | Mostly host | 2009 |
| Mason | Male | 43 | Doctor | Host | 2012 |
| Milles | Male | 33 | Advertising (unemployed) | Guest and Host | 2014 |
| Olga | Female | 23 | Teacher | Host | 2016 |
| Olivia | Female | 32 | Communications | Guest and host | 2009 |
| Paul | Male | 54 | Air force pilot (retired) | Host | 2015 |
| Ron | Male | 29 | Accountant | Host | 2014 |
| Roxana | Female | 23 | Student | Guest | 2017 |
| Samuel | Male | 37 | Lawyer | Host | 2012 |
| Teresa | Female | 26 | Nurse | Guest | 2017 |
| Zika | Female | 23 | Student | Guest | 2012 |
1 Notes: IT = information technology; N.D. = not disclosed.
Finally, we systematically searched for media reports about Couchsurfing and books written by couchsurfers. Although the media reports contain important information on how the general public has perceived Couchsurfing over time, the books written by couchsurfers contain detailed descriptions of the experiences they cocreate in the network (see the Web Appendix). Through this extensive fieldwork, we amassed a large volume of data in multiple formats, such as field notes, texts, videos, pictures, and audio files. A trained research assistant downloaded data related to value creation and experiences from the web pages we had observed and prepared them for coding and analysis using qualitative data analysis software.
We engaged closely with the entire data set. Consistent with inductive reasoning from grounded theory ([ 8]), we initially conducted emergent coding to identify how different aspects of Couchsurfing's consumer experiences create value for consumers and the platform firm. This approach made salient the multiple actions that consumers enact as they attempt to overcome the cocreation challenges. We refined our analysis and clustered these actions into types by identifying the similarities and differences and by noting the challenges consumers seemed to be solving when enacting each action. At this stage, it became clear to us that consumers were going beyond just collaborating with others to cocreate value for themselves and the community ([40]). They purposefully found workarounds for the challenges they faced when cocreating. This led us to use "orchestration" and "orchestration work," which refer to the coordination of cocreation by multiple actors ([19]) as sensitizing concepts for discussing the data. Thinking about orchestration work as a series of actions allowed us to identify theoretically meaningful patterns that were reciprocally adjusted to the literature ([24]).
In searching for patterns among how consumer actions worked to address the challenges of cocreation, we identified four orchestration mechanisms. We purposefully searched for cases that could dispute our framework, discussed discrepancies, and used several descriptors until we settled on those that clearly delineated the data patterns. In doing so, we adjusted our categories to progress toward a theoretical understanding of the phenomenon in a way that was consistent with the data.
Finally, with this grounded understanding of consumer orchestration work, we reconnected to existing theories of consumer experience, value cocreation, and the sharing economy, and then examined the premise that consumer orchestration actions could be mapped onto the known sources of value creation for firms ([56]). This last stage was an iterative process in which we independently classified excerpts, discussed data exemplars, and used different forms of data as triangulation tools while adjusting their interpretations to the literature ([43]).
As consumers attempt to address the challenges of cocreating experiences in the sharing economy, they engage in multiple, often overlapping actions. These actions and their underpinning mechanisms constitute consumer orchestration work, as they assist consumers in overcoming challenges to cocreating unique, valuable experiences for themselves. We identified four mechanisms of consumer orchestration work: C2C alignment, rewiring relations, trust investment, and network experimentation (see Table 2), clustering consumer actions into groups according to the key challenges that these actions aim to address. In this section, we introduce these mechanisms and their respective orchestration actions and illustrate them with examples from our data set. For the sources of numbered quotes cited in this section and additional examples, see the Web Appendix.
Graph
Table 2. Challenges, Roadblocks, Orchestration Mechanisms, and Actions.
| Cocreation Challenges | Common Sharing Economy Roadblocks | Orchestration Mechanisms | Types of Orchestration Actions |
|---|
| Challenge of Heterogeneity in CocreationHow can consumers cocreate experiences with partners who differ in resources, goals, and values? | Need to select suitable partners for cocreation | C2C Alignment:Enables platform consumers to align experiential elements (i.e., expectations, interactions, and responses) with those of heterogeneous cocreation partners. | Screening |
| Need to better understand current cocreation partners' preferences and expectations about cocreating experiences | Cueing |
| Need to accommodate variability in the interactions with cocreation partners | Flexing |
| Need to reduce the impact of mismatches and unclear preferences on cocreated experiences | Buffering |
| Challenge of Networked Sociality for CocreationHow can consumers reconcile communal and transactional relations with other platform consumers? | Need to establish desired connections with potential partners within the platform environment | Rewiring Relations:Enables consumers to use platform affordances to navigate and integrate the communal and transactional aspects of their relations. | Interest grouping |
| Need to reconnect cocreated experiences with individual and collective identity projects | Lifestyle signaling |
| Need to establish closer relationships within the platform collective | Enclaving |
| Need to balance transactional and communal relations within the platform collective | Reconciling |
| Challenge of Interpersonal TrustHow can consumers reduce the risk of cocreating experiences with strangers in one-off exchanges? | Need to signal one's integrity and assess that of potential cocreation partners | Trust Investment:Enables consumers to manage platform resources to mitigate the risk of engaging in one-off interactions with strangers. | Revealing |
| Need to signal the consistency of one's cocreation behavior | Cultivating reviews |
| Need to establish a higher level of interpersonal trust when cocreating riskier or intimate experiences with strangers | Scaffolding |
| Challenge of Personalizing CocreationHow can consumers explore new opportunities for collaborating with strangers to cocreate unique experiences through sharing economy platforms? | Need to expand cocreation beyond the limits established by available resources | Network Experimentation:Enables consumers to try new resources, roles, and goals when cocreating experiences and thus extends the possibilities for the cocreation of unique, personalized, and valuable experiences in the network. | Creative resourcing |
| Need to expand cocreation beyond the limits established by normative roles present in the platform | Role improvising |
| Need to stretch platform usage to meet personal goals when cocreating | Repurposing |
When platform consumers cocreate in the sharing economy, they have expectations regarding what they will be able to achieve and how. Because sharing economy consumers are highly heterogeneous, these expectations often differ largely between cocreation partners. For example, a Couchsurfing host, Mariam, struggled to understand a cocreation partner's expectations and to have her expectations fulfilled, as she shared on a Couchsurfing group:
I have been hosting just for a couple of weeks and the experiences were incredible until today. I accepted a female surfer from Ukraine. She is 29 and she seemed really nice when she sent me a request. But when she arrived, she is all shy, not talkative at all. I invited her to hang around (she said no because she wanted to charge her phone, which is acceptable), I invited her for pizza which she refused, I also invited her for a beer and she said no. I have been having these amazing surfers and I am not sure what to do now. Maybe I am missing something? (Initial post on thread "No interaction/adjusting interactions")
We uncovered several actions through which platform consumers (both guests and hosts) overcome common roadblocks to cocreation (such as those that Mariam describes) and navigate the challenge of heterogeneity. We detail these actions next, grouping them into similar types, as illustrated by examples from our data set.
"Screening" refers to actions that help consumers select the desired cocreation partners for an experience. These actions include applying search filters ("As a single guy, I do filter out if they're traveling with their partner" [ 1]), checking social media profiles ("For those with no references, I try to connect through Facebook or something where I can verify their IDs" [ 2]), or validating references and identity ("Do you know Bemelieu? He PMd me for a couch" [ 3]), among other actions.
"Cueing" refers to actions that cocreation partners or potential partners undertake to orient the cocreation of their experiences in desired ways. Cueing includes orchestration actions such as listing in one's profile expectations regarding partners ("I love cooking! We can cook together and share good recipes" [ 4]), messaging before a stay to fine-tune expectations ("Hi again, just so you know, I am working from eight to five but we have a spare key that you can borrow" [ 5]), and enacting welcoming rituals ("When someone arrived, someone in the house would show them around the house quickly, say 'if you can see it, you can use it,' and hand them a copy of the keys" [ 6]), among other actions.
"Flexing" refers to actions that imply concessions and adaptations that cocreation partners make concerning one another while interacting to cocreate their experiences. This group of orchestration actions includes proposing or accepting changes to planned activities ("Minni also hosted me a night as I was stuck in SF, even though his profile was on 'no couch.'" [ 7]), requesting/offering additional resources from/to a partner ("I said: OK, if you want to stay longer, stay, no worries" [ 8]), and making adjustments to one's environment, habits, or routine to accommodate partners' needs ("I had a host who gave me and my sister his bed and slept on the living room couch because he didn't want to disturb us when he [left] to work early in the morning" [ 9]), among other actions.
"Buffering" refers to actions taken by one or more partners to overcome heterogeneity impediments to value cocreation and to attenuate the potential loss of reputation due to setbacks in their cocreation of experiences with heterogeneous partners. Buffering includes actions such as establishing boundaries ("I tried to kiss her. She, was like, 'No, no, no—I don't want to make it awkward.'" [10]), apologizing ("I apologized for missing dinner that they had cooked for us" [11]), offering peacemaking gifts ("One girl, despite being certifiably crazy and her stay at my home being a little slice of hell, showed up with free passes to … Dirty Dancing at a theater near my work" [12]), or impeding further deterioration by ending the cocreation ("I've cut visits short, and only once have I just kicked someone out but that was due to some egregious s—" [13]).
These orchestration actions happen at the micro-level of cocreation, that is, within a particular cocreation experience. They allow consumers (both hosts and guests) to identify potential partners who are likely to be aligned with them and adjust to cocreation partners who are not. To capture this function, we categorized screening, cueing, flexing, and buffering actions under the mechanism C2C alignment. We define C2C alignment as the mechanism that enables platform consumers to overcome the challenge of heterogeneity by aligning experiential elements (i.e., expectations, interactions, and responses) with those of heterogeneous cocreation partners.
Take Hilkka, for example, a host who shared her cocreation knowledge in response to Mariam's question in the Couchsurfing "Advice for Hosts" group:
Everybody has different expectations. That's why I tell in my profile [cueing] that I like independent surfers who manage on their own in the downtown. Before I used to guide my guests, taking them to the city center and showing them around for a couple of hours. Having done sightseeing many, many times over the years, I appreciate it if my surfer will do it without me. If it seems that my surfer enjoys my company (and vice-versa), we spend more time together, otherwise it's enough for me to have some interaction in the evening, as well as in the morning at breakfast. There have been surfers who are a little shy, not too willing to have a chat with me, but I don't mind as long as they are friendly and respectful [flexing]. It is difficult to say how much (or little) interaction is good, it varies from case to case. It depends on the surfer and me, how we feel and get along, on mutual interests etc [screening]. Sometimes it's better to limit the interaction to small talk [buffering], although more often it's just great to learn what my surfer has to tell e.g., about her country, family and travels.
Hilkka's post highlights several actions[ 8] that she undertakes as a host to align her expectations with those of potential and actual guests to cocreate experiences, thereby offering an illustration of how the many actions of orchestration through C2C alignment are entangled and work for the common purpose of assuaging the potential differences among cocreators. The actions highlighted in Hilkka's account, when effective, leave cocreation partners with the impression that they "get along" and that the cocreated experiences were "great," despite the heterogeneity among cocreation partners.
As consumers partner to cocreate in the sharing economy, they develop platform-mediated relationships that are a hybrid of transactional exchanges and communal forms of sociality. Navigating such relations and shaping them in ways that are conducive to the creation of valuable experiences is challenging. As a guest, for example, Phil was on a journey through West Africa. While in Ghana, he needed information about Côte d'Ivoire, the country he would visit next. He chose to connect to other consumers on the Couchsurfing platform to obtain information:
I have joined groups and introduced myself, explaining why I was traveling to a particular country or city. Most guidebooks on Cote D'Ivoire are worthless. They are filled with information that is outdated and unreliable. Some have not been updated since the country was at war. ([32])
In planning his trip, Phil needed additional information, and it seemed to him that connecting to other guests and hosts on the Couchsurfing platform could address this need. However, some Couchsurfing consumers want the close, identity-shaping relations of communities; others, the impersonal, transactional relations of commercial settings; and others, a mix of both, depending on the situation. Thus, to achieve his instrumental goal of collecting information for his trip, Phil faced roadblocks associated with the need to form social relations with other couchsurfers whose social goals and interests were very different from his. Phil had to count on the platform's affordances to establish and maintain relationships in this hybrid environment. We identified four types of orchestration actions that help platform consumers overcome the roadblocks associated with the challenge of networked sociality.
"Interest grouping" refers to orchestration actions through which consumers use platform affordances to create explicit links to existing or imagined groups within the collective to access information and leverage shared values and interests. We found that the multiple interest groups hosted on Couchsurfing vary widely and can be based on identity ("Queer Couchsurfers, 53,070 members" [14]), the purpose of an experience ("Worldwide Volunteering, 59,402 members" [15]), information seeking ("Airlines: low-cost, budget, cheap flight, 83,504 members" [16]), location ("South America, 32,914 members" [17]), or needs ("Help! Need a place in Lyon for today" [18]), among other themes.
Most group participants are not committed to frequent participation, and most threads are started by new members. Unlike online communities, these groups do not cultivate communitas ([50]) or foster a sense of belonging among members. Instead, these fleeting associations are appealing because they help sharing economy participants locate sources of complementarities in the larger Couchsurfing network. For instance, those traveling to new places can pool local information and resources for cocreation that would otherwise be unavailable and identify potential cocreation partners with whom they can develop closer, more personal relationships.
"Lifestyle signaling" refers to orchestration actions that use platform affordances to help consumers connect cocreated experiences to a specific lifestyle. These actions consist of a subtler form of identification than interest grouping and allow consumers to connect identity projects to experiences cocreated in a network of which each participant is, at the same time, unique and interchangeable. Under this type, we include actions such as highlighting in one's profile skills that have been developed through participating on the platform ("As a keen amateur photographer, I enjoy chronicling our Couchsurfing adventures in pictures, and introducing our guests to studio photography in our little home studio" [19]) and showcasing identity or creative content that is based on one's experiences with the platform ("We have recently published a book about our Couchsurfing experiences" [20]), among others.
"Enclaving" refers to orchestration actions that use platform affordances to help create ephemeral communal spaces in which consumers experience the advantages of network proximity, despite the ephemeral nature of sharing economy interactions. Similar to consumers' efforts to build a hyper community ([21]), enclaving actions orchestrate experiences around the promise of temporary but intense communality. Enclaving actions include proposing and attending hangouts with strangers ("I have two hours before my train leaves. Wanna grab a beer?" [21]), establishing regularity in local meetings among strangers ("Weekly CS LYON Monday Meeting" [22]), and organizing extraordinary events ("A camp is a multiday event where couchsurfers from all of the world come together and hang out for a whole weekend" [23]).
"Reconciling" refers to orchestration actions that use platform affordances to attempt to reconcile relations based on quid pro quo and generalized reciprocity. These orchestration actions reinforce both the quid pro quo nature of exchanges and the ties among potential cocreation partners. These actions include toning down negative reviews as a way of preserving reputation and future relations while still maintaining the quid pro quo nature of the feedback ("I see a lot of people leave neutral references when they actually want to leave a negative reference but have mixed feelings" [24]), using the platform to achieve personal goals while strengthening communal bonds ("I am hosting in my city because I am kind of new here and it is difficult to meet people" [25]), and organizing activities out of self-interest and that enhance communal activities. Local tourist guides are often involved in organizing pub crawls via hangouts; while they do this for personal gains, they also help foster communal bonds or even seek advice from platform consumers about how to navigate conflicts between communal and transactional logics ("How do you handle situations like this [host seducing guest]? All these people seemed to like us and wanted us to stay with them. We on the other side felt in a very awkward situation" [26]).
These types of actions—interest grouping, lifestyle signaling, enclaving, and reconciling—are categorized under a mechanism we call "rewiring relations." We define rewiring relations as the mechanism that enables consumers to overcome the challenge of network sociality by using platform affordances to navigate and integrate the communal and transactional aspects of their relationships. The mechanism helps consumers leverage platforms' affordances and rework the sociality in the network to better suit their individual goals.
Consider how Phil describes his engagement with the utility of joining Couchsurfing groups nested within the Couchsurfing platform:
While I was still in Ghana, I joined the Cote D'Ivoire group and began searching through previous posts [interest grouping]…. Twenty minutes of browsing through the group and I know what to expect when traveling overland, where I should listen to reggae in Abidjan, and whether it's dangerous in the North of the country. I posted a question of my own about overland travel from Abidjan to Bamako and received some great advice. Within the Cote D'Ivoire group, I noticed several members were particularly active. Their profiles provided a lot of information about them, but what they said in the group was more revealing [lifestyle signaling]. I got a sense for who was proficient in English. I was hoping to unearth my French after several years of neglect, but I liked the idea of staying with someone who spoke English when I first arrived. One particular member spoke English and French and had posted some funny, enthusiastic, and informative messages. I contacted her, and she became my first host in Cote D'Ivoire [reconciling]. I've been staying with her and her boyfriend for almost three weeks. I got in on a meetup for reggae lovers at a bar called Parker Place. Missing my weekly dose of Patty Boom, my favorite reggae spot in DC, Parker Place has been an excellent stand-in, and it has allowed me to meet some awesome Ivorians who share the same musical tastes as me [enclaving]. ([32])
Phil's post highlights several orchestration actions that he enacted through the platform to benefit from being connected to the collective. His temporary association with the Côte d'Ivoire group helped him access updated and personal information on potential experiences ("overland travel from Abidjan to Bamako"). His sharing of activities and goals in the group ("introduced myself, explaining why I was traveling, describing my trips") helped him signal himself to others as a genuine couchsurfer and to find potential hosts who have desirable characteristics (i.e., spoke English and French and were "funny, enthusiastic, and informative"), and his attendance at events organized through the platform ("meetup for reggae lovers") allowed him to cocreate experiences with local couchsurfers who share his interests. These actions enable consumers (both guests and hosts) to navigate the challenge of network sociality to accommodate communal and transactional relations, which shapes sociality in the sharing economy collective in ways that enable the cocreation of valuable experiences.
When cocreating in the sharing economy, consumers face roadblocks associated with the challenge of interpersonal trust, which makes it difficult to cocreate valuable experiences while collaborating with other platform consumers whom they do not know. Guest Janaina, who opened a thread called "Trusting Your Host" on the "Advice for Surfers" group on the Couchsurfing website, explained how challenging it was for her to trust strangers in cocreating intimate experiences:
This will be my first time Couchsurfing alone overseas or couchsurfing at all I should say. So, my friends and family are very skeptical about Couchsurfing especially if it's a male being the host. Just looking for some advice on staying with a male; even if I read good reviews [I am] still a little nervous just because of a totally random person!… How do you really trust the host, leaving your bags there and everything? I almost feel safer just looking for a family or couple rather than a single person. Thanks.
These roadblocks to cocreation, Janaina noted, are not uncommon among couchsurfers. We identified three types of actions that platform consumers (guest and hosts) engage in that help them navigate the challenge of interpersonal trust in cocreating with strangers.
"Revealing" refers to actions that allow platform consumers to signal their integrity and assess that of potential cocreation partners. Revealing helps address a common trust roadblock that consumers face, which is determining how to assess the integrity of potential cocreation partners who may have different values, norms, and behaviors. Revealing includes requesting additional information from others ("For safety reasons, we want complete profiles and personal requests. We want to know who we are hosting and why they chose us" [27]), providing additional information about oneself in profiles, discussion forums, messages, and face-to-face interactions ("When a surfer sends me a link to his/her VLOG and after reviewing a few videos,… my decision to host is usually immediate" [28]), and asking friends for personal testimonies ("If you are new to [Couchsurfing] and don't have any reviews, ask your friends who use the service to write you a review and describe you as a friend" [29]).
A common roadblock that consumers face when cocreating with strangers is the need to demonstrate how reliable they are to potential cocreation partners. We found that consumers overcome this roadblock by engaging in orchestration actions that signal consistency in behavior and reliability. We refer to this type of action as "cultivating reviews." This includes actions such as hosting friends in exchange for reviews ("The host wants to be a surfer when they travel, so they host to build more reviews so that they can get accepted when they send out Couchsurfing requests" [30]), asking for reviews of a specific kind ("A host asked me to make sure I made specific comments about the freedom he gave to all guests staying at this place" [31]), and boosting reviews ("[Couchsurfer] asked me to leave him a reference so that he could achieve a better gender balance in his references" [32]).
"Scaffolding" refers to orchestration actions that allow cocreation partners to progressively trust one another. Scaffolding results in familiarity, security, and control, as it reduces trust gaps by allowing participants to make gradual assessments of their counterparts' reliability and integrity. Scaffolding actions include proposing short experiences before a stay ("Later he asked if he could stay tonight and I said yes, then he asked for a second night and I accepted that as well" [33]), meeting cocreation partners in public spaces first ("We met at the station and then headed to his home" [34]), using temporary gatherings as a way of getting to know partners ("I went to a hangout to try to find a couch" [35]), creating if/then rules for interactional behavior ("I don't even write a request to these men [who only host women]" [36]), restricting access to personal resources ("Laundry at the house is not on offer, but I can assist you in finding a coin laundry for you to get those clothes washed!" [37]), and pacing responses ("I see this constantly, mostly from young female surfers, where they say they are receiving multiple requests and will respond later or something similar to that" [38]), among other actions.
These types of actions—revealing, cultivating reviews, and scaffolding—enable consumers to reduce risks in cocreation. To capture this function, we categorize these actions under a mechanism we call "trust investment." We define trust investment as the mechanism that enables platform consumers to address the challenge of interpersonal trust by managing platform resources to mitigate the risk of engaging in one-off interactions with strangers in the sharing economy. In interpersonal relationships, trust develops when "one party has confidence in an exchange partner's reliability and integrity" ([30], p. 23); thus, it signals the degree to which a person can depend on others to do what they say they will. Whereas the actions of C2C alignment and rewiring relations may also occasionally end up reducing risk among cocreating consumers in a sharing economy collective, trust investment actions are enacted specifically to overcome the challenge of interpersonal trust.
Take, for example, how host Marie advised Janaina to host others to build a trustworthy track record rather than worrying about potential unsafe hosts:
[You should host, but] if you really really really cannot host, please show a CS member around … meet up and share a meal and conversations [scaffolding] … i.e., invest some of your time and resources,… And that way, get some references [cultivating reviews], your potential future host wants to be safe with you, too! as someone said above: if you request a couch from a host with references: you could double check with former guests about safety. but if you have none: how could I, your host, double check to make sure I am safe with you? after all you'll be invading all of my privacy - while all I could potentially invade would be your backpack :-)" (Marie, comment on the "Trusting your host" thread in the Couchsurfing "Advice for Surfers" group)
Marie recommends orchestration actions that work as trust investments. She recommends that the new couchsurfer make efforts to meet other consumers and share resources with them ("a meal and conversations") as a way to progressively build a reputation ("get some references") and signal her integrity and reliability on the platform. She also explains how references are not to be taken for granted, as they are indicators of a potential cocreation partner's reliability but should be mobilized to reveal further information ("doublecheck with former guests about safety") that could help reduce the risks.
Platform consumers seek personalized experiences in the sharing economy. To achieve these, they need to continuously work with cocreation partners, often improvising as they cocreate and adapting for personalization. Take, for example, the case of Nahim, a self-described "couchsurfing-drifter" (39) who was using Couchsurfing to find accommodation in Afghanistan. Given the country's weak security and deteriorated tourism infrastructure, Couchsurfing emerged in the region as an alternative (Kabul has approximately 1,000 Couchsurfing hosts), allowing local hosts to quench their thirst for cosmopolitan experiences and intrepid travelers to find hospitality. However, cocreating experiences in unconventional conditions is not easy. Nahim faced roadblocks associated with not knowing the hosts' culture, not being dressed adequately, and not understanding the language. We identified three types of orchestration actions that help platform consumers (both guests and hosts) overcome roadblocks associated with personalized experiences, such as those experienced by Nahim, and cocreating unique value outcomes in the sharing economy.
"Creative resourcing" refers to the type of actions whereby consumers introduce new resources to cocreate experiences that are unique and personalized and share these experiences on the platform. These actions include making new resources available for cocreation, which extends the range of cocreated experiences ("[Host] hosted me for two months at the university housing building. I had my own room, and even attended classes [laughs]" [40]) and offering different sets of resources within a traditional experience ("The [host] had coffins that went into the wall, would pull them out, and it became a bed" [41]), among others.
"Role improvising" refers to the type of actions whereby cocreation partners step into new roles and scripts while cocreating to extend the value potential of their experiences. We identified a series of actions for this type, including taking on new roles and scripts to enhance the range of cocreation activities ("[The host's] mother is sick, has cancer, so I … chose to spend more time with him, talk to him, just like a psychologist" [42]) and immersing oneself in these roles and scripts to prolong or intensify the cocreation experience ("we are still in touch, and [guest] invited me to his wedding in Brazil" [43]). Through these actions, cocreation partners move away from the normative guest and host roles to improvise as confidantes, psychologists, and stylists, among other roles.
"Repurposing" refers to orchestration actions whereby consumers introduce new goals or value propositions for interactions that are enabled by the platform. Under this label, we grouped actions such as using the network and the platform for cocreation purposes other than those endorsed by the platform firm (e.g., "sex surfing" [44]) and using current resources to repurpose activities toward achieving additional goals ("[Couchsurfers] will hire a bus to visit a remote area outside the city [as a Couchsurfing event]" [45]), among others.
These types of actions—creative resourcing, role improvisation, and repurposing—constitute a mechanism we call "network experimentation." We define network experimentation as the mechanism that enables platform consumers to overcome the challenge of personalizing cocreation by trying new resources, roles, and goals when cocreating experiences, thus extending the possibilities for the cocreation of unique, personalized, and valuable experiences. Through network experimentation, orchestration actions increase the field of potential expectations, interactions, and responses among cocreation partners in the network. Consider, for instance, how the couchsurfers in Afghanistan engaged in actions of network experimentation to help Nahim and other couchsurfers overcome the challenge of personalizing cocreation, as recounted in a blog post covering Nahim's travel experience:
[He] managed to hitchhike through Iraq by displaying a sign in Arabic to passing drivers, written by one of his hosts [role improvising]. After arriving in the western Afghanistan city of Herat, he became acquainted with some local members of the Taliban, whom he described as "actually really nice people." His disguise [to travel safely in Afghanistan] consisted of a white shalwar kameez (traditional Afghan clothing) and a taqiyah (cap for observant Muslims). The clothing was provided by his Couchsurfing hosts [creative resourcing], who also taught him how to pray to Mecca [role improvising], should the need arise. ([23])
Nahim's story highlights how orchestration actions that operate through the mechanism of network experimentation help consumers personalize cocreation with strangers. Nahim reported on the incorporation of new resources into cocreated experiences ("offering disguise clothes to help him reach his goals") and described how his hosts improvised by stepping into a new role ("writing signs in Arabic for his hitchhiking" and "teaching him how to pray"), which provided value outcomes beyond the usual Couchsurfing experience.
Overall, our findings point to four overarching mechanisms of orchestration work and 14 specific actions that consumers engage in to cocreate value in the hybrid setting of sharing economy platforms. Figure 1 consolidates consumer orchestration work, detailing the actions and mechanisms that help consumers cocreate unique and valuable experiences for themselves. Our focus thus far has been on examining the nature of consumer orchestration work on platforms. As we uncovered these mechanisms, we also found that they connected to well-known sources of value for platform firms. The next section explains how orchestration work creates value for platform firms.
Graph: Figure 1. How consumers orchestrate experiences and cocreate value in the sharing economy.
As orchestration work enables value creation for platform consumers (service providers and users), we found that it also appears to activate known sources of value for platform firms. In this section, we identify how consumer orchestration work leads to efficiency, complementarities, lock-in, and novelty ([56]) in the platform environment.
Efficiency, which refers to cost savings enabled by interconnections, is one of the primary value drivers for platform firms ([56]). Orchestration work contributes to increasing the efficiencies associated with cocreating on the platform. As consumers align their expectations, interactions, and responses with those of heterogeneous cocreation partners through C2C alignment, the cost of organizing exchanges among peer-service providers and users is reduced. C2C alignment is also likely to reduce costs arising from unsuccessful cocreation partnerships (e.g., consumer dissatisfaction and defection; reputation damage). Specifically, screening enhances consumers' likelihood of finding partners who cocreate in ways that these consumers consider desirable, thereby reducing the overall effort needed to integrate resources. Cueing allows cocreation partners to signal to one another the type of experience they envision, reducing the risk of misalignment during cocreation. Flexing allows cocreation partners to negotiate access to and provision of resources to better accommodate their individual needs. Buffering ensures that experiences gone wrong are dealt with quickly, preserving some of the cocreated value and preventing the escalation of tensions. These orchestration actions make consumers' cocreation in the sharing economy more efficient and create value for the platform firm by further reducing direct and indirect costs associated with organizing exchanges among platform consumers.
Complementarities, which refer to the "value-enhancing effect of interdependencies" ([56], p. 21), are another important source of value for the platform firm. Orchestration work leads to complementarities in cocreation by spurring beneficial interdependencies among platform consumers. Rewiring relations leads to complementarities by creating opportunities for generating synergies with others in the platform. Specifically, interest grouping and lifestyle signaling build on the idea that close-knit groups can exist in the sharing economy, even though we found, in line with prior research ([ 4]), that most consumers do not consider these collectives as being communities. Such groups can provide opportunities to establish desired relationships within the collective and reconnect identity projects to cocreated experiences, adding value to these experiences. Enclaving provides actors with opportunities to experience intense, albeit temporary, communal sociality at the local level. This intense sociality is likely to enhance opportunities for value cocreation through the platform. Reconciling helps consumers use the platform connections to achieve both individual and communal goals. These orchestration actions increase complementarities by capitalizing on the existing interdependencies in the network. As the network becomes more valuable to consumers, the platform firm that hosts it may be able to capture additional value in turn.
Lock-in, which refers to "business model elements that create switching costs or enhanced incentives for business model participants to stay and transact within the activity system" ([56], p. 21), is yet another source of value for the platform firm. One important way of generating lock-in in sharing economy platforms is to reduce the risk of cocreating with strangers ([53]). By improving interpersonal trust among platform consumers, orchestration work increases the pool of potential trustworthy cocreation partners, offering additional incentives for platform consumers to stay and transact within the platform. Trust investment actions help platform consumers become better at signaling and assessing the integrity and reliability of those with whom they will cocreate. Revealing increases opportunities for consumers to promote their ability to integrate resources and cocreate with others, promoting themselves as low-risk partners. Cultivating reviews is indicative of the likelihood of the success of future collaborations with cocreation partners. Scaffolding enables cocreation partners to slowly test their ability to cocreate together and adapt their behavior to reduce the risk in extended cocreation efforts. These orchestration actions promote lock-in by increasing the level of interpersonal trust that exists among platform consumers. A higher level of interpersonal trust is a source of value for platform firms because when perceived risk in cocreation is reduced, the platform is perceived as safer ([53]), and platform consumers are likely to increase their engagement for longer periods of time.
Novelty, which refers to "degree of business model innovation that is embodied by the activity system" ([56], p. 21), is also a source of value for the platform firm. Consumer cocreation becomes a source of innovation for the platform firm when it helps the platform incorporate resources, roles, and goals that are not only novel but also meaningful to its consumers. Actions of network experimentation enable platform consumers to explore new ways of cocreating unique experiences with more personalized value outcomes. Creative resourcing encourages platform consumers to share their creative use and integration of resources, resulting in a unique experience for those involved and in novel configurations of resources for cocreation by others in the network. Role improvising helps consumers step into new cocreation roles, allowing for diverse and unique sets of scripts to be available for platform consumers. Repurposing creates new opportunities for value cocreation by introducing new value propositions into the network; when these become ubiquitous (as is the case of "sex surfing"), they allow for a different set of repurposed activities to emerge. These orchestration actions lead to value for the platform firm by aggregating new resources, scripts, and purposes to the platform offering, which, in turn, becomes attractive to a larger number of potential consumers.
Finally, although we have noted clear links between the actions of each identified mechanism—C2C alignment, rewiring relations, trust investment, and network experimentation—and well-known sources of value for the firm ([ 2]), we highlight that the actions of each orchestration mechanism can activate multiple sources of value for firms. For example, although screening has the immediate effect of creating efficiencies through the selection of suitable partners, it may also lead to further complementarities in the platform if consumers identify cocreation partners who have resources that complement their own. Screening may also promote lock-in as consumers become increasingly skilled at identifying ideal partners on the platform and are likely to remain loyal to this platform; they can then continue using their screening skills to reduce the risks associated with cocreating via the platform. For more entrepreneurial platform consumers, screening can result in novelty for the platform firm when these consumers learn how to select partners who are more likely to propose or welcome atypical experiences or disseminate their novel, cocreated experiences to others on the platform. As with screening, other orchestration actions can also be linked to more than one known source of value creation for a firm ([ 2]). The capacity of each orchestration action to impact multiple sources of value for platform firms highlights the value-creating power of consumer orchestration work.
A key marketing issue for the managers of platform firms is how best to capture the value that is cocreated by platform consumers to achieve long-term sustainability in the sharing economy ([12]). Despite the enormous potential of consumer cocreation, many platform firms still approach cocreation from a traditional business mindset; they aim to manage the onstage aspects of consumers' experiences rather than assume a backstage role and focus on developing support processes and structures ([ 9]). We propose that platform firms can improve the value they obtain from consumer collectives in the sharing economy by understanding and supporting consumer orchestration work, which is geared toward helping consumers independently overcome cocreation roadblocks.
Platform consumers engage in orchestration work to address the challenges they face while cocreating experiences in the hybrid logics of the sharing economy. To best support consumer orchestration work, managers of platform firms must identify these roadblocks and know how to leverage orchestration actions and respective mechanisms. The case of Couchsurfing, with its low level of platform control over consumers' cocreation, allows us to better understand the types of actions and mechanisms that consumers put into place to overcome these challenges and cocreate value for themselves.
By facilitating C2C alignment, platform firms can help consumers overcome the challenge of cocreating with heterogeneous partners. For example, this challenge is evident with Airbnb, which, similar to Couchsurfing, has a vast and diverse network of consumers ([26]). Airbnb encourages hosts to be upfront in their online profiles and initial message exchanges. It also encourages guests to pay attention to cues provided by hosts. Drawing on our findings, Airbnb could further support consumers in dealing with the challenge of heterogeneity by encouraging screening through additional filters that more specifically account for expectations and preferred ways of cocreation (e.g., allowing guests to indicate their desired amount of contact/conversation with the host). Airbnb could further support cueing by reminding hosts and guests to complete profile tabs associated with preferred modes of interaction (e.g., by asking questions such as "Chat over coffee or no talk before breakfast?"). Furthermore, Airbnb could encourage its consumers to engage in flexing by educating them on the need to accommodate variability while cocreating (e.g., disseminating curated, user-generated how-to videos that address common misalignment issues, such as "Three ways to deal with guests who leave your kitchen messy"). Finally, Airbnb could help consumers reduce the impact of cocreation mismatches through buffering by providing guidelines on how consumers can curtail issues and recover from negative experiences (e.g., crowdsourcing a flowchart for common cocreation problems and ways to address them). These initiatives would allow this platform firm to increase opportunities for its consumers to pursue cocreation efficiencies while maintaining orchestration work under consumers' control.
Platform firms can help consumers overcome the challenge of networked sociality by helping them rewire relations for cocreation. For example, Tinder successfully developed a platform for high-involvement customer experiences ([42]), yet it has been criticized by consumers and the media for making someone's search for a partner too transactional. Tinder could address these critiques by helping its consumers reconcile the app's transactional logics (e.g., swipe left/right) with communal logics and the desire for more meaningful relationships. It could, for example, assist consumers with using the platform for interest grouping (e.g., offering tags for members who are into pet walking or cooking together), and assist consumers in lifestyle signaling by providing a means for them to connect over preferred hobbies or passions (e.g., forums for those into watching and discussing movies about art). It could also enable enclaving by supporting the creation of spaces that enable closer relationships among platform consumers (e.g., helping entrepreneurial consumers promote local singles' face-to-face night and facilitate these events by enabling communication about them on the platform). Tinder could also support consumers in their work of reconciling personal goals with the need for strengthening communal bonds. For instance, Tinder could create a forum for storytelling about successful and failed hookups, a bonding activity that often encourages sociality and sharing (communal goals) while equipping consumers to become better at finding dates for themselves (transactional). This type of forum currently happens outside this platform (e.g., on Reddit), which demonstrates that consumers already engage in orchestration work.
Overall, these initiatives could help consumers better utilize the power of social relations to address both their quid pro quo hookups and their need for meaningful social connections. Couchsurfing and other platforms can assist consumers in overcoming the challenge of interpersonal trust by helping them reduce the risk of cocreating experiences with strangers. At Uber, for instance, the perceived lack of authenticity (withholding one's true self) is a major challenge in building interpersonal trust ([15]).
While trust mechanisms such as protection insurance and identity verifications are fundamental to raising consumer trust in the platform and its consumers ([25]), platform firms can increase interpersonal trust by supporting and leveraging the trust investments made by consumers to overcome their interpersonal trust issues. Thus, supporting consumers' work to reveal more about themselves through profiles, tags, and prompts, safely and progressively (scaffolding), and helping them cultivate authentic reviews about their cocreation behavior, could increase interpersonal trust among consumers of the Uber platform.
Similarly, BlaBlaCar, a ride-sharing platform that matches empty car seats with potential passengers looking for long-distance rides, has a well-researched history of helping platform consumers (drivers and riders) overcome the interpersonal mistrust that usually exists when cocreating with strangers. BlaBlaCar supports revealing by encouraging consumers to share more information about themselves in their profiles and to connect this information with their existing online identity (e.g., Facebook profile). Research has found that 88% of BlaBlaCar consumers highly trust a member who has a complete digital profile, which is higher than the trust levels of colleagues or neighbors ([27]).
Our findings suggest that BlaBlaCar could further boost revealing by encouraging consumers to share more specific information about their cocreation preferences during interactions (e.g., ritualizing the sharing of cocreation stories as icebreakers, offering platform-specific personality quizzes, allowing consumers to associate their profile with a particular consumer persona—chosen from a pool of existing personas). Currently, BlaBlaCar asks its consumers to rate one another after having shared higher-stakes, real-life, offline experiences. Our findings suggest that BlaBlaCar could further leverage consumers' work of cultivating reviews by providing badges that consumers could win or gift each other for each review, or by allowing consumers to create additional questions to be asked by their reviewers. This would help future cocreation partners identify behavioral consistency (e.g., if John has great taste in music, he could get a "great music" badge and ask partners, "What did you think of your driver's soundtrack?" rather than the generic "What did you think about this experience?").
BlaBlaCar currently encourages its consumers to speak on the phone before agreeing on a ride. We suggest that the platform could further support consumers' scaffolding work by highlighting opportunities for them to progress into more trustworthy relationships with others in the platform. For example, BlaBlaCar could use notifications to prompt cocreation partners to discuss things ahead of time that could potentially lead to pain points along the customer journey (e.g., after agreeing on a ride and a day before the trip) to allow interpersonal trust to be built through multiple interactions. Overall, these initiatives would support consumer orchestration work aimed at increasing interpersonal trust among the platform's consumers.
Finally, platforms can help consumers overcome the challenge of personalizing cocreation with strangers by supporting them as they try new resources, roles, and goals when collaborating with others to cocreate experiences and, thus, extend the possibilities of creating unique and valuable experiences for themselves and others in the platform. Platform firms often propose new features and processes to continuously innovate and personalize the customer experience. Platforms can achieve these goals faster by leveraging the network experimentation conducted by consumers in their quest to create unique personalized experiences. For example, TaskRabbit, a platform that connects consumers who need help with local workers and specialists (e.g., plumbers, translators), has maintained a broad definition of what tasks can be offered or hired through the platform (thus helping consumers' innovations) and adopted some innovative skills offered by consumers (e.g., offers to wait in line for others) as standard categories in the platform. To further leverage consumers' improvising actions, TaskRabbit could confer badges to consumers who successfully break role boundaries, encourage consumers to identify the innovators among their cocreation partners, offer alternative profile categories (beyond their general label for service providers ["taskers"]), prompt consumers to teach each other something new while interacting, and encourage them to become local ambassadors.
To further support consumers in personalizing their cocreation, TaskRabbit should encourage consumers to engage in creative resourcing by actively prompting them to innovate or highlight the presence of new resources through badges, tags, and curations. For example, taskers, to describe all of their resources, often find workarounds to the limited options available on their platform profiles (e.g., offering American Sign Language as a skill-for-hire to showcase their language fluency). The platform could help consumers share these hacks with others through the TaskRabbit app or create a customizable language tag (rather than tags only for the most popular languages). Furthermore, TaskRabbit could support consumers' repurposing actions. For example, consumers may search for friendship or companionship through the platform, and the firm could create categories of social tasks to reflect these alternative purposes. Overall, by developing business models that support consumer experimentation actions, platform firms in the sharing economy can facilitate consumers' cocreation of experiences that are unique and valuable to participants. These innovations can eventually be picked up by the firm and help it continuously offer novelty to its stakeholders.
In the sharing economy, business models vary in terms of how much control they allow consumers to have over their cocreation activities. The decision of how much control to give to consumers is often a strategic one. The platform firm may prefer to keep a tight grip on consumers' interactions as a way of reducing costs, optimizing efficiencies, and enabling preset complementarities among partners. This type of firm-led orchestration of experiences works well in contexts in which experiences are expected to be similar or consistent (e.g., Kickstarter) and/or those in which heterogeneity does not affect the outcome of experiences (e.g., Lending Club). This also works well when sociality is not important (e.g., Zipcar), interpersonal trust is not required (e.g., Quirky), or experimentation is not valued (e.g., Lime). In these cases, the firm-led orchestration of experiences via traditional marketing controls (e.g., standard service quality metrics) should be preferred, as this saves costs and enables the firm to harvest the value of maximizing resource allocation and business scaling.
Other platform business models thrive on enabling unique experiences (e.g., Airbnb). Many require high consociality ([34]), as consumers need interaction to cocreate experiences (e.g., Uber); this often demands interpersonal trust (e.g., BlaBlaCar, TaskRabbit) and the alignment of consumers' expectations, actions, and responses (e.g., Tinder). It is for these groups of platform firms that consumer-led orchestration work dominates and the findings of this research are most relevant.
Orchestration work, although performed by consumers, can be influenced in important ways by the affordances of the platform. For example, the specific way in which C2C alignment manifests depends on the search filters, algorithms, and peer-to-peer communication tools available on the platform. Orchestration work is also shaped by how the platform firm incentivizes or limits consumers' interactions outside the platform. Importantly, Couchsurfing is a low-intermediation platform ([34]), in that it does not interfere much with what consumers are doing.
These findings align with [34] contention that marketers, to effectively hone and market their value propositions, must understand their market stakeholders. We support and extend their claim that each platform type has "particular forms of value creation that should focus managers' business investment decisions and resource deployment" ([34], p. 33) by considering how consumer-led value cocreation impacts each type differently. We argue that the platform types with business models that depend on a high degree of consumer interaction, such as Couchsurfing, Airbnb, and the others that [34] call "forums" and "matchmakers," are most likely to profit from delivering unique experiences. These platforms should focus on leveraging consumer-led orchestration actions by way of the mechanisms we have outlined. In contrast, platforms with a limited degree of consumer interaction—for example, Kickstarter and other "enablers" and "hubs" ([34], p. 20)—have less to harvest from consumer-led orchestration. In these cases, platform forms should be selective in terms of how they engage with consumer-led orchestration. These firms should consider whether each source of value creation (i.e., novelty, efficiency, lock-in, and complementarities) they wish to foster is better activated by consumer- or firm-led orchestration.
In line with prior research that examines how consumers cocreate value within a consumer collective (e.g., Schau, Muñiz, and Arnould 2009), we focus on identifying what consumers are doing to create value for themselves, and we see value for the firm as a welcomed, if not desired, outcome of consumer activity. However, the sharing economy's consumer collectives differ from the more communal collective contexts previously explored in the literature, such as online brand communities (Schau, Muñiz, and Arnould 2009) and interest-based communities ([ 7]; [46]). As such, most prior research has examined value cocreation in consumer collectives as an outcome of shared practices within the collective (e.g., [17]; [19]). In contrast, we identify the challenges to cocreation that are intrinsic to sharing platform collectives and map the emergent orchestration actions that consumers enact to surpass the roadblocks in cocreation and to have more valuable experiences. Thus, the orchestration work we identify here does not directly cocreate value. Instead, by overcoming the challenges to cocreation that are characteristic of these sharing platforms, these orchestration actions enable the conditions for the cocreation of value.
We also show how by engaging in orchestration work to enable the cocreation of value for themselves, the enterprising consumers ([14]) who cocreate in the sharing economy also support value creation for platform firms. Going beyond prior research, we explain how each orchestration mechanism may lead to multiple sources of value creation that are at the core of platform business models ([56]). This newly identified type of consumer entrepreneurial work advances the understanding of prosumer work (see, e.g., [57]). In orchestration work, platform consumers (both service providers and users) enact forms of governance that are traditionally associated with marketing managers. Through orchestration actions, enterprising consumers take control of the platform's value-creating activities.
Consumer orchestration work then explains how it is possible for platforms, such as Couchsurfing, to function well and generate value for millions of heterogeneous participants who are guided by hybrid logics despite maintaining limited control over the quality of service providers and consumers' experiences ([12]). We show how, for example, through trust investment orchestration actions (e.g., revealing additional information, cultivating reviews, scaffolding interactions), consumers complement platform-based mechanisms for reducing risk (e.g., rating systems, insurance for payments or damages) with interpersonal trust. Considering the myriad of ways in which consumers address trust-related roadblocks can extend knowledge on the interdependency of platforms' and consumers' reputations in the sharing economy ([11]).
We also note how, in the realm of cocreated experiences in the sharing economy, innovation is emergent ([35]); it comes largely from consumers' activity as they ideate, collaborate, and experiment in the network to develop solutions tailored to their needs ([ 1]). Our findings highlight the serendipitous nature of consumer orchestration, which favors the discovery of value potential in the collective ([31]), in turn bringing novelty to the consumer collective and the platform firm. This way of understanding innovation as an outcome of the orchestration of experiences can address recent calls for research on the specific mechanisms that drive consumer-led innovation in the sharing economy ([12]). In Table 3, we outline directions for future research on various areas connected to value cocreation by consumers.
Graph
Table 3. Future Research Directions.
| Research Areas | Directions for Future Research |
|---|
| Value cocreation in digital platforms | How does consumers' engagement in orchestration influence consumers' assessments of the value of cocreated experiences? How can consumer-led orchestration support systemic value creation (Figueiredo and Scaraboto 2016)? Are there instances in which consumer orchestration may lead to value destruction for the firm (Echeverri and Skålén 2011)? How can marketing controls be employed to help consumers overcome the four challenges of hybrid value cocreation logic? How can mechanisms of network experimentation be fostered as market-shaping devices for innovation (Nenonen, Storbacka, and Windahl 2019)? How can consumer orchestration be leveraged to more efficiently address consumers' needs, minimize resource waste, and turn consumer orchestration into a force for the social good (Rinne 2018)? How can the mechanism of trust investment be used to help build a digital reputation (Eckhardt 2020)? What sort of digitalized networked arrangements of artifacts, persons, processes, and interfaces might best support orchestration work (Ramaswany and Ozcan 2018)?
|
| Consumer experience in the sharing economy | How can firms design experiences with the ideal balance between company- and consumer-led orchestration to achieve optimal value creation? What are some pain points of the customer journey associated with the challenges of cocreating in the sharing economy? How can consumer-led orchestration be deployed to generate service innovations in the sharing economy (Tronvoll and Edvardsson 2020)?
|
| Sharing economy governance and policy | How has the onus of orchestrating cocreation been transferred to consumers as free work in the sharing economy (Zwick, Bonsu, and Darmody 2008)? How does orchestration work shape customers' risk perceptions in the sharing economy (Wang, Ma, and Wang 2020)? What economic and noneconomic rewards can be offered to consumers engaged in orchestration, thereby recognizing their engagements (Pazaitis, De Filippi, and Kostakis 2017)? Under what conditions are consumers more likely to engage in orchestration work or enact one type of action versus another? How can these conditions best be shaped? How does orchestration work influence consumers' relation to the platform (Zhou et al. 2021)? Does increased trust among cocreation partners lead to increased trust on the platform (Möhlmann and Geissinger 2018)? How can consumer orchestration in the sharing economy inform the regulation of business models in the sharing economy?
|
This article makes important contributions to research in marketing. First, it discusses the challenges that the hybrid logics of the sharing economy raise for consumers who are cocreating experiences. Second, it identifies orchestration as an important form of work conducted by platform consumers (service users and providers) to overcome these challenges. Third, it connects consumer orchestration work to known sources of value for platform firms and provides recommendations for platform firms to leverage the power of orchestration work. Finally, it offers a series of research questions that can inspire marketing researchers to further explore how consumers cocreate in the sharing economy.
Footnotes 1 Daiane Scaraboto and Bernardo Figueiredo contributed equally to the article.
2 Amber Epp
3 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
4 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The authors thank the Marketing Science Institute for funding the pilot study for this project through the MSI Research Grant #4-2002 MSI Customer Experience Initiative.
5 Daiane Scaraboto https://orcid.org/0000-0002-7658-9339
6 Online supplement:https://doi.org/10.1177/00222429211027777
7 Emic terms are indicated with italics in this section.
8 Types of actions are noted within brackets in this and other long quotes.
References Abhari Kaveh , Davidson Elizabeth J. , Xiao Bo. (2019), " Collaborative Innovation in the Sharing Economy: Profiling Social Product Development Actors Through Classification Modelling ," Internet Research , 29 (5), 1014 – 39.
Amit Raphael , Zott Christoph. (2001), " Value Creation in E-Business ," Strategic Management Journal , 22 (6/7), 493 – 520.
Arnould Eric J. , Rose Alexander S.. (2016), " Mutuality: Critique and Substitute for Belk's 'Sharing,' " Marketing Theory , 16 (1), 75 – 99.
Bardhi Fleura , Eckhardt Giana M.. (2012), " Access-Based Consumption: The Case of Car Sharing ," Journal of Consumer Research , 39 (4), 881 – 98.
Botsman Rachel. (2017), Who Can You Trust? How Technology Brought Us Together and Why It Could Drive Us Apart. London : Penguin.
Botsman Rachel. (2019), " Trust Leaps," Medium (February 20), https://medium.com/@rachelbotsman/trust-leaps-bae279d841a.
Caru Antonella , Cova Bernard. (2015), " Cocreating the Collective Service Experience ," Journal of Service Management , 26 (2), 276 – 94.
Charmaz Kathy. (2014), Constructing Grounded Theory. London : Sage Publications.
9 Chen Tom , Drennan Judy , Andrews Lynda , Hollebeek Linda D.. (2018), " User Experience Sharing: Understanding Customer Initiation of Value Co-Creation in Online Communities ," European Journal of Marketing , 52 (5/6), 1154 – 84.
Echeverri Per , Skålén Per. (2011), " Co-Creation and Co-Destruction: A Practice Theory-Based Study of Interactive Value Formation ," Marketing Theory , 11 (3), 351 – 73.
Eckhardt Giana M. (2020), " Playing the Trust Game Successfully in the Reputation Economy ," NIM Marketing Intelligence Review , 12 (2), 10 – 17.
Eckhardt Giana , Houston Mark , Jiang Baojun , Lamberton Cait , Rindfleisch Aric , Zervas Georgios. (2019), " Marketing in the Sharing Economy ," Journal of Marketing , 83 (5), 5 – 27.
Figueiredo Bernardo , Scaraboto Daiane. (2016), " The Systemic Creation of Value Through Circulation in Collaborative Consumer Networks ," Journal of Consumer Research , 43 (4), 509 – 33.
Fischer Eileen. (2020), "Association for Consumer Research Presidential Address," speech presented at the ACR Conference, Paris (accessed April 13, 2021), https://vimeo.com/465479728.
Frei Frances X. , Morriss Anne. (2020), " Begin with Trust ," Harvard Business Review , 98 (3), 112 – 21.
Frenken Koen , Schor Juliet. (2019), " Putting the Sharing Economy into Perspective, " in A Research Agenda for Sustainable Consumption Governance , Mont O. , ed. Chap. 8, Cheltenham, UK : Edward Elgar Publishing , 121 – 35.
Hartmann Benjamin J. , Wiertz Caroline , Arnould Eric J.. (2015), " Exploring Consumptive Moments of Value-Creating Practice in Online Community ," Psychology & Marketing , 32 (3), 319 – 40.
Helkkula Anu , Kelleher Carol , Pihlström Minna. (2012), " Practices and Experiences: Challenges and Opportunities for Value Research ," Journal of Service Management , 23 (4), 554 – 70.
Kelleher Carol , Wilson Hugh N. , Macdonald Emma K. , Peppard Joe. (2019), " The Score Is Not the Music: Integrating Experience and Practice Perspectives on Value Cocreation in Collective Consumption Contexts ," Journal of Service Research , 22 (2), 120 – 38.
Kocher Bruno , Morhart Felicitas , Zisiadis George , Hellwig Katharina. (2014), " Share Your Life and Get More of Yourself. Experience Sharing in Couchsurfing, " in Advances in Consumer Research, 42 , Cotte J. , Wood S. , eds. Duluth, MN : Association for Consumer Research , 510 – 11.
Kozinets Robert V. (2002), " Can Consumers Escape the Market? Emancipatory Illuminations from Burning Man ," Journal of Consumer Research , 29 (1), 20 – 38.
Kozinets Robert V. (2019), Netnography Revisited: Doing Ethnographic Research Online , 3rd ed. London : SAGE Publications.
Litchfield Liam. (2013), "On the Road: The Story of a Couchsurfing Drifter," The Beijinger (January 3), https://www.thebeijinger.com/blog/2013/01/03/road-story-couchsurfing-drifter.
Locke Karen , Feldman Martha , Golden-Biddle Karen. (2020), " Coding Practices and Iterativity: Beyond Templates for Analyzing Qualitative Data," Organizational Research Methods , (published online August 24), https://doi.org/10.1177/1094428120948600.
Luo Xueming , Tong Siliang , Lin Zhijie , Zhang Cheng. (2021), " The Impact of Platform Protection Insurance on Buyers and Sellers in the Sharing Economy: A Natural Experiment ," Journal of Marketing , 85 (2), 50 – 69.
Makkar Marian , Yap Sheau-Fen (Crystal). (2020), " Managing Hearts and Minds: Romanticizing Airbnb Experiences," Current Issues in Tourism (published online July 14), https://doi.org/10.1080/13683500.2020.1792855.
Mazzella Frédéric , Sundararajan Arun , d'Espous Verena B. , Möhlmann Mareike. (2016), " How Digital Trust Powers the Sharing Economy ," IESE Business Review , 26 (5), 24 – 31.
Minkiewicz Joanna , Evans Jody , Bridson Kerrie. (2014), " How Do Consumers Cocreate Their Experiences? An Exploration in the Heritage Sector ," Journal of Marketing Management , 30 (1/2), 30 – 59.
Möhlmann Mareike , Geissinger Andrea. (2018), " Trust in the Sharing Economy: Platform-Mediated Peer Trust, " in The Cambridge Handbook of the Law of the Sharing Economy , Davidson N. , Finck M. , Infranca J. , eds. Cambridge, UK : Cambridge University Press , 26 – 44.
Morgan Robert M. , Hunt Shelby D.. (1994), " The Commitment-Trust Theory of Relationship Marketing ," Journal of Marketing , 58 (3), 20 – 38.
Nenonen Suvi , Storbacka Kaj , Windahl Charlotta. (2019), " Capabilities for Market-Shaping: Triggering and Facilitating Increased Value Creation ," Journal of the Academy of Marketing Science , 47 (4), 617 – 39.
Paoletta Phil. (2010), " Why You Should Be Using Couchsurfing Groups," Go Backpacking (October 5), https://gobackpacking.com/couchsurfing-groups/.
Pazaitis Alex , De Filippi Primavera , Kostakis Vasilis. (2017), " Blockchain and Value Systems in the Sharing Economy: The Illustrative Case of Backfeed ," Technological Forecasting and Social Change , 125 , 105 – 15.
Perren Rebecca , Kozinets Robert V.. (2018), ��� Lateral Exchange Markets: How Social Platforms Operate in a Networked Economy ," Journal of Marketing , 82 (1), 20 – 36.
Peters Linda D. (2016), " Heteropathic Versus Homopathic Resource Integration and Value Cocreation in Service Ecosystems ," Journal of Business Research , 69 (8), 2999 – 3007.
Prahalad Coimbatore K. , Ramaswamy Venkat. (2004), " Cocreation Experiences: The Next Practice in Value Creation ," Journal of Interactive Marketing , 18 (3), 5 – 14.
Ramaswamy Venkat , Ozcan Kerimcan. (2018), " Offerings as Digitalized Interactive Platforms: A Conceptual Framework and Implications ," Journal of Marketing , 82 (4), 19 – 31.
Rinne April. (2018), " The Dark Side of the Sharing Economy," World Economic Forum (January 16), https://www.weforum.org/agenda/2018/01/the-dark-side-of-the-sharing-economy/.
Scaraboto Daiane. (2015), " Selling, Sharing, and Everything in Between: The Hybrid Economies of Collaborative Networks ," Journal of Consumer Research , 42 (1), 156 – 76.
Schau Hope J. , Muñiz Albert M. Jr. , Arnould Eric J.. (2009), " How Brand Community Practices Create Value ," Journal of Marketing , 73 (5), 30 – 51.
Sheepshanks Octavia. (2013), " Couchsurfing: More Than Just a Free Bed for the Night," The Independent (August 8), http://www.independent.co.uk/student/student-life/couchsurfing-more-than-just-a-free-bed-for-the-night-8751700.html.
Siebert Anton , Gopaldas Ahir , Lindridge Andrew , Simões Cláudia. (2020), " Customer Experience Journeys: Loyalty Loops Versus Involvement Spirals ," Journal of Marketing , 84 (4), 45 – 66.
Spiggle Susan. (1994), " Analysis and Interpretation of Qualitative Data in Consumer Research ," Journal of Consumer Research , 21 (3), 491 – 503.
Sundararajan Arun. (2016), The Sharing Economy: The End of Employment and the Rise of Crowd-Based Capitalism. Cambridge, MA : MIT Press.
Sundararajan Arun. (2019), " Commentary: The Twilight of Brand and Consumerism? Digital Trust, Cultural Meaning, and the Quest for Connection in the Sharing Economy ," Journal of Marketing , 83 (5), 32 – 5.
Thomas Tandy Chalmers , Price Linda L. , Schau Hope Jensen. (2013), " When Differences Unite: Resource Dependence in Heterogeneous Consumption Communities ," Journal of Consumer Research , 39 (5), 1010 – 33.
Thornton Patricia H. , Ocasio William , Lounsbury Michael. (2012), The Institutional Logics Perspective: A New Approach to Culture, Structure, and Process. Oxford, UK : Oxford University Press.
Tönnies Ferdinand. (2001), Community and Civil Society , Harris J. , ed. Cambridge, UK : Cambridge University Press.
Tronvoll Bård , Edvardsson Bo. (2020), "Explaining How Platforms Foster Innovation in Service Ecosystems," paper presented at Hawaii International Conference on System Sciences (HICSS) , (June 10), https://aisel.aisnet.org/hicss-53/da/service_science/3/.
Turner Victor W. (1969), The Ritual Process. London : Lowe & Brydone.
Vargo Stephen L. , Lusch Robert F.. (2016), " Institutions and Axioms: An Extension and Update of Service-Dominant Logic ," Journal of the Academy of Marketing Science , 44 (1), 5 – 23.
Von Richthofen Georg , Fischer Eileen. (2019), " Airbnb and Hybridized Logics of Commerce and Hospitality, " in Handbook of the Sharing Economy , Belk R.W. , Eckhardt G.M. , Bardhi F. , eds. Northampton, MA : Edward Elgar Publishing , 183 – 207.
Wang Shuai , Ma Shuang , Wang Yonggui. (2020), "The Role of Platform Governance in Customer Risk Perception in the Context of Peer-to-Peer Platforms," Information Technology for Development (published online October 2), https://doi.org/10.1080/02681102.2020.1827364.
Wittel Andreas. (2001), "Toward a Network Sociality," Theory , Culture & Society , 18 (6), 51 – 76.
Zhou Qiang (Kris) , Allen B.J. , Gretz Richard T. , Houston Mark B.. (2021), " Platform Exploitation: When Service Agents Defect with Customers from Online Service Platforms ," Journal of Marketing (published online February 22), https://doi.org/10.1177/00222429211001311.
Zott Christoph , Amit Raphael. (2017), " Business Model Innovation: How to Create Value in a Digital World ," GfK Marketing Intelligence Review , 9 (1), 18 – 23.
Zwick Detlev , Bonsu Samuel , Darmody Aron. (2008), " Putting Consumers to Work: Cocreation and New Marketing Govern-Mentality ," Journal of Consumer Culture , 8 (2), 163 – 96.
~~~~~~~~
By Daiane Scaraboto and Bernardo Figueiredo
Reported by Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 65- "How Do I Carry All This Now?" Understanding Consumer Resistance to Sustainability Interventions. By: Gonzalez-Arcos, Claudia; Joubert, Alison M.; Scaraboto, Daiane; Guesalaga, Rodrigo; Sandberg, Jörgen. Journal of Marketing. May2021, Vol. 85 Issue 3, p44-61. 18p. 4 Color Photographs, 3 Diagrams, 1 Chart. DOI: 10.1177/0022242921992052.
- Database:
- Business Source Complete
"How Do I Carry All This Now?" Understanding Consumer Resistance to Sustainability Interventions
Given the increasingly grave environmental crisis, governments and organizations frequently initiate sustainability interventions to encourage sustainable behavior in individual consumers. However, prevalent behavioral approaches to sustainability interventions often have the unintended consequence of generating consumer resistance, undermining their effectiveness. With a practice–theoretical perspective, the authors investigate what generates consumer resistance and how it can be reduced, using consumer responses to a nationwide ban on plastic bags in Chile in 2019. The findings show that consumer resistance to sustainability interventions emerges not primarily because consumers are unwilling to change their individual behavior—as the existing literature commonly assumes—but because the individual behaviors being targeted are embedded in dynamic social practices. When sustainability interventions aim to change individual behaviors rather than social practices, they place excessive responsibility on consumers, unsettle their practice-related emotionality, and destabilize the multiple practices that interconnect to shape consumers' lives, ultimately leading to resistance. The authors propose a theory of consumer resistance in social practice change that explains consumer resistance to sustainability interventions and ways of reducing it. They also offer recommendations for policy makers and social marketers in designing and managing sustainability initiatives that trigger less consumer resistance and thereby foster sustainable consumer behavior.
Keywords: consumer resistance; practice theory; social change; sustainable consumer behavior; sustainability intervention
The battle is not just being fought over the fate of a familiar modern convenience but over, for one side, our last vestiges of freedom and, for the other, the future of planet Earth. And fluttering above this battlefield like the tattered banner of a besieged army, amid a haze of misinformation, counter-arguments, and money, money, money, you'll find a single, flimsy, humble plastic bag.
One of the most important questions today for governments, marketers, and policy makers is how to foster sustainable consumer behavior. However, efforts to encourage sustainable consumer behavior with interventions such as water restrictions ([30]) and fees for using disposable coffee cups ([32]) often meet various forms of consumer resistance ([14]; [40]). Understanding why consumer resistance emerges is critical because such resistance undermines the effectiveness of sustainability interventions ([23]) and has significant implications for companies, consumers, and policy makers.
Although highly diverse and varied in scope, prevalent approaches to sustainability interventions often center on changing individual consumer behaviors ([20]). Early on, these approaches focused on the diffusion and adoption of planned social changes to convince individual consumers to alter their behavior (e.g., [21]). More recently, research in marketing and behavioral science that investigates behavioral, attitudinal, psychological, and social barriers to or drivers of behavioral change has informed policy to encourage individual consumers to act more sustainably ([19]; [29]; [54]; [55]). However, as [54], p. 34) note, sustainability interventions need to be embraced by large groups of people, as they differ "from traditional consumer behaviors in which the outcome is realized if the individual engages in the action alone." In this sense, individual resistance to behavioral change might arise due to habit ([51]), but sustainability interventions also provoke resistance when consumers reject "what is perceived as a power, a pressure, an influence, or any attempt to act upon one's conduct" ([35], p. 295).
Our purpose is to investigate consumer resistance in a sustainability context, defined as "the refusal to accept or support a sustainability intervention." We ask: What gives rise to consumer resistance to sustainability interventions, and how can consumer resistance be reduced? We approach these questions from a practice–theoretical perspective ([43]), which proposes that consumer behavior is not primarily determined by the individual but by the social practices through which they conduct their daily lives (e.g., eating, cooking, shopping). By conceiving of individual consumer behaviors as embedded in dynamic social practices, we can better understand how and why sustainability interventions are likely to face consumer resistance and ultimately fail.
We conducted a comprehensive, real-time study of Chile's 2019 nationwide ban on plastic bags. The ban was met with a high level of consumer resistance, evidenced by public manifestations of consumer resentment and extensive media coverage of consumers' refusal to accept the intervention. It constitutes a compelling case for investigating our research questions. Our findings show that consumer resistance to sustainability interventions emerges because the individual behaviors being targeted are not separate from, but rather embedded in, social practices. When interventions aim for individual behavioral change rather than social practice change, three major challenges emerge: ( 1) battles about who is responsible for making practices more sustainable, ( 2) unsettling emotionality brought about by the changing practice, and ( 3) the (un)linking of other practices involved in the change. These challenges generate consumer resistance that interferes with social practice change, which significantly undermines the effectiveness of the sustainability intervention.
We develop a theory of consumer resistance in social practice change that explains how the aforementioned challenges give rise to consumer resistance to sustainability interventions and how this resistance can be reduced. Drawing on our theory, we prescribe recommendations for policy makers and social marketers regarding how to design practice-based sustainability interventions to reduce resistance from the outset, as well as how to monitor and adjust these interventions to manage consumer resistance that may emerge later.
Marketing literature pertaining to sustainable consumer behavior has coalesced in its focus on how individual consumers should change their behaviors to be more sustainable ([20]). Early social marketing studies provided the foundations for this approach by conceptualizing sustainability as a "planned social change process" (e.g., [21]). More recently, the behavioral literature has profiled the behaviors of green consumers, informing the design of marketing interventions to encourage the adoption of relevant actions ([29]), such as choosing sustainably sourced products, conserving resources, and seeking more sustainable product disposal modes ([54]).
Research concludes that consumers will engage in more sustainable behaviors in response to specific messages ([29]; [56]), normative appeals ([55]), and priming ([19]). [54] SHIFT framework identified five psychological factors—social influence, habit formation, individual self-accounts, feelings and cognition, and tangibility—that can be leveraged in sustainability interventions. However, researchers also note the potential for obstacles, such as conflicts between sustainable behaviors and private goals ([22]), as well as skepticism, lack of support, or perceptions of unfairness ([ 4]).
Although they increase our understanding of sustainable consumer behavior, these approaches have tended to be based on an "individualistic understanding of both action and change" ([43], p. 142) that neglects the complex systems in which environmental issues are embedded ([23]). Furthermore, many sustainability interventions make individual consumers responsible for societal issues such as climate change and poverty ([ 2]; [12]; [13]; [24]; [42]). This approach, known as "responsibilization," is based on neoliberal ideology and involves government partnership with corporations to "encourage all citizens to become active and responsible consumer subjects...obliged to help solve pressing social issues through their everyday consumption choices" ([50], p. 255). Responsibilization assumes that individual consumers want to act responsibly and make moral choices to support an intervention's intended goals ([ 2]). However, consumers often resist such responsibilization ([10]), particularly when they experience physical, psychological, and/or philosophical discomfort. Therefore, effective sustainability interventions may require a shift away from responsibilizing individual consumers and toward shaping the social elements and systems of daily life, as implied by a practice–theoretical perspective ([45]).
Although several different theoretical approaches exist within the practice perspective (e.g., [36]; [39]; [47]), they all recognize that people, animals, materials, equipment, activities, norms, rules, values, and understandings are not independent but interacting units that constitute social practices and their performance ([33]). Social practices comprise "temporally evolving, open-ended sets of doings and sayings linked by practical understandings, rules, teleoaffective structure, and general understanding" ([37], p. 87). Continuous engagement in social practices, such as eating, cooking, shopping, driving, and reading, largely determines people's way of life and who they are ([36]). From this perspective, "behaviors are largely individuals' performances of social practices" ([45], p. 4). To apply such a perspective to sustainable consumer behavior, we build on [43] theory of the dynamics of social practice, which features five key premises.
First, social practices and their performance involve three broad groups of interacting elements: materials (e.g., equipment, tools, ingredients, bodies), competences (e.g., specific know-how, skills, shared practical understandings), and meanings (e.g., identities, symbols, norms, aspirations, ideas). Social practices depend on the interactions of these defining elements and thus cannot "be reduced to any one of these single elements" ([33], p. 250). Only when the elements are linked together, consistently and over time, do social practices come into existence and endure. Therefore, social practices are not fixed. Instead, these practices are dynamic, as they are produced and reproduced through their performance over time ([43]). As an illustration, the social practice of communicating with mobile phones comprises materials (e.g., phones, bodies, touchable screen), competences (e.g., typing, dialing, taking turns to speak, knowing proper times to call), and meanings (e.g., social closeness, convenience) that are linked every time someone makes a call.
Second, social practices are continuously carried out by multiple actors. Consumers, retailers, and other market actors are social practice "carriers" ([33]). As carriers, these actors produce, reproduce, and transform social practices by continuously linking elements in their performances of them ([ 3], p. 38). The practice's goals, meanings, and materials direct carriers to perform the doings and sayings of a given practice in specific ways ([37]). For example, mobile phone users reproduce the practice of mobile communication, and the practice influences how users communicate with friends, family, and colleagues (e.g., via texting).
Third, social practices also evolve and change through the making and breaking of links among their defining elements ([43]). Links are made and broken as a result of the introduction of new elements or the removal of existing ones. Such alterations require carriers to reconfigure the elements—that is, to develop and establish new links between them—for the practice to stabilize and endure. It is through this process of reconfiguring the links between modified elements across carriers that social practices evolve over time. For example, the introduction of mobile phones (new material element) changed the social practice of communicating. Mobile phones altered not only the material elements of the communication practice but also all its interacting elements, such as consumers' competences for handling mobile phones and the shared understanding of how and when communication should be performed.
Fourth, rather than existing in isolation, social practices are linked to other practices, forming nexuses of interacting practices ([17]) that together make up social life ([33]). Changes to some elements of a particular social practice therefore may require a reconfiguration of both its interacting elements and other, linked social practices. In the mobile phone example, the introduction of the new material element changed purchasing, repairing, emailing, and family practices (e.g., family video calls), each of which demanded new tools, skills, and know-how to perform.
Fifth, social practices have an inherent emotional dimension ([38]), "tied to the embodied and tacit aspects of everyday living" ([27], p. 372). This dimension provides practice carriers with a template for the acceptable beliefs, states, and feelings that they should express as part of the practice. Returning to our prior example, replacing landline phones with mobile devices altered the emotionality associated with different communication practices. Many users now regard voice calls negatively, as anxiety-inducing or intrusive, but text messages evoke more positive emotions related to efficiency or self-control.
These five tenets of practice theory highlight how practices can guide social life and consumption ([52]) and form the basis of our inquiry into consumer resistance to sustainability interventions in several key ways. First, this perspective considers the complexity associated with changing a ubiquitous social practice like shopping, which is linked to and intertwined with many other practices. Second, in this perspective, social practices constantly change and evolve, but their histories never disappear entirely ([43]). Carriers might draw on these histories and either adapt or fail to reconfigure practices when elements in a practice are misaligned ([30]; [47]). Third, this perspective allows us to consider resistance as an activity that interferes with the social practice change that is required by interventions. While individual in nature, such resistance can aggregate to cause even greater levels of disruption to the reconfiguration of the targeted practice ([53]). Finally, although [43] highlight that practices can change, the specific processes by which carriers reconfigure links and thereby change social practices remain unclear. Our findings extend this perspective to address this gap.
Plastic bags are a common target of sustainability interventions ([18]). Although they have become a symbol of an ecological crisis ([16]), plastic bags reached this status due to their mundane and widespread use ([46]). As an essential material element of the shopping practice, the bags also shape other practices, such as carrying, transporting, advertising, disposing of, and selling products ([15]).
~~~~~~~~
By July 2018, 127 countries had adopted restrictions on plastic bags, with laws that targeted their manufacture, retail distribution, use, and trade ([48]). Chile was the first South American country to ban the use of plastic bags nationally. Chilean policy makers argued the ban was simpler than other interventions that would require participation by stakeholders other than consumers (e.g., waste generators, producers' recycling efforts).[ 6] Thus, they began regulating the use of plastic bags in coastal areas in 2013 while initiating discussions of a nationwide ban. The law, approved in August 2018, applied throughout the country without exceptions ([ 9]). It required retailers to stop offering plastic bags to customers ([26]), which was done in two stages. During the first six-month adaptation period, retailers could provide two plastic bags per customer, and then the total ban was initiated in February 2019.
It may be tempting to assume that the implementation of the second-stage total ban signaled the success of the intervention, but our findings indicate this was not the case. As in many countries,[ 7] the ban prompted resistance in Chile ([ 8]), and some consumers struggled to accept, adjust to, and support it. Some even questioned its purpose, refusing to comply and challenging supporters ([25]). A later bill aimed to partially reverse the ban, ostensibly to restore consumer "dignity," by forcing retailers to provide at least one plastic bag per customer ([ 7]).
We collected archival, social media, interview, and ethnographic data related to the Chilean ban, starting in 2013 and lasting until four months after the implementation of the ban (i.e., June 2019). Web Appendix B summarizes these sources. To begin analysis, we undertook a descriptive exploration of the entire data set. The Spanish-speaking members of the author team identified prominent themes in the verbatim data (e.g., emotional reactions, relevant actors, meanings), which were discussed with the entire author team. Through this analysis, we gained an initial understanding of the shopping practice from consumers' and other carriers' perspectives, and we narrowed the focus to consumer resistance. Next, we developed etic codes, in accordance with analytical procedures commonly adopted in practice-based research (e.g., [11]; [30]; [47]). This coding stage focused on the processes of reconfiguring the shopping practice to build understandings of how consumers respond to sustainability interventions.
Similar to the procedures adopted by [ 5], we supplemented the initial practice–theoretical codes with emic terms (e.g., "proud," "angry," "hard," "unfair," "commercial interests") to reflect how consumers responded to changes in the shopping practice. Each Spanish-speaking author coded different types of data and discussed the coding to triangulate the findings among researchers and data sources ([ 1]). It became apparent during this round of analysis that consumers had expressed concerns about responsibility and manifested emotional responses to the sustainability intervention. This prominence of responsibilization and emotionality led us to focus on capturing these aspects. We then aggregated the emergent codes to develop meaningful themes that explain what gives rise to consumer resistance to sustainability interventions. In this iterative process, we moved between prior literature and our data, examining how existing concepts might explain or be challenged by the data ([44]). In the final stage of analysis, we examined selected excerpts (i.e., those simultaneously coded as particular reconfiguration processes and challenges) to identify how resistance interferes with practice change. This process continued until a set of theoretical concepts emerged that explained the phenomenon, allowing us to develop a theory of consumer resistance in social practice change. Throughout the process, we considered other types of consumer responses to the sustainability intervention (e.g., support, acceptance) and the roles of other actors (e.g., retailers) in reconfiguring the shopping practice. However, to keep the focus on consumer resistance to sustainability interventions, we did not integrate those aspects into our theory except when directly relevant (e.g., if consumers demanded retailers take responsibility). We provide evidence from the various sources to illustrate our coding in Web Appendix C. Our data references are provided in Web Appendix D.
We have suggested that consumer resistance to sustainability interventions arises because consumers are required to alter the social practice implicated by the intervention. Our findings, detailed in the next two sections, offer insights into that process. First, social practice change occurs through three recursive reconfiguration processes: sensemaking, accommodating, and stabilizing. Second, consumers encounter three challenges in reconfiguring the practice: responsibilization battles, unsettling emotionality, and the (un)linking of other practices. Each of these challenges disrupts the change process by creating different forms of consumer resistance that interfere with the reconfiguration processes—distracting sensemaking, discouraging accommodation, and delaying stabilization—which undermine the effectiveness of the intervention.
Our findings are summarized in Table 1 and further elaborated in the following sections. We begin by describing the three practice reconfiguration processes and follow this by documenting the three challenges to their effective unfolding, including how these challenges distract, discourage, and delay the social change process. We finish by offering a formal statement of this emergent theory that describes these insights in a more generalizable form.
Graph
Table 1. Understanding Consumer Resistance to Sustainability Interventions.
| Reconfiguration processesReconfigurationchallenges | SensemakingConsumers seek to understand and develop new meanings for the (reconfiguring) shopping practice. | AccommodatingConsumers develop new competences for using and handling the new materials (and meanings) involved in performing the shopping practice without disposable plastic bags. | StabilizingConsumers embody (at times with resignation) the changed practice, with more or less difficulty or speed. |
|---|
| Consumer resistance |
|---|
| Responsibilization battlesCarriers clash over who is responsible for reconfiguring the shopping practice. | Sensemaking is distracted as consumers divert sensemaking efforts towards responsibilization instead of reconfiguring the shopping practice. | Accommodation is discouraged as consumers question the motives and responsibility of each actor who introduces a new material involved in performing the practice. | Stabilization is delayed as consumers hesitate to commit to the reconfiguring practice without seeing commitment from other actors with whom they wish to share responsibility. |
| Unsettling emotionalityCarriers no longer feel completely attuned or "at home" with the shopping practice, which was previously familiar to them. This leads to anxiety and fear. | Sensemaking is distracted as consumers find it difficult to understand their unsettling emotions. | Accommodation is discouraged as consumers limit their use of new materials as well as their attempts to develop new competences to avoid experiencing unsettling emotions. | Stabilization is delayed as consumers may not want to stabilize the reconfiguring practice until they feel emotionally settled in it. |
| (Un)linking other practicesCarriers forge new or break existing connections between the shopping practice and other practices. | Sensemaking is distracted as consumer efforts are extended to other practices by making and breaking links between them. | Accommodation is discouraged as consumers direct their attention to the linked and unlinked practices, reducing their ability to accommodate elements within the reconfiguring practice. | Stabilization is delayed as consumers try to embody changes to (un)linked practices in addition to embodying changes to the reconfiguring practice. |
Our interviews and ethnographic incursions provide multiple similar descriptions of the shopping practice, which emphasize its mundane, routinized, and stable nature prior to the ban. Consumers easily reproduced the existing shopping practice without much effort, as described by one interviewee: "I normally check what's in the kitchen, a quick look to see what we need...and as I know the store layout by heart, I walk the aisles the same way, I go early when there's no one, I take one of my sons, I put things inside the cart...and only in plastic bags. The house was filled with plastic bags."1 Consumers took the availability of plastic bags for granted and counted on them to support other practices, such as waste disposal: "Before [the ban] I didn't bring anything to carry my purchases. In fact, if I needed five bags in a purchase, I grabbed five more for the garbage."2 When the ban challenged this shopping practice, we observed consumers seeking to change the practice through three reconfiguration processes. We present them separately for theorization but note that real-world reconfiguration processes are ongoing and recursive.
Carriers initially attempted to make sense of the changes to their shopping practice as required by the intervention. The plastic bag ban implied the loss of a material element of the shopping practice and many other practices. Many consumers started to consider substitute materials, as well as new competences they would need to continue performing their shopping practice, such as asking "how do I carry all this now?" (see Figure 1a). Consumers sought to understand and develop new meanings for the shopping practice as well. That is, the governmental campaign assigned negative meanings to plastic bags, portraying them as damaging to natural landscapes and animal life (see Figure 1b). This conflicted with the more conventional meanings in Chilean society, which regarded plastic bags as convenient, affordable, and widely used ([ 9]). The campaign did not extend the negative meanings to other, related materials though, so consumers had to find a way to resolve the contradiction that "in the meantime, everything continues to be wrapped in plastic...food...toilet paper...shampoo...etc. etc. etc."3
Graph: Figure 1. Illustrative social media posts.
Online and in supermarkets, consumers discussed the scope, purpose, and point of the ban to make sense of it. As noted by the checkout assistants, who pack bags for customers at the register, in the weeks following the implementation of the ban, "half [of the shoppers] think 'this is great for the planet' and half [of them] say 'this is a great business for the supermarket' that now sells bags rather than giving them away"4 (see also Figure 1c). These informants highlighted the difference between "the typical people who say 'this change is useless'"5 and others saying "this is a really good policy."6
We found that, after some initial sensemaking of the ban, carriers started to accommodate changes to the shopping practice, discuss the intervention and its impact, and develop new competences for using the new materials and meanings involved in performing the shopping practice without plastic bags. Retailers' and governmental communications focused on a single new competence: "Bring your own bag."10 However, we found evidence that consumers had additional competences associated with using disposable plastic bags while shopping, such as quickly placing products on the checkout belt, sorting products for a swift checkout, knowing how much to tip checkout assistants, and distributing loaded plastic bags in both hands to carry them easily into their cars (see Figure 2a). These competences were challenged when bags were limited (to two per customer) and eventually banned. Consumers also had to develop new skills for unloading purchases at home (e.g., using hard plastic boxes) and to design home storage options for their reusable bags (e.g., dedicated kitchen drawer; see Figures 2b and 2c). A local magazine offered tips for developing new competences, such as "when making your shopping list, get in the habit of always writing down 'reusable bags' as the first thing."11
Graph: Figure 2. Illustrative social media posts and photographs.
Cashiers also had to develop new competences for packing groceries into different types of materials (e.g., reusable bags, boxes, carts) and learn how to time their service provision accordingly, as some packing processes might take more time. Consumer interactions with these actors were also altered (e.g., when and how to provide the cashiers with the materials; if and how to pack the materials into a trolley), and thus new relational competences from consumers were required.
Furthermore, the law did not propose substitute material elements, and we found that consumers began experimenting with different substitutes for plastic bags (see Figure 3a). Media and social media actors also offered ideas: "#ByePlasticBags: The law that seeks to reduce the use of bags has already started...What do you think of this measure? What idea do you propose to replace the bags?"15 During the partial ban period, social marketing campaigns invited consumers to bring their own bags to stores but did not suggest the type of bag they should use. Retailers also proposed diverse alternative materials (see Figure 3b). Some supermarkets offered recyclable bags for sale, but because they contained 15% plastic, these were quickly denounced by Greenpeace as misleading.16 Other supermarkets offered cardboard boxes, fabric bags, reusable plastic bags, and paper bags, though some provided no alternatives. In searching for substitute materials and to develop competences, consumers accommodated the reconfigured shopping practice as carriers by attempting to become skillful shoppers once again: "I know I must carry a [reusable] bag in my backpack no matter what, because if I eventually want to buy something I need to know where to carry it."17
Graph: Figure 3. Illustrative social media posts.
As our analysis indicates, at some point, carriers started to embody (at times with resignation) changes to the shopping practice, with more or less difficulty or speed. The practice stabilized as it once again became familiar and routinized. At the time we concluded data collection, some consumers settled on a set of interconnected elements that would allow them to perform the reconfigured shopping practice skillfully, describing how they might "keep reusable bags in the car. When I get home and unload them, they go back to the car immediately"21 or noting "I haven't seen anyone else doing this...here we use garbage bags, those black ones that I purchase once a week. I purchase these bags, put my groceries in them, and when I take them out, I use the bags for the garbage."22 By regularly performing such behaviors, consumers supported the stabilization of the reconfigured practice. From a practice–theoretical perspective, we would expect that, as more consumers enter this stabilizing phase, their performances may converge into a new social version of the practice, which other consumers then start reproducing.
However, some consumers did not engage in stabilization immediately. We found evidence of consumers purchasing reusable bags on multiple shopping trips and accumulating them at home or "stealing" the disposable bags the supermarket provides for fruit and vegetables and repurposing them to carry purchases home (see Figure 3c). Such consumers continued trying to make sense of the intervention and develop sustainable shopping practices, but they also still experienced contradictions and misalignments in their performance, which impeded the stabilization of the shopping practice.
The ban on plastic bags forced carriers to reconfigure the shopping practice, and we found that this generated three challenges: responsibilization battles, unsettling emotionality, and the (un)linking of other practices. These challenges made practice change more difficult for consumers, leading to resistance. We describe each challenge and how it distracted, discouraged, and delayed the change process, leading to a recursive state of reconfiguration instead of stabilization.
Responsibilization battles emerged when carriers clashed over who was responsible for reconfiguring the shopping practice. In these battles, consumers who refused responsibilization challenged those who did not ("Are you an idiot or do you actually believe they removed the bags for the planet? To cut costs for companies, nothing else #ByePlasticBags"23) and vice versa ("I hope all those who are AGAINST the plastic bag ban choke on one! #ByePlasticBags"24). Retailers and government agencies were also pressed to take some responsibility ("Now retailers and supermarkets must give away eco bags. Not everything is revenue and profit. Do your share!!! #ByePlasticBags"25). These responsibilization battles unfolded in social media, the press, and retail spaces.
The battles evidenced consumers' discomfort due to responsibilization ([10]). We found that consumers experienced ( 1) physical discomfort from carrying fewer, larger, heavier bags ("I have to lift the bags and they are super heavy...because they are so large, I tend to load them too much and the truth is, I start feeling it in my back"26); ( 2) psychological discomfort due to social scrutiny of their performance of the shopping practice ("When I ask them, 'did you bring a bag?' they get upset, they resent it a bit"27); ( 3) financial discomfort caused by incurring the cost of replacing the plastic bags they previously got for free ("Always f—ing up the poorest and most vulnerable in our country, now paper bags are sold for $1000, $2000 and $3000 [Chilean] pesos"28); and ( 4) moral discomfort when they identified hypocrisy in corporations or government actors that profited from the change ("Supermarket B prospers and the consumer does not benefit at all. Customers now have to buy your bags and advertise your brand for free"29).
To resolve their discomfort and navigate the challenge of responsibilization battles, consumers resisted the plastic bag ban in various ways. Some consumers attempted to spread the responsibility ("@SupermarketA, @StoreA @DepartmentStore @StoreB and many more should give us bags and not sell them"30) or diffuse responsibilization claims ("No one forces you to buy a reusable bag from supermarkets, there are many people who have their small business selling bags, or you can make your own bag, carry a backpack, even carry your purchases in your hands when you don't have much stuff"31). Other consumers engaged in boycotts and retaliatory actions against both supermarkets and the government: "I also enjoy going to [Supermarket A] with a [Supermarket B] bag and going to [Supermarket B] with a [Supermarket A] bag, because I feel like supermarkets are benefiting from this law so this is my way of protesting against this. If I am forced to buy the bag, then I get to choose which one to use where."32
These responsibilization-provoked sources of resistance interfered with practice reconfiguration (see Table 1). They distracted consumers' sensemaking away from the shopping practice and toward other actors' intentions and behaviors, as exemplified in debates about government mandates involving supermarkets: "I don't understand why people are celebrating so much the stupidity and loss of freedom of #ByePlasticBags...Why weren't the supermarkets mandated to change the [disposable plastic] bags for biodegradable and compostable ones?"33
Such resistance also discouraged accommodation when consumers witnessed nonsupportive actions by other carriers whose motives they questioned. For example, consumers who tried to replace the banned bags with reusable bags or cardboard boxes often believed that supermarkets should support them: "Customers must be informed correctly. I bought a full trolley and when I got to the cashier I got the news that I cannot get bags, they did not have bags available to buy and the cashier tells me that the local manager said that giving cardboard boxes was inappropriate. They must provide solutions to the customers, put signs up warning them of the change."34 Finally, consumers delayed in committing to reconfiguring the shopping practice because they did not perceive sufficient commitment from other carriers with whom they wished to share responsibilities: "In part, the regulation of plastic bags is justified due to the contamination derived from them, but I believe that the ban does not solve the problem and unnecessarily burdens the customer with something that the shops should be responsible for."35
The ban also disrupted the affective structure of the shopping practice: Consumers as carriers no longer felt completely attuned or "at home" with their previously familiar practice. During reconfiguration processes, the shopping practice gets infused with an unsettling mix of negative and positive emotionality. Some consumers experienced anxiety and fear: "good heavens, what are we going to do?"36 and others grappled with the notion that "though I like nature and all this, the first week when I went to the supermarket and there were no bags, it was...'good God, the bags are over!' and I even got a bit angry like 'why are there no bags?'"37 For other carriers who still lack competence and therefore fail to perform the practice skillfully, reconfiguring the practice creates frustration and shame: "#ByePlasticBags I can't get used to this shit! ."38 As consumers reflected on their performance of the shopping practice, additional emotions emerged. Erratic or flawed performance (e.g., "people forget to bring bags or bring fewer than they need"39) prevented the changing practice from becoming "second nature" and added guilt and anger to its emotionality. The dynamic links between the modified elements (materials, competences, meanings) of the shopping practice (and other practices) further unsettled its emotionality (see Figure 4a). Consumers may have felt conflicted about performing well in one practice but not others:
Graph: Figure 4. Illustrative social media posts.
I have mixed feelings...too bad this will go on record...up to the very last minute [prior to the ban] I still asked for plastic bags. Now I imagine the little fish that's eating the plastic and I am committed, but my alternative is still to purchase a plastic bag for the garbage.40
However, the reconfiguration processes also offered numerous possibilities for performing the practice in ways that could be more effective or beneficial to carriers. Therefore, consumers could adopt more sustainable materials, become more competent, or derive more meaning from the practice. Such possibilities charged the shopping practice with positive emotions, such as hope, excitement, and pride ("You have to be calm and take it with humor, and that is all!!! We look cute carrying Cloth Bags!!! Hahaha #lookinglikegrandma !"41)
As the reconfiguration processes continued and consumers started shopping without disposable plastic bags, other emotions surfaced and became part of the unsettling emotionality. Pride characterized carriers who felt accomplished or creative in performing the practice (see Figure 4b) because they identified new substitute materials: "When you are offered a plastic bag at the farmers' market, but you open your backpack and say 'just in here please' #GoodbyeByePlasticBags" [accompanied by an image of Arnold Schwarzenegger looking at the horizon surrounded by animals and nature].42 This sense of pride was also reinforced by social marketing campaigns, such as one proclaiming "Chile is the 1st country in Latin America to say #ByePlasticBags in commerce!"43 Reconfiguration processes also prompted nostalgia: "When I was little and we shopped, they would wrap things in newspaper, there wasn't a plastic bag for sugar, it was paper."44
This mix of emotions we identified emerged during reconfiguration processes and, as suggested in prior research, became characteristic of the practice, providing consumers with a new (albeit changing) template for the beliefs and emotions they should express as part of that practice. We found that each shopping performance added to the emotionality of the practice, making it more volatile, complex, and tense.
In response to the challenge of unsettling emotionality, we found that consumers resisted the sustainability intervention by complaining that "to carry products in their hands is degrading"48 or claiming a "loss of dignity,"49 as well as engaging in more extreme acts such as "kicking checkout points, screaming at the cashiers, causing scandals, so the supermarket security guards have to be called."50 The resistance generated by the challenge of unsettling emotionality interfered with the ongoing practice configuration (see Table 1). It distracted sensemaking by diminishing consumers' cognitive capacity or ability to notice and make sense of important cues: "There is a feeling of disgust for having the responsibility of bringing our own bags. This increases our costs, and I don't see the benefits."51 Resistance also discouraged the accommodation of the reconfiguring practice, as consumers hesitated to handle new materials or develop new competences when they struggled with their emotions: "I am already getting used to having to carry the bag, but if I sometimes forget the bag, I have to buy a bag again. If I don't buy it and if there are a few things, I have to carry them in my hands and that's embarrassing...walking around with things in sight."52 Moreover, as consumers resisted in response to the challenge of unsettling emotionality, they tended to avoid repeating performances that prompted anxiety or fear, and this delayed practice stabilization: "Our family's initial reaction was very positive, as we understood the purpose. However, as soon as this ban started revealing the difficulties of this buying process, our view started changing and we now feel upset and uncomfortable, and seriously question the initiative...Isn't there an easier way?"53
The challenge of (un)linking other practices emerged because as the materials, competences, and meanings of the shopping practice underwent reconfiguration, they also forged new or broke existing connections between the shopping practice and other practices. For example, the ban disrupted domestic disposal of garbage because free disposable plastic bags, which represent a key material for both practices, were no longer available: "I used the supermarket bags to dispose of trash, now I need to buy trash bags because I still need to throw the trash out...Does anybody do this differently?"54
As the meaning of plastic bags evolved, we found that connotations of contamination and waste also extended to other retailing practices, such as product packaging ("wrapping eggplants in plastic film"55; see Figure 4c), selling reusable bags wrapped in plastic, "requiring that [consumers] use plastic bags to weigh fruit and bread,"56 and waste management efforts ("They could work on responsible waste management now, the producing companies MUST take care of the waste that remains when consuming their products. #wasteisadesignproblem @[sustainability ONG] knows about that."57) Once they face disruption to such links, consumers manifest resistance:
A gentleman once told me: "This is absurd! 2% of the country's pollution is plastic bags in the water. The rest is pure plastic that they continue selling. So what is the point? You get it?...Do you see how ridiculous this is? They are attacking 2% instead of attacking 30% through prohibiting other plastics, reducing that, or increasing these other things. This is more of a populist measure than anything else." I got to hear plenty of opinions from people [laughs].58
Moreover, consumers identified misalignments between the governmental discourse about the ban and the government's actions in other industries: "Everything is fine with the plastic bags...What about the coal mine in Patagonia?"59 or "#ByePlasticBags but [president] shrinks national parks for private exploitation, persists with + hydroelectric plants, mining, destruction of native forests with pine and eucalyptus plantations, there is no recycling, what we consume comes in plastic and is disposable, retail uses electricity for lights all day."60 Upon acknowledging the complexity of interrelated practices, consumers resisted the intervention, perceiving it as "absurd."
Consumer resistance in response to the challenge of (un)linking other practices also interfered with practice reconfiguration (see Table 1). It distracted sensemaking by requiring consumers to make sense of not just the focal shopping practice but also the broader nexuses with other practices and their elements (i.e., materials, competences, and meanings): "I don't understand how they can talk about #ByePlasticBags while still allowing tires. It must be because the bags contaminate 'in your face' while tire wear is invisible because their microparticles disappear in the air we breathe. #terriblelegislation."61 Similarly, such resistance discouraged consumers from accommodating elements within the shopping practice, as they would need to accommodate elements in linked and unlinked practices at the same time: "I went to the supermarket, good thing they eliminated plastic bags, I bought this [paper bag], but everything I am carrying inside it is in plastic packaging. What has changed from this? #ByePlasticBags #GoodBusiness."62 Finally, consumers' resistance delayed the stabilization of the reconfiguring practice because they were forced to embody changes to (un)linked practices in addition to embodying changes to the reconfiguring practice: "Today they didn't give me plastic bags at the supermarket, 10 fewer bags on the planet, but what I can't wrap my head around is that I had to purchase 10 of those black garbage bags for the bathroom and kitchen waste bins (I had never, ever purchased bags for this before). Something is not right!!! @EnvironmentMinistryChile #ByePlasticBags."64
In this section, we build on our findings to propose a theory of consumer resistance in social practice change. We now present this theory, illustrated in Figure 5, in broad terms to demonstrate its generalizability.
Graph: Figure 5. Consumer resistance in social practice change as required by interventions.
Practice theories, such as ours, conceive of individual behaviors as embedded in social practices. As such, we start from where consumers continuously and skillfully perform an existing practice by repeatedly linking its elements in a similar manner (see Figure 5, Existing Practice). However, when interventions (imposed or otherwise) occur that modify the basic elements of a practice, consumers must reconfigure the links across the modified elements to enable the social practice to develop and endure.
Consumers do this through three recursive reconfiguration processes (see Figure 5, Reconfiguration Processes): sensemaking, accommodating, and stabilizing. Consumers work to make sense of the modified elements to understand what the change means for the social practice in question and their continued performance of it ("what do we do now?"). They must also accommodate the modified elements while performing the changing social practice ("how do we do it now?"). Finally, consumers need to stabilize the changed practice by embodying the newly developed links between the modified elements in their performances ("this is how we will keep doing it from now on"). During reconfiguration, the links between the practice elements are provisional (dotted lines in Figure 5, Reconfiguration Processes), as consumers are not yet consistently engaging with the same elements in performing the changing practice.
Three major challenges emerge in social practice change: responsibilization battles, unsettling emotionality, and the (un)linking of other practices. These challenges generate consumer resistance that disrupts practice reconfiguration. Considering the nature of these reconfiguration processes, we identify how the dispersed consumer resistance interferes with each of them in a particular way. Sensemaking, which requires focused attention, emotional stability, and a manageable cognitive load ([36]), is distracted by consumer resistance. Accommodating, which involves experimentation, trial and error, and risk-taking to incorporate new materials into the changing practice, is discouraged by consumer resistance. Finally, stabilizing, which requires that consumers comfortably and consistently perform a new version of the practice ([30]; [47]), is delayed by consumer resistance. It is worth noting that, as reconfiguration processes are recursive, the ways in which consumer resistance disrupts them may overlap.
In this way, consumer resistance keeps the practice in a recursive state of reconfiguration, interfering with the desired change. Finally, when the reconfiguring practice becomes stable, the practice in question is reconfigured (see Figure 5, Reconfigured Practice): Consumers skillfully perform it again by continuously linking its modified elements in a similar manner. Taken together, our theory explains what gives rise to consumer resistance to interventions and how this resistance can be reduced by including consideration of social practice change.
Our theory of consumer resistance in social practice change has two main research implications. First, we advance marketing literature on sustainable consumer behavior by shifting the focus from individual consumer behavior to social practice change. Second, we advance theories of social practice change in marketing and social sciences by examining the role of consumer resistance and emphasizing the previously overlooked roles of responsibilization and emotionality.
Consumers often resist behavioral-focused interventions, particularly when they are made responsible for social issues (e.g., [10]), thereby undermining the effectiveness of the intervention. Our theory offers an explanation for this important problem. Extending prior research (e.g., [ 3]; [40]), we show how sustainability interventions disrupt social practices and explain that consumer resistance emerges because the individual behaviors being targeted are embedded in disrupted social practices. Specifically, we explain that, when interventions aim to change individual behaviors rather than social practices, they place excessive responsibility on consumers, unsettle their practice-related emotionality, and destabilize the multiple practices that interconnect to shape consumers' lives, ultimately leading to resistance. This theory offers a conceptual framework for better examining, understanding, and explaining consumer resistance to sustainability interventions and how this resistance can be reduced.
Our proposed theory also contributes to theories of social practice change in marketing and social sciences more generally. Whereas [43] highly influential theory shows convincingly that social practices change when links among their elements (i.e., materials, competences, meanings) are made or broken, their theory does not fully articulate what processes and challenges are actually involved in social practice change and what gives rise to consumer resistance in social practice change. Our theory advances existing social practice theories in three important ways. First, it shows that social practice change takes place through three recursive reconfiguration processes by which carriers reconnect the links among modified elements for a practice to endure. Second, it identifies three major challenges arising in reconfiguration processes. Third, it shows how these challenges generate consumer resistance, which disrupts the reconfiguration processes required by sustainability interventions and undermines its effectiveness.
In addition to articulating the practice reconfiguration processes, our theory extends current understandings of social practice change by shedding light on two subjects: responsibilization and emotionality. First, we explain why consumers resist responsibilization and provide evidence of how they do so, in the context of a social practice change. Our analysis shows that consumers resist responsibilization not only when they find it difficult to reconfigure their habituated social practices but also when they feel they are the primary carriers being tasked with the change. Thus, our findings extend [10] work by identifying other forms of discomfort that consumers experience in response to such allocations of responsibility. Moreover, by introducing "responsibilization battles," we identify the consequences of discomfort that go beyond the individual. The notion of responsibilization battles in social practice change is important because these battles are likely to become more frequent as consumers increasingly find themselves tasked with complex practice reconfigurations. Furthermore, when these battles occur publicly, such as through social media, they may amplify consumer resistance and outrage about sustainability interventions, potentially working through social contagion ([31]) to disrupt other social practices.
Second, we emphasize the role of emotionality in social practice change. We show that during reconfiguration, multiple, often conflicting emotions get linked to practices as consumers perform them. This notion adds to a practice–theoretical understanding of practice reconfiguration. We challenge the assumption that consumers simply accept responsibility assigned to them by government interventions. We find that consumer resistance is a disruptive force that pushes against consumers' desire to acclimatize to a new normal ([30]), and it can infuse reconfiguring practices with demoralizing emotions. Although emotionality is often a "blind spot" in social practice theory ([27]), examining its role offers a way to connect cultural and material explanations of social phenomena ([34]). By identifying the unsettling emotionality of social practice change, we help clarify the conflicts that often surround sustainability interventions ([46]). These go beyond individual reactions to routine disruptions or behavioral change and add insights about the role of collective emotions in sustainable consumer behavior ([54]).
If individual consumer behavior is determined by social practices beyond individual motivations or attitudes, then putting a sustainability intervention into effect is just a first step. Reconfiguring the practice should be the primary goal, which can lead to the broader aim of fostering sustainable consumer behavior. Our emergent theory offers a framework for designing and managing practice-based sustainability interventions, which makes it possible to explore methods to reduce consumer resistance that go beyond individual behavioral approaches. Our recommendations focus on two key aspects: how to ( 1) design practice-based sustainability interventions to reduce resistance at the outset and ( 2) monitor and adjust these interventions to manage consumer resistance that may emerge later. Using the plastic bag bans as an example, we offer a first set of recommendations for considerations that should be addressed prior to implementing the intervention, then a second set involving ways to monitor and adjust ongoing processes during practice reconfiguration. We outline the sets of recommendations in Figures 6 and 7.
Graph: Figure 6. Decision flowchart: planning and designing practice-based interventions.
Graph: Figure 7. Decision flowchart: monitoring and adjusting practice-based interventions.
First, when designing sustainability interventions, policy makers should identify the potential practice elements (i.e., materials, competences, and meanings) that will be disrupted and require reconfiguration. They can then introduce substitute elements that reflect the sustainability goal of the intervention, demonstrate how the new elements work, and provide advice regarding their use and assessment. For example, to replace disposable plastic bags, policy makers could present alternative forms of reusable bags, describing both their usage and their (positive) impact on the environment. Likewise, policy makers should identify the competences that consumers need to perform the changed practice, such as packing different types and sizes of reusable bags, choosing the right bags, or deciding where to store them. We advise policy makers to obtain consumers' perceptions of and reactions to the new practice before announcing the intervention. They can then include those insights in their planning and communication. Rather than relying exclusively on opinion polls, which often show strong support for interventions (see http://chaobolsasplasticas.cl/en/), deeper consumer insights should be gained through focus groups and ethnographic work (see [ 6]) to capture consumer experiences of the reconfiguration processes.
Second, policy makers should consider all practice carriers—beyond just consumers—and distribute responsibilities among them. For example, consumers may perform the shopping practice, but retailers and bag manufacturers set material arrangements for shopping, the government determines the rules for the commercial activity, and social marketers promote the meaning of sustainable consumption. Rather than banning bags, which eliminates retailers' responsibility for this aspect of the shopping practice, policy makers might assign retailers the task of developing sustainable alternatives. Similarly, to prevent retailers' opportunistic attempts to profit from the intervention (e.g., by selling reusable bags for profit), which threaten to irritate consumers because they perceive these tactics as hypocritical, policy makers might establish legal price limits for reusable bags or prohibit retailers from charging for a bag that features their brand logo.
Third, ethnographic studies might help policy makers determine and evaluate the potential emotional implications of an intervention. For retailers, the point of sale is generally where consumers experience performances laden with emotions. Planning to reduce those visible manifestations of unsettling emotionality may reduce their effects on consumer resistance. Social marketing campaigns and efforts at the point of sale (e.g., signals that indicate the shared responsibilities of multiple carriers, advice for accommodating the change) might reduce extreme negative manifestations, such as assaults on cashiers, abandoned shopping carts, or theft of plastic bags from the produce section. Consumers might also feel a sense of pride or other positive emotions if they can accomplish the shopping practice without plastic bags, so these positive emotions should be leveraged to reduce resistance, such as through the gamification ([28]).
Fourth, policy makers and social marketing institutions should identify which practices share materials, competences, or meanings with the targeted practice (e.g., waste practices, goods packaging, transportation), so they can anticipate other possible sources of resistance and act accordingly. Materials should be considered broadly, for example, a sustainability intervention targeting plastic bags should address links to other practices that also involve plastic—as a general substance, not necessarily just in the shape of disposable bags. Due to their broad goals, such as "promoting sustainable consumption," the scope of sustainability interventions tends to appear virtually endless. Consequently, consumers might link any intervention to other practices that they consider unsustainable (e.g., waste, mining). By establishing and communicating clear intervention boundaries, policy makers can establish a precise sequence of future interventions that can support the broader goal of sustainable consumer behavior.
Designing interventions that account for the aforementioned considerations may reduce consumer resistance at the outset, but policy makers must continue monitoring the reconfiguration processes to identify any emerging resistance, then make necessary adjustments to manage this resistance. These adjustments should focus specifically on how potential consumer resistance disrupts reconfiguration processes (i.e., distracting sensemaking, discouraging accommodation, or delaying stabilization) and should aim to refocus sensemaking, encourage accommodation, and accelerate stabilization.
First, consumer resistance may distract sensemaking during the reconfiguration process. To identify this resistance, note when consumers experience tensions and lack of focus while attempting to make sense of the intervention and its required changes. When consumer resistance manifests in this way, intervention efforts should remove or reduce these distractions. For example, communications could remind carriers of the intervention's scope, the distribution of responsibility, and specific benefits to them. By clearly communicating and reaffirming the boundaries around the intervention and its benefits, policy makers can reduce distraction and refocus sensemaking (e.g., establishing a roadmap for associated interventions). For example, to ensure benefits for consumers—often the most visible and numerous carriers of a practice—retailers might introduce limited-time discounts on eco-friendly garbage bags for shoppers who comply with the intervention by bringing reusable bags. If this incentive is not financially viable, retailers could consider other ways to encourage adoption (e.g., badges for early compliance). Policy makers might also build financial considerations (e.g., grants, funding) into the policy, then allow retailers to distribute the government-sponsored incentives to consumers.
Second, consumer resistance may discourage accommodation. To identify this resistance, observe consumers avoiding risks and restricting their experimentation with new materials, competences, and/or meanings. When consumer resistance manifests in this way, intervention efforts should focus on the challenges that trigger the discouragement. If consumers are struggling to develop competences due to unsettling emotionality, for example, additional educational programs might be helpful. At the point of sale, instruction banners might acknowledge initial forgetfulness, then offer sustainable alternatives for those shoppers who left their reusable bags at home. Policy makers should observe what alternatives become visible when consumers attempt to reconfigure the shopping practice, and they should use these insights to determine solutions that can be quickly and easily adopted. Alternatives that arise through reconfiguration efforts may be better suited to the market setting, even if they differ from the options predicted in the planning phase. Therefore, it is important to monitor and leverage consumer accommodation efforts.
Third, consumer resistance may delay stabilization. To identify this resistance, notice consumers grappling with how to comfortably embody the changes. To deal with these delays, intervention efforts should focus on removing barriers. Traditionally, testimonials and success stories have been recommended to foster consumer compliance to behavioral change ([54]). However, we find that consumers tend to be unwilling to stabilize a reconfiguring practice until they observe commitment from other actors. Thus, we propose that effective campaigns and forums should focus on other actors that consumers believe have not been adequately responsibilized. Other efforts to help consumers overcome the discomfort associated with stabilizing practices should refer to both the reconfiguring practice and any practices that have been newly (un)linked. Finally, in line with our recommendation that broader sustainability goals should be emphasized throughout the process, policy makers must ensure that any promises are met. Carriers will be more likely to stabilize reconfiguring practices if they know that their efforts are not moot when it comes to fostering more sustainable consumer behavior overall.
At the time an intervention is put in place, and thereafter, communications with carriers should be ongoing, describing the intervention's scope, importance, and responsibility assignments. When responsibilization, unsettling emotionality, and the (un)linking of other practices generate consumer resistance during the reconfiguration process, policy makers should prioritize identifying disruptions to ensure targeted responses to resistance. In this way, the process of designing and implementing interventions will remain appropriately dynamic and iterative, rather than static and linear.
Future research can address limitations in this study. First, we examine a ban on plastic bags. Despite its spread and importance, this empirical setting may differ from other contexts within the broader sustainability domain, such as those outlined in the United Nations' Sustainable Development Goals ([49]). Nevertheless, our emergent theory is relevant to intervention contexts that ( 1) result in significant changes to established practices; ( 2) are public, such that the intervention affects many consumers who might resist it; and ( 3) relate to changes that demand the involvement of multiple actors to reconfigure the practice. Additional research might apply this theory and investigate interventions that target other goals (e.g., interventions aimed at reducing smoking, drinking, obesity). Second, the intervention we study entails the elimination of a material (plastic bags), but reconfigurations of social practices could also vary in response to interventions that encourage new competences (e.g., recycling) or alterations to meaning (e.g., recycled drinking water). Continued research should address consumer resistance to interventions that target such practice elements and determine if consumer resistance and reconfiguration processes vary. Third, our comprehensive study mirrors the implementation of the plastic bag ban in real time. However, we did not assess the long-term outcomes of this sustainability intervention. We hope continued research will analyze consumer responses over time to gain additional insights into monitoring/adjusting strategies. Fourth, the sustainability intervention we study was a mandated governmental policy. Other organizations also propose sustainability interventions (e.g., Meat-Free Mondays), and the reconfiguration of social practices in response to marketing-led interventions (e.g., packaging-free product strategies) might differ in this context. We suggest adapting our theory to such research topics to develop insights into the roles of consumers and companies in the successful implementation of such interventions. Finally, the proposed theory provides a novel and comprehensive explanation for why consumers engage in resistance, and as such, it proposes several additional methods for reducing consumer resistance to interventions. However, facilitating more sustainable consumer behavior through interventions is a complex, multilayered effort that is likely to require contributions from multiple perspectives to be resolved satisfactorily. Continued research should consider how the proposed theory of consumer resistance in social practice change can be combined with other perspectives, such as the SHIFT framework ([54]), to clarify how consumer resistance to sustainability interventions can be reduced and, ultimately, to foster more sustainable consumer behavior.
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921992052 - "How Do I Carry All This Now?" Understanding Consumer Resistance to Sustainability Interventions
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921992052 for "How Do I Carry All This Now?" Understanding Consumer Resistance to Sustainability Interventions by Claudia Gonzalez-Arcos, Alison M. Joubert, Daiane Scaraboto, Rodrigo Guesalaga and Jörgen Sandberg in Journal of Marketing
Footnotes 1 Amber Epp
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The authors gratefully acknowledge the grant from UQ Global Strategy and Partnerships Seed Funding Scheme: Round Two 2018.
4 Claudia Gonzalez-Arcos https://orcid.org/0000-0001-5014-2427 Alison M. Joubert https://orcid.org/0000-0002-0774-9650 Daiane Scaraboto https://orcid.org/0000-0002-7658-9339 Jörgen Sandberg https://orcid.org/0000-0001-8008-5312
5 Online supplement: https://doi.org/10.1177/0022242921992052
6 The history of the legislation is available in Web Appendix A.
7 For example, a New York State ban on plastic bags similarly faced strong resistance from angry consumers and unhappy retailers ([41]).
References Atkinson Paul, Delamont Sara. (2005), "Analytic Perspectives," in The SAGE Handbook of Qualitative Research, 3rd ed., Denzin Norman K., Lincoln Yvonna S., eds. Thousand Oaks, CA: SAGE Publications, 821–40.
Bardhi Fleura, Eckhardt Giana M. (2017), "Liquid Consumption," Journal of Consumer Research, 44 (3), 582–97.
Blue Stanley, Shove Elizabeth, Carmona Chris, Kelly Michael P. (2014), "Theories of Practice and Public Health: Understanding (Un)Healthy Practices," Critical Public Health, 26 (1), 36–50.
Bolderdijk Jan W., Steg Linda, Woerdman Edwin, Frieswijk René, Groot Judith I. M. de. (2017), "Understanding Effectiveness Skepticism," Journal of Public Policy and Marketing, 36 (2), 348–61.
Bradford Tonya W., Boyd Naja W. (2020), "Help Me Help You! Employing the Marketing Mix to Alleviate Experiences of Donor Sacrifice," Journal of Marketing, 84 (3), 68–85.
Cayla Julien, Arnould Eric. (2013), "Ethnographic Stories for Market Learning," Journal of Marketing, 77 (4), 1–16.
CNN (2019), "Proyecto Pretende Obligar al Comercio a Entregar una Bolsa Gratis: 'Están Ahorrando a Costa de los Consumidores,'" (accessed September 5, 2020), https://www.cnnchile.com/pais/proyecto-comercio-una-bolsa-gratis-ahorrando-consumidores%5f20190404/.
8 Coleman Clayton. (2018), "Bans on Banning Bags: The Movement to End Single-Use Plastics Faces Significant Obstacles," Environmental and Energy Study Institute (September 6), https://www.eesi.org/articles/view/bans-on-banning-bags-the-movement-to-end-single-use-plastics-faces-signific.
9 Cristi María A., Holzapfel Camila, Nehls Medina, Veer Diamela de, Gonzalez Camila, Holtmann Geraldine, Honorato-Zimmer Daniela, Kiessling Tim, Muñoz Ailin L., Reyes Soledad N., Nuñez Paloma, Sepulveda Jose M., V´squez Nelson, Thiel Martin. (2020), "The Rise and Demise of Plastic Shopping Bags in Chile—Broad and Informal Coalition Supporting Ban as a First Step to Reduce Single-Use Plastics," Ocean and Coastal Management, 187 (1), 105079.
Eckhardt Giana M., Dobscha Susan. (2019), "The Consumer Experience of Responsibilization: The Case of Panera Cares," Journal of Business Ethics, 159 (3), 651–63.
Epp Amber M., Schau Hope J., Price Linda L. (2014), "The Role of Brands and Mediating Technologies in Assembling Long-Distance Family Practices," Journal of Marketing, 78 (3), 81–101.
Evans David. (2011), "Blaming the Consumer—Once Again: The Social and Material Contexts of Everyday Food Waste Practices in Some English Households," Critical Public Health, 21 (4), 429–40.
Giesler Markus, Veresiu Ela. (2014), "Creating the Responsible Consumer: Moralistic Governance Regimes and Consumer Subjectivity," Journal of Consumer Research, 41 (3), 840–57.
Gleim Mark, Lawson Stephanie J. (2014), "Spanning the Gap: An Examination of the Factors Leading to the Green Gap," Journal of Consumer Marketing, 31 (6), 503–14.
Hagberg Johan. (2016), "Agencing Practices: A Historical Exploration of Shopping Bags," Consumption Markets and Culture, 19 (1), 111–32.
Hawkins Gay. (2009), "More-Than-Human Politics: The Case of Plastic Bags," Australian Humanities Review, 46, 43–55.
Hui Allison, Schatzki Theodore, Shove Elizabeth. (2016), The Nexus of Practices: Connections, Constellations, Practitioners. New York: Taylor and Francis.
Jakovcevic Adriana, Steg Linda, Mazzeo Nadia, Caballero Romina, Franco Paul, Putrino Natalia, Favara Jesica. (2014), "Charges for Plastic Bags: Motivational and Behavioral Effects," Journal of Environmental Psychology, 40, 372–80.
Karmarkar Uma R., Bollinger Bryan. (2015), "BYOB: How Bringing Your Own Shopping Bags Leads to Treating Yourself and the Environment," Journal of Marketing, 79 (4), 1–15.
Kemper Joya, Ballantine Paul. (2019), "What Do We Mean by Sustainability Marketing?" Journal of Marketing Management, 35 (3/4), 277–309.
Kotler Philip, Zaltman Gerald. (1971), "Social Marketing: An Approach to Planned Social Change," Journal of Marketing, 35 (3), 3–12.
Kronrod Ann, Grinstein Amir, Wathieu Luc. (2012), "Go Green! Should Environmental Messages Be So Assertive?" Journal of Marketing, 76 (1), 95–102.
Little Vicki J., Lee Christina K. C., Nair Sumesh. (2019), "Macro-Demarketing: The Key to Unlocking Unsustainable Production and Consumption Systems?" Journal of Macromarketing, 39 (2), 166–87.
Luchs Michael G., Phipps Marcus, Hill Tim. (2015), "Exploring Consumer Responsibility for Sustainable Consumption," Journal of Marketing Management, 31 (13/14), 1449–71.
Masquelier Charles. (2017), Critique and Resistance in a Neoliberal Age: Towards a Narrative of Emancipation. London: Palgrave Macmillan UK.
Ministerio del Medio Ambiente (MMA) (2018), "Preguntas Frecuentes #ChaoBolsasPlásticas" (accessed April 24, 2019), https://mma.gob.cl/preguntas-frecuentes-chaobolsasplasticas/.
Molander Susanna, Hartmann Benjamin J. (2018), "Emotion and Practice: Mothering, Cooking, and Teleoaffective Episodes," Marketing Theory, 18 (3), 371–90.
Müller-Stewens Jessica, Schlager Tobias, Häubl Gerald, Herrmann Andreas. (2017), "Gamified Information Presentation and Consumer Adoption of Product Innovations," Journal of Marketing, 81 (2), 8–24.
Olsen Mitchell C., Slotegraaf Rebecca J., Chandukala Sandeep R. (2014), "Green Claims and Message Frames: How Green New Products Change Brand Attitude," Journal of Marketing, 78 (5), 119–37.
Phipps Marcus, Ozanne Julie L. (2017), "Routines Disrupted: Reestablishing Security Through Practice Alignment," Journal of Consumer Research, 44 (2), 361–80.
Plé Loïc, Demangeot Catherine. (2020), "Social Contagion of Online and Offline Deviant Behaviors and its Value Outcomes: The Case of Tourism Ecosystems," Journal of Business Research, 117, 886–96.
Poortinga Wouter, Whitaker Louise. (2018), "Promoting the Use of Reusable Coffee Cups Through Environmental Messaging, the Provision Alternatives, and Financial Incentives," Sustainability, 10 (3), 873–83.
Reckwitz Andreas. (2002), "Toward a Theory of Social Practices: A Development in Culturalist Theorizing," European Journal of Social Theory, 5 (2), 243–63.
Reckwitz Andreas. (2012), "Affective Spaces: A Praxeological Outlook," Rethinking History, 16 (2), 241–58.
Roux Dominique, Izberk-Bilgin Elif. (2018), "Consumer Resistance and Power Relationships in the Marketplace," in Consumer Culture Theory, Arnould Eric, Thompson Craig, eds. Thousand Oaks, CA: SAGE Publications, 295–317.
Sandberg Jörgen, Tsoukas Haridimos. (2015), "Practice Theory: What It Is, Its Philosophical Base, and What It Offers Organization Studies," in The Routledge Companion to Philosophy in Organization Studies, Mir Raza, Willmott Hugh, Greenwood Michelle, eds. New York: Routledge, 184–98.
Schatzki Theodore R. (2002), The Site of the Social: A Philosophical Account of the Constitution of Social Life and Change. University Park, PA: Pennsylvania State University Press.
Schatzki Theodore R. (2019), Social Change in a Material World: How Activity and Material Processes Dynamize Practices. London: Routledge.
Schatzki Theodore R., Cetina Karin K., Savigny Eike von. (2001), The Practice Turn in Contemporary Theory. London: Routledge.
Scheurenbrand Klara, Parsons Elizabeth, Cappellini Benedetta, Patterson Anthony. (2018), "Cycling into Headwinds: Analyzing Practices That Inhibit Sustainability," Journal of Public Policy and Marketing, 37 (2), 227–44.
Sheehan Kevin, Sullivan C. J., Fitz-Gibbon Jorge. (2020), "Angry U.S. Shoppers Slam New York Statewide Plastic Bag Move," News.com.au (March 2), https://www.news.com.au/finance/business/retail/angry-us-shoppers-slam-new-york-statewide-plastic-bag-move/news-story/e4514ec2ef551814734975d4819ea39f.
Shove Elizabeth. (2010), "Beyond the ABC: Climate Change Policy and Theories of Social Change," Environment and Planning A, 42 (6), 1273–85.
Shove Elizabeth, Pantzar Mika, Watson Matt. (2012), The Dynamics of Social Practice: Everyday Life and How It Changes. London: SAGE Publications.
Spiggle Susan. (1994), "Analysis and Interpretation of Qualitative Data in Consumer Research," Journal of Consumer Research, 21 (3), 491–503.
Spurling Nicola, McMeekin Andrew, Shove Elizabeth, Southerton Dale, Welch Daniel. (2013), "Interventions in Practice: Re-framing Policy Approaches to Consumer Behaviour," Sustainable Practices Research Group Report (September), http://www.sprg.ac.uk/uploads/sprg-report-sept-2013.pdf.
Sternbergh Adam. (2015), "The Fight over Plastic Bags Is About a Lot More Than How to Get Groceries Home," New York Magazine (July 15), http://nymag.com/intelligencer/2015/07/plastic-bag-bans.html.
Thomas Tandy C., Epp Amber M. (2019), "The Best Laid Plans: Why New Parents Fail to Habituate Practices," Journal of Consumer Research, 46 (3), 564–89.
United Nations (2018), "United Nations Environment Programme: Legal Limits on Single-Use Plastics and Microplastics: A Global Review of National Laws and Regulations" (accessed August 24, 2019), https://wedocs.unep.org/bitstream/handle/20.500.11822/27113/plastics_limits.pdf.
United Nations (2020), "Sustainable Development Goals" (accessed February 10, 2020), https://sustainabledevelopment.un.org/.
Veresiu Ela, Giesler Markus. (2018), "Neoliberalism and Consumption," in Consumer Culture Theory, Arnould Eric, Thompson Craig, eds. Thousand Oaks, CA: SAGE Publications, 255–75.
Verplanken Bas, Roy Deborah. (2016), "Empowering Interventions to Promote Sustainable Lifestyles: Testing the Habit Discontinuity Hypothesis in a Field Experiment," Journal of Environmental Psychology, 45, 127–34.
Warde Alan. (2005), "Consumption and Theories of Practice," Journal of Consumer Culture, 5 (2), 131–53.
Welch Daniel, Yates Luke. (2018), "The Practices of Collective Action: Practice Theory, Sustainability Transitions, and Social Change," Journal for the Theory of Social Behaviour, 48 (3), 288–305.
White Katherine, Habib Rishad, Hardisty David J. (2019), "How to SHIFT Consumer Behaviors to Be More Sustainable: A Literature Review and Guiding Framework," Journal of Marketing, 83 (3), 22–49.
White Katherine, Simpson Bonnie. (2013), "When Do (and Don't) Normative Appeals Influence Sustainable Consumer Behaviors?" Journal of Marketing, 77 (2), 78–95.
Winterich Karen P., Nenkov Gergana Y., Gonzales Gabriel E. (2019), "Knowing What It Makes: How Product Transformation Salience Increases Recycling," Journal of Marketing, 83 (4), 21–37.
~~~~~~~~
By Claudia Gonzalez-Arcos; Alison M. Joubert; Daiane Scaraboto; Rodrigo Guesalaga and Jörgen Sandberg
Reported by Author; Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 66- How Industries Use Direct-to-Public Persuasion in Policy Conflicts: Asymmetries in Public Voting Responses. By: Seiders, Kathleen; Flynn, Andrea Godfrey; Nenkov, Gergana Y. Journal of Marketing. Mar2022, Vol. 86 Issue 2, p126-146. 21p. 1 Diagram, 7 Charts. DOI: 10.1177/00222429211007517.
- Database:
- Business Source Complete
How Industries Use Direct-to-Public Persuasion in Policy Conflicts: Asymmetries in Public Voting Responses
Industries use persuasion strategies to gain public support when challenged by activist groups on consumer-relevant issues. This marketing practice, termed "direct-to-public persuasion," has received limited attention in the field, and thus we have little understanding of when such campaigns fail or succeed. This research introduces a theoretically derived and empirically supported framework that draws from multiple areas, including marketing persuasion, political campaign strategy, sociopolitical legitimacy, and perceptual fit, to identify important differences in the effectiveness of these persuasion strategies on attitudes and voting behavior. The multimethod approach includes a field study of ballot measure voting during a national U.S. election and three experimental studies. The findings contribute new knowledge of asymmetries in public response to industry and activist arguments. Stronger arguments from both sides lead to more favorable outcomes, but activist groups benefit most. Suspicion of activist arguments weakens the impact on attitudes and voting; industry argument suspicion has limited impact, though it does increase the likelihood of voter switching. A financial argumentation strategy works best for the industry side, while societal argumentation is more effective for the activist side. The insights offer guidance for industries and activist groups as argument strategy success is contingent on the side that uses it.
Keywords: direct-to-public persuasion; persuasion; political campaign strategy; political marketing; sociopolitical legitimacy; voting behavior
Industries often try to persuade the public at large to support their positions on prominent issues that are consequential for consumers and industry members. We term this phenomenon "direct-to-public persuasion," a marketing-driven political influence strategy used to convince the public that the industry is on the correct side of an issue. Industries "talk" to the public not only to sustain or gain advantageous conditions, such as preventing policy actions viewed as costly or restrictive, but also to align the public with practices that further industry objectives.
Industries use direct-to-public persuasion to influence public voting on ballot measures in state and local referendums, public opinion in advance of legislative policy maker decisions, and ongoing public attitudes toward the industry. We focus on conflict scenarios in which industries are challenged by activist groups—collectives that advocate for the public interest—in ballot measures (e.g., [67]). Such conflicts are commonly high-profile and a strategic priority for industries, which spend substantial funds to secure favorable outcomes and protect industry practices ([42]; [59]). The outcomes often have meaningful consequences for the public, including consumers, but the question of what drives voters to support one side or the other remains unanswered.
Table 1 presents examples of how industries and their activist opponents use direct-to-public persuasion to influence the public and gain support. The examples represent diverse issues, including food labeling, recycling, pharmaceutical drug pricing, and tobacco taxes. In the scenarios we examine in this article, industries defend the policy status quo to resist changing disputed practices (e.g., price limits on pharmaceuticals). However, in other settings, industries fight the status quo to expand or develop new markets.[ 5]
Graph
Table 1. Examples of Direct-to-Public Persuasion in Issue Referendums.
| Issue, Location, Year | Overview of Activist (Vote Yes) Side on Issue and Industry (Vote No) Side on Issue | Top Donors | Total Campaign Contributions | Example of Financial Argumentation Strategy | Example of Societal Argumentation Strategy | Result(% of Votes) |
|---|
| Renewable Energy Standards(Proposition 127)Arizona2018 | A "yes" vote supported requiring electric utilities in Arizona to acquire a certain percentage of electricity from renewable resources each year, with the percentage increasing annually from 12% in 2020 to 50% in 2030. | NextGen Climate Action, League of Conservation Voters | $24.1 million | The price of renewable energy continues to drop every day and is already competitive with coal and gas. | Cleaner air and water means healthier families—and less respiratory illnesses like asthma. | 31.4% |
| A "no" vote opposed requiring electric utilities in Arizona to acquire a certain percentage of electricity from renewable resources, thereby leaving in place the state's existing renewable energy requirements of 15% by 2025. | Pinnacle West Capital Corporation, Grand Canyon State Electric | $40.9 million | Corporate and industrial rates would rise over 100%, and residential ratepayers would see an average annual increase of $1,250. | The measure would dramatically harm Arizona's competitiveness and put our utilities' reliable delivery of power at risk. | 68.6% |
| Drug Price Standards(Proposition 61)California2016 | A "yes" vote supported regulating drug prices by requiring state agencies to pay no more than the U.S. Department of Veterans Affairs (VA) pays for prescription drugs. | AIDS Healthcare Foundation, California Nurses Association | $19.1 million | The proposition would save taxpayers billions of dollars in health care costs. | The proposition would provide better access to life-saving drugs. | 46.8% |
| A "no" vote opposed this regulation to require state agencies to pay no more than the VA pays for prescription drugs. | Pharmaceutical Research and Manufacturers of America, Merck & Co. Inc., Pfizer Inc. | $109.1 million | The proposition would increase state prescription drug costs. | The proposition would reduce patient access to medicines. | 53.2% |
| Tobacco Tax Increase(Proposition 56)California2016 | A "yes" vote favored increasing the cigarette tax by $2 per pack, with equivalent increases on other tobacco products and electronic cigarettes | Tom Steyer, California Hospitals Committee on Issues, Million Voter Project Action Fund | $35.5 million | The proposition would reduce tobacco-related health care costs and would help pay for those costs. | The proposition would prevent youth smoking and address tobacco marketing targeting a youth market. | 64.4% |
| A "no" vote opposed increasing the cigarette tax by $2 per pack, with equivalent increases on other tobacco products and electronic cigarettes. | Philip Morris USA, R.J. Reynolds Tobacco Company, Altria Client Services LLC | $71.0 million | The proposition would fund insurance companies and special interests more than it would fund treatments for smoking-related illnesses. | The proposition will not solve real problems like solving the drought and fighting violent crime. | 35.6% |
| Expansion of Bottle Deposits(Question 2)Massachusetts2014 | A "yes" vote would expand the state's beverage container deposit law to require deposits on all nonalcoholic drink containers, except beverages derived from dairy, infant formula, or medications. | Massachusetts Sierra Club, Elm Action Fund, Massachusetts Public Interest Research Group | $1.6 million | A yes vote equals big savings for towns' waste management costs. | A yes vote equals more recycling and less trash and litter. | 26.6% |
| A "no" vote would keep the state's beverage container deposit law in its current form. | American Beverage Association, Nestle Waters NA, Stop & Shop Supermarket Company | $9.6 million | The measure would cost nearly $60 million a year, more than three times the price of curbside programs. | Massachusetts should be a recycling leader, but Question 2 will keep us in the past. | 73.4% |
1 Notes: Information on ballot measures retrieved from https://ballotpedia.org/Ballot_Measures_overview.
Industries have resource advantages relative to activist opponents that suggest they should win policy issue battles, and they often do ([56]; [65]). Because industries are often in the advantageous role of defending the policy status quo, they can leverage the uncertainty of any policy change, whereas activist groups often have the more difficult task of convincing the public that the status quo is failing ([19]). Despite their advantages, industries do not always prevail, which may be due to an underlying predisposition of the public to be more skeptical of the industry and its motives ([ 8]; [45]; [62]). The fact that industries do frequently win suggests that they succeed in overcoming the public's skepticism with effective persuasion arguments ([44]). Our research examines the dynamics of competing campaigns to better understand and explain voting outcomes.
The industry-level practice of direct-to-public persuasion has received limited attention in the marketing field, and we know little about persuasion involving political campaigns with competing positions on an issue. The marketing literature has examined consumer responses to persuasion attempts in a variety of important contexts, and advertising research has examined aspects of argument effectiveness (e.g., [46]; [66]), but political arguments are believed to differ from commercial arguments ([27]). Political marketing research has largely focused on candidate elections rather than election voting on policy issues (see [35]; [38]), and the lobbying literature addresses influencing policy makers rather than influencing the public (see [43]; [67]). Further, although studies show that voters often switch their support during campaigns, and even a nominal shift can change the outcome ([44]; [50]), knowledge of what causes the switching is limited.
To expand knowledge of this understudied topic, we develop a framework that draws from multiple research areas, including marketing persuasion, political campaign strategy, sociopolitical legitimacy, and perceptual fit. We apply legitimacy theory to explain our predicted asymmetry of argument effectiveness across the competing sides and draw on perceptual fit theory to explain the efficacy of specific argumentation strategies on the public's attitudes and voting behavior. We use a multimethod approach that includes a field study of ballot measure voting on two policy issues during a national U.S. election and three experimental studies involving additional issues to test our ideas. The policy issues included prescription drug pricing, tobacco product taxes, renewable energy standards, and container recycling laws.
Our findings offer insight into how the public's attitudes and voting intentions change during the campaign when they are exposed to competing arguments from industry and activist groups. First, we uncover important asymmetries that should be considered in understanding and in using direct-to-public persuasion strategies. Specifically, stronger arguments from both industry and activist groups lead to more favorable attitudes and voting outcomes, but activist groups benefit more from stronger arguments. Greater argument suspicion weakens the impact on attitudes and voting, but primarily for activist groups. Therefore, overall, voters' evaluations of activist-side arguments have a greater impact than industry-side arguments on outcomes. Second, an exploration of vote switching found that industry argument suspicion has limited impact, though it does increase the likelihood of voter switching, particularly for voters who switch their support to the activist side. Finally, using follow-up experiments, we show that a financially focused argumentation strategy works best for the industry side, whereas a societally focused strategy is more effective for the activist side.
We contribute to marketing research with new knowledge of a marketing practice that influences voters and critical political outcomes. We add to the scope of persuasion research with insight into how voters respond to industry versus activist political campaigns, finding novel evidence of an asymmetric public response, where voter judgments of competing arguments—and the degree to which those arguments are strong or suspicious—help predict the relative impact of each side's campaign. We use sociopolitical legitimacy theory in a new way, proposing that an industry-activist legitimacy gap helps explain why industry argument strength has less impact, and apply persuasion theory to explain why voters give greater weight to suspicion of activist arguments. Our pre- and postelection recontact field study design captures individual voter switching, which is not commonly examined, and provides a rare indication that determinants differ for voter segments that flipped their allegiance from one side to the other. Finally, we offer the first known concrete guidance for the use of direct-to-public persuasion strategies with recommendations that differ for industries and activist groups, as we find that a given strategy's success is contingent on the side that uses it.
Figure 1 depicts the conceptual relationships we investigate. Our model predicts that the public's evaluations of industry- and activist-side arguments, in terms of argument strength and argument suspicion, play a central role in predicting outcomes including attitudes toward the issue, voting on the issue, and vote switching. These evaluations depend on each side's argumentation strategy. We begin by describing the two competing sides in the issue conflict scenarios we study and key distinctions between them, and then we develop the hypothesized asymmetric effects of argument strength and suspicion on the voting outcomes.
Graph: Figure 1. Conceptual model.
We define "competing sides on the issue" as the competitors in a policy issue conflict and focus on industry groups and activist groups, traditional policy adversaries ([ 8]). While industries comprise firms that produce a certain class of goods or services, public interest activist groups are political constituents working in a range of policy areas who claim to represent the public or collective good (e.g., [57]; [67]). Evidence suggests that the public responds differently to the industry and activist sides because of enduring skepticism about the undue influence of business interests ([25]; [49]. While attitudes toward individual industries vary, public favorability ratings of business have trended downward for decades and eroded further since the 2008 financial crisis ([ 1]; [68]).
A legitimacy theory perspective indicates that response to direct-to-public persuasion from the industry and activist sides should be asymmetric due to a legitimacy gap caused by the belief that industry-serving interests drive industry-side motives whereas commitment to the public good drives activist-side motives ([56]). Normative and sociopolitical legitimacy underlie the public's judgment of whether an industry practice is acceptable—meeting cultural and political norms—or should be sanctioned (e.g., [40]; [61]). Legitimacy judgments are likely to discourage many voters from initially supporting the industry side, but election polling and outcomes suggest that voters come to judge the industry more favorably during the course of a campaign. For example, in recent ballot measures that sought genetically modified organism labeling on food products (Oregon Measure 92 in 2014), regulations on dialysis clinic pricing (California Proposition 8 in 2018), and expansion of bottle deposit laws (Massachusetts Question 2 in 2014), the majority initially supported the activist side but switched to support the industry position, allowing the industry to win the vote (see https://ballotpedia.org/List_of_ballot_measures_by_year).
Argumentation is defined as a strategic approach used to justify a particular policy or political position and promote or challenge its implications ([ 6]; [20]). Effective argumentation is critical for industries battling controversy and for activist groups that confront them, as both strive to legitimize their positions ([21]; [39]). We explore the effectiveness of each side's argumentation on two key dimensions: argument strength and argument suspicion. Research on argument strength, including studies in competitive political contexts (e.g., [14]), has not concurrently examined the effects of argument suspicion.
We define "argument strength" as perceived persuasiveness determined by an individual's cognitive response, either favorable or unfavorable, to an argument ([ 4]; [71]). Empirical findings show that strong arguments consistently shift attitudes and beliefs but to varying degrees (e.g., [ 3]). Although argument strength has had limited study in real-world scenarios of competing arguments over time, some research reports that in the presence of a strong argument from one side, a weaker argument from the opposing side is viewed even more negatively and can actually backfire ([14]). As baseline predictions, we hypothesize:
- H1: Industry argument strength increases (a) favorable attitudes toward and (b) voting in favor of the industry side on the issue.
- H2: Activist argument strength increases (a) favorable attitudes toward and (b) voting in favor of the activist side on the issue.
We define "argument suspicion" as the perception that an argument is implausible or has a hidden intent (e.g., [13]; [24]). Although there is far less theoretical and empirical knowledge on the influence of argument suspicion than argument strength, marketing research on persuasion knowledge and skepticism toward advertising offers rich evidence that suspicion-driven responses to persuasion can have significant negative effects on attitudes and behaviors.
Consumers draw on persuasion knowledge, beliefs about how marketers influence consumers, to determine whether something is a persuasion attempt (e.g., [13]). In our research scenario, arguments have clear intent to persuade to vote no or vote yes, but whether the arguments' tactics are viewed as appropriate impacts their effectiveness (e.g., [34]). In some cases, voters may infer finer-level motives in an argument's tactics that trigger suspicion (e.g., Is the "vote no" side using arguments that demonize the "vote yes" side because it lacks more valid arguments?; [37]). Research on skepticism toward advertising messages has examined dispositional ad skepticism (e.g., [48]) as well as situational skepticism (e.g., [22]). Marketing studies suggest that when voters respond to an industry's arguments with skepticism, they are likely to scrutinize its underlying intentions, suspicious that they are deceptive as well as self-serving (see, e.g., [22]).
While this suggests that the public is inclined to view industry arguments as suspicious, other research suggests that activist arguments can also face public doubt, such as when they appear to exaggerate the problems and risks related to contested issues ([ 7]; [62]). As such, we expect voter evaluations of argument suspicion to be relevant for both sides on the issue. We predict the following:
- H3: Industry argument suspicion decreases (a) favorable attitudes toward and (b) voting in favor of the industry side on the issue.
- H4: Activist argument suspicion decreases (a) favorable attitudes toward and (b) voting in favor of the activist side on the issue.
While we predict that argument strength will have a favorable effect for both the industry and activist sides, there is some evidence suggesting that effects are not equally favorable. We suggest, based on the difference in sociopolitical legitimacy ([ 9]; [61]), that strong industry arguments will have a weaker effect relative to strong activist arguments. Due to their negative views of industry practices, voters' legitimacy judgments may trigger doubt of the underlying intent of industry arguments. As a result, voters may not be as influenced even by strong arguments to support the industry's side as by strong arguments to support the activist side. For example, voters assessing a proposition to cap drug prices (e.g., Proposition 61 in Table 1) may mistrust the motives underlying pharmaceutical industry pricing practices, which could diminish the impact of a strong industry argument. In contrast, because voters are likely to have more positive legitimacy judgments of activist groups (e.g., the AIDS Healthcare Association, which works to make life-saving drugs more affordable), the strength of their arguments will not be diminished. This type of voter response is consistent with research showing that when a prior ad triggers doubts about that ad's intent, the effects of argument strength in subsequent ads are diminished ([16]). We hypothesize:
- H5: Industry argument strength has a weaker positive effect than activist argument strength on (a) attitudes toward the issue and (b) voting on the issue.
Voters' responses to influence attempts reflect their persuasion beliefs, and they are likely to be far more guarded against industry arguments than activist arguments (see [37]). This initial disadvantage for the industry side may ultimately offer it an opportunity to use campaign tactics that lead voters to put more weight on activist argument suspicion and/or less weight on industry argument suspicion.
A method used by industries to increase voter suspicion of activist opponents is to intensify the public's doubt about activist-supported policy changes. Industries often argue that activist-side policy positions are impractical, ill-conceived, unrealistic, and costly ([ 6]; [31]). For instance, industry arguments might warn of hidden complexities or expenses, implying that activist arguments are not as genuine and aboveboard as they may seem ([19]). The tactic of using cogent, rational challenges to activist arguments is likely viewed as credible, as it seems cautious and shows due diligence. With this approach, the industry can heighten voters' formerly low persuasion knowledge of the activist side, increasing the weight of activist argument suspicion in voting decisions. This aligns with findings that previously trusted persuasion claims are further scrutinized when they are revealed to have hidden intentions (e.g., [13]), particularly in competitive contexts such as ours (e.g., [14]).
[44] study of a 2012 California ballot measure (Proposition 37) to require labeling genetically engineered foods supports this view. They found that industry advertising helped flip voter support for the activist side (around 70% less than four weeks prior to the election) to majority support for the industry. Industry advertising focused on loopholes that would make implementation difficult and costly, while activist advertising (which lost traction over time) focused on the ruthlessness and deceit of the large industries involved. Although research in ballot measure, referendum, and other issue-focused election scenarios is limited, similar tactics are described in other sources (e.g., [33]; [53]). Following this logic and evidence, we hypothesize:
- H6: Industry argument suspicion has a weaker negative effect relative to activist argument suspicion on (a) attitudes toward the issue and (b) voting on the issue.
Direct-to-public persuasion is most effective when it actually switches voter support from one side to the other. While brand switching is well examined, less is known about vote switching, when persuasion campaigns are short in duration and have a common deadline ([58]). An analysis of vote switching is needed to show whether voting outcomes are due to each side's arguments solidifying voters' initial support or causing them to switch their support over the course of a campaign ([38]).
There is limited guidance on whether the response asymmetry our model predicts would hold for voters who switched their support compared with those who did not or whether the drivers of switching could differ for those initially supporting the industry side compared with those initially supporting the activist side. Swings in public support are well documented by public opinion research, yet there is limited knowledge of how competing campaign arguments motivate voters to switch allegiance. Because research provides little insight on which to base predictions, we do not formally hypothesize effects on switching behavior but rather allow the empirical results to address the knowledge gaps.
To test our hypotheses, we conducted a large field study examining attitudes toward and voting on two California ballot measures involving distinct issues from different industries: Proposition 56, which proposed a tobacco tax increase, and Proposition 61, which proposed price limits on prescription drugs purchased for state programs (for examples of campaign media, see Web Appendix W1). These propositions featured highly salient, consumer-relevant issues in which the industries invested significantly in direct-to-public persuasion campaigns to defeat the measures. By investigating different issues and industries, we reduce the potential that our analyses capture idiosyncratic factors particular to one issue or one industry.
We examined the impact of industry and activist arguments on attitudes and voting using survey data we collected using a national consumer research panel of California voters in two waves: five weeks before (measuring respondent characteristics, preelection attitudes, and voting intentions) and one week after (measuring argument strength, argument suspicion, postelection attitudes, and self-reported voting) the November 2016 national election. We collected data for the same respondents in both waves to analyze the extent to which industry- and activist-side persuasion campaigns shifted individual voters' attitudes and voting.
We recruited respondents through Qualtrics, an online market research panel aggregator whose quality certification includes checking each person's IP address to exclude duplication and replacing those who finish the survey in less than one-third of the average completion time. Respondents were registered voters in California who indicated "I will vote" in the election in a screening question at the beginning of the preelection survey. After the election, we invited all 1,806 respondents of the preelection survey to answer the postelection survey. Of the 904 postelection survey respondents who voted in the election,[ 6] 58.4% were female, the mean age was 51 years old, 92.1% had some college education or higher, and 70.7% reported their socioeconomic status (SES) as middle to upper class.
In the preelection survey, we captured preelection attitudes toward the issue in the context of the drug price and tobacco tax propositions using a three-item ("wrong/right," "harmful/beneficial," and "unacceptable/acceptable"), seven-point, previously validated scale and voting intention on the issue for both propositions as a binary variable (0 = intend to vote "no," in favor of the industry side; 1 = intend to vote "yes," in favor of the activist side). In addition, we measured several individual-level characteristics, including political affiliation, need for orientation, and demographic characteristics (age, gender, education, and SES).
In the postelection survey, we measured our two dependent variables, attitudes toward the issue and voting on the issue, using the same multi-item scale used to capture preelection attitudes and a binary variable (0 = voted "no," 1 = voted "yes"), respectively. Respondents were then presented with two principal arguments from both the industry and activist sides for both propositions. To emphasize external validity, we identified the principal arguments from the California Official Voter Information Guide published by the office of the [12], which includes a set of Official Arguments formally submitted by each side (see Web Appendix W2). Respondents were asked to indicate their perceptions of argument strength and argument suspicion for each pair of industry arguments and activist arguments. We measured the two variables with two-item, seven-point, previously validated scales measuring the extent to which each of the arguments was convincing and aroused suspicion.[ 7] Finally, we measured familiarity with the arguments. For all multi-item scales, the mean of the items was used in the model estimation. Table 2 presents the descriptive statistics for all variables, and measurement details are provided in Web Appendix W2.
Graph
Table 2. Descriptive Statistics for Field Study Variables.
| Variables | Drug PriceProposition M (SD) | Tobacco TaxPropositionM (SD) | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 |
|---|
| 1. Attitudes toward issue | 4.25 (2.09) | 5.10 (2.10) | 1 | −.408 | .753 | .006 | −.561 | .724 | −.240 | .162 | −.224 | .064 | −.126 | .100 | .118 | .010 | −.865 | .672 |
| 2. Industry argument strength | 4.02 (1.78) | 3.64 (1.80) | −.448 | 1 | −.257 | −.157 | .406 | −.347 | .394 | −.009 | .099 | .027 | −.031 | −.071 | −.048 | .027 | .385 | −.345 |
| 3. Activist argument strength | 4.41 (1.85) | 4.54 (1.93) | .739 | −.332 | 1 | .109 | −.546 | .628 | −.147 | .319 | −.205 | .161 | −.150 | .040 | .117 | −.011 | −.707 | .555 |
| 4. Industry argument suspicion | 3.96 (1.73) | 4.16 (1.79) | .220 | −.255 | .289 | 1 | .045 | .079 | .062 | .025 | −.045 | −.020 | .008 | .041 | −.022 | −.073 | −.078 | .072 |
| 5. Activist argument suspicion | 3.84 (1.90) | 3.51 (2.01) | −.535 | .478 | −.488 | −.006 | 1 | −.438 | .228 | −.068 | .150 | −.040 | .001 | −.023 | −.085 | .000 | .524 | −.419 |
| 6. Preelection attitudes toward issue | 4.48 (1.86) | 4.94 (2.11) | .448 | −.272 | .433 | .184 | −.273 | 1 | −.215 | .091 | −.218 | .088 | −.120 | .112 | .134 | .012 | −.708 | .832 |
| 7. Familiarity with industry arguments | 4.88 (1.71) | 4.73 (1.79) | −.178 | .401 | −.091 | −.007 | .237 | −.112 | 1 | .378 | .046 | .120 | .138 | −.004 | .017 | −.118 | .268 | −.261 |
| 8. Familiarity with activist arguments | 5.24 (1.52) | 5.46 (1.53) | .255 | −.042 | .400 | .110 | −.081 | .087 | .359 | 1 | −.018 | .208 | .060 | .002 | .010 | −.028 | −.086 | .040 |
| 9. Political affiliation | 3.31 (2.15) | 3.30 (2.14) | −.240 | .140 | −.273 | −.139 | .181 | −.244 | .037 | −.115 | 1 | −.097 | .083 | −.111 | .011 | −.102 | .224 | −.182 |
| 10. Need for orientation | 5.32 (1.42) | 4.41 (1.78) | .057 | .025 | .106 | .059 | −.001 | .176 | .128 | .226 | −.057 | 1 | −.010 | −.008 | −.011 | −.037 | −.053 | .030 |
| 11. Age | 51.33 (15.60) | 51.02 (15.70) | −.109 | −.067 | −.156 | −.037 | .020 | −.134 | .060 | .055 | .069 | .115 | 1 | .003 | .066 | −.054 | .122 | −.134 |
| 12. Education | 3.23 (.58) | 3.23 (.58) | .044 | .025 | −.032 | .023 | .028 | .038 | −.007 | −.038 | −.109 | −.040 | −.001 | 1 | .321 | −.078 | −.087 | .107 |
| 13. SES | 2.83 (.88) | 2.85 (.88) | .020 | .043 | −.028 | −.008 | .013 | .005 | .042 | −.014 | .026 | −.064 | .083 | .305 | 1 | −.083 | −.145 | .129 |
| 14. Gender | M = 41.6%, F = 58.4% | M = 41.6%, F = 58.4% | −.032 | −.018 | .023 | −.028 | −.064 | −.035 | −.093 | −.014 | −.088 | .066 | −.052 | −.092 | −.087 | 1 | .026 | .006 |
| 15. Voting on the issue | Y = 52.3% N = 47.7% | Y = 68.2% N = 31.8% | −.789 | .437 | −.660 | −.191 | .482 | −.381 | .196 | −.207 | .249 | −.057 | .124 | −.070 | .019 | .026 | 1 | −.704 |
| 16. Preelection voting intention on issue | Y = 62.0% N = 38.0% | Y = 67.2% N = 32.8% | .381 | −.248 | .360 | .130 | −.274 | .722 | −.099 | .059 | −.211 | .135 | −.188 | −.027 | −.036 | −.021 | −.394 | 1 |
2 Notes: Correlation coefficients below diagonal relate to drug price proposition (>|.067| are significant at p < .05; N = 851). Correlation coefficients above diagonal relate to tobacco tax proposition (>|.066| are significant at p < .05; N = 871).SES = socioeconomic status.
We tested our hypotheses by estimating attitudes toward and voting on the issue as a function of argument strength and argument suspicion. For postelection attitudes toward the issue (Attitudest,i), we used an ordinary least squares (OLS) approach, specified as follows:
Graph
where the key predictor variables are industry argument strength (Industry_Argument_Strengtht,i), activist argument strength (Activist_Argument_Strengtht,i), industry argument suspicion (Industry_Argument_Suspiciont,i), and activist argument suspicion (Activist_Argument_Suspiciont,i). Covariates include preelection attitudes toward the issue (Attitudest − 1,i), familiarity with industry and activist arguments (Familiar_Industryt,i, Familiar_Activistt,i), political affiliation (Political_Affiliationi), need for orientation (Orientationi), age (Agei), gender (Genderi), education (Educationi), and socioeconomic status (SESi). For our voting on the issue outcome (Votet,i), we used a logit model with the same predictors and covariates as the attitudes toward the issue model, substituting preelection voting intentions (Votet − 1,i) in place of preelection attitudes.
Our analysis was subject to several forms of bias, which we wanted to resolve. First, because we measured our dependent variables and key predictor variables using the same instrument, we adopted [41] approach to test for common method variance using a method factor capturing need for orientation on an unrelated ballot measure included in the survey instrument. Correlation coefficients corrected for common method variance indicate that this does not significantly bias our data (see Web Appendix W3).
Second, we addressed endogeneity due to either reverse causality between argument strength and attitudes toward the issue or omitted variables using two-stage least squares estimation and comparing the estimated coefficients with the OLS estimated coefficients. We specified the first-stage equation to include a set of exogenous variables (the covariates from the Attitudest,i model) plus an instrumental variable, Reactancei, capturing individual i's resistance to a perceived threat to free choice or behavior. The instrument was relevant (i.e., correlated with argument strength) and met the exclusion criterion (i.e., uncorrelated with the error term of the Attitudest,i equation). In the second-stage equation, the estimated residuals from the first-stage regression were included in the Attitudest,i model to control for the endogenous regressors. Our test indicates that endogeneity does not bias the OLS coefficients (for details regarding the appropriateness of our instrument and model results, see Web Appendix W4).
Third, because a subset of respondents from the preelection survey opted to complete the postelection survey, we addressed selection bias using [30] two-step estimation of the effect of argument strength on attitudes toward and voting on the issue. In the first-stage selection equation, Votedt,i (binary variable indicating whether or not the respondent voted on the proposition in the election) is regressed on our covariates plus the conceptually unrelated variable Reactancei. The second-stage outcome equations for attitudes and voting were estimated using the Attitudest,i and Votet,i model specifications, respectively, conditioned on whether the respondent voted on the proposition in the election. From the estimated coefficients for the second-stage model, including the inverse Mills ratio, we conclude that selection bias did not adversely affect model estimation (see Web Appendix W5).
A model-free examination of the data offers preliminary evidence of our predicted relationships (see Web Appendix W6). Table 3 contains the results related to our hypothesized effects. Note that for our analysis, as industry is opposed to the proposition, negative coefficients indicate favorable attitudes toward and voting in favor of the industry side. We find consistent support for H1 and H2. Stronger industry arguments led to more favorable attitudes toward the industry side on both the drug price (βIndustry_Argument_Strength = −.178, p < .01) and tobacco tax (βIndustry_Argument_Strength = −.128, p < .01) propositions and increased the likelihood of voting in favor of the industry side on the drug price (βIndustry_Argument_Strength = −.623, p < .01; 65.1% likelihood of voting no[ 8]) and tobacco tax (βIndustry_Argument_Strength = −.476, p < .01; 61.7% likelihood of voting no) propositions. Similarly, stronger activist arguments led to more favorable attitudes toward this side for the drug price (βActivist_Argument_Strength = .630, p < .01) and tobacco tax (βActivist_Argument_Strength = .471, p < .01) propositions and increased the likelihood of voting in favor of this side for the drug price (βActivist_Argument_Strength = 1.300, p < .01; 78.6% likelihood of voting yes) and tobacco tax (βActivist_Argument_Strength = 1.206, p < .01; 77.0% likelihood of voting yes) propositions.
Graph
Table 3. Impact of Industry and Activist Argument Strength and Suspicion on Attitudes Toward and Voting on the Issue.
| Hyp. | Variables | Attitudes Toward the Issue | Voting on the Issue |
|---|
| Drug Price Proposition | Tobacco Tax Proposition | Drug Price Proposition | Tobacco Tax Proposition |
|---|
| Coefficient | Effect Size | Coefficient | Effect Size | Coefficient | Odds Ratio | Coefficient | Odds Ratio |
|---|
| Constant | 1.672*** | | 2.413*** | | –3.126*** | | –1.672* | |
| Predictor Variables | | | | | | | | |
| H1 | Industry argument strength | −.178*** | .035 | −.128*** | .026 | −.623*** | .536 | −.476*** | .621 |
| H2 | Activist argument strength | .630*** | .285 | .471*** | .202 | 1.300*** | 3.657 | 1.206*** | 3.339 |
| H3 | Industry argument suspicion | −.008 | .000 | −.026 | .002 | .194** | 1.214 | .078 | 1.082 |
| H4 | Activist argument suspicion | −.172*** | .037 | −.120*** | .026 | −.391*** | .676 | −.283*** | .753 |
| Covariates | | | | | | | | |
| Preelection attitudes toward the issue | .129*** | .025 | .349*** | .185 | | | | |
| Preelection voting intention on the issue | | | | | 1.233*** | 3.430 | 2.604*** | 13.520 |
| Familiarity: industry arguments | −.032 | .001 | −.027 | .001 | −.088 | .916 | −.092 | .912 |
| Familiarity: activist arguments | .023 | .001 | .003 | .000 | −.177* | .838 | −.221** | .801 |
| Political affiliation | −.007 | .000 | −.028* | .002 | −.072* | .931 | −.145** | .865 |
| Need for orientation | −.015 | .000 | −.047** | .005 | −.069 | .934 | −.005 | 1.005 |
| Age | −.002 | .001 | −.002 | .001 | −.006 | .994 | .002 | 1.002 |
| Gender | .213** | .006 | −.028 | .000 | .284 | 1.328 | .424* | 1.527 |
| Education | .192** | .007 | .117* | .003 | .692*** | 1.998 | −.090 | .914 |
| SES | .060 | .002 | −.009 | .000 | −.090 | .914 | .218 | 1.244 |
| N | 851 | | 871 | | 851 | | 871 | |
| R2 | .628 | | .703 | | .534 | | .574 | |
| AIC | | | | | 1179.760 | | 1091.398 | |
| H5 | Test of equality of industry vs. activist argument strength coefficients (t-value) | −4.077*** | | −2.821*** | | −3.551*** | | −3.400*** | |
| H6 | Test of equality of industry vs. activist argument suspicion coefficients (t-value) | −2.618*** | | −1.969** | | −2.249** | | −2.192** | |
- 3 *p < .10.
- 4 **p < .05.
- 5 ***p < .01.
- 6 Notes: One-tailed tests of significance. AIC = Akaike information criterion. SES = socioeconomic status. For attitudes toward the issue, negative coefficients indicate favorable attitudes toward the industry side on the issue (as industry opposes the proposed policy); positive coefficients indicate favorable attitudes toward the activist side on the issue (as activist side supports the proposed policy). For voting on the issue, negative coefficients indicate voting in favor of the industry side on the issue (as industry opposes the proposed policy); positive coefficients indicate voting in favor of the activist side on the issue (as the activist side supports the proposed policy).
Our results offer mixed support for H3, as suspicion of industry arguments had little impact on voting outcomes. It had no effect on attitudes toward either proposition, contrary to H3a. It significantly decreased the likelihood of voting in favor of the industry side for the drug price proposition (βIndustry_Argument_Suspicion = .194, p < .05; 45.2% likelihood of voting no), but not for the tobacco tax proposition, partially supporting H3b. By contrast, suspicion of activist arguments played a consistent role, fully supporting H4, resulting in less favorable attitudes toward that side for the drug price (βActivist_Argument_Suspicion = −.172, p < .01) and tobacco tax (βActivist_Argument_Suspicion = −.120, p < .01) propositions and decreased likelihood of voting in favor of that side for the drug price (βActivist_Argument_Suspicion = −.391, p < .01; 40.3% likelihood of voting yes on the proposition) and tobacco tax (βActivist_Argument_Suspicion = −.283, p < .01; 43.0% likelihood of voting yes) propositions.
To compare the relative impact of industry and activist argument strength in predicting our key outcomes, we used the delta method to test the equality of the estimated coefficients for the two variables ([28]). Specifically, we calculated t = (βIndustry_Argument_Strength − βActivist_Argument_Strength)/SE, where βIndustry_Argument_Strength and βActivist_Argument_Strength are the absolute value of the two parameter estimates and SE is the standard error of their difference calculated as Sqrt[Var(βIndustry_Argument_Strength) + Var(βActivist_Argument _Strength) − 2 × Cov(βIndustry_Argument_Strength, βActivist_Argument_Strength)]. The results provide full support for H5, indicating that activist argument strength has a significantly greater impact than industry argument strength on attitudes toward the drug price (t = −4.077, p < .01) and tobacco tax (t = −2.821, p < .01) propositions and on voting for the drug price (t = −3.551, p < .01) and tobacco tax (t = −3.400, p < .01) propositions. This is consistent with a comparison of effect sizes, which indicates that activist argument strength has a strong effect on attitudes toward both the drug price ( = .285) and tobacco tax ( = .202) propositions, while industry argument strength has a moderate-small effect for the drug price ( = .035) and tobacco tax ( = .026) propositions.[ 9]
Using the delta method to compare the relative impact of industry and activist argument suspicion in predicting attitudes and voting revealed that activist argument suspicion has a significantly greater impact than industry argument suspicion on attitudes toward the drug price (t = −2.618, p < .01) and tobacco tax (t = −1.969, p < .05) propositions and on voting for the drug price (t = −2.249, p < .05) and tobacco tax (t = −2.192, p < .05) propositions, in support of H6. A comparison of effect sizes reflects this difference, such that activist argument suspicion has a moderate-small effect on attitudes toward the issue for both the drug price ( = .037) and tobacco tax ( = .026) propositions, but industry argument suspicion has a negligible effect for the drug price ( = .000) and tobacco tax ( = .002) propositions. Effect sizes for all parameters and overall model statistics are reported in Table 3.
Although our survey respondents' characteristics generally align with the demographics of California voters in the 2016 election based on U.S. Census Bureau data in terms of gender, age, and income, our sample included higher proportions of individuals with higher education levels and whites/Caucasians and lower proportions of Hispanics (see Web Appendix W7). To help ensure that our results are representative of California voters, we performed a robustness check in which we weighted observations using an iterative proportional fitting procedure, which takes the marginal distributions of demographic variables from the California voter population and returns weights such that the marginal distributions of the demographic variables in the weighted survey data match those in the population ([69]). Compared with results for the unweighted models, our analysis finds consistent effects for 15 of the 16 total hypothesized effects of the impact of argument strength and argument suspicion on attitudes toward and voting on the issue for the two propositions (see Web Appendix W8).
Of the two ballot measures we study, the drug price proposition experienced significant swings in public support. Political polling in California leading up to the election indicated that 73% of registered voters supported the proposition in late July ([63]), but this decreased to 66% in mid-September (USC Dornslife/LA Times) and further to 51% in mid-October (Hoover Institute/YouGov). Following intensive pharmaceutical industry media spending in the six weeks before the election, the proposition was defeated in the November election with 48.6% of support (all polling data retrieved from https://ballotpedia.org/California%5fProposition%5f61,%5fDrug%5fPrice%5fStandards%5f(2016)).[10] Such a swing in support did not occur for the tobacco tax proposition.
We considered two scenarios. First, we focused on voters who intended to vote in favor of the activist side in the preelection survey (i.e., would vote yes on the proposition) and measured our dependent variable by identifying those who actually voted in favor of that side (voted yes) versus those who switched and actually voted in favor of the industry side (voted no) as measured in the postelection survey (1 = switched vote, 0 = did not switch vote). Second, we focused on those who intended to vote in favor of the industry side and measured the dependent variable by identifying those who actually voted in favor of the industry side versus those who switched and voted in favor of the activist side.
As Table 4 shows, argument strength and argument suspicion play important roles in vote switching (all results apply to both the drug price and tobacco tax propositions). Strong industry (activist) arguments significantly increased (decreased) the likelihood of voters switching support from the activist side to the industry side and decreased (increased) the likelihood of switching support from the industry side to the activist side. Consistent with our previous analyses, activist argument strength had a greater impact than industry argument strength in both scenarios. The results related to argument suspicion reveal a notable effect that may have been masked in our previous analyses. Consistent with our previous findings, suspicion of activist arguments significantly increased (decreased) voter switching to the industry (activist) side. However, in contrast with our previous findings, suspicion of industry arguments significantly increased the likelihood that voters who initially supported the industry switched their support to the activist side. This distinctive finding is bolstered by a test of equality on the impact of argument suspicion for the opposing sides, which indicates that suspicion of activist arguments does not play a significantly greater role than suspicion of industry arguments in this scenario.
Graph
Table 4. Impact of Industry and Activist Argument Strength and Suspicion on Vote Switching.
| Variables | Switched from Intending to Vote in Favor of Activist Side to Voting in Favor of Industry Side | Switched from Intending to Vote in Favor of Industry Side to Voting in Favor of Activist Side |
|---|
| Drug Price Proposition | Tobacco Tax Proposition | Drug Price Proposition | Tobacco Tax Proposition |
|---|
| Coefficient | Odds Ratio | Coefficient | Odds Ratio | Coefficient | Odds Ratio | Coefficient | Odds Ratio |
|---|
| Constant | .633 | | .553 | | −5.874*** | | −.140 | |
| Predictor Variables | | | | | | | | |
| Industry argument strength | .635*** | 1.887 | .394** | 1.483 | −.592*** | .554 | −.642*** | .526 |
| Activist argument strength | −1.275*** | .280 | −1.315*** | .268 | 1.360*** | 3.895 | 1.306** | 3.689 |
| Industry argument suspicion | −.155 | .856 | .118 | 1.126 | .294** | 1.342 | .400** | 1.487 |
| Activist argument suspicion | .463*** | 1.589 | .287** | 1.332 | −.256** | .774 | −.346** | .707 |
| Covariates | | | | | | | | |
| Familiarity: industry arguments | .202* | 1.224 | .130 | 1.138 | .056 | 1.057 | −.044 | .957 |
| Familiarity: activist arguments | .144 | 1.155 | .243 | 1.275 | −.239 | .787 | −.203 | .817 |
| Political affiliation | .120** | 1.127 | .011 | 1.011 | −.018 | .982 | .367*** | .693 |
| Need for orientation | .057 | 1.059 | −.010 | .990 | −.054 | .948 | .077 | 1.080 |
| Age | .007 | 1.007 | −.005 | .995 | −.006 | .994 | −.002 | .998 |
| Gender | −.163 | .850 | −.726** | .484 | .671** | 1.956 | .091 | .913 |
| Education | −.730*** | .482 | −.448 | .639 | .692** | 1.998 | −1.191** | .304 |
| SES | .169 | 1.184 | −.001 | .999 | .104 | 1.109 | .773** | 2.166 |
| N | 851 | | 871 | | 851 | | 871 | |
| R2 | .460 | | .237 | | .464 | | .432 | |
| AIC | 665.533 | | 352.953 | | 380.347 | | 298.456 | |
| Test of equality of industry vs. activist argument strength coefficients (t-value) | −2.525** | | −3.140*** | | −2.551** | | −1.971** | |
| Test of equality of industry vs. activist argument suspicion coefficients (t-value) | −1.970** | | 1.691* | | .151 | | .191 | |
- 7 *p < .10.
- 8 **p < .05.
- 9 ***p < .01.
- 10 Notes: One-tailed tests of significance. AIC = Akaike information criterion. SES = socioeconomic status. For attitudes toward the issue, negative coefficients indicate favorable attitudes toward the industry side on the issue (as industry opposes the proposed policy); positive coefficients indicate favorable attitudes toward the activist side on the issue (as activist side supports the proposed policy). For voting on the issue, negative coefficients indicate voting in favor of the industry side on the issue (as industry opposes the proposed policy); positive coefficients indicate voting in favor of the activist side on the issue (as the activist side supports the proposed policy).
Our field study findings affirm the relationships predicted in our framework and hypotheses. In the next section, we develop hypotheses predicting how two distinctive and prominent argumentation strategies, one focused on financial factors and one focused on societal factors, vary in effectiveness for the industry and activist sides. We then test our predictions using three experimental studies.
Prior research on perceptual fit suggests that argument effectiveness would vary depending on fit with the persuader (e.g., [60]). The positive effects of high fit between a firm or a brand and entities with which it is associated, such as sponsorships and cause-related marketing, are well established ([54]; [60]). Despite this, [ 6] document that opposing sides, including industry and activist groups, tend to use the same few strategies in the same policy battle, indicating that the fit between the persuader and the argument appears to be often overlooked. This raises an important question: Would a more focused approach be more effective and, if so, how should it be chosen?
To address this question, we examine argumentation strategies consistent with the image of the two sides in the conflicts we study. Financial argumentation is based on factors related to imposing or reducing financial benefits or costs, whereas societal argumentation is based on (nonfinancial) factors related to inhibiting or promoting shared societal goals or values. With financial argumentation, an industry may claim that there will be onerous financial costs for the public if current policy changes, whereas an activist group may argue that current policy needlessly burdens taxpayers ([19]; [67]). Using societal argumentation, an industry may claim that a policy change would weaken consumer access to important products or services, whereas an activist group may argue that current policy serves only a privileged few ([57]).
Fit between an organization and an argument is high when the two are perceived as congruent because they share similar intangible associations ([60]). Argumentation based on financial factors is congruent with how industries are viewed, as an industry's image is characterized by economic and operational aspects ([10]). Analysts' reports profile industries in terms of financial and market performance and industry actions directly affect the economic status of consumers (e.g., [ 2]). Because financial argumentation is more congruent with public expectations of industry, this strategy is a better fit with the industry side and will lead to more effective persuasion ([54]). As such, we predict that financial argumentation will be perceived as stronger and less suspicious when used by the industry side.
In contrast, argumentation based on societal factors is congruent with the image of activist groups, which generally are perceived as organizations that represent the collective good for citizens ([59]). These organizations build consensus on higher-level principles and fundamental values such as social justice ([ 7]; [56]). Because the activist side is focused on promoting shared societal goals or values, its image is more congruent with societal argumentation. As such, we predict that societal argumentation will be perceived as stronger and less suspicious when used by the activist side.
Although empirical evidence to inform our predictions is limited, research on organizational legitimacy provides some conceptual support by suggesting that industries battling to control policy most often assume a market and economic orientation, expressed by referencing price, competitiveness, and efficiency, whereas groups representing the public interest often assume a civic orientation, expressed by referencing collective interest, solidarity, and integrity ([10]). In summary, we predict:
- H7: The effect of argumentation strategy on argument strength and argument suspicion is moderated by the competing side on the issue, such that
- For the industry side, financial argumentation leads to higher argument strength and lower argument suspicion than societal argumentation.
- For the activist side, societal argumentation leads to higher argument strength and lower argument suspicion than financial argumentation.
- For financial argumentation, use by the industry side leads to higher argument strength and lower argument suspicion than use by the activist side.
- For societal argumentation, use by the activist side leads to higher argument strength and lower argument suspicion than use by the industry side.
Building on our previous predictions that argument strength and argument suspicion impact attitudes toward and voting on the issue, we expect these evaluations to mediate the effect of an argumentation strategy on voting outcomes. Specifically:
- H8: The mediating role of argument strength and argument suspicion depends on the competing side on the issue, such that
- For the industry side, financial (vs. societal) argumentation increases argument strength and decreases argument suspicion, which then increases favorable attitudes toward and voting in favor of the industry side on the issue.
- For the activist side, societal (vs. financial) argumentation increases argument strength and decreases argument suspicion, which then increases favorable attitudes towards and voting in favor of the activist side on the issue.
We test H7 and H8 in three experimental studies on different issues addressed in recent ballot measures (see Table 1): pharmaceutical drug price standards, which aligns with one of our field study contexts; recyclable bottle deposits; and renewable energy standards. In a controlled, randomized setting, we exposed participants to two opposing messages (one from each side on the issue) that used either financial or societal argumentation strategies, thus separating the effects of argumentation strategy from those of frequency of exposure ([14]).
We used a mixed experimental design with two manipulations administered at two points in time within the same survey. Participants were first introduced to a hypothetical scenario in which the residents of a U.S. state are scheduled to vote on a proposition to adopt statewide pharmaceutical drug price standards and were sequentially shown two messages related to the proposition. First, participants saw one of four messages in a 2 (argumentation strategy: financial vs. societal) × 2 (side on the issue: industry vs. activist) between-subjects experimental design (time 1). Next, in the same survey, we manipulated argumentation strategy within subject; depending on which argumentation strategy participants saw at time 1, they saw one of the two strategies (financial or societal) for the opposing side (time 2). In each condition, participants saw a different combination of arguments from each side (both financial; both societal; or one of each, in a different order), enabling us to examine perceptions of argument strength and suspicion of each strategy in the presence of opposing arguments.
In the introduction, participants were told that Proposition 25 asks voters to decide whether their state should adopt standards to restrict the amount that some state agencies and programs pay for selected prescription drugs (for stimuli, see Web Appendix W9). We then introduced each message one at a time, indicating that it is a message designed by supporters (opponents) of Proposition 25 to persuade voters to vote yes (no) and support (defeat) the proposition. We derived the arguments used in our manipulations from the official arguments in the California Official Voter Information Guide for Proposition 61 (Drug Price Standards, November 8, 2016).[11]
After participants viewed the two arguments, they indicated their attitudes toward the proposition and how likely they were to vote yes or no in the referendum (1 = "I will definitely vote NO and oppose this proposition," and 7 = "I will definitely vote YES and support this proposition"). Next, they saw the message viewed at time 1 and evaluated its argument strength and suspicion. After this, they undertook the same evaluations for the message viewed at time 2. We then assessed political affiliation, need for orientation, age, gender, education, and SES. Detailed descriptions of each of the measures appear in Web Appendix W10.
To enhance external validity, we recruited a nationally representative sample of registered voters. We partnered with Qualtrics to recruit 659 registered voters (49.68% female; Mage = 52.05 years; age range: 18–91 years). Qualtrics enforced quotas so that the sample was proportional to the U.S. population in terms of region, gender, age, household income, education, and ethnicity based on census data. Because the issue was voted on in California in 2016 and Ohio in 2017, we dropped 33 participants who indicated residency in these states, which left a sample of 626 participants for analyses. We conducted a pretest to confirm that our argumentation strategy manipulation was successful (for details, see Web Appendix W11).
Given the within-subject nature of our design, in which we measure strength and suspicion of the second argument after participants evaluate the first argument, carryover and anchoring effects inherent to the design limit the interpretation of the second message evaluation ([51]). Thus, to test the hypothesized effects in the presence of arguments from both sides, while ensuring that the argument strength and suspicion measures are not subject to bias, we use participants' evaluations of the first argument and control for the argument evaluated second.
We used PROCESS Model 58 ([29]) to test our full conceptual model (see Table 5). An OLS regression on argument strength, with argumentation strategy (1 = financial, 0 = societal), side on the issue (1 = industry, 0 = activist), and their interaction as predictors. Covariates were individual-level characteristics and the argumentation strategy participants saw at time 2. The analysis revealed significant negative main effects of argumentation strategy (bfinancial = −.529, p < .01) and side on the issue (bindustry = −.607, p < .01) and the predicted interaction effect (bfinancial × industry = .841, p < .01). A similar regression on argument suspicion revealed significant positive main effects of argumentation strategy (bfinancial = .384, p <= .05) and side on the issue (bindustry = .627, p < .01) and the predicted interaction effect (bfinancial × industry = −.744, p < .01).
Graph
Table 5. Response to Argumentation Strategies from Competing Sides: Drug Price Standards Proposition.
| | Dependent Variables |
|---|
| Hyp. | Predictor Variables | Coefficient (Effect Size,) |
|---|
| Moderating Effect of Issue Side on Argumentation Strategy | Argument Strength |
| Financial argumentation strategy (time 1, 0 −1) | −.529*** (.001) |
| Industry side (time 1, 0−1) | −.607*** (.003) |
| Financial argumentation strategy × Industry side | .841*** (.026) |
| Intercept | 4.104*** |
| R2 | .123 |
| Model significance | F(10, 611) = 8.581*** |
| Moderating Effect of Issue Side on Argumentation Strategy | Argument Suspicion |
| Financial argumentation strategy (time 1, 0−1) | .384** (.000) |
| Industry side (time 1, 0−1) | .627*** (.007) |
| Financial argumentation strategy × Industry side | −.744*** (.017) |
| Intercept | 3.747*** |
| R2 | .044 |
| Model significance | F(10, 611) = 2.851*** |
| Effects of Argument Strength and Argument Suspicion | Attitudes Toward the Issue | Voting on the Issue |
| Financial argumentation strategy (time 1, 0−1) | .014 (.000) | .019 (.000) |
| Argument strength | .589*** (.032) | .631*** (.045) |
| Argument suspicion | −.266*** (.051) | −.211*** (.050) |
| Industry side (time 1, 0 −1) | .766 (.000) | .621 (.004) |
| Argument strength × Industry side | −.683*** (.073) | −.681*** (.065) |
| Argument suspicion × Industry side | .533*** (.210) | .509*** (.186) |
| Intercept | 2.134*** | 2.016*** |
| R2 | .330 | .299 |
| Model significance | F(13, 608) = 23.063*** | F(13, 608) = 19.977*** |
| H8a | Conditional indirect effect through argument strength, industry | −.029 [−.120,.032] | −.015 [−.100,.053] |
| H8a | Conditional indirect effect through argument suspicion, industry | −.096 [−.230, −.006] | −.107 [−.261, −.007] |
| H8b | Conditional indirect effect through argument strength, activist | −.312 [−.564, −.101] | −.334 [−.578, −.110] |
| H8b | Conditional indirect effect through argument suspicion, activist | −.102 [−.245, −.005] | −.081 [−.207, −.001] |
| Index of moderated mediation (argument strength) | .282 [.059,.541] | .319 [.086,.576] |
| Index of moderated mediation (argument suspicion) | .006 [−.158,.174] | −.026 [−.203,.130] |
| Simple Effects of Issue Side on Argument Strength and Suspicion | Argument Strength | Argument Suspicion |
| H7a | Simple effects of financial (vs. societal) argumentation for industry side | .312* | −.360** |
| H7b | Simple effects of financial (vs. societal) argumentation for activist side | −.529*** | .384** |
| H7c | Simple effects of industry (vs. activist) side for financial argumentation | .234 | −.117 |
| H7d | Simple effects of industry (vs. activist) side for societal argumentation | −.607*** | .627*** |
- 11 *p < .10.
- 12 **p < .05.
- 13 ***p < .01.
- 14 Notes: Results from PROCESS Model 58, which includes need for orientation, political affiliation, age, gender (male), education, SES, and argumentation strategy at time 2 (0−1) as covariates; 95% confidence intervals reported for conditional indirect effects; n = 622 for these analyses due to missing values on age and SES covariates.
Tests of the simple effects of argumentation strategy for each side on the issue showed that for the industry side, financial argumentation led to marginally significantly higher argument strength (bfinancial = .312, p < .10) and significantly lower argument suspicion (bfinancial = −.360, p < .05), as compared with societal argumentation, in support of H7a. For the activist side, societal argumentation led to significantly higher argument strength (bfinancial = −.529, p < .01) and lower argument suspicion (bfinancial = .384, p < .05), as compared with financial argumentation, supporting H7b (negative coefficients indicate enhanced effects and positive coefficients indicate diminished effects because side on the issue is coded as industry = 1 and activist = 0).
Tests of the simple effects of side on the issue for each argumentation strategy revealed that, for financial argumentation, use by the industry side was not significantly different than use by the activist side for both argument strength (bindustry = .234, p > .10) and argument suspicion (bindustry = −.117, p > .10), contrary to H7c. However, in support of H7d, for societal argumentation, use by the activist side led to significantly higher argument strength (bindustry = −.607, p < .01) and significantly lower argument suspicion (bindustry = .627, p < .01) than use by the industry side.
For the full moderated mediation model, for the industry side, argument suspicion mediated the effects of argumentation strategy on attitudes and voting (conditional indirect effect on attitude: b = −.096, 95% confidence interval [CI]: [−.230, −.006]; conditional indirect effect on voting: b = −.107, 95% CI: [−.261, −.007]), but argument strength did not (conditional indirect effect on attitude: b = −.029, 95% CI: [−.120,.032]; conditional indirect effect on voting: b = −.015, 95% CI: [−.100,.053]), partially supporting H8a. For the activist side, both argument strength (conditional indirect effect on attitude: b = −.312, 95% CI: [−.564, −.101]; conditional indirect effect on voting: b = −.334, 95% CI: [−.578, −.110]) and argument suspicion (conditional indirect effect on attitude: b = −.102, 95% CI: [−.245, −.005]; conditional indirect effect on voting: b = −.081, 95% CI: [−.207, −.001]) mediated the effects of argumentation strategy on attitudes and voting, fully supporting H8b.[12]
To assess the generalizability of our findings across different issues, we conducted two additional experimental studies. Both studies used the same design as Experimental Study 1. Experimental Study 2 (331 U.S.-based participants from the Prolific Academic panel; 44.11% female; Mage = 33.19 years, age range: 18–73 years) focused on expanding a state's bottle deposit law to require deposits for all nonalcoholic, noncarbonated drinks (Expansion of Bottle Deposits Initiative, Massachusetts, November 4, 2014). Experimental Study 3 (340 U.S.-based participants from Prolific Academic; 50.88% female; Mage = 34.70 years, age range: 18–73 years) focused on increasing a state's renewable energy standards, requiring electricity providers to obtain at least 50% of their electricity from renewable sources by 2030 (Renewable Energy Standards Initiative, Arizona, November 6, 2018). The stimuli are presented in Web Appendix W9 and pretests to confirm our argumentation strategy manipulations are included in Web Appendix W11.
Results from the two studies were largely identical to those of Experimental Study 1 (see Table 6). For both propositions, as expected and in line with Experimental Study 1, for the industry side, financial argumentation led to higher argument strength and lower argument suspicion than societal argumentation, while for the activist side, societal argumentation led to higher argument strength and lower argument suspicion than financial argumentation. Comparing side on the issue effects for each argumentation strategy, for societal argumentation, use by the activist side led to higher argument strength and lower suspicion than use by the industry side but, as in Experimental Study 1 and contrary to predictions, participants did not perceive financial argumentation as significantly stronger or less suspicious when used by the industry versus the activist side. Across both propositions, argument strength mediated the effect of argumentation strategy on attitudes toward and voting on the issue for both the industry and activist sides. Argument suspicion also had significant mediating effects for the industry and activist sides across both propositions with two exceptions. For the renewable energy standards proposition, argument suspicion mediated the effect of argumentation strategy on voting but not on attitudes toward the industry side, and argument suspicion had a significant mediating effect on attitudes but not on voting for the activist side.[13]
Graph
Table 6. Response to Argumentation Strategies from Competing Sides: Bottle Deposits and Energy Standards Propositions.
| | Bottle Deposits Proposition | Energy Standards Proposition |
|---|
| Hyp. | Predictor Variables | DV: Coefficient (Effect Size,) | DV: Coefficient (Effect Size,) |
|---|
| Moderating Effect of Issue Side on Argumentation Strategy | Argument Strength | Argument Strength |
| Financial argumentation strategy (time 1, 0 −1) | −.911*** (.001) | −.747*** (.040) |
| Industry side (time 1, 0−1) | –1.959*** (.138) | –2.566*** (.168) |
| Financial argumentation strategy × Industry side | 1.575*** (.075) | 2.573*** (.165) |
| Intercept | 3.842*** | 5.186*** |
| R2 | .232 | .322 |
| Model significance | F(10, 310) = 9.386*** | F(10, 322) = 15.355*** |
| Moderating Effect of Issue Side on Argumentation Strategy | Argument Suspicion | Argument Suspicion |
| Financial argumentation strategy (time 1, 0 −1) | .587*** (.001) | .538** (.004) |
| Industry side (time 1, 0−1) | 1.658*** (.169) | 2.054*** (.195) |
| Financial argumentation strategy × Industry side | −.982*** (.032) | –1.334*** (.054) |
| Intercept | 2.837*** | 2.287*** |
| R2 | .201 | .242 |
| Model significance | F(10, 310) = 7.819*** | F(10, 322) = 10.300*** |
| Effects of Argument Strength and Argument Suspicion | Attitudes | Voting | Attitudes | Voting |
| Financial argumentation strategy (time 1, 0−1) | .451*** (.041) | .503*** (.029) | .205 (.024) | .024 (.007) |
| Argument strength | .428*** (.008) | .602*** (.008) | .208** (.024) | .430*** (.017) |
| Argument suspicion | −.273*** (.010) | −.234** (.002) | −.271*** (.005) | −.068 (.001) |
| Industry side (time 1, 0 −1) | .546 (.015) | 1.180 (.003) | .675 (.008) | 2.215** (.012) |
| Argument strength × Industry side | −.817*** (.145) | –1.068*** (.149) | −.522***(.126) | −.833*** (.192) |
| Argument suspicion × Industry side | .648*** (.376) | .756*** (.358) | .383*** (.320) | .255* (.311) |
| Intercept | 3.579*** | 2.172** | 5.173*** | 3.057** |
| R2 | .477 | .460 | .480 | .497 |
| Model significance | F(13, 307) = 21.558*** | F(13, 307) = 20.119*** | F(13, 319) = 22.654*** | F(13, 319) = 24.249*** |
| H8a | Conditional indirect effect through argument strength, industry | −.258 [−.492, −.081] | −.309 [−.623, −.075] | −.573 [−.900, −.270] | −.735 [–1.144, −.365] |
| H8a | Conditional indirect effect through argument suspicion, industry | −.148 [−.326, −0010] | −.205 [−.461, −.010] | −.088 [−.238,.035] | −.148 [−.353, −.006] |
| H8b | Conditional indirect effect through argument strength, activist | −.390 [−.672, −.158] | −.549 [−.895, −.222] | −.155 [−.335, −.024] | −.321 [−.607,.097] |
| H8b | Conditional indirect effect through argument suspicion, activist | −.161 [−.344, −.021] | −.137 [−.350, −.004] | −.146 [−.319, −.017] | −.037 [−.177,.108] |
| Index of moderated mediation (argument strength) | .132 [−197,.473] | .239 [−.209,.669] | −.417 [−.775, −.086] | −.413 [−.891, 026] |
| Index of moderated mediation (argument suspicion) | .013 [−.213,.245] | −.068 [−.365,.226] | .057 [−.140,.263] | −.111 [−.365,.088] |
| Simple Effects of Issue Side on Argument Strength and Suspicion | Argument Strength | Argument Suspicion | Argument Strength | Argument Suspicion |
| H7a | Simple effects of financial argumentation for industry side | .664*** | −.394* | 1.826*** | −.796*** |
| H7b | Simple effects of financial argumentation for activist side | −.911*** | .587*** | −.747*** | .538** |
| H7c | Simple effects of industry side for financial argumentation | −.383 | .676 *** | .006 | .719*** |
| H7d | Simple effects of industry side for societal argumentation | –1.959 *** | 1.658 *** | –2.566*** | 2.054 *** |
15 Notes: Results from PROCESS Model 58, which includes need for orientation, political affiliation, age, gender (male), education, SES, and argumentation strategy at time 2 (0 −1) as covariates; for bottle deposits study, 10 participants from MA, where proposition was voted on, were dropped from analyses; for the energy standards study, 7 participants from AZ were dropped from analyses; 95% confidence intervals reported for conditional indirect effects; * p < .10, ** p < .05, *** p < .01.
In summary, across the three propositions, we obtained consistent support for the mediating role of argument strength and argument suspicion (H8a and H8b). Our results consistently show that societal argumentation is more effective for the activist side: it is more effective than financial argumentation for the activist side (H7b), and is perceived as stronger and less suspicious when used by the activist side than the industry side (H7d). For financial argumentation, results were somewhat mixed: as expected, financial argumentation was more effective than societal argumentation for the industry side (H7a). However, it was not perceived as significantly stronger and less suspicious when used by the industry versus the activist side (H7c). We summarize the results across all experimental studies in Web Appendix W12.
We contend that battles between industry and activist groups over consumer-related issues tend to be less partisan than other political scenarios such as candidate elections. For example, major early polls reported 77% of Democrats and 70% of Republicans supported Proposition 61, the drug price standards proposition examined in our field study and Experimental Study 1 ([52]; [63]).
To gain insight into the impact of partisanship in our context, we performed additional regression analyses to examine the extent to which political affiliation moderates the effects of side on the issue and argumentation strategy on argument strength and argument suspicion across the three issue contexts. Overall, our results indicate that only one of the six interaction effects between political affiliation and argumentation strategy was significant, but all six interaction effects of political affiliation and side on the issue were significant (see Web Appendix W13). The strongest Democrats perceived activist argumentation more favorably than industry argumentation, but there were no significant differences in how Republicans perceived the two sides' argumentation. These results suggest that political affiliation may play a role in how voters view the industry and activist sides. This is consistent with recent polls indicating that the majority of conservatives (liberals) have a positive (negative) view of big business ([55]) and is in line with current polarization along party lines in many domains in the United States ([ 8]; [57]).
We find support for our direct-to-public persuasion model across a field study and three experimental studies examining four policy conflict scenarios recently voted on in U.S. state ballot measures. Our findings show that industry and activist arguments play a key role in voting decisions, but industry arguments have less impact than activist arguments. Stronger arguments from both sides lead to more favorable outcomes, but activist groups benefit most. Similarly, industry argument suspicion has limited influence, except for voters who switch their support to the activist side. While societal argumentation is the preferred strategy for the activist side and financial argumentation is preferred for the industry side, the industry's competitive advantage is far less pronounced than expected.
Our exploration of direct-to-public persuasion increases the breadth of persuasion theory within the marketing domain. Persuasion campaigns in political settings—in which opinions are often strongly held, outcomes are win or lose, and consequences are societal rather than tied to individual consumers—are more complex in their effects than persuasion in commercial settings ([27]; [58]). As there is little theoretical grounding to build on, our research offers new ways of understanding this unique form of persuasion and adds to the scope of sociopolitical legitimacy theory, research on voting behavior, and perceptual fit theory.
Our findings related to an asymmetric public response in industry versus activist conflicts caution against assumptions that competing campaigns have an equal opportunity to persuade the public. By addressing unexplored differences in public response to competing campaigns, including legitimacy differences, our research updates persuasion theory by examining how and why arguments from competing sides affect voting behavior. Our results show that competition in politics is uneven, encompassing not just resource imbalances that may advantage one side, but also asymmetries across sides in the extent to which persuasion knowledge plays a role in voters' response (see [17]; [37]). Because the effects of dueling campaign arguments over time are not well known, yet are central to political outcomes, our findings highlight an important direction for future persuasion research.
Although theoretical guidance on the impact of argumentation on vote switching behavior is scarce, our results reveal an unexpected role of industry argument suspicion in that, although it has limited effect when voters are considered in aggregate, it is a key driver for the segment of voters who switch their support to the activist side. This finding is novel for persuasion research, as it indicates that the industry argument suspicion effects that drive a change in voter support over the course of a campaign differ somewhat from those driving overall voter support. This implies that response to the competing sides is not only asymmetric but nuanced and that research needs to consider various outcome measures to capture underlying factors.
Our finding that the industry side is most effective with a financial argumentation strategy while the activist side is most effective with societal argumentation offers new theoretical support regarding the need for fit between the persuader's identity and arguments. This complements the perceptual fit literature that has traditionally addressed the fit between a company's actions and claims ([64]) or a company's identity and cause-related marketing choices ([54]). We contribute to perceptual fit theory by finding that fit does not have equal importance for both sides, as an image-congruent argumentation strategy is vital for the industry, but voters may be more tolerant of a broader strategy for the activist side.
Our results offer practical guidance that differs for the industry versus activist sides related to the substance of their arguments and their goals of acquiring or retaining supporters (see Table 7). The findings are relevant for marketers and practitioners who design and implement direct-to-public campaigns with industry associations, corporations, public interest advocacy and activist groups, consultancies, and other policy-focused coalitions ([59]; [65]).
Graph
Table 7. Key Findings and Guidance for Industries and Activist Groups.
| Key Findings for Winning Public Support | Guidance for Industry Practice | Guidance for Activist Group Practice |
|---|
| Substance of Principle Arguments |
| Strong arguments benefit both sides, but activist argument strength has a greater impact than industry argument strength | Develop campaign arguments focused on diminishing the strength of activist arguments | Develop campaign arguments with ironclad strength and the potential to preempt industry arguments |
| Suspicion of activist arguments negatively affects outcomes, but industry argument suspicion has limited impact | Take an aggressive approach, raising skepticism of activist tactics, but avoid triggering a backlash | Take a defensive approach, preserving the legitimacy of activist group practices and arguments |
| Industry arguments with a financial focus are perceived as stronger and less suspicious; activist group arguments with a societal focus are stronger and less suspicious | Focus on a financial argumentation strategy based on factors related to imposing or reducing financial benefits or costs | Focus on a societal argumentation strategy based on (nonfinancial) factors related to inhibiting or promoting shared societal goals or values |
| Acquiring and Retaining Voter Support |
| Industry argument suspicion is only a key factor for voters who switch their support to the activist side | Focus on acquiring supporters by targeting the segment of voters whose initial support for the activist side is tentative | Emphasize retention of early supporters by reenforcing public perceptions that activist group legitimacy is higher |
| When both sides use financial argumentation, the industry side's competitive advantage over the activist side is limited | Build voter support with a narrow argumentation strategy focused on specific financial implications to voters | Retain voter support with a broader argumentation strategy focused on societal factors and counter industry claims with financial arguments |
While intuitive that both sides benefit from strong arguments, it is most critical for the activist side to be certain that voters will view its arguments as strong, as this is the effect with the highest impact on voting outcomes. This implies that it should be more effective for the industry to focus on diminishing the strength and increasing suspicion of activist arguments. As suspicion of industry arguments has limited impact, the industry has some license to encourage skepticism of activist tactics. This approach may require finesse, however, so as not to trigger a backlash if the issue is salient and the industry is controversial ([57]).
Preelection polls often indicate majority support for the activist side, suggesting that the public is initially drawn to a public interest perspective. When the activist side has this early advantage, it needs to retain initial supporters by ensuring that its arguments are strong and by utilizing counterarguments to protect its legitimacy. In contrast, the industry side must focus on acquiring voters and build support with arguments that highlight the hidden complexities and costs of the proposed policy change. As industries often succeed in acquiring supporters during a campaign, as voting nears, the activist side may need to shift its focus in order to cultivate a switching segment.
Our results show that the fit between an argumentation strategy and the identity of the persuader is key, especially for the industry side, which is safer emphasizing financial arguments closely aligned to its business sector profile. The activist side has some degrees of freedom as it is viewed positively by the public with either a financial or societal argumentation strategy. While societal argumentation is likely to have the most favorable impact for the activist side, financial argumentation may be necessary to challenge overstated cost analyses used by industry.
In summary, our results suggest the industry side does best to follow an aggressive approach of attacking activist-side tactics while using a narrow argumentation strategy. Conversely, the activist side does best by vigilantly preserving its legitimacy in reaction to any industry attacks, while using a broader argumentation strategy.
Given the consequential outcomes of the ballot measure venue we study and continued increases in funding ($1.24 billion in 2020; https://ballotpedia.org/2020%5fballot%5fmeasures), it is surprising that there is limited emphasis on the transparency of industry and organizational support, such as comprehensive information about donors and the extent of their financial support. Our findings indicate that response to campaign arguments may be driven by legitimacy perceptions of the opposing sides, suggesting that policy makers should facilitate increased information to voters. As the public is primarily informed and persuaded by advertising during political campaigns, including the ballot initiative process, advertising law changes might be the most effective approach to reform ([53]; [70]).
The Federal Election Commission regulates campaign finance law and political advertising, but its authority does not extend to ballot measure advertising, which is largely a state issue ([26]). California, under the jurisdiction of the California Fair Political Practices Commission (www.fppc.ca.gov), has among the strictest regulation for ballot measure advertising, following the logic that it is less clear who is responsible for these ads compared with candidate election ads. Because California already requires that top donors be explicitly listed in television, electronic media, and print advertisement as well as in mass mailings and robocalls, it can serve as a model for other states' ballot measure commissions to develop more comprehensive disclosure and increased transparency requirements. Alternatively, the Federal Election Commission could take steps to facilitate more unified federal standards for ballot measure advertisements. Without federal guidelines, consumers will continue to experience a patchwork approach to information that is unequal across states, making it difficult to assess the impact of direct-to-public persuasion on voters' decisions and policy outcomes.
Our investigation addresses one type of direct-to-public industry persuasion but not others, such as the more indirect route of shaping legislator decisions by influencing public opinion (for an examination of this issue, see [11]) or expensive direct-to-public industry image campaigns. In our field study, both ballot measures took place in one U.S. state and, thus, lack geographic diversity to mitigate potential unobserved influences. While our experimental studies address well-established argumentation strategies, they represent a partial picture of argumentation used to justify (or sanction) industry practice to the public.
There is an important need for more field studies, natural experiments, and other real-world persuasion research initiatives, as well as more research explaining asymmetric public response ([17]). We conducted a posttest to investigate our position that the public perceives that activist groups are more motivated than the industry side to serve the public interest (see Web Appendix W14). While our findings confirm this assertion, future experimental research should examine the extent to which this perception is a driver of voting outcomes. Future studies should explore whether our results indicating that societal argumentation strategies create substantially higher suspicion when used by the industry side could be attenuated for industries with a high proportion of companies with strong corporate social responsibility reputations. Although our findings across four consumer-relevant issues suggest that our hypotheses are likely to generalize across other high-salience issues, further research should investigate whether asymmetry may be muted in contests over lower salience issues.
We believe our findings will largely generalize to other contested issue scenarios, including situations where ( 1) the industry is challenging rather than defending the status quo and ( 2) industries battle other industries. When an industry tries to establish or expand a market for growth, it may challenge the policy status quo. A recent example includes the petroleum and gas industry using political clout to expand fracking onto previously protected Native American land in New Mexico ([47]). In these cases, the asymmetric impact of activist arguments having greater impact than industry arguments may be even more pronounced because the industry's self-interested motivation for market expansion may be more obvious to the public. Further, financial argumentation may actually backfire by magnifying the industry-serving objectives unless public attitudes have shifted over time to align with industry initiatives—for example, to promote casino development and marijuana legalization across the United States ([32]; [36]). In situations where industries battle other industries, when there is a legitimacy gap and one industry's public image is more favorable than the other's (e.g., renewable energy industries fighting fossil fuels industries over energy subsidies and standards; see [25]), arguments by the higher legitimacy industry could be more influential in driving outcomes. However, if there is little or no legitimacy gap (e.g., two agricultural industries battling over source of origin labeling laws), asymmetrical effects would not manifest.
Despite the fact that direct-to-public persuasion occurs across a variety of industries and scores of issues and has been prevalent for decades, further research is needed because the many dimensions that characterize this form of marketing may represent significant boundary conditions on the impact of argumentation on voting outcomes. Our research offers an important step forward in advancing knowledge of this underexamined area of marketing.
Footnotes 1 Jan-Benedict E.M. Steenkamp
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Online supplement:https://doi.org/10.1177/00222429211007517
5 Companies and industry groups can be involved in the activist side on the issue. However, because this represents a small proportion of the total financial support for that side's campaign, we focus on the situation in which industry and activist opponents face one another.
6 A 50% recontact rate is the response quota recommended by the panel provider based on a two-wave data collection design and an intercontact time of six weeks. This recontact rate is typical for this type of study ([18]).
7 We report details on the operationalization of all variables and results of a confirmatory factor analysis supporting the reliability and validity of multi-item scales in Web Appendix W2. Cronbach's alpha or Pearson correlation coefficient for each scale is greater than.75, and the average variance extracted exceeds the minimum recommended score of.50, supporting the reliability of each scale ([5]). The scales exhibit discriminant validity, as the squared correlation between each pair of constructs is less than the average variance extracted for each construct ([23]).
8 Likelihood of voting (%) calculated as exp(β/[1 + exp(β)].
9 We define effect size ( ) strength using [15], p. 283) thresholds, where.01 = small effect,.06 = medium effect, and.14 = large effect.
Although state and federal laws require formal documents disclosing the financial sponsors of media and advertising campaigns, the sponsors disclosed in the ads themselves are not explicitly identified as industry or activist groups but rather as "Yes on 61" and "No on 61." From extensive media coverage and voter information, we assume the public was aware that industry was a major financial supporter of the "No on 61" side.
Financial argumentation, industry: "Proposition 25 will hurt people in our state financially because it would increase the prices of other needed medicines, as drug companies make up for lost profits"; Financial argumentation, activist: "Proposition 25 will save some agencies money because it would restrict the amount they pay for selected prescription drugs"; Societal argumentation, industry: "Proposition 25 will reduce patients' access to needed medicines because it would create a new bureaucratic approval process"; Societal argumentation, activist: "Proposition 25 will save patients' lives because it would provide better access to life-saving drugs."
Across our studies, the conditional indirect effects of argumentation strategy on attitudes and voting via argument strength and suspicion have the same sign for the industry and activist sides, and as a result, most indices of moderated mediation are not significant. In our model, the effects of argumentation strategy on argument strength and argument suspicion depend on the side on the issue (crossover interaction) and the effects of argument strength and argument suspicion on attitudes and voting also depend on the side on the issue (crossover interaction). Because the overall pattern of path coefficients at each step of the mediation model is that the industry and activist sides are mirror images, the products of the path coefficients have the same sign, which results in the indices of moderated mediation not being significant (because these indices are calculated as products of the path coefficients). Although a change in argumentation strategy produces a similar effect on attitudes or voting for both sides, the process by which argumentation strategy impacts attitudes or voting is very different for the industry and activist sides.
In addition, we tested our model by pooling Experimental Studies 1–3 (n = 1,280) and including type of proposition as a covariate. Results provided even stronger support for our mediation model than results from the individual experiments. All conditional indirect effects of argumentation strategy through argument strength and argument suspicion were significant for both dependent variables, and the simple effect results were fully consistent with the individual experiments.
References American National Election Studies, University of Michigan, and Stanford University (2020), " ANES 2020 Time Series Study ," Inter-University Consortium for Political and Social Research (July 19, 2021) , https://electionstudies.org/data-center/2020-time-series-study/.
Amiram Dan , Kalay Alon , Sadka Gil. (2017), " Industry Characteristics, Risk Premiums, and Debt Pricing ," Accounting Review , 92 (1), 1 – 27.
Arceneaux Kevin. (2012), " Cognitive Biases and the Strength of Political Arguments ," American Journal of Political Science , 56 (2), 271 – 85.
Areni Charles S. , Lutz Richard J.. (1988), " The Role of Argument Quality in the Elaboration Likelihood Model ," in Advances in Consumer Research , Vol. 15 , Houston Michael J. , ed. Provo, UT : Association for Consumer Research , 197 – 202.
Bagozzi Richard P. , Yi Youjae. (1988), " On the Evaluation of Structural Equation Models ," Journal of the Academy of Marketing Science , 16 (1), 74 – 94.
Baumgartner Frank R. , Berry Jeffrey M. , Hojnacki Marie , Leech Beth L. , Kimball David C.. (2009), Lobbying and Policy Change: Who Wins, Who Loses, and Why. Chicago : University of Chicago Press.
Baur Dorothea , Palazzo Guido. (2011), " The Moral Legitimacy of NGOs as Partners of Corporations ," Business Ethics Quarterly , 21 (4), 579 – 604.
Berry Jeffrey M. , Wilcox Clyde. (2018), The Interest Group Society , 6th ed. New York : Routledge.
Bitektine Alex. (2011), " Toward a Theory of Social Judgments of Organizations: The Case of Legitimacy, Reputation, and Status ," Academy of Management Review , 36 (1), 151 – 79.
Boltanski Luc , Thévenot Laurent. (2006), On Justification: Economies of Worth. Princeton, NJ : Princeton University Press.
Burstein Paul. (2003), " The Impact of Public Opinion on Public Policy: A Review and an Agenda ," Political Research Quarterly , 56 (1), 29 – 40.
California Secretary of State (2016), Official Voter Information Guide. Sacramento : California Secretary of State Elections Division.
Campbell Margaret C. , Kirmani Amna. (2000), " Consumers' Use of Persuasion Knowledge: The Effects of Accessibility and Cognitive Capacity on Perceptions of an Influence Agent ," Journal of Consumer Research , 27 (1), 69 – 83.
Chong Dennis , Druckman James N.. (2010), " Dynamic Public Opinion: Communication Effects over Time ," American Political Science Review , 104 (4), 663 – 80.
Cohen Joshua. (1988), Statistical Power Analysis for the Behavioral Sciences. New York : Routledge Academic.
Darke Peter R. , Ritchie Robin B.. (2007), " The Defensive Consumer: Advertising Deception, Defensive Processing, and Distrust ," Journal of Marketing Research , 44 (1), 114 – 27.
Druckman James N. , Fein Jordan , Leeper Thomas J.. (2012), " A Source of Bias in Public Opinion Stability ," American Political Science Review , 106 (2), 430 – 54.
Druckman James N. , Peterson Erik , Slothuus Rune. (2013), " How Elite Partisan Polarization Affects Public Opinion Formation ," American Political Science Review , 107 (1), 57 – 79.
Drutman Lee. (2015), The Business of America Is Lobbying: How Corporations Became Politicized and Politics Became More Corporate. Oxford, UK : Oxford University Press.
Dunn William N. (2012), Public Policy Analysis , 5th ed. London : Routledge.
Eesley Charles , DeCelles Katherine A. , Lenox Michael. (2016), " Through the Mud or in the Boardroom: Examining Activist Types and Their Strategies in Targeting Firms for Social Change ," Strategic Management Journal , 37 (12), 2425 – 440.
Forehand Mark R. , Grier Sonya. (2003), " When Is Honesty the Best Policy? The Effect of Stated Company Intent on Consumer Skepticism ," Journal of Consumer Psychology , 13 (3), 349 – 56.
Fornell Claes , Larcker David F.. (1981), " Evaluating Structural Equation Models with Unobservable Variables and Measurement Error ," Journal of Marketing Research , 18 (1), 39 – 50.
Fransen Marieke L. , Smit Edith G. , Verlegh Peeter W.J.. (2015), " Strategies and Motives for Resistance to Persuasion: An Integrative Framework ," Frontiers in Psychology , 6 , 1201.
Gallup (2019), " Confidence in Institutions " (accessed February 15, 2021), https://news.gallup.com/poll/1597/confidence-institutions.aspx.
Glazer Emily , Haggin Patience. (2019), " Political Ads Are Flourishing Online. Few Agree How to Regulate Them ," The Wall Street Journal (November 15), https://www.wsj.com/articles/as-political-ad-spending-balloons-online-consensus-on-regulation-is-elusive-11573813803.
Gordon Brett R. , Lovett Mitchell J. , Shachar Ron , Arceneaux Kevin , Moorthy Sridhar , Peress Michael , et al. (2012), " Marketing and Politics: Models, Behavior, and Policy Implications ," Marketing Letters , 23 (2), 391 – 403.
Greene William H. (2003), Econometric Analysis , 5th ed. Upper Saddle River, NJ : Prentice Hall.
Hayes Andrew F. (2018), Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression Based Approach , 2nd ed. New York : The Guilford Press.
Heckman James J. (1979), " Sample Selection Bias as a Specification Error ," Econometrica , 47 (1), 153 – 61.
Hiatt Shon R. , Grandy Jake B. , Lee Brandon H.. (2015), " Organizational Responses to Public and Private Politics: An Analysis of Climate Change Activists and U.S. Oil and Gas Firms ," Organization Science , 26 (6), 1769 – 86.
Humphreys Ashlee. (2010), " Megamarketing: The Creation of Markets as a Social Process ," Journal of Marketing , 74 (2), 1 – 19.
Hussain Suhauna. (2020), " Uber, Lyft Push Prop. 22 Message Where You Can't Escape It: Your Phone ," Los Angeles Times (October 8), https://www.aol.com/news/uber-lyft-push-prop-22-120027093.html.
Isaac Mathew S. , Grayson Kent. (2017), " Beyond Skepticism: Can Accessing Persuasion Knowledge Bolster Credibility? " Journal of Consumer Research , 43 (6), 895 – 912.
Jung Minah H. , Critcher Clayton R.. (2018), " How Encouraging Niceness Can Incentivize Nastiness: An Unintended Consequence of Advertising Reform ," Journal of Marketing Research , 55 (1), 147 – 61.
Kees Jeremy , Fitzgerald Paula , Dorsey Joshua D. , Hill Ronald Paul. (2020), " Evidence-Based Cannabis Policy: A Framework to Guide Marketing and Public Policy Research ," Journal of Public Policy & Marketing , 39 (1), 76 – 92.
Kirmani Amna , Campbell Margaret C.. (2009), " Taking the Target's Perspective: The Persuasion Knowledge Model ," Social Psychology of Consumer Behavior , 297 – 316.
Klein Jill G. , Ahluwalia Rohini. (2005), " Negativity in the Evaluation of Political Candidates ," Journal of Marketing , 69 (1), 131 – 42.
Klein Jill Gabrielle , Craig Smith N. , John Andrew. (2004), " Why We Boycott: Consumer Motivations for Boycott Participation ," Journal of Marketing , 68 (3), 92 – 109.
Lamin Anna , Zaheer Srilata. (2012), " Wall Street vs. Main Street: Firm Strategies for Defending Legitimacy and Their Impact on Different Stakeholders ," Organization Science , 23 (1), 47 – 66.
Lindell Michael K. , Whitney David J.. (2001), " Accounting for Common Method Variance in Cross-Sectional Research Designs ," Journal of Applied Psychology , 86 (1), 114 – 21.
Maisel L. Sandy , Berry Jeffrey M.. (2010), " The State of Research on Political Parties and Interest Groups ," in The Oxford Handbook of American Political Parties and Interest Groups , Sandy Maisel L. , Berry Jeffrey M. , eds. Oxford, UK : Oxford University Press , 3 – 18.
Martin Kelly D. , Josephson Brett W. , Vadakkepatt Gautham G. , Johnson Jean L.. (2018), " Political Management, Research and Development, and Advertising Capital in the Pharmaceutical Industry: A Good Prognosis? " Journal of Marketing , 82 (3), 87 – 107.
McFadden Brandon R. , Lusk Jayson L.. (2013), " Effects of Cost and Campaign Advertising on Support for California's Proposition 37 ," Journal of Agricultural and Resource Economics , 38 (2), 174 – 86.
McKinsey and Company (2021), " The Next Normal Arrives: Trends that will Define 2021 – and Beyond ," in McKinsey Global Publishing. New York : McKinsey & Co.
Meyers-Levy Joan , Malaviya Prashant. (1999), " Consumers' Processing of Persuasive Advertisements: An Integrative Framework of Persuasion Theories ," Journal of Marketing , 63 (4_suppl1), 45 – 60.
Nelson Cody. (2020), " 'Their Greed Is Gonna Kill Us': Indian Country Fights Against More Fracking ," The Guardian (June 10), https://www.theguardian.com/us-news/2020/jun/10/new-mexico-fracking-navajo-indian-country.
Obermiller Carl , Spangenberg Eric , MacLachlan Douglas L.. (2005), " Ad Skepticism: The Consequences of Disbelief ," Journal of Advertising , 34 (3), 7 – 17.
Pew Research Center (2019), " Views of the Economic System and Social Safety Net ," in American Trends Panel Wave 53. Washington DC : Pew Research Center.
Phillips Joan M. , Urbany Joel E. , Reynolds Thomas J.. (2008), " Confirmation and the Effects of Valenced Political Advertising: A Field Experiment ," Journal of Consumer Research , 34 (6), 794 – 806.
Podsakoff Philip M. , MacKenzie Scott B. , Lee Jeong-Yeon , Podsakoff Nathan P.. (2003), " Common Method Biases in Behavioral Research: A Critical Review of the Literature and Recommended Remedies ," Journal of Applied Psychology , 88 (5), 879 – 903.
Public Policy Institute of California (2016), PPIC Statewide Survey: Californians and Their Government. San Francisco : Public Policy Institute of California.
Richards Timothy , Allender William , Fang Di. (2013), " Media Advertising and Ballot Initiatives: The Case of Animal Welfare Regulation ," Contemporary Economic Policy , 31 (1), 145 – 62.
Robinson Stefanie Rosen , Irmak Caglar , Jayachandran Satish. (2012), " Choice of Cause in Cause-Related Marketing ," Journal of Marketing , 76 (4), 126 – 39.
Saad Lydia. (2019), " Do Americans Like or Dislike 'Big Business'? " Gallup (accessed December 26, 2019) , https://news.gallup.com/poll/270296/americans-dislike-big-business.aspx?.
Schlozman Kay L. , Verba Sidney , Brady Henry E.. (2012), The Unheavenly Chorus: Unequal Political Voice Political Voice and the Broken Promise of American Democracy. Princeton, NJ : Princeton University Press.
Scott John C. (2018), Lobbying and Society: A Political Sociology of Interest Groups. Cambridge, UK : Polity Press.
Shachar Ron. (2009), " The Political Participation Puzzle and Marketing ," Journal of Marketing Research , 46 (6), 798 – 815.
Sheingate Adam. (2016), Building a Business of Politics: The Rise of Political Consulting. Oxford, UK : Oxford University Press.
Simmons Carolyn J. , Becker-Olsen Karen L.. (2006), " Achieving Marketing Objectives Through Social Sponsorships ," Journal of Marketing , 70 (4), 154 – 69.
Suddaby Roy , Bitektine Alex , Haack Patrick. (2017), " Legitimacy ," Academy of Management Annals , 11 (1), 451 – 78.
Trumbull Gunnar. (2012), Strength in Numbers: The Political Power of Weak Interests. Cambridge, MA : Harvard University Press.
Tulchin Research and Strategic Consulting (2016), " California Statewide Survey Shows Strong Support for Proposition 61 ," Tulchin Research Poll Results (July 28) , https://tulchinresearch.com/wp-content/uploads/2016/08/California-RX-Drug-Pricing-362-A-PUBLIC-MEMO-7-16-Final.pdf.
Wagner Tillmann , Lutz Richard J. , Weitz Barton A.. (2009), " Corporate Hypocrisy: Overcoming the Threat of Inconsistent Corporate Social Responsibility Perceptions ," Journal of Marketing , 73 (6), 77 – 91.
Walker Edward T. (2014), Grassroots for Hire: Public Affairs Consultants in American Democracy. Cambridge, UK : Cambridge University Press.
Wan Echo Wen , Rucker Derek D. , Tormala Zakary L. , Clarkson Joshua J.. (2010), " The Effect of Regulatory Depletion on Attitude Certainty ," Journal of Marketing Research , 47 (3), 531 – 41.
Waterhouse Benjamin C. (2014), Lobbying America. Princeton NJ : Princeton University Press.
Werner Timothy. (2015), " Gaining Access by Doing Good: The Effect of Sociopolitical Reputation on Firm Participation in Policy Making ," Management Science , 61 (8), 1989 – 2011.
Wonnacott Thomas H. , Wonnacott Ronald J.. (1990), Introductory Statistics , Vol. 5. New York : John Wiley & Sons.
Xu Alison Jing , Moorman Christine , Qin Vivian Yue , Rao Akshay. (2020), " Four More Years: Presidential Elections, Comparative Mind-Set, and Managerial Decisions ," Academy of Management Journal , 63 (5), 1370 – 94.
Zhao Xiaoquan , Strasser Andrew , Cappella Joseph N. , Lerman Caryn , Fishbein Martin. (2011), " A Measure of Perceived Argument Strength: Reliability and Validity ," Communication Methods and Measures , 5 (1), 48 – 75.
~~~~~~~~
By Kathleen Seiders; Andrea Godfrey Flynn and Gergana Y. Nenkov
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 67- How Marketing Perks Influence Word of Mouth. By: Lisjak, Monika; Bonezzi, Andrea; Rucker, Derek D. Journal of Marketing. Sep2021, Vol. 85 Issue 5, p128-144. 17p. 2 Charts, 1 Graph. DOI: 10.1177/0022242921991798.
- Database:
- Business Source Complete
How Marketing Perks Influence Word of Mouth
This research illustrates how marketing perks can be leveraged to spur WOM. Specifically, this research introduces a previously overlooked yet practically relevant dimension on which perks differ: "contractuality," defined as the extent to which consumers perceive a perk to be conditional on specific behaviors and contingencies dictated by a company. Importantly, consumers can perceive the exact same perk as more versus less contractual depending on the way it is conferred, structured, or framed. This research demonstrates that low-contractuality perks can be more effective than high-contractuality perks at fostering WOM in the absence of explicit incentives to do so. In particular, low-contractuality perks are more likely than high-contractuality perks to convey a relational signal that motivates consumers to help the company by sharing WOM. Seven experiments, two of which were conducted in the field, support this hypothesis and illustrate conditions under which the effect attenuates or reverses. On the whole, this work offers insights into how managers might design interventions that foster WOM, and it reveals potential trade-offs of commonly used high-contractuality perks.
Keywords: contractuality; marketing perks; relational value; word of mouth; consumer–brand relationship
Word of mouth (WOM) is arguably the most influential means of persuasion and is one of the main predictors of a company's growth ([59]). Studies by Nielsen consistently find that WOM is the single most trusted source of information for consumers ([53]), and other reports suggest that between 50% and 74% of consumers act in response to others' recommendations ([14]; [39]). Indeed, WOM has been estimated to drive as much as $6 trillion in consumer spending per year ([ 4]), and 64% of marketers consider it the most effective form of advertising ([74]).
Given its persuasiveness, marketers invest in campaigns to stimulate WOM. These campaigns typically take the form of referrals ([62]) or seeding programs ([10]) and involve giving consumers an incentive—a reward or a direct payment—to generate WOM. Such "incentivized" WOM ([77]) can increase sales through market expansion and purchase acceleration ([29]; [48]). However, incentivized WOM can also have drawbacks. Incentives can reduce the perceived sincerity of the sender ([71]; [73]), undermining the persuasiveness of the message ([ 5]). Incentives can also cause consumers to worry about being perceived as having ulterior motives ([38]), hampering their willingness to engage in WOM ([31]).
Because incentivizing WOM has drawbacks, marketers often seek to fuel WOM without explicit incentives ([63]; [70]). In this work, we propose that tactics commonly used by companies to serve objectives other than WOM can be tailored to foster WOM without resorting to explicit incentives. Specifically, we focus on the use of "perks," a term we use to refer to gifts or benefits given to customers that range from free items ([44]) to preferential treatment ([57]) to financial benefits ([49]). For example, Starbucks offers consumers free items once they spend a certain amount of money. Peet's Coffee gives consumers a free item in recognition of their birthday. We argue that, although such perks provide no direct incentive for consumers to share WOM, under the proper circumstances, they can increase consumers' likelihood to engage in WOM.
We focus on an overlooked but practically relevant dimension that perks differ on: "contractuality." We suggest that perks can be perceived as more versus less contractual depending on whether they are contingent on consumers performing specific behaviors dictated by a company. We further suggest that contractuality carries a relational signal that influences consumers' desire to support the company, which in turn affects consumers' likelihood to engage in WOM in the absence of explicit incentives to do so. We also identify conditions that moderate the influence of contractuality on WOM and can even lead to such offerings backfiring.
Our work contributes to both theory and practice by providing insights into how companies can leverage existing perks to promote nonincentivized WOM. In doing so, this research contributes to the WOM literature ([ 7]; [21]) by documenting a novel and practically actionable antecedent of WOM: the level of contractuality of a perk. Moreover, it contributes to the marketing perks literature ([49]; [57]) by examining when and why perks foster WOM. Specifically, it suggests that perks are more likely to fuel WOM when they carry a relational signal that motivates consumers to support the company. Thus, this research lays the foundation for practitioners to understand potential trade-offs of using high- versus low-contractuality perks.
We propose that perks differ in terms of perceived contractuality. We use the term "contractuality" to refer to the degree to which consumers perceive a perk to be given to them in exchange for engaging in specific behaviors dictated by a company, such as filling out a survey, making a certain number of purchases, or shopping within a specified time period. We argue that consumers can perceive the exact same perk (e.g., a free coffee) as more versus less contractual depending on the way it is conferred, structured, or framed. A perk should be perceived as high in contractuality when it is conferred, structured, or framed in a way that explicitly makes it conditional on contingencies dictated by the company. In contrast, a perk should be perceived as lower in contractuality when it is conferred, structured, or framed in a way that makes it less explicitly conditional on specific contingencies established by the company.
To illustrate, consider these examples. Dunkin' and Subway have required consumers to spend a set sum of money or make a certain number of purchases to earn free items. This framing is higher in contractuality, as the consumer is explicitly required to perform an action to receive the perk. In contrast, Chick-fil-A and Panera have awarded consumers free food or beverages after repeated purchases, but without any explicit contract as to when these perks would be awarded. This framing is lower in contractuality, as the consumer is not explicitly required to perform an action to earn the perk. Rather, the company observes the consumer and chooses when to provide the perk. As these examples illustrate, a perk can be identical in its cost to the company, provide the same benefit to the consumer (e.g., a free meal), and require the same amount of consumer effort (e.g., number of visits) but still vary in contractuality depending on how the company awards it.
As another example, consider gifts such as birthday freebies. Some companies (e.g., Starbucks, Au Bon Pain) have required consumers to come in on the day of their birthday or spend a minimum amount of money to obtain the freebie, resulting in offerings that have a high degree of contractuality. In contrast, other companies (e.g., Pete's Coffee, Sephora) have given consumers a free item for their birthday with limited or no contingencies—for example, it can be used even after one's birthday—resulting in offerings that are lower in contractuality. Again, although such perks might have the same value (e.g., a free coffee) and require the same effort from the consumer, their perceived contractuality differs as a function of the way the company structures them. Table 1 provides additional examples of how perks used by companies across a variety of industries and markets can differ in terms of contractuality depending on the way they are conferred, structured, or framed. A survey confirmed that consumers perceived these perks to vary in their contractuality (see Web Appendix A).
Graph
Table 1. Examples of Brands That Have Used the Same Perks in Ways That Differed in Terms of Perceived Contractuality.
| Low Contractuality | High Contractuality |
|---|
| PerkDescription | CompanyExamples | PerkDescription | CompanyExamples |
|---|
| Frequent purchaseperks | Perks are awarded at the discretion of a company, without providing consumers with clear rules (e.g., surprise and delight reward programs). | Chick-fil-AHyattMastercard | Perks are awarded in exchange for specific behaviors according to clear rules known to consumers (e.g., point-based reward programs). | American AirlinesHotels.comMarriott Bonvoy |
| Birthday perks | Gifts do not require a minimum purchase and can be redeemed within a large time window, or there is no time restriction at all. | DSWPete'sSephora | Gifts require a minimum purchase or can be redeemed only within a restricted time window. | Au Bon PainDairy QueenStarbucks |
| Perks for new moms | Perk box that only requires signing up for the baby registry or other free programs. | BuyBuy BabyEnfamilTarget | Perk box that requires signing up for a monthly subscription or a subscription and a minimum purchase. | AmazonHonestWall-Mart |
| Back-to-school perks | Gifts given when consumers provide a valid school email address or create a free account online. | Back2SchoolMicrosoftTiconderoga | Gifts given with a minimum purchase or subscription. | AmazonApple |
In short, we argue that perks vary along a contractuality continuum, and consumers can perceive perks as more versus less contractual depending on the way a company confers, structures, or frames them. Importantly, as we will show, contractuality is orthogonal to other dimensions on which perks might differ, such as their expectedness (e.g., [72]), their value (e.g., [20]), or the effort needed to obtain them (e.g., [40]).
How does a perk's perceived contractuality influence WOM? Intuitively, one might think that any perk, irrespective of its perceived contractuality, should increase WOM, possibly by inducing positive mood ([36]) or triggering physiological arousal ([ 9]). For example, to the extent a perk creates excitement, this excitatory state should boost WOM ([ 7]). Alternatively, one could argue that perks high in contractuality might spur greater WOM because they are characterized by clear behavior–reward contingencies ([15]).
We suggest that perks lower in contractuality foster greater WOM than perks higher in contractuality. We propose that this result occurs because, compared with perks higher in contractuality, perks lower in contractuality carry a relational signal that motivates consumers to support the company. More specifically, our conceptual explanation builds on three key premises: ( 1) contractuality influences perceived relational value, ( 2) perceived relational value influences people's desire to help a company, and ( 3) the desire to help a company can drive WOM.
As to the first premise, research suggests that people have a deeply rooted need to feel valued and appreciated ([46]). [45] coined the term "relational value" to refer to this fundamental psychological experience and posited that it is the keystone of relationship formation and maintenance. Prior literature suggests that people experience heightened relational value when others engage in intentional acts that signal benevolence toward them ([13]; [47]) and are devoid of ulterior motives ([64]; [66]). To illustrate, in dyadic relationships, people feel heightened relational value when their partners express benevolence by supporting them ([13]). In organizational settings, employees feel heightened relational value when employers engage in benevolent behaviors such as offering family-supportive work practices ([43]). In marketing, we propose that a perk will heighten relational value when it is characterized by lower contractuality. A perk that is lower in contractuality has no explicit strings attached, and consumers should thus be more likely to perceive it as an act of benevolence compared with a perk higher in contractuality.
As to the second premise, several streams of research suggest that experiencing high relational value induces people to engage in behaviors aimed to support and benefit others. According to sociometer theory ([46]), individuals have a sociometer that monitors the social environment for cues about their relational value (i.e., signals that people appreciate a relationship with an individual). Signals that one's relational value is high motivate people to engage in behaviors that benefit those who express appreciation ([19]; [61]). People who perceive others to value them experience a sense of security that triggers a focus on supporting those others ([13]; [52]). For example, organizational support theory suggests that feeling valued and appreciated is a crucial determinant of employees' commitment and loyalty to an organization ([26]; [65]). The more employees feel valued, the more they help and support their companies (for a meta-analysis see [43]).
As to the third premise, prior research has identified a plethora of psychological motives that drive sharing (for a review, see [ 8]), including the desire to self-enhance ([21]), restore control ([56]), signal expertise ([54]), and foster social connections ([16]; [25]). Key to our research, prior literature suggests that sharing may also arise from an altruistic desire to help the company ([33]; [68]). If experiencing relational value triggers a desire to engage in behaviors aimed to support and benefit a company, consumers may act on this motive by engaging in positive WOM when given a chance to do so.
In summary, we propose that perks associated with lower contractuality heighten relational value, which increases consumers' propensity to share WOM about the company in the absence of any explicit incentive to do so. In contrast, perks associated with higher contractuality are less likely to heighten relational value and thus are less likely to foster WOM. Stated formally, we propose a relational value account with the following hypotheses:
- H1 : Perks lower (vs. higher) in contractuality increase WOM.
- H2 : Perks lower (vs. higher) in contractuality signal higher relational value. Higher relational value, in turn, heightens consumers' motivation to help the company, which explains the increase in WOM predicted in H1.
While the effect of contractuality on WOM might be multiply determined, we aim to conceptually and empirically differentiate our relational value account from two alternative accounts that might produce similar effects: reciprocity and reactance. Of note, the goal in this research is not to argue that relational value is the only mechanism at play, but rather to evaluate whether relational value plays a unique role above and beyond other processes.
First, our relational value account is distinct from a reciprocity account. Reciprocity refers to a social norm whereby people ought to repay a favor in kind ([17]). Reciprocity operates via feelings of indebtedness and obligation ([58]), which people are motivated to rid themselves of by returning the favor ([30]; [51]). In contrast, our relational value account posits that low-contractuality perks foster WOM via a sense of being valued and appreciated that does not trigger feelings of indebtedness and obligation. We empirically test the role of reciprocity in Experiments 2–5 and a follow-up study.
Second, our relational value account also differs from a reactance account. According to reactance theory ([11]), when people feel that their behavioral freedom has been restricted or threatened, they experience an unpleasant state of arousal that can lead them to engage in behaviors aimed to re-establish their freedom ([27]; [75]). Because high-contractuality perks are conferred in a way that makes contingencies salient, consumers may perceive them as a threat to their personal freedom and thus experience reactance. Consequently, people might be motivated to restore their threatened freedom by engaging in behaviors aimed to punish the company, such as refraining from spreading positive WOM. We empirically test the role of reactance in Experiments 2–5.
The benefit of low-contractuality perks on WOM hinges on consumers interpreting a perk that is low in contractuality as a signal of relational value (i.e., a signal that the company values them). Factors that undermine this relational signal should attenuate or reverse the effect of low contractuality on WOM. Specifically, perks low in contractuality should no longer signal relational value if consumers attribute them to ulterior motives ([12]; [50]).
- H3 : The positive effect of lower (vs. higher) contractuality on WOM is attenuated or reversed when ulterior motives are salient.
We investigate two distinct circumstances that should increase the salience of ulterior motives. First, consumers may attribute a low-contractuality perk to ulterior motives when a company provides it before rather than after purchase ([28]; [35]). For example, [12] found that participants who received a flattering comment from a salesperson before purchase, as opposed to after, viewed the salesperson as less sincere. Relatedly, when consumers receive a low-contractuality perk before rather than after purchase, relational value may be undermined, thereby attenuating the effectiveness of the offering at fostering WOM.
Second, consumers may attribute lower contractuality perks to ulterior motives when the offering comes from a disliked brand. Research on the sinister attribution error ([42]) suggests that people attribute nice gestures to ill intent when they come from others who are disliked or distrusted. For example, [50] showed that in a context characterized by distrust, receiving a flattering comment reduced trust toward the person making the comment. Thus, when a perk comes from a disliked or distrusted company, low-contractuality perks might raise suspicion of ulterior motives more than high-contractuality perks. That is, whereas high-contractuality perks imply a contract and thus make the motive for the company's behavior clear, this is not the case for low-contractuality perks. People who dislike a company may assume the worst and conclude that the company is being manipulative. As such, consumers' dislike toward a company may lead them to interpret low-contractuality perks as being driven by ulterior motives, yielding a negative effect on WOM.
Finally, low-contractuality perks might involve a trade-off, in that they might foster less compliance with a direct request. For example, if a company wants consumers to complete a customer satisfaction survey, offering a high-contractuality perk might be more effective and efficient than offering a low-contractuality perk. Simply put, perks higher in contractuality might be better incentives for consumers to engage in a specific behavior than perks lower in contractuality because they make behavior–reward contingencies clear and salient ([67]; [69]). However, as we propose, perks lower in contractuality might be more effective than perks higher in contractuality at fostering WOM. Thus, from a managerial standpoint, it is important to recognize that a trade-off may exist between incentivizing first order responses and nudging second-order responses such as engaging in WOM. Formally:
- H4 : Although low-contractuality perks might be more effective at fostering WOM, they might be less effective than high-contractuality perks at stimulating compliance with a request for a specific behavior.
We conducted seven experiments (six main and one follow-up) to test our hypotheses. Experiment 1, a field study, tests whether awarding a perk in a less contractual way increases WOM. Experiments 2 and 3 test the proposed relational value account via mediation by using different operationalizations of contractuality and WOM. Experiments 4 and 5 provide evidence for the relational value account via moderation. Finally, Experiment 6, a second field study, tests the idea that low-contractuality perks can be more effective at fostering WOM yet less effective at stimulating compliance with a request for a desired behavior.
We analyzed data only after all responses were collected. We report all conditions, manipulations, data exclusions, and measures related to our hypotheses. Unless reported, no participant was excluded. In the field studies (Experiments 1 and 6), we targeted a number of approximately 50–100 responses per cell on the basis of funds and participants' availability. In the MTurk studies (Experiments 2–5), the targeted number of responses per cell was 100–200.
Experiment 1 tested whether the perceived contractuality of a perk influences consumers' likelihood to share WOM (H1). To test this hypothesis, we conducted a field experiment in a mom-and-pop bakery. We picked this bakery because it has a strong social media presence and is described by bloggers as being an "Insta-worthy" spot because of its aesthetically pleasing interiors and unique desserts. After completing a survey, customers of the bakery received a free macaroon awarded in a more or less contractual way. Then, they were encouraged to engage with the bakery by taking pictures to post on social media. We predicted that customers who received the macaroon in a less contractual manner would be more likely to engage in WOM.
We worked in conjunction with the owners of the bakery to conduct this field study. A research assistant posing as an employee of the bakery approached 101 customers (80 female). The assistant asked them to complete a short survey on an iPad to help the bakery better understand their preferences. Customers were asked to indicate their favorite macaroon flavor from the bakery assortment and suggest new flavors the bakery could offer. They were also asked to indicate whether they currently followed the bakery on any social media platform and to provide their Instagram account handle.
Upon completing the survey, customers returned the iPad to the research assistant. At this point, the research assistant pulled out a basket filled with same-flavored macaroons and randomly administered one of two treatments. In one condition, the research assistant invited customers to pick a macaroon, explaining that the bakery had committed to give a free macaroon as a gift in exchange for filling out the survey. Thus, in this condition, the contractual nature of the perk was salient (i.e., high contractuality). In the other condition, the research assistant invited customers to pick a macaroon, explaining that it was a gift from the bakery. Thus, in this condition, the perk was not portrayed as being given out of contractual obligation (i.e., low contractuality). Note that customers in both conditions received an identical gift that was equally unexpected, thus any effect observed is unlikely to stem from differences in the perceived value of the perk or in how surprising the perk was.
Finally, the research assistant encouraged customers to engage with the bakery by taking a photo and posting it on social media. The research assistant unobtrusively recorded whether or not customers took a picture.
A study conducted with 100 participants recruited from Amazon Mechanical Turk (MTurk) (41 female, mean age = 37.16 years, SD = 12.27) confirmed that the low-contractuality perk was perceived as lower in contractuality (M = 4.63, SD = 2.04) than the high-contractuality perk (M = 6.31, SD =.90, F( 1, 98) = 28.89, p <.001) on a three-item scale that measured the extent to which participants perceived that the perk was given to them ( 1) conditionally on doing something specific, ( 2) in exchange for doing something specific, and ( 3) contingently on engaging in a particular behavior specified by the bakery (1 = "not at all," and 7 = "very much;" α =.93).
We ran a binary logistic regression in which we regressed whether customers took photos to post on social media (coded as 1 = "yes" and 0 = "no") onto contractuality (coded as 1 = "low in contractuality" and 0 = "high in contractuality"), controlling for whether they already followed the bakery on social media prior to the intervention (coded as 1 = "yes" and 0 = "no"). The analysis revealed a marginal effect of following the company prior to the intervention (exp(b) =.39, Wald ( 1, N = 101) = 3.82, p =.051) and a significant effect of contractuality (exp(b) = 2.38, Wald ( 1, N = 101) = 4.24, p =.039). With regard to contractuality, 62.7% of customers in the low contractuality condition took a picture after receiving the free macaroon, whereas only 44% of customers in the high contractuality condition did so.
As an ancillary measure, a week after the intervention, we tracked the Instagram account handles customers provided in the survey to test whether they had started following the bakery on Instagram. A total of 19.6% of customers in the low contractuality condition started following the bakery after the intervention, whereas only 8% of customers in the high contractuality condition did so (exp(b) = 3.37, Wald ( 1, N = 101) = 3.53, p =.060). No other effects were significant.
These results suggest that a perk's perceived contractuality influences consumers' propensity to share WOM (H1). Visitors of the bakery who received a free macaroon were more likely to take pictures to post on social media after receiving a perk awarded in a less rather than more contractual way. We observed these results even though customers received the exact same perk (i.e., a free macaroon) and the perk was equally unexpected in both conditions.
Experiment 2 sought to provide convergent evidence for the effect via a different operationalization of contractuality. In addition, we tested whether the effect of contractuality on WOM is serially mediated by consumers' perceptions of relational value and motivation to help the company (H2). Finally, we tested our theoretical account vis-à-vis reciprocity and reactance. To explore reciprocity as a mechanism, we measured how indebted and obligated people felt after receiving a perk that varied in contractuality. To explore reactance as a mechanism, we measured whether people felt forced or limited in their freedom. We also included additional measures related to surprise, effort, and perceived value of the perk.
Participants (n = 398, 170 female, mean age = 37.19 years, SD = 13.01) recruited via MTurk completed an online survey in exchange for monetary compensation. Only participants who passed an initial attention check were eligible to participate. Participants were presented with a study exploring common consumption experiences. They read a scenario that involved ordering food from a restaurant. Then, they were randomly assigned to two experimental conditions in which we manipulated the perceived contractuality of a perk.
In the high contractuality condition, participants were told that their order was delivered and were given a note informing them that the restaurant included a $15 bonus gift card that they could redeem in the next three days between 10 a.m. and noon. In the low contractuality condition, participants received the same information, except the bonus gift card had no redemption limitations (see Web Appendix B for all stimuli). We expected this manipulation to affect perceived contractuality because the redemption requirements for the high-contractuality gift card were more specific and restrictive than those for the low-contractuality gift card.
To measure WOM, we adopted a procedure used in prior research ([ 6]). Participants were asked, if they were on social media later that day, whether they would share any comments about their experience with the restaurant (1 = "yes," and 0 = "no"). Participants who indicated they would write a comment were invited to actually write the comment as if they were posting it on social media.
We measured relational value by asking participants to indicate to what extent they felt the restaurant ( 1) valued them and ( 2) appreciated them on a seven-point scale (1 = "not at all," and 7 = "very much;" r =.88). We measured motivation to help the company by asking participants to indicate the extent to which they would feel motivated to ( 1) benefit the restaurant and ( 2) support the restaurant (1 = "not at all," and 7 = "very much;" r =.81).
To assess reciprocity, we asked participants if they felt ( 1) indebted and ( 2) obligated toward the restaurant (1 = "not at all," and 7 = "very much;" r =.81). To assess reactance, we asked them if they felt that ( 1) their freedom was limited by the restaurant and ( 2) the restaurant forced them to do something (1 = "not at all," and 7 = "very much;" r =.88).
To assess surprise, we asked participants to report how ( 1) unexpected and ( 2) unpredictable they thought it was to receive the free gift card, as well as how ( 3) surprised and ( 4) amazed they would feel upon receiving the free gift card (1 = "not at all," and 7 = "very much;" α =.85). To measure perceived effort, participants reported whether they felt that getting the free gift card was ( 1) effortful, ( 2) difficult, and ( 3) hard (1 = "not at all," and 7 = "very much;" α =.89). To measure the perceived monetary value of the perk, participants indicated how valuable they thought the free gift card was (1 = "not high at all," and 7 = "very high").
We measured contractuality using the same three items described in Experiment 1 (α =.93). To conclude, participants reported their demographics.
A one-way analysis of variance (ANOVA) revealed that participants in the low contractuality condition perceived the free gift card to be given in a less contractual way (M = 3.35, SD = 1.95) than those in the high contractuality condition (M = 4.87, SD = 1.54; F( 1, 396) = 74.24, p <.001, ηp2 =.158).
A chi-square analysis on likelihood to generate WOM revealed a significant effect of contractuality (χ2 ( 1, N = 398) = 17.17, p <.001, Cramer's V =.21): 71% of participants in the low contractuality condition were willing to share WOM about the restaurant, relative to 51% in the high contractuality condition (see Table 2). Moreover, of participants who indicated they would share a comment on social media, all but four participants wrote a complete comment. Excluding these four participants from the analysis does not change the results (see Web Appendix C).
Graph
Table 2. Means (Standard Deviations) of WOM and Psychological Processes as a Function of Contractuality.
| Experiment 2 | Experiment 3 |
|---|
| Contractuality | Low | High | Low | High |
|---|
| WOM | 71% (45%) | 51% (50%)** | 6.23 (.81) | 5.89 (.94)** |
| Process | | | | |
| Relational value | 6.19 (.88) | 5.84 (1.17)** | 6.18 (.85) | 5.80 (1.02)** |
| Motivation to help | 6.05 (1.04) | 5.79 (1.15)* | 5.86 (.99) | 5.50 (1.11)** |
| Alternatives | | | | |
| Reciprocity | 4.00 (1.66) | 3.68 (1.83) | 3.81 (1.83) | 3.50 (1.85) |
| Reactance | 2.01 (1.48) | 2.28 (1.69) | 2.10 (1.57) | 2.72 (1.95)** |
| Surprise | 6.12 (.97) | 5.89 (1.12)* | 4.31 (1.47) | 4.36 (1.64) |
| Effort | 2.35 (1.57) | 2.60 (1.58) | 2.72 (1.44) | 3.04 (1.55) |
| Monetary value | 4.55 (1.17) | 4.64 (1.14) | 5.31 (1.27) | 5.45 (1.14) |
1 *p <.05. **p <.01. Notes: Relational value and motivation to help serially mediated the effect of contractuality on WOM, even when controlling for alternative processes.
2 A series of parallel mediation analyses with relational value along with each alternative process as parallel mediators revealed that relational value mediated the effect, whereas the alternative process did not. The only exception is reactance, which mediated the effect in Experiment 3 but not in Experiment 2.
As an ancillary analysis, we examined whether contractuality influenced the valence of participants' comments. To this end, two coders blind to hypotheses and conditions coded the number of positive and negative thoughts respondents wrote in the 242 comments. Given an interrater reliability of 91.45%, we averaged the coders' ratings. An index of WOM valence was computed by subtracting the number of negative thoughts from the number of positive thoughts (e.g., [50]). Thus, the higher the score on this index, the more positive participants' thoughts were overall. Participants who received the perk that was lower in contractuality shared more positive WOM (M = 2.39, SD = 1.24) than participants who received the perk that was higher in contractuality (M = 1.71, SD = 1.23, F( 1, 240) = 17.78, p <.001, ηp2 =.069).
A one-way ANOVA on relational value revealed a significant effect of contractuality (F( 1, 396) = 11.85, p <.001, ηp2 =.029): Participants in the low contractuality condition reported greater perceptions of relational value (M = 6.19, SD =.88) than participants in the high contractuality condition (M = 5.84, SD = 1.17). Contractuality also had a significant effect on motivation to help (F( 1, 396) = 5.66, p =.018, ηp2 =.014): Participants in the low contractuality condition reported greater motivation to help the restaurant (M = 6.05, SD = 1.04) than those in the high contractuality condition (M = 5.79, SD = 1.15).
To test whether changes in relational value and motivation to help underlie the effect of contractuality on WOM, we ran a serial mediation analysis using Model 6 in PROCESS ([32]). We entered the contractuality manipulation as the independent variable (1 = "low contractuality" and 0 = "high contractuality"), likelihood to generate WOM as the dependent variable, and relational value and motivation to help as serial mediators. Consistent with our prediction, the analysis (based on 5,000 bootstrap samples) revealed that relational value and motivation to help serially mediated the effect of contractuality on WOM (b =.14, 95% CI =.037 to.294). The results remained unchanged after controlling for reciprocity, reactance, surprise, effort, and monetary value (b =.05, 95% CI =.002 to.135). In addition, we tested whether any of these processes mediated the effect of contractuality on WOM in parallel. We found that while relational value significantly mediated the effect of contractuality on WOM, reciprocity, effort, surprise, reactance, and monetary value did not. These results, along with an analysis of the effect of contractuality on these alternative processes, are reported in Web Appendix D.
Experiment 2 provides further support for the idea that contractuality influences the effectiveness of perks at fostering WOM (H1). This experiment also shows that the relational signal carried by low-contractuality perks and the resultant motivation to help the company explain this effect (H2) above and beyond reciprocity, reactance, surprise, effort, and perceived value. Low contractuality, by virtue of increasing perceived relational value, triggers a desire to help the company that increases the spreading of positive WOM.
As a follow-up to these results, we conducted an additional experiment to test our relational value account vis-à-vis reciprocity. Specifically, we randomly assigned 453 participants to three conditions: ( 1) a high-contractuality perk, ( 2) a low-contractuality perk, and ( 3) a low-contractuality perk with salient reciprocity norms. Participants in all conditions were asked whether they would engage in WOM in two sequential asks. For the first ask, they were asked if they would write a review (yes vs. no). For the second ask, they were asked whether they would make a positive comment on social media (1 = "not likely at all," and 7 = "very likely"; see details in Web Appendix E).
If our theorizing is correct, the low-contractuality perk should drive WOM via relational value, which is our proposed process. In contrast, the low-contractuality perk with salient reciprocity norms should drive WOM via feelings of obligation, which is a process through which reciprocity operates. Of importance, these two processes should lead to similar outcomes on the first WOM ask but differentiated outcomes on the second WOM ask, thereby revealing their distinct nature. To elaborate, on the first ask, both the low-contractuality perk and the low-contractuality perk with salient reciprocity norms should increase WOM relative to the high-contractuality perk, as both heightened relational value and feelings of obligation motivate participants to engage in WOM. However, we expected a different result on the second ask: Willingness to engage in WOM should be higher in the low-contractuality perk condition compared to the low-contractuality perk with salient reciprocity norms condition. We expected this to occur because relational value motivates consumers to support the brand, which should cause participants' willingness to engage in WOM to sustain into the second request. In contrast, feelings of obligation motivate participants to rid themselves of their "debt" of obligation, a goal that could be satiated by completing only the first WOM ask.
Consistent with these predictions, we found that participants in both the low contractuality condition (62%) and the low contractuality with salient reciprocity norm condition (50%) were more likely to agree to the first WOM ask compared with participants in the high contractuality condition (30%, ps <.001). Critically, this pattern diverged on the second WOM ask. As expected, among participants who agreed to engage in the first WOM ask, those in the low contractuality condition were significantly more likely to engage in the second WOM ask (M = 5.04, SD = 1.73) compared with those in the low contractuality with salient reciprocity norm condition (M = 4.25, SD = 1.85; p =.002) and those in the high contractuality condition (M = 4.36, SD = 1.64, p =.019). Among those who did not agree to engage in the first WOM ask, condition did not influence the likelihood of engaging in the second WOM ask (ps >.60). Together, these results show that whereas a low-contractuality perk that triggers relational value leads to sustained WOM, a low-contractuality perk that triggers a mere reciprocity norm does not, further revealing that relational value differs from reciprocity (for details, see Web Appendix E).
Experiment 2 found that low contractuality fosters WOM by virtue of increasing perceived relational value, which in turn triggers a desire to help the company that increases the spreading of positive WOM. However, perceiving relational value might also induce people to share WOM to self-enhance. Simply put, people might want to tell others about the special treatment they received in order to boost the self in the eye of others. If this is the case, the effect of contractuality on WOM would still hold, but its underlying psychological mechanism would be different. We tested this possibility in Experiment 3.
Participants (n = 303, 155 female, mean age = 36.85 years, SD = 12.38) recruited via MTurk completed an online survey in exchange for monetary compensation. They were told the goal of the study was to explore common consumption experiences. In all conditions, participants read that they were in the process of placing an order with a restaurant. Then, they were randomly assigned to two experimental conditions in which we manipulated the perceived contractuality of a perk by varying the requirements necessary to obtain the perk.
In the high contractuality condition, participants were told that, when placing their order, the restaurant promised they would receive a free dessert if they completed a survey. In the low contractuality condition, participants were told that, when placing their order, the restaurant asked them to complete a survey. In both conditions, participants were told they completed the survey. To keep the expectedness of the perk parallel, in the low contractuality condition, participants were also told that they knew from prior experience that the restaurant often included a free dessert with their order, so they expected to receive one. All participants read that after a short time their order was delivered, along with a note that informed them that the restaurant had included a free dessert. To conclude, participants reported their demographics.
Participants indicated how likely they would be to ( 1) talk positively about this restaurant to their friends and ( 2) say positive things about the restaurant to others on a seven-point scale (1 = "not likely at all," and 7 = "very likely;" r =.81).
We measured relational value (r =.83) and motivation to help the restaurant (r =.77) using the same measures as in Experiment 2 (1 = "not at all," and 7 = "very much").
To measure motivation to self-enhance, we asked participants to indicate the extent to which they felt motivated to ( 1) showcase that they got a free dessert and ( 2) showcase their relationship with the restaurant (1 = "not at all," and 7 = "very much"; r =.71). We also measured reciprocity (r =.88), reactance (r =.84), surprise (α =.88), effort (α =.84), and perceived monetary value using the same measures as in Experiment 2.
We measured contractuality using the same three items described in Experiment 1 (α =.90).
A one-way ANOVA revealed that participants perceived the perk to be given in a less contractual way in the low contractuality condition (M = 4.14, SD = 1.77) than in the high contractuality condition (M = 5.50, SD = 1.33; F( 1, 301) = 57.32, p <.001, ηp2 =.160).
A one-way ANOVA on likelihood to generate WOM revealed a significant effect of contractuality (F( 1, 301) = 11.05, p =.001, ηp2 =.035; Table 2). As predicted, participants in the low contractuality condition were more likely to engage in WOM (M = 6.23, SD =.81) than those in the high contractuality condition (M = 5.89, SD =.94).
A one-way ANOVA on relational value revealed a significant effect of contractuality (F( 1, 301) = 12.62, p <.001, ηp2 =.040): Participants in the low contractuality condition reported greater perceptions of relational value (M = 6.18, SD =.85) than participants in the high contractuality condition (M = 5.80, SD = 1.02). Consistent with our theoretical account, contractuality also had a significant effect on motivation to help (F( 1, 295) = 8.96, p =.003, ηp2 =.029, six responses missing): participants in the low contractuality condition reported greater motivation to help the restaurant (M = 5.86, SD =.99) than participants in the high contractuality condition (M = 5.50, SD = 1.11).
To test whether relational value and motivation to help underlie the effect of contractuality on WOM, we ran a serial mediation analysis ([32]; Model 6). As predicted, the analysis (based on 5,000 bootstrap samples) revealed that relational value and motivation to help serially mediated the effect of contractuality on WOM (b =.09, 95% CI =.037 to.158). This result remained significant after controlling for reciprocity, reactance, surprise, effort, and monetary value (b =.05, 95% CI =.011 to.096). In addition, we tested whether any of these processes mediated the effect of contractuality on WOM in parallel with relational value. Whereas relational value mediated the effect of contractuality on WOM, reciprocity, effort, surprise, and monetary value did not (see Web Appendix F). The only exception was reactance, which mediated the effect of contractuality on WOM (b =.02, 95% CI =.002 to.056) along with relational value (b =.27, 95% CI =.121 to.416), albeit with a smaller influence.
To test whether relational value could also induce people to share WOM by increasing their motivation to self-enhance rather than their motivation to help the company, we conducted a serial mediation analysis using Model 81 in PROCESS ([32]). We entered the contractuality manipulation as the independent variable (1 = "low contractuality" and 0 = "high contractuality"), likelihood to generate WOM as the dependent variable, relational value as Mediator 1, and motivation to help and motivation to self-enhance as Mediators 2. This analysis revealed that relational value increased WOM by motivating participants to help the company (contractuality → relational value → motivation to help the company → WOM; b =.08, 95% CI =.032 to.153), rather than by motivating participants to self-enhance (contractuality → relational value → motivation to self-enhance → WOM; b =.01, 95% CI = −.010 to.034).
Experiment 3 provides further evidence for the notion that contractuality influences the effectiveness of a perk at fostering WOM because of the relational signal carried by low-contractuality perks and the resultant motivation to help the company. We again found this account persisted above and beyond alternative accounts.
Experiment 4 tested whether the beneficial effect of lower contractuality on WOM is attenuated when a perk is attributed to ulterior motives, which would thus undermine the relational signal of a perk lower in contractuality (H3). To manipulate the likelihood that participants would attribute a low-contractuality perk to ulterior motives, we followed prior literature ([12]; [50]) and varied whether the perk was offered before purchase (ulterior motive salient) or after purchase (ulterior motive not salient). We reasoned that if we provided a low-contractuality perk before purchase, participants would be less likely to perceive it as a signal that the brand values consumers, as alternative attributions will be salient. As such, the beneficial effect of lower-contractuality perks on WOM should be attenuated.
Participants (n = 405, 182 female, mean age = 37.25 years, SD = 11.63) recruited via MTurk completed an online survey in exchange for monetary compensation. Because the study was about auto insurance renewal, we recruited only participants who indicated having purchased auto insurance before and thus could relate to the experimental stimuli ([76]). Moreover, only participants who passed an initial attention check were eligible to participate. Participants were randomly assigned to conditions in a 2 (contractuality: high vs. low) × 2 (timing: before vs. after purchase) between-participants design. In all conditions, participants were told that they called an agent who walked them through several options and upgrades while reviewing their auto insurance.
In the high contractuality condition, participants read that the agent offered a free car care kit if they were willing to fill out a survey, which they agreed to. In the low contractuality condition, the agent simply offered a free car care kit.
Half of the participants read that the more versus less contractual perk (i.e., the free car care kit) was offered before the purchase decision, whereas the remaining half read that the perk was offered after the purchase decision ([12]; [50]).
Next, participants indicated how likely they would be to talk positively about this insurance company to others (1 = "not at all likely," and 7 = "very likely"). We also measured alternative processes related to surprise, effort, reciprocity, and reactance using the same measures as in Experiments 2 and 3. In addition, we measured perceived monetary value by asking participants to indicate the value of the car care kit (from $1 to $50). At the end of the study, participants reported demographics.
A 2 (contractuality: high vs. low) × 2 (timing: after purchase vs. before purchase) ANOVA on likelihood to engage in WOM revealed a significant main effect of contractuality (F( 1, 401) = 40.39, p <.001, ηp2 =.092) and no effect of timing (F( 1, 401) = 2.21, p =.138). More importantly, the analysis yielded a significant contractuality × timing interaction (F( 1, 401) = 5.51, p =.019, ηp2 =.014). Planned contrasts revealed that when the agent provided the perk after the purchase—when the salience of an ulterior motive should be lower—participants reported greater likelihood to engage in WOM when the offering was low in contractuality (M = 5.82, SD =.97) than when it was high in contractuality (M = 4.80, SD = 1.21; t(401) = 6.15, p <.001), conceptually replicating the previous studies. However, when the agent provided the perk before purchase—when the salience of an ulterior motive should be higher— this effect was attenuated, although it remained statistically significant (low contractuality: M = 5.37, SD = 1.19 vs. high contractuality: M = 4.90, SD = 1.34; t(401) = 2.84, p =.005). From a slightly different perspective, giving a low-contractuality perk before (vs. after) purchase significantly decreased WOM (t(401) = −2.72, p =.007). In contrast, timing did not influence the effectiveness of the high-contractuality perk in shaping WOM (t(401) =.61, p =.546).
We also conducted a 2 (contractuality) × 2 (timing) analysis of covariance (ANCOVA) with likelihood to engage in WOM as the dependent variable and surprise, effort, monetary value of the perk, reciprocity, and reactance as covariates to control for alternative processes. The contractuality × timing interaction remained significant (F( 1, 396) = 4.63, p =.032). Contrast analysis revealed that whereas a low-contractuality perk generated more WOM than a high-contractuality perk when given after purchase (t(396) = 3.92, p <.001), this was not the case when the perk was given before purchase (t(396) = 1.10, p =.274; for details, see Web Appendix G).
Experiment 4 provides convergent evidence that perks lower in contractuality are more likely to trigger WOM than perks higher in contractuality (H1). Moderation by timing of the offering suggests that this is less likely to occur when low-contractuality perks are attributed to ulterior motives, which is consistent with our relational value account that consumers must feel a company actually values them (H3).
In Experiment 5, we aimed to further test our process via moderation. Prior research suggests that when consumers dislike a person, a positive gesture such as a flattering comment might be viewed with suspicion ([50]). As such, when consumers interact with a disliked company, they may be more suspicious of lower- compared to higher-contractuality perks. High-contractuality perks have explicit contingencies that make the company's motive transparent (e.g., buy ten coffees to get one for free). In contrast, low-contractuality perks lack explicit contingencies and thus may introduce ambiguity with regard to the company's motive. Therefore, consumers who dislike a company may default to sinister attributions for the company's behavior ([42]), such as the company being manipulative. As such, consumers' dislike or distrust toward a company may lead them to perceive low-contractuality perks as being driven by ulterior motives, undermining such perks' positive effect on WOM (H3).
A total of 417 participants (194 female, mean age = 40.80 years, SD = 12.61; 11 responses missing) recruited via MTurk completed an online survey in exchange for monetary compensation. Because the study was about auto insurance renewal, we used the same recruitment procedures as in Experiment 4. Participants were randomly assigned to conditions in a 2 (contractuality: high vs. low) × 2 (company: control vs. disliked) between-participants design.
All participants read a scenario in which they had to renew their auto insurance. In the disliked company condition, participants read that they received a call from a telemarketing agent working for a direct selling insurance company they disliked, who talked them into buying auto insurance. In the control condition, participants read that they called an agent from an insurance company and decided to buy auto insurance. A pretest with 102 respondents from the same population confirmed that the liking manipulation worked as intended (for details, see Web Appendix H).
We used the same contractuality manipulation as in Experiment 4, whereby participants either read that the agent offered a free car care kit for filling out a survey, or that the agent simply offered a free car care kit.
Next, we asked participants to report how likely they would be to talk positively about this insurance company to others (1 = "not at all likely," and 7 = "very likely"). We also measured alternative processes related to reciprocity, reactance, surprise, effort, and perceived monetary value using the same measures as in Experiments 2, 3 and 4. Next, participants completed an attention check to verify they read the contractuality manipulation. Specifically, participants were asked what they had to do to get the free car care kit (fill out a survey vs. nothing). At the end of the study, participants reported demographics.
Forty-eight participants failed the attention check, which left us with 369 participants. We report results without any exclusions in Web Appendix I.
A 2 (contractuality: high vs. low) × 2 (company: control vs. disliked) ANOVA on likelihood to engage in WOM revealed a significant main effect of company (F( 1, 365) = 254.19, p <.001, ηp2 =.41): Participants reported greater likelihood to engage in WOM in the control company condition (M = 4.99, SD = 1.36) than in the disliked company condition (M = 2.77, SD = 1.47). The effect of contractuality was nonsignificant (F( 1, 365) = 1.40, p =.238, ηp2 =.004). More importantly, the analysis yielded a significant interaction between perk contractuality and company (F( 1, 365) = 22.95, p <.001, ηp2 =.059; see Figure 1).
Graph: Figure 1. Likelihood to engage in WOM as a function of contractuality and disliked company (Experiment 5).Notes: Error bars represent ± 1 standard error.
Follow-up contrasts revealed that when participants imagined interacting with the company in the control condition, they reported greater likelihood to engage in WOM when the perk was low in contractuality (M = 5.46, SD = 1.20) than when it was high in contractuality (M = 4.60, SD = 1.37; t(365) = 4.29, p <.001), consistent with prior studies. However, when participants imagined interacting with a disliked company, they reported a lower likelihood to engage in WOM when the perk was low in contractuality (M = 2.48, SD = 1.42) than when it was high in contractuality (M = 3.00, SD = 1.47; t(365) = −2.51, p =.012), resulting in a reversal of the focal effect.
To probe the role of alternative processes, we conducted a 2 × 2 ANCOVA with likelihood to engage in WOM as the dependent variable and surprise, reciprocity, reactance, effort, and perceived value as covariates. The contractuality × company interaction remained significant (F( 1, 351) = 14.36, p <.001; nine responses missing). Follow-up contrasts revealed that when participants imagined interacting with the company in the control condition, they reported marginally greater likelihood to engage in WOM when the perk was low in contractuality than when it was high in contractuality (t(351) = 1.84, p =.066), whereas the opposite was true when participants imagined interacting with a disliked company (t(351) = −3.33, p =.001; for details, see Web Appendix J).
Experiment 5 documented a reversal of the beneficial effects of low contractuality on WOM. Specifically, we found that when the low-contractuality perk was given by a disliked company, participants' likelihood to share positive WOM decreased below that of receiving a high-contractuality perk.
Experiment 6 sought to provide further evidence that contractuality influences WOM by examining social media likes in a real-world setting ([34]). In addition, we aimed to test the idea that contractuality may come with a trade-off: despite being more effective at fostering WOM, low-contractuality perks might be less effective than high-contractuality perks as incentives aimed to induce compliance with a request for a desired behavior.
To achieve these aims, we conducted a field study in collaboration with a university-affiliated bookstore that was seeking to encourage shoppers to fill out a customer satisfaction survey. In the high contractuality condition, shoppers were promised a gift card for filling out the survey. In the low contractuality condition, shoppers were asked to fill out the survey without any mention of the gift card. To test whether high contractuality is more effective at stimulating a desired behavior (i.e., filling out a survey), we recorded whether shoppers agreed to complete the survey. Upon completing the survey, shoppers in both conditions received a gift card. To test whether low contractuality is more effective at generating WOM, after a time delay, shoppers were invited to like the bookstore on social media. We predicted that whereas high contractuality would be more effective at increasing participation in the survey, low contractuality would be more effective at increasing social media likes.
A research assistant posing as an employee approached shoppers at a university bookstore. Because the study aimed to explore WOM on Instagram for a university-affiliated bookstore, the research assistant was instructed to approach university students and only administer the treatment to those who confirmed having an Instagram account.
In the high contractuality condition, the research assistant asked shoppers if they would be willing to complete a five-minute survey for the bookstore and informed them that the bookstore had committed to give them a $5 gift card if they did so. In the low contractuality condition, the research assistant only asked shoppers if they would be willing to complete a five-minute survey for the bookstore, without mentioning the $5 gift card. The research assistant recorded the treatment condition they were in, as well as whether they agreed to complete the survey. If shoppers agreed, the research assistant handed them an iPad and asked them to complete the survey.
Of all participants approached, 149 agreed to complete the survey (71 female, mean age = 20.47 years, SD = 3.07). Although our original plan for data collection was to have 160–200 shoppers fill out the survey and share their social media handle (i.e., 80–100 shoppers per cell), we were ordered to stop data collection due to the COVID-19 outbreak. We therefore chose to end the study and analyze the responses we acquired. In the survey, shoppers were asked to indicate why they were coming to the bookstore, why they had decided to shop in the store versus online, and their satisfaction with the bookstore on several dimensions (i.e., level of service, pleasantness of the atmosphere, selection of products, and ease of findings products; 1 = "poor," and 7 = "excellent").
The survey also contained questions to assess participant characteristics. The goal of these measures was to probe for a possible selection bias that testing our trade-off hypothesis inherently introduced. Namely, those who agreed to comply with the request to fill out the survey in the high contractuality condition might have differed systematically from those who agreed to comply with the request in the low contractuality condition. If this occurred, differences in WOM could not be attributed exclusively to our experimental manipulation of the contractual nature of the perk. Specifically, we measured potential differences in competence and intelligence (r =.76). We also assessed whether the two groups might differ in altruism by asking participants to what extent they saw themselves as altruistic and generous (r =.58) on a seven-point scale (1 = "not at all," and 7 = "very much"). To conclude the survey, shoppers reported demographic information and provided their Instagram account handle. Participants returned the iPad upon completing the survey.
In the high contractuality condition, the research assistant told participants that, as mentioned earlier, they were entitled to a $5 gift card because they completed the survey. In the low contractuality condition, the assistant informed participants that the bookstore was giving them a $5 gift card. Thus, in both the high and low contractuality conditions, the gift card was equivalent in value and was awarded only after completing the same survey. However, the high contractuality condition made it clear that the gift card was given to participants for completing the survey, whereas the low contractuality condition presented the gift card as a gift. The research assistant then gave everyone who completed the survey the $5 gift card.
Shoppers then continued their visit to the bookstore. After a time delay of several minutes, a second research assistant posing as a bookstore employee intercepted participants. The second research assistant was blind to the treatment condition participants had been assigned to. The research assistant told shoppers that the bookstore was trying to increase its social media presence on Instagram and invited them to check out the bookstore handle and like some recent social media posts. The research assistant then gave shoppers the bookstore Instagram handle so they could engage with the bookstore if they wanted to.
A week after the intervention, we logged into the bookstore's Instagram account and counted the number of social media likes the bookstore received from customers who provided their Instagram account handles in the survey.
A separate 2 (contractuality: low vs. high) between-participants pretest with 100 participants recruited from MTurk (42 female, mean age = 35.42 years, SD = 11.67) confirmed that the low-contractuality perk was perceived as lower in contractuality (M = 5.64, SD = 1.18) than the high-contractuality perk (M = 6.45, SD =.77, F( 1,98) = 16.22, p <.001) on the same contractuality scale used in previous studies.
We excluded six participants who did not fit the recruitment criteria, as they were visitors who were not affiliated with the university. In addition, three participants did not provide valid Instagram handles and were thus excluded. We were left with 193 participants, 140 of whom agreed to take the survey.
A chi-square analysis revealed that 78% of shoppers in the high contractuality condition complied with the request to complete the survey, whereas 67% complied with the request in the low contractuality condition (χ2 ( 1, 193) = 2.80, p =.094, Cramer's V =.12). Although this difference was marginally significant, the results are consistent with the notion that high-contractuality perks can be more effective than low-contractuality perks at incentivizing behaviors unrelated to WOM.
We counted the number of posts shoppers liked on Instagram. Because the distribution of likes was skewed, we first log-transformed the number of likes ([23]). A one-way ANOVA showed that shoppers in the low contractuality condition (M = 3.60, SD = 4.06) liked marginally more posts than those in the high contractuality condition (M = 2.66, SD = 3.88, F( 1, 138) = 3.30, p =.072, ηp2 =.023).
To address the possibility of a selection bias inherent in our experimental procedure, we first examined dispositional differences among shoppers in the high versus low contractuality conditions. The analysis revealed that shoppers did not differ on the basis of their intelligence or altruism (ps >. 25). Shoppers also did not differ in terms of their satisfaction with the bookstore (ps >.20), with the exception of level of service, in which those in the high contractuality condition reported marginally greater satisfaction (M = 5.84, SD = 1.81) than those in the low contractuality condition (M = 5.17, SD = 2.39, F( 1, 138) = 3.54, p =.062), a result that is inconsistent with the idea that the observed effect on WOM was due to participants in the low (vs. high) contractuality condition being more satisfied with the bookstore. Furthermore, we conducted the analysis on WOM while controlling for demographic information (age and gender) as well as dispositional differences (intelligence and altruism) and satisfaction with the bookstore (four dimensions). The analysis revealed a marginally significant effect of age (F( 1, 128) = 3.67, p =.058, ηp2 =.028, two responses missing). More importantly, contractuality had a significant effect on WOM (F( 1, 128) = 4.64, p =.037, ηp2 =.034, two responses missing), suggesting that individual differences alone cannot account for the focal effect.
The results suggest that contractuality may sometimes involve a trade-off between the desired behavior a company wants to stimulate and WOM generation. Specifically, our findings show that whereas high contractuality tended to yield a higher response rate on the survey, low contractuality was more effective at increasing the average number of social media likes. Furthermore, low-contractuality perks generated a greater number of total likes (227) than high-contractuality perks (205) and had a lower cost per like ($1.39 vs. $1.88). These results suggest that the widely used practice of imbuing perks with contractuality can come at a cost.
Companies often engage in efforts aimed to spur WOM. Typically, these efforts involve the use of incentives (e.g., paid WOM agents, "refer-a-friend" promotions). Although incentivized WOM can be beneficial, it also has risks and costs for companies (e.g., lower message persuasiveness). As such, marketers have begun to seek ways to promote WOM in the absence of direct incentives. In this spirit, this research examines how marketers can tailor common marketing perks that are already used to serve other objectives to fuel WOM.
Six main experiments and one supplemental experiment show that perks lower in contractuality foster more WOM than perks higher in contractuality. This result occurs because low-contractuality perks are more likely to convey a relational signal than perks higher in contractuality. This effect was robust across different operationalizations of contractuality, different perks, different measures of WOM, and different populations. Our studies also show that when ulterior motives for the perk were salient, the beneficial effect of lower (vs. higher) contractuality was attenuated or even reversed.
This work contributes to two streams of research. First, it contributes to the WOM literature ([ 7]) by identifying both a new psychological process and a practical trigger of WOM. Whereas past research has mostly focused on examining the benefits (e.g., [29]; [62]) and costs (e.g., [38]; [71]) of incentivized WOM, the present work provides insight into how companies can use existing perks to promote WOM that is not directly incentivized. This research shows that perks perceived as lower in contractuality carry a relational signal, which increases consumers' desire to support the brand via WOM. By identifying contractuality as a novel antecedent of WOM, this research provides a response to [77] call for understanding levers companies can use to facilitate nonincentivized WOM.
Second, our research contributes to work on marketing perks ([49]; [57]). Whereas prior literature has mostly focused on how to structure perks to increase sales (e.g., [40]) and customer satisfaction (e.g., [37]), the present research is the first to examine how perks can be effectively framed to promote WOM. Furthermore, this research draws attention to an unexplored psychological mechanism that perks can activate: relational value. Research in social psychology ([60]) shows that when one appraises another to have high relational value, it triggers a cascade of processes, including compassionate goals toward the partner ([13]) and feelings of gratitude ([ 3]), which work in concert to foster positive interpersonal relationships ([60]). This work provides the first evidence that marketing perks can signal relational value.
Our research suggests that marketers can encourage WOM with easily implementable pivots that lower the contractuality of their perks. For example, contractuality can be lowered by reducing the restrictiveness of a perk, reducing the salience of a perk's contingencies, or framing a perk as a gift from the company rather than a prize earned through a customer's effort. In each of these instances, companies do not have to change what they are offering—they only need to change the way consumers perceive the perk.
Our research also cautions against the idea that low contractuality is always beneficial. Our results suggest that sometimes consumers may be suspicious of perks low in contractuality, with potentially negative consequences on WOM. This seems particularly likely to occur when low-contractuality perks come from disliked or distrusted brands, or when they are awarded before rather than after purchase. For example, many consumers do not like utility providers or financial institutions (e.g., [ 1]). To the extent that dislike prompts consumers to make hostile attributions toward benevolent gestures, such companies might be better off using perks that are higher in contractuality or using low-contractuality perks in conjunction with other interventions, such as positioning themselves as underdogs (e.g., [55]). Moreover, to the extent that awarding a low-contractuality perk before purchase may make ulterior motives salient, companies may be better off timing them so that they are awarded after the purchase.
Finally, our research suggests that although high-contractuality perks can be effective at motivating consumers to engage in a desired behavior, they may also carry an opportunity cost. Indeed, the field study conducted at the university bookstore showed that although shoppers were more likely to comply to a request (i.e., filling out a survey) when given a high- (vs. low-) contractuality perk, they were less likely to share WOM about the company. These results suggest that when designing perks, marketers need to be cognizant of the potential trade-off between incentivizing first-order responses and nudging second-order responses such as WOM.
One may wonder how the findings observed in this research relate to intrinsic and extrinsic motivation ([22]). Specifically, perhaps offerings higher in contractuality trigger extrinsic motivation, which undermines participants' willingness to perform future behavior without an incentive (i.e., such offerings undermine intrinsic motivation). However, if this were true, we would not have expected to find mediational evidence via relational value. Furthermore, the finding that people are more likely to engage in WOM about a disliked company when contractuality is high rather than low seems also inconsistent with such an account. Nonetheless, future research could examine whether conditions exist in which contractuality might affect the nature of consumers' motivation.
Our research differs from prior literature on exchange and communal relationships ([ 2]; [18]). In exchange relationships, partners give each other benefits with the expectation of receiving a comparable benefit in return, whereas in communal relationships, no such expectation is present. The main premise in this literature is that once people are in a certain type of relationship (communal or exchange), they evaluate a partner's action on the basis of whether they conform to or violate the norms of the relationship. In contrast, our theory does not hinge on the premise that people are in a particular type of relationship with a brand, but only on the idea that when people receive a perk in a low-contractualilty manner, they interpret it as a benevolent behavior that heightens their relational value.
Might the effect of low contractuality on WOM be explained via liking? While feeling valued by a company might increase liking toward the company, prior literature (e.g., [13]) shows that relational value does more than increase liking: it activates compassionate goals. In other words, relational value operates through a motivational route that might entail positive attitudes, but it ought not be reduced to positive attitudes alone. Thus, we contend that the effects of low-contractuality perks on WOM are contingent on such perks heightening consumers' motivation to help the brand rather than merely increasing their liking for the brand.
Future research could examine other circumstances under which perks higher versus lower in contractuality increase WOM. For example, prior research has argued that perks might strengthen the relationship between a consumer and a company if the acquisition of the perk is effortful ([41]). A perk that has low value yet requires considerable effort to be earned could prompt consumers to conclude that they really care about the company. In this research, we intentionally held effort constant, but future research could allow effort to vary. High-contractuality perks might be more effective at increasing commitment than low-contractuality perks to the extent they require more effort on consumers' part and/or when the effort-to-benefit ratio is extremely salient ([24]). Similarly, additional research could examine whether our conceptual framework applies to other marketing tactics. For example, do less-contractual return policies and products (e.g., devices that are more compatible) convey a relational signal that spurs WOM?
Finally, future research could examine whether marketing offerings that tap into distinct WOM motives have a differential effect on WOM persuasiveness. In the present research, participants appeared motivated to generate WOM because they wanted to help the firm after the firm gave them a marketing offering that was lower in contractuality. However, marketing offerings could also prompt consumers to generate WOM by way of triggering the desire to self-enhance ([21]). For example, preferential treatment that signals high status could boost a customer's desire to self-enhance in front of others. Future research could explore how different motivations for sharing WOM ultimately shape how persuasive the WOM is to others.
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921991798 - How Marketing Perks Influence Word of Mouth
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921991798 for How Marketing Perks Influence Word of Mouth by Monika Lisjak, Andrea Bonezzi and Derek D. Rucker in Journal of Marketing
Footnotes 1 Jonah Berger
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 https://doi.org/10.1177/0022242921991798
References Accenture (2013), "The New Energy Consumer Handbook ," https://www.accenture.com/%5facnmedia/Accenture/next-gen/insight-unlocking-value-of-digital-consumer/PDF/Accenture-New-Energy-Consumer-Handbook-2013.pdf.
Aggarwal Pankaj. (2004), " The Effects of Brand Relationship Norms on Consumer Attitudes and Behavior ," Journal of Consumer Research , 31 (1), 87 – 101.
Algoe Sara B. , Haidt Jonathan , Gable Shelly L.. (2008), " Beyond Reciprocity: Gratitude and Relationships in Everyday Life ," Emotion , 8 (3), 425 – 29.
Analytic Partners (2014), "Landmark Study Reveals How Word-of-Mouth Correlates with Rise in Sales," (November 18), https://analyticpartners.com/news-blog/2014/11/word-of-mouth-interactions-correlate-with-rise-in-sales/
5 Barasch Alixandra , Berman Jonathan Z. , Small Deborah A.. (2016), " When Payment Undermines the Pitch: On the Persuasiveness of Pure Motives in Fundraising ," Psychological Science , 27 (10), 1388 – 97.
6 Beck Jonathan M. , Voorhees Clay M. , Fombelle Paul W. , Lemon Katherine N.. (2020), " Automated Electronic Word of Mouth Suggestions from the Firm: Untapped Potential or Inevitable Backlash? " working paper, Marketing Science Institute.
7 Berger Jonah. (2011), " Arousal Increases Social Transmission of Information ," Psychological Science , 22 (7), 891 – 93.
8 Berger Jonah. (2014), " Word of Mouth and Interpersonal Communication: A Review and Directions for Future Research ," Journal of Consumer Psychology , 24 (4), 586 – 607.
9 Berger Jonah , Milkman Katherine L.. (2012), " What Makes Online Content Viral? " Journal of Marketing Research , 49 (2), 192 – 205.
Berger Jonah , Schwartz Eric M.. (2011), " What Drives Immediate and Ongoing Word of Mouth? " Journal of Marketing Research , 48 (5), 869 – 80.
Brehm Jack W.. (1966), A Theory of Psychological Reactance. New York : Academic Press.
Campbell Margaret C. , Kirmani Amna. (2000), " Consumers' Use of Persuasion Knowledge: The Effects of Accessibility and Cognitive Capacity on Perceptions of an Influence Agent ," Journal of Consumer Research , 27 (1), 69 – 83.
Canevello Amy , Crocker Jennifer. (2010), " Creating Good Relationships: Responsiveness, Relationship Quality, and Interpersonal Goals ," Journal of Personality and Social Psychology , 99 (1), 78 – 106.
Celen Bogachan , Kariv Shachar , Schotter Andrew. (2010), " An Experimental Test of Advice and Social Learning ," Management Science , 56 (10), 1687 – 701.
Cheatham Lauren B. , Tormala Zakary L.. (2017), " The Curvilinear Relationship Between Attitude Certainty and Attitudinal Advocacy ," Personality and Social Psychology Bulletin , 43 (1), 3 – 16.
Chen Zoey. (2017), " Social Acceptance and Word of Mouth: How the Motive to Belong Leads to Divergent WOM with Strangers and Friends ," Journal of Consumer Research , 44 (3), 613 – 32.
Cialdini Robert. (2006), Influence: The Psychology of Persuasion, Revised Edition. New York : Harper Business.
Clark Margaret S. , Mills Judson. (1979), " Interpersonal Attraction in Exchange and Communal Relationships ," Journal of Personality and Social Psychology , 37 (1), 12 – 24.
Collins Nancy L. , Feeney Brooke C.. (2000), " A Safe Haven: An Attachment Theory Perspective on Support Seeking and Caregiving in Intimate Relationships ," Journal of Personality and Social Psychology , 78 (6), 1053 – 73.
Daryanto Ahmad , de Ruyter Ko , Wetzels Martin , Patterson Paul G.. (2010), " Service Firms and Consumer Loyalty Programs: A Regulatory Fit Perspective of Reward Preferences in a Health Club Setting ," Journal of the Academy of Marketing Science , 38 (5), 604 – 16.
De Angelis Matteo , Bonezzi Andrea , Peluso Alessandro M. , Rucker Derek D. , Costabile Michele. (2012), " On Braggarts and Gossips: A Self-Enhancement Account of Word-of-Mouth Generation and Transmission ," Journal of Marketing Research , 49 (4), 551 – 63.
Deci Edward L. , Koestner Richard , Ryan Richard M.. (1999), " A Meta-Analytic Review of Experiments Examining the Effects of Extrinsic Rewards on Intrinsic Motivation ," Psychological Bulletin , 125 (6), 627 – 68.
Dewitte Siegfried , Bruyneel Sabrina , Geyskens Kelly. (2009), " Self-Regulating Enhances Self-Regulation in Subsequent Consumer Decisions Involving Similar Response Conflicts ," Journal of Consumer Research , 36 (3), 394 – 405.
Dodson Joe A. , Tybout Alice M. , Sternthal Brian. (1978), " Impact of Deals and Deal Retraction on Brand Switching ," Journal of Marketing Research , 15 (1), 72 – 81.
Dubois David , Bonezzi Andrea , De Angelis Matteo. (2016), " Sharing with Friends Versus Strangers: How Interpersonal Closeness Influences Word-of-Mouth Valence ," Journal of Marketing Research , 53 (5), 712 – 27.
Eisenberger Robert , Huntington Robin , Hutchison Steven , Sowa Debora. (1986), " Perceived Organizational Support ," Journal of Applied Psychology , 71 (3), 500 –5 07.
Fitzsimons Gavan J. , Lehmann Donald R.. (2004), " Reactance to Recommendations: When Unsolicited Advice Yields Contrary Response ," Marketing Science , 23 (1), 83 – 94.
Friestad Marian , Wright Peter. (1994), " The Persuasion Knowledge Model: How People Cope with Persuasion Attempts ," Journal of Consumer Research , 21 (1), 1 – 31.
Godes David , Mayzlin Dina. (2009), " Firm-Created Word-of-Mouth Communication: Evidence from a Field Test ," Marketing Science , 28 (4), 721 – 39.
Gouldner Alvin W.. (1960), " The Norm of Reciprocity: A Preliminary Statement ," American Sociological Review , 25 (2), 161 – 78.
Haenlein Michael , Libai Barak. (2017), " Seeding, Referral, and Recommendation: Creating Profitable Word-of-Mouth Programs ," California Management Review , 59 (2), 68 – 91.
Hayes Andrew F.. (2013), Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach. New York : The Guilford Press.
Hennig-Thurau Thorsten , Gwinner Kevin P. , Walsh Gianfranco , Gremler Dwayne D.. (2004), " Electronic Word-of-Mouth via Consumer-Opinion Platforms: What Motivates Consumers to Articulate Themselves on the Internet? " Journal of Interactive Marketing , 18 (1), 38 – 52.
Hoffman Donna L. , Fodor Marek. (2010), " Can You Measure the ROI of Your Social Media Marketing? " Sloan Management Review , 52 (1), 41 – 49
Isaac Mathew S. , Grayson Kent. (2017), " Beyond Skepticism: Can Accessing Persuasion Knowledge Bolster Credibility? " Journal of Consumer Research , 43 (6), 895 – 912.
Isen Alice M. , Simmonds Stanley F.. (1978), " The Effect of Feeling Good on a Helping Task That Is Incompatible with Good Mood ," Social Psychology , 41 (4), 346 – 49.
Jiang Lan , Hoegg JoAndrea , Dahl Darren W.. (2013), " Consumer Reaction to Unearned Preferential Treatment ," Journal of Consumer Research , 40 (3), 412 – 27.
Jin Liyin , Huang Yunhui. (2014), " When Giving Money Does Not Work: The Differential Effects of Monetary Versus In-Kind Rewards in Referral Reward Programs ," International Journal of Research in Marketing , 31 (1) 107 – 16.
Keller Fay Group (2012), " The Power of Consumer-to-Consumer Recommendation," https://rewardstream.com/blog/consumer-to-consumer-recommendations-2012.
Kivetz Ran. (2005), " Promotion Reactance: The Role of Effort–Reward Congruity ," Journal of Consumer Research , 31 (4), 725 – 36.
Kivetz Ran , Simonson Itamar. (2003), " The Idiosyncratic Fit Heuristic: Effort Advantage as a Determinant of Consumer Response to Loyalty Programs ," Journal of Marketing Research , 40 (4), 454 – 67.
Kramer Roderick M.. (1998), " Paranoid Cognition in Social Systems: Thinking and Acting in the Shadow of Doubt ," Personality and Social Psychology Review , 2 (4), 251 – 75.
Kurtessis James N. , Eisenberger Robert , Ford Michael T. , Buffardi Louis C. , Stewart Kathleen A. , Adis Cory S.. (2017), " Perceived Organizational Support: A Meta-Analytic Evaluation of Organizational Support Theory ," Journal of Management , 43 (6), 1854 – 84.
Laran Juliano , Tsiros Michael. (2013), " An Investigation of the Effectiveness of Uncertainty in Marketing Promotions Involving Free Gifts ," Journal of Marketing , 77 (2), 112 – 23.
Leary Mark R.. (2005), " Sociometer Theory and the Pursuit of Relational Value: Getting to the Root of Self-Esteem ," European Review of Social Psychology , 16 (1), 75 – 111.
Leary Mark R. , Baumeister Roy F.. (2000). " The Nature and Function of Self-Esteem: Sociometer Theory ," in Advances in Experimental Social Psychology , Zanna Mark P. , ed. Cambridge, MA : Academic Press , 1 – 62.
Lemay Edward P. Jr , Clark Margaret S. , Feeney Brooke C.. (2007). " Projection of Responsiveness to Needs and the Construction of Satisfying Communal Relationships ," Journal of Personality and Social Psychology , 92 (5), 834 – 53.
Libai Barak , Muller Eitan , Peres Renana. (2013), " Decomposing the Value of Word of Mouth Seeding Programs: Acceleration Versus Expansion ," Journal of Marketing Research , 50 (2), 161 – 76.
Liu Peggy , Lamberton Cait , Haws Kelly L.. (2015), " Should Firms Use Small Financial Benefits to Express Appreciation to Consumers? Understanding and Avoiding Trivialization Effects ," Journal of Marketing , 79 (3), 74 – 90.
Main Kelley J. , Dahl Darren W. , Darke Peter R.. (2007), " Deliberative and Automatic Bases of Suspicion: Empirical Evidence of the Sinister Attribution Error ," Journal of Consumer Psychology , 17 (1), 59 – 69.
McCullough Michael E. , Kimeldorf Marcia B. , Cohen Adam D.. (2001), " An Adaptation for Altruism? The Social Causes, Social Effects, and Social Evolution of Gratitude ," Current Directions in Psychological Science , 17 (4), 281 – 85.
Mikulincer Mario , Shaver Phillip R. , Gillath Omri , Nitzberg Rachel A.. (2005), " Attachment, Caregiving, and Altruism: Boosting Attachment Security Increases Compassion and Helping ," Journal of Personality and Social Psychology , 89 (5), 817 – 39.
Nielsen (2015), " Global Trust in Advertising: Winning Strategies for an Evolving Media Landscape ," (September 28) , https://www.nielsen.com/us/en/insights/report/2015/global-trust-in-advertising-2015/.
Packard Grant , Wooten David B.. (2013), " Compensatory Knowledge Signaling in Consumer Word-of-Mouth ," Journal of Consumer Psychology , 23 (4), 434 – 50.
Paharia Neeru , Keinan Anat , Avery Jill , Schor Juliet B.. (2010), " The Underdog Effect: The Marketing of Disadvantage and Determination Through Brand Biography ," Journal of Consumer Research , 37 (5), 775 – 90.
Peluso Alessandro M. , Bonezzi Andrea , De Angelis Matteo , Rucker Derek D.. (2017), " Compensatory Word-of-Mouth: Advice as a Device to Restore Control ," International Journal of Research in Marketing , 34 (2), 499 – 515.
Reczek Rebecca Walker , Haws Kelly L. , Summers Christopher A.. (2014), " Lucky Loyalty: The Effect of Consumer Effort on Predictions of Randomly Determined Marketing Outcomes ," Journal of Consumer Research , 41 (4), 1065 – 77.
Regan Dennis T.. (1971), " Effects of a Favor and Liking on Compliance ," Journal of Experimental Social Psychology , 7 (6), 627 – 39.
Reichheld Frederick F.. (2003), " The One Number You Need to Know ," Harvard Business Review , 81 (12), 46 – 54.
Reis Harry T. , Clark Margaret S. , Holmes John G.. (2004), " Perceived Partner Responsiveness as an Organizing Construct in the Study of Intimacy and Closeness ," in The Handbook of Closeness and Intimacy , Mashek Debra J. , Aron Arthur , eds. Mahwah, NJ : Lawrence Erlbaum Associates , 201 – 25.
Reis Harry T. , Shaver Phillip. (1988), " Intimacy as an Interpersonal Process ," in Handbook of Personal Relationships , Duck Steve , ed. Chichester, UK : John Wiley and Sons , 367 – 89.
Ryu Gangseog , Feick Lawrence. (2007), " A Penny for Your Thoughts: Referral Reward Programs and Referral Likelihood ," Journal of Marketing , 71 (1), 84 – 94.
Schmitt Philipp , Skiera Bernd , Bulte Christophe Van den. (2011), " Referral Programs and Consumer Value ," Journal of Marketing , 75 (1), 46 – 59.
Schopler John , Thompson Vaida D.. (1968), " Role of Attribution Processes in Mediating Amount of Reciprocity for a Favor ," Journal of Personality and Social Psychology , 10 (3), 243 – 50.
Shore Lynn M. , Shore Ted H.. (1995), " Perceived Organizational Support and Organizational Justice ," in Organizational Politics, Justice, and Support: Managing the Social Climate of the Workplace , Cropanzano Russell S. , Kacmar K. Michele , eds. Westport, CT : Quorum , 147 – 97.
Sirdeshmukh Deepak , Singh Jagdip , Sabol Barry. (2002), " Consumer Trust, Value, and Loyalty in Relational Service Exchanges ," Journal of Marketing , 66 (1), 15 – 37.
Skinner Burrhus F.. (1938), The Behavior of Organisms: An Experimental Analysis. New York : Appleton-Century.
Sundaram D. Suresh , Mitra Kaushik , Webster Cynthia. (1998), " Word-of-Mouth Communications: A Motivational Analysis ," in Advances in Consumer Research , Vol. 25 , Alba Joseph W. , Wesley Hutchinson J. , eds. Provo, UT : Association for Consumer Research , 527 – 31.
Thorndike Edward L.. (1898), " Animal Intelligence: An Experimental Study of the Associative Processes in Animals ," The Psychological Review: Monographs Supplements , 2 (4), 1 – 109.
Trusov Michael , Bucklin Randolph E. , Pauwels Koen. (2009), " Effects of Word-of-Mouth Versus Traditional Marketing: Findings from an Internet Social Networking Site ," Journal of Marketing , 73 (5), 90 – 102.
Tuk Mirjam A. , Verlegh Peeter W. J. , Smidts Ale , H. J. Wigboldus Daniel. (2009), " Sales and Sincerity: The Role of Relational Framing in Word-of-Mouth Marketing ," Journal of Consumer Psychology , 19 (1), 38 – 47.
Valenzuela Ana , Mellers Barbara , Strebel Judi. (2010) " Pleasurable Surprises: A Cross-Cultural Study of Consumer Responses to Unexpected Incentives ," Journal of Consumer Research , 36 (5), 792 – 805.
Verlegh Peeter W. J. , Ryu Gangseog , Tuk Mirjam A. , Lawrence Feick. (2013), " Receiver Responses to Rewarded Referrals: The Motive Inferences Framework ," Journal of the Academy of Marketing Science , 41 (6), 669 – 82.
Whitler Kimberly A.. (2014), " Why Word of Mouth Marketing Is the Most Important Social Media ," Forbes (July 17) , https://www.forbes.com/sites/kimberlywhitler/2014/07/17/why-word-of-mouth-marketing-is-the-most-important-social-media/?sh=5f9bfc354a8c.
Wicklund Robert A.. (1974), Freedom and Reactance. New York : John Wiley and Sons.
Wu Eugenia C. , Moore Sarah G. , Fitzsimons Gavan J.. (2019), " Wine for the Table: Self-Control, Group Size, and Choice for Self and Others ," Journal of Consumer Research , 46 (3), 508 – 27.
You Ya , Vadakkepatt Gautham G. , Joshi Amit M.. (2015), " A Meta-Analysis of Electronic Word-of-Mouth Elasticity ," Journal of Marketing , 79 (2), 19 – 39.
~~~~~~~~
By Monika Lisjak; Andrea Bonezzi and Derek D. Rucker
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 68- How Physical Stores Enhance Customer Value: The Importance of Product Inspection Depth. By: Zhang, Jonathan Z.; Chang, Chun-Wei; Neslin, Scott A. Journal of Marketing. Mar2022, Vol. 86 Issue 2, p166-185. 20p. 1 Diagram, 6 Charts, 3 Graphs. DOI: 10.1177/00222429211012106.
- Database:
- Business Source Complete
How Physical Stores Enhance Customer Value: The Importance of Product Inspection Depth
The authors investigate the role of the physical store in today's multichannel environment. They posit that one benefit of the store to the retailer is to enhance customer value by providing the physical engagement needed to purchase deep products—products that require ample inspection for customers to make an informed decision. Using a multimethod approach involving a hidden Markov model of transaction data and two experiments, the authors find that buying deep products in the physical store transitions customers to the high-value state more than other product/channel combinations. Findings confirm the hypotheses derived from experiential learning theory. A moderated serial mediation test supports the experiential learning theory–based mechanism for translating physical engagement into customer value: Customers purchase a deep product from the physical store. They reflect on this physical engagement experience, and because it is tangible, concrete, and multisensory, it enables them to develop strong learning about the retailer. This experiential knowledge precipitates repatronage and generalizes to future online purchases in the same category and in adjacent categories, thus contributing to higher customer value. This research suggests that multichannel retailers use a combination of right-channel and right-product strategies for customer development and provides implications for experiential retail designs.
Keywords: customer relationship management; experiential learning; hidden Markov model; multichannel retail; offline store; product inspection depth; sensory marketing
The online channel has assumed a dominant role in many industries, the result of a 15% annual growth in e-commerce (U.S. [52]). In this environment, retailers seem ambivalent about the role of physical stores. Industry surveys show that many consumers still prefer them: "Half of the shoppers surveyed stated that they preferred to shop with online retailers who also operated physical stores" ([47], p. 7; see also [ 7]; [43]). However, many retailers such as Macy's, Walgreens, and Bath & Body Works are closing physical stores ([41]). At the same time, digital-native online retailers such as Amazon, Alibaba, Blue Nile, Warby Parker, Bonobos, Google, and Indochino are opening them ([57]). Given these mixed messages, we find it natural to ask, "What in fact is the value of the physical store?"
To answer this question, we first recognize that consumers buy products, not channels. They must decide which products to buy in which channels. [31] advocate for studying this joint product/channel decision. They maintain that the product and channel choice processes are "intertwined" (p. 320) and that "there is significant academic and managerial motivation for the studying the interrelationships between brand and channel choice" (p. 320). Although, strictly speaking, they discuss channel and brand choice, their message can also be interpreted as calling on academics to research "the consumer's decision of where and what to buy" (p. 329).
Our research draws on this call to action and the seismic shifts in the retail landscape to propose and empirically examine a central hypothesis: physical stores can enhance customer profitability by providing the physical engagement that customers value when purchasing "deep" products. We argue that products differ in the amount of inspection customers need to make a purchase. Some products require relatively "shallow" inspection, where a picture and written description suffice (e.g., a mobile phone charger), whereas other products require "deep" inspection, such as touch and physical interaction (e.g., a shirt). Drawing from experiential learning theory (ELT), our thesis is that customers value physical engagement when buying deep products and that the store provides such physical engagement, creating a favorable learning experience that increases repatronage. The lesson is that the product/channel purchase combination—deep product purchases in-store—develops more profitable customers.
Our research objective is therefore to investigate the following questions: ( 1) Does buying deep products in the physical store[ 5] enhance customer value more than other product/channel combinations? ( 2) If so, what are the implications for the customer's future channel choices?
We adopt a multimethod approach to answer these questions. In Study 1, we analyze customer-level transaction data for 50,387 customers of a large multichannel retail chain that sells outdoor recreation gear, sporting goods, and clothing. We first classify products as "deep" and "shallow" on the basis of the novel concept of product inspection depth that builds on [23] concept of digital and nondigital attributes and [37] findings around the importance of haptic information to consumer experiences. We then use a hidden Markov model (HMM) to examine consumers' product/channel choice dynamics and uncover two latent states: ( 1) a low-value state characterized by lower purchase frequency and profitability and ( 2) a high-value state characterized by higher purchase frequency and higher profitability. We find that customers are more likely to transition to the high-value state and remain there to the extent they have purchased deep products in the physical store.
To replicate these findings and understand the underlying process, we conduct two lab experiments. Study 2 verifies that the deep product/physical store combination produces the highest repatronage intentions. Moderated serial mediation analysis supports the following process proposed by ELT: concrete experience → reflection on physical engagement → hypothesized learning → repatronage. This finding suggests that purchasing deep products in-store provides consumers with concrete, tangible, multisensory experiences that enable them to reflect on and then generate hypothesized learning that encourage them to repatronize the retailer, thus enhancing customer value.
Experiential learning posits that on gaining experience, consumers generalize their learning beyond the contexts they have experienced. Study 3 verifies that once customers have purchased a deep product in a physical store, they are more amenable to purchasing the same as well as adjacent deep products online from the retailer in the future. They thus generalize from the retailer's store to its website and to adjacent deep product categories.
Our findings support the trend of online retailers establishing an offline presence to enhance customer value by providing customers with concrete, tangible, multisensory experiences. This corroborates that many consumers still prefer to buy in-store. A recent survey found that "if given the opportunity, 71% of consumers said they would even prefer to shop at an Amazon store over Amazon.com" ([49]). Another survey revealed that "50% of shoe buyers, 64% of sports equipment buyers, 59% of furniture buyers and 68% of jewelry buyers still prefer the physical store" ([43]). Thus, despite two decades of innovation aimed at making e-commerce more engaging, many consumers still perceive the benefits of buying deep products in physical stores.
Practitioner quotes support consumers' needs for concrete, tangible, multisensory experiences, which facilitate physical engagement, for why online retailers seek physical store presence:
[Alibaba's physical stores are] providing an option for consumers to physically inspect, touch and feel products before purchase. This appears to be the right strategy. The physical store should attract … online shoppers, who want a more human shopping experience (Trefis [51]).
Many stores will be giving customers an experience and providing insight and information. Want new pants? Find your size and the style you like in the store. Looking for a new smartphone or tablet? Try them out at a store and have a clerk walk you through the different features ([ 9]).
With certain products, seeing and feeling makes a difference … even the most elegant descriptions and images can't replace the feel of organic, high thread count cotton sheets ([57]).
These quotes echo our proposition that the physical store provides customers with the engagement they value when purchasing deep products. The fact that online retailers are pursuing an offline presence today validates our findings and supports our physical engagement theme. In turn, our findings support this trend. Our findings also suggest that online retailers who cannot afford the investment required for a physical presence should mimic the physical engagement found in stores and make the digital experience more concrete, tangible, and multisensory.
In summary, we offer three key findings. First, buying deep products in the physical store increases long-term customer value. Second, consumers who purchase deep products in-store are subsequently more likely to buy the same and adjacent deep product categories online. Third, an important underlying mechanism is experiential learning. We also find that direct mail marketing encourages customers to purchase deep products in-store and increases customer value, suggesting that retailers can onboard new customers and revive lapsed customers via the promotion of deep products and in-store purchasing.
These findings make the following contributions. First, we empirically identify an important role of physical stores, which is to enhance customer value by providing the physical engagement needed to purchase deep products. Second, we provide evidence that physical stores fulfill this role through experiential learning. Third, we advance theory by introducing the notion of "product inspection depth" and highlighting the distinction between physical engagement and digital engagement. Finally, we expand multichannel research that has focused on channel choice ([50]; [53]; [58]) by demonstrating that management of the joint product/channel decision is crucial for better understanding customer behavior.
We aim to identify which product/channel purchase combination enhances customer value more than other combinations. We posit that customers value physical engagement when buying deep products. The store provides this engagement, creating a favorable learning experience that increases patronage and customer profitability.
We distinguish products in terms of inspection depth, defined as the degree to which customers examine the product to make an informed purchase decision. Inspection depth can be ordered along a continuum: ( 1) pictures and descriptions are adequate, ( 2) visual inspection of the product is needed, ( 3) touching the product is needed, and ( 4) interaction with the product is needed (e.g., trying on, testing). We refer to products that require less inspection as "shallow products"; those that require more inspection we refer to as "deep products."
The concept of product inspection depth follows from the literature that examines how extensively the customer must examine a product to make an informed purchase. [45] stress the importance of physical inspection in the fashion industry. [14] find the inability to inspect shoes, DVD players, flowers, and food is an impediment to patronizing online stores, but not so for books and toothpaste. [33] suggest that profits for the multichannel firm decrease when consumers find it important to inspect the product before purchase. Product inspection depth is rooted in [30] theory of search and experience goods. Search goods can be evaluated prior to purchase, whereas experience goods need to be consumed to be evaluated. Relatedly, [23], pp. 487–88) advance the concept of digital attributes, which can be "communicated online" versus nondigital attributes, which "can only be evaluated through physical inspection."
Product inspection depth synthesizes these ideas yet differs in important ways. For example, it differs from experience/search in that a deep product does not have to be consumed to be evaluated—it just needs to be inspected. It extends digital attributes, such as appearance, into the physical domain (e.g., "I will try on this clothing to see how it actually looks on me"). Furthermore, it is not confounded with price. For instance, a pair of shoes requires deeper inspection than a (higher-priced) computer, whose specifications can be read off a product description. Inspection depth is particularly relevant to multichannel shopping, where the ability to inspect differs by channel.
To acquire product inspection depth, consumers need to physically examine, inspect, or even interact with the product. That is, the customer has to physically engage with the product. More generally, the literature defines engagement as "a behavioral manifestation toward the brand, beyond purchase" ([54], p. 253). This manifestation can include touching and examining the product, reading descriptions, watching a demonstration, reading reviews, and interacting with a sales representative.
The literature further delineates two forms of engagement: digital and physical ([55]). Digital engagement entails nonphysical actions such as seeing the product on a printed image. Formally, we define "physical engagement" as when the customer goes beyond visual inspection to gain multisensory knowledge of the product (e.g., by touching and using it). [38] link physical touch to object valuation, and [39] link touch to persuasion, suggesting that physical engagement can increase customer satisfaction. The potential for physical engagement differs by channel. The physical versus digital distinction is important because physical stores offer both physical and nonphysical engagement, whereas the online channel only offers nonphysical (digital) engagement.
The product inspection depth consumers acquire through physical engagement defines a shopping experience. What do consumers learn from this experience? This is the bailiwick of experiential learning theory (ELT). David Kolb developed ELT as a synthesis of work by Lewin, Piaget, and others ([15], [16]; [17]; [28]). It is relevant for our purposes because it translates experience into learning. In Kolb's words, "Learning is the process whereby knowledge is created through the transformation of experience" ([16], p. 38), and "Knowledge results from the combination of grasping and transforming experience" ([16], p. 41).
ELT is a process whereby people learn in four recursive stages: experiencing, reflecting, thinking, and acting ([17], p. 194). People first experience something concrete or tangible. They then reflect on the experience. Reflection enables people to hypothesize what they have learned from the experience and the extent to which this learning generalizes beyond the recent experience. People then act (i.e., test this hypothesis when the opportunity arises). The action provides more experience, and the four-stage process repeats.[ 6]
A review of the ELT literature reveals four themes we draw on in forming our hypotheses. First, experiential learning builds on concrete and tangible experiences ([16]). [16], p. 21) notes, "The emphasis is on here-and-now concrete experience," and "Immediate personal experience is the focal point for learning, giving life, texture, and subjective personal meaning to abstract concepts."
Second, experiential learning is a feedback process: Consumers develop hypothesized learning from experience. They use subsequent experiences to test how well these hypotheses generalize, modifying them as needed. "Learning is described as a process whereby concepts are derived from and continuously modified by experience" ([16], p. 26).
Third, experiential learning draws on the five senses. Under the rubric "sensory marketing," researchers have connected experience to the five senses—touch, taste, smell, hearing, and seeing ([18]). Thus, three findings from sensory marketing prove critical to our hypothesis development. First, touch experience influences attitudes. Research has found that touch leads to more confident conclusions ([37]), is more persuasive ([39]), generates affect ([38]), and influences quality judgments ([ 1]). Second, multisensory experiences are more effective in generating learning than are single sensory experiences ([19]). Finally, touch can be an end in itself or a gateway to enhancing other senses, particularly visualization ([35]). Consider the case of buying jewelry and wristwatches. Touching the jewelry or the watch to feel its texture and weight distribution is valuable in itself, but it also enhances visualization, as one can pick it up, view it in natural light from various angles, and try it on to see how it looks and feels. Touching and seeing naturally go together.
Figure 1 integrates product inspection depth, engagement, experiential learning, and customer value. The process starts with a shopping experience, characterized by the type of product the consumer buys (deep vs. shallow) and the purchase channel. This initiates the experiential learning process. Iterating through ELT yields learning that consumers want to confirm and determine if it generalizes to other contexts. These tests take the form of future shopping behaviors, which serve as the experiences that initiate future ELT cycles.
Graph: Figure 1. Experiential learning: translating engagement to future behavior.
Deep products require inspection, touch, and trial. The store provides this physical engagement, and the shopping experience therefore is concrete, tangible, and multisensory—prerequisites for effective experiential learning. These factors in synchrony encourage deeper reflection, stronger hypothesized learning and potential generalizations, and, ultimately, more action.
Importantly, we assume that providing physical engagement for a deep product purchase enables the consumer to make a more informed decision. The experience, therefore, will be positive, encouraging favorable learning that propel the consumer toward repatronage and higher customer value. We therefore hypothesize,
- H1: Purchasing deep products in the physical store is associated with higher future customer value more than any other product/channel purchase combination.
ELT contributes to the translation of deep products/in-store to future customer value. It suggests the following mechanism underlying H1: customers purchase a deep product from the physical store. They reflect on this physical engagement experience, and because it is tangible, concrete, and multisensory, it enables them to develop strong hypotheses of what they learned about the retailer. Because they are satisfied with their purchase, this learning is favorable, precipitating repatronage.
In summary, the mechanism is as follows: customers purchase a deep product in-store → they reflect on the physical engagement experience → they hypothesize favorable learning → they repatronize the retailer. This parallels ELT's process of experience → reflection → hypothesized learning → action. We thus predict:
- H2: Experiential learning contributes to the process by which purchasing deep products in the physical store is associated with higher future customer value.
Two assumptions underlie H1 and H2. First, the customer purchases a satisfying product. It is possible that despite the concrete experience, the customer emerges dissatisfied. Experiential learning is still at play, but the consumer learns that this retailer is not suitable, and customer value declines. Second, the customer's favorable experience spills over to both the retailer and the brand. Thus, H1 and H2 assume the customer emerges from the deep/offline experience with favorable learning that transfers to the retailer.
Importantly, H1 is comparative; it maintains that deep/offline purchasing enhances customer value more than any other product/channel combination for the following reasons. First, the online channel entails only one sense—visual. Thus, the online experience is less concrete, less tangible, and not multisensory, and the consumer learns less. Second, shallow/offline purchasing has the potential to provide a concrete, tangible, and multisensory experience because the store offers this opportunity. However, by definition, shallow products do not require this physical engagement. As a result, the customer does not reflect enough to generate strong hypothesized learning, and repatronage is not enhanced as much.
ELT posits that customers will test the extent to which their learning generalize beyond their experience to date. We consider two types of generalization. First, customers may generalize to a new channel. Consider customers buying a shirt in the physical store, and assume that they hypothesize from this experience that the retailer is a good place to buy shirts. They can then see how well this learning generalizes to another channel by purchasing deep products online from the retailer's website. This lets them test for generalization while taking advantage of the convenience of the website. As elaborated in our discussion of H1 and H2, customers who purchase shallow products in the physical store do not learn enough to encourage them to explore whether to purchase deep products online. In other words, these consumers are less apt to experiment with the website if they need a shirt. We thus hypothesize:
- H3: Purchasing deep products in the physical store is associated with buying deep products online in the future, compared with purchasing shallow products in the physical store.
Second, learning inferred by deep/offline customers can generalize not only to new channels but also to new products. Product generalization can occur because the initial purchase experience allows customers to learn about the retailer's overall product quality and product fit—factors that are especially important for deep products. Obviously, it is easier to generalize to something that is most related to the current context, and we believe the strongest impact of buying a shirt offline will be buying a shirt online in the future. But the generalization could extend to adjacent deep products such as sweaters, coats, and other apparel, and possibly to different product categories altogether. We thus hypothesize:
- H4: Purchasing a particular deep product in the physical store is associated with buying adjacent deep product categories online in the future.
Study 1 uses retail transactional data and an HMM to verify H1 and H3. We test whether the translation of product/channel into customer value is most favorable for the deep/store combination (H1) and whether this combination encourages future usage of the online channel (H3). We conduct two randomized experiments in Studies 2 and 3 to replicate Study 1's findings. In addition, Study 2 tests the hypothesized learning mechanism (H2), and Study 3 tests product generalization (H4).
Our data are from a national retailer that sells outdoor recreation gear, sporting goods, and clothing in 140 retail stores and on its website. The data chronicle customer-level purchase occasions from January 2003 to July 2005. For each purchase occasion, we observe stockkeeping units (SKUs) purchased, price, purchase amount (dollars spent on the entire order), cost of goods sold, channel choice, timing, and returns. We know each customer's zip code and tenure with the retailer.
Our sample contains 50,387 customers buying more than 30,000 SKUs on 585,577 purchase occasions, an average of 12 purchase occasions per customer. Table 1 shows that online purchases make up 11.1% of purchase occasions. Because we are interested in dynamics, we select only customers who had at least two purchase occasions. Of these, 8,391 are "new customers," with 80,751 purchase occasions, acquired after the beginning of the data set. Deep products constitute 54.7% of purchases, with an average spend of $56.21 per purchase occasion. The firm uses direct mail to communicate new styles and special events. On average, customers receive 19.6 direct mail pieces annually. The retailer does not target on the basis of past purchases. Prices and cost of goods sold do not vary between online and offline. Only 5.6% of customers purchased the same SKU more than once, either on the same purchase occasion or over multiple purchase occasions. So there are few instances of rebuying the same product. Half of the customers shop in a single channel; the others shop both in-store and online. Consistent with previous research, multichannel shoppers are more profitable (see Web Appendix Table W1.1).
Graph
Table 1. Descriptive Statistics per Customer.
| Mean | SD | Min | Mdn | Max |
|---|
| Total number of purchase occasions | 11.62 | 8.67 | 5 | 9 | 128 |
| Tenure (weeks) | 498.23 | 65.12 | 4.33 | 507 | 708 |
| Total purchase amount per purchase occasion | $102.81 | $60.13 | $2.70 | $92 | $913 |
| Purchase amount of deep products per purchase occasion | $56.21 | $42.64 | $0 | $48 | $757 |
| Purchase amount of shallow products per purchase occasion | $46.60 | 42.47 | $0 | $37 | $865 |
| Interpurchase time between purchase occasions (days) | 75.37 | 36.00 | 2 | 71 | 186 |
| Number of direct mailings received within 7 days of purchase | .46 | .26 | 0 | 0 | 1 |
| Number of direct mailings received within 14 days of purchase | 1.07 | .43 | 0 | 1 | 2 |
| Annual number of direct mailings received | 19.6 | 5.21 | 12 | 16 | 26 |
| Percentage of purchases in-store | 88.87% | .19 | 0 | 1 | 1 |
| Percentage of purchases online | 11.13% | .19 | 0 | 0 | 1 |
The retailer categorizes the 30,000 SKUs into ten "specialties," such as camping, travel, cycling, snow sports, and clothing, followed by 373 "classes" or categories, such as jackets, pants, and shorts, and finally specific SKUs. For details, see Web Appendix Table W2.1.
Three independent judges rated each of the 373 product categories on inspection depth as well as digital or nondigital (Web Appendix W3) on a scale of 1 to 7. Intercoder reliability for inspection depth is.92. This suggests that the inspection depth concept is robust across coders and covers dimensions such as touch and interaction. The correlation between inspection and digital/nondigital ratings is.77, thus offering discriminant validity from digital/nondigital.[ 7] Not surprisingly, clothing and footwear generally have high ratings, but there is much variation within a specialty (Web Appendix Figure W3.1). This suggests that specialties are not perfect indicators of inspection depth. For example, within footwear, hiking boots are rated 7, women's sandals are rated 4, and insoles are rated 2 (on a 7-point scale). Deep products are not necessarily more expensive than shallow products; the correlation between price and inspection depth is.12 and insignificant. We categorize products as deep or shallow using a median split. This enables us to model purchase amounts of each type on each purchase occasion.[ 8]
Our key hypothesis is that buying deep products in the physical store increases future customer value. New customers provide model-free evidence for this. Table 2 shows four cohorts of new customers defined by when they are acquired. Profits for the one-year period after acquisition differ depending on the first product/channel choices; buying deep products offline as the first purchase yields the highest profit, consistent with H1.
Graph
Table 2. Model-Free Evidence: First Product/Channel Choice and One-Year Profit.
| Customer Cohort | First Purchase Deep Product Offline | First Purchase Deep Product Online | First Purchase Shallow Product Offline | First Purchase Shallow Product Online |
|---|
| Cohort 1: Jan.–Apr., 2003 | $77.75 | $69.51 | $65.29 | $62.75 |
| Cohort 2: May–Aug., 2003 | $82.38 | $72.98 | $71.54 | $59.40 |
| Cohort 3: Sep. –Dec., 2003 | $77.30 | $72.78 | $69.19 | $62.71 |
| Cohort 4: Jan.–Apr., 2004 | $82.60 | $74.77 | $75.63 | $73.08 |
Figure 2 illustrates dynamics. It depicts product and channel choices for new customers' first purchase and the same set of new customers on their eighth purchase. We see that 5.76% of first purchases are deep products bought online. This increases to 8.45% by the eighth purchase. This finding is consistent with H3 and shows that new customers' buying patterns evolve and alleviates the concern that new customers' purchase patterns are set before the retailer acquires them.
Graph: Figure 2. Model-free evidence of product/channel choice evolution.
It is still possible that new customers self-select into the relationship with the retailer. We address this in Study 1 by ( 1) developing a model designed to flexibly detect dynamics and control for unobserved customer heterogeneity, ( 2) separately analyzing new and existing customers via robustness checks, and ( 3) conducting a propensity scoring analysis. Studies 2 and 3 alleviate the self-selection concern using random treatment assignment in experiments.
We use a multivariate HMM to study customers' joint decisions for channel choice, purchase amount, and interpurchase time. HMMs are often employed to study the dynamics of customer–firm relationships (e.g., [21]; [24]; [26]; [29]; [32]; [46]; [61]; [62]). HMMs incorporate experiential learning in that they include dynamics and feedback and model changes in behavior arising from experience (reflected in "latent states"). HMMs also identify the drivers of these dynamics by studying customer transitions between latent states. A simpler model such as regression would have difficulty capturing these dynamics and rich insights. Furthermore, HMMs can distinguish temporal dynamics from customer heterogeneity. We will model time-invariant customer heterogeneity as well as dynamics.
We use the HMM to discern when the customer is in a high- versus low-value state and predict transitions between these latent states by descriptors such as customers' previous purchases of deep versus shallow products and their use of offline versus online channels. In this way, the HMM provides tests for H1 and H3.
Specifically, we model a customer's purchase occasion by four interrelated decisions: ( 1) channel choice, ( 2) purchase amount (in dollars) of deep products, ( 3) purchase amount (in dollars) of shallow products, and ( 4) purchase timing (in terms of interpurchase time). These four dependent variables not only paint a rich and multifaceted picture of customer behaviors beyond overall purchase amount but also enable us to calculate customer value within a particular time frame. We incorporate covariates in customers' utility functions and thus predict customers' decisions at each purchase occasion. We include covariates in transition functions to predict how customers migrate between latent states.
Because HMMs are popular in marketing, we detail the specific components of our HMM to Web Appendix W4. Next, we highlight the covariates in the specification.
H1 and H3 both predict that previous channel choices influence future customer value. We include the customer's cumulative number of channel choices prior to purchase occasion j, offline_choices(j − 1) and online_choices(j − 1),[ 9] for the store (offline) and website (online), respectively. This is consistent with [22], [48], and [59].[10]
Direct mail is the only firm-initiated marketing activity in the data and did not promote specific channels. Discussion with management revealed that communications were not customer-targeted, and we test and confirm that there is no endogenous relationship between customers' past purchase behavior and the likelihood of receiving direct mail. Thus, it reflects a baseline advertising effect. The variable marketingj equals the number of mailings the customer received within 30 days prior to purchase occasion j. We use a squared term, , to capture decreasing returns.[11]
The variable holidayj indicates whether purchase occasion j occurs within two days preceding the following gift-giving holidays: Mother's Day, Father's Day, Valentine's Day, Christmas Eve, and New Year's Day. Because last-minute online shopping risks that a gift will not arrive on time, people may be more likely to shop offline when it is very close to these holidays.
We measure customer spend (in dollars) on deep and shallow products on previous purchase occasion j − 1 (deep_amount[j − 1] and shallow_amount[j − 1]). Both H1 and H3 predict that the type of product purchased is crucial for determining future customer value.
Tenure (tenurej) denotes how long the customer has been purchasing from the retailer as of purchase occasion j. It represents the length of the relationship.
We calculate cumulative returns prior to the current purchase occasion (in dollars) and distinguish between deep and shallow product returns (deep_returns[j − 1] and shallow_returns[j − 1]). [40] find that product returns enhance customer relationships.
We define interpurchase_timej as the length of time between the previous and current purchase occasion. Shorter interpurchase time indicates more frequent purchases and, thus, a more valuable customer. Longer times could indicate lapses in loyalty or changes in lifestyle, which could influence subsequent channel or product choice.
HMMs model latent states and estimate "transition functions" that predict how the customer migrates in and out of these states over time. HMMs characterize each state by its own set of utility functions—in our case, one for each of the four customer decisions we model. Following HMM conventions, we include covariates expected to have an immediate impact in the utility functions and covariates expected to have a long-term impact in the transition functions.
Accordingly, we include previous channel/product decisions—offline_choices(j − 1), online_ choices(j − 1), deep_amount(j − 1), and shallow_amount(j − 1)—in all four utility functions to reflect ELT's dictum that customers test what they learn from experience by taking action—all four behaviors are actions. We include marketingj and in the utility functions because direct mail can have an immediate reminder impact. We expect channel choice to be driven by proximity to a holiday. Thus, holidayj enters in the utility function for channel choice.
We also include previous channel/product decisions (offline_choices[j − 1], online_choices[j − 1], deep_amount[j – 1], shallow_amount[j − 1]) in the transition functions to test our hypotheses that these influence future customer value. Direct mail is advertising that can have long-term effects, so we include marketingj and in the transition equations. Drawing on [40], we include returns (deep_returns(j − 1] and shallow_returns[j − 1]) in the transition functions. As noted previously, long customer tenure and short interpurchase times could proxy for a strong long-term relationship, so we include tenurei and interpurchase_timej in the transition functions.
Note that we use lagged variables in the utility and transition functions to capture dynamics (how previous decisions drive current decisions and state transitions). We are particularly interested in how previous channel and product choices determine future customer value.
We do not include current prices as a covariate. From a theoretical standpoint, including current prices in the purchase timing and channel choice models would make the difficult-to-support assumption that customers make these decisions based on prices they do not observe until after they make those decisions. As a robustness check, we included monthly fixed effects and a monthly price index of top 30 best-selling items as covariates in the utility equations. Results, most importantly those pertaining to H1 and H3, were substantively the same.[12]
Capturing customer heterogeneity is crucial for distinguishing temporal dynamics from time-invariant customer heterogeneity ([13]). We do this by adding latent class segmentation to the HMM. This allows the coefficients for the transition functions and the four utility functions to vary across segments.
We use Markov chain Monte Carlo (MCMC) methods for estimation and use the adaptive Metropolis procedure ([ 3]) to improve mixing and convergence. We use the first 24 months of data for training and the last seven months for testing. We obtain our estimates from the last 50,000 draws from an overall MCMC run of 200,000 iterations. We assessed convergence by monitoring the time-series of the MCMC draws.
To determine the number of HMM states and the number of segments, we consider the in-sample log-marginal density, deviance information criterion, and predictive log-likelihood on the validation sample. Drawing on these criteria, we find that a two-state, two-segment HMM exhibits the best performance. It captures dynamics and heterogeneity while keeping model complexity in check. Therefore, we adopt this model. Web Appendix Table W4.1 shows these criteria for various permutations of the HMM. The table shows that the two-state, two-segment HMM is better than models that do not include dynamics, do not include heterogeneity, or include heterogeneity but as a continuum rather than discrete segments.[13]
We used methods described in [27] to infer each customer's state membership at each purchase occasion. We assign customers to the state with the highest probability to which they belong (in our two-state case, greater than.5). Table 3 shows average customer characteristics for each state. The interpretation is clear: customers in state 1 have longer interpurchase times (i.e., buy less frequently) and generate less revenue and profit. We label state 1 "low-value" and state 2 "high-value."
Graph
Table 3. Description of the Two HMM States.
| State 1Low Value | State 2High Value |
|---|
| Percentage who are new customers | 62% | 38% |
| Mean spend per purchase occasion | $91 | $127 |
| Mean interpurchase time (days) | 59 | 46 |
| Percentage of deep products purchased in-store | 88% | 71% |
| Percentage of purchases online | 11.92% | 14.27% |
| Revenue per customer | $901 | $1,078 |
| Profit per customer | $138 | $169 |
We assign customers to the latent segment to which they have the highest probability of belonging. Table 4, Panel A, shows that segment 1 is more profitable than segment 2, has a more balanced mix between in-store and online buying, transitions more quickly to the high-value state, and lives closer to the retail store. The channel mix and higher profitability suggest that segment 1 is the "multichannel segment." Table 4, Panel B, shows that customers in this segment buy more deep products, especially when they move to the high-value state. Table 4, Panel C, shows that customers in this segment are more likely to migrate to the high-value state and remain once they get there. Segment 2 focuses on offline—92% of these customers' purchase occasions are in-store (95% when they are in the low-value state and 76% when they are in the high-value state). Table 4, Panel C, further shows that segment 2 customers are less likely to transition to the high-value state. Accordingly, we label segment 1 "multichannel" and segment 2 "offline."
Graph
Table 4. Segment Descriptions and Migration Probabilities.
| A: Segment Descriptions | Multichannel | Offline |
|---|
| Percentage of customers | 23% | 77% |
| Number of years customer of the retailer | 1.5 | 7.0 |
| Distance from closest stores (miles) | 23.57 | 43.52 |
| Percentage of purchases in-store | 68% | 92% |
| Percentage of $ spent in-store | 43% | 49% |
| Number of purchase occasions until migrate to high-value state | 4.8 | 7.4 |
| Revenue per customer | $1,281 | $736 |
| Profit per customer | $209 | $101 |
| B: Segment Descriptions by State |
| Multichannel | Offline |
| Low Value | High Value | Low Value | High Value |
| $ spent per purchase occasion | $103 | $139 | $81 | $119 |
| Deep $ spent per purchase occasion | $57 | $75 | $40 | $59 |
| Shallow $ spent per purchase occasion | $46 | $64 | $41 | $60 |
| Mean interpurchase time (days) | 51 | 44 | 62 | 51 |
| Likelihood of purchase online (conditional on buying a deep product) | 15% | 43% | 11% | 26% |
| Revenue per customer | $975 | $1,245 | $710 | $993 |
| Likelihood of online purchase occasion | 27% | 87% | 5% | 24% |
| Profit per customer | $155 | $228 | $84 | $144 |
| Mean product depth when buying online (1–7 scale) | 2.4 | 4.3 | 2.1 | 3.9 |
| Mean product depth across all purchase occasions (1–7 scale) | 3.9 | 4.6 | 3.8 | 4.2 |
| C: Probability of Migrating from State to State, from Purchase Occasion j to j +1, by Segment |
| Multichannel Segment | Offline Segment |
| Low Value(j + 1) | High Value (j + 1) | Low Value(j + 1) | High Value (j + 1) |
| Low value (j) | 46.23% | 53.77% | 72.05% | 27.95% |
| High value (j) | 13.22% | 86.78% | 54.06% | 45.94% |
Table 5 shows the parameter estimates for the transition functions. A positive coefficient for the low-value state means that customers in that state are more likely to move from low value to high value as the covariate increases; a positive coefficient for customers in the high-value state means they are more likely to stay high value. In both segments, previous choices of the offline channels drive customers to high value or keep them there. In contrast, online choices decrease the likelihood that low-value customers move to the high-value state, as well as the chance they remain high value. Marketing drives customers to the high-value state and keeps them there. Purchasing deep products drives customers to the high-value state and keeps them there, whereas purchasing shallow products drives customers to the low-value state and keeps them there. Longer tenure transitions customers to high value, whereas longer interpurchase times move customers to low value. Importantly, the transition equations reveal plenty of dynamics that vary by segment.
Graph
Table 5. HMM Transition Functions Parameter Estimates.
| Multichannel Segment | Offline Segment |
|---|
| Low-Value State | High-Value State | Low-Value State | High-Value State |
|---|
| Parameter | Mean | SD | Mean | SD | Mean | SD | Mean | SD |
|---|
| Intercept | .4034 | (.008) | .8160 | (.009) | −.8244 | (.007) | .1486 | (.014) |
| Offline_choices(j − 1) | .0690 | (.016) | .2923 | (.008) | .5884 | (.011) | .9726 | (.007) |
| Online_choices(j − 1) | −1.9515 | (.004) | −.0708 | (.013) | −.2920 | (.004) | −.0723 | (.019) |
| Marketingj | .4038 | (.004) | .0689 | (.005) | .2608 | (.007) | .2139 | (.004) |
| Marketingj2 | −.0635 | (.009) | −.0111 | (.009) | −.0397 | (.015) | −.0294 | (.003) |
| Deep_amount(j − 1) | .2374 | (.004) | .3964 | (.005) | .4023 | (.010) | .2721 | (.005) |
| Shallow_amount(j − 1) | −.5612 | (.006) | −.2416 | (.004) | .0286 | (.012) | .0888 | (.007) |
| Tenure(j − 1) | .1607 | (.005) | .1234 | (.008) | .4690 | (.009) | .8466 | (.007) |
| Interpurchase timej | −.4391 | (.007) | −.2185 | (.006) | −.4717 | (.007) | −.3881 | (.005) |
| Deep_return(j − 1) | 1.1861 | (.007) | −.1345 | (.011) | −.3357 | (.004) | .1167 | (.014) |
| Shallow_return(j − 1) | .8613 | (.004) | .3316 | (.006) | −.3077 | (.009) | −.4951 | (.006) |
| Initial State and Segment Probabilities |
| Low-value state | .588 | | .412 | | .267 | | .733 | |
| Segment | .222 | | | | .778 | | | |
1 Notes: Positive coefficient means variable increases transition among customers in the low-value state and increases the probability of staying in high-value if the customer is already there.
Table 6 contains parameter estimates for the four decision models (i.e., the utility functions) for both segments and both states. We also calculated marginal effects (Web Appendix W6), which are consistent with the estimates in Table 6.
Graph
Table 6. Parameter Estimates for the Four Decision Utility Functions.
| A: Decision to Choose Offline Rather than Online Channela |
|---|
| Multichannel Segment | Offline Segment |
|---|
| Low-Value State | High-Value State | Low-Value State | High-Value State |
|---|
| Parameter | Mean | SD | Mean | SD | Mean | SD | Mean | SD |
|---|
| Intercept | 1.1857 | (.010) | .3884 | (.007) | 2.2406 | (.027) | 1.3280 | (.027) |
| Offline_choices(j − 1) | .8685 | (.003) | −.3282 | (.009) | .1574 | (.006) | 1.1498 | (.009) |
| Online_choices(j − 1) | .7893 | (.010) | .1641 | (.007) | .9579 | (.006) | .0561 | (.011) |
| Marketingj | .8612 | (.015) | −.5749 | (.009) | .7030 | (.007) | .1705 | (.010) |
| Marketingj2 | .0244 | (.012) | .0552 | (.007) | −.1246 | (.015) | −.0194 | (.011) |
| Holidayj | .5220 | (.004) | .1769 | (.006) | .5449 | (.011) | .6596 | (.004) |
| Deep_amount(j − 1) | −.8965 | (.005) | −.7168 | (.008) | −1.0957 | (.021) | .0592 | (.006) |
| Shallow_amount(j − 1) | .7063 | (.009) | −.5750 | (.008) | .4369 | (.005) | .1228 | (.012) |
| B: Deep Product Purchase Amount |
| Multichannel Segment | Offline Segment |
| Low-Value State | High-Value State | Low-Value State | High-Value State |
| Parameter | Mean | SD | Mean | SD | Mean | SD | Mean | SD |
| Intercept | −.3086 | (.010) | .5059 | (.003) | .0756 | (.004) | 1.7441 | (.026) |
| Offline_choices(j − 1) | .7237 | (.012) | .0949 | (.005) | .2403 | (.008) | .1535 | (.008) |
| Online_choices(j − 1) | −.1664 | (.011) | −.0918 | (.011) | −.3241 | (.006) | −.0253 | (.008) |
| Marketingj | .8412 | (.006) | .8029 | (.006) | .3594 | (.004) | .2003 | (.012) |
| Marketingj2 | −.1373 | (.005) | −.1295 | (.011) | −.0921 | (.015) | −.0261 | (.007) |
| Deep_amount(j − 1) | .7765 | (.021) | .2501 | (.008) | .7159 | (.008) | .2423 | (.004) |
| Shallow_amount(j − 1) | .0179 | (.009) | −.0104 | (.006) | .1602 | (.007) | −.0056 | (.006) |
| C: Shallow Product Purchase Amount |
| Multichannel Segment | Offline Segment |
| Low-Value State | High-Value State | Low-Value State | High-Value State |
| Parameter | Mean | SD | Mean | SD | Mean | SD | Mean | SD |
| Intercept | −.7080 | (.012) | .9356 | (.007) | .5469 | (.010) | 1.7792 | (.010) |
| Offline_choices(j − 1) | .5049 | (.007) | .3081 | (.017) | 1.5740 | (.007) | .1758 | (.008) |
| Online_choices(j − 1) | .1751 | (.008) | .0222 | (.020) | .2712 | (.010) | .1758 | (.008) |
| Marketingj | .1060 | (.008) | .0424 | (.016) | .6788 | (.008) | .0100 | (.007) |
| Marketingj2 | −.0531 | (.009) | −.515 | (.009) | −.0801 | (.015) | −.0092 | (.003) |
| Deep_amount(j − 1) | 1.6638 | (.007) | .2450 | (.003) | −.0315 | (.016) | .0034 | (.003) |
| Shallow_amount(j − 1) | .4455 | (.010) | .0668 | (.011) | .5239 | (.011) | .2516 | (.006) |
| D: Interpurchase Time Between Purchase Occasionsb |
| Multichannel Segment | Offline Segment |
| Low-Value State | High-Value State | Low-Value State | High-Value State |
| Parameter | Mean | SD | Mean | SD | Mean | SD | Mean | SD |
| Intercept | −.0228 | (.008) | .2083 | (.003) | −.1176 | (.007) | .1181 | (.010) |
| Offline_choices(j − 1) | .1623 | (.011) | .0786 | (.005) | .2700 | (.005) | 1.1314 | (.054) |
| Online_choices(j − 1) | −.1899 | (.011) | .4226 | (.004) | −.1683 | (03015) | .2242 | (.016) |
| Marketingj | .1003 | (.012) | .2382 | (.007) | .2826 | (.011) | .0271 | (.009) |
| Marketingj2 | −.0041 | (.009) | −.0388 | (.010) | −.0554 | (.013) | −.0096 | (.005) |
| Deep_amount(j − 1) | .2784 | (.006) | .0515 | (.010) | .0242 | (.006) | .0147 | (.003) |
| Shallow_amount(j − 1) | −.7165 | (.014) | −.4262 | (.006) | .1064 | (.006) | .0243 | (.003) |
- 2 aPositive coefficient means variable increases likelihood of choosing offline channel.
- 3 bPositive coefficient means variable decreases interpurchase time.
Table 6, Panel A, displays the utility functions for the channel choice decision. The offline and online previous choice coefficients are mostly positive, suggesting that previous purchase in either channel encourages customers to buy in-store for their next purchase. Interestingly, higher previous deep product purchases mostly encourage customers to buy online, whereas shallow product purchases encourage them to buy offline. Marketing primarily stimulates offline purchases, with the exception of multichannel/high-value purchases. Holiday shopping more likely takes place in the physical store, as expected. Overall, we find strong effects of previous channel choice, previous spending by product type, and marketing.
Table 6, Panels B and C, show the utility functions for deep and shallow purchase amount. In general, previous offline purchase increases both deep and shallow spend, whereas previous online purchase has the opposite impact. Marketing increases deep and shallow spend, with a stronger impact on deep products. Previous deep product spend begets higher spend for both deep and shallow products. Shallow has the same direction of impact, albeit relatively smaller. Table 6, Panel D, shows that, for all segments and states, previous offline purchase decreases interpurchase time, possibly because the customer is more satisfied and thus buys again sooner.
Our model relies on temporal precedence, the association between previous and subsequent decisions, to support a causal interpretation of the results. As we show next, these estimated dynamics suggest that buying deep products in the physical store creates higher customer value and encourages customers to buy online in the future. However, as with any model of field data, we cannot claim the model unequivocally establishes causation. This is one reason we rely not only on Study 1 but also on controlled experiments that use randomization.
H1 hypothesizes that purchasing deep products offline is more likely to transition the customer to the high-value state than any other product/channel combination. The transition function estimates in Table 5 support this. Coefficients for deep_amount(j − 1) (row 7) and offline_choices(j − 1) (row 3) are positive in all four transition functions. Coefficients for online_ choices(j − 1) (row 4) are negative, so online purchases make it less likely that the customer will transition to high value. The coefficients for shallow_amount(j − 1) (row 8) are negative—or, at best in the offline segment, positive but far smaller in magnitude than the coefficients for deep_amount(j − 1). Overall, we find that buying deep products offline is most positively associated with transitioning to the high-value state (i.e., to higher customer value). Study 1 thus supports H1.
H3 proposes that customers who buy deep products offline are more likely to purchase deep products online in the future. We have shown that deep/offline purchasing drives customers to the high-value state. Table 3 then shows that these high-value customers purchase a higher percentage of their deep products online (vs. in-store) than do the low-value customers (29% vs. 11%, row 4). Table 4, Panel B, further illustrates this at the segment level: high-value multichannel customers make 43% of their deep product purchases online, compared with 15% if they are low value. The same holds true for the offline segment: these customers purchase 26% of their deep products online if they are in the high-value state, compared with 11% if they are in the low-value state. Study 1 thus also supports H3.
We ran the model only on existing customers (41,996 customers, 504,826 purchase occasions). The substantive results were very similar to those for the full data set, indicating that the evolution of customer behavior based on channel/product experiences exists for both new and existing customers.
We tested whether the results in Table 2 are due to self-selection—that is, that new customers who start by purchasing deep products in-store already prefer the retailer. We conducted propensity score matching with new customers whose first purchase is deep products in-store as the treatment group and all other new customers as potential controls. We matched on ( 1) demographic variables extracted from each customer's zip code and ( 2) variables calculated using data from the first two months after the initial purchase. These included distance to store, city versus rural, average prices paid, the gender category of the products purchased, purchase of children's products, number of categories purchased per visit, and average interpurchase time. We then calculated subsequent profit, excluding those first two months.
The average treatment effect (incremental value among new customers with deep/in-store as the first purchase) is +$132.74, suggesting that purchasing deep products in-store generates higher long-term customer value (Web Appendix W7). The Rosenbaum test ([44]) states that our treatment effect is significant even if the impact of an unobserved covariate were to increase the odds ratio of buying deep products offline by 50% (Γ = 1.5; see Web Appendix Tables W7.2 and W7.3).
We ran the model with products classified as digital/nondigital instead of deep/shallow. The same substantive results hold for the digital/nondigital classification, with slightly worse model fit and prediction compared with the deep/shallow classification (deviance information criterion = 796,253 for digital/nondigital vs. 792,594 for deep/shallow, predictive likelihood = −198,237 for digital/nondigital vs. −194,899 for deep/shallow). This suggests that the deep/shallow product categorization yields a better-fitting model but similar findings to a "digital/nondigital" categorization. It also suggests that the sensory-rich inspection depth concept is particularly relevant in multichannel environments.
Our analysis categorizes deep and shallow products using a median split. We examined how much our results would change if we used different split thresholds. We reran the model using thresholds ranging from 30% to 70% (i.e., from a 30th percentile rating used to classify a product as deep up to a 70th percentile threshold). Web Appendix W8 indicates that the 50/50 (and 60/40) splits provide the best fit and performance.[14] In addition, our substantive results hold up between 30% or 70% thresholds, suggesting that the results are robust within a reasonable range of rules for classifying products as deep versus shallow. Finally, the predictive likelihood for the 30% threshold is better than that of 70% threshold, suggesting that it is safer to classify shallow as deep than deep as shallow.
The positive coefficient for marketingj in Table 5 (row 5) suggests that direct mail marketing moves customers to the higher-value state or keeps them there if they are already in a high-value state. A reasonable strategy is to target marketing to increase the probability that customers are in the high-value state. The question is, which customers should be targeted on the basis of their current state, segment membership, and previous product/channel choice?
We conducted a simulation to investigate this question. Details are in Web Appendix W10. We used model parameters to simulate purchase behavior over a 30-month horizon. In the base case, marketing is set so that each customer receives two direct mail pieces per month. In the "+1" case, we increased this to three per month. One could use dynamic programming to optimize targeting, but our purpose is simply to demonstrate the potential of targeting.
Profits in the base case were $127.36 per customer ($131.21 under the +1 strategy).[15] We found that the +1 strategy increased the probability of transitioning to or staying in the high-value state, "Prob(Hi)," for all customers except multichannel customers who currently are in a high-value state. They already have a high probability of staying high value (86.78%), so there is not much to gain by increasing marketing. A key result is that the gain in Prob(Hi) is largest for customers who just bought shallow products. For example, we found the gain for offline-segment low-value customers who just bought shallow offline is 29.20% − 24.67% = 4.53%, while the gain for those who just bought deep offline is 36.39% − 33.27% = 3.12%. This finding is consistent with H1: deep offline purchases naturally boost customers to a high-value state, so they have less need for marketing.
H1, supported by the HMM, proposes that one "sweet spot" for generating future customer value is for the customer to purchase deep products in the physical store. Study 2 employs a lab experiment to replicate H1 and test H2, the ELT mechanism we propose underlies it: deep product purchased in-store → physical engagement → favorable learning → repatronize the retailer.
Our sample is 411 Amazon Mechanical Turk subjects. The average online and store patronage experience, age, and gender are statistically equal across treatment groups (details in Web Appendix Table W11.1).
We use a 2 (deep vs. shallow product) × 2 (physical store vs. online) between-subjects design with random assignment to treatment. We used "sports shirt" for the deep product and "portable cell phone charger" for the shallow product.
The survey (Web Appendix W11) instructed subjects that there is "a new sports and outdoor-gear retailer in town" and that "you have never shopped at this retailer." We then asked, "Now, imagine that you shop at this retailer for the first time. You visit its physical store (website) and buy a sports shirt (portable cell phone charger) that costs $40. Please take a few minutes to describe in detail the specific steps you would have taken to purchase the sports shirt (portable cell phone charger) in this new physical store (on this new website)." We provided a text box for subjects to write a description of the steps they would have undertaken.
We then asked subjects to state their intention to shop at this retailer again, using the [ 4] three-item repatronage scale (e.g., "I would be willing to buy from this retailer again in the future" [1 = "Strongly disagree," and 7 = "Strongly agree"]). We asked subjects to rate how much they thought they would have learned about the retailer's "product offerings and quality, as a result of this experience," on a seven-point scale. This was followed by a question asking subjects to rate the product they bought on product inspection depth (1 = "Picture and description would be adequate," 2 = "Visual inspection of actual product needed," 3 = "Touch of product needed," and 4 = "Interaction of the product needed [e.g., trying on, testing the features]"). This last step enables us to verify the deep versus shallow product manipulation.
We begin with the product manipulation check. Results indicate that the sports shirt attained statistically higher means on inspection depth relative to the portable cell phone charger (M = 2.42 vs. M = 1.64; t(409) = 6.88, p < .001).
Figure 3 shows mean repatronage intentions by treatment. The 2 × 2 interaction analysis using analysis of variance demonstrates that both deep product (F( 1, 407) = 6.94, p < .01) and store condition (F( 1, 407) = 6.18, p < .05) positively contribute to higher repatronage. Central to our prediction, the deep × store interaction also positively contributes to higher repatronage (F( 1, 407) = 4.82, p < .05). A planned contrast indicates the deep/store treatment clearly evoked the highest repatronage intentions among the three other conditions (Mdeep/store = 5.38 vs. Mdeep/online = 4.88, Mshallow/store = 4.86, and Mdeep/online = 4.83; F( 3, 407) = 6.09, p < .001), consistent with H1. Web Appendix Tables 11.2A–D provide additional details of the analysis of variance and contrast tests.
Graph: Figure 3. The effect of first product/channel purchase combination on repatronage (Study 2).
We measured the extent to which subjects reflected on the physical engagement component of the experience by analyzing how they articulated the experience in their written descriptions. We recruited two research assistants blind to the research agenda to rate a physical engagement variable, "try_touch," from each respondent's description. We instructed the research assistants to read each description of the customers' shopping experiences. We told the research assistants that some shopping experiences involve elements of touching, trying on, and feeling and instructed them to code try_touch as 1 if the description contains words, synonyms, or themes related to "try," "touch," or "feel"; alternatively, we instructed them to code try_touch as 0 in the absence of these themes. Intercoder correlation was.92.
The following are examples of subjects' descriptions that clearly suggest physical engagement:
I would look at the size of it. I would feel its texture. I would test it out. I would see how it would look on me. I would see if my favorite team is on the shirt.
I would go into the store and do quite a bit of browsing first. I would allow myself extra time in this store as it is my first time going. I would familiarize myself with the brand and touch everything to test out its quality.
I would go into the store and browse shirts. I might try on the shirt before buying it, unless I was sure it would fit and/or I didn't want to spend extra time. Then I'd buy it.
We tested the proposed ELT mechanism using try_touch to measure physical engagement. We used Preacher and Hayes's PROCESS Model 83 ([12]) for moderated serial mediation, which combines serial mediation (Model 6) and moderated mediation (Model 7). Following our theory, we used store versus online as the "X variable" moderated by deep versus shallow product (the "W variable").[16] PROCESS generated estimates and standard errors via bootstrapped sampling with 5,000 iterations. The ELT-based mechanism we propose (deep/store purchase → physical engagement → learning → repatronize retailer) is the "indirect effect" reflecting the mediating role of physical engagement and learning on the relationship between a deep/store purchase and repatronage intentions.
We first conducted several preliminary analyses to demonstrate the value of moderation and mediation. We first regressed store on repatronage, which yielded a significant direct effect (bstore = .273, p = .013). Then, we added the deep/shallow moderator to this model, yielding a significant interaction effect of deep × store (bdeep × store = .472, p = .029) but rendering the main effect of store insignificant (bstore = .031, p = .84; bdeep = .047, p = .76). Similarly, the direct effect of store becomes insignificant once try_touch and learning are added as regressors (bstore = −.143, p = .146; btry_touch = .425, p < .001; blearning = .442, p < .001).
The results of Model 83 regarding moderated serial mediation show that the direct effect of store on repatronage is insignificant (−.143, 95% confidence interval [CI] = [−.336,.0501]). The index of moderated mediation, however, is.188 (95% CI = [.100,.293]), supporting the moderating effect of deep versus shallow product on the impact of the store versus online on repatronage. For the deep product, the conditional indirect effect for store on patronage is.219 (95% CI = [.122,.332]); for the shallow product, it is significantly weaker at.031 (95% CI = [.011,.059]). The results suggest that both the proposed moderation and mediation are at work and that store purchase increases patronage through physical engagement and learning, more so when purchasing a deep product. This confirms H2's prediction that ELT contributes to the mechanism translating deep/store purchasing into repatronage.
We conducted two robustness checks to reinforce this analysis. First, recall that the deep/store condition stands out and the other three essentially are equal. We therefore created a deep/store dummy variable equal to 1 if the subject was in the deep/store condition and 0 otherwise. Although regressing deep/store on repatronage yields a significant direct effect (b = .524, p < .001), the direct effect becomes insignificant after we account for the proposed mediation. Serial mediation analysis (PROCESS Model 6) yields a significant indirect effect: deep/store → try_touch → learning → repatronage. Details are in Web Appendix W11 (Table W11.3).
Second, we reran Models 83 and 6, switching the order of try_touch and learning, estimating an alternative process: store → learning (moderated by deep product) → try_touch → repatronage for Model 83 and deep/store → learning → try_touch → repatronage for Model 6. While the global model fit and the total indirect effects are unsurprisingly the same across both orderings ([42]), the indirect effects of serial mediation are quite different. For instance, our proposed ordering yields.1882 (SE = .048) for the deep products' index of moderated mediation in Model 83; the alternative ordering yields an index of.0138 (SE = .009). Model 6 highlights this difference more saliently. Whereas the proposed ordering yields a serial mediation indirect effect of.1258 (SE = .03), or 30% of the total indirect effect, the alternative ordering yields the serial mediation indirect effect of.0154 (SE = .006), which translates to 3.7% of the total indirect effect. We acknowledge that [42], p. 698) cautions against trying to infer the correct mediation order by the aforementioned tests. He advocates that the ordering be based on "strong evidence from logic, theory, and prior research that the hypothesized casual direction is more plausible than indicated alternatives" ([42], p. 697). We believe the application of ELT we used to specify the try_touch → learning ordering satisfies this requirement.
In summary, Study 2 replicates Study 1's finding that the deep/store purchase combination produces the highest repatronage, in support of H1. It also tests the ELT mechanism as proposed by H2. The moderated serial mediation results, and the robustness checks, are consistent with the ELT mechanism.
Study 3 tests our predictions related to generalization of learning. We aim to replicate the HMM's support for H3, which predicts that purchasing deep products in-store increases the likelihood of purchasing deep products online, compared with purchasing shallow products in-store. Further, this study tests H4, which proposes that the impact of purchasing a specific deep product in-store generalizes to related, adjacent deep products. As noted previously, generalization is an important component of experiential learning and could be very powerful for retailers. It demonstrates the temporal interplay between offline and online channels and suggests that deep/store purchase of a particular product "spills over" to higher likelihood of purchasing adjacent deep products online when the customer repatronizes the retailer.
Study 3 is a two-treatment between-subjects design. The two treatments were deep product/in-store and shallow product/in-store. We randomly assigned 414 participants from Qualtrics Consumer Panel to the two treatments. To ensure that participants were relevant for our context, we asked Qualtrics to screen them based on age (between 20 and 60 years old), household income (minimum of $30,000; 50% of sample needs to have at least $60,000 household income), and e-commerce experience (need to have purchased a product online at least once in the past six months).
We first told subjects that a new sports and outdoor-gear retailer has opened in town. We then asked them to imagine their first shopping trip to this retailer's physical store, where they purchased either a sport shirt (deep product) or a battery charger (shallow product). We then asked them how likely they would be to purchase each of four products from the retailer's website in the future. These four products included two deep products, a sport shirt and a sweater (a product adjacent to the shirt), and two shallow products, a battery charger and an activity tracker watch (neither of which are adjacent to a shirt). Finally, as in Study 2, near the end of the study, after the key measures were collected, we asked the respondents to rate the shirt or the battery charger for product inspection depth, depending on their assigned conditions, to enable us to check the deep versus shallow product manipulation. Details of the questionnaire are in Web Appendix W12.
Manipulation checks confirmed that the shirt was perceived to be deeper on the four-point product inspection depth scale than the charger (shirt = 2.26 (SE = .09), charger = 1.65 (SE = .07); t(412) = 5.51, p < .01).
Figure 4 shows that first purchasing a shirt offline increases intentions to subsequently purchase a shirt online, compared with if the first purchase is a charger offline (Moffline_shirt = 4.35, SE = .14 vs. Moffline_charger = 3.62, SE = .13; t(412) = 3.7, p < .01), confirming H3.
Graph: Figure 4. The effect of first deep product in-store purchase on future online purchase likelihoods, compared with shallow product purchase (Study 3).
We next tested for spillover. H4 posits that generalization will be to adjacent products. Indeed, the results show that the shirt treatment leads to a higher online purchase likelihood for the adjacent sweater than does the charger treatment (Moffline_shirt = 3.88, SE = .14 vs. Moffline_charger = 3.45, SE = .13; t(412) = 2.3, p < .05). This supports H4. While spillover extends from shirt to sweater, it does not extend to the charger and watch—two shallow, nonadjacent products. The likelihood of buying a charger online next time is nonsignificant between the two treatments (Moffline_shirt = 4.48, SE = .15 vs. Moffline_charger = 4.88, SE = .14; t(412) = −1.9, p > .05). Similarly, the difference between purchasing an activity tracker online subsequently is nonsignificant (Moffline_shirt = 4.4, SE = .14 vs. Moffline_charger = 4.5, SE = .14; t(412) = −.5, p > .05). In conclusion, we have support for H4: buying deep products in-store generalizes to adjacent products. Additional regression analyses corroborate these findings and are in Web Appendix W12.
In summary, Study 3 demonstrates that buying deep products in-store encourages customers subsequently to purchase deep products online more than does buying a shallow product in-store, in support of H3. Furthermore, buying a shirt in-store increases the likelihood of not only buying a shirt online in the future but also buying an adjacent product (a sweater) online. This supports H4. Study 3 overall testifies to the value of ELT as a theory for understanding the impact of deep/store purchases on future customer value.
We set out to study the role of the physical store in today's multichannel retailing environment. Our thesis was that the physical store increases customer value by providing physical engagement when customers buy deep products. We drew on ELT to formulate four hypotheses: H1 suggests that deep product/in-store purchases increase customer value more than any other product/channel combination. H2 suggests that ELT provides a mechanism that contributes to this effect. H3 posits that customers who purchase deep products in-store are more likely to purchase online from the retailer in the future. H4 proposes that customers will repurchase not only the original deep product but also related, adjacent deep products online. H1 and H2 follow because the store delivers a tangible, concrete, multisensory experience, which facilitates the physical engagement beneficial for buying deep products. This precipitates effective experiential learning. H3 and H4 follow from the generalization phenomenon proposed by ELT.
We evaluated these four hypotheses using an HMM applied to field data (Study 1) and two lab tests (Studies 2 and 3). The results support the following conclusions: ( 1) purchasing deep products in-store increases future customer value more than any other product/channel combination (H1), ( 2) the ELT mechanism (deep product/in-store purchases → physical engagement → favorable learning → repatronize the retailer) contributes to this increase (H2), and ( 3) purchasing deep/in-store increases the likelihood of purchasing the focal and adjacent deep products online in the future (H3 and H4).
Our work has several implications for future research. First, researchers should consider the product/channel combination in studying consumer decisions in a multichannel context. We focus on the physical store and deep products, but the bigger picture is for future research to study product and channel choices together. Not doing so may incur a lost opportunity—the insights in our article would have been diminished if we had focused solely on channel or product.
Second, the product inspection depth concept extends theory regarding product categorizations such as digital versus nondigital by incorporating details related to both physical and visual inspection. We believe this is a valuable concept for studying and comparing products because inspection depth varies appreciably across products, and channels differ in the degree of inspection they can provide. We hope researchers will apply and perfect this concept.
Third, we demonstrate the applicability of ELT to developing customer relationships. This importantly extends the domain to which ELT has been applied. The pivotal role of the customer experience ([60]) cannot be understated.
Fourth, our reliance on ELT proved fruitful, but [25] posit that decision making can be a dual process, drawing on cognitions and feelings. ELT prescribes an involved process of reflection, hypotheses, and repetition, by which consumers form cognitions. ELT does not directly tap feelings. Under this umbrella are concepts such as trust, commitment, affect, and emotions advanced by [34] and [ 5]. These concepts provide additional theories for our results and thus need testing.
Study 1's data are from 2005, yet our findings are in evidence today. The last decade has witnessed a marked increase in the opening of physical stores by online retailers, despite myriad changes in the retailing environment. This attests that our findings are not ephemeral. The general lesson of our research is for retailers to create a concrete, tangible, and multisensory experience for customers buying deep products. This sets the stage for favorable experiential learning and increased customer value. Retailers can do this in numerous ways:
First, when retailers find that a customer is buying online but is decreasing in value, we suggest a promotion for deep products in-store. Our marketing simulations show that there is potential to increase customer value through direct marketing. Second, retailers should facilitate physical engagement for deep products through merchandising and training sales personnel to walk customers through the engagement (e.g., by helping customers try and use deep products in-store). Third, retailers cannot and should not infer product inspection depth solely from predefined product categories, because there is much variation in inspection depth with a particular category. Rather, management should infer inspection depth using our proposed measures or expert, independent judges. Fourth, we recommend retailers use a deep/offline onboarding strategy for new customers. They should use acquisition channels and product promotion strategies that encourage the first purchase to be deep products in-store.
Our general lesson applies to recent developments in retailing. For example, showrooming ([10]) starts with customers in-store, where the retailer can provide physical engagement. However, the retailer may lose customers who use their smartphones in-store to find the product elsewhere. There are two possible solutions. Retailers can train sales reps to attend to customers buying deep products and equip reps with a mobile device to place orders, or retailers can provide customers with an app to "lock them in." Interestingly, [55] find that store-oriented customers are good targets for retailer apps. Similarly, "buy online, pickup in-store," a form of "web-rooming," can get customers to the store where they can physically engage with additional products to the ones they ordered online.
Our central thesis also has implications for retail loyalty program design and its real-time management. Loyalty programs may be more effective if they provide incentives for customers to shop in-store, such as extra reward points for in-store shopping, particularly when it comes to deep products. These programs need to provide the data, system, and incentives needed to route customers to the physical store when needed, such as when customer value is waning.
Our work has limitations that suggest opportunities for future investigation. First, our hypotheses assume that physical stores provide effective physical engagement and favorable experiential learning. Our results could have turned out differently if our focal retailer's stores did not provide satisfactory experiences. This meant that our hypotheses were nontrivial and falsifiable, and future work should investigate the store features that best provide physical engagement. Second, we could not observe and thus could not incorporate customers' category expertise or preference evolution due to product consumption. Two customers who bought similar tents could have different camping experiences, which would partially determine future purchases that are outside of the firm's control. Third, with transactional data typical in customer relationship management research, we assumed that customers make channel and product decisions jointly. Future work using more granular data could examine the sequence of channel and product choices and explore situations such as planned versus serendipitous purchases. Fourth, we do not know exactly when a customer makes the decision of what and where to purchase. A shopping list study would be useful for future research. Finally, our observational data were from one retailer—a specialist in outdoor products. Future work should consider other product categories.
Guided by our mantra to create physical engagement to enhance customer value, we next discuss additional fertile areas for research.
Is the traditional physical store, with its requisite square footage and inventory, the best way to sell deep products? As fulfillment logistics have gotten faster, more orders can be placed via in-store kiosks and staff-assisted online ordering and can be fulfilled quickly. This suggests stores do not need to allocate a large space for inventory and can use that real estate to help convert the physical store to a showroom. The question is which is best—physical store or showroom?—in terms of customer preferences and the financials.
As the physical store's value lies in facilitating physical engagement through deep products, stores may not need to carry a large assortment (not all sizes and not all colors). The counterargument is that stores should carry a broad assortment because customers might want to physically engage with specific sizes/colors. Future research should study and resolve this tension.
A trained and empathetic sales team would play a key role in delivering physical engagement to the customer. We noted this in discussing showrooming, where the sales rep was needed to keep the customer "on course" with the retailer. [20] also delineate an important role for staff in developing customer value. This would suggest full staffing of physical stores (see [10]). However, this strategy is expensive, and customers may be quite capable of physically engaging on their own.
As noted in formulating H1, we assume the favorable learning the customer gains from physical engagement transfer to the store. One way the retailer might ensure this is to emphasize its private label. Nordstrom, L.L.Bean, and Warby Parker are good examples.
Physical engagement is a particular challenge for online retailers. What combination of videos, chat, user testimonials, virtual reality, augmented reality, and other interactive features should be deployed to mimic in-store physical engagement?
Although the current state of augmented reality technology is probably not realistic enough to fully capture physical engagement ([ 6]), future work could examine how this technology can satisfactorily do so for certain product categories, consumption contexts, and consumer segments.
One recent notable example is the virtual tasting experiences offered by Wine.com, an online wine retailer. This program encourages customers to order a featured wine ahead of time, then go online and, either live or via recorded videos, taste the featured wine alongside its winemaker or renowned critics such as Steven Spurrier. This program not only can result in immediate sales of the featured wines and fills the vacuum created by the closure of physical tasting rooms during COVID-19 but, according to our theory, also has the potential to facilitate sensory engagement (in this case, visual, audio, taste, and smell). Research could investigate whether such creative digital efforts can translate into long-term loyalty towards the online retailer.
Retailers might consider updating their customer segmentation schemes to reflect customers' needs for physical engagement. [36] show that the need for haptic (touch) experiences is an individual trait. An interesting line of future research is to investigate whether this trait evolves over time and what drives this evolution. Retailers could thus explore ways to identify physical engagement-prone customers and design their stores and websites accordingly.
Retailers may want to explore other avenues to physically engage customers in-store, such as through cultural events or prosocial efforts. For example, they may celebrate ethnic holidays and dedicate a certain percentage of sales to charitable causes. Can these be turned into forms of physical engagement? A retailer can advertise online its efforts to fight COVID-19, but it can also demonstrate in-store the personal protective equipment its donations buy for the community. Future work could investigate whether these various in-store efforts generate customer interactions and affect that enhance customer value.
These speculations, along with the more direct implications stated previously, demonstrate the richness of marching to a simple yet powerful message: Retailers can increase customer value by providing physical engagement when selling deep products. We hope both researchers and practitioners will leverage this message in future work.
Footnotes 1 David Schweidel
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Online supplement:https://doi.org/10.1177/00222429211012106
5 For parsimony, we use "offline" or "store" to denote "physical store."
6 While often not explicitly stated, experiential learning plays a key role in marketing models. [11] capture it by a lagged experience variable reflecting "state dependence" ([13]). In Bayesian learning models, [8] find that experience is a more important driver of future purchases than advertising.
7 Although building a new construct is not the main contribution of this research, a.77 correlation is within the range of discriminant validity for new constructs in the marketing literature (see, e.g., [2]).
8 Another approach would be to use a continuous measure of the product inspection depth of a shopping basket. However, this would present practical challenges. For example, where would a basket of five inexpensive deep products and one expensive shallow product fit along a continuum? Many different combinations of deep and shallow product purchases could yield the same measure on a continuum, decreasing the diagnostic value of a continuous metric.
9 To simplify the notation, we do not use a customer subscript (e.g., i), for a specific customer.
To further test the recency effects of experience, we tried various temporal decay specifications on the cumulative experiences. The decay parameters are not significant, and the specification without decay fit the data best.
We do not have information on content of the marketing and do not know which products are featured. Examination of some of the physical fliers and discussions with the management indicate that the retailer featured both deep and shallow products.
We also do not have product attribute measures for all available products (in typical consumer choice models, one would study a single category—e.g., automobiles—and attributes would include horsepower, color, price, miles per gallon, etc.). The common attribute we explicitly measure and incorporate, drawing on our theory, is product inspection depth, which we use to define our purchase amount dependent variables. As further robustness check for alternative price specifications, we conducted the following robustness checks: (1) Adding monthly fixed effects, (2) Creating a product basket-based measure of price index per [56]. In particular, we identified the top 30 selling brands and constructed a monthly price index. Both robustness checks account for price fluctuations over time. The substantive results remain the same. We find that the price index is not significant, which we believe is due to the aggregation of prices, combined with insufficient temporal price variation. Note that the price index is significant in [56] because the retailers in their context were explicitly engaged in a price war. This is not the case in our context.
The HMM assumes independence of the four decisions conditional on the latent state. We estimated a two-state, two-segment model that correlated channel choice and purchase amount using copulas. The results indicated that our more parsimonious conditional independence model produces better predictive performance (Web Appendix W5). In addition, we examined the correlations of the residuals post hoc and found that they are small and insignificant. The corroborating evidence between this result and the copula robustness check alleviates concerns that the unobserved components of our utility equations are correlated.
We find that 50/50 and 60/40 yield the same results because the data are such that no deep product is reclassified when moving from 50th to 60th percentile.
Profits are sensitive to marketing costs, which we assumed were $.60 per direct mail piece. At $.75, the base and +1 strategies would have equal profit. However, the purpose of this simulation is to illustrate targeting implications, not profits per se.
As a check, we flipped the model and used "store versus online" as the moderator of deep versus shallow. This yielded the same conclusions.
References Ackerman Joshua M. , Nocera Christopher C. , Bargh John A.. (2010), " Incidental Haptic Sensations Influence Social Judgments and Decisions ," Science , 328 (5986), 1712 – 15.
Ailawadi Kusum L. , Lehmann Donald R. , Neslin Scott A.. (2003), " Revenue Premium as an Outcome Measure of Brand Equity ," Journal of Marketing , 67 (2), 1 – 17.
Atchadé Yves F. (2006), " An Adaptive Version for the Metropolis Adjusted Langevin Algorithm with a Truncated Drift ," Methodology & Computing in Applied Probability , 8 (2), 235 – 54.
Baker Julie , Parasuraman Albert , Grewal Dhruv , Voss Glenn B.. (2002), " The Influence of Multiple Store Environment Cues on Perceived Merchandise Value and Patronage, Journal of Marketing, 66 (2), 120–41.
Bowden Jana Lay-Hwa. (2009), " The Process of Customer Engagement: A Conceptual Framework ," Journal of Marketing Theory and Practice , 17 (1), 63 – 74.
Conger Kate. (2020), " Does the Shoe Fit? Try It on With Augmented Reality ," The New York Times (December 20), https://www.nytimes.com/2020/12/22/technology/augmented-reality-online-shopping.html.
Creswell Julie. (2017), " The Incredible Shrinking Sears ," The New York Times (August 17), https://www.nytimes.com/2017/08/11/business/the-incredible-shrinking-sears.html.
Erdem Tulin , Keane Michael P. , Sun Baohong. (2008), " A Dynamic Model of Brand Choice When Price and Advertising Signal Product Quality ," Marketing Science , 27 (6), 1111 – 25.
Gaudin Sharon. (2016), " Thanks to Tech, Stores Are Evolving into Showrooms ," Computerworld (January 18), https://www.computerworld.com/article/3023345/thanks-to-tech-stores-are-evolving-into-showrooms.html.
Gensler Sonja , Neslin Scott A. , Verhoef Peter C.. (2017), " The Showrooming Phenomenon: It's More Than Just About Price ," Journal of Interactive Marketing , 38 , 29 – 43.
Guadagni Peter M. , Little John D.C.. (1983), " A Logit Model of Brand Choice Calibrated on Scanner Data ," Marketing Science , 2 (3), 203 – 38.
Hayes Andrew F. (2018), Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach , 2nd ed. New York : The Guilford Press.
Heckman James J. (1981), "Heterogeneity and State Dependence," in Studies in Labor Markets , Rosen S. , ed. Chicago : University of Chicago Press , 91 – 139.
Kacen Jacqueline J. , Hess James D. , Chiang Wei-Yu Kevin. (2013), " Bricks or Clicks? Consumer Attitudes Toward Traditional Stores and Online Stores ," Global Economics and Management Review , 18 (1), 12 – 21.
Kolb David A. (1971), Individual Learning Styles and the Learning Process. Cambridge, MA : MIT Press.
Kolb David A. (1984), Experience as the Source of Learning and Development. Upper Saddle River, NJ : Prentice Hall.
Kolb Alice Y. , Kolb David A.. (2005), " Learning Styles and Learning Spaces: Enhancing Experiential Learning in Higher Education ," Academy of Management Learning & Education , 4 (2), 193 – 212.
Krishna Aradhna. (2012), " An Integrative Review of Sensory Marketing: Engaging the Senses to Affect Perception, Judgment and Behavior ," Journal of Consumer Psychology , 22 (3), 332 – 51.
Krishna Aradhna , Elder Ryan S. , Caldara Cindy. (2010), " Feminine to Smell but Masculine to Touch? Multi-Sensory Congruence and Its Effect on the Aesthetic Experience ," Journal of Consumer Psychology , 20 (4), 410 – 18.
Kumar V. , Pansari Anita. (2016), " Competitive Advantage Through Engagement ," Journal of Marketing Research , 53 (4), 497 – 514.
Kumar V. , Sriram S. , Luo Anita , Chintagunta Pradeep. (2011), " Assessing the Effect of Marketing Investments in a Business Marketing Context ," Marketing Science , 30 (5), 924 – 40.
Kumar V. , Venkatesan Rajkumar. (2005), " Who Are the Multichannel Shoppers and How Do They Perform? Correlates of Multichannel Shopping Behavior ," Journal of Interactive Marketing , 19 (2), 44 – 62.
Lal Rajiv , Sarvary Miklos. (1999), " When and How Is the Internet Likely to Decrease Price Competition? " Marketing Science , 18 (4), 485 – 503.
Li Shibo , Sun Baohong , Montgomery Alan L.. (2011), " Cross-Selling the Right Product to the Right Customer at the Right Time ," Journal of Marketing Research , 48 (4), 683 – 700.
Loewenstein George F. , Weber Elke U. , Hsee Christopher K. , Welch Ned. (2001), " Risk as Feelings ," Psychological Bulletin , 127 (2), 267 – 86.
Luo Anita , Kumar V.. (2013), " Recovering Hidden Buyer-Seller Relationship States to Measure the Return on Marketing Investment in Business-to-Business Markets ," Journal of Marketing Research , 50 (1), 143 – 60.
MacDonald Iain L. , Zucchini Walter. (1997), Hidden Markov and Other Models for Discrete-Valued Time Series. Boca Raton, FL : CRC Press.
McLeod Saul A.. (2017), " Kolb's Learning Styles and Experiential Learning Cycle ," (October 24), https://www.simplypsychology.org/learning-kolb.html.
Montoya Ricardo , Netzer Oded , Jedidi Kamel. (2010), " Dynamic Allocation of Pharmaceutical Detailing and Sampling for Long-Term Profitability ," Marketing Science , 29 (5), 909 – 24.
Nelson Phillip. (1970), " Information and Consumer Behavior ," Journal of Political Economy , 78 (2), 311 – 29.
Neslin Scott A. , Jerath Kinshuk , Bodapati Anand , Bradlow Eric T. , Deighton John , Gensler Sonja , et al. (2014), " The Interrelationships Between Brand and Channel Choice ," Marketing Letters , 25 (3), 319 – 30.
Netzer Oded , Lattin James M. , Srinivasan V.. (2008), " A Hidden Markov Model of Customer Relationship Dynamics ," Marketing Science , 27 (2), 185 – 204.
Ofek Elie , Katona Zsolt , Sarvary Miklos. (2011), " Bricks and Clicks: The Impact of Product Returns on the Strategies of Multichannel Retailers ," Marketing Science , 30 (1), 42 – 60.
Pansari Anita , Kumar V.. (2017), " Customer Engagement: The Construct, Antecedents, and Consequences ," Journal of the Academy of Marketing Science , 45 (3), 294 – 311.
Peck Joann. (2011), " Does Touch Matter? Insights from Haptic Research in Marketing ," in Sensory Marketing: A Confluence of Psychology, Neuroscience and Consumer Behavior Research , Krishna Aradhna , ed. New York : Routledge , 47 – 62.
Peck Joann , Childers Terry L. (2003a), " Individual Differences in Haptic Information Processing: The 'Need for Touch' Scale ," Journal of Consumer Research , 30 (3), 430 – 42.
Peck Joann , Childers Terry L. (2003b), " To Have and to Hold: The Influence of Haptic Information on Product Judgments ," Journal of Marketing , 67 (2), 35 – 48.
Peck Joann , Shu Suzanne B.. (2009), " The Effect of Mere Touch on Perceived Ownership ," Journal of Consumer Research , 36 (3), 434 – 47.
Peck Joann , Wiggins Jennifer. (2006), " It Just Feels Good: Customers' Affective Response to Touch and Its Influence on Persuasion ," Journal of Marketing , 70 (4), 56 – 69.
Petersen J. Andrew , Kumar V.. (2009), " Are Product Returns a Necessary Evil? Antecedents and Consequences ," Journal of Marketing , 73 (3), 35 – 51.
Peterson Hayley. (2020), " More than 3,600 Stores Are Closing in 2020 as the Retail Apocalypse Drags On. Here's the Full List ," Business Insider (May 27), https://www.msn.com/en-us/lifestyle/lifestyle-buzz/more-than-3-300-stores-are-closing-in-2020-as-the-retail-apocalypse-drags-on-here-s-the-full-list/ss-BB13V2cJ.
Pieters Rik. (2017), " Meaningful Mediation Analysis: Plausible Causal Inference and Informative Communication ," Journal of Consumer Research , 44 (3), 692 – 716.
PwC (2017), " Total Retail Survey: 10 Retailer Investments for an Uncertain Future ," (accessed August 5, 2021), https://www.pwc.com/gx/en/industries/assets/total-retail-2017.pdf.
Rosenbaum Paul R. (2002), Observational Studies , 2nd ed. New York : Springer.
Sachdeva Ishita , Goel Suhsma. (2015), " Retail Store Environment and Customer Experience: A Paradigm ," Journal of Fashion Marketing and Management , 19 (3), 290 – 98.
Schweidel David A. , Bradlow Eric T. , Fader Peter S.. (2011), " Portfolio Dynamics for Customers of a Multiservice Provider ," Management Science , 57 (3), 471 – 86.
Squire Patton Boggs , Retail Kantar. (2015), " The Multichannel High Street: Winning the Retail Battle in 2015 ," (accessed August 6, 2021), https://www.squirepattonboggs.com/-/media/files/insights/publications/2015/01/the-multichannel-high-street/themultichannelhighstreetwinningtheretailbattlein2015.pdf.
Stone Merlin , Hobbs Matt , Khaleeli Mahnaz. (2002), " Multichannel Customer Management: The Benefits and Challenges ," Journal of Database Marketing , 10 (1), 39.
Taylor Glenn. (2015), " 85% of Consumers Prefer to Shop in Physical Stores ," Retail TouchPoints (June 15) , https://www.retailtouchpoints.com/topics/shopper-experience/85-of-consumers-prefer-to-shop-in-physical-stores.
Thomas Jacquelyn S. , Sullivan Ursula Y.. (2005), " Managing Marketing Communications with Multichannel Customers ," Journal of Marketing , 69 (4), 239 – 51.
Trefis Team (2016), " Alibaba Opens A Physical Store: Eyeing A Broader Market? " Forbes (January 18), https://www.forbes.com/sites/greatspeculations/2016/01/18/alibaba-opens-a-physical-store-eyeing-a-broader-market/#6cb5bb8344fa.
U.S. Census Bureau (2020), " Quarterly Retail E-Commerce Sales: 4th Quarter 2019 ," (February 19), https://www2.census.gov/retail/releases/historical/ecomm/19q4.pdf.
Valentini Sara , Montaguti Elisa , Neslin Scott A.. (2011), " Decision Process Evolution in Customer Channel Choice ," Journal of Marketing , 75 (6), 72 – 86.
Van Doorn Jenny , Lemon Katherine N. , Mittal Vikas , Nass Stephan , Pick Doreén , Pirner Peter , et al. (2010), " Customer Engagement Behavior: Theoretical Foundations and Research Directions ," Journal of Service Research , 13 (3), 253 – 66.
Van Heerde Harald J. , Dinner Isaac M. , Neslin Scott A.. (2019), " Engaging the Unengaged Customer: The Value of a Retailer Mobile App ," International Journal of Research in Marketing , 36 (3), 420 – 38.
Van Heerde Harald J. , Gijsbrechts Els , Pauwels Koen. (2015), " Fanning the Flames? How Media Coverage of a Price War Affects Retailers, Consumers, and Investors ," Journal of Marketing Research , 52 (5), 674 – 93.
VanStory Beth. (2019), " Are the Online Retailers Opening Stores Losing Their Minds? " Digital Commerce 360 (October 24), https://www.digitalcommerce360.com/2019/10/24/are-the-online-retailers-opening-stores-losing-their-minds/.
Venkatesan Rajkumar , Kumar V. , Ravishanker Nalini. (2007), " Multichannel Shopping: Causes and Consequences.," Journal of Marketing , 71 (2), 114 – 32.
Verhoef Peter C. , Donkers Bas. (2005), " The Effect of Acquisition Channels on Customer Loyalty and Cross-Buying ," Journal of Interactive Marketing , 19 (2), 31 – 43.
Verhoef Peter C. , Lemon Katherine N. , Parasuraman A. , Roggeveen Anne , Tsiros Michael , Schlesinger Leonard A.. (2009), " Customer Experience Creation: Determinants, Dynamics, and Management Strategies ," Journal of Retailing , 85 (1), 31 – 41.
Zhang Jonathan Z. , Netzer Oded , Ansari Asim. (2014), " Dynamic Targeted Pricing in B2B Relationships ," Marketing Science , 33 (3), 317 – 37.
Zhang Jonathan Z. , Watson George F. IV , Palmatier Robert W. , Dant Rajiv P.. (2016), " Dynamic Relationship Marketing ," Journal of Marketing , 80 (5), 53 – 75.
~~~~~~~~
By Jonathan Z. Zhang; Chun-Wei Chang and Scott A. Neslin
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 69- How Political Identity Shapes Customer Satisfaction. By: Fernandes, Daniel; Ordabayeva, Nailya; Han, Kyuhong; Jung, Jihye; Mittal, Vikas. Journal of Marketing. Feb2022, p1. DOI: 10.1177/00222429211057508.
Ahead of Print- Database:
- Business Source Complete
Record: 70- Identifying Market Structure: A Deep Network Representation Learning of Social Engagement. By: Yang, Yi; Zhang, Kunpeng; Kannan, P.K. Journal of Marketing. Jul2022, Vol. 86 Issue 4, p37-56. 20p. 3 Diagrams, 4 Charts, 8 Graphs, 1 Map. DOI: 10.1177/00222429211033585.
- Database:
- Business Source Complete
Identifying Market Structure: A Deep Network Representation Learning of Social Engagement
With rapid technological developments, product-market boundaries have become more dynamic. Consequently, competition for products and services is emerging outside the product-market boundaries traditionally defined based on Standard Industrial Classification and North American Industry Classification System codes. Identifying these fluid product-market boundaries is critical for firms not only to compete effectively within a market but also to identify lurking threats and latent opportunities outside market boundaries. Newly available big data on social media engagement presents such an opportunity. The authors propose a deep network representation learning framework to capture latent relationships among thousands of brands and across many categories, using millions of social media users' brand engagement data. They build a brand–user network and then compress the network into a lower-dimensional space using a deep autoencoder technique. The authors evaluate this approach quantitatively and qualitatively and visually display the market structure using the learned representations of brands. They validate the learned brand relationships using multiple external data sources. They also illustrate how this method can capture the dynamic changes of product-market boundaries using two well-known events—the acquisition of Whole Foods by Amazon and the introduction of the Model 3 by Tesla—and how managers can use the insights that emerge from this analysis.
Keywords: artificial intelligence; deep representation learning; social media; competitive market structure; big data
Firms compete in a market to satisfy the specific needs of consumers in the market. The market and the competing products make up a "product-market," with the boundary defining the brands competing within that market. Market structure is defined on the basis of these product-markets and their (possibly overlapping) boundaries. Identifying the product-market boundary and examining the strength of competition between brands within the product-market have long been important issues with strategic implications for next-generation product design, product positioning, new customer acquisition, and pricing and promotion decisions. Rapid changes to the competitive environment, however, have made identifying the product-market boundaries increasingly challenging. With technological advances, the product-market boundaries themselves are changing and competitive threats and opportunities are emerging outside of narrowly defined product-market boundaries.
Numerous recent competitive events support the idea that product-market boundaries are highly fluid. For example, the digital camera product-market was upended by technological developments in smartphone categories. Similarly, Tesla, which initially entered the product-market of high-end automobiles with an innovative fuel technology, has since rolled out products for the lower-end market, thereby changing competition in the lower-end product-market as well. Amazon, previously an online platform, essentially crossed product-market boundaries when it acquired Whole Foods and entered the offline product-market. In many such situations, product-market boundaries based on traditional Standard Industrial Classification and North American Industry Classification System codes are inadequate indicators of emerging threats and opportunities. Given the potential for new and unforeseen relationships between brands, managers need deeper insights into the fluid product-market boundaries to be able to spot potential competitors and complements, identify cross-promotion strategies, and develop firm-level strategies.
These observations naturally lead to several important questions: How can managers accurately identify potential threats and opportunities? If a competitive threat emerges from a different market, how can managers proactively anticipate such threats? How can we answer these questions and derive marketing insights using easy-to-obtain and publicly available data? Our article aims to answer these questions using large-scale (>100 million) social media user engagement data (likes and comments) spanning several thousands of brands in different product/service categories.
Over the years, academics and practitioners have contributed significantly to developing various methods to define and identify market structure (see the review by [46]]). These include survey-based methods such as brand concept maps ([19]) and the Zaltman metaphor elicitation technique ([53]), methodologies based on observational purchase data (e.g., brand switching) ([21]; [37]), consideration sets ([42]), and scanner-based purchase data ([10]; [36]; [45]). Within the online context, researchers have used unstructured user click streams ([32]), online search logs ([24]; [42]), and customer reviews ([26]). Many of these methods use data from the bottom of the purchase funnel, such as evaluation- and purchase-stage data, and thus assume that the product-market boundaries are prespecified. Even those methods that use data from the top of the funnel at the awareness or preevaluation stage, such as forum discussions ([35]) and hashtags ([33]), define a product-market boundary first and then examine the competition within the prespecified product-market to make these methods implementable. Thus, many of the methods are unable to capture changes to the product-market boundaries and/or the impact that a brand from outside the boundary may have on brands within a product-market.
Our methodology creates a more inclusive representation of brands by examining brand–user relationships at the top of the purchase funnel. Unlike the extant methods for identifying market structure that use data from consumers' lower funnel activities such as purchase data, brand switching, price comparison data, or consideration data that prespecify boundaries (e.g., [10]; [21]; [37]; [42]; [50]), we use upper-funnel user–brand engagement data (such as liking and commenting on brand posts) from social media that spans product-markets. At the lower end of the purchase funnel, consumers winnow down the brands they consider to a few substitutes. Thus, interactions at this stage are not as informative of the broader (and possibly complementary) linkages between the brands across product-markets, which are captured more easily at the upper funnel. For example, a consumer considering travel may consider hotel or Airbnb options, airline options or travel intermediaries. At this early stage (the upper end of the funnel), understanding such user–brand linkages could be more informative of the broader relationships between the brands on a continuum from substitutes to complements. Our methodology uses such upper-funnel user–brand engagement data to identify these latent relationships among a large number of brands.
Many extant studies in market structure, including those mentioned previously and those using big data technologies (e.g., [ 7]; [12]; [26]; [35]; [42]), view the competing/complementary brands as brand–brand networks. That is, they specify the relationship between any two brands using similarity metrics derived from brand switching, co-occurrences, and word embeddings, without directly modeling the entities (customers, individual consideration sets, or individual reviews) that give rise to such similarities. Our methodology based on brand–user networks considers both brands and users as primitives and uses as input the relationship in terms of each user's liking and commenting on brands. The essential difference between these approaches and our methodology is that extant research considers aggregate data of relationships between brands (brand–brand) as input, whereas our methodology considers the disaggregate individual-level relationships between users and brands (brand–user) as input.
The distinction becomes more salient when a product-market boundary is not prespecified. Consider, for example, User 1, who likes United Airlines and Hyatt, while User 2 likes Southwest Airlines and Hyatt. When the product-market is prespecified as "airline brands," information about the users liking the Hyatt brand is discarded. As a result, information that could provide insights into the relationship between United Airlines and Southwest Airlines through their relationships with Hyatt is not considered. However, when we do not prespecify the product-market boundaries, we are able to leverage all such information and create a more accurate representation of the brands.
From this premise, we first construct a large-scale brand–user network based on user engagement on brands' social media public fan pages. Then, we propose a deep network representation learning method to discover relationships within the data. Specifically, we use a deep learning method suitable for ( 1) handling large data efficiently and ( 2) learning complex patterns from data effectively (see [ 2]; [49]). The process leads to a low-dimensional representation (i.e., a vector) for each brand and each user by training a deep autoencoder on the network data. The deep autoencoder is similar to traditional dimensionality reduction methods such as principal component analysis (PCA) in capturing latent factors in data with few dimensions. It is, however, very different from those methods in that it uses a nonlinear transformation function to learn the latent patterns in data while reducing the noise in the data. In our context, the deep autoencoder can preserve the first-order (user–brand direct connection) and the second-order (two users connecting to the same brand, or one user connecting to two different brands) network topologies. As a result, brands with network structural equivalence are located closer together in the representation space, while brands with dissimilar network structures are located further away from each other. This method also projects users and brands onto the same dimensional space, which can be used for many different follow-up analyses. We use an illustrative example (in Figure 1) to demonstrate how network representation learning works.
Graph: Figure 1. An illustration of deep network representation learning.
Suppose we have three brand nodes (B1, B2, and B3) and five user nodes (U1, U2, U3, U4, and U5) in a network. Our network representation learning approach aims to find a function that maps each node into a low-dimensional vector (e.g., three dimensions, for the sake of illustration) while the network structural information is preserved maximally. That is, when nodes exhibit similar structures (first order and/or second order), they are projected onto similar vectors and located closer in the reduced three-dimensional embedding space. Because U1 engaged with B1, we expect the vector representation of B1 and U1 to be close. Similarly, B2 is closer to B1 than to B3 because B2 shares more common users with B1 than with B3. Because B2 has connections to U4 and U5, this makes B2 lean toward them.
We establish the face validity of our approach through the identification of product-market boundaries. Our analysis of the brand–user engagement data of over 5,000 brands and nearly 26 million users reveals product-market boundaries with high face validity—grouping of specific categories, high-end brands, and overlaps. We then conduct external validation checks using additional sources including survey and Google search trend data. The market structure derived using our approach is highly correlated with those derived using external data sources. Our approach also overcomes common limitations in extant methods such as data sparsity. Our event studies on Amazon's acquisition of Whole Foods and Tesla's introduction of the Model 3 illustrate how our methodology captures the changes in product-markets associated with these events. We also discuss how the market structure maps can reveal opportunities and threats facing a brand. For instance, our market structure identifies Disney Cruise Line and Hyatt—two brands outside the airline market—as proximal brands to Southwest Airlines. Such findings provide opportunities for Southwest, as it can target those who like Disney Cruise and Hyatt in social media, cross-promote its brand by teaming up with Disney Cruise and/or Hyatt on each other's websites, or launch coalition loyalty programs.
Our article contributes to product-market research by leveraging the information embedded in big data of user–brand engagement networks to identify product-markets without having to prespecify boundaries. User–brand engagement network data at a high level in the purchase funnel (interest phase), together with deep learning techniques, provide us with insights at a greater scale and level of detail than extant methods. Our ability to map a large number of brands and precisely visualize brand relationships using learned vector representations enables managers to identify opportunities and threats that lie beyond product-market boundaries. Moreover, our method satisfies the three elements widely regarded as essential to successful real-world applications of artificial intelligence: data, algorithm, and computing power ([ 2]). In this article, we leverage deep learning and a network representation learning (algorithm) to understand market structure using large-scale social media data (data). This model implementation is efficient under NVIDIA P100 graphics processing unit, with Tensorflow as the backend framework (computing power). In summary, our study is an apt illustration of how artificial intelligence can be used to tackle a traditional marketing problem and provide richer insights for mangers in a rapidly changing competitive environment.
Extant work in identifying competitive market structures dates to the 1970s (e.g., [ 8]; [20]), when diary panel–based brand-switching purchase data and survey-based consumer judgments of substitution in use or similarities were used to construct market structure maps. These studies depended on customer data generated either at a late stage of the customer journey or at the very beginning of the journey. The increased availability of scanner-panel data of purchases, market structure models with marketing mix (e.g., [ 5]; Kannan and Wright 1991), and dynamic market structure models (e.g., [10]) provided more detailed insights into interbrand relationships and competition. Approaches such as brand concept maps ([19]) and the Zaltman metaphor elicitation technique ([53]) relied on data collected using surveys and, therefore, were effort intensive. Given the scaling issues with the maximum likelihood–based models and the limitations of survey data, the product-market boundaries were prespecified generally at the industry level so that a smaller number of brands within an industry could be analyzed. The advent of online sources, such as review platforms, social media platforms, and clickstream data, dramatically increased the volume and variety of data for market structure studies, especially at the awareness, search, and consideration stages of the customer journey ([24]; [26]; [35]; [42]; see Tables 1 and 2). Even with a large volume of data, these studies predefine the product-market boundaries at the industry level to make the analyses viable.
Graph
Table 1. Comparison of Different Types of Work on Market Structure Discovery.
| Primary/Survey Data | Text Mining | Social Tag–Based | Search Data | Shopping Data | Social Engagement |
|---|
| Data volume | Small | Large | Large | Large | Very large | Very large |
| Data veracity | Authentic | Noisy | Moderately noisy | Moderately noisy | Authentic | Moderately noisy |
| Privacy preserving | Yes | Yes | Yes | No (need to insert a tracking pixel) | Yes | Yes |
| Data availability | Low (need to do survey) | High (publicly available) | High (publicly available) | Low (need to insert a tracking pixel) | Low (need to partner with retailers) | High (publicly available) |
| Data preprocessing cost | Low (use consideration set directly) | High (text mining is error-prone) | High (text mining is error-prone) | Low (use consideration set directly) | Low (use product co-occurrence) | Low (use network raw data) |
Graph
Table 2. Summary of Difference Among Extant Literature on Market Structure Discovery.
| Kim, Albuquerque, and Bronnenberg (2011) | Lee and Bradlow (2011) | Netzer et al. (2012) | Ringel and Skiera (2016) | Culotta and Cutler (2016) | Nam, Joshi, and Kannan (2017) | Gabel, Guhl, and Klapper (2019) | Our Study |
|---|
| Objective | To visualize user search behavior and understand market structure | To visualize competitive market structure using text mining on customer review | To visualize competitive market structure using text mining on forum discussion | To understand asymmetric competition in the product categories | To infer attribute-specific brand ratings | To analyze user-generated tags for marketing research | To leverage NLP and ML for analyzing market structure | To propose a novel deep network representation learning framework for market structure |
| Prespecifying market category | Yes | Yes | Yes | Yes | No | No | No | No |
| Network types | Brand–brand | Brand–brand | Brand–brand | Product–product | Brand–brand | Brand–brand | Product–product | Brand–user |
| Brands/products | 62 products, 4 brands | 9 brands | 169 products, 30 brands | 1,124 products | 200 brands | 7 brands | 133 categories, 30,763 products | 5,478 brands |
| Consumers/users | N.A. | N.A. | 76,587 | 100,000+ | 14.6 million | N.A. | N.A. | 25,992,832 |
| Data sources | Amazon | Customer review at Epinions | Online discussion forum | Product comparison website | Twitter | Social tagging platform Delicious | Retailer | Facebook public fan page |
| Data type | Consumer search | Text | Text | Consumer search | Network | Social tags | Shopping baskets | Network |
| Brand association methodology | Consideration set | Text-mining | Text-mining | Consideration set | Network learning | Network learning | Network learning | Network learning |
| Brand relationship asymmetry | Yes | No | No | Yes | No | No | Yes | Yes |
| Dimension reduction | Yes | Yes | No | No | No | Yes | Yes | Yes |
| External validation | N.A. | N.A. | Purchase data,survey | Survey | Survey | Brand concept map (survey) | N.A. | Event study,survey, Google search trends |
1 Notes: NLP = natural language processing; ML = machine learning; N.A. = not applicable.
There are other studies where the product-market boundaries are not predefined: [11] using online reviews, [33] using social tags, and [ 7] using Twitter hashtags. More recently, using word embeddings, [12] analyze customers' market baskets of items purchased on shopping trips. Still, from a methodological perspective all these studies use brand–brand networks—a distinct disadvantage, as we discussed previously. Our methodology uses brand–user networks, and the scale at which we analyze the data is much larger than any of the extant methods (cf. [12]).
We analyze social media engagement data in the form of user–brand links. Social media platforms such as Facebook, Twitter, and Instagram host public fan pages created by firms to facilitate communication with customers and promote products. The user–brand engagement could be in the form of a user liking a post by the brand, sharing a brand post, or commenting on a brand post. Because each of these likes, shares, and comments/posts is a user–brand link in our study, it is important to understand what they represent. Surveys of fans of brands have revealed many reasons as to why users "like" a brand or post/share comments. Positive motivations for interacting with a brand include to support a brand they like, to get a coupon or discount, to receive regular updates from the brand, to participate in contests, to share personal experiences, to share their interests/lifestyles with others, to research brands, to imitate a friend who likes the brand, or to act on a recommendation from another fan ([25]; [34]; [39]; [41]). Conversely, users may also leave negative comments to hurt a brand in favor of its rival brand ([17]).
In our approach, we make a minimal assumption by creating a user–brand link regardless of the type of engagement (like, share, or comment/post). This assumption is based on the rationale that users interacting with a brand online exhibit their interest toward the brand to some extent. Thus, the two brands are related to one another on a spectrum ranging from substitutes to independents to complements. Prior research has examined such contexts and studied the impact of user engagement on brand image and customer purchase intentions with mixed results ([16]; [34]). [31] use a field experiment to find that users who liked a gym brand online were likely to become members of that gym offline. In another field experiment setting, [18] find that "liking" is simply a symptom of a positive brand attitude and does not imply the fan is any more loyal to the brand or any more likely to purchase the brand. In addition, it is only when users who liked the brand are targeted using promotional communication by the firm that purchase probabilities increase. Thus, for our research purposes we treat a like or a comment/post as exhibiting an interest toward the brand at the beginning of the customer journey. Such a tendency for users to connect to brands is generally interpreted as interest and may indicate broader (e.g., offline) interactions ([ 7]; [25]; [34]; [35]), which is consistent with our treatment. Our proposed approach is also consistent with research in social network analysis suggesting that social network structure equivalence reflects value/interest homophily and can be used to measure social proximity ([29]).
Social network platforms, such as Facebook, Instagram, and Twitter, can be abstracted as a network containing business (firm) accounts and individual user accounts. Firms use the public fan pages of business accounts to communicate with their customers and fans. Users interact with brands and with each other in different ways, such as commenting, liking, sharing, and following. To discover latent relationships among brands, we propose a deep network representation learning framework with the following steps.
We specify a set of brands that is of interest in the social network platform. We then download all available user engagement data from the brands' public fan pages covering an appropriate time window based on managerial interest. A user engagement is defined as either liking or commenting on a firm's post on its public fan page. Note that for the sake of privacy, we do not attempt to collect any personal information of users. Rather, the only user information we obtain is the unique user identifier, assigned by platform, and the user's public engagement activities, consistent with recent studies on social media marketing ([17]; [23]).[ 5] Moreover, different platforms may have their own specific data policy. For example, Facebook does not permit collecting personal information from individuals who liked a given page. Such data restrictions and potential ethical concerns do come at a research cost, as we are unable to verify how representative they are of the population at large.
We start with a cleansing operation to remove spurious users. We then construct a brand–user network including all selected brands and all users engaging with them. A brand node and a user node are connected if the user engages with the brand. The strength of an edge between a brand node and a user node is the engagement frequency.
The deep network representation learning algorithm represents each node (brand or user) as a low-dimensional vector, also known as a node embedding. Embedding techniques are not new in marketing. For example, [49] adopt pretrained word embeddings, where each word is represented as a low-dimensional vector, to extract insights from textual reviews. However, our node embeddings are trained via an unsupervised deep autoencoder. This representation learning is essential to data-driven analysis, and the learned low-dimensional embeddings are useful for the downstream task of identifying and visualizing the product-markets.
The objective in using an autoencoder is to learn the representation of the data so that each node can be represented in a lower-dimensional space while the network structure between users and brands is preserved. It trains the network to ignore the "noise" in the data and focus on the primary latent structure. The autoencoder reduces the dimensionality of the input data to a "bottleneck" (the reduced encoding) and, using the reduced encoding as input, reconstructs a representation of the original data. Learning occurs through backpropagation of the loss (see detailed definition in Web Appendix WA1) to achieve a reconstructed representation as close as possible to the original representation. We are interested in the bottleneck-reduced encoding for developing market structure. In essence, we can compare the dimensionality reduction functionality of the autoencoder with that of PCA. Whereas in PCA the reduced dimensions are linear combinations of the input variables, the reduced dimensions in autoencoder are nonlinear and nonorthogonal, which is achieved through nonlinear activations of the neurons, allowing the model to learn more powerful generalizations than PCA can.
In our application, the autoencoder works on the large brand–user network in an attempt to preserve the network structure such that ( 1) nodes that are directly connected have similar vectors (are closer to each other) in the reduced embedding space, and ( 2) nodes that are not directly connected but share structural equivalence (such as many common neighbors) are also similar in the embedding space. These two types of similarity are referred to as the first-order (direct connection) similarity and the second-order (network structural equivalence) similarity. Formally, we denote the aforementioned network as , where represents a set of n brand nodes, represents user nodes, and represents all links between users and brands. indicates an engagement between user and brand . Given such a network , the network representation aims to learn a mapping function , where . are called brand embedding and user embedding, respectively. A commonly used embedding dimensionality is 300 ([30]; [40]). The objective of the mapping function is to develop appropriate embeddings so that the brand proximities, brand–user proximities, and user proximities exhibited in the original network are preserved as much as possible in the reduced embedding space. (Technical details of the autoencoder methodology and parameter tuning are discussed in Web Appendix WA1.) Representing brands as dense low-dimensional vectors allows us to capture brand relations from multiple facets, as opposed to using unique vectors for each user and each brand as in a network adjacent matrix representation. An example is illustrated in Figure 1.
Drawing on vector representation for brands and users, we use learned embeddings to efficiently compute similarity among brands and to visualize natural clusters of related brands. Finding similar brands to a focal brand can be achieved by a nearest-neighbor search based on the widely used cosine similarity, which measures the cosine of the angle between two vectors and has a range [−1, 1]. Visualizing natural clusters of related brands can be achieved by a dimension reduction method, such as t-distributed stochastic neighbor embedding (t-SNE; [30]), which projects high-dimensional data into a low-dimensional space (e.g., two or three dimensions).[ 6] It has been used for visualization in a wide range of applications and is especially well-suited for visualizing high-dimensional representations learned from deep neural networks. t-SNE preserves the distance of data points well, such that data points nearby in a high-dimensional space (d = 300 in our case) would be close in a lower-dimensional (e.g., two-dimensional) space, while distant data points would be further apart in a lower-dimensional space. Thus, we observe that related brands are surrounding each other in the reduced two-dimensional space after t-SNE.
We use Facebook as our empirical benchmark, as it is one of the largest and most representative online social network platforms. (Our model can be generalized to other similar social network platforms.) To collect Facebook data, we first obtain a list of U.S. brands with the most followers from the social media marketing website Socialbakers.[ 7] Facebook public fan pages are categorized into several groups on Socialbakers, such as Brands, Celebrities, Community, Entertainment, Media, Place, Society, and Sport. We focus on the "Brands" category because it covers a wide range of industries and is more interesting to marketers. On Facebook, every brand is associated with a category chosen from the predefined Facebook option when creating the public page. This category label is solely determined by the brand and is aligned with its core business (e.g., Walmart is in the category of "retail," Amazon is in the "ecommerce" category). In total, we obtain 5,478 different brands, covering 25 different categories. The largest brand, in terms of number of followers, is Walmart, with 30 million followers. The smallest brand is Bladz Jewelry in the "fashion" category, with 100,000 followers. Figure 2 shows the histogram of number of followers of brand Facebook page. We observe that the data set contains brands with varying popularity, making it representative of brands on Facebook.
Graph: Figure 2. Histogram of number of followers of 5,478 Facebook brands.
On Facebook, firms post on their public fan pages and allow users to comment, like, and share posts. The posts become an important marketing channel for businesses to interact with their customers. We use Facebook Graph API[ 8] to download all activities visible on a brand page such as posts by the brand administrator, as well as posts by users, including comments and likes on brand posts. It is worth emphasizing that to ensure privacy protection, we do not download any user profile information or examine the content of user comments. All engagement activities are represented by unique user identifiers, regardless of whether the user has a public or private Facebook profile, and brand identifiers. The data set collected for this study covers the period from January 1, 2017, through January 1, 2018. In total, we obtain 106,580,172 user–brand engagement activities from 25,992,832 unique users. Because prior research has shown that online interaction is a reflection of broader and even offline interaction ([38]), given the scale of user online engagement in this study, we believe it is a good proxy of how the overall consumer population perceives these brands.
To ensure data quality and robust results (i.e., that the comments on Facebook brand pages reflect genuine user experiences, opinions, and interactions with brands), we design a set of rules, following [54], to remove fake users and their corresponding activities. For example, we find one user who liked posts across 475 different brands. As most users are likely to be interested in far fewer brands, we remove users who like posts on more than 200 brands, which accounts for.01% of the total users and 1.6% of the total user–brand engagement. We also remove users who posted duplicate comments containing URL links. Table 3 describes the data details. The brands' degree distribution (number of connections) exhibits a scale-free distribution (shown in Figure 3), a well-documented phenomenon in most social networks.
Graph: Figure 3. Degree distribution of brands in the user–brand network.
Graph
Table 3. Data Description and Statistics.
| Number of brands | 5,478 |
| Number of users | 25,992,832 |
| Number of unique user–brand interactions | 36,927,613 |
| Number of like interactions | 87,876,623 |
| Number of unique user–brand like interactions | 29,611,805 |
| Number of comment interactions | 18,703,549 |
| Number of unique user–brand comment interactions | 7,612,358 |
| Total number of user–brand interactions | 106,580,172 |
In this section, we extensively evaluate the market structure derived from our approach from both quantitative and qualitative perspectives. We also validate the derived market structure using two external data sources: consumer survey and Google search trend.
With the learned brand representation vectors, we can visualize how the brands are grouped and focus on local fine-grained brand proximity. We use t-SNE to obtain market structure visualization by reducing the learned 300-dimensional brand representations to obtain the associated 2-dimensional visualization map. Figure 4 presents the global structure of the brands in our Facebook data. Each data point in the figure denotes a brand belonging to one of the 25 categories, and each category is indicated by a different color. We interpret the visualization as follows: the closer any two brands are in the figure, the more similar their brand representations are in the 300-dimensional space (see Figure 4). The color codes in the map indicate brands in the same Facebook category, with the category label selected by the brands themselves on Facebook.
Graph: Figure 4. The global structure among brands.
The global Facebook brand market structure map yields several interesting observations. First, there are clear grouping patterns into clusters, particularly between brands in the same industry (points with the same color tend to be in a group). For example, Cluster 1 in Figure 4 (expanded in Figure 5) includes nonluxury domestic and imported automobile brands such as Toyota, Nissan, and Mazda, as well as some automobile accessories brands such as Michelin, DENSO, and Auto Parts. Note that in our data we have several luxury automobile brands such as BMW, Mercedes-Benz, Audi, Tesla, and Maserati, which are not close to the brands in Cluster 1. In fact, they are clustered in a different region of the map with other luxury brands such as Chanel, Gucci, and Cartier. Such a separation between luxury car brands and nonluxury car brands further confirms that brand representation learned from our approach captures latent semantics in multiple dimensions not only on the industry dimension but also on the price and luxury dimensions. The strength of our methodology lies in its ease of capturing these relationships on a single map, which it does by locating thousands of brands in the market structure map and highlighting the complex and possibly overlapping product-market boundaries characterizing these brands. We present a robustness check for different visualization methods in Web Appendix WA5.
Graph: Figure 5. Enhanced view of Clusters 1 (top left), 2 (top right), 3 (bottom left), and 4 (bottom right).
We provide an enhanced view of the four clusters in Figure 5 to examine the fine-grained local market structures. Panel A displays automobile brands along with automobile accessories and motorcycle brands at the top. Panel B displays premium vacation resort brands, such as The Signature at MGM Grand and the Coconut Bay Beach Resort & Spa. Panels C and D contain airline brands and cosmetic brands, respectively. Taken together, these maps provide face validity to our methodology in terms of core brands making up an industry and the overlaps among product-markets.
While visual mapping is sufficient to provide a gestalt picture of all 5,000 plus brands in the aggregate, it does not provide the actual distance between the brand vectors in the reduced dimension space. Because identifying proximal brands for substitute/complement analysis is a critical task in marketing decisions ([ 8]), we focus on identifying proximal brands from the perspective of a focal brand. In doing so, we offer a new perspective that reflects the nature of the varied relationships ranging from substitutes to complements in the social network space.
In this illustration, we choose United Airlines and Southwest Airlines from the airlines category and Audi USA and Nissan from the automobile category, as these brands are generally regarded as having different consumer bases and belonging to different submarkets. Each of the four brands is referred to as a focal brand, and we find their top ten proximal brands according to cosine similarity. Table 4 provides several interesting insights. First, our method is able to capture specific brand latent characteristics. For example, Southwest Airlines is generally considered a low-budget airline compared with United. The brands most proximal to Southwest Airlines and United reflect this difference. The proximal brands for Southwest Airlines are JetBlue, Frontier Airlines, and Allegiant, while the most proximal brands for United are major domestic and international airlines, such as American Airlines, Delta, Lufthansa, All Nippon Airways, Air China, LATAM Airlines, and Air New Zealand. Similar results also are identified in the automobile industry. Second, we observe asymmetric competition (see [42]). For example, Southwest Airlines is the fourth-most-proximal brand to United Airlines, while United Airlines ranks sixth in the set of top proximal brands to Southwest Airlines.
Graph
Table 4. Top 10 Proximal Brands to Each Focal Brand.
| Rank | Focal Brand |
|---|
| United | SouthwestAirlines | Audi USA | Nissan |
|---|
| 1 | American | JetBlue | Mercedes-Benz USA | Mazda |
| 2 | Delta | Frontier | BMW USA | Toyota |
| 3 | Lufthansa | Allegiant | Land Rover | Volkswagen |
| 4 | Southwest | Delta | Lexus | Kia Motors America |
| 5 | Alaska | Alaska | Chevrolet Camaro | Subaru of America |
| 6 | All Nippon | United | Maserati USA | Chrysler |
| 7 | Air China | Airfarewatchdog | Kawasaki USA | Fiat |
| 8 | LATAM | American | Firestone Tires | Jaguar |
| 9 | Air New Zealand | Virgin America | Tesla | Alfa Romeo |
| 10 | Airfarewatchdog | Hyatt | Ram Trucks | Klim |
Third, unlike prior market structure analysis, where proximal brands are usually from the same industry as the focal brand, the top most proximal brands derived from our analysis are from different industries. For example, a brand called "Airfarewatchdog" is proximal to both United and Southwest Airlines. Airfarewatchdog is a deal-finder for flight tickets and has a large follower base (over 1 million) on Facebook. Traditional market analysis would simply ignore this brand, as it is not an airline. Further, it is also interesting to see that Southwest Airlines is closer to Airfarewatchdog than to United, which may indicate that the fans of Southwest Airlines are more likely to use a deal finder before purchasing flight tickets; thus, Airfarewatchdog could be a complement to Southwest when customers look for cheap flights on that site and end up at Southwest, or it could potentially compete with Southwest. In either case, Southwest could focus more on this site and examine the nature of the relationship.
Our market structure map can help managers identify brands outside of the product-market that are close to a specific brand and, thus, identify opportunities and threats posed by different brands. Take the airline product-market (Figure 5, Panel C) as an example. In our analysis, Disney Cruise Line and Hyatt are two brands outside of the airline product-market but are identified as proximal brands to Southwest but not for United. These proximal locations simply are due to a greater number of users in our data set liking both Southwest and Hyatt ( 2,709) versus the number of users liking both United and Hyatt (954). Similarly, a greater number of users like both Southwest and Disney Cruise Line ( 3,050) than like both United and Disney Cruise (729).
Such findings can provide opportunities for Southwest, as it could target users who like Disney Cruise and Hyatt on social media. Southwest could cross-promote these brands by teaming up with Disney Cruise and/or Hyatt on each other's websites and launch coalition loyalty programs. From the viewpoint of other hotel chains that are competitors to Hyatt, these could be potential threats, so gleaning such insights early on may help them take proactive actions. Such opportunities/threats are difficult to identify when product-markets are prespecified, and they cannot be obtained easily through other means.
Our user engagement data set contains top 5,478 primarily large brands, ranked by their popularity (number of followers as of data collection period) on Facebook. A key question is whether our proposed approach is still able to identify meaningful market structure for smaller brands. If they can find the right position in the product-market structure, smaller brands have the potential to increase consumer awareness and interest in their brands ([13]), which could lead to a permanent benefit in terms of competitive advantage ([47]). Therefore, to test whether our methodology is able to capture relationships among large brands as well as small and local business brands, we add a set of smaller brands to the original data set. Specifically, we focus on the "Travel" category, as it includes many small, local travel agencies, and their followers on Facebook range from a few hundred to a few thousand on average. In total, we have 241 travel brands. Figure 6, Panels A and B, plot the distribution of the number of followers of these travel brands and shows that it is quite diverse.
Graph: Figure 6. Size and market structure of 241 travel brands.
Upon applying our methodology to the enlarged data set, we observe (Figure 6, Panel B) that these 241 travel brands are predominantly located in two areas. This pattern indicates that the latent brand relationship is well captured, even when brands have few engagement activities due to their smaller user bases. In a brand–brand network, such a small number of shared user bases could result in a failure to capture proximal locations, in essence treating them as noise.
The market structure uncovered for these small businesses by identifying their proximal brands has good face validity. For example, "The Luxury Travel Expert" is an information portal for luxury travel and premium tours, with about 11,000 followers on Facebook as of our data collection period. Most posts receive fewer than ten comments and likes. The top proximal brands based on the cosine similarity are Smithsonian Journeys, The Peninsula Beverly Hills, Peter Sommer Travels, Quasar Expeditions, and DuVine Cycling. It is noteworthy that these are small travel brands that focus on expert-led, small-group, luxury, and premium tours. The results further confirm that our deep network representation learning method is generalizable to both small and large brands. This analysis also allows brand marketing managers to identify business opportunities. For example, in our analysis, the two brands The Luxury Travel Expert and The Peninsula Beverly Hills are quite close. The former is an information portal for luxury travel and premium tours, and the latter is a five-star luxury hotel. Therefore, the marketing manager of the Peninsula Beverly Hills could promote the brand on the information portal website to attract users from The Luxury Travel Expert to expand its customer base.
Extant methods typically predefine the product-market boundary to derive market structure and brand relationships. In contrast, we allow product-market boundaries to emerge from the data. Therefore, a natural question is whether it is necessary to have a broader range of brands from other industries to derive a highly precise market structure for a specific industry. Although managers would typically focus on engagement data for their brands and for brands within the same industry, how does engagement data from brands in different categories help? To answer this question, we choose the "auto" category and only use the engagement data from the auto brands to derive the market structure. In the data set, we have 163 auto brands, including cars and car accessories brands (e.g., tires, oil), with 2.7 million user engagements in total. The analysis shows (Figure 7, Panels A and B) that structures with reasonable face validity still emerge using only the auto brands data. For example, the top left corner in Figure 7, Panel B, presents a cluster of imported auto brands such as Kia Motor America, Toyota, and Nissan. However, compared with the derived auto brand market structure learned from using all brand data, as shown in Figure 5, Panel A, the market structure is less clustered and more ambiguous.
Graph: Figure 7. Visualization of market structure of using engagement data only from "auto" brands. Notes: The right panel is the zoomed in visualization with BMW as centroid.
Next, we compare the market structure using the engagement data from the auto brands alone with that from all brands across categories in a qualitative manner. Specifically, we choose the brand FMF Racing, which is a company that develops dirt bike exhausts for off-road or racing motocross riding. Using the engagement data from the auto brands alone, the top proximal brands are Lucas Oil, KTM USA, Yamaha Motor, Arctic Cat, Two Brothers Racing, Phoenix Pro Scooters, Auto Alliance, Valvoline USA, Lance Camper, and Castrol. Some are related to off-road motocross riding, while others are not. For example, Lucas Oil, Valvoline USA, and Castrol are global automotive oil brands.
In contrast, the top ten proximal brands to FMF Racing emerging from using all categories of data are KTM USA, Polaris Snowmobiles, Fox Racing, Mickey Thompson Performance Tires & Wheels, Two Brothers Racing, King Shocks, Arctic Cat, Addictive Desert Designs, NISMO, Skunk2 Racing, and MBRP performance exhaust. Upon further investigation, we find that they are all related to off-road motocross riding. These results indicate that our approach with engagement data from brands across industries can learn better brand representation and thus reveal a highly precise market structure.
To assess the external validity of our approach, we conduct a survey on Amazon Mechanical Turk (MTurk), which is a reliable source for data collection and marketing analytics ([43]). Prior market structure literature has also administered brand perception survey on MTurk ([ 7]). Following this prior study, we surveyed 28 automobile brands (after ignoring the other 150 brands that are related to motorcycle or automobile accessory such as tires, parts, and oil). Specifically, we recruited 500 MTurk participants, each of whom was required to be in the United States and have a good MTurk record (successful completion of at least 100 assignments with a minimum 95% rate of approval). Each participant was asked to rate the similarity between a focal automobile brand and the other 27 automobile brands on a scale of one to five. To avoid fatigue due to information overload, each participant was randomly assigned to work on one task. Participants were also asked to indicate their age, gender, and whether they owned an automobile. Details of the participants' demographics information and the survey design are presented in Web Appendix WA6.
In the survey, participants could choose "N/A" if they were not aware of the automobile brands. Brand recognition rate was 88.2%, implying that 11.8% of ratings were not applicable due to lack of brand awareness. We aggregated the survey data and built a 28 × 28 matrix, where each cell represented the pairwise brand similarity, and denoted it as the "survey matrix." We also used the brand representations learned from our approach to construct another 28 × 28 matrix of brand similarity, which we denoted as the "deep learning–based matrix." The correlation between two matrices is significantly positive (r = .385, p = .000). This result provides additional evidence on the validity of our deep learning–based approach for market structure identification. We also did an additional check where we calculated the correlation between the survey response and that constructed by our approach but using only automobile (within-industry) data. The correlation is.152 (p < .05), which is not as substantial and significant as the correlation between the survey response and our approach using all industry data. We present the market structure learned from the survey data in Web Appendix WA7 as an external validity.
To provide further external validity of our approach, we use Google Trends data to identify market structure and examine how it aligns with our approach of using online social media users' brand engagement. Google Trends provides an interest score for every search query across regions and languages, as measured by an aggregated search volume over time. A higher interest score means that queries are more popular in a specific region and time. Google Trends data have been widely used by industry ([44]) and academia ([ 6]; [ 9]; [22]; [48]) to address marketing and economic problems (e.g., competitive analysis). Researchers have also shown that this score is consistent with consumers' purchase interest in general ([ 6]; [ 9]).
To determine relative popularity for every pair of brands, we make a search query consisting of two brand names—for example, "Toyota BMW" for the brands Toyota and BMW. For every brand pair, we can obtain an interest score returned by Google. For example, in the United States in 2017, the interest score is 13 and 85 for the query "Toyota BMW" and "Toyota Honda," respectively. This indicates that consumers in general are more interested in searching Toyota and Honda together, compared with searching Toyota and BMW together.
In the first validation exercise, we focus on the airline industry and the derived market structure. We have 19 airline brands in our data set, including U.S. domestic airlines and international airlines (Figure 5, Panel C). For every brand pair, we first obtain a Google search interest score in the U.S. region in 2017 (the same as our engagement data period). Then, following previous work ([35]), we calculate the similarity between two brands A and B as , where is the set of all brands (e.g., 19 here). [35] use the co-occurrence of two brands in an online discussion forum instead of a Google search interest score. We also calculate similarity for every pair of 19 airline brands using 300-dimensional vectors derived from our deep network representation learning on the engagement data using cosine similarity.
To check whether the two aforementioned similarity scores are similar to each other, we calculate their Pearson's two-tailed correlation between two sets of 361 (= 19 × 19) similarity scores. It is significantly and highly correlated ( ). This indicates that our social engagement-based market structure is similar to that derived from Google Trends. Because prior studies have shown that the Google search data have a high correlation with a consumer's actual purchase interest ([ 6]; [ 9]), we can conclude that users' social engagement with brands also contains valuable information for deriving brand relationships.
In the second validation exercise, we focus on the travel industry, including not only major travel brands but also many small and local travel brands (see the "Large Brand Versus Small Brand" subsection). There are 241 travel brands in the data set. Similar to the first validation exercise, for every brand pair, we obtain a Google search interest score in the United States in 2017 (the same as our engagement data period). Among the 241 travel brands, Google Trends does not return scores for 90 brands (i.e., showing "your search doesn't have enough data to show here"), which results in data for 151 remaining brands. Although individual brands show a considerable amount of search, only four brand pairs return nonempty interest scores.[ 9] This data sparsity may be attributed to the uniqueness of the travel category. Many of the travel brands are local/small businesses, such as the travel agencies "Spirit of Boston" and "Historic Philadelphia." Naturally, they do not receive as many queries as large brands. Moreover, consumers may search travel agency brands in different queries, but they very rarely search two travel brands in the same query. Therefore, there is not enough data for Google to aggregate and return the cosearch score. This analysis highlights the limitation of the cosearch-based approach, which is likely to suffer from the data sparsity issue. In contrast, our approach built on large-scale brand–user social engagement data can provide valuable marketing insights not only for large international brands but also for small local brands.
In a practical setting, marketing managers may need to quantitatively determine the quality of derived market structure maps, based on which they can infer actionable insights. We evaluate the conceptual maps using a standard metric—silhouette score ([ 1])—which has been adopted in prior market structure literature ([12]). The silhouette coefficient is calculated using the mean intracluster distance (a) and the mean nearest-cluster distance (b) for each sample, as . The values of silhouette score range between −1 and 1 (1 being the best and −1 the worst). Values near 0 indicate overlapping clusters. Negative values generally indicate that a sample has been assigned to the wrong cluster. Recall that our approach can naturally group brands that have similar representations in the high-dimensional space. An ideal market structure would favor brands that are concentrated and exhibit clean cluster structures. We conduct K-means clustering on the brand representations and compute the mean silhouette coefficient of all samples.
In the "Within-Industry Market Structure Analysis" subsection, we qualitatively show that our approach—without prespecifying product-market—reveals more interesting and coherent brand insight than using brand engagement data within only one industry. Next, we vary the number of clusters in K-means and calculate the silhouette coefficient of different methods. The result in Figure 8 shows that our approach using all brand engagement data consistently achieves better clustering than using only the automobile brand engagement data. For example, when we cluster 168 automobile brands into two clusters (i.e., K = 2 in K-means), our approach achieves a silhouette coefficient of.334, while the approach using only the automobile engagement data has a low silhouette coefficient of.043. The silhouette coefficient of our approach gradually converges to.10 as the number of clusters increases. In contrast, the approach using only automobile industry data stays near.01, indicating a poor separation among automobile brands. This analysis not only confirms the superiority of our approach without prespecifying product-market boundary but also enables marketing managers to determine the quality of derived market structure maps.
We have shown that our approach can derive good market structure with large-scale social engagement data. In practice, it is easy to obtain a relatively comprehensive set of brands across different categories and associated user engagement from social media marketing platforms, such as Socialbakers. However, a marketing manager may not have enough resources to collect as large a data set as we have, raising the question as to whether our approach is sensitive to the size of data for obtaining a good market structure. To answer this question, we calculate the correlation between the similarity of pairwise brands generated using the full data set and that generated by a fraction of data selected at random. Figure 9 presents this result, showing that the correlation reaches over.90 when 40% data is used (and.7 when 12.5% data is used), and it starts to converge to the market structure generated by using the full data. This analysis suggests that our approach is relatively robust to the amount of engagement data used. Marketing managers can use this analysis as guidance to determine the amount of data resources needed. In addition, we examine how the number of prespecified industries affects the robustness of market structure maps and present the analysis in Web Appendix WA8.
Graph: Figure 8. The clustering silhouette coefficient of 168 automobile brands.
Graph: Figure 9. Pearson correlation between the similarity of pairwise brands generated using a percentage of full data and the full data.
In studying market structure, there is a lack of ground truth about the identified structure, that is, an understanding of what the "true" structure is, which makes demonstrating the performance of various proposed methods challenging. We introduce an alternative approach, adopted from network analysis literature ([27]), to evaluate the identified market structure. An identified market structure is a function of the brand representation, and so an accurate representation is more likely to identify valid market structures. This approach is supported by prior research showing a strong relationship between brand image and the characteristics of a brand's supporters and followers ([ 7]; [25]; [34]). If a network learning method were capable of accurately representing network nodes accounting for these relationships between brands and users, then it would be able to predict the future links between brands and users accurately. Therefore, we use a cross-validation procedure under a link prediction research design, where we predict the most likely newly formed links of user–brand engagement in an out-of-sample network given the brand vectors and user vectors learned from a training network. This research design is widely used in the network analysis community to evaluate network clustering algorithm performance ([27]; [51]). In our context, we use the user–brand interactions from the first half of the time span in our data to build a training network (G0,1) and use the second half to build a testing network (G1,2). The likelihood of a link formation is measured by the proximity of a learned brand vector and a learned user vector. Note that link prediction performance is significantly correlated with the quality of learned vectors, given the assumption that a better network representation learning can predict new interactions between users and brands with a high degree of accuracy. We provide details of the link prediction experiments in Web Appendices WA2–WA4. Overall, our analysis shows that ( 1) link prediction using representation learned from our brand–user network performs better than a reduced brand–brand network (a widely used method in extant approaches), ( 2) deep learning–based methods learn better representation than shallow machine learning methods, and ( 3) our deep learning–based model is robust and able to handle sparse networks as compared with baselines.
Market structure evolves over time and can change dramatically, especially under an unexpected industry shock. Whether our proposed method can be adaptively learned is also of interest as it could provide useful insights to marketing practitioners. In this section, we analyze how market structure changes under exogenous shocks by analyzing two case studies: Amazon acquiring Whole Foods and Tesla introducing the Model 3. We take a before-and-after strategy where we use data for the three months pre- and postevent announcement day and calculate the change in distance from the focal brand (i.e., Amazon and Tesla) to other representative brands selected from the same category. The purpose of the event study is to examine how a focal brand relationship with other brands changes as a major event occurs. Specifically, for Amazon–Whole Foods, we select several brands from the retail and e-commerce category, and for Tesla, we select several brands from the automobile category. We calculate the change between focal brand 's representation and target brand 's representation before and after the specific event using cosine similarity: . Therefore, positive numbers indicate a similarity increase, whereas negative numbers indicate a decrease in similarity.
Amazon acquired Whole Foods in June 2017. This acquisition has had a significant impact on the grocery and retail industries. At the time, it was widely believed that Amazon planned to use its acquisition of Whole Foods to enter the online grocery delivery business. Amazon and Whole Foods ran separate Facebook pages. After the merger of the two firms, we see from Figure 10 that Amazon is more proximal to retail brands as measured by cosine similarity, while the proximity to other relevant brands decreases slightly. For example, the cosine similarity between Amazon and Lowe's Home Improvement decreases by.184. In contrast, the cosine similarity between Amazon and other super-market retailer brands increases. Among them, proximity of Amazon to Whole Foods increases by.202, and between Amazon and Kroger by.165. As inferred from our data-driven model, Amazon even becomes more proximal to Walmart, indicating that Amazon's competitive market structure landscape has shifted. By further examining our data, we find that, after the Whole Foods acquisition, the number of common users who interact with both Amazon and Whole Foods on their public Facebook pages increases. Some Amazon users posted comments on Whole Foods' fan page mentioning Amazon. For example, in the Whole Foods post, "Here are 6 New Healthy Products Coming to Whole Foods in March," a user who had liked an Amazon post earlier commented, "You mean AMAZON... as they bought Whole Foods...right?" This direct link between Amazon and Whole Foods leads the deep autoencoder to increase the proximity between the two brands. Moreover, in another Whole Foods post, a user who had liked a Kroger post earlier posted, "The quality has gone downhill and prices have soared.... You've made Kroger look appealing...." Although we do not find that this user has ever interacted with Amazon's Facebook page before, her interaction with Whole Foods leaves an implicit connection between Amazon and Kroger, which can be captured by the deep autoencoder. In short, after Amazon acquired Whole Foods, online social media users who are Amazon's fans pay more attention to Whole Foods, and users who are fans of other supermarket brands engage more with Whole Foods due to the acquisition event. As a result, the deep autoencoder captures the dynamics and updates the brand representation accordingly.
Graph: Figure 10. Similarity change of Amazon to other brands in retail and e-commerce industry.
The acquisition by Amazon has an impact on the market structure of Whole Foods as well. In Figure 11, we consider Whole Foods as the focal brand and calculate the change in proximities to other brands before and after the acquisition. Drawing on the results, we observe that Whole Foods' proximity to other retail brands such as Target, Walmart, and Best Buy increases. Among them, the proximity to Amazon increases the most due to the increase in the number of common users between them. In contrast, Whole Foods' proximity to supermarket brands such as Goya Foods, Enjoy Life Foods, and HelloFresh decreases slightly. Second, the magnitude of change in proximity values is smaller than those of Amazon to other brands. This seems to indicate that the acquisition has had less impact on Whole Foods, as it is still positioned around other supermarket brands, while Amazon is expanding closer to the grocery retail category.
Graph: Figure 11. Similarity change of Whole Foods to other brands in retail and e-commerce industry.
Although this analysis is retrospective, it highlights that our approach offers managers a series of multiple snapshots of the structure over time to measure a brand's relative position change, thus identifying potential market structure change. Suppose a supermarket chain brand A observes that Amazon is moving closer to A's position on the map. This may indicate that Amazon is getting more engagements (likes or comments) from A's customers. Given that one motivation of liking a brand's Facebook post is to receive some benefit from the brand (e.g., coupon, discount), it could further indicate that Amazon is conducting effective promotional marketing campaigns on social media. No matter the underlying reasons, the increasing proximity of Amazon on the brand map can at least provide an early warning to A's marketing managers to the potential threat. Late response to the competition may harm the brand and eventually the whole business.
Next, we validate the case study of Amazon's Whole Foods acquisition using Google Trends data. Similar to the first external validity exercise, we choose 29 "retail" brands (including Walmart, Target, Macy's, Best Buy, Walgreens, Lowe's, Whole Foods, IKEA, Sears, 7-Eleven, Dollar General, Sam's Club, Dollar Tree, CVS Pharmacy, Aldi, Barnes & Noble, Costco, Kroger, Meijer, Safeway, Office Depot, Rite Aid, Albertsons, ShopRite, and The Fresh Market) plus Amazon and obtain their interest scores for every brand pair in the United States in 2017. Note that we exclude some small retail brands such as Goya Foods because their Google cosearch interest scores with other brands are mostly 0, indicating insufficient search data for the brand.
The Pearson's two-tailed correlation between two sets of 900 (= 30 × 30) similarity scores is significantly high before ( ) and after ( ) acquisition. This result confirms the external validity of our social engagement–based method. We observe that for Amazon, before the Whole Foods acquisition, the most similar brands were Barnes & Noble, Macy's, and Best Buy. After the acquisition, the most similar brands are Whole Foods, Barnes & Noble, and Macy's. For Whole Foods, before the acquisition the most similar brands were The Fresh Market, Albertsons, and ShopRite. After the acquisition, the most similar brands are The Fresh Market, Amazon, and Safeway.
We obtain further search interest data for one year after the acquisition (June 2017 to June 2018) to examine whether the market structure change is sustained for a long period after the acquisition announcement. For Amazon, the most similar brands are still Whole Foods, Barnes & Noble, and Macy's. Other grocery "retail" brands such as Kroger and The Fresh Market become more similar to Amazon than before the acquisition. For Whole Foods, the most similar brands are The Fresh Market, Safeway, ShopRite, and Amazon. Because Whole Foods is still Amazon's most similar brand among these retailer brands, this indicates that for Amazon, the acquisition impact holds for the extended period of analysis. It seems that the acquisition has less of an impact on Whole Foods, as Whole Foods is still positioned around other supermarket brands. All findings are consistent with our case study using social engagement data, which provides external validity to our results.
Tesla sells two types of sedans: the Model S and the Model 3. The Model S is a luxury premium sedan with a larger range of acceleration and customization options, while the Model 3 is designed as a more affordable mass-market electric vehicle. The Model S can cost over $100,000 depending on the configuration, while the Model 3 costs approximately $35,000. After the announcement of the new Model 3, we see that Tesla becomes more distant from luxury car brands and moves closer to nonluxury car brands. We can see in Figure 12 that the cosine similarity between Tesla and the luxury car brand Maserati decreases by.209. Similar trends exist between Tesla and other high-end or luxury car brands such as BMW, Mercedes-Benz, Audi, and so on. Meanwhile, Tesla becomes more proximal to Kia, Mazda, and other more affordable car brands.[10]
Graph: Figure 12. Similarity change of Tesla to other selected brands in the auto industry.
In the previous analysis, we compute the distance change between the focal brand (i.e., Amazon or Whole Foods) and other brands before and after the acquisition. We can see that there is a significant increase in similarity between Amazon and Whole Foods after the acquisition. However, whether this distance change is caused by the acquisition or other unobserved factors, such as the difference of data split and/or noise, still remains unclear. Therefore, we conduct further analysis by randomly splitting all data before the acquisition into two parts (i.e., d1 and d2, with d1 before d2). We then measure the distance between Amazon and Whole Foods using d1 and d2 separately. We repeat this process 30 times using different data cuts in the preacquisition data. The average distances between the two brands across using all d1s and d2s are.228 and.232, respectively. The two-tailed t-test on the distance is.055, which indicates that there is no statistically significant difference between the distances between Amazon and Whole Foods before the acquisition in different cuts of the preacquisition data. Accordingly, the substantial increase in similarity between Amazon and Whole Foods is not attributed to sample differences.
We perform a similar process on Tesla's introduction of the Model 3. In particular, we choose one nonluxury brand, Mazda, and compute its distances to Tesla before the event using various data splits. The average distances between Mazda and Tesla across using all d1s and d2s are.185 and.191, respectively, with a p-value of.076. This seems to indicate there is no statistically significant difference between Mazda and Tesla when the cutting point of data varies before the event. Therefore, we conclude that after Model 3's announcement, Tesla becomes more similar to nonluxury automobile brands on the social media platform. Note that we also conduct analyses on Tesla and other automobile brands, and the results are consistent.
Our proposed approach examines millions of user engagements with thousands of brands and focuses on the early stage of the customer journey. This allows for visualization of potentially overlapping product-market boundaries across many categories and helps managers identify latent threats and potential opportunities, which cannot be done with extant methods that focus on later stages of the customer journey (lower levels of the purchase funnel) within categories. As an example, for Southwest, is Airfarewatchdog a potential competitor that might draw visitors away, or is it a complementor that would increase visits to Southwest? Having identified the overlapping market with Airfarewatchdog, Southwest could invest more attention to evaluate the exact nature of this relationship. If Airfarewatchdog is a competitor, then Southwest might focus on developing strategies to differentiate itself and channel visitors to its website exclusively. If it is a complementor, then Southwest might run display ad campaigns on Airfarewatchdog's website. In addition, both Disney Cruise and Hyatt are closely associated with Southwest, with common users who like these brands on social media; therefore, Southwest could run mutually beneficial joint and cross-promotions with these other brands. In fact, all these brands could join in a dynamic coalition loyalty ecosystem built around a fluid partnership of products, services, and experiences, thereby providing a unifying customer value proposition that could be difficult to compete against ([ 4]). Identifying such unusual or unforeseen insights is the greatest advantage of our approach.
Another important strategic use of our market structure maps is to identify competitors and complementors across industries and track how these relationships change over time. While [15] apply text analysis to 10-K statements to identify such grouping based on product descriptions that the firms provide, we provide a more dynamic structure based on actual customer/user social media activities. Moreover, our market structure map is more forward looking and predictive of emerging competition and complementors and more proactive than those based on 10-K statements, which can be viewed as reactive. Because [14] show that merging firms with more similar product descriptions in their 10-Ks results in more successful outcomes, using our market structure maps to identify merger-and-acquisitions targets (firms sharing common users) may have similar benefits. We leave this for future research.
The power of our method lies in its ability to capture the dynamic changes in market structure. Because the maps are based on the analysis of big data that can be collected in a relatively short window of time, our methodology can track changes in their relative position when firms introduce new products, new promotions, and new marketing initiatives. The case studies that we highlighted provide good illustrations of this. In addition, although we have not analyzed this in the article, firms can deploy our method to enhance their social network-based marketing efforts by better targeting specific potential customers, because user nodes in the network are also learned and represented as vectors in the same multidimensional space as brands. Our link prediction design demonstrates a possible use for targeting. Finally, our proposed method is generalizable to other similar platforms if we can construct a brand–user network from public fan pages' engagement data. We implemented our proposed method using NVIDIA P100 graphics processing unit, with Tensorflow as the back-end deep learning framework. For future research to replicate or practitioners to adopt, we have provided details regarding data collection, data cleaning, and deep model architecture, and model the fine-tuning process in Web Appendix WA1.
While marketing analytics techniques have extensively used consumer personal data to derive valuable insights, they raise many privacy and ethical concerns. How to balance these two important aspects has become a key consideration for many marketing scholars ([ 3]; [52]). Our approach provides a useful example. The only input to our network representation learning method is the brand–user network, which can be publicly obtained from brands' social media page.
Our research has some limitations. Given the nature of our data, our method cannot examine stockkeeping unit–level competition as is done by some of the extant methodologies using lower funnel data. From this perspective, we recommend our methodology as a complement to extant methods and for higher-level brand strategies and tactics. Future work could examine how perceptual maps vary by customer segment using lower-funnel data such as purchase frequency and purchase amount. Second, our analysis is conducted on one social network, Facebook. Even though Facebook is one of the largest online social networks, with billions of users and thousands of brands, it is likely that users on different platforms exhibit different engagement behavior, and some of the research findings may not be generalized to other platforms. For example, it is reported that Instagram users and Facebook users fall into different age groups (Pew Research Center 2021). We could apply the same technique to other social media platforms and compare findings. Finally, each link in the user–brand network is created when the user engages with the brand on the public page. Facebook has introduced various reaction emotions to the platform to allow users interact with brands in different ways, such as "Like," "Love," "Care," "Haha," "Wow," "Sad," and "Angry." Future work could build a multirelational network to deeply capture brand–user engagement heterogeneity.
sj-pdf-1-jmx-10.1177_00222429211033585 - Supplemental material for Identifying Market Structure: A Deep Network Representation Learning of Social Engagement
Supplemental material, sj-pdf-1-jmx-10.1177_00222429211033585 for Identifying Market Structure: A Deep Network Representation Learning of Social Engagement by Yi Yang, Kunpeng Zhang and P.K. Kannan in Journal of Marketing
Footnotes 1 Koen Pauwels
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship and/or publication of this article.
4 P.K. Kannan https://orcid.org/0000-0003-0738-0766
5 In fact, obtaining detailed user personal information for marketing analysis (e.g., political targeting) is controversial and subject to ethical concerns, such as those raised during the Facebook–Cambridge Analytica data scandal.
6 Note that other emerging dimension reduction methods, such as uniform manifold approximation and projection (UMAP; [28]), can be applied for visualizing high-dimensional data in a low-dimensional space. Similar to t-SNE, UMAP is good at preserving global structure and is efficient for a large and noisy data set. Web Appendix WA5 presents our comparison of different visualization methods.
7 Socialbakers is a social media marketing company offering a marketing software-as-a-service platform called the Socialbakers Suite. It includes data from Facebook, Twitter, and YouTube (https://www.socialbakers.com/).
8 https://developers.facebook.com/docs/graph-api/.
9 Those brand pairs are (Carnival Cruise Line, Disney Cruise Line), (Carnival Cruise Line, Princess Cruises), (Princess Cruises, Celebrity Cruises), and (Interval International, RCI).
Such analysis has descriptive uses for managers, as we have discussed. If many such events are tracked to measure how they impact the structure, they could be used in a prescriptive sense.
References Pew Research Center (2021), "Social Media Fact Sheet," (April 7), https://www.pewresearch.org/internet/fact-sheet/social-media/.
Agrawal Ajay , Gans Joshua , Goldfarb Avi. (2018), Prediction Machines: The Simple Economics of Artificial Intelligence. Boston: Harvard Business Press.
Bleier Alexander , Goldfarb Avi , Tucker Catherine. (2020), " Consumer Privacy and the Future of Data-Based Innovation and Marketing ," International Journal of Research in Marketing , 37 (3), 466 – 80.
Boudet Julien , Huang Jess , Rothschild Phyllis , von Difloe Ryter. (2020), " Preparing for Loyalty's Next Frontier: Ecosystems ," McKinsey & Company (March 5) , https://www.mckinsey.com/business-functions/marketing-and-sales/our-insights/preparing-for-loyaltys-next-frontier-ecosystems.
Carpenter Gregory S. , Lehmann Donald R.. (1985), " A Model of Marketing Mix, Brand Switching, and Competition ," Journal of Marketing Research , 22 (3), 318 – 29.
Choi Hyunyoung , Varian Hal. (2012), " Predicting the Present with Google Trends ," Economic Record , 88 (S1) , 2 – 9.
Culotta Aron , Cutler Jennifer. (2016), " Mining Brand Perceptions from Twitter Social Networks ," Marketing Science , 35 (3), 343 – 62.
Day George S. , Shocker Allan D. , Srivastava Rajendra K.. (1979), " Customer-Oriented Approaches to Identifying Product-Markets ," Journal of Marketing , 43 (4), 8 – 19.
Du Rex Yuxing , Kamakura Wagner A.. (2012), " Quantitative Trendspotting ," Journal of Marketing Research , 49 (4), 514 – 36.
Erdem Tülin. (1996), " A Dynamic Analysis of Market Structure Based on Panel Data ," Marketing Science , 15 (4), 359 – 78.
France Stephen L. , Ghose Sanjoy. (2016), " An Analysis and Visualization Methodology for Identifying and Testing Market Structure ," Marketing Science , 35 (1), 182 – 97.
Gabel Sebastian , Guhl Daniel , Klapper Daniel. (2019), " P2V-MAP: Mapping Market Structures for Large Retail Assortments ," Journal of Marketing Research , 56 (4), 557 – 80.
Hanssens Dominique M. , Pauwels Koen H. , Srinivasan Shuba , Vanhuele Marc , Yildirim Gokhan. (2014), " Consumer Attitude Metrics for Guiding Marketing Mix Decisions ," Marketing Science , 33 (4), 534 – 50.
Hoberg Gerard , Phillips Gordon. (2010), " Product Market Synergies and Competition in Mergers and Acquisitions: A Text-Based Analysis ," Review of Financial Studies , 23 (10), 3773 –3 811.
Hoberg Gerard , Phillips Gordon. (2016), " Text-Based Network Industries and Endogenous Product Differentiation ," Journal of Political Economy , 124 (5), 1423 – 65.
Hoffman Donna L. , Novak Thomas P. , Kang Hyunjin. (2017), " Let's Get Closer: Feelings of Connectedness from Using Social Media, with Implications for Brand Outcomes ," Journal of the Association for Consumer Research , 2 (2), 216 – 28.
Ilhan Behice Ece , Kübler Raoul V. , Pauwels Koen H.. (2018), " Battle of the Brand Fans: Impact of Brand Attack and Defense on Social Media ," Journal of Interactive Marketing , 43 (August) , 33 – 51.
John Leslie K. , Emrich Oliver , Gupta Sunil , Norton Michael I.. (2017), " Does 'Liking' Lead to Loving? The Impact of Joining a Brand's Social Network on Marketing Outcomes ," Journal of Marketing Research , 54 (1), 144 – 55.
John Deborah Roedder , Loken Barbara , Kim Kyeongheui , Monga Alokparna Basu. (2006), " Brand Concept Maps: A Methodology for Identifying Brand Association Networks ," Journal of Marketing Research , 43 (4), 549 – 63.
Kalwani Manohar U. , Morrison Donald G.. (1977), " A Parsimonious Description of the Hendry System ," Management Science , 23 (5), 467 – 77.
Kannan P.K. , Sanchez Susan M.. (1994), " Competitive Market Structures: A Subset Selection Analysis ," Management Science , 40 (11), 1484 – 99.
Kannan P.K. , Wright Gordon P.. (1991), " Modeling and Testing Structured Markets: A Nested Logit Approach ," Marketing Science , 10 (1), 58 – 82.
Kim Jun B. , Albuquerque Paulo , Bronnenberg Bart J.. (2011), " Mapping Online Consumer Search ," Journal of Marketing Research , 48 (1), 13 – 27.
Kim Ho , Hanssens Dominique M.. (2017), " Advertising and Word-of-Mouth Effects on Pre-Launch Consumer Interest and Initial Sales of Experience Products ," Journal of Interactive Marketing , 37 , 57 – 74.
Kübler Raoul V. , Colicev Anatoli , Pauwels Koen H.. (2020), " Social Media's Impact on the Consumer Mindset: When to Use Which Sentiment Extraction Tool? " Journal of Interactive Marketing , 30 (May), 136–55.
Kuksov Dmitri , Shachar Ron , Wang Kangkang. (2013), " Advertising and Consumers' Communications ," Marketing Science , 32 (2), 294 – 309.
Lee Thomas Y. , Bradlow Eric T.. (2011), " Automated Marketing Research Using Online Customer Reviews ," Journal of Marketing Research , 48 (5), 881 – 94.
Liben-Nowell David , Kleinberg Jon. (2007), " The Link-Prediction Problem for Social Networks ," Journal of the American Society for Information Science and Technology , 58 (7), 1019 – 31.
McInnes Leland , Healy John , Melville James. (2018), " UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction ," arXiv preprint , https://arxiv.org/abs/1802.03426.
McPherson Miller , Smith-Lovin Lynn , Cook James M.. (2001), " Birds of a Feather: Homophily in Social Networks ," Annual Review of Sociology , 27 (1), 415 – 44.
Mikolov Tomas , Chen Kai , Corrado Greg , Dean Jeffrey. (2013), " Efficient Estimation of Word Representations in Vector Space ," arXiv , https://arxiv.org/abs/1301.3781.
Mochon Daniel , Johnson Karen , Schwartz Janet , Ariely Dan. (2017), " What Are Likes Worth? A Facebook Page Field Experiment ," Journal of Marketing Research , 54 (2), 306 – 17.
Moe Wendy W.. (2006), " An Empirical Two-Stage Choice Model with Varying Decision Rules Applied to Internet Clickstream Data ," Journal of Marketing Research , 43 (4), 680 – 92.
Nam Hyoryung , Joshi Yogesh V. , Kannan P.K.. (2017), " Harvesting Brand Information from Social Tags ," Journal of Marketing , 81 (4), 88 – 108.
Naylor Rebecca Walker , Lamberton Cait Poynor , West Patricia M.. (2012), " Beyond the 'Like' Button: The Impact of Mere Virtual Presence on Brand Evaluations and Purchase Intentions in Social Media Settings ," Journal of Marketing , 76 (6), 105 – 20.
Netzer Oded , Feldman Ronen , Goldenberg Jacob , Fresko Moshe. (2012), " Mine Your Own Business: Market-Structure Surveillance Through Text Mining ," Marketing Science , 31 (3), 521 – 43.
Novak Thomas P.. (1993), " Log-Linear Trees: Models of Market Structure in Brand Switching Data ," Journal of Marketing Research , 30 (3), 267 – 87.
Novak Thomas P. , Stangor Charles. (1987), " Testing Competitive Market Structures: An Application of Weighted Least Squares Methodology to Brand Switching Data ," Marketing Science , 6 (1), 82 – 97.
Pauwels Koen , van Ewijk Bernadette. (2020), " Enduring Attitudes and Contextual Interest: When and Why Attitude Surveys Still Matter in the Online Consumer Decision Journey ," Journal of Interactive Marketing , 52 (November) , 20 – 34.
Pelletier Mark J. , Horky Alisha Blakeney. (2015), " Exploring the Facebook Like: A Product and Service Perspective ," Journal of Research in Interactive Marketing.
Pennington Jeffrey , Socher Richard , Manning Christopher D.. (2014), " Glove: Global Vectors for Word Representation ," in Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing. Stroudsburg, PA: Association for Computational Linguistics, 1532 – 43.
Pereira Hélia Gonçalves , Salgueiro Maria de Fátima , Mateus Inês. (2014), " Say Yes to Facebook and Get Your Customers Involved! Relationships in a World of Social Networks ," Business Horizons , 57 (6), 695 – 702.
Ringel Daniel M. , Skiera Bernd. (2016), " Visualizing Asymmetric Competition Among More Than 1,000 Products Using Big Search Data ," Marketing Science , 35 (3), 511 – 34.
Rousseeuw Peter J.. (1987), " Silhouettes: A Graphical Aid to the Interpretation and Validation of Cluster Analysis ," Journal of Computational and Applied Mathematics , 20 (November) , 53 – 65.
Sheehan Kim Bartel , Pittman Matthew. (2016), Amazon's Mechanical Turk for Academics: The HIT Handbook for Social Science Research. Irvine, CA: Melvin & Leigh Publishers.
Shimshoni Yair , Efron Niv , Fink Michael , Segalis Eyal , Patton Brian , Levin Michal , et al. (2015), " Campaign and Competitive Analysis and Data Visualization Based on Search Interest Data ," Google Patents, https://patents.google.com/patent/US9043302.
Shugan Steven M.. (1987), " Estimating Brand Positioning Maps Using Supermarket Scanning Data ," Journal of Marketing Research , 24 (1), 1 – 18.
Shugan Steven M. (2014), " Market Structure Research ," in The History of Marketing Science , Russell S. Winer and Scott A. Neslin, eds. Singapore: World Scientific Publishing, 129 – 64.
Slotegraaf Rebecca J. , Pauwels Koen. (2008), " The Impact of Brand Equity and Innovation on the Long-Term Effectiveness of Promotions ," Journal of Marketing Research , 45 (3), 293 – 306.
Stephen Andrew T. , Galak Jeff. (2012), " The Effects of Traditional and Social Earned Media on Sales: A Study of a Microlending Marketplace ," Journal of Marketing Research , 49 (5), 624 – 39.
Timoshenko Artem , Hauser John R.. (2019), " Identifying Customer Needs from User-Generated Content ," Marketing Science , 38 (1), 1 – 20.
Urban Glen L. , Johnson Philip L. , Hauser John R.. (1984), " Testing Competitive Market Structures ," Marketing Science , 3 (2), 83 – 112.
Van der Maaten Laurens , Hinton Geoffrey. (2008), " Visualizing Data Using T-SNE ," Journal of Machine Learning Research , 9 (November), 2579 –2 605.
Wang Daixin , Cui Peng , Zhu Wenwu. (2016), " Structural Deep Network Embedding ," in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining - KDD '16. New York: Association for Computing Machinery , 1225 – 34.
Wieringa Jaap , Kannan P.K. , Ma Xiao , Reutterer Thomas , Risselada Hans , Skiera Bernd. (2021), " Data Analytics in a Privacy-Concerned World ," Journal of Business Research , 122 , 915 – 25.
Zaltman Gerald , Coulter Robin Higie. (1995), " Seeing the Voice of the Customer: Metaphor-Based Advertising Research ," Journal of Advertising Research , 35 (4), 35 – 51.
Zhang Kunpeng , Bhattacharyya Siddhartha , Ram Sudha. (2016), " Large-Scale Network Analysis for Online Social Brand Advertising ," MIS Quarterly , 40 (4), 849–68.
~~~~~~~~
By Yi Yang; Kunpeng Zhang and P.K. Kannan
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 71- Increasing Organ Donor Registrations with Behavioral Interventions: A Field Experiment. By: Robitaille, Nicole; Mazar, Nina; Tsai, Claire I.; Haviv, Avery M.; Hardy, Elizabeth. Journal of Marketing. May2021, Vol. 85 Issue 3, p168-183. 16p. 1 Diagram, 4 Charts, 1 Graph. DOI: 10.1177/0022242921990070.
- Database:
- Business Source Complete
Increasing Organ Donor Registrations with Behavioral Interventions: A Field Experiment
Although prior research has advanced our understanding of the drivers of organ donation attitudes and intentions, little is known about how to increase actual registrations within explicit consent systems. Some empirical evidence suggests that costly, labor-intensive educational programs and mass-media campaigns might increase registrations; however, they are neither scalable nor economical solutions. To address these limitations, the authors conducted a field experiment (N = 3,330) in Ontario, Canada, testing the effectiveness of behaviorally informed promotion interventions as well as process improvements. They find that intercepting customers with materials targeting information and altruistic motives at the right time, along with streamlining customer service, significantly increased registrations. Specifically, the best-performing intervention, prompting perspective taking through reciprocal altruism ("If you needed a transplant would you have one?"), significantly increased new registration rates from 4.1% in the control condition to 7.4%. The authors followed up with seven posttests (total N = 3,376) to find support for their theoretical predictions and to explore the mechanisms through which the interventions may have operated. This article provides evidence for low-cost, scalable marketing solutions that increase organ donor registrations in a prompted choice context and has important implications for public policy and societal welfare.
Keywords: behavioral science; field experiment; nudge; organ donation; public policy
Current statistics on organ donation point to an ever-increasing demand yet inadequate supply of available donors. For example, in Canada, more than 4,400 people are waiting to receive lifesaving organ transplants ([15]). Similarly, in the United States, there are over 113,000 individuals currently on the transplant waiting list, and 22 people die each day waiting ([23]). Concerningly, the gap between those needing transplants and those receiving them continues to widen ([23]). One way to address the growing demand for transplantable organs is to increase the number of individuals who register as donors ([16]). To illustrate, in the United States, the "conversion" rate for registered donors who have died and are medically suitable for organ donation is nearly 100% ([72]).
Low registration rates are especially prevalent in countries with explicit consent registration policies—that is, individuals must opt in to become organ donors—compared with countries with presumed consent policies—where individuals are organ donors by default but can opt out ([34]). Although changing the default appears to be a promising intervention ([66]), the impact on actual donations has been mixed due to, among other things, uncertainties about a deceased person's donation preferences ([21]; [48]). Furthermore, changing registration policies involves implementation challenges and ethical considerations surrounding informed consent ([27]). To date, most jurisdictions have maintained their existing policies ([58]), thus prompting the following question: What can be done within explicit consent systems to improve organ donor registration rates?
Prior research has identified factors predicting organ donation attitudes and intentions, such as having adequate information about organ donation as well as altruistic motives (for reviews, see [25]] and [53]]). However, attitudes and intentions do not consistently translate into actual registrations ([54]). In Canada, where we conducted our study, even though the vast majority of Canadians (90%) are in favor of organ donation, and 81% say they themselves would be willing to register ([33]), only 23% have actually registered their decision to become an organ donor ([15]).
Furthermore, the limited work focusing on registrations has largely tested elaborate and costly interventions outside of organ donor registration systems (e.g., testing workplace education programs and mass-media campaigns; for reviews, see [26]] and [30]]). Finally, in a recent article on living organ donation, [14] emphasized that promotional messages, despite being the primary focus of most charitable giving research (e.g., [24]; [40]; [41]; [55]; [83]), are only one aspect of the marketing mix that can be employed to solicit donations. Through a qualitative study, they outlined how the entire marketing mix—product, price, promotion, place, process, and people—may be employed to reduce experiences of sacrifice in the complex and cumbersome process to encourage living organ donations.
Our article contributes to the limited empirical evidence for low-cost and scalable marketing solutions to increase actual in-person organ donor registrations in current explicit consent systems. In addition, this research contributes to our understanding of how to employ multiple elements of the marketing mix to help achieve the objectives of nonprofit organizations ([14]). Specifically, our field experiment demonstrates how intercepting customers with promotional materials at the right time (an information brochure and perspective-taking prompts), along with other process improvements (streamlined customer service that includes additional time to review the promotional materials and a simplified form), can increase new organ donor registrations. By leveraging behavioral science to design our marketing interventions, we contribute to the understanding of how to reduce the intention–action gap in the context of organ donation, improve public policy, and enhance societal welfare.
Organ donation systems typically take one of two forms: explicit consent and presumed consent. In explicit consent systems, individuals have to enroll in the organ donor registry (i.e., opt in). The specific process can vary, but it usually occurs when people obtain or renew identification (e.g., driver's license) at a local government office such as the Department of Motor Vehicles. Although many countries have recently made online registries available, to date the majority of registrations still take place offline ([22]). For example, in Ontario, where our research was conducted, 85% of registrations in 2016 occurred in person at ServiceOntario centers (i.e., Department of Motor Vehicles equivalent; [75]).
Within explicit consent systems, one technique used to "nudge those who are willing donors into becoming registered donors" is mandating or "prompting" choice ([72], p. 125). In prompted choice contexts, customer service agents ask individuals whether they would like to register their consent to be a donor. Prompting forces individuals to decide, instead of waiting for them to actively volunteer their consent unsolicited, which can help overcome procrastination, inertia, and limited attention ([72]). However, even when prompting is implemented, many jurisdictions continue to have low organ donor registration rates ([22]; [36]). For example, at the time of our field experiment, only 24% of the 12 million eligible Ontarians were registered, despite using prompted choice ([74]).
With the rise of behavioral science and nudging in policy, one solution that has received attention is changing legislation from explicit consent to presumed consent, where individuals are considered organ donors by default but can opt out ([27]). Recent evidence finds that donation rates are approximately 30% higher, on average, in countries with presumed consent systems ([66]), though default policies were argued to be only one factor among many that determined donation rates (e.g., systems for obtaining family consent, transplant infrastructure, religious beliefs). In fact, some countries even observed a decrease in donations when moving to presumed consent ([ 5]; [21]). To date, very few countries have chosen to change their default policy to presumed consent (e.g., Singapore, the United Kingdom, Argentina, the Netherlands; [58]), as doing so can present several challenges. These include ( 1) a significant investment of time and money ([48]), ( 2) ethical concerns relating to informed consent and individual autonomy ([42]; [43]), and ( 3) ambiguity for the surviving family about the deceased's wishes ([12]). Together, these factors lead [72], p. 121) to conclude,
We favor the policy of prompted choice because there is no evidence that a viable alternative system would save more lives (and hence is superior in terms of the interests of Patients), and because we think that it does the best job of respecting the rights and interests of Potential Donors and Families. At the same time, we favor more nudges, and better choice architecture, to improve the prompting.
To date, research focused on actual organ donor registrations remains rare. [30] recently conducted a narrative review of all empirical research measuring actual registrations. Although they identified 24 studies, the authors concluded that many suffered from methodological weakness including selection bias, confounds, and self-reported dependent variables. As a result, the authors could not conduct a meta-analysis or provide clear prescriptions for how to improve registrations. In fact, only eight studies were found to be methodologically robust, and even among these, the majority were conducted outside the current registration systems and tested interventions that were relatively complex, costly, and labor intensive. For example, interventions tested included town halls with expert panels ([ 2]), mass-media campaigns ([61]), and workplace lunch-and-learn programs with presentations by transplant recipients and donor family members ([52]). Though some interventions proved promising (e.g., educational programs), they were neither scalable nor economical ways to improve registration rates within existing explicit consent systems.
One notable exception is recent work by [60], who conducted a field experiment testing the effects of adding persuasive messaging (e.g., using reciprocal altruism or social norms) to an online prompt to join the national organ donor registry in the United Kingdom. The authors found that their reciprocal altruism message ("If you needed an organ transplant would you have one? If so please help others.") performed best and increased individuals' sign-ups from 2.3% in the control condition to 3.1%. This study was the first to illustrate the potential for low-cost, scalable interventions, in general, and persuasive messages, specifically, to improve actual organ donor registrations. However, it was unable to distinguish between new and existing donors. In addition, it was conducted outside of the typical organ donor registration system. For example, it was conducted online at a time when most transactions were done in person ([81]), after drivers completed their government transactions. Given both the novelty and practical importance of these findings, there are several opportunities to extend this research that are worth pursuing. For instance, what might this effect look like for in-person transactions and on only new registrations? Would these findings replicate when applied within the more typical explicit consent registration system?
Following an early release of these findings ([10]), [49] wanted to test the effectiveness of reciprocal altruism persuasive messages on registration intentions in both online and in-person contexts. The authors found that reciprocal altruism primes significantly increased intentions to register online but had no such effect in person. Moreover, no significant effects were found on proxies for donation behavior (i.e., whether participants accessed optional information on organ donation), regardless of mode of delivery. Although this study did not measure actual registrations, it provides some support for the use of reciprocal altruism messages in the organ donation context, while also calling into question whether such messages would be effective for in-person registrations.
Taken together, although prior research on organ donation suggests that targeting altruistic motives and information may be promising, we know little about how to encourage actual, new, in-person organ donor registrations, especially in a low-cost and easy-to-scale manner. We designed our field experiment to explore these opportunities.
Promotional materials are commonly used by for-profit and nonprofit organizations to inform, persuade, and motivate actions. For example, nonprofit organizations can employ promotional tools when individuals are in the deliberation stage of a donation to help them proceed to the actual decision stage ([14]). We built on prior research to develop and test promotional materials to increase new organ donor registrations by providing information (with an information brochure) and enhancing altruistic motives (with perspective-taking prompts). We supported our interventions with improvements to streamline the registration process (i.e., additional time to review the promotional materials, intercepting customers at the time of decision, and a simplified form).
Most theories of human judgment and decision making argue that individuals make decisions on the basis of declarative knowledge—facts and information—that comes to mind at the time of decision making (for reviews, see [32]], [84]], and [85]]). The information can be obtained from external sources (such as an information brochure) or retrieved internally from long-term memory. Information can also increase procedural knowledge—the knowledge of how to perform a specific task ([59]). Studies have shown that providing information targeting each of these types of knowledge encourages action (e.g., [51]; [57]; [70]; [78]). In the context of organ donation, declarative information has been shown to be effective at enhancing attitudes, especially when framed positively (e.g., "one individual organ donor can donate organs [e.g., heart, lungs, kidneys, liver] to eight other people"; [56]). Messages providing procedural information about how to become a donor were particularly effective at enhancing attitudes when individuals were unaware of these details ([44]). Building on the aforementioned research, we predict that providing individuals with promotional material (i.e., an information brochure containing declarative and procedural information), specifically at the point when they are deciding, will make that information salient to them and increase new organ donor registrations.
Altruistic motives arise from empathy toward others and have been found to drive prosocial behavior across multiple domains ([ 6]; [ 7]; [ 9]; [79];). Research finds that one effective way to evoke altruistic motives is through perspective taking, considering a situation from a different point of view ([82]). In the context of organ donation specifically, perspective taking correlates with positive attitudes and willingness to register ([18]; [46]), though we do not yet know if perspective taking can be employed to reliably increase actual registrations. Moreover, perspective-taking manipulations have been tested almost exclusively in the lab until recently, and therefore, it is unclear how their effects would translate to field settings (cf. [37]; [60]).
We sought to test the effectiveness of enhancing altruistic motives on organ donor registrations in the field with three differing perspective-taking prompts: imagine other (IO), imagine self (IS), and reciprocal altruism (RA). The imagine other prompt asks individuals to consider how others would feel in a situation, enhancing one's pure altruistic motives to help ([ 7]). Alternatively, asking individuals to imagine oneself in the situation—imagine self—can increase both one's altruistic as well as self-interested motives ([17]), which some have suggested may be even more effective for encouraging prosocial behavior ([ 9]; e.g., [ 3]; [29]). Perspective-taking prompts can also encourage prosocial behaviors by making additional psychological concepts salient (e.g., reciprocity). For example, recent research has found that the reciprocal altruism prompt "If you needed an organ transplant would you have one? If so please help others" significantly increased online organ donor registrations ([60]). Such a statement evokes both self-interest and reciprocity by pointing out that that if individuals are willing to accept an organ, they should also donate ([39]; [67]; [77]). Given that prior research has found that benefits to the self as well as benefits to others can drive prosocial behaviors ([ 7]; [13]), we predicted that prompting these three types of perspective taking (imagine other, imagine self, and reciprocal altruism), would increase actual organ donor registration rates.
Our field experiment was conducted over a 2.5-week period (March–April 2014) in one preselected ServiceOntario location in Ontario, Canada—an explicit consent jurisdiction with prompted choice. To maximize the generalizability of our findings, we carefully considered the choice of location for the experiment. The specific location chosen has a sizable population (one of the largest and busiest centers in the province) that is demographically representative of Ontario's total population on several preselected characteristics including age, income, education, and religion.[ 7]
In 2014, when our experiment was conducted, all Ontarians were required to visit ServiceOntario centers in person for almost any public service including driver's license, health card, and photo identification transactions, thus reducing sampling bias concerns.[ 8] Each individual who visited this service center was a participant in our experiment (N = 3,330), and all participants visiting on a given day were exposed to the same experimental condition or phase. Because the timing of the phases and conditions was defined before the start of the experiment, neither the center, nor individual service agents, nor the researchers had control over which individuals received each condition. On average, 214 individuals visited the center each day. New donor registrations were measured using the service center's computer system. For each individual, we observed the type of transaction(s) they completed, the service agent they saw, and whether they registered during that visit as a new organ donor (yes/no). No identifying information about the participants and service agents was shared with the researchers to maintain privacy and protection of all parties involved.
The standard in-person registration process in Ontario is similar to that of many prompted choice jurisdictions. Upon arrival at ServiceOntario, individuals are given a number at the reception desk and wait until their number is called. Once called on, individuals perform the transaction(s) they came for at a service agent's counter, and during these transactions they are prompted to register. That is, they are asked by the service agent if they would like to register their consent as an organ donor today. Only if they affirm, they are then given the center's standard organ donor registration form to complete on the spot (Figure W1–1 in the Web Appendix).
The standard organ donor registration form is a black-and-white full-page document consisting of three sections (for a visual of the standard form, see Figure 1). The left-hand column primarily presents legal and procedural information about organ donation (e.g., "You have the right to decide whether or not to consent to the donation of your organs and tissue"). Although this information is meant to inform consumers, it is handed to consumers only after they agree to register and thus comes too late in the process. On the right-hand column of the document, the top portion serves to collect personal information from the individual (e.g., name, address, date of birth). It is important to note that for in-person registrations, that information is redundant as customers just completed another transaction in which they provided that information (e.g., renewing a driver's license) and therefore unnecessarily lengthens the process. Finally, on the bottom right-hand side, individuals are asked to indicate their consent and sign the form.
Graph: Figure 1. Registration form layouts.
We created five experimental conditions for our field experiment: A control condition that involved two process changes (time and simplification) and four promotion intervention conditions (information brochure and three perspective-taking prompts). The process changes made in the control condition were also present in all of the experimental conditions, enabling us to test for improvements specifically resulting from our promotion interventions. For an overview of the persuasive materials tested, see Table 1; for the forms and brochure, see Web Appendix W1.
Graph
Table 1. Overview of Promotional Materials.
| Experimental Condition | Promotional Materials and Perspective-Taking Prompts |
|---|
| Control | No promotional materials or perspective-taking prompts |
| Information | Standard organ donation information brochure |
| Reciprocal altruism | "If you needed a transplant, would you have one?If so, please help save lives and register today." |
| Imagine self | "How would you feel if you or someone you loved needed a transplant and couldn't get one?Please help save lives and register today." |
| Imagine other | "How do you think people feel when they, or someone they love, need a transplant and can't get one?Please help save lives and register today." |
Our control condition included two process changes designed to streamline customer service and support our interventions tested in the experimental conditions. First, individuals were handed the organ donor registration form with their waiting number when they arrived at the reception desk to allow adequate time to read, process, and consider the materials (vs. during their transaction[s] at the agent's counter). An additional benefit of handing the form out in advance is that it reduces the variation in registrations that may be caused by the individual service agents[ 9] and ensures that every customer is handed a registration form.
Second, to increase the salience of our interventions, we created a simplified version of the organ donor registration form[10] (for a comparison of the forms, see Figure 1; for the simplified form, see Figure W1–2 in the Web Appendix). In addition, behavioral research consistently finds that reducing the effort required to perform an action, or even just simplifying content (i.e., reducing "sludge"; [69]; [71]), can increase the number of people taking action ([11]; [68]). In creating our simplified version, we first removed all material from the standard form that was not required, legally or practically, for in-person transactions. As a result, the simplified form retained only the consent questions and a place to sign the form, focusing individuals on the decision at hand. In addition, we added a colored banner on top to add visual appeal, which, importantly, provides a location to highlight the perspective-taking prompts in three of the experimental conditions. Finally, we printed this smaller form on a half sheet of cardstock paper because the thicker, sturdy paper would enable individuals the option to complete the form without counter space (e.g., while waiting).
In all experimental conditions, participants received this simplified form upon arrival at the reception desk. These two process changes—extra time and simplified form—ensured that we could better capture the effect of our key behavioral interventions. Stated differently, without this streamlined customer service, our interventions might not be able to improve registration rates, because they may be overlooked and/or would come too late (i.e., after responding to the prompt to register). The control condition serves as a clean, conservative benchmark that enables us to quantify the effect of our interventions.
Behavioral researchers have argued that policy interventions are more likely to be successful if you consider their timing and prompt people when they are most likely to be receptive ([11]). In our information condition (info), we aimed to intercept customers and provide information at the right point in time. Although this condition presents promotional material, it is primarily a test of improving process; Ontario's standard organ donation brochure (see Figure W1–3 in the Web Appendix) includes detailed declarative (e.g., "1 organ donor can save 8 lives") and procedural information (e.g., "Registering is easy. Ask at the counter or do it online.") presented in a visually appealing and easy-to-process way (i.e., cleanly organized and relatively large font). This brochure is always readily available to take from self-serve brochure stands in the waiting area of ServiceOntario centers. It is also mailed to individuals with their driver's license renewal notice. However, in this condition, we tested the impact of handing out the brochure along with the simplified form to all customers when they arrived at the reception desk. By providing this information while individuals were waiting and deciding, we predicted that this would increase their likelihood of reading the brochure; the salience of the information during decision making ([84]); and, in turn, registrations.
The other three experimental conditions targeted altruistic motives using perspective-taking prompts. Participants were handed the simplified registration form with one prompt printed in the colored box at the top of the form (Table 1). First, our reciprocal altruism prompt, "If you needed a transplant, would you have one?" (adapted from [60]]), leveraged self-interest, empathy, and reciprocity. Our second prompt, imagine self, leveraged self-interest and empathy by stressing that without enough registered donors, they (the reader) or their loved ones might not have a transplant available if needed. Finally, our third prompt, imagine other, leveraged empathy by highlighting that without enough registered donors, others might not have a transplant available if needed. We conducted a pretest to confirm that each prompt induced the correct perspective taking as intended (see Web Appendix W2).
Our experimental conditions were each run consecutively for three business days. In addition, we added two phases, each pre- and postexperiment, in which we ( 1) measured registrations with the standard registration process (standard process phases) and ( 2) included time for acclimating service agents to the procedural changes caused by the experiment and informing them registrations would be tracked, a jurisdictional requirement (acclimation phases; for more details, see Web Appendix W3). During the acclimation phases, visitors were given our new simplified form but otherwise the service center followed the standard registration process. That is, the registration form was handed to individuals during their transactions at the service agent's counter only if they agreed to the registration prompt. Therefore, in total, data collection spanned an eight-week period beginning on February 24, 2014. For an overview, see Table 2.
Graph
Table 2. Timeline of Field Experiment.
| | Registration Process | | | |
|---|
| Experiment Condition | Phase/Condition (in Order Tested) | Materials | Timing | Agents Aware of Tracking | Duration (Days) | Sample Size (N) |
|---|
| N | Preexperiment standard process phase (SP) | Standard form | During a transaction | N | 12 (Mon–Sat) | 2,631 |
| N | Preexperiment acclimation phase (A) | Simplified form | During a transaction | Y | 3 (Mon–Wed) | 650 |
| Y | Control condition (C) | Simplified form | At reception desk (i.e., more time) | Y | 3 (Thu–Sat) | 659 |
| Y | Information condition (Info) | Simplified form + Brochure | At reception desk (i.e., more time) | Y | 3 (Mon–Wed) | 679 |
| Y | Reciprocal altruism condition (RA) | Simplified form w/ RA prompt | At reception desk (i.e., more time) | Y | 3 (Thu–Sat) | 608 |
| Y | Imagine self condition (IS) | Simplified form w/ IS prompt | At reception desk (i.e., more time) | Y | 3 (Mon–Wed) | 735 |
| Y | Imagine other condition (IO) | Simplified form w/ IO prompt | At reception desk (i.e., more time) | Y | 3 (Thu–Sat) | 649 |
| N | Postexperiment acclimation phase (A2) | Simplified Form | During a transaction | Y | 6 (Mon–Sat) | 1,251 |
| N | Postexperiment standard process phase (SP2) | Standard form | During a transaction | N | 11 (Mon–Sat) | 2,181 |
| TOTAL | | | | | | 10,043 |
20022242921990070 Notes: The chosen location operates six days a week, Monday through Saturday. Due to required messaging sent to service agents informing them it was "the last week" of the organ pilot, A2 was run for one business week (six days) instead of three days. Also, although the standard process phases were planned for two weeks each, one day during SP2 landed on a holiday and the service center was closed. Therefore, we have data for only 11 instead of 12 days.
To analyze the impact of the field experiment on new organ donor registrations, we start by presenting model-free evidence. Here, we adopt a-two-part approach. First, to test the effectiveness of our behavioral interventions, we compare the registration rates of each experimental condition with that in our control condition. Second, to explore the impact of our field experiment process changes (additional time and simplified form) relative to the standard registration process, as well as the impact of acclimating service agents to the experiment, we compare the registration rates in the pre- and postexperimental phases with that in the control. We then present logistic regressions that control for time-varying factors, such as the day of the week and the available agents. Next, we present a set of validity checks and robustness checks that test our hypotheses using alternative modeling specifications. Finally, we present a summary of seven follow-up posttests to provide support for our theoretical predictions and to explore the potential mechanisms through which our interventions may have operated.[11]
The gray-shaded bars in Figure 2 illustrate how registrations were affected by our interventions (see also Table W1–1 in the Web Appendix). A joint F-test confirmed that there were statistically significant differences between the conditions (F = 5.285, p <.001). New organ donor registrations were highest in the reciprocal altruism condition (7.4%; 95% confidence interval [CI] = ±2.08%). In fact, reciprocal altruism was the only condition that significantly increased donor registrations (Δ = 3.30%, p =.012, Cohen's h =.143) relative to our control condition (4.10%; 95% CI = ±1.51%). That said, registration rates in the reciprocal altruism condition were not significantly greater than any of our other interventions at the 5% level (info: Δ = 1.51%, p =.279; IS: Δ = 2.36%, p =.076; IO: Δ = 2.40%, p =.070).
Graph: Figure 2. New organ donor registration rates, raw means.Notes: White bars represent pre-post experimental phases, and gray bars represent our experimental conditions. Conditions are presented in order of implementation. Error bars represent ±1 SE.
We found that the registration rate in our control condition was not significantly different from that in any of the pre- and postexperimental phases (all ps >.18).
Because each experimental condition was run for three consecutive days, we account for potential differences across days when treatments were applied. To do so, we ran a fixed-effects logistic regression using all individuals who engaged in a transaction during our entire eight-week data collection period (i.e., during the experimental and pre- and postexperimental phases; N = 10,043). In this analysis, we controlled for day of week fixed effects and agent serving each individual ("agent") fixed effects.[12] Our experimental control condition served as the baseline. To be conservative, we use robust standard errors and have clustered all standard errors at the daily level as this was the unit where treatments were assigned ([ 1]). The dependent variable was whether an individual registered as a new organ donor (see Table 3). Results are presented in terms of odds ratios (ORs), that is, the odds that an individual registered as a new organ donor given a particular treatment (e.g., information condition) compared with the odds of the individual registering in the control condition.
Graph
Table 3. Organ Donor Registration Results (ORs) and Robustness Checks.
| M | R1 | R2 | R3 |
|---|
| Preexperiment standard process phase | .86 | .84 | .82 | .86 |
| (.12) | (.19) | (.22) | (.13) |
| Preexperiment acclimation phase | 2.09** | 2.09** | 2.28** | 2.03** |
| (.50) | (.50) | (.59) | (.47) |
| Control condition | 1.00 | 1.00 | 1.00 | 1.00 |
| (.) | (.) | (.) | (.) |
| Information condition | 1.99*** | 2.02** | 2.23** | 1.87** |
| (.40) | (.48) | (.62) | (.38) |
| Reciprocal altruism condition | 1.85*** | 1.88*** | 2.14*** | 1.74*** |
| (.12) | (.33) | (.48) | (.17) |
| Imagine self condition | 1.81* | 1.87 | 2.19 | 1.99** |
| (.44) | (.70) | (.92) | (.51) |
| Imagine other condition | 1.25 | 1.29 | 1.55 | 1.31* |
| (.14) | (.44) | (.56) | (.17) |
| Postexperiment acclimation phase | .76 | .80 | .92 | .75 |
| (.12) | (.39) | (.47) | (.13) |
| Postexperiment standard process phase | .86 | .93 | 1.20 | .78 |
| (.10) | (.65) | (.86) | (.11) |
| Time trend (days) | | 1.00 | .99 | |
| | (.02) | (.02) | |
| Customers per agent | | | | .95 |
| | | | (.02) |
| Day-of-week fixed effects | Yes | Yes | Yes | Yes |
| Agent fixed effects | Yes | Yes | Yes | Yes |
| Agent × day-of-week fixed effects | No | No | Yes | No |
| N | 10,027 | 10,027 | 9,056 | 10,027 |
- 30022242921990070 *p <.05.
- 40022242921990070 **p <.01.
- 50022242921990070 ***p <.001.
- 60022242921990070 Notes: Column M is our main specification, column R1 adds a linear time trend, column R2 adds agent × day-of-week fixed effects, and column R3 adds customers per agent. Standard errors are robust and clustered at the daily level. Our dependent variable is registration as a new organ donor (i.e., consent = 1, no consent = 0).
The results of this analysis appear in column 1 of Table 3 (overall model: χ2 (d.f. = 39) = 1,978.19, p <.001). As with our model-free results, we found that being exposed to the reciprocal altruism prompt significantly increased the odds of registering compared with the control condition (OR = 1.84, p <.001). After controlling for day of week and agent effects, we found two additional conditions to be significant. Compared with the control, the odds of registering were also significantly higher in the information condition (OR = 1.99, p <.001) and imagine self condition (OR = 1.81, p =.015). Follow-up pairwise Wald comparison tests under our main specification (i.e., agent and day-of-week fixed effects) show that these three interventions did not significantly differ from one another in their effectiveness (all ps >.699; for details, see Table W1–2 in the Web Appendix, columns "Info" and "RA").
Although we did not observe a significant difference in the model-free results, after adding agent and day of week fixed effects, we find that registrations were significantly higher in the preexperiment acclimation phase (OR = 2.09, p =.002) compared with control. All other comparisons with our control condition remained nonsignificant.
During the preexperiment acclimation phase, attention was being drawn to organ donor registrations, service agents were exposed to our new form for the first time, and service agents were informed that organ donor registration rates were being tracked during this pilot period. Critically, for the interpretation of our subsequent experimental conditions, registration rates declined immediately afterward. Nevertheless, to test whether these specific process changes impacted registration rates, the following section presents a validity check that decomposes the different elements of our interventions. Moreover, we subsequently test additional controls to account for changes in agent's behavior with a series of robustness checks.
Given the constraints of our government partners, we were limited in the number of experimental conditions we could run. As a result, we were unable to counterbalance and test each element of our interventions individually. It was for this reason that we created an experimental control condition: to serve as a clean benchmark against which we compared our interventions. However, a result of this approach is that compared with the center's standard process, each of our interventions comprised a combination of multiple changes. For example, unlike the standard process, those in the information condition received additional time, a simplified form, and an information brochure (see Table 2). Our analysis formally separates the following seven elements: ( 1) the standard process, ( 2) the simplified form with agents aware of tracking, ( 3) additional time, ( 4) information brochure, ( 5) reciprocal altruism prompt, ( 6) imagine self prompt, and ( 7) imagine other prompt. We calculated the effect of each these elements using a fixed-effects logit model predicting the odds of registering (see Table 4).
Graph
Table 4. Analysis of Organ Donor Registrations (Odds Ratios): Validity Check Decomposing the Elements of the Interventions.
| Coefficient | Standard Error |
|---|
| Standard process (SP) | 1 | (.) |
| + Simplified form + Agents aware of tracking | 1.234 | (.245) |
| + Additional time | 1.001 | (.187) |
| + Information brochure | 1.697* | (.356) |
| + Reciprocal altruism prompt | 1.870*** | (.115) |
| + Imagine self prompt (change from RA to IS) | .833 | (.215) |
| + Imagine other prompt (change from IS to IO) | .814 | (.224) |
| Day-of-week fixed effects | Yes | |
| Agent fixed effects | Yes | |
| Agent × day-of-week fixed effects | No | |
| N | 10,027 | |
- 10022242921990070 *p <.05.
- 200022242921990080 **p <.01.
- 300022242921990100 ***p <.001.
- 400022242921990100 Notes: Our dependent variable is registration as organ donor (i.e., consent = 1, no consent = 0). SP = preexperiment standard process phase, RA = reciprocal altruism condition, IS = imagine self condition, IO = imagine other condition.
This analysis illustrates that the process changes initiated during the preexperiment acclimation phase (i.e., using the simplified form and making service agents aware of the fact that we were tracking registration rates) by themselves did not significantly increase an individual's odds of registering in comparison to the standard process, nor did handing the materials out in advance to provide more time. However, as predicted, and in line with our previous results, we find that adding a reciprocal altruism prompt did significantly increase individuals' odds of registering (OR = 1.870, p <.001). Similarly, providing an information brochure significantly increased individuals' odds of registering (OR = 1.697, p =.012). Finally, the results presented in Table 4 show that changing from the reciprocal altruism prompt to the imagine self prompt had no significant impact on the odds of registering (OR =.833, p =.479), nor did changing from the imagine self to the imagine other prompt (OR =.814, p =.456).
In this subsection, we perform a series of robustness checks to confirm that our results are robust to alternative sets of controls, in particular a time trend, agent–day of week interactions, and the number of customers per agent (for the inclusion of type of transaction controls such as health card or driver's license, see Web Appendix Table W1–4; for the use of different "baseline" conditions, see Web Appendix Table W1–5).
We incorporated a linear time trend in column R1 of Table 3 to address a potential confound related to history and ensure our results were not being driven by seasonality or a change in agents' behavior due to the experiment ([63]). Our results are robust to the inclusion of a time trend, and if anything, increased the estimated effect of the information and reciprocal altruism interventions. The estimated effect of imagine self also increased, but the effect was not significant due to a higher standard error.
In column R2 of Table 3, we incorporated agent–day of week fixed effects to account for the possibility that certain agents may perform better on certain days of the week. Again, our results are robust to the addition of this control.
Another potential confound is how busy the center was. Customers may have been more likely to register on days with longer wait times, as they would have more time to attend to the materials. Conversely, agents may be more likely to promote organ donation transactions when wait times are short. To account for this, we controlled for the ratio of the number of customers to the number of agents working on a given day in column R3 of Table 3. Our results remain robust to the addition of this control.
To provide support for our hypotheses and test possible alternative mechanisms, we conducted seven online posttests (five preregistered) with 3,376 North American participants.[13] The purpose of these experiments was to explore the process through which our interventions may have been operating. In all of the posttests, participants were randomly presented the actual materials of one of our experimental conditions (between-participants design), they were asked a series of questions aimed at measuring their reactions toward these stimuli, and our experimental control condition served as the point of comparison. To increase confidence in significant findings, we sought to replicate them across posttests. Next, we discuss the main results from our posttests (for an overview of measures and findings see Appendix W4).
Posttest 1 examined whether our interventions may have affected perceptions of risk. It is possible that our materials focused participants on specific aspects of risk, such as the risk of not having a transplant available if needed, leading to self-interested motivations to act ([17]). Conversely, our interventions may have shifted participants' focus away from the risks and onto the benefits of registering ([80]). We assessed the perceived risk of five factors: ( 1) needing a transplant in the future, ( 2) being able to get a transplant if needed, ( 3) encountering an organ donation shortage, ( 4) the medical system treating organ donors fairly, and ( 5) the medical system allocating organ donations efficiently. We found no evidence that our interventions significantly changed risk perceptions compared with control (all ps >.44).
In posttest 2a (N = 403, MTurkers) and 2b (N = 364, Ontario students), we examined an important assumption of our reciprocal altruism prompt: that individuals envision themselves in a position where they are accepting help, which then increases their likelihood of reciprocating (see [77]]). Specifically, we hypothesized that the presence versus absence of the prompt "If you needed a transplant, would you have one?" would significantly increase registration intentions, both when participants were explicitly asked to answer the prompt or—as in our field experiment—when they were simply presented the prompt. However, individuals' registration intentions—that is, their hypothetical registration likelihood (seven-point scale), consent decision (indicating "I would consent to help save lives by becoming an organ and tissue donor"), and exclusions decision (indicating "I would wish to donate any needed organs and tissue")—were not significantly different from the control at the 5% level (all ps ≥.091). In hindsight, these null effects are likely the result of the aforementioned organ donation intention–action gap that many countries, including Canada, observe. In support of this, we found extremely high registration likelihood ratings and consent rates even without the reciprocal altruism prompt (posttest 2a: Mcontrol = 5.55 and Mcontrol = 99.0%, respectively; posttest 2b: Mcontrol = 5.57 and Mcontrol = 96.6%, respectively). Finally, as a manipulation check, we explored whether people envisioned themselves accepting a transplant, if they needed one. Indeed, when forced to respond to the reciprocal altruism prompt, almost all participants answered with "yes" (posttest 2a: 92 of 99, posttest 2b: 87 of 88).
Posttests 3–7 aimed to examine a combination of ( 1) perceptions of the materials (e.g., educational, thought-provoking, focused on self vs. others); ( 2) the thoughts and feelings evoked from our interventions, including positive and negative emotions, feelings of sympathy, comfort registering; and ( 3) the extent to which our interventions affected participants' general views on organ donation (i.e., the importance and norms of registering). Finally, we conducted an internal meta-analysis ([45]) focusing on each of the measures assessed in at least two posttests. Next, we present the reliable insights from this meta-analysis.
First, in terms of perceptions of our promotional materials, we found that indeed the brochure was viewed as significantly more educational (β =.84, 95% CI = ±.36; p <.001). In addition, it was seen as more emotionally positive (β =.67, 95% CI = ±.23; p <.001) and less emotionally negative (β = −.37, 95% CI ±.27; p =.008). These results are in line with our hypothesis that the information condition (brochure) would make additional information salient.
Second, in terms of thoughts and feelings, we found that our perspective-taking prompts evoked significantly stronger feelings of sympathy (in fact, all of our interventions, including the information condition, did so; βinfo =.63, 95% CI = ±.37; βRA =.52, 95% CI = ±.37; βIS =.60, 95% CI = ±.37; βIO =.76, 95% CI = ±.37; all ps ≤.006). In addition, all but the reciprocal altruism intervention (p =.17) were viewed to be significantly more focused on others (βinfo =.44, 95% CI = ±.27; βIS =.53, 95% CI = ±.27; βIO =.55, 95% CI = ±.27; all ps ≤.003; no difference for focus on self: all ps ≥.10). These results support that our perspective-taking prompts in the field experiment likely evoked stronger altruistic motives.
In terms of alternative mechanisms, it is possible that our interventions may have stimulated new considerations ([84]), impacting registrations in our field experiment. For example, in our posttest meta-analysis we found that all of our interventions, except the imagine self prompt (p =.113), were rated as significantly more thought-provoking (βinfo =.45, 95% CI = ±.30; βRA =.34, 95% CI = ±.30; βIO =.43, 95% CI = ±.30; all ps ≤.027). It is also possible that providing information caused individuals to feel more knowledgeable ([31]), feel greater comfort with the decision to register ([50]), and feel more prepared to register ([78]), positively affecting registration rates. However, in our meta-analysis, we found no evidence that our interventions influenced feeling knowledgeable (all ps ≥.407), comfort registering (all ps ≥.260), or preparedness (all ps ≥.518). Finally, our interventions could have changed general perceptions of how important it is to donate ([ 6]) or how much it is the right thing to do (norms; [28]), in turn increasing registrations in our field experiment. However, we found no evidence for such effects in our meta-analysis (all ps ≥.300; for other nonsignificant measures that were explored, including feelings and perceptions of ethicality, see Table W4–3 in the Web Appendix).
The results of our field experiment support our prediction that marketing interventions grounded in behavioral science, targeting information and altruistic motives, would significantly increase new organ donor registrations in a prompted choice context. While our interventions did not significantly differ in effectiveness from one another in the majority our analyses,[14] our reciprocal altruism intervention ("If you needed a transplant, would you have one? If so, please help save lives and register today") was the best performing. It led to the highest registration rates and was the only condition to significantly increase registrations compared with our control condition consistently across all analyses, including our model-free results. After including relevant controls (e.g., day-of-week and agent fixed effects), we found that our information and imagine self interventions also significantly increased registrations compared with control.
Our posttests provide some initial evidence for the mechanisms driving our interventions. As we predicted, all of our perspective-taking prompts induced greater feelings of sympathy compared with control ([ 8]), and our information condition was rated as more educational. Moreover, all of our perspective-taking prompts, except reciprocal altruism, were perceived to focus more on others. The posttests also suggested some additional mechanisms through which our interventions may have been operating. For example, our brochure was found to be more emotionally positive and less emotionally negative than the control. Previous research has shown that declarative information is especially effective at increasing organ donation attitudes when framed positively ([56]), which may have contributed to its success. The brochure also increased feelings of sympathy and focus on others, suggesting that it may have targeted altruistic motives as well ([ 6]). All our interventions, aside from imagine self, were viewed to be more thought provoking than our control, suggesting that they may have changed what individuals were considering when deciding ([84]). Other mechanisms tested (e.g., risk perceptions, comfort registering, importance of donating, norms) were not supported. Although our posttests exposed participants to the actual materials used in our field experiment, it is critical to note that they were conducted outside the in-person, actual organ donor registration context (e.g., online MTurk and online university student samples) limiting our ability to draw conclusions about what occurred in our field experiment. Furthermore, previous research has shown that organ donation attitudes and intentions often do not translate into actual behavior ([47]). For these reasons, it is important to exercise caution when drawing conclusions from these posttests.
This field experiment contributes to the literature by testing marketing interventions to increase new organ donor registrations within the prevalent explicit consent systems. Prior organ donation research has primarily focused on factors that influence intentions and attitudes ([27]); however, the shortage of registered donors appears to be primarily a problem of inaction ([64]). To date, a small number of studies have documented some positive impact on actual registrations from elaborate education programs and mass-media campaigns ([30]), yet they provide little insight into how to increase new registrations within explicit consent systems in an economical and scalable way. This research contributes to the limited empirical evidence for low-cost and scalable marketing solutions, targeting knowledge and altruistic motives, to overcome the intention–action gap and improve registrations within the current systems. Specifically, we find that in our best-performing condition, prompting perspective taking through reciprocal altruism ("If you needed a transplant, would you have one? If so, please help save lives and register today") significantly increased registration rates from 4.1% in the control condition to 7.4%, an 80% increase.
Our work also contributes to our understanding of how to employ multiple elements of the marketing mix to help achieve the objectives of nonprofit organizations. Research on charitable giving has primarily focused on promotional strategies to solicit donations of time, money, and blood (e.g., [38]; [62]). Recently however, [14] qualitatively examined how the entire marketing mix could be employed, more broadly, to encourage living organ donations. We expand on this research by empirically testing interventions to support the more common request to register as a deceased organ donor. We demonstrate that intercepting customers with promotional materials at the right time, along with streamlined customer service—additional time and a simplified form—significantly increased new organ donor registrations. Importantly for practitioners, this streamlined organ donation process ensured that every customer was exposed to the materials and had ample time to consider them and complete the form. It also reduced the burden on the individual service agents to prompt registrations and reduced agent-caused variation in registrations. We obtained preliminary evidence, based on a small postexperiment survey, suggesting that our process changes may have reduced the time it took for individuals to register.[15] Processing transactions faster would save ServiceOntario time and money and could also lead to happier customers, as they would have shorter wait times. Moreover, other research has found that giving individuals something to read while waiting can make the time go by faster and increase satisfaction ([35]). Therefore, these simple changes (i.e., reducing "sludge"; [71]) may help increase not only organ donor registrations but also the efficiency of the registration process as well as customer satisfaction.
This work also expands our understanding of how altruistic motives can be leveraged to increase prosocial behavior. While prior research has largely tested perspective taking in the lab ([37]), we demonstrate its effectiveness in the field. In addition, we replicate and extend the generalizability of using reciprocal altruism to improve organ donation behavior ([60]) by demonstrating its effectiveness at encouraging new and in-person registrations within the typical registration system and in a different national culture. We predicted that the reciprocal altruism prompt would be especially effective due to its combined focus on benefits for self and others ([39]). While it did perform best, it is important to reiterate that our three perspective-taking prompts (reciprocal altruism, imagine self, and imagine other) did not significantly differ in effectiveness from each other across the majority of our analyses. The fact that all of our prompts improved registrations, albeit to varying degrees and not consistently significantly different from control, suggests that each of these different forms of perspective taking might be viable solutions for motivating organ donor registrations, though it would be important to test variations of their conceptualization in the field.
Our results highlight the importance of carefully considering the process and timing of delivering promotional materials. For example, the fact that our information condition was successful (after controlling for day of the week and agent) is noteworthy because the brochure that we provided was the standard one that was both readily available in self-serve stands at ServiceOntario centers (throughout the experiment) and was mailed with all drivers' license renewal notices along with a standard registration form. Therefore, it is important for managers to intercept customers at the right point in time.
For practitioners, our results highlight the critical importance of testing interventions in the field and measuring their realized impact. For one, individuals in our pretest (Web Appendix W2) did not accurately predict the relative effectiveness of our interventions. Moreover, it would have been reasonable to predict that merely providing individuals with a simplified form along with additional time to decide would increase donor registrations (i.e., our control condition; [11]), but our results revealed otherwise. Finally, in a recent meta-analysis, [19] suggested that nudge interventions tested in academic research tend to have a significantly larger effect than those tested "at scale" by government "nudge units." While our field experiment is similar to that of academic research in terms of sample size and effect size, many other features of our experiment suggest it is similar to research of nudge units: we tested elements of simplification, had relatively representative sampling, and our findings were released to the general public by our government partners, irrespective of the results. Together, these observations highlight the importance for researchers and policy makers to consider issues such as statistical power, selection effects, and characteristics of the interventions when planning at-scale implementations of interventions from academic research.
Although this experiment provides us with low-cost, scalable marketing solutions to increase actual organ donor registrations, we were unable to test the mechanisms driving our interventions and did not have the opportunity to replicate the study over a longer period or at a larger scale. For example, to maintain customers' privacy, we could not survey the individuals who came to the center or collect any personal information about them, and therefore, it was not feasible to administer manipulation checks or process measures. While our posttests provided some initial process evidence, they can only be considered as indicative, at most, because they were conducted outside the in-person registration context (e.g., online student samples, MTurk workers), and measured feelings, thoughts, and intentions, which do not always map onto behavior ([54]). Future research should systematically examine the underlying mechanisms driving interventions within field settings to facilitate generalizations.
In addition, due to timing and design constraints we were unable to counterbalance each component of our interventions with a fully-crossed and randomized design. Although we estimated the unique impact of the elements of our interventions, future research could experimentally manipulate each to explore whether they had additive or interactive effects. For example, it may be worthwhile to formally test how information and perspective-taking work in combination with one another, as they may be more effective together than either condition alone.
Finally, another area for future research would be to examine the role that customer service agents play in encouraging organ donor registrations. First, our data revealed significant effects of agents on registrations. Second, registrations in the preexperiment acclimation phase (a hybrid between the center's standard process and our subsequently introduced experimental process) were significantly higher than the preexperiment standard process phase and, after including day-of-week and agent fixed effects, also higher than our experimental control condition. In the preexperiment acclimation phase, service agents encountered our new simplified form for the first time and received formal communication from their manager drawing attention to the organ donation task and making them aware that registrations would be tracked. These factors may have led to excitement or changes in agents' behavior that in turn led to increased registrations. In all of our experimental conditions, we took care to limit any variation in registrations caused by the service agents by handing the form out in advance (i.e., when individuals arrived at the registration desk). However, these findings suggest that maximizing agent effectiveness could be another interesting avenue for future research aiming to increase organ donation rates.
With thousands currently on the transplant waiting list, the need for organ donors is urgent, and as the population ages, the demand is only predicted to increase further. One way to address the ever-growing demand is to increase the number of individuals registered as donors within the prevalent explicit consent systems, in which low registration rates are especially common. In our field experiment we were able to increase new in-person registrations in a prompted choice context using easy-to-scale, low-cost[16] promotion interventions supported by process improvements. We were able to do so without imposing on individuals' freedom, raising ethical concerns (i.e., changing defaults), or passing new legislation. To illustrate the potential impact of our findings, if we were to assume that everything held constant over time and we introduced our best-performing intervention (reciprocal altruism) throughout Ontario, we could expect roughly 225,000 additional new registrations annually. While quantifying the effect of increased registrations on the ultimate goal—lives saved or enhanced—is challenging (see, e.g., [20]), research has shown that those who register their consent are significantly more likely to actually donate than those who have not ([16]). Specifically, the majority of registered individuals in Ontario who become eligible to donate are ultimately converted to donors ([73]), and [76] advertises that one single donor may save up to eight lives and enhance as many as 75 more lives.
Compared with the center's standard registration process, all of our interventions significantly increased organ donor registration rates (see Table W1–1 and W1–2 in the Web Appendix). We recommended Ontario implement our reciprocal altruism intervention and track its performance for three reasons: ( 1) this intervention was successful in increasing registrations in both the United Kingdom ([60]) and our field study, ( 2) it significantly increased registrations compared with the control across all our analyses, and ( 3) it avoids the costs associated with printing additional brochures. In 2016, the Ontario government adopted our recommendation partially by introducing a somewhat simpler organ donor registration form with the reciprocal altruism prompt province-wide (compare Figure W1–7 with Figure W1–4 in the Web Appendix). For policy makers who want to use our insights to improve organ donor registrations in a similar context, we recommend implementing as many of the design elements of our marketing materials as possible (e.g., colored banner on cardstock), along with implementing the supporting process changes (e.g., simplified form, intercepting customers at the right time, providing time to attend to the materials). Together, we believe this research not only informs our understanding of how marketing can be leveraged to improve nonprofits' goals but also offers insights that could benefit society by increasing organ donor registrations.
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921990070 - Increasing Organ Donor Registrations with Behavioral Interventions: A Field Experiment
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921990070 for Increasing Organ Donor Registrations with Behavioral Interventions: A Field Experiment by Nicole Robitaille, Nina Mazar, Claire I. Tsai, Avery M. Haviv and Elizabeth Hardy in Journal of Marketing
Footnotes 1 Conceived experiments: NR NM CT EH. Designed experiments: NR NM CT EH. Directed experiments: NR NM CT EH. Data analysis: NR NM CT AH. Interpreted the results: NR NM CT AH. Wrote the first draft: NR. Edited the manuscript: NR NM CT AH.
2 Robert Meyer
3 The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: At the time of the field experiment, Nicole Robitaille was employed as behavioral scientist at the Behavioural Insights Unit (BIU), Government of Ontario, and Elizabeth Hardy was the BIU's manager. Nina Mazar and Claire Tsai were uncompensated academic advisors with the BIU.
4 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by a DI McLeod Grant from the Smith School of Business, Queen's University awarded to NR, a TD Bank Group Research Fund awarded to BEAR at the Rotman School of Management, University of Toronto, that NM codirected at the time of the research, and a SSHRC Insight Grant awarded to CT.
5 Nicole Robitaille https://orcid.org/0000-0002-3323-4423 Nina Mazar https://orcid.org/0000-0001-8248-654X
6 Online supplement: https://doi.org/10.1177/0022242921990070
7 For the area served by this ServiceOntario center [Ontario overall], the average age was 39.6 years [41.0 years], 30% [29%] had a postsecondary degree, 60% [65%] identified as Christian, and the median household income was $83,018 [$74,287] ([65]).
8 While more transactions can now be completed online, all Ontarians still need to visit in person for some services (e.g., new photos on government-issued identification).
9 Background data revealed that registration rates varied by service agents, which could result from inconsistent prompting (ServiceOntario does not monitor agents' prompting behavior).
The simplified form was approved as a form that the Ontario government could implement.
Complete pretest and posttest materials and data are available on OSF (https://osf.io/m3kuj/). Anonymized data from the field experiment are available from the authors upon request.
Although individuals in our experimental conditions were given the simplified form in advance, limiting the effect of agents on registrations, we control for agent fixed effects because there was noteworthy variation in agents' number of processed registrations.
Six of our seven posttests were conducted during the COVID-19 pandemic, which may have impacted data quality ([4]).
In our logit model (i.e., after controlling for day-of-week and agent effects), we find that the reciprocal altruism and information conditions significantly increased registrations relative to the imagine other condition (info: OR = 1.60, p =.033, RA: OR = 1.48, p <.001; for details, see Table W1–2 in the Web Appendix, column "IO").
A voluntary postexperiment survey completed by ten of the service agents (all were invited) revealed an estimated time-savings of 2.3 minutes per registration. Given the low response rate, we are cautious to draw conclusions from this survey. Future research should carefully examine actual and perceived wait times as well as customer satisfaction.
Designing (a one-time cost), printing, and shipping new forms for the field experiment cost less than $3,000.
References Abadie Alberto, Athey Susan, Imbens Guido W., Wooldridge Jeffrey. (2017), "When Should You Adjust Standard Errors for Clustering?" Working Paper 24003, National Bureau of Economic Research.
Alvaro Eusebio M., Siegel Jason T., Jones Sara P. (2011), "Increasing Organ Donor Registration Rates by Providing an Immediate and Complete Registration Opportunity: An Experimental Assessment of the IIFF Model," Psychology, Health & Medicine, 16 (6), 686–94.
Andreoni James. (1989), "Giving with Impure Altruism : Applications to Charity and Ricardian Equivalence," Journal of Political Economy, 97 (6), 1447–58.
Arechar Antonio A., Rand David G. (2020), "Turking in the Time of COVID," working paper, PsyArXiv.
Arshad Adam, Anderson Benjamin, Sharif Adnan. (2019), "Comparison of Organ Donation and Transplantation Rates Between Opt-Out and Opt-In Systems," Kidney International, 95 (6), 1453–60.
Batson C. Daniel. (1987), "Prosocial Motivation: Is It Ever Truly Altruistic?" in Advances in Experimental Social Psychology, Berkowitz Leonard, ed. New York: Academic Press, 65–122.
Batson C. Daniel, Early Shannon, Salvarani Giovanni. (1997), "Perspective Taking: Imagining How Another Feels Versus Imaging How You Would Feel," Personality and Social Psychology Bulletin, 23 (7), 751–58.
Batson C. Daniel, Sager Karen, Garst Eric, Kang Misook, Rubchinsky Kostia, Dawson Karen. (1997), "Is Empathy-Induced Helping Due to Self-Other Merging?" Journal of Personality and Social Psychology, 73 (3), 495–509.
Batson C. Daniel, Shaw Laura L. (1991), "Evidence for Altruism: Toward a Pluralism of Prosocial Motives," Psychological Inquiry, 2 (2), 107–22.
Behavioural Insights Team (2013), "Applying Behavioural Insights to Organ Donation: Preliminary Results from a Randomised Controlled Trial," research report (accessed February 2, 2021), https://www.bi.team/wp-content/uploads/2015/07/Applying_Behavioural_Insights_to_Organ_Donation_report.pdf.
Behavioural Insights Team (2014), "EAST: Four Simple Ways to Apply Behavioural Insights," research report, https://www.bi.team/wp-content/uploads/2015/07/Applying_Behavioural_Insights_to_Organ_Donation_report.pdf.
Beshears John, Choi James J., Laibson David, Madrian Brigitte C. (2008), "How Are Preferences Revealed?" Journal of Public Economics, 92 (8/9), 1787–94.
Böckler Anne, Tusche Anita, Singer Tania. (2016), "The Structure of Human Prosociality: Differentiating Altruistically Motivated, Norm Motivated, Strategically Motivated, and Self-Reported Prosocial Behavior," Social Psychological and Personality Science, 7 (6), 530–41.
Bradford Tonya Williams, Boyd Naja Williams. (2020), "Help Me Help You! Employing the Marketing Mix to Alleviate Experiences of Donor Sacrifice," Journal of Marketing, 84 (3), 68–85.
Canadian Blood Services (2019), "About," (accessed June 25, 2019), https://organtissuedonation.ca/en/about.
Christmas Ashley Britton, Mallico Eric J., Burris Gary W., Bogart Tyson A., Norton Harry James, Sing Ronald F. (2008), "A Paradigm Shift in the Approach to Families for Organ Donation: Honoring Patients' Wishes Versus Request for Permission in Patients with Department of Motor Vehicles Donor Designations," Journal of Trauma and Acute Care Surgery, 65 (6), 1507–10.
Cialdini Robert B., Schaller Mark, Houlihan Donald, Arps Kevin, Fultz Jim, Beaman Arthur L. (1987), "Empathy-Based Helping: Is It Selflessly or Selfishly Motivated?" Journal of Personality and Social Psychology, 52 (4), 749–58.
Cohen Elizabeth L., Hoffner Cynthia. (2013), "Gifts of Giving: The Role of Empathy and Perceived Benefits to Others and Self in Young Adults' Decisions to Become Organ Donors," Journal of Health Psychology, 18 (1), 128–38.
DellaVigna Stefano, Linos Elizabeth. (2020), "RCTs to Scale: Comprehensive Evidence from Two Nudge Units," Working Paper 27594, National Bureau of Economic Research.
DeRoos Luke J., Marrero Wesley J., Tapper Elliot B., Sonnenday Christopher J., Lavieri Mariel S., Hutton David W., et al. (2019), "Estimated Association Between Organ Availability and Presumed Consent in Solid Organ Transplant," JAMA Network Open, 2 (10), e1912431.
Domínguez Javier, Rojas J.L. (2013), "Presumed Consent Legislation Failed to Improve Organ Donation in Chile," Transplantation Proceedings, 45 (4), 1316–17.
Donate Life America (2018), "2018 Annual Update," report (accessed April 29, 2020), https://www.donatelife.net/wp-content/uploads/2018/09/DLA%5fAnnualReport.pdf.
Donate Life America (2020), "Organ, Eye and Tissue Donation Statistics," (accessed April 29, 2020), https://www.donatelife.net/statistics/.
Fajardo Tatiana M., Townsend Claudia, Bolander Willy. (2018), "Toward an Optimal Donation Solicitation: Evidence from the Field of the Differential Influence of Donor-Related and Organization-Related Information on Donation Choice and Amount," Journal of Marketing, 82 (2), 142–52.
Falomir-Pichastor Juan M., Berent Jacques A., Pereira Andrea. (2013), "Social Psychological Factors of Post-Mortem Organ Donation: A Theoretical Review of Determinants and Promotion Strategies," Health Psychology Review, 7 (2), 202–47.
Feeley Thomas Hugh, Moon Shin-Il. (2009), "A Meta-Analytic Review of Communication Campaigns to Promote Organ Donation," Communication Reports, 22 (2), 63–73.
Ferguson Eamonn, Murray Catherine, O'Carroll Ronan E. (2019), "Blood and Organ Donation: Health Impact, Prevalence, Correlates, and Interventions," Psychology and Health, 34 (9), 1073–104.
Fisher Robert J., Ackerman David. (1998), "The Effects of Recognition and Group Need on Volunteerism: A Social Norm Perspective," Journal of Consumer Research, 25 (3), 262–75.
Gershon Rachel, Cryder Cynthia, John Leslie K. (2020), "Why Prosocial Referral Incentives Work: The Interplay of Reputational Benefits and Action Costs," Journal of Marketing Research, 57 (1), 156–72.
Golding Sarah Elizabeth, Cropley Mark. (2017), "A Systematic Narrative Review of Effects of Community-Based Intervention on Rates of Organ Donor Registration," Progress in Transplantation, 27 (3), 295–308.
Hart Julian T. (1965), "Memory and the Feeling-of-Knowing Experience," Journal of Educational Psychology, 56 (4), 208–16.
Higgins E. Tory. (1996), "Knowledge Activation: Accessibility, Applicability, and Salience," in Social Psychology: Handbook of Basic Principles, Tory Higgins E., Kruglanski Arie W., eds. New York: Guilford Press, 133–68.
Ipsos (2006), "Most Canadians (81%) Willing to Donate Their Organs in the Case of Their Death," (July 11), https://www.ipsos.com/en-ca/most-canadians-81-willing-donate-their-organs-case-their-death.
Johnson Eric J., Goldstein Daniel. (2003), "Do Defaults Save Lives?" Science, 302 (5649), 1338–39.
Katz Karen L., Larson Blaire M., Larson Richard C. (1991), "Prescription for the Waiting-in-Line Blues: Entertain, Enlighten, and Engage," MIT Sloan Management Review, 32 (2), 44–53.
Kessler Judd B., Roth Alvin E. (2014), "Don't Take 'No' for an Answer: An Experiment with Actual Organ Donor Registrations," Working Paper 20378, National Bureau of Economic Research.
Ku Gillian, Wang Cynthia S., Galinsky Adam D. (2015), "The Promise and Perversity of Perspective-Taking in Organizations," Research in Organizational Behavior, 35, 79–102.
Lacetera Nicola, Macis Mario, Slonim Robert. (2013), "Economic Rewards to Motivate Blood Donations," Science, 340 (6135), 927–28.
Landry Donald W. (2006), "Voluntary Reciprocal Altruism: A Novel Strategy to Encourage Deceased Organ Donation," Kidney International, 69 (6), 957–59.
Leipnitz Sigrun, de Vries Martha, Clement Michel, Mazar Nina. (2018), "Providing Health Checks as Incentives to Retain Blood Donors: Evidence from Two Field Experiments," International Journal of Research in Marketing, 35 (4), 628–40.
Liu Wendy, Aaker Jennifer. (2008), "The Happiness of Giving: The Time-Ask Effect," Journal of Consumer Research, 35 (3), 543–57.
MacKay Douglas. (2015), "Opt-Out and Consent," Journal of Medical Ethics, 41 (10), 832–35.
MacKay Douglas, Robinson Alexandra. (2016), "The Ethics of Organ Donor Registration Policies: Nudges and Respect for Autonomy," American Journal of Bioethics, 16 (11), 3–12.
McIntyre Pat. (1990), "Perceptions of Mexican-Americans and Anglo-Americans Regarding Organ Donation Advertisements," in Organ Donation and Transplantation: Psychological and Behavioral Factors, Shanteau James, Harris Richard Jackson, eds. Washington, DC: American Psychological Association, 97–107.
McShane Blakeley B., Böckenholt Ulf. (2017), "Single-Paper Meta-Analysis: Benefits for Study Summary, Theory Testing, and Replicability," Journal of Consumer Research, 43 (6), 1048–63.
Milaniak Irena, Wilczek-Rużyczka Ewa, Przybyłowski Piotr. (2018), "Role of Empathy and Altruism in Organ Donation Decisionmaking Among Nursing and Paramedic Students," Transplantation Proceedings, 50 (7), 1928–32.
Morgan Susan E., Miller Jenny, Arasaratnam Lily A. (2002), "Signing Cards, Saving Lives: An Evaluation of the Worksite Organ Donation Promotion Project," Communication Monographs, 69 (3), 253–73.
Noyes Jane, McLaughlin Leah, Morgan Karen, Walton Philip, Curtis Rebecca, Madden Susanna, et al. (2019), "Short-Term Impact of Introducing a Soft Opt-Out Organ Donation System in Wales: Before and After Study," BMJ Open, 9 (4), e025159.
O'Carroll Ronan E., Haddow Lorna, Foley Laura, Quigley Jody. (2017), "If You Needed an Organ Transplant Would You Have One? The Effect of Reciprocity Priming and Mode of Delivery on Organ Donor Registration Intentions and Behaviour," British Journal of Health Psychology, 22 (3), 577–88.
Parker Jeffrey R., Lehmann Donald R., Xie Yi. (2016), "Decision Comfort," Journal of Consumer Research, 43 (1), 113–33.
Peterson Dane K., Pitz Gordon F. (1988), "Confidence, Uncertainty, and the Use of Information," Journal of Experimental Psychology: Learning, Memory, and Cognition, 14 (1), 85–92.
Quinn Michael T., Alexander G. Caleb, Hollingsworth Diane, O'Connor Kate Grubbs, Meltzer David. (2006), "Design and Evaluation of a Workplace Intervention to Promote Organ Donation," Progress in Transplantation, 16 (3), 253–59.
Radecki Carmen M., Jaccard James. (1997), "Psychological Aspects of Organ Donation: A Critical Review and Synthesis of Individual and Next-of-Kin Donation Decisions," Health Psychology, 16 (2), 183–95.
Radecki Carmen M., Jaccard James. (1999), "Signing an Organ Donation Letter: The Prediction of Behavior from Behavioral Intentions," Journal of Applied Social Psychology, 29 (9), 1833–53.
Reed Americus, II, Aquino Karl, Levy Eric. (2007), "Moral Identity and Judgments of Charitable Behaviors," Journal of Marketing, 71 (1), 178–93.
Reinhart Amber Marie, Marshall Heather M., Feeley Thomas Hugh, Tutzauer Frank. (2007), "The Persuasive Effects of Message Framing in Organ Donation: The Mediating Role of Psychological Reactance," Communication Monographs, 74 (2), 229–55.
Robitaille Nicole, House Julian, Mazar Nina. (2020), "Effectiveness of Planning Prompts on Organizations' Likelihood to File Their Overdue Taxes: A Multi-Wave Field Experiment," Management Science, (published online November 10), https://doi.org/10.1287/mnsc.2020.3744.
Saab Sammy, Saggi Satvir S., Akbar Mizna, Choi Gina. (2018), "Presumed Consent: A Potential Tool for Countries Experiencing an Organ Donation Crisis," Digestive Diseases and Sciences, 64 (5), 1346–55.
Sadler D. Royce. (1989), "Formative Assessment and the Design of Instructional Systems," Instructional Science, 18 (2), 119–44.
Sallis Anna, Harper Hugo, Sanders Michael. (2018), "Effect of Persuasive Messages on National Health Service Organ Donor Registrations: A Pragmatic Quasi-Randomised Controlled Trial with One Million UK Road Taxpayers," Trials, 19 (1), 513.
Sanner Margareta A., Hedman Håkan, Tufveson Gunnar. (1995), "Evaluation of an Organ-Donor-Card Campaign in Sweden," Clinical Transplantation, 9 (4), 326–33.
Sargeant Adrian, Woodliffe Lucy. (2007), "Individual Giving Behavior: A Multi-Disciplinary Review," in The Routledge Companion to Nonprofit Marketing, Sargeant Adrian, Wymer WalterJr., eds. London: Routledge, 111–44.
Shadish William R., Cook Thomas D., Campbell Donald T. (2002), Experimental and Quasi-Experimental Designs for Generalized Causal Inference. Boston: Houghton Mifflin.
Siegel Jason T., Alvaro Eusebio M., Crano William D., Gonzalez Amelia V., Tang Julia C., Jones Sara P. (2010), "Passive-Positive Organ Donor Registration Behavior: A Mixed Method Assessment of the IIFF Model," Psychology, Health and Medicine, 15 (2), 198–209.
Stats Canada (2020), "Data," (accessed September 29, 2020), https://www150.statcan.gc.ca/n1/en/type/data.
Steffel Mary, Williams Elanor F., Tannenbaum David. (2019), "Does Changing Defaults Save Lives? Effects of Presumed Consent Organ Donation Policies," Behavioral Science and Policy, 5 (1), 69–88.
Stijnen Mandy M.N., Dijker Anton J.M. (2011), "Reciprocity and Need in Posthumous Organ Donation: The Mediating Role of Moral Emotions," Social Psychological and Personality Science, 2 (4), 387–94.
Sunstein Cass R. (2013), Simpler: The Future of Government. New York: Simon & Schuster.
Sunstein Cass R. (2020), "Sludge Audits," Behavioural Public Policy(published online January 6), https://doi.org/10.1017/bpp.2019.32
Swann William B.Jr, Gill Michael J. (1997), "Confidence and Accuracy in Person Perception: Do We Know What We Think We Know About Our Relationship Partners?" Journal of Personality and Social Psychology, 73 (4), 747–57.
Thaler Richard H. (2018), "Nudge, Not Sludge," Science, 361 (6401), 431.
Thaler Richard H., Sunstein Cass R. (2020), Nudge: The Final Edition: Improving Decisions About Money, Health, and the Environment. New York: Penguin.
Toews Maeghan, Caulfield Timothy. (2016), "Evaluating the 'Family Veto' of Consent for Organ Donation," CMAJ, 188 (17/18), E436–37.
Trillium Gift of Life Network (2014), "Trillium Gift of Life Network Annual Report 2013/14," (accessed April 29, 2020), Available at: https://www.giftoflife.on.ca/resources/pdf/Trillium%5fAR%5f13-14%5fEng%5fAccessible%5fWEB.pdf.
Trillium Gift of Life Network (2017), "Trillium Gift of Life Network Annual Report 2016/17," report (accessed April 29, 2020), https://www.giftoflife.on.ca/resources/pdf/Trillium%5fAR%5f16-17%5fENG%5faccess%5f10%5f12%5f2017.pdf.
Trillium Gift of Life Network (2020), "About Donation," (accessed December 21, 2020), https://beadonor.ca/about-donation.
Trivers Robert L. (1971), "The Evolution of Reciprocal Altruism," Quarterly Review of Biology, 46 (1), 35–57.
Tsai Claire I., Klayman Joshua, Hastie Reid. (2008), "Effects of Amount of Information on Judgment Accuracy and Confidence," Organizational Behavior and Human Decision Processes, 107 (2), 97–105.
Tusche Anita, Böckler Anne, Kanske Philipp, Trautwein Fynn Mathis, Singer Tania. (2016), "Decoding the Charitable Brain: Empathy, Perspective Taking, and Attention Shifts Differentially Predict Altruistic Giving," Journal of Neuroscience, 36 (17), 4719–32.
Tversky Amos, Kahneman Daniel. (1981), "The Framing of Decisions and the Psychology of Choice," Science, 211 (4481), 453–58.
U.K. Cabinet Office (2013), "Government Digital Strategy: December 2013," policy paper (December 10), https://www.gov.uk/government/publications/government-digital-strategy/government-digital-strategy.
Underwood Bill, Moore Bert. (1982), "Perspective-Taking and Altruism," Psychological Bulletin, 91 (1), 143–73.
Winterich Karen Page, Mittal Vikas, Aquino Karl. (2013), "When Does Recognition Increase Charitable Behavior? Toward a Moral Identity-Based Model," Journal of Marketing, 77 (3), 121–34.
Wyer Robert S.Jr. (2008), "The Role of Knowledge Accessibility in Cognition and Behavior," in Handbook of Consumer Psychology, Haugtvedt Curtis P., Herr Paul M., Kardes Frank R., eds. New York: Lawrence Erlbaum Associates, 31–76.
Wyer Robert S.Jr, Srull Thomas K. (1989), Memory and Cognition in its Social Context. Hillsdale, NJ: Lawrence Erlbaum Associates.
~~~~~~~~
By Nicole Robitaille; Nina Mazar; Claire I. Tsai; Avery M. Haviv and Elizabeth Hardy
Reported by Author; Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 72- Inefficiencies in Digital Advertising Markets. By: Gordon, Brett R.; Jerath, Kinshuk; Katona, Zsolt; Narayanan, Sridhar; Shin, Jiwoong; Wilbur, Kenneth C. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p7-25. 19p. DOI: 10.1177/0022242920913236.
- Database:
- Business Source Complete
Inefficiencies in Digital Advertising Markets
Digital advertising markets are growing and attracting increased scrutiny. This article explores four market inefficiencies that remain poorly understood: ad effect measurement, frictions between and within advertising channel members, ad blocking, and ad fraud. Although these topics are not unique to digital advertising, each manifests in unique ways in markets for digital ads. The authors identify relevant findings in the academic literature, recent developments in practice, and promising topics for future research.
Keywords: digital advertising; ad effect measurement; advertising channel friction; ad blocking; ad fraud
Digital advertising markets offer unprecedented innovations to marketers. Businesses can now advertise to finely targeted sets of individuals with customized commercial messages at specific locations and times in a variety of formats. Compared with traditional advertising, digital ads promise better targeting and relevance, personalized ad content, programmatic sales based on real-time auctions, and measurement of the co-occurrence of individual consumer ad exposures with a variety of online and offline response behaviors. These features have fundamentally altered marketers' spending: digital advertising revenues reached $108 billion in 2018, up 117% over 2014 ([67]), and they are expected to grow 19% and surpass cumulative traditional advertising revenues in 2019 ([47]).
Digital advertising markets sell a wide variety of search and display advertising opportunities to marketers. Although digital advertising has existed since 1995, market structures are still changing rapidly. For example, a census of marketing technology firms showed an increase from 150 in 2011 to 7,040 in 2019 ([22]). The IAB Tech Lab recently introduced a series of broad-based initiatives, including a new real-time bidding standard, and Google moved from second-price to first-price auctions for display ads. Publishers have introduced a variety of ad formats, with spending typically following consumer attention and media usage: after initial growth in desktop display and search, recent growth has been more concentrated in social networking, video, audio, and mobile ads.
However, there are indicators that unregulated markets for digital advertising have experienced problems. For example, the European Union has fined Google more than $9 billion in three antitrust cases, and the U.S. Federal Trade Commission fined Facebook $5 billion after it broke a 2012 consent order (Case 19-cv-2184). Prominent politicians have criticized the industry and proposed structural reforms. New privacy laws mandate transparency and consent requirements for data-driven advertising and user identification practices.
Several comprehensive reviews of digital platform markets have advised new regulation. For example, the [10] made 23 recommendations, including that "a specialist digital platforms branch be established" to proactively monitor digital markets, enforce laws, conduct inquiries and recommend actions to address consumer harm and market failure. [64] reached similar conclusions, noting "a lack of understanding among policy-makers." Similar conclusions have been reached by the European Commission Directorate-General for Competition; the U.K. Department of Digital Culture, Media & Sport; and the U.K. Digital Competition Expert Panel; moreover, the U.S. Federal Trade Commission and the U.K. Competition and Markets Authority, among others, are conducting ongoing investigations.
The purpose of this article is to review four prominent features that may limit digital advertising markets from reaching their maximum allocative efficiency (i.e., the extent to which markets distribute digital advertising opportunities to the agents that value the opportunities most, as in [58])[ 3]:
- Ad effect measurement is the estimation of incremental effects of advertisements on consumer behaviors. Ignorance or uncertainty about ad effects may inefficiently distort advertisers' reservation prices, demand, budgets, and bids for ads.
- Organizational inefficiencies occur within advertising organizations and between advertisers and their self-interested agencies; these inefficiencies may lead to suboptimal advertising decisions.
- Ad blocking is a consumer technology that prevents ads from being displayed. Ad blockers may inefficiently appropriate advertising revenues and harm publishers' incentives to provide content.
- Ad fraud is a collection of practices that misrepresent advertising inventory or disguise machines as humans in order to steal advertising expenditures. Most industry estimates indicate that fraud takes 10%–30% of total digital advertising revenue.
These features are important: 75% of brand marketers indicated that ad effect measurement is the leading "threat to digital ad budgets in 2019," and 69% of agency professionals reported that ad fraud as the leading threat ([15]). We dedicate one section to each topic, selectively describing relevant scientific literature and credible industry knowledge and then identifying opportunities for future research. Although we discuss each issue in isolation, they may interact to exacerbate some of the inefficiencies we describe. We close with a broader discussion of policy-relevant research opportunities.
It is important to note that these four inefficiencies predate digital advertising markets. More than 100 years ago, an executive famously claimed that half of his ads were working but lamented his inability to determine which half. The first newspaper advertising agent represented advertisers but was paid commissions by publishers ([33]). Before online advertising, ad blocking affected television ad markets via the remote control, video cassette recorder, and digital video recorder. Numerous media outlets have fraudulently overreported circulation numbers to increase ad prices. Traditional advertising markets developed partial solutions; for example, the Audit Bureau of Circulation verifies print publishers' audience data. However, all four inefficiencies manifest differently in digital advertising markets than in traditional advertising markets and therefore are likely to require different solutions.
We write for several audiences. Many aspects of digital advertising markets remain poorly understood, so we describe opportunities for future research. We hope these suggestions appeal to academics as well as the rapidly growing number of scientists employed by large technology companies ([ 9]). We also write for business practitioners who seek to understand digital advertising markets, to make individual campaigns more efficient, or to develop new ventures that can enhance market efficiency. Finally, we aim to help policy makers understand root causes of inefficiencies in digital advertising markets. Extant investigations have given less attention to the four inefficiencies discussed subsequently than to other issues (e.g., antitrust, brand safety, disinformation, consumer privacy, market transparency). We do not take a position on whether or how governments should regulate digital advertising markets, but we hope this article will help inform policy makers about certain key issues.
Advertising effect measurement is the process of quantifying the incremental effect of an ad on consumer behavior. The incremental (or causal or marginal) effect of an ad is the incremental number of outcomes obtained as a result of the campaign, such that these outcomes would not have occurred in the absence of the campaign.
Firms measure ad effects to quantify the incremental return of their marketing investment—to determine whether their money was well spent—and use this information to help guide future marketing decisions. Other motivations may include satisfying internal reporting requirements and benchmarking one campaign's performance against others.
Broadly speaking, campaign outcomes can be divided into two groups: branding and direct response (DR; or performance). Brand advertising seeks to enhance the firm's long-term prospects by improving consumers' perceptions, attitudes, and awareness of the brand. However, quantifying these metrics, and linking them to profits, can be difficult. In contrast, a DR campaign is about driving immediate results on a conversion metric, such as purchases, store visits, registrations, downloads, or webpage visits. Our discussion of ad effects primarily involves DR campaigns, although it is important to recognize that brand advertising represents a significant portion of online ads.[ 4]
In many ways, advertising measurement boils down to answering a simple question: Did the campaign work? And yet advertising measurement is anything but simple. The purpose of this section is to highlight common measurement challenges, to review measurement techniques, and to propose questions for future research. This review is intentionally selective and should not be viewed as exhaustive, as our strategy is to convey the challenges and methods insofar as they explain market inefficiencies and motivate suggestions for future work.
The practice of ad measurement is hardly new. However, there is a sense that digital advertising measurement presents a mix of both old and new challenges. These challenges include, but are not limited to, the following:
- Measurement and data availability. Despite the volume and granularity of data available, many advertisers are still unable to connect ad exposures to outcomes at the individual level. Among other reasons, this is due to long purchase cycles, unobserved stages of consumer decision making, and a lack of access to distribution channel members' customer data. Proving the connection between measurable outcomes, such as clicks or likes, and bottom-line metrics, such as sales, is another common struggle.
- Strategic advertiser behavior. Marketers often target their advertising (with regard to, e.g., timing, characteristics, location) based on expectations of future demand; for example, a car dealer could advertise a temporary price reduction before a holiday weekend. Ad treatment often correlates with ad response—and possibly other marketing actions, such as temporary price reductions—creating a confounding correlation between ad levels and outcomes ([11]).
- Strategic platform behavior. Digital platforms optimize their advertising delivery. For example, if a digital advertising platform is paid per click, the platform attempts to show ads to consumers with high predicted click probabilities, creating another source of confounding variation.
- Strategic consumer behavior. Consumers may pay attention to, or withhold attention from, advertisements according to ingrained habits or their valuation of the ad content ([13]; [122]). As a result, it may be difficult for advertisers to distinguish incremental effects of ads from consumers' baseline propensities to attend to the brand's ads. An implication is that many ad experiments recover something closer to an intent-to-treat effect rather than average treatment effects or treatment-on-the-treated effects.
- The complexity of ad effects. Advertising effects may be nonlinear in the number and types of ads a consumer sees. Marginal effects of ads vary with wear-in, wear-out, or weariness ([25]; [108]); competitor advertising ([35]; [111]); and ads in other media (e.g., [72]; [81]; [93]). The inability to measure all of a consumer's advertising exposures makes it difficult to obtain a fully accurate view of ad effects in many settings.
The severity of these challenges varies across campaigns, but all represent significant obstacles to quantifying ad effects reliably and accurately.
Next, we discuss the two major approaches to measuring ad effects: designing strategies to create experimental data ex ante and analyzing observational data ex post. Randomized experiments have become increasingly common as more online ad platforms provide such capabilities to advertisers, although usage still appears to be limited. In contrast, methods that rely on observational data are often more accessible, and thus, marketers have more broadly adopted them. The next two subsections discuss the strengths and weaknesses of each approach with brief discussions of some representative work. The final subsection discusses some promising directions for future research.
Advertising experiments are widely considered the gold standard to estimate the effects of a marketing action on consumer behavior. Experiments randomly allocate units (e.g., consumers, markets) across treatment and control conditions. With sufficient sample sizes, on average the only difference between conditions is advertising in the treatment condition. Incremental conversions can be calculated by comparing outcomes between the advertising treatment condition to conversions in the no-advertising control condition.
By creating exogenous variation by design, randomized experiments address a subset of the measurement problems discussed previously. A randomly allocated control group directly addresses the problem of strategic advertiser behavior by inducing exogenous variation in treatment. Furthermore, most online ad platforms implement experiments in a balanced manner across conditions so as to neutralize the issue of strategic platform behavior.
The academic literature increasingly uses experiments to understand advertising effects, including influential collaborations between academics and firms. The pioneering work includes a collection of papers by Randall Lewis, David Reiley, and Justin Rao, who explored the effectiveness of digital display advertising on Yahoo.com by randomizing the ad shown to different visitors ([84]; [80]).[ 5] This work was among the first to make clear both the benefits and difficulties of relying on large-scale online experiments for the purpose of measuring ad effects. [85] provide an excellent review of the early academic literature on measuring digital advertising effects.
Sometimes the unit of randomization is at the geographic market level, rather than the individual consumer. In these cases, researchers have often relied on quasi-experiments or matched market tests to generate suitable treatment and control groups ([74]; [123]).
[20] document a particularly influential geo-experiment, in which eBay halted Google search advertising in a sample of cities and continued search ads in other cities. Prior to the study, executives at eBay believed sponsored search was effective at generating incremental sales. However, the results indicated a return on advertising spending (ROAS) of −63%, mainly because consumers could easily substitute organic search results for eBay when the sponsored links were removed. This was possible because eBay's organic links ranked highly and competitors' keyword ads appeared infrequently.
eBay's market leadership position makes it natural to wonder whether these results would generalize to firms with weaker market positions. This observation motivated at least two subsequent studies. First, [32] used the same market-level research design to evaluate paid search at Edmunds.com, for which organic search results do not enjoy the same high rank as eBay. The authors found that half of paid traffic was lost when branded paid search was turned off. Second, [113] also found that the result did not hold similarly for most other online retailers, as competitors could poach branded keyword searchers when the focal brand does not purchase ads on its own keywords. These two studies suggest that the findings in [20] could be limited to companies with a similar market position as eBay, and they highlight the challenges in generalizing results from any single experiment to other settings.
Major digital platforms increasingly offer tools for advertisers to run experiments to measure incrementality on their platforms. These tools are usually offered freely, though sometimes only to sufficiently large advertisers, with the goal of helping them improve their advertising outcomes—the implicit assumption being that this will lead to increased advertising spend on the platform. A nonexhaustive list of platforms that offer experimentation tools includes AdRoll, Facebook, Google, JD.com, MediaMath, Microsoft, and YouTube.[ 6] Although these tools are increasingly popular, widespread adoption by advertisers may take some time.[ 7]
[114] provide some systematic public evidence about advertisers' usage of experiments. The authors examine online retailers' search advertising expenditures at Bing.com during the period when the results from [20] were publicized in the trade press and popular media. [114] found that 11% of firms whose branded keyword ads were not regularly purchased by competitors (similar to eBay) discontinued brand search advertising. However, the incidence of experimental advertising variation was essentially unchanged and was uncorrelated with the size of individual firms' advertising effects. In summary, some firms reduced their ad spend on their own branded keywords after the eBay study results were publicized; but it appears that they did so without running their own tests first.
A common problem with ad experiments is that many ad effects are small and require surprisingly large sample sizes to achieve reasonable statistical power. [82] reported the results of 25 digital display advertising field experiments with samples of 500,000 people or more. Despite large sample sizes, most experiments produced confidence intervals for ROAS wider than 100%, with the smallest confidence interval exceeding 50%. Advertising response data are highly variable, making it difficult to separate advertising's influence on conversions from unobserved factors. The difficulty of measuring small advertising effects is a fundamental problem for advertising measurement. Whereas a product's price acts like a hammer on consumer behavior, advertising is perhaps closer to a feather in a strong wind.
Cost is another barrier. Every consumer held out of the treatment group is a consumer who does not see the ad. If the ad is profitable, the firm may be unwilling to bear the opportunity cost of not treating consumers in the control group. Compounding this problem, some digital ad experiments serve consumers in the control group with public service announcements at the advertiser's expense. When ad effects are unknown, one might reasonably regard experiment costs as either positive or negative, depending on one's prior about the ad effects.
Lower costs of experimentation could increase the number of ad experiments that are run (e.g., [109]). [70] proposed "ghost ads" as a method to "identify ads in the control group that would have been the focal advertiser's ads had the consumer been in the treatment group." Google implemented ghost ads and has reduced experimentation costs by an order of magnitude. [71] present a meta-study of 432 field experiments at Google that used the ghost ads design. Ghost ads appears to be gaining traction within the advertising industry ([95]).
[107] propose another approach to lowering ad experiment costs in the context of retargeting (i.e., targeting ads to previous website visitors). They use a single-impression public service announcement campaign to tag all consumers who were eligible for retargeting ads. This created a single control group, offering a valid baseline to estimate treatment effects of multiple varying advertising intensities.
As low-cost experimentation technologies continue to be developed and gain wider acceptance, we expect advertisers to increasingly adopt—and perhaps even demand—experimentation services on other advertising platforms.
Most firms either do not or cannot measure ad effects using experimental or quasi-experimental methods. Instead, they rely on observational data collected in the normal course of business. In such cases, academics and firms have developed many techniques to estimate ad effects from observational field data.
Probably the most common approach among firms is to analyze the market-level relationship between sales and advertising. This top-down approach is known as a marketing mix model (MMM), or similarly a Media Mix Model, which dates back to at least the 1960s ([21]). The model typically takes the form of a time-series or panel regression of aggregate sales on aggregate marketing spending, or impressions, in each advertising medium. Additional controls include factors such as macroeconomic conditions, weather, seasonality, other marketing mix variables (e.g., price), and competitor activity. Once estimated, the model can be used to measure ROAS and to forecast sales at various levels of marketing spending in different channels.
[27] present an excellent discussion of the uses of MMMs, highlighting pitfalls and identifying opportunities for improvement. There are several arguments in favor of MMMs. First, the scope at which they operate—aggregate spending—allows MMMs to play a key role in a firm's budgeting process, helping the firm divide its market budget across media. It would be difficult for many firms to implement a sufficient number of experiments to trace out ad effects across spending levels and media; in contrast, an MMM relies only on historical data. Second, the firm does not bear any opportunity costs of holding back advertising from a control group. Third, firms are more often capable of satisfying the technical and data requirements of MMMs. Fourth, MMMs can help socialize the overall concept of advertising measurement within the organization, which can help build internal support for more rigorous measurement practices in the future.
In the absence of exogenous variation, these methods rely on the assumption that they produce valid incremental effects. However, given the challenges of advertising measurement raised earlier, we should expect that some of the variation in advertising levels is not independent of the error term, as firms often divide advertising budgets across time, markets, and products in expectation of future demand. Furthermore, ad spending in different media tends to be highly correlated, which makes it difficult to isolate channel-specific ad effects. To be clear, these concerns represent significant hurdles to the potential causal interpretation of effects obtained from observational data. That said, in many situations this is the most feasible path advertisers can take, so many make do with these imperfect ad effect estimates to inform their decision making.
In contrast to settings with aggregate data, another common paradigm is to use individual-level data (or cookie-level data) to measure advertising effects. In these cases, advertisers often apply multi-touch attribution models, which are bottom-up approaches that seek to assign credit (to attribute) to each ad that preceded a user's conversion.[ 8] For example, suppose that a user clicks on a sponsored search ad, then clicks on a display ad, and then makes buys the advertised product online. To what extent did either ad contribute to generating the conversion? The most common approach is a last-click attribution model that assumes that the display ad is solely responsible for the conversion, because it occurred last. Many variations on this style of model exist, and a survey indicates their usage is widespread ([51]).
A related measurement approach using individual-level data compares the conversion rates of users who were exposed to an ad campaign with users who were not exposed (e.g., [ 2]; often called "lift"). Lift methods rely on consumer characteristics data and model the joint likelihood of exposure and conversion to estimate the ad effect. The hope is that, after controlling for enough characteristics, the two groups are similar to the point where the ad effect can be given a causal interpretation. However, because exposure is influenced by advertiser and platform targeting strategies, conversion differences between groups may be caused by strategic behavior or unobserved characteristics, even in the absence of advertising.
[55] demonstrate the difficulty of using observational data to estimate valid incremental ad effects. They reanalyzed 15 Facebook ad experiments comprising 1.6 billion ads served to 500 million users. Despite having rich data, they were unable to recover true ad effects accurately or consistently using observational methods. Even the sign of the bias was unpredictable: most ad effects were overestimated, but some were underestimated. As the authors put it, "even good data prove inadequate to yield reliable estimates of advertising effects."
Fortunately, it is possible to use observational data to address some of the advertising measurement problems. In general, quasi-experimental approaches make use of specific information about the timing of events or the data-generating process to lend a causal interpretation to the estimates and may require additional assumptions. Causality is not free.
One such approach is found in [94]. The data they analyzed came from a casino that had previously targeted promotions to consumers based on their observed gaming behaviors. The authors exploited knowledge of the firm's previous targeting policy to obtain unbiased estimates of consumer response to promotions. Using the estimates to segment consumers, they show, via a field experiment, that the segmentation scheme contributed to a targeting policy that generated higher incremental profits than alternate policies.
Another strategy is to take advantage of local randomization with respect to time or some other variable. [87] measured changes in brand website traffic and conversions in narrow two-minute windows around the airing times of television advertisements. They based the assumption of quasi-random advertisement treatment times on a detailed understanding of television networks' sales practices, which resulted in quasi-random ordering of advertisements within commercial breaks. This helped ensure that systematic differences between traffic and conversions in pre-ad and post-ad windows of time could be attributable solely to the presence of the TV ads. [96] studied position effects in search advertisements by using local randomness in position when competing advertisers were very close to the threshold of winning or losing the bid for a particular position. This local randomness allows for any differences in subsequent behaviors by consumers to be attributed to the position the ad was in.
Finally, advertising exposure may be influenced by factors that are arguably independent of consumers' preferences for the brands being advertised. [59] implement this strategy, using quasi-random geographic variation in exposure to ads during the Super Bowl to study the effects of ads in the beer and soda categories. Probability of exposure to Super Bowl ads depends on the performance of a consumer's local team. Super Bowl ads are planned at the national level many months before the event, so advertising levels should be independent of shocks to local audience sizes.
These examples demonstrate that advertisers can measure some ad campaign effects without relying on randomized experiments. However, extra work is required to identify these situations, and they may not cover all the scenarios relevant to advertising decision making.
Measuring the returns to all of a firm's digital advertising expenditures is a daunting task, especially if the goal is to do so continuously and to immediately incorporate insights into the next advertising decision. Competent, expensive, large-scale attempts to estimate ad effects have failed to measure effects with reasonable precision. Following are some questions that researchers could consider for the future and some brief thoughts on how to make progress on them:
- Advertisers can improve the usefulness of observational models. One strategy is to practice continuous experimentation across geographies, media, and groups of consumers, building more random variation into the data ([129]). For example, some digital advertising platforms offer targeting at the zip code level, which could enable greater statistical power than traditional market-level randomizations. It would be useful to understand how firms could efficiently design this experimentation and how to incorporate it into existing models.
- Firms can produce and leverage larger panel data sets to improve statistical power. Most advertising experiments are implemented in traditional static split/test designs. However, large panel data sets may offer increased statistical power ([18]). For example, if the timing of advertising treatment can be randomized, treatment/control comparisons could potentially be made within individual consumers, thereby removing confounding individual fixed effects. A few firms can track both ad exposures and conversions in single-source customer data (e.g., Apple, AT&T, Google, Verizon). Market research firms and internet service providers are probably best positioned to create single-source panel data that measure both ad exposures and ad responses for a large number of brands, although doing so would require compliance with data privacy rules. Concerns may also remain about endogenous person-/time-based targeting if ads are delivered nonrandomly.
- Advertising agencies could develop approaches that integrate ad effect measurement with advertising procurement. One such approach, described in [86] and implemented by an agency called Nanigans, involves manipulating advertisement bids to induce exogenous variation in advertisement selection. [126] addressed a similar problem at JD.com by using a multi-armed bandit approach to combine continuous experimentation with optimal exploitation of experimental findings.
- It may be possible to develop new quasi-experimental designs for digital advertising. A variety of such strategies have been used in other media. For example, television ad effects have recently been estimated using such instrumental variables as advertising treatment discontinuities at geographic television market borders ([111]), exogenous shocks to TV ad demand during elections ([115]), exogenous TV ad insertion timing (e.g., [41], among others), and assignments of TV networks to local cable channel numbers ([19]). It may be possible to find and exploit similar exogenous discontinuities in digital advertising markets to measure causal ad effects in observational data. For example, [60] show that, under particular assumptions, variation in ad viewability can help measure ad effects, considering that approximately 45% of ads rendered never appear in the viewable portion of a user's screen.
- Firms may be able to integrate the results from observational models and randomized experiments. Some firms use experiments for tactical decisions (e.g., does this new ad design outperform the previous design?), whereas they rely on MMMs for strategic decisions (e.g., how much budget should we allocate to search versus display?). The information contained in the experiment could help inform the aggregate model, but it is not as simple as feeding the results of the experiment into the MMM. One possibility is that effects from experiments could be imposed as informative priors when estimating the MMM. Whatever the solution, many marketing measurement firms are actively searching for a unified measurement model to help deliver on this broad goal.
- Managers may be able to embed experimentation within their profit-maximizing objective. The estimation of causal effects produces an input—the ad effect—intended to aid a manager's decision-making process. Ideally, the decision to experiment and to integrate experimental findings into profit maximization can help the manager to make use of the results for subsequent marketing decisions. [49] provide a recent example of such a framing.
- Digital ad sellers may require techniques to validate their experimentation platforms. An increasing number of platforms offer experimentation services to help measure ad effects. Experimentation on ad platforms typically allows the advertiser to achieve greater scale, and the platform's control of the experiment should help it develop higher-quality measurement technology. However, ad platforms also face a possible credibility problem, as advertisers may worry that the platforms have an incentive to inflate ad effectiveness estimates.[ 9] To date, we are unaware of any systematic external audits or validations of the experimental tools that platforms provide to advertisers.
- As more platforms offer experimentation as a service and more advertisers take advantage of this capability, it is unclear how marketers should interpret the treatment effects obtained when these effects depend on competitors' advertising policies. An experiment delivers the ad effect that is incremental conditional on all other marketing activities in the market. [88] provide a framework for interpreting ad effects obtained under parallel experimentation. Future work could provide guidance on how to best generalize experimental ad effects, which may require the advertiser to form beliefs about competitors' future policies.
In summary, advertising effect measurement can increase advertising profits, but good solutions are often unavailable. Ad effects are typically small, and conversions are highly variable, so conclusive experiments are costly and rare. Observational data are plentiful, but statistical analyses often do not uncover causal effects because advertisements are allocated strategically in ways that correlate with treatment. Although many people may share the intuition that it should be possible to design large-scale studies to systematically estimate ad effects, it is not yet clear when that goal will be achieved.
Classical economic theories of market efficiency assume that purchasers allocate budgets to maximize their own welfare. However, markets for digital ads are intermediated by specialists whose self-interested actions may distort advertising decisions.
Principal/agent problems and moral hazard are well-known sources of economic inefficiency. Generally speaking, asymmetric information enables agents to extract inefficient information rents from principals ([78]). Perhaps the best-known example is the dead-weight loss due to double-marginalization in retail pricing, which results from the divergence of manufacturer and retailer interests in a distribution channel. However, contracting problems have received limited attention in the specific context of digital advertising markets. For example, many marketers ask digital advertising agencies to evaluate their own performance; under what conditions and contracts are agencies incentivized to report truthfully?
We discuss two types of organizational inefficiencies arising in the digital advertising ecosystem. Intrafirm inefficiencies may result from misalignments between corporate officers, functional departments, or business units within a firm due to different incentives or strategic objectives. Interfirm inefficiencies may occur between marketing organizations and their external agencies, between competing agencies serving the same marketing client, between complementors within a value chain, or between colluding purchasers and sellers of advertising.
Chief financial officers often set marketing budgets and review marketing financial performance, a common source of tension within a firm ([92]). Many finance executives believe that marketing executives should be able to measure return on advertising spend ([50]). As explained previously, such beliefs may be misplaced. Misalignment of internal beliefs and incentives may distort marketing tactics. Marketing may be perceived as a short-run lever that can be used to meet particular financial targets. Marketing managers may hire outside consultants to shift blame away from themselves in the event of poor or unmeasurable outcomes. For example, they may use retargeting campaigns, which send advertisements to shoppers who previously viewed the firm's website and therefore have a higher expected baseline propensity to purchase. This creates a positive correlation between ad exposure and conversion. Lacking better causal metrics, managers may incorrectly present inflated lift metrics derived from correlations ([79]; [107]).
Another potentially unintended consequence of an internal desire to measure ad effects is a distortion between short-run and long-run objectives. It is seldom possible to reliably estimate advertising effects on long-run outcomes, such as brand attitudes ([41]). In contrast, direct response campaigns focus on short-run actions that are easier to reliably link to ad exposure. Tension between long- and short-run objectives is difficult to quantify, but it is possible that underinvesting in brand-building advertising could reduce long-run profits. Also pertinent in this context is the potential misalignment between the firm's time horizon and that of the manager, with the manager often taking a more short-run view than the firm as a whole.
A third type of intrafirm inefficiency may result from misalignment between functional groups within the organization. Previous research has shown that poor integration of marketing and sales teams may lead to suboptimal advertising policies and customer focus ([61]; [117]). Marketing goals may conflict with other functional groups as well, such as when the marketing objectives are misaligned with those of the procurement department. For example, procurement may use a reverse auction to award a contract to an advertising agency to meet a goal of minimizing expenditures. The low-bidding agency may then provide lower-quality service or fail to fully realize marketing objectives ([97]).
Finally, distinct business units, brands or campaigns within the same firm may enter rivalrous bids on the same advertising inventory. For example, [96] show that after a large retailer acquired three of its rivals, the four brands continued to compete with each other by entering rivalrous bids in advertising keyword auctions.
Many digital advertisers have replaced the traditional agency-of-record model with a complex array of external agencies. When a consumer requests a web page, any subset of the following players may be involved in delivering each digital display ad impression: publisher, ad server, supply-side platform, ad network, ad exchange, demand-side platform, multiple data management platforms, third-party verifiers, ad agency, and, finally, the advertiser ([31]). Each agent takes a cut of the advertising transaction. The Association of National Advertisers (ANA; [ 7]) concludes that "58 cents of each dollar ultimately purchased media inventory and audience exposure from a publisher, with 42 cents of programmatic investment consumed by supply chain data and transaction fees." Therefore, the share of digital ad spend going to intermediaries is nearly triple the traditional 15% agency commission, although it is split between many more entities in the programmatic advertising supply chain.
In addition to the programmatic marketplace, advertisers can purchase guaranteed inventory. In such transactions, the advertiser specifies the parameters of the purchase (e.g., targeted demographics, number of ads to be delivered, time frame) and the agency quotes a price. However, [73] documented widespread obfuscation of agency markups in such transactions. Arbitrage may reduce market efficiency due to moral hazard and asymmetric information between advertiser and agency, although agencies have countered that they can add value to low-cost advertising inventory by applying analytics.
Another type of inefficiency may arise when advertisers fail to coordinate agencies whose work complements each other or generates spillovers across marketing channels. As one example, "last-click" attribution models incentivize digital agencies to compete over credit for attributed conversions. As another example, a large body of research has found significant spillovers between advertising placements in traditional media and digital activities such as brand search, website traffic, online sales, and social media conversations.[10] Few marketers are known to coordinate their advertising activity to account for such spillovers. An integrated channel would consider the effect of traditional advertising on volume of online search, types of keyword searched, search advertising expenditure, and sales, and would quantify spillovers to allocate advertising budgets across media ([77]). However, the typical approach is for marketers to hire specialist agencies for each individual medium (e.g., traditional advertising, search engine marketing, social networking, search engine optimization, website analytics), each of whom operates independently and competes for a larger share of the marketer's advertising budget.
Another type of inefficiency related to advertising agencies is how to properly align incentives to produce efficient advertising creative content. On one hand, advertising agencies' creative ideas can be stolen during the pitch process, especially when clients work with multiple agencies ([34]). Therefore, some agencies may not reveal the full idea during the contracting process, potentially leading to adverse selection or underinvestment in idea production ([62]). On the other hand, advertisers have difficulty evaluating the quality of agencies' creative strategies due to the complexity and high dimensionality of the messaging problem. Although messaging content is still largely driven by human intuition, there is increasing movement toward algorithmic production and validation of advertising messages in digital environments.
Three recent trends affect the market for advertising agency services. First, a long-term trend has emerged toward in-housing, in which marketers create internal advertising agencies. In-housing is used primarily when firms have internal creative abilities or straightforward advertising requirements ([63]). A recent [ 8] survey of large advertisers shows that 78% of advertisers had in-house agencies in 2018, up from 42% in 2008, and that 90% of those advertisers continue to work with external agencies in addition to their internal teams. Second, digital platforms have simplified their customer interfaces, a move that may encourage some advertisers to forgo agency services. For example, Google recently made "smart display campaigns" the default interface to Google Ads. The streamlined purchase process relieves the advertiser of control over bidding, ad placement, user targeting and final control over the ad creative. However, advertisers using this interface may pay advertising costs that exceed advertising revenues, because they no longer can enter a maximum bid per click. Third, there is an emerging trend toward large advertisers requesting log-level files (i.e., impression-level data) from supply chain partners such as advertising exchanges and publishers ([116]). Data sharing enables advertisers to monitor partner agencies and suppliers, trim wasteful spending, and find undervalued inventory.
Another type of interfirm inefficiency may occur when advertisers compete in advertising auctions with complementors within the same value chain. For example, Hilton and Expedia both sell Hilton rooms in New York, and Expedia takes a commission when it sells a Hilton room; yet both place search ads on keywords like "hotel New York." Cooperative advertising policies may even exacerbate this competition: Hilton may subsidize the advertising expenditures of Expedia, even though this directly increases Hilton's keyword advertising costs by making the keyword auctions more competitive. [24] and [68] examine such practices.
Finally, there are inefficiencies that may increase advertisers' profits at the expense of overall market efficiency. One such effect may occur when competing advertisers use a common advertising agency, as doing so may coordinate marketing policies or facilitate information sharing ([124]). Another may occur when the structure of programmatic advertising marketplaces facilitates collusion between purchasers. Digital advertisements are sold in high-frequency auctions with repeated contact among bidders, creating ripe conditions for collusive bidding strategies and punishments for defection. [38] found that advertising agency consolidation is associated with decreased costs in keyword auctions when merging agencies represent competing bidders. There has been a recent trend toward global consolidation of advertising agencies ([99]), suggesting that such effects may be considerable.
We suggest several important research opportunities within this area.
- Recent theoretical work in information economics has introduced Bayesian persuasion models and the analysis of strategic information design ([75], [16]). Bayesian persuasion models characterize optimal informational strategies (i.e., what message to send, to whom, and when or how much information to reveal) to influence receivers' decisions. This seems to be a promising theoretical framework on which to build an understanding of optimal advertising strategies for advertising content, targeting, and platform design ([53]; [91]; [130]).
- Principal/agent– and theory of the firm–style analyses have rarely been applied to advertising agencies. It may be that such settings can be understood with straightforward applications of canonical models. However, given the particular institutional details, we suspect there is substantial scope to extend existing theories in the process of applying them to advertising phenomena. Future research could include normative guidelines about how optimal contract terms depend on features of a market and the contracting parties. Contract terms likely affect incentive alignment, effort elicitation, payment models, in-housing/outsourcing decisions, and agency selection and coordination.
- There are opportunities to further develop and improve techniques to optimize digital advertising content and context. Machine learning has demonstrated progress in identifying sentiments ([23]) across a wide variety of unstructured data including text (e.g., [98]), images (e.g., [89]), and video (e.g., [100]). The meaning of advertising content depends on the context ([103], [104]), such that algorithms are needed to detect and appropriately understand the context and work with multimodal data ([118]). Similar opportunities may be available when a marketer can access log files from multiple partners, as the data may be used to optimize advertising frequency, quantify the impact of context on conversion, optimize procurement techniques, or evaluate ad agency services.
- Perhaps the most promising area for research involves designing auction and advertising allocation systems that more effectively align the incentives of the parties involved (e.g., [65]). [69] propose a cost-per-incremental-action pricing model for advertising, with the goal of aligning all participants' objectives on profit maximization for the advertiser, thereby reducing the misaligned incentives common in advertising pricing models. Additional research is needed to test such models in the field and to adapt them to various digital advertising markets.
Early digital advertising efforts developed intrusive formats such as pop-up ads or autoplaying audio/video ads. This led to consumer demand for ad blockers, applications that allow users to passively block advertising from showing up in their browsers. Most ad blockers came in the form of free-to-use browser extensions enforcing a set of community-defined rules for ads. Recent estimates of ad blocking prevalence vary from 8% to 47% ([110]). In mid-2019, Google incorporated ad blocking features into its market-leading Chrome browser, while continuing to enable third-party ad blockers.
The phenomenon of ad blocking highlights an important tension in the current digital advertising ecosystem. Consumers are flooded by ads, most of which are not interesting or relevant. Prior to ad blocking technology, consumers had no choice but to tolerate the ads; most users were not willing to pay for most content, so content providers relied on advertising revenues. Ad blocking threatens this model, as advertisers' most sought-after consumer segments are often the most likely to install ad blockers.
The most popular ad blockers work by intercepting browser requests to lists of known ad servers, so advertisers are not charged for blocked ads. However, some ad blockers are more aggressive: they only block ads after they are requested, thereby wasting the advertiser's money. One ad blocker, AdNauseam, even clicks on all blocked ads, in order to inject noise into the user data that underpins digital advertising.
Platforms such as web browsers and mobile operating systems enable or restrict ad blocking services. However, platforms that sell ads have conflicting incentives. Their revenues depend on consumer experience but also advertising exposures. It is thus not surprising that critics argue that Google blocks relatively few ads and interferes with competing ad blockers. Mobile operating systems assert more control over applications than traditional desktop operating systems. Apple and Google have rules about apps that interfere with how other apps work, making it difficult to offer blocking of in-app ads.
Ad-supported publishers are perhaps most harmed by ad blocking. Facebook and some other sites deliver ads and content from the same servers, preventing ad blockers from identifying which page elements are ads. Other publishers seek to detect ad blocker usage and withhold content from consumers who do not allow ads; [131] found that 30% of the top 10,000 websites detect ad blockers. Some publishers, including Facebook, invest heavily in website engineering to circumvent ad blockers and unblock ads. Other publishers explicitly request that users whitelist the site (i.e., selectively disable their ad blocker), so the site's ads may be displayed. Software developers have responded with "ad blocker-blocker-blockers," which enables the consumer's browser to obfuscate ad blocker usage, thereby preventing publishers from withholding content.
The overall effect of ad blocking on advertisers, consumers, and publishers is unclear. Using the circulation spiral theory of newspapers (e.g., [54]), one might predict that ad blockers will put advertising-supported media out of business. By reducing ad revenues, ad blockers increase direct-access prices and thereby reduce publishers' incentives to produce high-quality content. Taken to its extreme, ad blockers would destroy the ad-supported internet and thereby become irrelevant, as there would be no more ads to block. A contrary argument holds that ad blocking serves consumer interests: some ad blockers even position themselves as a consumer movement ([76]). Ad blockers say they only block the most intrusive ads and that users can whitelist any site they want. In fact, ad blockers whitelist some sites by default.
This brings us to an important question: How do the ad blockers make money? Most ad blockers do not directly charge users for downloads or ad blocking services. The typical model ad blockers use is to demand payment from publishers in exchange for whitelisting-by-default. If the publisher pays and conforms to some guidelines about ad formats and intrusiveness, the ads will be displayed to ad block users by default. Large publishers typically pay 30% of ad revenue that would otherwise be blocked; small publishers can be whitelisted for free if they agree to follow programs such as the "acceptable ads" standard.
Publishers have argued that such practices amount to extortion, and these allegations have resulted in numerous lawsuits. Some of the most important cases have taken place in Germany where the company that makes Adblock Plus, Eyeo, is based. The German Supreme Court has ruled that ad blocking and the practice of soliciting payment for whitelisting is legal ([57]).
The phenomenon of ad blocking raises a range of future research opportunities.
- It would be worthwhile to study how consumers trade off the intrinsic benefits of content consumption with advertising quantity, advertising nuisance, and subscription prices. [66] ran experiments that showed that increased advertising on Pandora.com directly decreased consumers' site usage and increased consumer subscriptions. It would be interesting to measure similar effects in other contexts; to condition on the nuisance level of the ads served; and to estimate effects of ads on adoption dates of ad blockers, among other relevant outcome variables. It also would be interesting to study how design elements within publisher requests for whitelisting might correlate with response metrics.
- There may be better ways for browsers and ad blockers to improve mechanisms to allocate attention. Although browsers that specialize in privacy and ad blocking have not realized high market shares, some notable experiments are ongoing. For example, the browser Brave implements the "Basic Attention Token" to reward users for looking at ads and send money to publishers the user likes. New designs might take hints available from recent literature on privacy, which shows that consumer valuations of privacy respond strongly to framing and context ([ 3], [128]).
- Recent work by [112] shows that, as the proportion of a site's visitors who use ad blockers increases, the site's quality declines. Further research is needed. How does a site's revenue loss from ad blocking compare with payments to ad blockers? When do sites switch from ads to paid subscription models, unblockable "native advertising," or other blocking-proof business models? Similar questions can be asked for advertisers, especially those whose target customer segments are most likely to use ad blockers. How do ad blocking and whitelisting affect ad placements and ad prices?
- It would be interesting to understand how ad blocking changes product market outcomes. A few recent theoretical pieces address the interaction between publishers and ad blockers ([56]) and differentiation between publishers ([39]), but there are more directions to explore. For example, how do ad blockers compete with each other? Blocking more ads than a rival ad blocker may attract more consumers, and it may also impact the value to publishers of paying to be whitelisted by default. It would be fruitful to quantify these trade-offs, as they likely rely on the extent of consumer multihoming, both across publishers and across ad blockers.
- There is also the question of how advertisers react to ad blocking. [30] argue that ad blocking can incentivize higher quality ads when ads send informative signals. Other fundamental questions should also be addressed. If ad blocking reduces the overall supply of advertising space, this necessarily raises the overall price of reaching customers. The increase could be disproportionate for more ad-averse consumers. How does that affect advertisers? Intuitively, the direct effect should be negative, but it is possible that limited advertising also reduces price competition ([44]) or increases the effects of ads that are not blocked. Future research could dig deeper to examine how other dimensions of advertising content are affected by ad blocking. If a brand has to pay more to reach a consumer, the brand might change how much it focuses on communicating a single piece of information (e.g., price, product feature) or alter the mix of entertaining and informative content contained in the ad.
- Platforms' business models may incentivize them to set ad blocking defaults or limits in mobile operating systems and desktop browser software. Overall, it seems likely that platforms selling ads will restrict ad blocking more than platforms that charge consumers directly for devices and software, in order to limit harm to advertising revenues.
Digital advertising fraud is a collection of practices that misrepresent advertising inventory or disguise machines as humans to steal advertising budgets. It is fundamentally difficult or impossible to measure, but it appears to be widespread.[11][48] reported that recent estimates of fraud vary from $6.5 to $19 billion, and that fraudsters target high-price ad impressions, new markets, and new media. Adobe reported in 2018 that about 28% of website traffic showed "strong non-human signals." The IAB Tech Lab reported that "just 59.8% of clicks could be confirmed as human traffic" ([120]). [102] estimated that 10%–15% of programmatic desktop advertising was invalid traffic. [37] reported that "28% of global mobile media budgets are wasted on fraud." [121] reported that 95% of marketing executives surveyed said that digital media must become more reliable and 21% said they have cut ad spend due to inaccurate, questionable, or false reporting.
We know of six basic motivations to commit advertising fraud.
- Publishers may overreport or misrepresent audience metrics to increase ad revenues.
- Advertisers may click competitors' ads to harm rivals' market inferences, ad budget, or brand awareness, or to reduce competition in ad auctions by depleting a rival's budget.
- A firm may commit detectable advertising fraud that ostensibly benefits a rival, with the goal of inducing a platform or other intermediary to punish the rival.
- Advertising market intermediaries, including agencies, ad networks, and demand- and supply-side platforms, may misrepresent the inventory or the bids they have available in an effort to alter ad prices, ad sales, or commissions. They also may deviate from contracts to manipulate ad auctions, as in the recent bid caching scandal ([36]).
- Publishers or agencies may distribute advertisements surreptitiously or insert false information into advertisers' conversion tracking systems to claim undue credit from advertisers who pay publishers per conversion achieved.
- Firms or individuals may create fraudulent profiles on social networks to falsify measures of influence, ad clicks, or seemingly organic discussions (astroturfing).
Collectively, there are dozens of known schemes to commit ad fraud.
Fraud techniques that use machines to mimic human behavior are often enabled by "botnets," which consist of a central server and a host of malware-infected computers. The server directs connected machines to take specific actions in ways that resemble their owners' organic behaviors, making this machine activity difficult to distinguish from human activity.
Security researchers have publicized several ad hoc botnet discoveries. For example, the discovery of the 3ve botnet in 2018 led to the first criminal charges filed for advertising fraud. The indictment stated that "more than 1.7 million infected computers...download[ed] fabricated webpages and load[ed] ads."[12]
Another set of fraud techniques misrepresents advertising inventory, as several publishers have shown. The Financial Times did not sell inventory in programmatic advertising marketplaces, but it found fraudulent FT.com ad inventory offered in 25 ad exchanges, with fraudulent video inventory exceeding truly available inventory by a factor of 30. The Guardian purchased intermediaries' Guardian.com video inventory on open advertising exchanges and found that 72% of ads purchased falsely claimed to occur on Guardian.com.
Large digital platforms have been inconsistent in their claims about digital advertising fraud. Google states in its advertiser-facing webpages that "we protect you from invalid activity and advertising fraud" and describes a variety of approaches. However, internal communications filed as evidence in a 2017 lawsuit indicated that Google recalled fraudulent funds from publishers but lacked an internal system to return those funds to defrauded advertisers and that Google's advertiser contracts did not require the company to offer refunds to defrauded advertisers ([ 5]). Facebook states that "we believe fake accounts are measured correctly within the limitations to our measurement systems." However, Facebook does not specifically define fake accounts, although its community standards do prohibit "accounts that are fake" without elaboration. Facebook's Community Standards Enforcement Reports show that Facebook "took action on" 6.6 billion fake accounts in the 12 months ending in September 2019 but did not specify the actions taken. Facebook has restated advertising metrics numerous times ([101]).
There are two types of efforts to address digital advertising fraud. The first increases market transparency and accountability to reduce the incidence of advertising supply chain participants from stealing from each other. For example, the IAB Tech Lab defined the ads.txt standard in 2017 to enable publishers to publicly identify authorized advertising sellers and resellers, so that buyers can audit ads.txt files to avoid unauthorized sellers. Publisher adoption is widespread. Newer standards ads.cert, sellers.json, and the OpenRTB Supply Chain Object enable similar disclosures by other market participants. Anecdotally, it appears that these efforts have meaningfully reduced fraud borne by supply chain participants.
The other type of antifraud effort helps advertisers avoid paying for fraudulent ads ex ante and to seek reimbursement for fraud detected ex post. Most importantly, there are a range of firms that specialize in ad fraud detection, including some that partner with large digital advertising platforms to directly analyze some data. There are also industry collaborations, such as the Trustworthy Accountability Group, which publishes a monthly Data Center IP List, a "common list of [Internet Protocol] addresses with invalid traffic coming from data centers where human traffic is not expected to originate." In addition, industry-standard contractual language has been developed with the intent of helping advertisers obtain relief for fraudulent ads. Measurement difficulties make it challenging to know how advertiser-facing fraud has changed over time. Advertiser concern and awareness about fraud have increased, but at the same time, fraudsters have grown more sophisticated and discovery of large-scale botnets has increased.
The issues underlying digital ad fraud are nontransparency and nonverifiability of human recipients of ads. Advertising delivery systems seldom require proof of humanity; for example, publishers do not require users to solve a reCAPTCHA before serving ads. Ads are delivered when a computer requests a page from a web server, placing ad exposures in the set of activities that are trivial for botnets to perform programmatically. As Vinton Cerf, co-creator of the internet protocol, put it, "We didn't focus on how you could wreck this system intentionally." Digital ad fraud is often described as a cat-and-mouse game in which fraudsters develop new tactics when previous tactics have been neutralized.
We emphasize the role of botnets because evasion of detection is of central importance to those who would commit advertising fraud. Fraud that can be detected can be reversed, with charges reverting back to the client marketers and fraudsters being excluded from established advertising networks. Therefore, competent ad fraud is, by definition, fraud that is sufficiently unlikely to be detected. Major advertising networks say that they seek to detect advertising fraud, but they do not fully explain the specific techniques they use. Fraud detectors are hamstrung in gaining advertisers' trust: sharing their precise fraud detection algorithms could help unscrupulous actors avoid fraud detection. There is also a credibility problem: [127] proved that, under general conditions, digital advertising platforms' revenues may rise when they fail to detect ad fraud, even when advertisers adjust their bids based on expected fraud levels. In summary, small advertisers are forced to trust in partners' efforts to detect fraud, even though the partners get paid every time their detection algorithms fail.
The academic literature on ad fraud is relatively sparse. Only a few theoretical studies examine how advertising market conditions and business models affect various players' incentives to engage in fraud (e.g., [29]). A more sizable literature in computer science proposes algorithms to detect particular fraud schemes (e.g., [14]). Finally, some empirical papers relate fraud prevalence to other market conditions to offer guidance to firms. For example, [46] relate marketers' monitoring efforts to online affiliate fraud. They show that monitoring by outside specialists is more effective at punishing clear violations of marketer rules, whereas internal marketing staff are more effective at punishing borderline rule violations.
More research is needed, particularly the following five areas:
- Theoretical analyses have seldom considered the economic antecedents and consequences of digital ad fraud. It would be interesting to model digital advertising as a credence good as many advertisers cannot reliably measure ad effects. Two open questions are how credence goods markets function when delivery of the service is only partially verifiable and how optimal contracting terms depend on the properties of the transaction. A related question is whether there are strategies buyers or market makers can adopt independently to improve market efficiency in such settings. Model-based predictions could help to identify opportunities to improve market regulations.
- Another area of opportunity is to help platforms and fraud detection firms design better approaches to detect and reverse ad fraud. In general, the topic of how to attack machine learning algorithms and how to resist attack is a topic of active research (e.g., [125]). Consideration of adversarial methods may help firms anticipate and defend against future developments in fraud technology. It also may help identify opportunities to incentivize humans to prove their humanity in order to better measure when ads are being delivered to people as opposed to bots.
- It may be interesting to adapt market designs and regulations from other contexts. For example, securities fraud may have been affected by government regulations or dispersion of trading across competing stock markets. There may also be lessons available from other examples of fraud, such as insurance fraud or identity theft. Successful efforts to detect, prevent, reverse, or deter fraudulent activity in other settings may have implications for effective management of digital advertising fraud.
- It is important to understand empirical predictors and impacts of fraud. If institutional ethics guidelines allow, researchers could simply run experiments by purchasing various types of botnet actions, as they are available cheaply online, and then observe how fraudulent traffic changes market outcomes. Ideally, an approach might be developed that enables individual agencies or advertisers to audit platforms', publishers' and fraud detection firms' claims about audiences delivered and fraud prevented.
- Another approach could use certain events as discontinuities in quasi-experimental designs, such as privacy regulation changes, publishers' adoption dates of ads.txt or other transparency standards, or major public disclosures of botnets or ad fraud schemes. Examining how market data and indicators of fraudulent activity change around event times could show how fraud responds to incentives and distorts markets.
We close by highlighting two hypotheses related to ad fraud and allocative efficiency. First, there are trade-offs between consumer privacy, market transparency, and fraud detection. In particular, the more signals there are about human identity and activity, the more information is available to potentially distinguish humans from machines. Therefore, it seems possible that increasing consumer privacy may also hinder detection of ad fraud and market transparency. Second, we think the application of blockchain technology is promising in digital advertising markets. However, widespread adoption of new standards requires coordinated action. Will the walled gardens adopt blockchain or other antifraud technologies? As an example, Ads.txt adoption was boosted significantly when Google adopted and promoted the standard, but Google did not actively support the standard until after a significant number of publishers adopted it. To date, several blockchain-based solutions have been proposed for digital ad markets, but we are not aware of any that have achieved significant traction. New standards may need to anticipate the chicken-and-egg problem to get ad purchasers, sellers and intermediaries on board.
We have discussed each source of inefficiency in isolation, but they may interact to exacerbate market-level inefficiencies. A prominent example is when advertising agencies' private incentives depart from marketing principals' profit incentives to measure credible ad effects. This interaction could help explain the relative scarcity of experimental evidence about advertising effects and the frequent misinterpretation of observational methods as yielding causal ad effects.
Other harmful interactions between inefficiencies appear likely. Agencies may misreport the extent of ad blocking or ad fraud to their clients, especially if negative information might reduce the client's advertising budget and, consequently, the agency's commission. Another prominent example is that ad fraud or ad blocking may contribute to the statistical challenge of measuring ad effects by reducing the number of valid advertising exposures delivered to consumers in treatment and control cells of an experiment.
Several high-profile regulatory reports have recommended new regulations for digital advertising markets. Those reports have described many important topics extensively (e.g., antitrust, brand safety, disinformation, consumer privacy, market transparency). However, they have covered the four themes treated in this article in less detail. We now consider the question of how research could help inform broader policy considerations, in addition to the four individual themes treated previously.
Ultimately, digital advertising markets exist to create value for the two types of agents they aim to connect: consumers and marketers. Consumers supply attention and realize utility from advertisements, media content, and advertised products. Marketers buy ads and realize profits from product sales. All other relevant parties are intermediaries enabling interactions between consumers and marketers. Therefore, consumer welfare and producer profits should be the two primary concerns in designing government policy.
It is particularly important to understand how advertisements affect consumer search for information about products and prices. Some advertising may enhance product-market outcomes for consumers, whereas other ads may suppress product search and reduce competition in the product market. It is also important to uncover when advertising may have direct negative impacts on consumer utility, as is suggested by the substantial literature on advertising avoidance and the extensive use of ad blockers. There is existing literature on these topics, but more comprehensive and systematic evidence is needed to guide policy due to the heterogeneity among consumers, advertisers, and the variety of advertising channels and messages.
The ad effect measurement section explains some significant obstacles to profit-maximizing advertising decisions. An important topic that is not well documented is the variability in how different types of firms make different types of advertising decisions. Some advertisers are known to be more sophisticated and make more data-driven decisions than others, but we are aware of little public evidence about the financial returns to data-driven advertising decision making. Similar questions can be posed about estimating the returns to using specific new tools in the advertising process, such as automated ad content generation, programmatic advertising purchases, fraud detection, and brand safety monitoring. Another open question is how different types of firms should structure their external advertising agencies and what contracting mechanisms might be optimal in what situations. Differences in advertiser objectives, capabilities, ownership structures, and firm size may help explain why some firms make more data-driven decisions than other firms. Overall, there is substantial scope to shed more light on the antecedents and consequences of advertiser behavior.
There are not many examples of successful regulation in multisided platform industries like digital advertising markets. It is unclear how to optimally trade off conflicting goals among various players. One relevant trade-off may occur between consumers' interests, such as privacy, and advertisers' interests, such as fraud detection. Another relevant trade-off may occur between large platforms' interests in acquiring early-stage companies and potential competitors' incentives to enter and innovate.
It would also be interesting to consider what metrics regulators can use to measure the efficiency of digital advertising markets; greater economic activity in advertising markets does not necessarily indicate greater economic welfare ([119]). It seems likely that regulations that entrench incumbents may deter potential entrants, so entry and innovations might be potential indicators of well-functioning markets. It may be that ad blocking installations decrease with market performance for consumers and therefore might underpin relevant metrics. Perhaps advertiser usage of experimentation platforms, and subsequent changes in advertising spending, could indicate how digital advertising markets are serving advertisers' interests.
Finally, a question remains whether regulatory regimes will be robust to platforms' active preparations to be regulated. For example, some technology companies have been anticipating antitrust regulation for years ([ 9]). If regulatory policy is not robust to adversarial interference, it could exacerbate the problems it is meant to resolve.
Digital advertising markets offer numerous innovations over traditional advertising. Perhaps most notable is the immediacy with which a small business can target and advertise to local consumers, along with the increased variety of advertising formats, relevance, market structures, and interactivity. However, as with most technological innovations, the shift to digital advertising has produced costs as well as benefits. We have reviewed four common issues that are likely to hinder allocative efficiency in digital advertising markets. Most marketers either do not or cannot measure incremental ad effects; uncertainty about ad effects may distort market demand for advertising. Numerous intermediaries separate marketers from publishers, each of which takes a cut of advertising expenditures and has its own private information and incentives, leading to asymmetric information and moral hazard. Consumers use ad blocking software to passively prevent advertisements from being displayed, resulting in misappropriated advertising revenues and reduced incentives to provide media content. Advertising fraud misrepresents advertising opportunities and directs ad exposures to machines to steal advertising budgets.
This discussion is aimed to help academics, scientists employed by advertising companies, policy makers, and executives as they seek to understand modern digital advertising markets. We do not claim that these four issues are necessarily more important than others; we have focused on them because we believe that they are less well understood than other topics that have received more careful study. Such topics as antitrust, brand safety, disinformation, consumer privacy, and market transparency are also important policy considerations. We have not taken a position on whether or how digital advertising markets should be regulated, but we do recommend that policy makers have full information and hope this article helps them achieve this goal. Overall, we hope this survey of market inefficiencies in digital advertising markets helps develop scientific literature and better inform the policy-making process. Many opportunities remain, and much work has yet to be done.
Footnotes 1 Declaration of Conflicting Interests The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: Brett R. Gordon was previously a contractor and is currently an employee at Facebook; he donates all resulting income to charity. The authors have no other conflicts of interest to report.
2 Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
3 We focus on allocative efficiency because it is a classic and fundamental measure of market performance and is a frequent criterion economists use to recommend and evaluate government policies. We do not intend to discount other policy objectives such as consumer welfare, distributional concerns, competitiveness, or innovation.
4 Advertisers sometimes rely on other metrics to evaluate ad effects, such as "copy testing" metrics or social media engagement metrics. In display and search campaigns, some advertisers focus on the number of ad clicks, even though that measure is directly proportional to total campaign expenditure and may come at the expense of free clicks on organic search results.
5 Field experiments in marketing go back more than 30 years (e.g., in direct mail [[12]; [28]]; television advertising [[45]; [90]]).
6 Google: See [70] and https://www.thinkwithgoogle.com/intl/en-gb/marketing-resources/data-measurement/a-revolution-in-measuring-ad-effectiveness/. Facebook: See [55] and https://www.facebook.com/business/help/552097218528551?helpref=uf%5fpermalink. Microsoft: https://about.ads.microsoft.com/en-us/blog/post/july-2019/experiments-test-your-campaign-changes-with-confidence. YouTube: https://www.blog.google/products/ads/new-tools-creative-storytelling-youtube/. AdRoll: [4]. MediaMath: [26]. JD.com: [43].
7 Platform- or business-specific factors may complicate an advertiser's ability to align the conditions in an experiment with regular, "business as usual," advertising conditions. For example, an advertiser may need to set up multiple campaigns in the experimental platform to test different conditions. Each campaign may make differential use of the advertiser's account history, which could alter the campaign's performance and affect the external validity of the experiment's results.
8 See [1] and [17] for theoretical analyses of multitouch attribution.
9 There are also challenges and conflicting incentives if platforms try to estimate interactions between ads on competing platforms. For example, it would be difficult for Google to credibly offer advertisers a tool to estimate interactions between Google ads and Facebook ads (see, e.g., https://marketingland.com/where-is-google-attribution-256098).
See, for example, [40], [52], [72], [83], [87], [42], among others, for effects of TV ads on digital behaviors. [81] document spillovers from digital display advertising to search behaviors.
Skepticism is appropriate when interpreting fraud measurements. Fraud is defined by intention, which is not directly observable in ad data. Indirect measures of ad fraud trade off subjective risks of false positive observations of fraud against false negative detections of valid ad exposures. Fraud detection firms might overreport fraud to attract business, whereas ad sellers might underreport fraud to reassure clients.
The indictment further stated, "To create the illusion that real human internet users were viewing the advertisements loaded onto these fabricated websites, the defendants programmed the datacenter servers to simulate the internet activity of human internet users: browsing the internet through a fake browser, using a fake mouse to move around and scroll down a webpage, starting and stopping a video player midway, and falsely appearing to be signed into Facebook. Furthermore, the defendants...[made] it appear that the datacenter servers were residential computers belonging to individual human internet users who were subscribed to various residential internet service providers" (https://www.justice.gov/usao-edny/pr/two-international-cybercriminal-rings-dismantled-and-eight-defendants-indicted-causing).
References Abhishek Vibhanshu, Despotakis Stylianos, Ravi R. (2017), "Multi-Channel Attribution: The Blind Spot of Online Advertising," working paper, Carnegie Mellon University.
Abraham Magid. (2008), "The Off-Line Impact of Online Ads," Harvard Business Review, 86 (4), 28.
Acquisti Alessandro, John Leslie K., Loewenstein George. (2013), "What Is Privacy Worth?" The Journal of Legal Studies, 42 (2), 249–74.
AdRoll (2019), "How Incremental Lift Testing Uncovers the Truth Behind Your Results." https://www.adroll.com/blog/marketing-analytics/how-incremental-lift-testing-uncovers-the-truth-behind-your-results.
AdTrader (2018), "Why AdTrader Is Suing Google and Why Most Advertisers Might Never Get that Chance Again," https://medium.com/adtrader/why-adtrader-is-suing-google-and-why-most-advertisers-might-never-get-that-chance-again-38e688d53872.
Adweek (2019), "Results of IAB Tech Lab's Blockchain Pilot Shows How Many Digital Ads Are Seen by Humans," https://web.archive.org/web/20190213033426/https://www.adweek.com/digital/results-of-iab-tech-labs-blockchain-pilot-shows-how-many-digital-ads-are-seen-by-humans/.
Association of National Advertisers (ANA) (2017), "Programmatic: Seeing through the Financial Fog," http://www.ana.net/getfile/25070.
ANA (2018), "The Continued Rise of the In-House Agency," https://www.ana.net/miccontent/show/id/rr-2018-in-house-agency.
Athey Susan, Luca Michael, (2019), "Economists (and Economics) in Tech Companies," Journal of Economic Perspectives, 33 (1), 209–30.
Australian Competition and Consumer Commission (2019), "Digital Platforms Inquiry: Final Report," white paper.
Barajas Joel, Akella Ram, Holtan Marius, Flores Aaron. (2016), "Experimental Designs and Estimation for Online Display Advertising Attribution in Marketplaces," Marketing Science, 35 (3), 465–83.
Bawa Kapil, Shoemaker Robert W. (1987), "The Effects of a Direct Mail Coupon on Brand Choice Behavior," Journal of Marketing Research, 24 (4), 370–76.
Becker Gary S., Murphy Kevin M. (1993), "A Simple Theory of Advertising as a Good or Bad," The Quarterly Journal of Economics, 108 (4), 941–64.
Behdad Mohammad, Barone Luigi, Bennamoun Mohammed, French Tim. (2012), "Nature-Inspired Techniques in the Context of Fraud Detection," IEEE Transactions on Systems and Man, 42 (6), 1273–90.
Benes Ross. (2019), "Agency Pros Say Fraud Is Biggest Threat to Their Budgets." https://www.emarketer.com/content/agency-pros-say-fraud-is-biggest-threat-to-their-budgets.
Bergemann Dirk, Morris Stephen, (2016), "Information Design, Bayesian Persuasion, and Bayes Correlated Equilibrium," American Economic Review, 106 (5), 586–91.
Berman Ron. (2018), "Beyond the Last Touch: Attribution in Online Advertising," Marketing Science, 37 (5), 771–92.
Berman Ron, Feit Elea McDonnell. (2018), "Enhancing Power of Marketing Experiments Using Observational Data," working paper, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3140631.
Biswas Shirsho, Dube Jean-Pierre, Simonov Andrey. (2019), "Channel Positions and TV Ad Effects," unpublished manuscript, University of Chicago.
Blake Tom, Nosko Chris, Tadelis Steven, (2015), "Consumer Heterogeneity and Paid Search Effectiveness: A Large-Scale Field Experiment," Econometrica, 83 (1), 155–74.
Borden Neil H. (1964), "The Concept of the Marketing Mix," Journal of Advertising Research, 4 (2), 2–7.
Brinker Scott. (2019), "Marketing Technology Landscape Supergraphic (2019): Martech 5000 (actually 7,040)," https://web.archive.org/web/20190801223536/https://chiefmartec.com/2019/04/marketing-technology-landscape-supergraphic-2019/.
Cambria Erik, Poria Soujanya, Gelbukh Alexander, Thelwall Mike. (2017), "Sentiment Analysis Is a Big Suitcase," IEEE Intelligent Systems, 32 (6), 74–80.
Cao Xinyu, Ke T. Tony. (2019), "Cooperative Search Advertising," Marketing Science 38 (1), 44–67.
Chae Inyoung, Bruno Hernan A., Feinberg Fred M. (2019), "Wearout or Weariness? Measuring Potential Negative Consequences of Online Ad Volume and Placement on Website Visits," Journal of Marketing Research, 56 (1), 57–75.
Chalasani Prasad, Buchalter Ari, Thiagarajan Jaynth, Winston Ezra. (2017), "Counterfactual-Based Incrementality Measurement in a Digital Ad-Buying PlatformZ," arXiv preprint arXiv:1705.00634.
Chan David, Perry Michael. (2017), " Challenges and Opportunities in Media Mix Modeling," technical report, Google Inc. https://ai.google/research/pubs/pub45998.
Chapman Randall G. (1986), "Assessing the Profitability of Retailer Couponing with a Low-Cost Field Experiment," Journal of Retailing 62 (1), 19–40.
Chen Min, Jacob Varghese S., Radhakrishnan Suresh, Ryu Young U. (2015), "Can Payment-per-Click Induce Improvements in Click Fraud Identification Technologies?" Information Systems Research, 26 (4), 754–72.
Chen Yuxin, Liu Qihong. (2019), "Signaling Through Advertising When Ad Can Be Blocked," working paper, New York University.
Choi Hana, Mela Carl F., Balseiro Santiago, Leary Adam. (2019), "Online Display Advertising Markets: A Literature Review and Future Directions," working paper, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3070706.
Coviello Lorenzo, Gneezy Uri, Goette Lorenz. (2017), "A Large-Scale Field Experiment to Evaluate the Effectiveness of Paid Search Advertising," working paper, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3061698.
Crouse Megan Corinn. (2010) Business Revolution: The Ad Agency. http://pabook2.libraries.psu.edu/palitmap/AdCo.html.
Dan Avi. (2014), " Rethinking the Agency-of-Record Model," Forbes (April 15) https://www.forbes.com/sites/avidan/2014/04/15/rethinking-the-agency-of-record-model/.
Danaher Peter J., Bonfrer Andre, Dhar Sanjay. (2008), "The Effect of Competitive Advertising Interference on Sales for Packaged Goods," Journal of Marketing Research, 45 (2), 211–25.
Davies Jessica. (2018), "Ad Tech's Bid-Caching Controversy, Explained," Pixalate accessed April 2020, https://digiday.com/media/ad-techs-bid-caching-controversy-explained.
Davies Jessica. (2019), "Ghost Sites, Domain Spoofing, Fake Apps: A Guide to Knowing Your Ad Fraud," https://web.archive.org/web/20190206032634/https://digiday.com/media/ghost-sites-domain-spoofing-fake-apps-guide-knowing-ad-fraud/.
Decarolis Francesco, Rovigatti Gabriele. (2019), "From Mad Men to Maths Men: Concentration and Buyer Power in Online Advertising," working paper, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3428421.
Despotakis Stylianos, Ravi R., Srinivasan Kannan. (2017), "The Beneficial Effects of Ad Blockers," working paper, Carnegie Mellon University.
Dinner Isaac M., Heerde Harald J. Van, Neslin Scott A. (2014), "Driving Online and Offline Sales: The Cross-Channel Effects of Traditional, Online Display, and Paid Search Advertising," Journal of Marketing Research, 51 (5), 527–45.
Du Rex, Joo Mingyu, Wilbur Kenneth C. (2019), "Advertising and Brand Attitudes: Evidence from 575 Brands over Five Years," Quantitative Marketing and Economics, 17 (3), 257–323.
Du Rex, Linli Xu, Wilbur Kenneth C. (2019), "Immediate Responses of Online Brand Search and Price Search to TV Ads," Journal of Marketing, 83 (4), 81–100.
Du Ruihuan, Zhong Yu, Nair Harikesh S., Cui Bo, Shou Ryan. (2019), "Causally Driven Incremental Multi Touch Attribution Using a Recurrent Neural Network," https://arxiv.org/abs/1902.00215.
Dukes Anthony, Gal-Or Esther. (2003), "Negotiations and Exclusivity Contracts for Advertising," Marketing Science, 22 (2), 222–45.
Eastlack Joseph O., Rao Ambar G. (1989), "Advertising Experiments at the Campbell Soup Company," Marketing Science, 8 (1), 57–71.
Edelman Benjamin, Brandi Wesley. (2015), "Risk, Information and Incentives in Online Affiliate Marketing," Journal of Marketing Research, 52 (1), 1–12.
eMarketer (2019 a), "Digital Ad Fraud 2019," https://web.archive.org/web/20190824032234/%20https://www.emarketer.com/content/digital-ad-fraud-2019.
eMarketer (2019 b), "US Digital Ad Spending 2019," https://www.emarketer.com/content/us-digital-ad-spending-2019.
Feit Elea McDonnell, Berman Ron. (2019), "Test & Roll: Profit-Maximizing A/B Tests," Marketing Science, 38 (6), 1038–58.
Fitz Grant. (2018), "It's Time to Review the Advertising Budget," https://www.cfo.com/budgeting/2018/05/time-to-review-advertising-budget.
Forrester Consulting (2014), "Cross-Channel Attribution Is Needed to Drive Marketing Effectiveness," https://think.storage.googleapis.com/docs/forrester-cross-channel-attribution_research-studies.pdf.
Fossen Beth L., Schweidel David A. (2017), "Television Advertising and Online Word-of Mouth: An Empirical Investigation of Social TV Activity," Marketing Science, 36 (1), 105–23.
Fuchs William, Öry Aniko, Skrzypacz Andrzej. (2015), "Transparency and Distressed Sales Under Asymmetric Information," Theoretical Economics, 11 (3), 1103–44.
Gabzewicz Jean J., Garelli Paolo G., Sonnac Natalie. (2007), "Newspapers' Market Shares and the Theory of the Circulation Spiral," Information Economics and Policy, 19 (3–4), 405–13.
Gordon Brett R., Zettelmeyer Florian, Bhargava Neha, Chapsky Dan. (2019), "A Comparison of Approaches to Advertising Measurement: Evidence from Big Field Experiments at Facebook," Marketing Science, 38 (2), 193–225.
Gritckevich Aleksandr, Katona Zsolt, Sarvary Miklos. (2019), "Ad Blocking," working paper, Columbia University.
Ha Anthony. (2018), "German Supreme Court Dismisses Axel Springer Lawsuit, Says Ad Blocking Is Legal," https://techcrunch.com/2018/04/20/adblock-plus-v-axel-springer.
Harberger Arnold C. (1954), "Monopoly and Resource Allocation," The American Economic Review, 44 (2), 77–87.
Hartmann Wesley R., Klapper Daniel. (2018), "Super Bowl Ads," Marketing Science, 37 (1), 78–96.
Hill Daniel N., Moakler Robert, Hubbard Alan E., Tsemekhman Vadim, Provost Foster, Tsemekhman Kiril. (2015), "Measuring Causal Impact of Online Actions Via Natural Experiments: Application to Display Advertising," in 21th ACM SIGKDD International Conference. New York : Association for Computing Machinery.
Homburg Christian, Jensen Ove, Krohmer Harlem. (2008), "Configurations of Marketing and Sales: A Taxonomy," Journal of Marketing, 72 (2), 133–54.
Horsky Dan, Horsky Sharon, Zeithammer Robert. (2016), "The Modern Advertising Agency Selection Contest: A Case for Stipends to New Participants," Journal of Marketing Research, 53 (5), 773–89.
Horsky Sharon. (2006), "The Changing Architecture of Advertising Agencies," Marketing Science, 25 (4), 367–83.
House of Lords Select Committee on Communications (2019), " Regulating in a Digital World," white paper.
Hu Yu Jeffrey, Shin Jiwoong, Tang Zhulei. (2016), "Incentive Problems in Performance-Based Online Advertising Pricing: Cost per Click vs. Cost per Action," Management Science, 62 (7), 2022–38.
Huang Jason, Reiley David H., Riabov Nickolai M. (2018), "Measuring Consumer Sensitivity to Audio Advertising: A Field Experiment on Pandora Internet Radio," http://www.davidreiley.com/papers/PandoraListenerDemandCurve.pdf.
Internet Advertising Bureau (IAB) (2019), " IAB Internet Advertising Revenue Report," white paper.
Jerath Kinshuk, Tony Ke T., Long Fei. (2018), "The Logic and Management of 'Digital Co-op' in Search Advertising," working paper, Columbia University.
Johnson Garrett, Lewis Randall A. (2015), "Cost per Incremental Action: Efficient Pricing of Advertising," working paper, https://ssrn.com/abstract=2668315.
Johnson Garrett, Lewis Randall A., Nubbemeyer Elmar I. (2017 a), "Ghost Ads: Improving the Economics of Measuring Online Ad Effectiveness," Journal of Marketing Research, 54 (6), 867–84.
Johnson Garrett, Lewis Randall A., Nubbemeyer Elmar I. (2017 b), "The Online Display Ad Effectiveness Funnel & Carryover: Lessons from 432 Field Experiments," working paper, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2701578.
Joo Mingyu, Wilbur Kenneth C., Cowgill Bo, Zhu Yi. (2014), "Television Advertising and Online Search," Management Science, 60 (1), 56–73.
K2 Intelligence (2016), "An Independent Study of Media Transparency in the U.S. Advertising Industry." https://www.ana.net/content/show/id/industry-initiative-media-transparency.
Kalyanam Kirthi, McAteer John, Marek Jonathan, Lin Lifeng. (2017), "Cross Channel Effects of Search Engine Advertising on Brick & Mortar Retail Sales: Meta-Analysis of Large Scale Field Experiments on Google.com," Quantitative Marketing and Economics, 16 (6), 1–42.
Kamenica Emir, Gentzkow Matthew. (2011), "Bayesian Persuasion," American Economic Review, 126 (4), 2590–615.
Katona Zsolt, Sarvary Miklos. (2018), " Eyeo's Adblock Plus: Consumer Movement or Advertising Toll Booth?" Berkeley-Haas Case Series B5912.
Kim Alex, Balachander Subramanian. (2017), " Coordinating Traditional Media Advertising and Search Advertising," working paper, University of California, Riverside.
Laffont Jean-Jacques, Martimort David. (2001), The Theory of Incentives: The Principal-Agent Model. Princeton, NJ : Princeton University Press.
Lambrecht Anja, Tucker Catherine E. (2013), "When Does Retargeting Work? Information Specificity in Online Advertising," Journal of Marketing Research, 50 (5), 561–76.
Lewis Randall A, Reiley David H. (2014), "Online Ads and Offline Sales: Measuring the Effect of Retail Advertising via a Controlled Experiment on Yahoo!," Quantitative Marketing and Economics, 12 (3), 235–66.
Lewis Randall A., Nguyen Daniel. (2015), "Display Advertising's Competitive Spillovers to Consumer Search," Quantitative Marketing and Economics, 13 (2), 93–115.
Lewis Randall A., Rao Justin M. (2015), "The Unfavorable Economics of Measuring the Returns to Advertising," The Quarterly Journal of Economics, 130 (4), 1941–73.
Lewis Randall A., Reiley David H. (2013), " Down-to-the-Minute Effects of Super Bowl Advertising on Online Search Behavior," in EC '13 Proceedings of the fourteenth ACM Conference on Electronic Commerce. New York : Association for Computing Machinery, 639–56.
Lewis Randall A., Rao Justin M., Reiley David H. (2011), "Here, There, and Everywhere: Correlated Online Behaviors Can Lead to Overestimates of the Effects of Advertising," in Proceedings of the 20th International Conference on World Wide Web, 157–66.
Lewis Randall A., Rao Justin M., Reiley David H. (2015), "Measuring the Effects of Advertising: The Digital Frontier." in Economic Analysis of the Digital Economy, Goldfarb Avi, Greenstein Shane M., Tucker Catherine E., eds. Chicago : The University of Chicago Press, 191–218.
Lewis Randall A., Wong Jeffrey. (2018), "Incrementality Bidding and Attribution." http://causaleffect.io/lewis2018.pdf.
Liaukonyte Jura, Teixeira Thales, Wilbur Kenneth C. (2015), "Television Advertising and Online Shopping," Marketing Science, 34 (3), 311–30.
Lin Xiliang, Nair Harikesh S., Sahni Navdeep S., Waisman Caio. (2019), "Parallel Experimentation in a Competitive Advertising Marketplace," https://arxiv.org/abs/1903.11198.
Liu Liu, Dzyabura Daria, Mizik Natalie. (2018), " Visual Listening In: Extracting Brand Image Portrayed on Social Media," in The Workshops of the Thirty-Second AAAI Conference on Artificial Intelligence, 71–77.
Lodish Leonard M., Abraham Magid, Kalmenson Stuart, Livelsberger Jeanne, Lubetkin Beth, Richardson Bruce, Stevens Mary Ellen. (1995), "How TV Advertising Works: A Meta-Analysis of 389 Real World Split Cable TV Advertising Experiments," Journal of Marketing Research, 32 (2), 125–39.
Mayzlin Dina, Shin Jiwoong. (2011), "Uninformative Advertising as an Invitation to Search," Marketing Science, 30 (4), 666–85.
McKinsey & Company (2013), "How CMOs Can Get CFOs on Their Side," https://www.mckinsey.com/business-functions/marketing-and-sales/our-insights/how-cmos-can-get-cfos-on-their-side.
Naik Prasad A., Raman Kalyan. (2003), "Understanding the Impact of Synergy in Multimedia Communications," Journal of Marketing Research, 40 (4), 375–88.
Nair Harikesh S., Misra Sanjog, Hornbuckle IV William J., Mishra Ranjan, Acharya Anand. (2017), "Big Data and Marketing Analytics in Gaming: Combining Empirical Models and Field Experimentation," Marketing Science, 36 (5), 699–725.
Nanigans (2019), "TIL: Ghost Ads," https://www.nanigans.com/2019/01/25/til-ghost-ads/.
Narayanan Sridhar, Kalyanam Kirthi. (2015), "Measuring Position Effects in Search Advertising: A Regression Discontinuity Approach," Marketing Science, 34 (3), 388–407.
Neff Jack. (2015), " The CMO's Guide to Agency Procurement," AdAge, May 13. https://adage.com/article/cmo-strategy/cmo-s-guide-agency-procurement/298536.
Netzer Oded, Feldman Ronen, Goldenberg Jacob, Fresko Moshe. (2012), "Mine Your Own Business: Market-Structure Surveillance Through Text Mining," Marketing Science, 31 (3), 521–43.
Pathak Shareen. (2018), "Marketing 2019: The Year of Consolidation Arrives in Force," https://digiday.com/marketing/marketing-2019-year-consolidation-arrives-force/.
Pérez-Rosas Veronica, Mihalcea Rada, Morency Louis-Phillippe. (2013), "Utterance-Level Multimodal Sentiment Analysis," in Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, 1, 973–82.
Peterson Tim. (2017), "FAQ: Everything Facebook Has Admitted About Its Measurement Errors," MarketingLand.com, https://marketingland.com/heres-itemized-list-facebooks-measurement-errors-date-200663.
Pixalate (2018), "Ad Fraud (IVT) Benchmarks: 10%-15% of Desktop Traffic Was Invalid in Q2 2018," https://web.archive.org/web/20190207193901/https://blog.pixalate.com/ad-fraud-ivt-benchmarks-desktop-q2-2018.
Poria Soujanya, Cambria Erik, Hazarika Devamanyu, Majumder Navonil, Zadeh Amir, Morency Louis-Phillippe. (2017), "Context-Dependent Sentiment Analysis in User-Generated Videos," in Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, 1, 873–83.
Rafieian Omid, Yoganarasimhan Hema, (2019), "Targeting and Privacy in Mobile Advertising," Marketing Science, https://www.msi.org/reports/targeting-and-privacy-in-mobile-advertising/.
Rao Justin M., Simonov Andrey. (2019), "Firms' Reactions to Public Information on Business Practices: The Case of Search Advertising," Quantitative Marketing and Economics, 17 (2), 105–34.
Sahni Navdeep N. (2015), "Effect of Temporal Spacing Between Advertising Exposures: Evidence from an Online Field Experiment," Quantitative Marketing and Economics, 13 (3), 203–47.
Sahni Navdeep N., Narayanan Sridhar, Kalyanam Kirthi. (2019), "An Experimental Investigation of the Effects of Retargeted Advertising: The Role of Frequency and Timing," Journal of Marketing Research, 56 (3), 401–18.
Schmidt Susanne, Eisend Martin. (2015), "Advertising Repetition: A Meta-Analysis on Effective Frequency in Advertising," Journal of Advertising Research, 44 (4), 415–28.
Schwartz Eric M., Bradlow Eric T., Fader Peter S. (2017), "Customer Acquisition via Display Advertising Using Multi-Armed Bandit Experiments," Marketing Science, 36 (4), 500–22.
Searls David. (2019), "Is Ad Blocking Past 2 Billion Worldwide?" http://blogs.harvard.edu/doc/2019/03/23/2billion/.
Shapiro Bradley. (2018), "Positive Spillovers and Free Riding in Advertising of Prescription Pharmaceuticals: The Case of Antidepressants," Journal of Political Economy, 126 (1), 381–437.
Shiller Benjamin, Waldfogel Joel, Ryan Johnny. (2018), "The Effect of Ad Blocking on Website Traffic and Quality," RAND Journal of Economics, 49 (1), 43–63.
Simonov Andrey, Nosko Chris, Rao Justin M. (2018), "Competition and Crowd-Out for Brand Keywords in Sponsored Search," Marketing Science, 37 (2), 200–15.
Simonov Andrey, Rao Justin M. (2019), "Firms' Reactions to Public Information on Business Practices: Case of Search Advertising," Quantitative Marketing and Economics, 17 (2), 105–34.
Sinkinson Michael, Starc Amanda. (2019), "Ask Your Doctor? Direct-to-Consumer Advertising of Pharmaceuticals," Review of Economic Studies, 86 (2), 836–81.
Sluis Sarah. (2019), "Marketers Are Going Straight to Exchanges for Ultimate Data Transparency," https://adexchanger.com/data-exchanges/marketers-are-going-straight-to-exchanges-for-ultimate-data-transparency/#close-olyticsmodal.
Smith Timothy M., Gopalakrishna Srinath, Chatterjee Rabikar. (2006), "A Three-Stage Model of Integrated Marketing Communications at the Marketing–Sales Interface," Journal of Marketing Research, 43 (4), 564–79.
Soleymani Mohammad, Garcia David, Jou Brendan, Schuller Björn, Chang Shih-Fu, Pantic Maja. (2017), "A Survey of Multimodal Sentiment Analysis," Image and Vision Computing, 65, 3–14.
Stigler Committee on Digital Platforms (2019), "Final Report," https://www.judiciary.senate.gov_imo_media_doc_market-2Dstructure-2Dreport-2520-2D15-2Dmay-2D2019.pdf.
Swant Marty. (2019), "Results of IAB Tech Lab's Blockchain Pilot Shows How Many Digital Ads Are Seen by Humans," AdWeek (February 11), https://www.adweek.com/digital/results-of-iab-tech-labs-blockchain-pilot-shows-how-many-digital-ads-are-seen-by-humans/.
Sweeney Erica. (2018), "Study: 21% of Marketers Pull Back Ad Spend due to Poor Digital Measurement," https://web.archive.org/web/20190730133601/https://www.marketingdive.com/news/study-21-of-marketers-pull-back-ad-spend-due-to-poor-digital-measurement/522199/.
Tuchman Anna E., Nair Harikesh S., Gardete Pedro M. (2018), "Television Ad Skipping, Consumption Complementarities and the Consumer Demand for Advertising," Quantitative Marketing and Economics, 16 (2), 111–74.
Vaver Jon, Koehler Jim. (2011), " Measuring ad Effectiveness Using Geo Experiments," technical report, Google Inc.
Villas-Boas J. Miguel. (1994), "Sleeping with the Enemy: Should Competitors Share the Same Advertising Agency?" Marketing Science, 13 (2), 190–202.
Wang Xianmin, Li Jing, Kuang Xiaohui, Tan Yu-an, Li Jin. (2019), "The Security of Machine Learning in an Adversarial Setting: A Survey," Journal of Parallel and Distributed Computing, 130 (August), 12–23.
Waisman Caio, Nair Harikesh S., Carrion Carlos, Xu Nan. (2019), "Online Inference for Advertising Auctions," https://arxiv.org/abs/1908.08600.
Wilbur Kenneth C., Zhu Yi. (2009), "Click Fraud," Marketing Science, 28 (2), 293–308.
Winegar Angela G., Sunstein Cass R. (2019), "How Much Is Data Privacy Worth? A Preliminary Investigation," Journal of Consumer Policy, 42, 425–40.
Zantedeschi Daniel, Feit Elea McDonnell, Bradlow Eric T. (2017), "Measuring Multichannel Advertising Response," Management Science, 63 (8), 2706–28.
Zhong Zemin. (2018), "A Model of Search Design," working paper, University of Toronto.
Zhu Shitong, Hu Xunchao, Qian Zhiyun, Shafiq Zubair, Yin Heng. (2018), "Measuring and Disrupting Anti-Adblockers Using Differential Execution Analysis," in Network and Distributed Systems Security (NDSS) Symposium, February 18–21, San Diego, CA.
~~~~~~~~
By Brett R. Gordon; Kinshuk Jerath; Zsolt Katona; Sridhar Narayanan; Jiwoong Shin and Kenneth C. Wilbur
Reported by Author; Author; Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 73- Influencer Marketing Effectiveness. By: Leung, Fine F.; Gu, Flora F.; Li, Yiwei; Zhang, Jonathan Z.; Palmatier, Robert W. Journal of Marketing. Jul2022, p1. DOI: 10.1177/00222429221102889.
Ahead of Print- Database:
- Business Source Complete
Record: 74- Informational Challenges in Omnichannel Marketing: Remedies and Future Research. By: Cui, Tony Haitao; Ghose, Anindya; Halaburda, Hanna; Iyengar, Raghuram; Pauwels, Koen; Sriram, S.; Tucker, Catherine; Venkataraman, Sriraman. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p103-120. 18p. 3 Charts. DOI: 10.1177/0022242920968810.
- Database:
- Business Source Complete
Informational Challenges in Omnichannel Marketing: Remedies and Future Research
Omnichannel marketing is often viewed as the panacea for one-to-one marketing, but this strategic path is mired with obstacles. This article investigates three challenges in realizing the full potential of omnichannel marketing: ( 1) data access and integration, ( 2) marketing attribution, and ( 3) consumer privacy protection. While these challenges predate omnichannel marketing, they are exacerbated in a digital omnichannel environment. This article argues that advances in machine learning and blockchain offer some promising solutions. In turn, these technologies present new challenges and opportunities for firms, which warrant further academic research. The authors identify both recent developments in practice and promising avenues for future research.
Keywords: attribution; blockchain; machine learning; omnichannel; privacy
Despite the prevalence of new advertising and promotional channels and significant investments in data and technology, marketers are still struggling to generate and to prove sales results in an increasingly omnichannel world.
—Eric Solomon, Senior Vice President, Nielsen ([62])
Channels have traditionally been viewed as intermediaries that facilitate distribution and transfer of products from manufacturers to their customers.[ 3] Prior to the commercialization of the internet and subsequent digitization innovations, firms usually employed one type of channel such as a physical store, a call center, or a catalog. However, there were also instances where firms employed multiple channels to serve their customers. For example, firms such as L.L. Bean, Sears, and Lands' End sold their products in brick-and-mortar stores, in catalogs, and by phone. This practice gave birth to the idea of multichannel marketing. Subsequently, the idea of multichannel marketing moved beyond product fulfillment to include a whole gamut of interactions between a firm and its customers. [60], p. 96) define multichannel marketing as the "design, deployment, coordination, and evaluation of the channels to enhance customer value through effective customer acquisition, retention, and development." Therefore, in a multichannel context, although customers may interact with the firm across multiple channels before a conversion occurs, the firm's focus is on managing and optimizing the performance of each channel separately.
The presence of multiple channels can alter how customers gather product information (e.g., [ 5]; [79]) and where they purchase these products ([64]). In addition, a portfolio of channels allows customers to self-select into their preferred channel at each stage of the purchase journey ([10]; [82]), thereby allowing the firm to access a larger base of customers. Furthermore, when an online retailer expands into offline channels, the firm may also see some benefits of complementarity (e.g., [ 8]; [49]). As a result, operating additional channels might result in customers increasing their purchases ([48]).
With continuing growth in digitization, consumers today interact with firms across online, mobile, and offline media channels. This, in turn, has led to a shift toward "omnichannel" marketing, which emphasizes a unified consumer experience rather than just facilitating transactions. Furthermore, as [74] study indicates, the growing popularity of omnichannel marketing has been fueled by the idea that the different stages of the customer journey can be decoupled and delivered by various entities. In effect, for firms, omnichannel marketing entails managing a combination of different types of channels such that they align well with the way their customers search, purchase, and consume their products and share those experiences ([ 3]).
[81], p. 176) define omnichannel as the "synergetic management of the numerous available channels and customer touchpoints, in such a way that the customer experience across channels and the performance over channels is optimized." In the ideal scenario, customers interact seamlessly with the firm across channels both internal and external to the firm, and the firm has full information on all customer touchpoints to provide a single unified experience across channels.
However, this ideal faces several important hurdles in reality. As retailers adopt omnichannel marketing, it presents its own set of challenges and opportunities for the suppliers and other distribution channel partners. [ 3], p. 120) note that omnichannel marketing "often encompasses not just the channels of distribution through which a supplier's products reach the consumer but also the channels of communication—owned, paid, and earned."
As we see it, this important observation made in [ 3] does not fit within the scope of [81] definition of omnichannel marketing. We broaden the scope of previous definitions and define omnichannel marketing as the synergistic management of all customer touchpoints and channels both internal and external to the firm to ensure that the customer experience across channels as well as firm-side marketing activity, including marketing-mix and marketing communication (owned, paid, and earned), is optimized for both firms and their customers. Thus, whereas [81] emphasize experience over transactions and [ 3] emphasize communications over sales, our view of omnichannel marketing considers sales, experience, and communications. Note that the synergistic management of touchpoints and experiences might affect important outcomes for firms, such as market share, profits, and customer lifetime value ([ 6]). The exact objective function is likely to vary across firms and its and customers' life cycle.
Given its promise, it is not surprising that firms have invested heavily in omnichannel marketing. The transformation to omnichannel marketing has gained prominence in a wide range of industries, including consumer packaged goods such as Unilever, fashion retailers such as Bonobos, service providers such as Bank of America, restaurants such as Starbucks, and pharmacies such as Walgreens. However, firms also need to consider the cost of implementing customer integration (for details, see [20] and [34]). In the end, firms have to assess whether additional costs are commensurate with the expected benefits of undertaking omnichannel marketing. Our treatment of omnichannel marketing in this article focuses more on the customer side and the ensuing impact on revenues rather than on the supply-side costs that firms may incur in achieving such integration.
Despite the promise of omnichannel marketing to manage how firms interact with their customers to drive growth, foster innovation, and improve long-term performance, we posit that this potential has not been fully realized. In our view, there are three main interrelated challenges that have prevented omnichannel marketing from realizing its full potential:
- Data Challenges: To fully realize the potential of omnichannel marketing, firms need information on all their interactions with each customer during the different stages of the customer journey. We include consideration of the communications between the firm and its customers, activities where the customers interact with the firm (or its partners) while gathering information, making a purchase, receiving the product, making a return, and receiving postpurchase service. Such data might not be readily available or easily usable.
- Marketing Attribution Challenges: For optimizing the customer experience across all channels, firms need to know the impact of various touchpoints on behavior and measure the return on investment of its marketing spend. In our opening quote from Eric Solomon, this is captured as "prove sales results." Such analysis may be challenging when the effect of a touchpoint can transcend multiple stages in the purchase funnel, when several occur concurrently, or when consumers go back and forth between different stages in their path-to-purchase journey.
- Customer Privacy Challenges: The promise of omnichannel marketing relies on using data on all the interactions between the firm and its customers. However, this can come at the cost of infringing on customer privacy. Therefore, an important challenge for a firm is determining how to embrace an omnichannel strategy while respecting consumers' privacy.
Each section of this article elaborates on these challenges and discusses recent attempts to address them. We then propose promising avenues for future research in these areas.
Firms such as REI carefully plan for their customer experience to be unified across all touchpoints. While REI has a large physical footprint, it is mobile-centric and encourages its customers to use the app. For instance, if a customer clicks on a product in an email from REI and installs the mobile application, the app will note which nearest store has the product in stock. In addition, when customers visit a store, they are strongly encouraged to join the store Wi-Fi, log into the app, and check product availability. Disney and Bank of America are examples of other companies that have carefully integrated the customer experience across different channels ([27]).
One of the main challenges that a firm might face in realizing the full potential of omnichannel marketing pertains to availability and usability of such data from various touchpoints. We can broadly classify such data-related challenges along two key dimensions: ( 1) gaining access to these data and ( 2) integrating these data from different sources. We elaborate on these points in the following subsections. The first column in Table 1 summarizes the key issues in each of these two dimensions.
Graph
Table 1. Data-Related Challenges, Remedies, and Future Research.
| Data Challenges | Data Remedies | Future Data Research |
|---|
| Gain Data Access | |
| 1. Within the firm, information on various contact points by the same customer resides in silos. | 1. Deploy federated learning to construct joint machine-learning model while keeping parties' servers' training data private. | 1. Which machine learning methods are optimal and generalizable to impute missing pieces of information? |
| 2. Many customer touchpoints not owned by firm. | 2. Track customers on third parties: walled garden platforms, legacy media agencies, or syndicated providers. | 2. What is the optimal means to collate information from different parties spanning different touchpoints? |
| Aggregate Data Across Sources |
| 3. Different databases use different rules, data formats, and reporting standards. | 3. Deploy probabilistic tracking when information from different databases cannot be clearly combined. | 3. What is the impact of data sharing and probabilistic tracking on consumers (price), firms, and policy makers (welfare)? |
| 4. Data sources differ in reliability. | 4. Use permissioned blockchains to allow firms to control who can see data and validate transactions. | 4a. How can firms incentivize internal and external partners to participate in blockchains?4b. Do blockchain-enabled omnichannel marketing efforts increase or soften competition? |
As noted previously, in omnichannel marketing, firms interact with their customers at multiple touchpoints, some within the firm and some beyond it. Within the firm, often, information on various contact points by the same customer resides in silos. As a result, a given unit might not even know what data are being collected by other units. For example, a firm's e-commerce platform team may not know what information about a customer exists in other divisions within the firm, and vice versa. Thus, the first bottleneck for effective omnichannel marketing is knowing what kind of data exist on the same customer within the firm ([87]). The extent to which a firm is siloed depends on how it approaches the role of data-driven marketing. In some organizations, the role is centralized within a large data science team. In others, the individuals are spread out among smaller units that might specialize in that area.
Beyond the firm, the problem is compounded. For example, many of the touchpoints for a consumer interested in an automobile are not controlled by the manufacturer, which might use paid, owned, and earned media to engage with customers; provide product information; and possibly entice them to visit the distribution channel (i.e., its local dealership). Subsequent interactions such as test drives and price negotiations occur at these dealerships. However, neither the manufacturer nor the retailer has a complete view of the multiple interactions; worse, they may not even know whether such interactions occurred. Thus, even if a firm is efficient in cataloging what data exist on a customer in each silo of the firm, it may not know what data exist on the same customer beyond the firm.
When a firm is aware of all the data that exist on a customer within (and even outside) the firm, the second challenge is the right to use it ([83]). One of the reasons for this bottleneck is that complicated administrative procedures can make data sharing between different departments with the same company very difficult, if not impossible. For example, in financial companies, one set of investments being made by customers may not be reported to other parts of the company. In addition, in industries such as health care and finance, regulations might impose restrictions on sharing of data across units. For example, [57] showcase the presence of data silos in the context of health care. They find that even within a hospital system, there is evidence of incomplete sharing of patient and clinical data.
Even if firms can surmount the challenges related to awareness of and access to data, managers still need to integrate the data to produce insights. There are two main problems that can arise with such integration. First, because each touchpoint with the customer may be managed by different entities (both within and outside the firm), they may be stored in different databases, using different rules, data formats, and reporting standards. As a result, it can be extremely challenging to match data on the same customer across different touchpoints ([61]; [73]).
The second problem is that data from diverse sources may differ in terms of their reliability. For example, the sales department within a firm might have accurate information on the various interactions it had with the customer. However, the information on the other interactions assembled by the marketing department might be less accurate, perhaps because its data are more aggregated and/or acquired from third-party vendors with their own rules and market definitions that may not overlap completely with those used by the firm. Similarly, data on some interactions might be missing some key information, which could arise, for example, from a firm's internal infrastructural limitations. For instance, a firm's interactions with its customers' via its call center/customer support channel often requires manual entry of the details of customers' inquiries, which makes it prone to transcription errors. This is in contrast to sales transactions channels, where state-of-the-art point-of-sale information technology systems reliably automate the process of obtaining reliable data on customers' purchase history and product returns.
As noted previously, gaining access to data on different customer touchpoints can be difficult even if such data reside within the same organization. In such settings, is it possible to fuse customer data together without having to transport them across various departments within an organization?
In the past few years, developments in artificial intelligence (AI) have addressed this problem. One such example is federated learning. Unlike standard machine-learning practice, in which the training data sit on one machine or in a data center, federated learning enables multiple parties to use data from multiple decentralized data servers to collaboratively construct a machine-learning model while keeping their respective servers' training data private ([45]). Over the course of several training iterations, the shared models get exposed to a significantly wider range of data than any single organization or department possesses in-house. Such an approach would be valuable in situations where regulations, such as those in the context of health care, preclude business units within a firm from sharing data.
Additional challenges are introduced when moving from situations where data reside within a company to those where outside entities own some of the customer information. This warrants reconsidering the boundary of the firm. Firms can form strategic partnerships or engage in acquisitions to ensure access to data. There are two broad situations where such partnerships have proved to be fruitful. The first situation involves tracking known customers on the so-called third-party "walled garden" platforms (Google, Facebook, and Amazon). Platforms such as Facebook and Google now allow firms to import their own "first-party" data, such as lists of email addresses or phone numbers. This can help firms identify consumers with whom they have previously had contact. Similarly, e-commerce platforms such as Amazon's "Amazon Publisher Services" enable a firm to understand how its customers engage on Amazon across products. Another example of a successful data partnership is the acquisitions of large data brokers by the legacy media agencies. In particular, the acquisitions of Epsilon by Publicis and Acxiom by IPG are two prominent mergers and acquisitions that have the potential to enable highly personalized omnichannel customer experiences when data from the data brokers are combined with the vast scale and breadth of complementary agency services. That said, the recent decisions by Google and Apple to stop supporting open-source identifiers such as third-party cookies and the identifier for advertisers can erode some of the benefits from these remedies.
The second situation pertains to tracking known customers and prospects across the open web. There have been some positive developments wherein third-party data providers enable retailers to track consumers' engagement with ad platforms such as Amazon, Apple, Facebook, Google, Verizon, and Walmart, among others. For instance, data brokers such as Experian, Acxiom, and LiveRamp have allowed firms to match information such as email addresses or cookies with other data sets, such as spending and demographic information. These examples point to the growing set of choices available for marketers and advertisers of all sizes to access and integrate customer data from different sources to successfully execute their omnichannel marketing campaigns.
An additional challenge is that even if firms can access data from several sources, they may face instances where some of the information is missing. New advancements in AI and novel predictive algorithms offer promising avenues for addressing these challenges. For example, in online purchases, product returns are a serious threat to the profitability of manufacturers and retailers, especially in the case of experience goods such as clothing. [25] have recently developed a machine-learning-based approach to predict the probability that an item will be returned. In a similar vein, many companies are now monitoring the use of products and enhanced product fulfillment even before the customer shows a need. For instance, Amazon has patented "anticipatory" shipping to cut down delivery times by predicting what buyers are going to buy before they buy it. This trend of using predictive models to forecast customer behavior might enable AI-powered companies to ship products to consumers before they are ordered ([ 2]). While these algorithms have been developed to predict purchase and consumption behavior to curate products and content, they can also be used to identify missing pieces of information in the data. For example, if a firm observes purchase information, but not the consumption or product return information, the predictive power of such algorithms can be used to fill these data voids.
There are two main ways that firms currently track consumers across devices and media that the firm controls. The first is deterministic tracking, which occurs when the firm can identify a consumer from multiple databases. For example, a subscriber of The New York Times would log in to both the website and the app using the same email login, allowing for perfect identification of the same user.
However, it is common for firms to encounter situations where they cannot match customers across different databases. For example, a website that did not have a subscription model and did not require a login would not be able to easily track whether it was the same consumer visiting its desktop website, mobile website, or application. As cookie deletion becomes more prevalent, it will become increasingly difficult to track the same consumer returning to the website. Under such situations, probabilistic tracking is a promising approach to identify consumers as they browse across different devices. As the name suggests, probabilistic matching allows firms to use algorithms to probabilistically identify and track the same user across multiple touchpoints. Drawbridge, which was recently acquired by LinkedIn last year, is an example of a firm that uses probabilistic tracking. To implement probabilistic tracking, marketers have the option of deploying machine-learning models trained on user location data, triangulated from multiple devices. This would enable them to identify the best model for probabilistic matching.
A novel set of technologies that have the potential to help track customer data and its integrity are blockchain technologies, such as those inspired by smart contracts and shared tamper-evident ledgers. Blockchain-based solutions offer a way to coordinate among different entities in the supply chain (e.g., different sources within a channel or even different channels per se).[ 4] A key feature of blockchain solutions to this challenge is an attempt to bring all the data into one protected location. If the standards are enforced when the data are entered, a well-designed blockchain system can provide data integrity as well. The data recorded in a blockchain may easily be made accessible to the participants.
Blockchain technologies have been developed mostly in response to the success and popularity of Bitcoin, in which all transactions are stored in a blockchain. Bitcoin's novelty was in creating a reliable digital currency system without any need for a centralized trusted party who would protect against copying of digital assets ([37]). This is an example of a permissionless blockchain, as it operates without any gatekeepers—and thus, the number and identity of the participants is not known. A central feature of this type of blockchain is a shared ledger, which is reconciled among the participants via a consensus mechanism ([36]). In contrast, permissioned blockchains allow firms to control who can see their data and validate the transactions ([36]). From a firm's point of view, the key advantages of using a permissioned blockchain as opposed to a more regular means of storing data is that blockchain offers more data integrity, because by the nature of shared ledger, there cannot be discrepancy when two users see the same piece of data.
Permissioned blockchains require some asymmetry in authority because there must be a trusted party or consortium to give permissions to access the system.[ 5] The level of involvement of the trusted party in maintaining the records would depend on the structure of the system. The trusted parties may be either a private company or a government agency. It is important to note that while permissionless blockchains can be slow and expensive, permissioned blockchains are much faster and cheaper. In the world of digital ads, Lucidity is such a player, constructing and running a permissioned blockchain and controlling access to it. It is a trusted party in a similar way that Google is a trusted party in running keyword auctions.
Participants may be punished for "misbehavior" outside of the blockchain (e.g., with fines, access restrictions) and their permission to participate revoked. While there is still a need for a method to reach agreement between the participants, there is no need for such demanding consensus systems as with permissionless systems. However, it is important to emphasize that permissioned blockchains can also be viewed as a more efficiently run distributed database, rather than a distinctly different way of managing data. A distributed database is a database where multiple parties can make an entry (e.g., Google Docs, Dropbox). Here, the "multiple parties" are the parties representing different channels. From a firm's point of view, the key advantages of using a permissioned blockchain as opposed to a more regular means of storing data is that blockchain offers more data integrity, because by the nature of shared ledger, there cannot be discrepancy when two users see the same piece of data.
There are several advantages for storing data and safeguarding their integrity that result from adopting a blockchain. Blockchain-based systems can help with standardization and unification of data, leading to better data integrity in digital supply chains, such as in the adtech and martech world ([30]; [33]). The current opaque and fragmented adtech supply chain does not allow for seamless cross-validation of ad campaign data from the different entities in the ecosystem that sit between the brand and the publisher, such as the demand-side platform, supply-side platform, ad exchanges and data management platform, that would ascertain the veracity of the data. One problem omnichannel advertisers often face is the reconciliation of a transaction in a given ad campaign when mapping it from a brand to a publisher—ensuring that the raw campaign data for a given transaction is the same across the different entities (e.g., the demand-side platform, ad exchange, and the supply-side platform) in the adtech supply chain ([33]). A blockchain-related solution could allow for proper ad engagement tracking that will lead to more precise digital attribution. Higher data quality achieved through transparency and unification of data streams from the different entities in the adtech ecosystem will allow firms not only to track delivered messages but also to set up smart contracts to automatically execute intricate programmatic advertising strategies and eliminate redundancy and irrelevance, to the benefit of both the advertiser and the customer. With data standardization and integration across different parts of the adtech supply chain, marketing messages in an omnichannel environment can be delivered consistently and data can be verified.[ 6]
The adoption of blockchain-based data management systems can affect how customer data are combined and integrated in many other areas as well. Omnichannel marketers typically have a complex supply chain consisting of physical stores, home delivery, online browsing, and online commerce, all of which comprise a complex network of data points on different systems and in different entities. Despite the advances made, in today's world, retail agreements are largely manual and based on proprietary systems. To get integrated views of the inventory and the customer, this complex world of data and transactions needs to be merged. For example, if a retailer pilots a blockchain solution to trace the cotton being used for a line of T-shirts, its internal system needs to be able to communicate with its cotton suppliers' and contract manufacturers' systems with a high degree of automation and accuracy to enable full end-to-end supply chain visibility.
In this context, blockchain-related systems offer several business benefits for retailers and their partners in the supply chain, both upstream and downstream, as they gather information from multiple channels in one system, inducing standardization and unification of data.[ 7] With transparent, real-time data access enabled by a shared database, retailers will know where their stock is at any point in time in that complex supply chain and where their customers interact with them at any touchpoint in that path to purchase. This real-time knowledge can lead to a faster, more transparent, and end-to-end integrated supply chain. Although the database is shared, it is not visible in its entirety by all players, thereby mitigating any privacy concerns.
Finally, the smart contracting feature of blockchains—due to automated execution of agreements—can drastically reduce the transaction costs within supply chains, thereby potentially lowering the cost of goods sold.[ 8][39] highlight that blockchains could allow firms to use "micropayments to motivate consumers to share personal information—directly, without going through an intermediary." Such forms of micropayments could significantly negate the need for firms to pay third parties such as Google or Facebook to share customer information, as is currently undertaken by omnichannel firms. The extent to which this will enhance customer welfare will depend on the degree to which firms can use this information to provide the most relevant products or services for consumers. In summary, the increased integrity of the data resulting from standardization and unification through blockchain-related solutions also brings an indirect benefit by supplying both higher-quality data for advanced data analysis and predictive analytics about customers.
While many of the advancements discussed in the previous subsection have significantly improved firms' ability to acquire and utilize disparate data to have a unified view of a customer/prospect, they also present an interesting set of challenges and opportunities for future research. First, building on the work of [25], how can one decide which machine-learning methods may be best and are generalizable to impute missing pieces of information using data already available to the firm? One challenge with typical imputation algorithms is that they are context-specific. For instance, [16] model the incomplete information problem faced by credit card companies by using the interpurchase time distributions. While the model works well for a credit card application, its use may be limited for other applications where interpurchase times are less regular. Developing a more general approach that accommodates situations that do not occur periodically is a promising opportunity for future research.
Second, to aggregate and manage data from different firms and/or units within a firm that track different customer touchpoints, it might be useful to have matchmakers who can deliver that function. Firms such as A.C. Nielsen have been successful in delivering this for a part of the customer journey. However, increasing the scope of such data collection efforts would require significant changes in how these data integration platforms are designed. In this regard, future research could discuss the optimal design of matchmakers/platforms that will collate information from different parties spanning different customer touchpoints.
Third, what is the impact of data sharing within and across firms on consumers (e.g., prices they pay), firms (e.g., supply-chain efficiency, profit margins), and policy makers (e.g., market structure, efficiency, overall surplus)? [15] suggest that the answer might depend on the precision of customer-level information. They model two firms that each have their own set of loyal (price-insensitive) customers and are competing with prices for switchers. Each firm can classify its own loyal customers and switchers correctly with some probability (this is the imprecision in targeting). The key insight from their study is that while individual marketing is feasible but imprecise, improvements in targetability can be a win-win for competitors. The intuition behind this result is that when a firm becomes better at distinguishing its loyal customers from the switchers, it is motivated to charge a higher price to the former group. However, targeting is imperfect. Therefore, firms can make mistakes such as classifying price-sensitive switchers as price-insensitive loyal customers and charging them a higher price. These mistakes allow the competitor to acquire the mistargeted customers without lowering prices and, thus, reduce the rival firm's incentive to cut prices. Therefore, the study reveals that firms may be better off sharing information with their competitors. However, the kinds of incentives that will facilitate data sharing are still unclear. In this regard, it would be worthwhile to explore what kinds of mechanisms should be put in place to incentivize firms to share data with their up- and downstream partners as well as with their competitors.
Fourth, if one were to deploy blockchains, how could one incentivize internal and external partners to participate in the blockchains? The existing commercial success stories typically rely on the strength of large players—for example, Walmart uses its bargaining power to force all its suppliers to use its blockchain. For such an incentive design problem, one needs to measure and quantify the economic benefits enabled by blockchain technology in interorganizational environments. These benefits include the decentralized management of digital assets, the algorithmic enforcement of agreements in the form of software programs, and the verification of data records in an adversarial environment. These benefits can incentivize internal and external partners to work collaboratively on the development and deployment of different blockchain-based solutions for their interorganizational environments. Certain applications of blockchain technology such as smart contracts could significantly influence the level of challenges and transaction costs between upstream and downstream partners within a supply chain. Smart contracts can also be adopted to reduce routine processes to a set of articulated conditions and facilitate frictionless execution. Research should consider whether these actions would mean that blockchain can have a measurable impact on transaction costs, firm boundaries, and interfirm governance.
Fifth, a blockchain's decentralized consensus feature can eliminate information asymmetry as a barrier to entry and facilitate greater competition ([19]). Increased competition can, in turn, enhance welfare and consumer surplus. However, decentralized consensus affords greater information transparency, which, in turn, can foster tacit collusion. Tacit collusion can, in turn, result in higher prices and erode consumer surplus. Consequently, would blockchain-enabled omnichannel marketing efforts result in increasing or softening competition?
Unlike multichannel marketing, where marketing investments are optimized on a channel-by-channel basis, in an omnichannel setting, such optimization needs to be done jointly across all distribution and communication channels ([89]). This becomes challenging when the purchase funnel has many stages and/or is traversed by customers in a nonsequential manner, as is often the case in the digital economy. That is, a customer might begin their search process in a brick-and-mortar store, form an initial consideration set, and then at some point in the near future restart their search process on a website leading up to a new consideration set and eventually make a purchase.
Before omnichannel marketers can optimize their marketing efforts across various customer touchpoints, they need to understand the effectiveness and role of each touchpoint in the consumer decision journey and its incremental role on the overall sales conversion ([43]). Attribution is more complicated in an omnichannel setting because consumers self-select into different channels, and part of the difference in response to marketing interventions might be a result of such self-selection ([58]). As a result, inferring the causal effect of interventions, which is essential for attribution, might be difficult or probably even impossible. The potential number of communication paths is incredibly large, and there is no way to have sufficient causal variation. Not surprisingly, the Marketing Science Institute (MSI) has consistently highlighted attribution as the number-one priority in its research priorities since 2016.
Attribution-related bottlenecks in omnichannel marketing stem from three key sources. First, a touchpoint in the customer journey might have an effect on multiple subsequent stages in the purchase funnel. Even if each marketing intervention can be uniquely linked to a transition from one stage in the purchase funnel to the next, it might not be appropriate to view the effect of the intervention as being restricted within the boundaries of a stage in the purchase funnel. For example, if search advertising resulted in a customer clicking on it and arriving at a firm's website, should it be given credit only for reaching the website or also for all subsequent on-site activities, including purchase, either in the same session or at a later point in time? There are two potential implications of this challenge. One implication pertains to the contract between the advertising platforms (and/or publishers) and the advertiser. The price that the advertiser is charged (and/or should be willing to pay) needs to reflect the downstream impact of the exposure. This issue is not specific to the context of omnichannel marketing. A second implication, which is more relevant in the context of omnichannel marketing, regards the appropriate allocation of resources across different touchpoints. For instance, the impact of a marketing intervention in one channel at an early stage in the purchase funnel might interact with the impact of another intervention in a different channel, possibly at a subsequent stage.
Second, consumers may be interacting with the firm via multiple touchpoints simultaneously. For example, there is ample evidence that people frequently consume several media at the same time (see [22]; [50]; [51]; [77]). Multihoming in digital platforms is a well-documented phenomenon. In such settings, marketing efforts are likely to be concurrently directed at the consumer across different channels ([31]; [32]; [59]; [69]). Under such a scenario, the challenge is to apportion credit among different omnichannel marketing activities for a conversion. As noted previously, this requires firms to reconsider the design of contracts as well as the appropriate allocation of resources across different touchpoints.
Third, many attribution methods are largely focused on quantifying which touchpoint gets credit when a purchase happens. However, if a purchase does not happen, which touchpoint(s) needs to be held accountable? The question of what is ineffective as a marketing touchpoint should take priority in a firm's marketing measurement approach, as that is an appropriate place to start the conversation around reallocation of marketing budgets from one channel to another. This can become more problematic if that touchpoint's failure to drive purchase also led other touchpoints to fail. For example, if a customer had a poor retail store experience, it might lead them subsequently to decide against buying products on the firm's mobile app; however, identifying that chain of causality can be challenging. A related problem arises when a firm uses only a subset of potential touchpoints. Under such a scenario, the effectiveness of unused touchpoints cannot be assessed. Together, these two scenarios highlight some key limitations of the traditional multitouch attribution (MTA) approaches.
Fourth, another challenge with attribution is when the data belonging to different stages of the purchase funnel are aggregated at different levels. For example, television advertising investments may be available only at the market level, while search information may be available at the individual level ([42]; [46]). Therefore, although we can infer whether an individual customer was exposed to search advertising, we may not have equivalent information for television advertising. Consequently, we can potentially relate actions by individual customers to their search behavior, but not for television advertising. The first column in Table 2 summarizes the key issues related to each of these challenges.
Graph
Table 2. Attribution-Related Challenges, Remedies, and Future Research.
| Attribution Challenges | Attribution Remedies | Future Attribution Research |
|---|
| Across Multiple Touchpoints |
| 1. Estimate downstream and interaction impact of each touchpoint when it has an effect on multiple subsequent stages in the purchase funnel. | 1a. Assess touchpoints' long-term impact and synergies in marketing-mix model.1b. Deploy hidden Markov models to assess the impact of various channels at different stages of the decision process. | 1. What is the value of assembling a rich data set that tracks customers across different stages of the purchase funnel and links them to various interactions between the firm and customers at each of these stages? |
| 2. Apportion credit among different omnichannel marketing activities when customers interact with them simultaneously. | 2. Develop MTA models to attribute the individual-level purchase conversion to exposures to individual marketing messages. | 2. How can firms exploit differences in flexibility among communication channels to change communication touchpoints on short notice for attribution? |
| 3a. Identify ineffective marketing touchpoints based on purchases that did not occur if the failure of that touchpoint to drive purchase also led other touchpoints to fail.3b. Identify effectiveness of unused touchpoints if a firm uses only a subset of potential touchpoints. | 3. Undertake carefully curated randomized field experiments and leverage advanced machine learning (e.g., multi-armed bandits) and econometric methods to evaluate the effectiveness of marketing interventions. | 3a. What is the value of obtaining verifiable fine-grained data on consumer exposure to touchpoints via blockchain technology in improving attribution?3b. Can we develop modeling approaches that are scalable to touchpoints that are large in dimensionality? |
| Across Aggregation Levels |
| 4. Identify the set of marketing touchpoints that each customer is exposed to when the data on these exposures are aggregated at different levels. | 4. Develop models that combine information on touchpoints across different levels of aggregation. | 4. Can we develop approaches that can integrate MTA (individual) with aggregate marketing-mix models? |
How should firms resolve the first attribution challenge—that the effect of a marketing intervention can carry over to subsequent stages? One way to address this problem is to employ extant methods that have focused on modeling long-term effects (e.g., [24]; [38]; [41]; [56]; [70]). While traditional attribution modeling has used aggregate metrics (e.g., overall TV ad budget, number of website visits, net social media sentiment), more recent research uses individual-level path-to-purchase data. This has enabled researchers to obtain a richer understanding of carryover and spillover effects across channels ([21]; [31]; [47]; [68]).
[ 1] model customers' states in their decision processes using a hidden Markov model to assess the impact of various channels at different stages of the decision process. [ 4] propose a graph-based attribution model that maps the sequential nature of customer paths as first- and higher-order Markov walks and shows the idiosyncratic channel preferences (carryover) and interaction effects both within and across channel categories (spillover). [88] develop a hierarchical Bayesian model for individual differences in purchase propensity and marketing response across channels, finding that catalogs have a substantially longer-lasting purchase impact on customer purchase than emails.
The second challenge pertains to the case in which firms might employ multiple touchpoints simultaneously (i.e., within each stage in the purchase funnel) and/or when consumers might be multihoming. In such settings, firms tend to use heuristics such as first touch and last touch to infer attribution. In recent years, several "digital native" companies have developed intricate ways to uncover and influence online consumer decision journeys and attribute the individual-level purchase conversion to the individual exposure to specific marketing messages. As a result, MTA has come into prominence in recent years ([48]). This body of research has demonstrated the limits of heuristics such as last- and first-click attribution shortcuts. For example, [23] find evidence that last-click attribution can underestimate the effectiveness of some types of interventions and lead to suboptimal budget allocation. In addition, research has explored mapping and visualizing different consumer journeys in the digital space across display and search ads ([31]), examining the impact of offline channel opening on consumers' online shopping behaviors or vice versa ([10]; [28]; [49]; [63]) and developing more efficient ways to analyze and store big data ([13]).
However, MTA runs into problems when companies also use more traditional marketing communication channels such as TV, radio, print, and billboards, as even digital native companies such as Amazon and Kayak do. Individual-level exposure and response data are either not available for these channels or their collection is severely constrained by costs and/or privacy concerns.[ 9] Likewise, MTA typically does not account for nonpaid influences on individual consumers, such as online and offline word of mouth ([26]).
Next, we consider the third issue related to attribution: understanding the effectiveness of unsuccessful and unexplored interventions. To this end, advertisers are increasingly undertaking carefully curated randomized field experiments and leveraging advanced machine learning and econometric methods to evaluate the effectiveness of marketing interventions. Methods such as multi-armed bandits ([67]) have the potential to address some of these challenges. Multi-armed bandit experimentation is good for situations where conditions can change over time. This is essentially an optimization-driven approach where the omnichannel marketer creates a series of ads, which can be delivered to users by running multiple concurrent combinatorial tests of the creative, and offers to find the combinations that deliver the best results (e.g., click, conversion, revenue) ([76]). Multi-armed bandit experimentation can be slower than traditional A/B testing, but it is more robust in dynamic contexts and thus has the potential to lead to a more reliable digital attribution analysis.
While these innovations in attribution modeling have significantly improved firms' ability to assign credit to a specific marketing touchpoint, several challenges remain, which serve as the basis for future research. First, attribution models still cannot link the transition across stages of the purchase funnel to a single marketing intervention. They typically presume that the impact of the previous intervention stops with the next step within the purchase funnel and that this impact does not carry over to subsequent steps within the funnel. This assumption is inconsistent, for example, with aggregate-level findings that content-related (vs. content-separated) ads generate site traffic that is more likely to convert in the add-to-cart and checkout stages ([23]). This attribution challenge can be addressed by assembling a rich data set that tracks customers across different stages of the purchase funnel and can link them to their various interactions with the firm at each of these stages. If such data have sufficient variation in terms of the extent of firm–customer interactions at different stages of the purchase funnel, we should be able to map the short- and long-term impacts of marketing interventions at different stages of the purchase funnel and beyond.
Second, in many settings, omnichannel marketers may have access to customer-level data for some channels and only aggregate data for other channels. There is a well-established tradition in marketing that combines aggregate and disaggregate data ([11]; [12]; Chintagunta, Gopinath, and Venkataraman [17]; [18]; [66]; [75]). These studies have shown that the combination of customer-level and aggregate data (usually market-level sales data) allows for a better, much richer understanding of consumer heterogeneity than either micro or macro data alone. To the best of our knowledge, we are unaware of any attribution models that leverage aggregate and disaggregate data.
Third, as omnichannel marketers adopt technologies such as blockchain, these firms will realize greater transparency and more reliable integration of consumer data across touchpoints within and outside the firm. Precise MTA modeling and empirical analyses require access to atomic user-level data, some of which come from touchpoints on assets owned by the firm (e.g., the data that the brand may own about a consumer surfing on its website or mobile app) and some from touchpoints on external sources (platform-owned data about a consumer created when that consumer interacts with the brand's ads on Google, Instagram, Amazon, and others). Examples of such granular information include details about the various touchpoints in the consumer path to purchase, the sequence of touchpoints, the kind of content published on a given touchpoint and time spent interacting with that content, the kind of ads (e.g., search, display, video) on a given touchpoint and the time spent interacting with ads, the time lag between different touchpoints, and how frequently the consumer visited that touchpoint in the past. Such fine-grained omnichannel data about consumer response to digital advertising eventually need to be verified, collated, and made accessible. In implementing marketing-mix and attribution models, it is important to verify the various customer touchpoints. Blockchain technologies can serve this purpose. This naturally warrants a better understanding of how the attribution effects change (in terms of both magnitude and reliability) with and without blockchain-enabled marketing platforms.
Fourth, as discussed previously, one challenge relates to assessing the effectiveness of unexplored intervention options. Because marketers can potentially have a plethora of intervention options, exploring the effectiveness of each of these options presents a unique challenge. Approaches that balance the trade-off between exploration and exploitation (e.g., the multi-armed bandit approach) have proved to be promising ways to address this issue. However, their ability to scale to a large set of alternatives faced by a typical decision maker is unclear. Developing approaches that are scalable to interventions that are large in dimensionality might be a worthwhile avenue for future research.
Fifth, the channels through which firms interact with their customers may differ in terms of the flexibility of contracts. For example, let us consider the communication touchpoints that a firm may employ to inform its customers about products. Historically, television advertising contracts are negotiated in advance and are largely irreversible ([86]). In contrast, keyword advertising can be changed instantaneously. Low flexibility limits how quickly a firm can experiment with the nature and volume of its interactions with customers, which is required for attribution. In instances where firms concurrently use multiple channels with varying levels of flexibility, can one exploit the differential flexibility as a new source of identification for attribution?
Until recently, questions of privacy and questions of channel structure were far removed from each other. This is because, in general, channel management was associated with a lack of insight into customers' desires, purchases, and feedback. Lack of insight was very much bound up with the lack of data as firms had different experiences with different aspects of consumer behavior.
However, in the omnichannel environment, which relies on a fully integrated view of the various customer touchpoints, privacy issues are becoming a crucial question in any discussion of channel management. The ability to use first-party data and match them with external activity on digital touchpoints not owned by the firms is both novel and attractive for firms, but such practices have been challenged by privacy activists ([80]). In particular, control of a customer's data that may give insight into future sales opportunities is something that, in theory, should be available to all channel participants due to the widespread nature of a customer's digital footprint. However, in practice, channel conflicts can arise when one channel partner claims ownership over these data and tries to exclude other channel partners. Such claims often rely on certain interpretations of privacy regulations and customer privacy preferences. As such, customer privacy concerns can often be in surprising conflict with channel coordination.
There are several reasons why privacy will become an important factor in omnichannel marketing. First, the types of products sold via omnichannel marketing will expand. At the moment, many of the key examples of omnichannel marketing are products, such as coffee, that tend to have short customer decision journeys and for which customers are generally untroubled if their shopping habits are visible to others. Omnichannel marketing may ultimately be most useful, however, for high-involvement products that involve many stages of deliberation and research by the customer. Often, high-involvement products fall into sectors that most naturally give rise to privacy concerns, such as health and finance. Consumers may not be troubled if Starbucks can link coffee-browsing profiles across an app and a store, but they might feel differently about their blood-pressure profile being linked to their features via facial recognition.
Second, as technological capacity improves, the trade-off between personalization and privacy concerns will sharpen. Existing research has emphasized that there are natural trade-offs between a customer's acceptance of personalization and the degree of their privacy concerns and sense of control over their data ([29]; [78]; [85]). Given the natural technological challenges of merely tracking a customer across different touchpoints in their customer decision, as of yet most technological investments have been focused on syncing and tracking. However, once this natural technology barrier has been resolved, firms will soon have to face key decisions about how much personalization they attempt, and how acceptable such personalization will be, given customer privacy concerns. For example, one of the primary goals of matching omnichannel marketing to the customer journey is to link earlier stages in the decision process with prior purchase decisions. However, will customers find it acceptable for firms to remind them of their prior purchase decisions or their product search history across different digital touchpoints?
This leads to three major potential challenges for firms aiming to conduct effective omnichannel marketing while being mindful of consumer privacy concerns. The first challenge is that customers may not be willing to allow the focal firm to collect, parse, and sync their data across devices and touchpoints for use in marketing. The marketing literature has emphasized that one way of addressing this natural privacy concern is to improve perceived consumer control over data. Typically, it is the combination of lack of control and perceived privacy intrusion that is most problematic in customers' minds ([78]). Therefore, many managerial solutions to these constraints imposed on omnichannel marketing by customer privacy concerns may come in the form of improving customer control over their data.
The second challenge is that customers may not be willing to allow other firms that they interact with in their decision journey to collect, parse, and sync their data across devices and share these data with the focal firm. In general, omnichannel marketing has focused on questions of how to piece together disparate fragments of customer data ([61]), in the absence of privacy concerns. However, as of yet, little research has investigated how firms can share customer data with channel partners in a way that reflects consumer privacy concerns.
The third challenge is that regulators may not be willing to allow firms to share, sync, and collect customer data across different firms, devices, and touchpoints. Since May 2018, firms throughout the world have had to grapple with the General Data Privacy Regulation (GDPR), a European Union (EU) regulation designed to ensure that firms document that they have obtained consent from customers to use their data. One of the most striking novelties of this regulation is its global reach. For example, if a Malaysian website served EU citizens, then it is subject to the regulation and needs to ensure that its use of cookies was compliant. Furthermore, penalties for contravening the regulation are large—4% of worldwide turnover. There are already examples of how such regulation has restrained firms' attempts at omnichannel marketing. Firms such as JD Wetherspoon, a restaurant chain, had to take steps antithetical to the ambitions of an omnichannel retailer, such as deleting over 800,000 email addresses and halting email marketing, in anticipation of the regulation ([54]).
Although the GDPR is focused on EU data subjects, there is some evidence that even firms based in the United States are choosing to implement its strictures rather than go through the complex process of identifying which website visitors are or are not affected ([55]). By contrast, the new California Privacy Act in the United States could potentially affect U.S. firms directly. Because the California Privacy Act has some data-use restrictions that resemble that of the GDPR, there may be similar negative effects on firms' ability to pursue omnichannel strategies in the United States. However, at the time of writing of this article, the act is still being litigated and its actual effects are uncertain.
Another effect of the GDPR for omnichannel marketing has been its impact on firms' ability to engage in probabilistic matching. Probabilistic tracking uses data on the visit (e.g., the IP address, the device used, the browser used, the timing, the location) to predict whether it is the same customer. The GDPR has restricted the collection of IP addresses as potentially personally identifiable information. As such, the regulation has restricted one of the major ways that probabilistic matching is done. It has also given incentives to firms to pursue more deterministic forms of tracking, such as forcing the use of login credentials, which may, in turn, be more privacy-intrusive than probabilistic tracking methods.
Many of the potential costs of this regulation for omnichannel markets stem from its focus on obtaining and documenting consent. This means that firms are prioritizing their use of technologies such as customer data platforms for compliance reasons, rather than focusing on the potential for such technologies to provide a more complete picture of a customer or theorizing how that customer might feel about the combination of data the firm is collecting. Customer data platforms are therefore being marketed as a way of tracking the consent status and origins of disparate pieces of information about a customer, rather than their initial aim of enabling seamless omnichannel marketing. It is not clear, however, whether documentation of compliance with the law supplants the ideal use of such technology, which is to ensure that firms track customers across the decision journey in a manner that makes customers feel comfortable. The first column in Table 3 summarizes the key issues related to each of these challenges.
Graph
Table 3. Privacy-Related Challenges, Remedies, and Future Research.
| Privacy Challenges | Privacy Remedies | Future Privacy Research |
|---|
| 1. Customers unwilling to allow the focal firm to collect, parse, and sync their data across devices, touchpoints for marketing in high-involvement settings. | 1a. Make predictions about a customer's likely future purchases based on aggregated actions of other customers instead of storing data about a particular customer.1b. Use blockchain technology to provide incentives to customers in the form of a share in the profits derived from using their data. | 1a. How can researchers build a predictive model whose suggestions are unlikely to be perceived as intrusive?1b. What are the types of industries, products, and patterns of consumer behavior for which offering incentives (facilitated by blockchain technology) will encourage customers to share their data? |
| 2. Customers are unwilling to allow the other firms that they interact with to share their data with the focal firm. | 2. Develop data exchange platforms that allow organizations to match data sets with deidentified information and without ever having to leave the firm's secured servers. | 2a. What are the effects of deploying methods such as blockchain-enabled federated learning architecture on tempering privacy concerns and implementing more efficient omnichannel marketing programs?2b, What are the adverse consequences of identifiable data in inducing algorithmic biases and discriminatory practices? |
| 3. Regulators are unwilling to allow firms to share and sync customer data across different firms, devices, and touchpoints. | 3. Use regulations such as GDPR to give customers control of their data. | 3. What is the extent of privacy regulation compliance among firms and what are its implications for consumer welfare and the firm–consumer relationship? |
In general, the technological frontier on marketing is at odds with maintaining customer privacy. In this subsection, we discuss the source of this tension and offer potential future remedies.
Recent advances in machine learning and other predictive technologies are primarily focused on allowing firms to make predictions about an individual customer's future behavior. This contrasts with previous marketing analytics, which have been focused on predicting aggregate behavior. To address privacy concerns while conducting omnichannel marketing, a firm can either try to guarantee not to predict behavior using only an individual's data or, if they do predict behavior at the individual level, try to ensure that these data and predictions are anonymized. For example, rather than storing data about a particular customer, a firm could make predictions about customers' likely purchase path going forward on the basis of the aggregated actions of other customers. Alternatively, a firm could ensure that all data it stores about an individual are anonymized and depersonalized.
We argue, though, that eventually privacy in omnichannel marketing will become less a question of where data are stored and more a question of whether a customer feels that the predictions made by data are intrusive. Although predictive analytics can be conducted in a way that focuses on using aggregated, anonymized, and depersonalized data, it is not clear that it directly addresses customer privacy concerns, even if it is compliant with privacy regulation. For example, imagine that a customer is browsing a web supermarket storefront, and a predictive analytics suite that uses privacy-compliant aggregated and anonymized data that associates mobile data with desktop website–based data predicts that, in line with her browsing behavior, she is also likely to be interested in contraception. The customer may still find such a suggestion intrusive, even though the suggestion itself was made using privacy-compliant analytics.
As another example, in the world of adtech, Data Republic is a data exchange platform that allows organizations to deidentify and match data sets without personally identifiable information ever having to leave the firm's secured servers. Again, privacy compliance is focused on the question of how and where data are stored and how anonymous the data are when stored.
Blockchain technology may provide customers better (or at least decentralized) ownership rights over their data. In advertising, an example of this is Brave, a "privacy browser" that is combined with blockchain-based digital advertising. The underlying idea is that Brave users will own the rights to their data and share in the profits of firms advertising to them ([14]). The role of blockchain technology is to allow the immutability of "basic attention tokens," which is the currency by which Brave users are rewarded for their attention to advertising. Although Brave has solved some concerns, recently it has been criticized for trying to monetize its users' attention through steering their browsing behavior ([72]).
Although this example is focused on advertising rather than full omnichannel marketing, it does illustrate the potential challenges of using blockchain technology to resolve privacy concerns in a context where multiple firms are trying to track users across multiple touchpoints. The challenges that exist between blockchain technology and data privacy requirements include, at a minimum, the following three use cases: ( 1) different perspectives on anonymity and pseudonymity, ( 2) identification of data controllers and data processors in various blockchain technology implementations and how they affect the applicability of various data protection and privacy laws, and ( 3) reconciling transaction immutability and data preservation in blockchain applications with individuals' rights.
First, it is often believed that transparency afforded by blockchain-related solutions may help mitigate such consumer concerns by giving consumers information on how advertisers have used their data ([30]; [84]). Blockchains are often designed so that all transactions are visible to everyone. They are pseudonymized, meaning that only addresses are visible on the blockchain, and anyone can get an unlimited number of addresses. Still, even in this system, it is possible to identify individuals by examining transactions linked by the addresses ([35]) and statistically predicting the characteristics and identity of an individual by combining these transaction data. Furthermore, it would be very difficult to prevent the visible information from being copied and used in a different way on a different system. Therefore, current blockchain technology that emphasizes visibility and the reduction of asymmetric information may not prevent marketers from selling customer data.
Second, blockchain technology's distributed peer-to-peer network architecture can also put it at odds with data privacy laws such as the GDPR and California Consumer Privacy Act. This is because a law such as the GDPR relies on the idea of centralized controller-based data processing or a distinct firm that oversees and manages data processing. By contrast, blockchain is explicitly decentralized, and part of its merit is that there is not a single controlling firm or body. This disconnect can make it difficult to reconcile current data protection laws with blockchain's other core elements, such as the lack of centralized control, immutability, and perpetual data storage. Regulatory guidance on reconciling this and other potential conflicts is currently a work in progress.
Finally, many of the privacy concerns associated with blockchain stem from the fact that its major virtue is to ensure data integrity and ensure that data are immutable. However, preserving data in an immutable form is itself a privacy challenge.
As we have discussed, blockchain technology can be either permissionless or permissioned. Typically, permissionless blockchains are explicitly decentralized without a governing or controlling body. One potential solution to some of these challenges of protecting privacy in a blockchain environment is to move to permissioned blockchains, such as the IBM technology used by Walmart. IBM Food Trust is a permissioned blockchain that Walmart's suppliers of leafy greens are required to use. However, unlike the more traditional permissionless blockchain, simply participating in the blockchain does not provide any visibility into the data recorded there. Walmart has access to all the information, but suppliers can see only the information they have provided themselves. Such blockchain-based systems provide only constrained transparency, however. The information in the blockchain is more transparent to Walmart than the previous record-keeping methods. The suppliers obtain more information than before, but the system is not fully transparent for them. In other words, concerns about data visibility can be addressed by moving blockchain toward a permissioned format, which loses some of the unique benefits of decentralized blockchains that have often attracted blockchain enthusiasts. However, it is not clear that the permissioned blockchain format addresses issues of immutability of data or the fact that blockchain is essentially a technology focused on preserving and ensuring the integrity of data, which naturally puts it at tension with privacy.
Our discussion highlights that although it is possible to use tools such as machine learning and blockchain to address privacy concerns, the use of these technologies creates different privacy concerns. This insight suggests fruitful avenues for future research. We highlight several possibilities.
First, is there a way of using predictive analytics in a manner that is conscious of customers' likely privacy preferences? For example, is it possible to build a predictive model that ensures that any suggestions made in an omnichannel context are never likely to be perceived as intrusive? To achieve this goal requires a deep understanding of what customers consider a privacy-invasive touchpoint or suggestion in an omnichannel context ([ 7]). We highlight that this kind of research—whether it be done through surveys, data analysis, or A/B testing—is going to be crucial to ensure that predictive analytics are not just privacy-compliant but actually privacy-conscious. Toward this direction of future research, [52] build on the principle of location data obfuscation to provide a framework that allows, for example, a reduction in a firm's probability of being able to infer a customer's home address, with no reduction in actual targeting accuracy for advertising.
Second, can research uncover ways to emulate existing blockchain-based ecosystems in an omnichannel context? For example, can a firm use blockchain to create a token that establishes a currency allowing the consumer to be rewarded for sharing their data as a part of an omnichannel marketing effort? More ambitiously, is there a way that multiple firms can coordinate around a single-token-based scheme to help kick-start a larger ecosystem? As with any time firms work together, there will be interorganizational challenges, especially if the firms are competitors and these proposals involve sharing data. These interorganizational challenges may lead to useful theoretical modeling opportunities for marketing academics. For example, theory work could examine what would give rise to incentive-compatibility issues in a blockchain-fueled data exchange system in an omnichannel context, which would uncover the likelihood and drivers of firms being willing to share data with competitors and channel partners. This would illustrate the types of industries, products, and patterns of consumer behavior offering the largest incentive-compatibility issues in terms of data sharing.
Third, how successful are adtech initiatives that have helped omnichannel marketers become privacy-regulation compliant? Are they inherently just a cost that interrupts the accurate processing of information, or are there benefits in terms of enhanced consumer trust of that firm? For example, if a firm offers an array of privacy-compliance tools, does it actually have a measurable effect on the consumers' relationship to the firm, in terms of measurable purchase behavior or measured attitudinal change? The recent spate of privacy regulation, and in particular regulation in California, has led to a large number of startups that are trying to help firms comply with new regulations ([40]). These vendors span functionalities such as activity monitoring, assessment management, consent management, data discovery, data mapping, deidentification, and privacy management. Each of these functionalities is likely to be core to a privacy-compliant omnichannel future. Yet these are also technologies whose role the academic marketing community knows little about. It strikes us that useful partnerships between academics and firms in this space could help provide an early assessment of the usefulness of such tools, and how to improve them, for firms, consumers, and regulatory compliance.
Fourth, as discussed in the previous section, recent developments in machine learning aim to provide privacy controls. For example, "federated learning" trains a machine-learning algorithm across multiple decentralized devices such as mobile phones that hold local data samples, without exchanging the data. These leakages can stem from loopholes in collaborative machine-learning systems, whereby an adversarial participant can infer membership as well as properties associated with a subset of the training data. [44] propose a blockchained federated learning (BlockFL) architecture, where the local-learning model updates are exchanged and verified using a blockchain. Might such developments temper privacy concerns and lead to more efficient omnichannel marketing programs?
Fifth, public policy has thus far focused on the deleterious effects of machine-learning-induced algorithmic biases in the form of racial or gender discrimination. Scant research or policy has examined the use of personal information in algorithms. For example, does greater transparency into customers' path-to-purchase journey, even with the explicit consent of the customer, result in the unintended consequence of giving omnichannel firms room to price discriminate efficiently and, in doing so, erode consumer welfare? This would be particularly problematic if these data led groups of different socioeconomic backgrounds or different races to pay different prices. As a starting point, it would be useful for research to document the extent to which having more individualized data leads to more price discrimination and, if so, whether that price discrimination appears to be associated with any historically disadvantaged groups.
How does omnichannel marketing differ from how firms have interacted with consumers in the past? In this article, we argue that, to realize the full potential of omnichannel marketing, firms need to track the same consumer across multiple channels. Obtaining such a 360-degree view of the customer experience would require hitherto unimagined consumer tracking capacity by firms. We have highlighted the root causes of three key sources of informational challenges that might prevent firms from realizing the potential of omnichannel marketing—data access/integration, marketing attribution, and protection of consumers' privacy—and discuss how emerging technologies such as machine learning and blockchain can help address these challenges. We establish that while these technologies have promise as solutions, they also create new challenges and opportunities. In addition, we discuss fruitful avenues for future research in each of the three challenge areas. Next, we highlight several possibilities of future research that integrate the three areas.
First, obtaining a 360-degree view of the customer experience, on the one hand, while maintaining customer privacy, on the other, seem to be at odds. However, a firm might need only a subset of information on customer touchpoints to make effective inferences about attribution. If some of these data that firms might not need for attribution are also those for which customers have serious privacy concerns, the firm could collect only the subset that is useful for its internal purposes, thus giving customers a semblance of privacy. Identifying such data represents a potential win-win and therefore is a useful area of research. This is likely a process that will need to be ongoing as consumer education and government regulation increases.
Second, related to the previous point, are there some types of data that are only needed in the short run for attribution purposes about which customers have privacy concerns? Identifying such data is a useful area of research from a public policy perspective, as countries could mandate potentially attractive regulations limiting retention of such data.
Third, while more information is always beneficial to the firm from the perspective of managing customer experience, there may be diminishing returns. Therefore, might it be worthwhile to quantify the incremental benefit of additional data or data from multiple sources for attribution? If we believe it is the combination of data that represents the greatest privacy risk, it would be beneficial for future research to identify instances where there are swift diminishing returns to incremental data in companies, as these data could be removed from regular collection.
Fourth, could there be a marketplace for consumer data that results in fair valuation while preserving privacy, thus creating a win-win situation? Many consumers are increasingly willing to share their personal data (e.g., their location) with brands in return for some economic incentives (e.g., discounts). This comes from the belief that their data are their asset, and just like a property right, they should be able to exchange this asset with brands for monetary compensation from marketers ([39]). Some consumers, however, hesitate to participate because they believe that brands and marketers may not appropriately compensate them for their data. Future research could consider how platform design can inspire consumer confidence and how various mechanisms, such as auction, might be useful in clearing such a market.
Fifth, can blockchain-based technologies be used in facilitating the market for customer information? The hope is that when such a blockchain-based marketplace emerges, consumers will have a transparent overview of how their data are valued and which brands might be willing to enter an exchange with them. It would be beneficial for future research to identify the hurdles (both from consumers and firms) to participating in such markets, and how to overcome them.
In summary, our thesis is that while omnichannel marketing promises to open up new opportunities for firms, firms need to be cognizant of the tension between obtaining a 360-degree view of the customer (and the challenges therein) and alleviating concerns about loss of privacy. We hope that our article helps spearhead future research solving these challenges in omnichannel marketing.
Footnotes 1 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
2 Funding The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The first author's research is partially supported by the NSFC Project No. 71832008.
3 [65], p. 334) identify three types of channel intermediaries: distribution channels, transactional channels, and communication channels. The distribution function is rooted in realizing efficiency ([71]) and often involves functions such as sorting, inventory holding, assortment management, and so on. Transaction channels "facilitate economic exchanges between buyers and sellers," while communication channels inform buyers about "the availability and features of the seller's product or service." Unless stated otherwise, in this article, we assume that channels serve all three intermediation functions.
4 For a discussion of blockchain technologies and their impact on operations management, see [9].
5 As an example, consider TradeLens, the shipping blockchain started by IBM and Maersk, which also has added several competitors to the system ([53]).
6 An important caveat to keep in mind is that the larger digital platforms will need to be appropriately incentivized to adopt a blockchain-based mechanism that can alleviate issues of data inconsistency across supply chain and opacity in how money is shared between the different entities that sit between the brand and the publisher.
7 The visibility here does not need to mean that all players see all entries in the shared database. For example, for the blockchain solution developed by IBM and used by Walmart to operate its supply chain for leafy greens, only Walmart and selected validators have access to all the data. The supplier can see only the data related to its interaction with the supply chain, but not competitors'. However, information stored in the blockchain can be available upon request (e.g., for auditing, allowing the consumer to check the provenance of a particular head of lettuce by scanning a QR code).
8 Blockchain-enabled smart contracts are virtual agreements that remove the need for validation, review, or authentication by intermediaries ([19]).
9 The problems arise because with traditional analog media, it would be difficult to match individual customers and their touchpoints with the firm. This is somewhat aided by the advent of programmatic television and addressable television markets, but there are still many media (e.g., billboards) for which it is nearly impossible to get individual data.
References Abhishek Vibhanshu, Fader Peter, Hosanagar Kartik. (2015), "Media Exposure Through the Funnel: A Model of Multi-Stage Attribution," https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2158421.
Agrawal Ajay, Gans Joshua, Goldfarb Avi. (2018), Prediction Machines: The Simple Economics of Artificial Intelligence. Boston : Harvard Business Review Press.
Ailawadi Kusum L., Farris Paul W. (2017), "Managing Multi- and Omnichannel Distribution: Metrics and Research Directions," Journal of Retailing, 93 (1), 120–35.
Anderl Eva, Becker Ingo, Wangenheim Florian von, Schumann Jan H. (2016), "Mapping the Customer Journey: Lessons Learned from Graph-Based Online Attribution Modeling," International Journal of Research in Marketing, 33 (3), 457–74.
Ansari Asim, Mela Carl F., Neslin Scott A. (2008), "Customer Channel Migration," Journal of Marketing Research, 45 (1), 60–76.
Ascarza Eva, Fader Peter S., Hardie Bruce G.S. (2017), " Marketing Models for the Customer-Centric Firm," in Handbook of Marketing Decision Models, 2nd ed., Wierenga Berend, van der Lans Ralf, eds. Cham, Switzerland : Springer International, 297–329.
Athey Susan, Catalini Christian, Tucker Catherine. (2017), " The Digital Privacy Paradox: Small Money, Small Costs, Small Talk," No. w23488. National Bureau of Economic Research.
Avery Jill, Steenburgh Thomas J., Deighton John, Caravella Mary. (2012), "Adding Bricks to Clicks: Predicting the Patterns of Cross-Channel Elasticities over Time," Journal of Marketing, 76 (3), 96–111.
Babich Volodymyr, Hilary Gilles. (2020), "Distributed Ledgers and Operations: What Operations Management Researchers Should Know About Blockchain Technology," Manufacturing and Service Operations Management, 22 (2), 223–40.
Bell David R., Gallino Santiago, Moreno Antonio. (2018), "Offline Showrooms in Omnichannel Retail: Demand and Operational Benefits," Management Science, 64 (4), 1629–51.
Berry Steven, Levinsohn James, Pakes Ariel. (2004), "Differentiated Products Demand Systems from a Combination of Micro and Macro Data: The New Car Market," Journal of Political Economy, 112 (1), 68–105.
Besanko David, Dubé Jean-Pierre, Gupta Sachin. (2003), "Competitive Price Discrimination Strategies in a Vertical Channel Using Aggregate Retail Data," Management Science, 49 (9), 1121–38.
Bradlow Eric T., Gangwar Manish, Kopalle Praveen, Voleti Sudhir. (2017), "The Role of Big Data and Predictive Analytics in Retailing," Journal of Retailing, 93 (1), 79–95.
Brave (2019), "Brave Software and TAP Network Put Blockchain to Work with New Partnership," press release (February 26), https://brave.com/brave-tap-Blockchain/.
Chen Yuxin, Chakravarthi Narasimhan, John Zhang Z. (2001), "Individual Marketing with Imperfect Targetability," Marketing Science, 20 (1), 23–41.
Chen Yuxin, Steckel Joel H. (2012), "Modeling Credit Card Share of Wallet: Solving the Incomplete Information Problem," Journal of Marketing Research, 49 (5), 655–69.
Chintagunta Pradeep K., Gopinath Shyam, Venkataraman Sriram. (2010), "The Effects of Online User Reviews on Movie Box Office Performance: Accounting for Sequential Rollout and Aggregation Across Local Markets," Marketing Science, 29 (5), 944–57.
Christen Markus, Gupta Sachin, Porter John C., Staelin Richard, Wittink Dick R. (1997), "Using Market-Level Data to Understand Promotion Effects in a Nonlinear Model," Journal of Marketing Research, 34 (3), 322–34
Cong William L., He Zhiguo. (2019), "Blockchain Disruption and Smart Contracts," Review of Financial Studies, 32 (5), 1754–97.
Coughlan Ann T. (2011), " Marketing Channel Strategy," in Marketing Strategy. Hoboken, NJ : John Wiley and Sons.
Dalessandro Brian, Perlich Claudia, Stitelman Ori, Provost Foster. (2012), " Causally Motivated Attribution for Online Advertising," ADKDD '12: Proceedings of the Sixth International Workshop on Data Mining for Online Advertising and Internet Economy Article 7 (August 12), Beijing, 1–9.
Danaher Peter J., Dagger Tracey S. (2013), "Comparing the Relative Effectiveness of Advertising Channels: A Case Study of a Multimedia Blitz Campaign," Journal of Marketing Research, 50 (4), 517–34.
De Haan Evert, Wiesel Thorsten, Pauwels Koen. (2016), "The Effectiveness of Different Forms of Online Advertising for Purchase Conversion in a Multiple-Channel Attribution Framework," International Journal of Research in Marketing, 33 (3), 491–507.
Dekimpe Marnik G., Hanssens Dominique M. (1999), "Sustained Spending and Persistent Response: A New Look at Long-Term Marketing Profitability," Journal of Marketing Research, 36 (4), 397–412.
Dzyabura Daria, Kihal Siham El, Hauser John R., Ibragimov Marat. (2019), "Leveraging the Power of Images in Predicting Product Return Rates," working paper, Stern School of Business, New York University.
Fay Brad, Keller Ed, Larkin Rick, Pauwels Koen. (2019), "Deriving Value From Conversations About Your Brand," MIT Sloan Management Review, 60 (2), 72–77.
Fontanella Clint. (2020), "15 Examples of Brands With Brilliant Omni-Channel Experiences," blog entry, HubSpot (February 28), https://blog.hubspot.com/service/omni-channel-experience.
Forman Chris, Ghose Anindya, Goldfarb Avi. (2009), "Competition Between Local and Electronic Markets: How the Benefit of Buying Online Depends on Where You Live," Management Science, 55 (1), 47–57.
Ghose Anindya. (2017), Tap: Unlocking the Mobile Economy. Cambridge, MA : MIT Press.
Ghose Anindya. (2018), "What Blockchain Could Mean for Marketing," Harvard Business Review (May 14), https://hbr.org/2018/05/what-blockchain-could-mean-for-marketing.
Ghose Anindya, Todri Vilma. (2016), "Towards a Digital Attribution Model: Measuring the Impact of Display Advertising on Online Consumer Behavior," MIS Quarterly, 40 (4), 889–910.
Godfrey Andrea, Seiders Kathleen, Voss Glenn B. (2011), "Enough Is Enough! The Fine Line in Executing Multichannel Relational Communication," Journal of Marketing, 75 (4), 94–109.
Gordon Brett R., Jerath Kinshuk, Katona Zsolt, Narayanan Sridhar, Shin Jiwoong, Wilbur Kenneth C. (2021), "Inefficiencies in Digital Advertising Markets," Journal of Marketing, 85 (1), 7–25.
Gustafson Krystina. (2017), "An Overwhelming Number of Retailers Are Losing Money Chasing Amazon," CNBC (February 23) https://www.cnbc.com/2017/02/22/an-overwhelming-amount-of-retailers-are-losing-money-chasing-amazon.html.
Haeringer Guillaume, Halaburda Hanna. (2018), "Bitcoin: A Revolution? " in Digital Economy, Ganuza J., Llobet G., eds. Madrid : FUNCAS, 397–422.
Halaburda Hanna. (2018), "Blockchain Revolution Without the Blockchain," Communications of the ACM, 61 (7), 27–29.
Halaburda Hanna, Sarvary Miklos. (2016), Beyond Bitcoin: The Economics of Digital Currencies. New York : Palgrave McMillan.
Hanssens Dominique M., Pauwels Koen. (2016), "Demonstrating the Value of Marketing," Journal of Marketing, 80 (6), 173–90.
Harvey Campbell R., Moorman Christine, Toledo Marc. (2018), "How Blockchain Can Help Marketers Build Better Relationships with Their Customers," Harvard Business Review (October), https://hbr.org/2018/10/how-blockchain-can-help-marketers-build-better-relationships-with-their-customers.
International Association of Privacy Professionals (2019), "2019 Privacy Tech Vendor Report," research report (accessed October 15, 2020), https://iapp.org/media/pdf/resource_center/2019TechVendorReport.pdf.
Jedidi Kamel, Mela Carl F., Gupta Sunil. (1999), "Managing Advertising and Promotion for Long-Run Profitability," Marketing Science, 18 (1), 1–22.
Joo Mingyu, Wilbur Kenneth C., Cowgill Bo, Zhu Yi. (2014), "Television Advertising and Online Search," Management Science, 60 (1), 56–73.
Kannan P.K., Reinartz Werner, Verhoef Peter C. (2016), "The Path to Purchase and Attribution Modeling: Introduction to Special Section," International Journal of Research in Marketing, 33 (3), 449–56.
Kim Hyesung, Park Jihong, Bennis Mehdi, Kim Seong-Lyun. (2018), "On-Device Federated Learning via Blockchain and Its Latency Analysis," https://arxiv.org/abs/1808.03949.
Konecný Jakub, McMahan H. Brenda, Ramage Daniel, Richtárik Peter. (2016), "Federated Optimization: Distributed Machine Learning for On-Device Intelligence," https://arxiv.org/abs/1610.02527.
Lee Hyeong-Tak, Venkataraman Sriraman. (2019), "TV Advertising and Online Search—Understanding with- and Across-Brand Spillovers Using Clickstream Data," working paper.
Li Hongshuang, Kannan P.K. (2014), "Attributing Conversions in a Multichannel Online Marketing Environment: An Empirical Model and a Field Experiment," Journal of Marketing Research, 51 (1), 40–56.
Li Jing, Konus Umut, Pauwels Koen, Langerak Fred. (2015), "The Hare and the Tortoise: Do Earlier Adopters of Online Channels Purchase More?" Journal of Retailing, 91 (2), 289–308.
Liang Yilong, Qian Yue, Cui Tony H., John George, Yang Shilei. (2019), How Offline Experience Changes Online Behavior of Member-Customer Segments," working paper, University of Minnesota.
Liaukonyte Jura, Teixeira Thales, Wilbur Kenneth C. (2015), "Television Advertising and Online Shopping," Marketing Science, 34 (3), 311–30.
Lin Chen, Venkataraman Sriram, Jap Sandy D. (2013), "Media Multiplexing Behavior: Implications for Targeting and Media Planning," Marketing Science, 32 (2), 310–24.
Macha Meghanath, Li Beibei, Foutz Natasha Z., Ghose Anindya. (2019), "Perils of Location Tracking? Personalized and Interpretable Privacy Preservation in Consumer Trajectories," working paper, Carnegie Mellon University.
Maersk (2019), "TradeLens Blockchain-Enabled Digital Shipping Platform Continues Expansion with Addition of Major Ocean Carriers Hapag-Lloyd and Ocean Network Express," press release (July 2), https://www.maersk.com/news/articles/2019/07/02/hapag-lloyd-and-ocean-network-express-join-tradelens.
Manthorpe Rowland. (2017), "Wetherspoons Just Deleted Its Entire Customer Email Database—On Purpose," Wired (July 3), https://www.wired.co.uk/article/weatherspoons-email-database-gdpr.
Marthews Alex, Tucker Catherine. (2019), "Privacy Policy and Competition," research report, Brookings (December), https://www.brookings.edu/wp-content/uploads/2019/12/ES-12.07.19-Marthews-Tucker.pdf.
Mela Carl F., Gupta Sunil, Lehmann Donald R. (1997), "The Long-Term Impact of Promotion and Advertising on Consumer Brand Choice," Journal of Marketing Research, 34 (2), 248–61.
Miller Amalia R., Tucker Catherine. (2014), "Health Information Exchange, System size and Information Silos," Journal of Health Economics, 33 (January), 28–42.
Mulpuru Susharita. (2011), "The Purchase Path of Online Buyers," Forrester Report (March 16), https://www.forrester.com/report/The+Purchase+Path+Of+Online+Buyers/-/E-RES58942.
Naik Prasad, Raman Kalyan. (2003), "Understanding the Impact of Synergy in Multimedia Communications," Journal of Marketing Research, 40 (4), 375–88.
Neslin Scott A., Grewal Dhruv, Leghorn Robert, Shankar Venkatesh, Teerling Marije L., Thomas Jacquelyn S., et al. (2006), "Challenges and Opportunities in Multichannel Customer Management," Journal of Service Research, 9 (2), 95–112.
Neumann Nico, Tucker Catherine, Whitfield Timothy. (2019), "Frontiers: How Effective Is Third-Party Consumer Profiling? Evidence from Field Studies," Marketing Science, 38 (6), 918–26.
Nielsen (2018), "The Nielsen CMO Report 2018," research report (June 6), https://www.nielsen.com/us/en/insights/report/2018/cmo-report-2018-digital-media-roi-measurement-omnichannel-marketing-technology/.
Pauwels Koen, Neslin Scott A. (2015), "Building with Bricks and Mortar: The Revenue Impact of Opening Physical Stores in a Multichannel Environment," Journal of Retailing, 91 (2), 182���97.
Pauwels Koen, Leeflang Peter S., Teerling Marije L., Eelko Huizingh K.R. (2011), "Does Online Information Drive Offline Revenues? Only for Specific Products and Consumer Segments!" Journal of Retailing, 87 (1), 1–17.
Peterson Robert A., Balasubramanian Sridhar, Bronnenberg Bart J. (1997), "Exploring the Implications of the Internet for Consumer Marketing," Journal of the Academy of Marketing Science, 25 (4), 329–46.
Petrin Amil. (2002), "Quantifying the Benefits of New Products: The Case of the Minivan," Journal of Political Economy, 110 (4), 710–29.
Schwartz Eric M., Bradlow Eric T., Fader Peter S. (2017), "Customer Acquisition via Display Advertising Using Multi-Armed Bandit Experiments," Marketing Science, 36 (4), 500–22.
Shao Xuhui, Li Lexin. (2011), " Data Driven Multi-Touch Attribution Models," in Proceedings of the 17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, San Diego (August, 21–24).
Sridhar Shrihari, Sriram S. (2013), "Is Online Newspaper Advertising Cannibalizing Print Advertising?" Quantitative Marketing and Economics, 13 (4), 283–318.
Sriram S., Kalwani Manohar. (2007), "Optimal Advertising and Promotion Budgets in Dynamic Markets with Brand Equity as a Mediating Variable," Management Science, 53 (1), 46–60.
Stern Louis W., EI-Ansary Adel I., Coughlan Anne T. (1996), Marketing Channels, 5th ed. Upper Saddle River, NJ : Prentice Hall.
Stevens Robert. (2020), "Privacy Browser Brave Under Fire for Violating Users' Trust," Decrypt (June 6), https://decrypt.co/31522/crypto-brave-browser-redirect.
Stuart Greg, Rubinson Joel, Bakopoulos Vassilis. (2017), "The Case for Multi-Touch Attribution in the Age of People-Based Marketing: Why It Matters (and Why It Is Hard)," Applied Marketing Analytics, 3 (3), 226–38.
Teixeira Thales S., Piechota Greg. (2019), Unlocking the Customer Value Chain: How Decoupling Drives Customer Disruption: Currency. New York : Currency.
Tenn Steven. (2006), "Avoiding Aggregation Bias in Demand Estimation: A Multivariate Promotional Disaggregation Approach," Quantitative Marketing and Economics, 4 (4), 383–405.
Thomas Ian. (2017), "Solving the Attribution Conundrum with Optimization-Based Marketing," blog entry Lies, Damned Lies (January 25) https://www.liesdamnedlies.com/2017/01/solving-the-attribution-conundrum-with-optimization-based-marketing.html.
Tonietto Gabriela N., Barasch Alixandra. (2020), "Generating Content Increases Enjoyment by Immersing Consumers and Accelerating Perceived Time," Journal of Marketing, (published online September 10), DOI:10.1177/0022242920944388.
Tucker Catherine E. (2014), "Social Networks, Personalized Advertising, and Privacy Controls," Journal of Marketing Research, 50 (5), 546–62.
Van Nierop, Johannes E.M., Leeflang Peter S.H., Teerling Marije L., Eelko Huizingh K.R. (2011), "The Impact of the Introduction and Use of an Informational Website on Offline Customer Buying Behavior," International Journal of Research in Marketing, 28 (2), 155–65.
Venkatadri Giridhari, Lucherini Elena, Sapiezynski Piotr, Mislove Alan. (2019), "Investigating Sources of PII Used in Facebook's Targeted Advertising," Proceedings on Privacy Enhancing Technologies, 2019 (1), 227–44.
Verhoef Peter C., Kannan P.K., Inman J. Jeffrey. (2015), "From Multi-Channel Retailing to Omnichannel Retailing: Introduction to the Special Issue on Multi-Channel Retailing," Journal of Retailing, 91 (2), 174–81.
Vinhas Alberto S., Anderson Erin. (2005), "How Potential Conflict Drives Channel Structure: Concurrent (Direct and Indirect) Channels," Journal of Marketing Research, 42 (4), 507–15.
Wathne Kenneth H., Heide Jan B. (2000), "Opportunism in Interfirm Relationships: Forms, Outcomes and Solutions," Journal of Marketing, 64 (4), 36–51.
Werbach Kevin. (2018), The Blockchain and the New Architecture of Trust. Cambridge, MA : MIT Press.
White Tiffany, Zahay Debra, Thorbjornsen Helge, Shavitt Sharon. (2008), "Getting Too Personal: Reactance to Highly Personalized Email Solicitations," Marketing Letters, 19 (1), 39–50.
Wilbur Kenneth C. (2008), "A Two-Sided, Empirical Model of Television Advertising and Viewing Markets," Marketing Science, 27 (3), 356–78.
Wilder-James Edd. (2016), "Breaking Down Data Silos," Harvard Business Review (December 5), https://hbr.org/2016/12/breaking-down-data-silos.
Zantedeschi Daniel, Feit Eleanor M., Bradlow Eric T. (2017), "Measuring Multichannel Advertising Response," Management Science, 63 (8), 2706–28.
Zhang Sha, Pauwels Koen, Peng Chenming. (2019), "The Impact of Adding Online-to-Offline Service Platform Channels on Firms' Offline and Total Sales and Profits," Journal of Interactive Marketing, 47, 115–28.
~~~~~~~~
By Tony Haitao Cui; Anindya Ghose; Hanna Halaburda; Raghuram Iyengar; Koen Pauwels; S. Sriram; Catherine Tucker and Sriraman Venkataraman
Reported by Author; Author; Author; Author; Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 75- Innovation Imprinting: Why Some Firms Beat the Post-IPO Innovation Slump. By: Wies, Simone; Moorman, Christine; Chandy, Rajesh K. Journal of Marketing. Aug2022, p1. DOI: 10.1177/00222429221114317.
Ahead of Print- Database:
- Business Source Complete
Record: 76- Investigating the Effects of Excise Taxes, Public Usage Restrictions, and Antismoking Ads Across Cigarette Brands. By: Wang, Yanwen; Lewis, Michael; Singh, Vishal. Journal of Marketing. May2021, Vol. 85 Issue 3, p150-167. 18p. 7 Charts, 1 Graph. DOI: 10.1177/0022242921994566.
- Database:
- Business Source Complete
Investigating the Effects of Excise Taxes, Public Usage Restrictions, and Antismoking Ads Across Cigarette Brands
The prevalence of strong brands such as Coca-Cola, McDonald's, Budweiser, and Marlboro in "vice" categories has important implications for regulators and consumers. While researchers in multiple disciplines have studied the effectiveness of antitobacco countermarketing strategies, little attention has been given to how brand strength may moderate the efficacy of tactics such as excise taxes, usage restrictions, and educational advertising campaigns. In this research, the authors use a multiple discrete-continuous model to study the impact of antismoking techniques on smokers' choices of brands and quantities. The results suggest that although cigarette excise taxes decrease smoking rates, these taxes also result in a shift in market share toward stronger brands. Market leaders may be less affected by tax policies because their market power allows strong brands such as Marlboro to absorb rather than pass through increased taxes. In contrast, smoke-free restrictions cause a shift away from stronger brands. In terms of antismoking advertising, the authors find minimal effects on brand choice and consumption. The findings highlight the importance of considering brand asymmetries when designing a policy portfolio on cigarette tax hikes, smoke-free restrictions, and antismoking advertising campaigns.
Keywords: advertising; antismoking; cigarette excise taxes; cigarette marketing; smoke-free restrictions; public policy
While the goal of marketing is usually to boost purchase rates and strengthen relationships between consumers and brands, countermarketing is an increasingly common strategy for reducing the consumption of "vice" goods such as cigarettes. Countermarketing activities may include excise taxes that increase consumer costs, usage constraints that restrict public consumption, and advertising that highlights product dangers. Cigarette countermarketing has seemingly been effective, as U.S. smoking rates have dropped from 44% in 1950 to 14% in 2011 ([16]). In addition, countermarketing is now increasingly applied in other categories that may create health risks, such as soft drinks and fast food ([36]).
A notable feature of many "vice" categories is that they are dominated by very strong or high-equity brands. For example, the Interbrand Top 100 brands list has often included Coca-Cola, McDonald's, Budweiser, and Marlboro. However, economic and public health research on countermarketing effectiveness has largely ignored the role of brands. This is an oversight in that the perceived importance of branding and marketing is demonstrated by advocacy groups' and regulators' efforts to limit brand advertising. Although almost all previous branding research has focused on the value of strong brands in forming and maintaining brand–consumer relationships, it is reasonable to speculate that strong brands might also affect the efforts of advocacy groups and regulators to disrupt these relationships and reduce consumption.
The marketing literature discusses a variety of benefits that accrue to strong brands. Strong brands may have advantages in terms of increased customer loyalty, diminished price sensitivity, wider distribution, heightened consumer awareness, and other benefits ([ 1]; [ 4]). For example, brands may provide symbolic benefits that increase the value of public consumption, and there may be strong psychological bonds between a brand and its customers (Fournier 1998). Furthermore, stronger brands might enjoy greater channel power that results in wider distribution and customer awareness. A prominent example of an effort to reduce brand power is the Australian government's attempt to limit the influence of branding through mandating plain packaging without any iconography for tobacco products starting in December of 2012 ([15]; [59]).
An important aspect of the literature on branding is that brand strength may be manifested through different mechanisms. Critically, the different dimensions of brand strength may protect brands against or make brands more vulnerable to specific countermarketing tactics. For instance, if brands provide benefits by conveying status or glamor, the most effective regulations may be different than if brand strength involves deeper psychological bonds that influence loyalty or price sensitivity. This insight highlights the importance of including brand-level effects for alternative countermarketing activities in an empirical specification.
Our research investigates how the interplay between branding and countermarketing activities influences consumers' consumption of cigarettes. The tobacco industry provides an important and useful context for our research for several reasons. First, tobacco consumption causes significant economic costs and adverse health consequences. Cigarette smoking has been estimated to cause 480,000 premature deaths each year in the United States, and it imposes health care costs and productivity losses of about $300 billion each year ([16]). Second, this industry has been the target of a significant amount of countermarketing activities that affect consumer decision making. For instance, taxes increase consumer prices, smoking bans make public consumption less convenient, and educational advertising campaigns highlight adverse health consequences. In addition, as countermarketing tactics are largely determined at the state level in the United States, there is a significant variation in policies across states. This variation facilitates identification of the effectiveness of different countermarketing techniques. Third, advocacy groups and regulators are currently using experience from the tobacco category to guide efforts in other categories. For example, there is significant interest in using countermarketing techniques to reduce obesity ([36]; [38]; [51]; [52]). Fourth, differences in brand equity in the cigarette category afford an opportunity to study the interplay between countermarketing techniques and brand power.
Vice categories such as cigarettes are also of interest because they highlight the existence and incentives of diverse stakeholders within a category. These diverse perspectives are relevant to consider because groups with different goals may adjust strategies in response to different regulatory approaches. For instance, the literature on countermarketing ([18]) has primarily focused on the effectiveness of regulations in reducing smoking. This perspective is concerned with identifying successful tactics for regulators by tallying smoker quit rates. While these analyses are important, they are incomplete. In addition to quit rates, governments may be interested in the impact of policies on tax revenues, and consumers may suffer economic consequences.
Beyond regulators and consumers, brand manufacturers are often overlooked as relevant participants in the category. This omission affects our knowledge on two aspects of this issue. First, firms wish to select the most effective strategies for their environment. Second, firms and brands vary in terms of their characteristics, distribution strength, and awareness. These factors may lead to different regulatory tactics with asymmetric effects across the category. For example, some brands may have pricing or distributional power that allows them greater flexibility in managing the tax pass-through to consumers. Policies that limit the public consumption of cigarettes may also be relevant because cigarettes have long been considered a prototypical example of a badge product, one used to project social status ([ 8]). Thus, prohibitions on public consumption may vary in effectiveness on the basis of brand strength. How different dimensions of brand strength influence the effectiveness of countermarketing techniques remains an open research topic.
To investigate the relationship between branding and countermarketing, we assemble a data set that includes a consumer panel of cigarette purchases for a six-year period from 2005 to 2010, retail scanner data from 2006 to 2010, and a comprehensive data set on state-level cigarette taxes, state-level smoke-free restrictions, and national antismoking advertising campaigns. We conduct our analysis using a multiple discrete-continuous choice model of smokers' monthly brand and quantity decisions. Our empirical specification is designed to evaluate if and how cigarette excise taxes, smoke-free restrictions, and antismoking advertising campaigns influence cigarette purchase decisions asymmetrically across a variety of brands and composites of brands based on price tier. Our research also includes an analysis of tax pass-through, highlighting the role of brand positioning and channel characteristics. The pass-through analysis is used in a series of counterfactual simulations that assess the full effects of alternative countermarketing techniques across different stakeholders.
Our results show that the effects of antismoking interventions vary significantly across brands. For example, the demand model reveals that Marlboro is relatively less affected by tax increases but relatively more affected by usage restrictions. The resistance to taxes is driven by Marlboro's ability to pass through less of the tax increases than most other brands. This effect may be due to market share based on economies of scale or distribution strength that leads retailers to limit price increases of their highest-volume brand. In the case of usage restrictions, results show that high-equity brands incur more negative effects, and our speculation is that public prohibitions make it more difficult for consumers to garner symbolic or image-based benefits through consumption of high-equity brands. In regard to antismoking advertising, we find that these communications have relatively little effect overall but do have a slightly above-average impact on Marlboro. In terms of category evolution, our results offer an explanation for why Marlboro's relative market share has increased dramatically over time. During our observation window, cigarette excise taxes almost doubled. This aggressive tax policy has shifted demand toward the category leader.
We conduct a series of policy experiments to assess the differential effects of alternative countermarketing policies across stakeholders including regulators, consumers, and brands. We find that a 100% tax increase yields a 30% increase in quit rate, but it imposes significant costs to consumers and only increases tax receipts by about 28%. In contrast, an aggressive smoke-free policy increases quit rates by 9% and reduces tax revenues by 6%. With usage restrictions, consumers may experience inconvenience and reduced symbolic benefits but do not incur economic costs. In general, we find that stronger brands tend to be more resilient to tax increases and more susceptible to usage restrictions. Collectively, our simulations show that the choice of countermarketing tactics greatly impacts relative quit rates, consumer costs, government revenues, and brands' market shares.
To frame our research, we consider selected literature on countermarketing from the fields of economics, public health, and marketing. Economics and public health have significant traditions of studying countermarketing effectiveness, and these disciplines typically rely on surveys rather than actual customer behavior; therefore, marketing issues are usually neglected. The marketing literature focused on tobacco countermarketing has used a variety of experimental and empirical methods to examine consumer response to countermarketing. In addition to research on tobacco control, we also review literature related to the possible interactions between countermarketing and branding. In our review, we put explicit focus on branding topics that may lead to asymmetric effects of countermarketing tactics for stronger versus weaker brands.
The economics literature on smoking has relied on large-scale surveys and reduced-form models to investigate the role of individual countermarketing tactics on consumption ([17]; [23]; e.g., [10]). Of the various countermarketing instruments, excise taxes and pricing have received the most attention in the economics literature. Cigarette excise taxes are implemented at the pack level and are included in retail prices ([21]). These taxes typically include a federal and state component. In general, researchers have found that excise taxes have a significant impact on smoking rates. The price elasticity of cigarette demand is generally found to be about −.4 (see [18]).
Antismoking advocates have been increasingly successful in implementing "smoke-free" restrictions such as prohibitions against smoking in bars, restaurants, and public places. These interventions reduce convenience and increase time costs by forcing smokers outdoors. Smoke-free restrictions have increased in prevalence over time. In the year 2000, approximately 50% of the U.S. population was potentially affected by clean-air smoking policies. By 2008 this percentage had grown to over 70%. Research on smoke-free air policies has yielded mixed results. [25] find that voluntary workplace restrictions lead to minor reductions in smoking. [14] and [ 3] find that smoke-free laws have no impact on smoking behavior. However, these studies all rely on self-reports collected via surveys.
There is a significant literature on the impact of marketing communications on cigarette purchases. For example, [47] find that cigarette brand advertising elasticity is.28. Several marketing studies provide lab-based experimental evidence on the effectiveness of antismoking ad messages ([ 6]; [44]; [45]). For example, [44] use experimental methods to study how smoking scenes in movies elicit different emotional reactions depending on whether an antismoking message was shown before the film. Other research attempts to quantify the relationship between levels of antismoking advertising and quitting behaviors. [60] show that an increase of 390 monthly gross rating points leads to a.3% decline in smoking prevalence in Australia. [26] claim that an increase of 5,000 gross rating points annually increases the odds of quit attempts by 21% in New York City.
There is also a growing marketing literature that evaluates consumer-level purchasing data. In terms of pricing and taxes, [20] examine the effect of Marlboro's permanent 1993 price cut on brand choice; [31] investigate the elasticity of demand for temporary versus permanent price adjustments. Gordon and Sun find that short-term price elasticity is smaller than the long-term elasticity. While these marketing studies illustrate the roles of pricing and promotion on brand-tier choice and incidence, they consider only limited elements of countermarketing. [61] investigate the relative effectiveness of cigarette excise taxes, antismoking advertising, and smoke-free restrictions on category sales. They examine the consequences of the countermarketing mix on product substitution among products with varying nicotine levels but do not consider branding effects. In general, this literature pays little attention to the issues of branding and consumer loyalty. In a notable exception, [15] use secondary data to measure the causal effect of the Australian antibranding legislation at both the cigarette category level and the brand-strength tier level. They find that the elimination of branding elements results in greater price sensitivity to increases for premium and mainstream brands.
Some research investigates the impact of countermarketing on other categories. [30] measure the cross-category spillover effects of selling tobacco products on the revenue generated by nontobacco categories. In addition to tobacco, there is growing interest in using countermarketing techniques to reduce obesity ([36]; [38]; [51]; [52]).
The literature on smoking cessation has largely ignored the impact of branding on efforts to reduce cigarette consumption. This is an oversight given that marketing researchers have found that brand–consumer relationships have significant effects on consumer decision making ([ 5]; Fournier 1998; [35]). The Australian plain tobacco packaging policy has yielded significant results related to the importance of packaging and branding. For example, [24] find that the elimination of branding elements reduces the perceived attractiveness of cigarette packages and affects brand choice. They find that eliminating branding reduces consumer perceptions that the look of their cigarette package "says something good about them" or "is fashionable." In addition, [59] find that the health-oriented warnings mandated by the Australian policy result in an increase in smokers concealing or hiding packages.
A negative link between removing visual branding elements and consumption intentions is intuitive. Cigarettes have often been referred to as badge products, as cigarette consumption frequently involves displays of branded packages in public settings such as bars and nightclubs. There are multiple streams of the marketing literature relevant to the value a badge brand may give a consumer. [33] discusses how brands can act as vehicles for expressing psychological and social traits. [34] suggests that brands provide a means for consumers to express their self-concepts. For instance, consumers might choose Marlboro to associate the rugged brand image with themselves ([ 2]).
Brands can also serve as a focal point for communities of consumers ([39]; [42]). Brand communities are groups largely based on admiration and preference for a focal brand. For these communities to exist, consumption and brand preferences must be publicly expressed so that members can identify each other. A notable example of a consumption community built around a cigarette brand was Marlboro Lights in the U.K. market. The economics literature also includes work that emphasizes the social-signaling benefits of conspicuous consumption ([ 7]).
The role of brands as instruments for expressing self-identity or as a focal point for a consumption community is potentially relevant to the effectiveness of smoke-free air policies. These policies are primarily designed to limit cigarette consumption in public venues. By limiting public consumption, these policies may limit the value that brands provide to consumers. However, there remains an outstanding question as to whether the impact of such policies will vary across types of brands if some brands provide greater symbolic benefits.
Brand strength may also operate through other mechanisms that affect how tax increases are passed through to consumers by the retailer. Specifically, tax increases may have differential effects across brand price tiers due to differences in price sensitivity and distribution channel power ([ 1]). First, higher-equity brands may be more able to pass through greater percentages of tax increases simply because consumers are less price-sensitive for these brands. In fact, given that taxes will tend to shrink the entire category, it is possible that stronger brands may even choose to implement price increases to make up for lost volume. Second, if stronger brands charge higher prices than other brands, then imposing constant per pack taxes will result in lower percentage price increases. Third, awareness and broad distribution may offer benefits in terms of larger market shares and economies of scale—advantages that can accrue to higher-equity brands if retailers wish to maintain prices on especially important brands within a category.
The addictive nature of cigarettes carries its own implications for the design of a consumer demand model. Because nicotine is an addictive substance, much of the repeat buying of cigarettes is driven by physical addiction. However, it is also possible that some type of attitudinal loyalty exists in the category. The key point is that in a category such as cigarettes that includes powerful brands, purchase feedback effects such as brand loyalty, satiation rates, and addiction effects need to be included in any empirical specification.
The review of the existing literature on antismoking effectiveness highlights several salient empirical issues and research gaps. Researchers have investigated the effects of taxes, usage restrictions, and negative advertising. However, these variables have seldom been evaluated simultaneously, and there is still debate about the effectiveness of interventions such as smoke-free restrictions. Therefore, it is critical that any empirical specification include the complete set of countermarketing tactics.
The discussion of branding and consumer issues highlights key considerations for an empirical specification. In terms of branding, the cigarette category includes many brands that vary in terms of price, market share, brand personality, and distribution power. As discussed previously, there are theoretical reasons to believe that different types of brands may be differentially affected by alternative policy interventions. There are also important aspects of brand loyalty that need to be incorporated in an empirical specification. For example, brand loyalty and other purchase feedback effects may be relevant for modeling brand choice. Even basic elements of consumer choice such as whether consumers purchase single or multiple brands need to be considered.
Our research objectives necessitate the use of multiple data sets. To understand consumer-level decisions about brand choices and quantity consumed over time, we use panel data of individual smoker purchases. Given the large number of brands in the category, we supplement the individual-level data with market-level data to identify the price environment faced by consumers. To study the effects of countermarketing, we assemble information on taxes, antismoking ads, and smoke-free restrictions from governmental agencies and nongovernmental organizations.
The individual smoker panel for our study is from the Nielsen Consumer Panel for the six-year period between January 2005 and December 2010.[ 7] The Consumer Panel provides each household with an optical scanner for scanning the barcodes of all consumer packaged goods they purchase, regardless of the outlet. The data, therefore, include purchases from supermarkets, convenience stores, drug stores, gas stations, and other outlets. This broad coverage is important because, unlike the product categories often studied in the literature (i.e., those primarily sold in supermarkets), smaller retail outlets account for a significant proportion of cigarette sales.
We select a sample of smokers for our study by applying the following ordered criteria: ( 1) keep only single smokers that stayed in the Nielsen Consumer Panel for all six years, ( 2) keep smokers that made at least 20 cigarette purchases, and ( 3) keep smokers that had cigarette purchases in 2005, the beginning of our observation window. The three selection criteria result in a panel of 422 single smokers that were potentially in the process of quitting smoking or did quit smoking over the six-year period. We use 2005 as an initialization period and the years 2006–2010 for estimation.
Table 1 shows that approximately 22% of smokers quit smoking, where quitting is defined as individuals with no purchases during the final 12 months of the observation window. The median cigarette purchase interval is about once per month. On average consumers purchase 23 packs and spend an average of $75 on cigarettes per month. To ensure representativeness, we cross-validate our sample's demographics and cigarette consumption patterns against the 2009–2010 CDC National Adult Tobacco Survey. We show in Web Appendix W1 that the 422 single smokers in our estimation sample had similar demographic distributions[ 8] and consumption levels as those in the CDC National Survey.
Graph
Table 1. Smoker Cigarette-Purchase Summary.
| Mean (SD) | Median |
|---|
| Quit rate | 22.27% | — |
| Purchase interval (in months) | 1.69 (1.41) | 1.20 |
| Monthly cigarette spending (cond. on purchase) | 75.74 (55.99) | 63.64 |
| Monthly cigarette packs (cond. on purchase) | 22.68 (16.17) | 20.05 |
20022242921994570 Notes: Quitting is defined as no cigarette purchase during the final 12 months of the observation window.
A primary benefit of the Consumer Panel is that we can observe consumers' brand and quantity choices. However, the 422 smoker panelists purchased more than 170 cigarette brands. We use the following approach to facilitate the analysis: First, we select the top four cigarette brands (in terms of purchase volume) within our purchase panel: Marlboro, Basic, Winston, and Virginia Slims. Next, we aggregate the remaining brands into three categories—premium, mainstream, and economy—on the basis of average national retail price (see Web Appendix W2).
To implement the brand categorization scheme, we obtain information on cigarette prices and quantities sold at the Universal Product Code/store/week level between January 2006 and December 2010 from the Nielsen Retail Scanner Data. We construct brand-level data at the monthly level for each store by aggregating the Universal Product Code–level data at a set of 3,874 retail stores across 46 states. These stores are selected because they have complete price and sales information for the four brands and three price-tier categories. Table 2 presents the average prices of the four individual brands and the three categories of brands. The average price differentials between premium and mainstream brands, and between the mainstream and economy brands, are $.70. Marlboro, Basic, and Winston are priced similarly to the mainstream brand category, while Virginia Slims is priced similarly to the premium category.
Graph
Table 2. Cigarette Brand Prices and Purchases.
| Per Pack Price ($) | Monthly Brand Choice Prob. | Monthly Purchased Packs Conditional on Brand Choice | Unconditional Brand Share |
|---|
| Marlboro | 4.251 (1.156) | 15.23% | 21.180 (18.377) | 19.49% |
| Basic | 4.194 (1.283) | 5.71% | 20.601 (15.878) | 7.11% |
| Virginia Slims | 4.797 (1.307) | 4.65% | 20.828 (16.103) | 5.85% |
| Winston | 4.295 (1.221) | 3.47% | 27.301 (20.012) | 5.72% |
| Premium tier | 4.867 (1.183) | 11.70% | 20.260 (19.397) | 14.33% |
| Mainstream tier | 4.103 (1.134) | 19.89% | 21.644 (19.361) | 26.02% |
| Economy tier | 3.563 (1.131) | 13.14% | 27.055 (23.051) | 21.48% |
30022242921994570 Notes: Standard deviations are in parentheses. "Monthly brand choice probability" refers to the probability a brand will be chosen in the estimation sample, and "monthly packs" refers to the number of packs conditional on purchase. The multiplication of brand choice probabilities and conditional brand purchase quantity provides the unconditional brand shares among the seven brands.
Table 2 also includes data on brand choice and consumption. In Table 2 and the other exhibits in this section, all results related to price are from the 3,874-store Retail Scanner Data, and all results related to brand choice and consumption are calculated using the 422 smokers in the Consumer Panel. In terms of choice, the mainstream category has the highest monthly choice probability at 20%, followed by Marlboro at 15%, economy brands at 13%, and premium brands at 12%. Basic, Virginia Slims, and Winston have choice probabilities of less than 6%. Conditional on brand choice, smokers typically purchase in the range of 20 to 30 packs per month. There is some variation in the average brand purchase quantity. Winston and the economy brands are purchased in slightly larger quantities. The multiplication of brand choice probabilities and conditional brand purchase quantity provides the unconditional brand shares among the seven brands.
The figure in Web Appendix W3 gives additional insight by illustrating the distribution of consumption levels and brand shares for different monthly consumption levels. It shows that conditional on buying any cigarettes, 28% of smokers buy more than 30 packs per month, 22% buy between 20 and 30 packs, 28% buy between 10 and 20 packs, and 22% purchase fewer than 10 packs. Conditional on monthly purchase quantities, the relative brand shares across the seven brand categories vary. Notably, Marlboro captures a substantial share at all levels of consumption. This implies that Marlboro is the dominant brand among both casual and regular smokers.
Web Appendix W4 presents data related to brand loyalty and switching. If we define the category in terms of the seven brands and brand categories, Web Appendix W4a shows that over the five years of the data window, 30% of smokers stick with one brand, 28% have purchased only two brands, and about 41% have purchased more than two brands.
Web Appendix W4b explores whether multiple brand purchases occur within a month or over time. In 89% of months, smokers purchase only one out of seven cigarette brands. In the other 11% of months, they purchase two or more cigarette brands. Therefore, although the majority of brand switching happens over time, multibrand purchasing within a month is still meaningful. Web Appendix W4c shows that 60% of smokers engage in multiple brand purchases within a month at some point over the observation window. This pattern suggests a need for our demand model to accommodate multiple brand purchase and quantity decisions within a decision period.
Our investigation's critical interventions are countermarketing tactics such as cigarette excise taxes, smoke-free restrictions, and antismoking advertising. Figure 1 shows the evolution of cigarette purchases for the sample of 422 single smokers and the three countermarketing programs' levels over time. Specifically, Figure 1, Panel A, shows that cigarette consumption declines over time for the sample. The average monthly purchase quantity drops from 20 packs in 2006 to 10 packs in 2010.
Graph: Figure 1. Purchased quantities and antismoking techniques over time.Notes: Panel A illustrates the unconditional number of packs per month. In Panel C, smoke-free restrictions are enforced in part or all of the eight locations including restaurants, bars, hospitals, private workplaces, government workplaces, grocery stores, hotels, and motels.
Cigarette excise taxes are from the "Tax Burden on Tobacco" report ([43]), which collects detailed information on federal, state, and local tax rates and effective dates. Figure 1, Panel B, and Table 3 show the evolution of the taxes faced by panelists. The jump in taxes during 2009 is from an increase in the federal tax from $.39 to $1.01 per pack in April 2009. The other changes in taxes are due to changes in state and local taxes.
Graph
Table 3. Antismoking Techniques Summary.
| Variable | Mean | SD | Min | Max | N |
|---|
| Tax per pack ($) | 1.762 | .902 | .460 | 6.860 | 25,320 |
| Federal tax per pack ($) | .607 | .296 | .390 | 1.010 | 25,320 |
| State tax per pack ($) | 1.118 | .728 | .070 | 4.350 | 25,320 |
| County tax per pack ($) | .032 | .225 | 0 | 2.000 | 25,320 |
| City tax per pack ($) | .004 | .053 | 0 | .68 | 25,320 |
| Smoke-free restriction level | 3.702 | 3.573 | 0 | 8 | 25,320 |
| Antismoking ads ($) | 535,932 | 1,331,248 | 0 | 5,191,064 | 25,320 |
40022242921994570 Notes: Smoke-free restrictions are enforced in part or all of the eight locations (restaurants, bars, hospitals, private workplaces, government workplaces, grocery stores, hotels, and motels).
To measure smoke-free restrictions, we collected smoke-free air policy information for eight common venues defined as restaurants, bars, hospitals, private workplaces, government workplaces, grocery stores, hotels, and motels from the CDC's state tracking studies. In each venue, smoke-free restrictions are assigned one of two values: 0 for no restriction and 1 for a complete restriction. We sum the number of smoke-free restrictions in the eight venues to describe a state's smoke-free restriction level. Figure 1, Panel C, shows the evolution of smoke-free restrictions. Smoke-free restrictions dramatically increased between 2006 and 2008.
We also obtained the U.S. monthly spending on antismoking campaigns from Kantar Media. Figure 1, Panel D, shows nationwide monthly spending on antismoking advertising. This figure highlights a significant antismoking advertising campaign that accompanied the federal tax hike in 2009. Overall, expenditures on antismoking ads averaged $535,932 per month over the observation period.
We use zip codes to match the taxes and smoke-free restrictions to each smoker. For simplicity, we assume that a smoker purchases only from stores located in the same state where they live and that match the federal, state, and local cigarette excise taxes, respectively. Smoke-free restrictions are matched to each smoker on the basis of the state where they live. The 422 single smokers in our estimation sample cover 46 states. There is substantial variation in the two tactics across states and over time. At the start of our observation window, the tax per package varied from a low of $.46 in South Carolina to a high of $3.30 in New York. At the end of 2010, tax per package varied from $1.18 in Missouri to $6.86 in New York. In terms of smoke-free restrictions, there were relatively few restrictions at the start of the data window. However, by the end of 2010, 19 of the 46 states had complete smoking bans in all eight venues.
Model-free evidence related to the relationship between countermarketing and branding is also possible. For example, based on the Nielsen retailer scanner data, the figure in Web Appendix W5 shows that every state increased its per pack tax rates over the five-year observation window and that the market share of Marlboro also increased in almost all states over the five years. For example, in South Carolina, Marlboro's market share increased by 51% as the state per pack tax rates increased by $4.47. Web Appendix W5 also plots the distribution of Marlboro's market shares by year and shows increases over time in response to greater taxes. These analyses suggest asymmetric effects of antismoking techniques across brands, with the market leader consistently gaining share as taxes increase.[ 9]
We use a multiple discrete-continuous choice model ([12], [13]; [50]; [54]) to model smokers' monthly cigarette purchase-quantity decisions within the seven brand categories. The model provides a parsimonious structural approach for investigating the purchase of multiple combinations of brand choices and quantities. The analysis quantifies the effect of the three antismoking techniques on brand choices and purchase quantities while allowing for the possibility of asymmetric effects across brands.
As there are J brands in the choice set, a smoker i can xijt of brand j in period t. We drop the subscripts i and t and specify the monthly cigarette purchase utility to a smoker as the sum of the utilities obtained from purchasing xj packs of each cigarette brand as
U(x)=ϕ1ln x1+∑j=28γjϕj{ln(xjγj+1)},1
where U(x) is a quasiconcave, increasing, and continuously differentiable function with respect to the purchase quantity vector x ( for all j). The first good x1 is the outside good that is always consumed. Term ϕj is the baseline marginal utility that represents the utility of choosing brand j at the point of zero purchase:
∂U(x)∂xj=ϕj(xjγj+1)−1.2
The marginal rate of substitution between any brand k and l at the point of zero purchase of both goods is . We parameterize smoker i's baseline utility of purchasing brand j in period t as
ϕijt=exp[β0i+Brandjβ1ij+β2ij×Brandj×SFit+β3ij×Brandj×ln(1+AntiAdSt)+β4iBrandLoyalijt+∑tYeartβ5,t+∊ijt],3
where the Brandj terms are brand dummies, and β1ij is smoker i's intrinsic preference for brand j. The term SFit represents the level of smoke-free restrictions faced by smoker i in period t. We allow the impact of smoke-free restrictions to vary across brands (β2ij). The term AntiAdS t denotes the antismoking national advertising stock that a smoker is exposed to in period t. We let the antismoking advertising stock evolve as .[10] The effect of antismoking advertising also potentially varies across brands (β3ij). The term BrandLoyalijt is a dummy variable indicating purchase or not of brand j in the last period. The coefficient of BrandLoyalijt represents brand choice state-dependence. We also include year fixed effects in the baseline utility to account for any trend in smoking. The term ∊ijt is an extreme value distributed error term that is i.i.d. with a scale parameter of σ. The baseline utility for the outside good ϕi0t is normalized to 1.
Term γ in Equation 2 is a satiation (or translation) parameter.[11] A larger γ value indicates a stronger preference (or lower satiation) for cigarette brand j. All else equal, smokers would purchase more of brand j if γ is larger. Including a satiation parameter specific to brand j allows for the model to yield corner solutions where only brand j is chosen. We parameterize the satiation term as . The satiation parameter is brand-specific and is also a function of past cigarette consumption and addiction. We formulate a cigarette addiction stock as , where the addiction decay ρ2 is evaluated with a grid search of the.1 intervals from 0 to 1. This formulation is consistent with [ 9] theoretical model of addiction behaviors, where past consumption of addictive goods such as cigarettes increases the desire for present consumption. In our specification, we include the past consumption stock in the satiation parameter. A positive value for δ2i indicates that past consumption of cigarette brand j increases the marginal utility of consuming an additional package of brand j.
In each period, a smoker chooses an optimal set of cigarette purchase quantities x over the J brands, which solves the following Lagrangian condition. We drop subscripts i and t in the following equation:
L=ϕ1lnx1+∑j=28γjϕj{ln(xjγj+1)}−λ(∑j=18xjPj−E),4
where λ is the Lagrangian multiplier associated with the budget constraint E, and Pj is the tax-inclusive cigarette price per pack of brand j. The Khun–Tucker first-order conditions for optimal purchase quantities can be derived as follows:
∊j=V1−Vj+∊1 if xj*>05
∊j<V1−Vj+∊1 if xj*=0.6
Thus, the indirect utility Vj is written as
Vj=Zj′β−ln(xj*γj+1)−lnPjandV1=−lnx1*,7
where indicates the deterministic component in the baseline marginal utility in Equation 3. Note that ∊j is an extreme value distribution error term that is independently distributed across brands with a scale parameter of σ. Thus, the scale parameter σ is the negative inverse of the price coefficient of lnPj.
All the parameters in the baseline marginal utility and satiation terms are random coefficients that follow a multivariate normal distribution where . Conditional on θi, the probability that a smoker i purchases a nonzero quantity of cigarette packs of M of the J brands and zero cigarette packs of the remaining J-M brands in period t is
Li(θi,xit1*,xit2*,...,xitM*, 0,..., 0)=∫|J|×∏l=2Mg(V1−Vl+∊1σ)×∏s=M+1JG(V1−Vs+∊1σ)d∊1.8
The probability combines the integrals capturing a combination of extreme value density functions g(.) for the nonzero purchased brands and extreme value cumulative distribution functions G(.) for the zero purchased brands. We follow [12] and show in Web Appendix W7 that conditional on θi the multiple discrete-continuous probability in Equation 8 has the closed form in Equation 9:
Li(θi,xit1*,...,xitM*, 0,..., 0)=1Pit1 1σM−1[∏l=1M1xitl*+γitl][∑l=1MPitl(xitl*+γitl)][∏l=1MeVitl/σ(∑j=1JeVitj/σ)M](M−1)!,9
where for the first outside option. Letting , the unconditional probability that a smoker makes the observed sequence of choices over T periods is the integral over all values of random coefficients θ:
L(xi*)=∫θ∏t=1TLi(θ,xit*)dF(θ),10
where F is the multivariate cumulative normal distribution. The simulated maximum likelihood approach is used for estimation. We use a scrambled version of the Halton sequence to draw realizations of θ. Additional estimation details are in Web Appendix W8. We also show in Web Appendix W9 that the closed-form solution of this multiple-discreteness continuous model collapses to the multinomial choice model in the case when M = 1 (i.e., only one brand is chosen).
The model parameters are identified through variation in different aspects of the environment and consumer decision-making. The satiation parameters are identified from the observable differences in purchase quantities when only a single brand is purchased. [12] shows that the role of the satiation parameters is to allow for corner solutions where some optimal quantity of a single brand is purchased. Observations involving the purchases of two brands provide variation that allows for the identification of the parameters in the baseline utility via the marginal rate of substitution between pairs of brands.
Year fixed effects control for any trends or cultural shifts outside the model that influence smoking rates. Controlling for external trends helps facilitate the identification of the effects of the countermarketing techniques. The relationship between cigarette purchase probabilities and temporal/cross-sectional variation in cigarette excise taxes identifies the price coefficient. The relationship between across-brand variation in choice probabilities and temporal/across-state variation in smoke-free restrictions identifies the brand-specific smoke-free parameters. The relationship between across-brand variation in choice probabilities and temporal variation in antismoking ads identifies the brand-specific antismoking coefficients.
Web Appendix 10 reports model comparisons for estimating the likelihood function using a baseline model and several extensions. We start with a baseline model that contains only brand intercepts, year dummies, cigarette prices, and satiation parameters. We then add each additional term discussed previously and examine the relative impact on model performance. The full model has the best performance in log-likelihood (−121,327), Akaike information criterion (242,792), and Bayesian information criterion (243,354). The remainder of this section focuses on the full model.
We report the estimation results in Table 4. The first set of parameters is the mean estimates of intrinsic brand preferences in the marginal utility. These parameters range from −6.765 for mainstream brand categories and −6.780 for Marlboro to −7.200 for Winston. The ratio of marginal utilities at the point of zero purchase is the marginal rate of substitution between any two brands. For example, the marginal rate of substitution between Marlboro and economy brands is . Controlling for other factors, higher marginal utility implies that a smoker can increase overall utility by consuming Marlboro rather than an economy brand at the point of zero purchases of both brands. In addition, there is little individual heterogeneity in the intrinsic preferences of the seven brands. For example, the standard deviation in the intrinsic brand preference for Marlboro is 1.337 compared to the mean Marlboro (−6.780).
Graph
Table 4. Estimation Results of Cigarette Brand Choice and Quantity Decisions.
| Variable | Estimate (SE) | SD (SE) |
|---|
| Intercept in the Baseline Utility ϕk | | |
| Marlboro | −6.780 (.066) | 1.337 (.079) |
| Basic | −7.108 (.076) | 1.139 (.091) |
| Winston | −7.200 (.082) | .830 (.107) |
| Virginia Slims | −6.861 (.074) | .714 (.097) |
| High-price tier | −6.769 (.069) | 1.310 (.083) |
| Mainstream tier | −6.765 (.064) | 1.565 (.076) |
| Low-price tier | −7.117 (.068) | 1.506 (.078) |
| Brand Loyalty in the Baseline Utility ϕk | | |
| Brand loyalty | 4.704 (.053) | 1.769 (.052) |
| Smoke-free restrictions in the baseline utility ϕk | | |
| Smoke-free on Marlboro | −.034 (.007) | .056 (.009) |
| Smoke-free on Basic | −.007 (.010) | .018 (.013) |
| Smoke-free on Winston | −.030 (.011) | .030 (.015) |
| Smoke-free on Virginia Slims | −.026 (.009) | .001 (.013) |
| Smoke-free on high-price tier | −.002 (.007) | .006 (.009) |
| Smoke-free on mainstream tier | −.022 (.005) | .020 (.008) |
| Smoke-free on low-price tier | −.020 (.007) | .001 (.009) |
| Antismoking Ads in the Baseline Utility ϕk | | |
| Antismoking ads on Marlboro | −.136 (.037) | .071 (.052) |
| Antismoking ads on Basic | −.078 (.049) | .002 (.063) |
| Antismoking ads on Winston | −.163 (.064) | .058 (.085) |
| Antismoking ads on Virginia Slims | −.031 (.053) | .048 (.072) |
| Antismoking ads on high-price tier | −.027 (.039) | .034 (.053) |
| Antismoking ads on mainstream tier | −.085 (.032) | .038 (.045) |
| Antismoking ads on low-price tier | −.080 (.037) | .047 (.049) |
| Year Dummies in the Baseline Utility ϕk | Yes | Yes |
| Satiation Parameters γk (reparam as exp(.))a | | |
| Intercept Marlboro | .081 (.077) | .639 (.082) |
| Intercept Basic | .158 (.093) | .720 (.104) |
| Intercept Winston | .360 (.108) | 1.262 (.134) |
| Intercept Virginia Slims | .237 (.100) | 1.234 (.124) |
| Intercept high-price tier | −.012 (.081) | .768 (.088) |
| Intercept mainstream tier | .060 (.075) | .625 (.080) |
| Intercept low-price tier | .288 (.081) | .702 (.087) |
| Category-specific addiction stock | .465 (.014) | .059 (.017) |
| Price Coefficient | | |
| Sigma (inverse of the coef. of – ln Price) | .844 (.007) | |
- 50022242921994570 a Satiation parameters suggest the satiation preference, with a larger number referring to a stronger preference (lower satiation). The carryover parameter is found to be.3 for advertising stock and.8 for addiction stock.
- 60022242921994570 Notes: We model the brand choice and purchase quantity decisions of seven brands and one outside good for 422 individuals over five years at monthly level. Estimates in bold are significant at the 5% level. The baseline utility is the marginal utility at the point of zero purchase. The carryover parameters are determined by grid search from.1 to 1 at.1 interval. Results with continuous time trend are provided in Web Appendix W11.
The brand-loyalty term has a significant positive coefficient (4.704). The model comparison illustrates that brand loyalty has a greater impact on model fit than other variables. In our context, brand loyalty is operationalized as a brand purchase dummy in the last period. The positive coefficient implies that brand choice in the previous period increases the marginal utility of purchase in the current period.
The next block of Table 4 reports estimates for the smoke-free restriction coefficients across brands. Smoke-free restrictions have significant negative effects on five of the seven brands. Interestingly, we find that smoke-free restrictions have the greatest influence on Marlboro purchases (−.034) relative to lower-priced brands such as Basic (−.007) or the economy brand category (−.020). As smoke-free restrictions become stricter and more prevalent, preference for Marlboro is significantly reduced compared with other brands. The smoke-free findings are especially interesting from public health and marketing perspectives. The literature on intrinsic versus image-related motivations provides a possible explanation for our finding that strong brands such as Marlboro are more susceptible to usage restrictions. Intrinsic motivations are derived from internally focused concerns such as economic costs or health fears ([48]). Image-related motivations, in contrast, describe the motivation to reduce smoking because of how one is perceived by others ([27]; [40]; [56]). Because they impact public consumption, smoke-free restrictions reduce this type of image motivation. The stronger brand–consumer relationship between Marlboro and smokers may make customers less responsive to intrinsically motivated countermarketing tactics such as tax hikes (price increase) but more responsive to smoking bans that impact image-related consumption.
The antismoking advertising results suggest that these education campaigns have relatively little effect on reducing cigarette consumption. The antismoking advertising coefficients are significant for four of the seven brands. We acknowledge that the observation window from 2006 to 2010 is not a period with intensive antismoking campaigns relative to previous time periods, such as 1998 to 2003. It is possible that earlier educational campaigns had already successfully educated the public about potential health hazards.[12]
Table 4 also includes satiation parameter estimates. Higher values of the satiation parameter mean that less satiation occurs when consumers purchase the brand. With less satiation, a smoker will purchase greater quantities. Because the model includes individual heterogeneity, the mean satiation parameter estimates need to be considered together with the estimated random coefficients of the satiation parameters. For example, Marlboro has a below-average mean satiation parameter estimate (1.084 = e.081). It also has a relatively large standard-deviation estimate (1.895 = e.639) as compared with the mean estimate (1.748 = 1.895/1.084). This pattern suggests that there is a great deal of individual heterogeneity in the satiation effect of Marlboro. This finding is consistent with the purchase quantity distributions in Web Appendix W3, where we observe sizable populations who consume small purchase quantities (i.e., 1–10 packs per month) and large purchase quantities (more than 30 packs). This implies that Marlboro has a large segment based on preference but also attracts a segment of more casual users. The cigarette addiction stock has a significant positive effect on the satiation parameters. It suggests that past cigarette consumption increases the marginal utility of consuming cigarettes ([ 9].
Finally, we consider the estimated coefficient of lnPrice, . While cigarette excise taxes are applied equally at the per pack level, the same tax hike may differentially impact brands because brand manufacturers and their retailers may vary in terms of their ability and willingness to pass through tax increases to consumers. We use weekly store pricing data for each brand at the 3,874 retailer stores in the Nielsen Retail Scanner data to analyze pass-through practices. We calculate pass-through by regressing the cigarette prices per pack of brand j in week t at store s on combined federal, state, and local taxes per pack as in Equation 11.
Pricejst=α1+Taxstα2+Brandj×Taxstα3j+αj+αs+αt+δjst.11
To determine brand-level pass-through, we interact the tax rate with the four individual brands and three brand categories. For estimation, we use Marlboro as the baseline. Brand, store, year, and weekly fixed effects are included. The coefficient α3j, associated with cigarette taxes per pack, measures the pass-through rate for brand j compared with the baseline Marlboro. A coefficient of 1.0 indicates full pass-through, whereby a $1 increase in cigarette taxes per pack will increase cigarette per-pack prices by $1. Table 5, Panel A, presents the pass-through regression results. The tax pass-through coefficient of Marlboro is.916. This is the lowest pass-through compared with the other cigarette brands. The coefficients of the interaction of tax and the other six brands indicate that the pass-through rates vary from approximately 94% to 110% across the seven brands. In descending order, the ranking of pass-through rates is Marlboro (91.6%), the economy brand category (94%), the mainstream brand category (95.4%), the premium brand category (96%), Winston (102%), Virginia Slims (108%), and Basic (109%). These pass-through estimates are broadly consistent with the range of the pass-through rates from 80% to 118% reported in the literature on cigarette excise taxes ([32]).
Graph
Table 5. Linear Regression Results of Pass-Through Rates.
| Variable | A: DV = Price ($) | B: DV = Log (Price) |
|---|
| Estimate | Robust SE | p-Value | Estimate | Robust SE | p-Value |
|---|
| Intercept | 2.133 | .005 | <.001 | 1.137 | .001 | <.001 |
| Basic | −.331 | .004 | <.001 | −.056 | .001 | <.001 |
| Virginia Slims | .291 | .005 | <.001 | .121 | .001 | <.001 |
| Winston | −.262 | .006 | <.001 | −.052 | .001 | <.001 |
| High price | .479 | .005 | <.001 | .144 | .001 | <.001 |
| Mainstream | −.114 | .004 | <.001 | −.024 | .001 | <.001 |
| Low price | −.638 | .005 | <.001 | −.205 | .001 | <.001 |
| Tax | .916 | .005 | <.001 | / | | |
| Tax × Basic | .172 | .003 | <.001 | / | | |
| Tax × Virginia Slims | .167 | .004 | <.001 | / | | |
| Tax × Winston | .106 | .003 | <.001 | / | | |
| Tax × High price | .046 | .004 | <.001 | / | | |
| Tax × Mainstream | .038 | .002 | <.001 | / | | |
| Tax × Low price | .024 | .003 | <.001 | / | | |
| Log tax | — | | | .312 | .001 | <.001 |
| Log tax × Basic | — | | | .092 | .001 | <.001 |
| Log tax × Winston | — | | | .066 | .001 | <.001 |
| Log tax × Virginia Slims | — | | | .020 | .001 | <.001 |
| Log tax × High price | — | | | −.030 | .001 | <.001 |
| Log tax × Mainstream | — | | | .027 | .001 | <.001 |
| Log tax × Low price | — | | | .099 | .001 | <.001 |
| Store fixed effects | Yes | | | Yes | | |
| Year fixed effects | Yes | | | Yes | | |
| Week fixed effects | Yes | | | Yes | | |
| Adjusted R-square | .9503 | | | .9445 | | |
| N observations | 7,050,680 | | | 7,050,680 | | |
70022242921994570 Notes: DV = dependent variable. The analyses are conducted at the brand-store-week level with seven brands in 3,874 stores over five years ( 7,050,680 observations). Marlboro is the omitted baseline brand. Standard errors are clustered at store-brand level. Panel A investigates dollar pass-through rates. The coefficient of cigarette taxes represents the dollar change in cigarette prices per one dollar change in cigarette excise taxes. Marlboro has the lowest pass-through rate compared with all other cigarette brands (92% vs. 94%–110%). Panel B investigates the pricing change percentage. The coefficient of logarithm tax represents the percentage change in cigarette prices per 1% increase in cigarette excise taxes. Premium brand categories and Marlboro have the lowest percentage increase in prices (28% and 31%) compared with other brands (33% to 41%).
The low pass-through of Marlboro versus the high pass-through of the higher-equity brands is an interesting finding. The economics literature includes multiple studies that consider tax pass-through rates. In general, the literature has found that pass-through is lower for products with lower demand elasticity ([29]). Lower demand elasticity is likely to be related to the brand equity construct. Although our finding of a lower-than-average pass-through for Marlboro is consistent with economic theory, the results for other brands do not seem to vary on the basis of brand equity. There may be a number of reasons for this, including the manufacturer's desire to capture optimal levels of revenue premium ([ 5]), the retailer's desire to maximize category contribution across its whole brand portfolio, and strategic considerations by both players.
Excise tax pass-through rates are likely determined by a combination of brand and retailer strategy. On the one hand, as tax hikes reduce overall category consumption, higher-equity brands may exploit the lower price sensitivity of consumers for their products ([ 1]) and raise prices. This approach would trade off increased margins with lost volume. On the other hand, for a dominant brand such as Marlboro, retailers may wish to minimize pass-through for the category leader to maintain overall volume. In this case, retailers may engage in promotions or reduce their own margins. These types of institutional considerations cannot be observed but are consistent with the observed industry trends.
Tax hikes also cause discrete jumps in prices that may create a sticker-shock effect. Sticker shocks occur when the relative price increase also influences consumer reaction ([11]). Table 5, Panel B, reports the results of a log-log specification of the pass-through analyses, revealing the percentage price increase caused by increased taxes for each brand. The results suggest that a 100% tax increase translates into a 31% price increase for Marlboro. Most of the other brands, with the exception of the premium category, have significantly higher percentage increases in price. A 100% tax increase leads to a 41% price increase for Basic, 40% for the economy category, 38% for Winston, 34% for the mainstream category, 33% for Virginia Slims, 31% for Marlboro, and 28% for the premium category. The higher relative price increase provides a possible explanation for why economy cigarette brands show the greatest loss of volume over time. In contrast, the relatively low markup for Marlboro might explain why Marlboro has higher volumes following the tax increase.
The preceding model results suggest that the effects of countermarketing tactics vary across brands. In this section, we conduct a set of simulations that evaluate the effects of alternative countermarketing tactics on a variety of usage and revenue metrics. These counterfactuals highlight the consequences of countermarketing activities on various category stakeholders and for different types of brands. The policy experiments involve five years of simulated monthly cigarette consumption decisions for 422 consumers with identical initial states as the estimation sample. We follow the simulation procedures proposed by [46], Equations 15 and 16). Details are provided in Web Appendix W12.
Table 6 reports the results of six countermarketing scenarios, including a 100% increase in tax rates, a shift toward maximum smoke-free restrictions in all states, and a doubling of antismoking advertising. We also evaluate more incremental scenarios including a 37.6% increase in taxes, a single additional smoke-free restriction, and a policy that employs the combination of a 37.6% tax increase and an additional smoke-free restriction. We assume that brands maintain the same level of pass-through estimated in the preceding section for the tax increase scenarios. For each of the six scenarios, we calculate changes in category volume, spending on cigarettes, quit rates, packs purchased per day for active smokers, and government tax revenues. We also calculate the sales impact on cigarette brands and brand price-tier categories.
Graph
Table 6. Counterfactual of the Three Antismoking Techniques.
| 100% Tax Increase | Max Smoke-Free Restriction | Double Antismoking Ads | 37.6% Tax Increase | One-Level Smoke-Free Increase | 37.6% Tax + One-Level Smoke-Free Increase |
|---|
| Smokers | | | | | | |
| Consumption vol. | −35.95% | −7.51% | −1.14% | −16.36% | −1.21% | −17.45% |
| Spending ($) | −13.78% | −6.58% | −1.52% | −5.39% | −1.07% | −6.48% |
| Packs a daya | −32.06% | −5.49% | −.81% | −14.74% | −.91% | −15.54% |
| Quit rateb ppt. | 10.65 | 3.25 | 1.86 | 3.95 | .48 | 4.49 |
| Quit rateb chg. | 29.92% | 9.17% | 5.21% | 11.08% | 1.35% | 12.61% |
| Government | | | | | | |
| Tax revenue ($) | 27.89% | −6.19% | −1.46% | 15.05% | −1.01% | 13.80% |
| Brand Sales | | | | | | |
| Marlboro | −33.08% | −14.03% | −2.76% | −14.67% | −2.20% | −16.68% |
| Basic | −43.28% | n.s. | n.s. | −20.10% | n.s. | −20.19% |
| Winston | −41.88% | −16.34% | −4.69% | −19.70% | −2.63% | −21.97% |
| Virginia Slims | −37.11% | −11.23% | n.s. | −16.89% | −1.92% | −18.64% |
| High-price | −27.70% | n.s. | n.s. | −12.06% | n.s. | −11.89% |
| Mainstream | −33.52% | −7.34% | −1.44% | −15.03% | −1.12% | −16.05% |
| Low-price | −42.68% | −8.53% | n.s. | −20.01% | −1.47% | −21.31% |
- 80022242921994560 a Packs a day before quitting calculates the average daily cigarette consumption in months when there are cigarette purchases.
- 90022242921994560 b Quit rate is defined as the percentage of smokers who quit within five years, and quitting is defined as no cigarette purchase during the final 12 months of the observation window.
- 100022242921994560 Notes: n.s. = nonsignificant at the 95% confidence interval. The table compares counterfactual results of seven policy simulations. The 37.6% tax increase corresponds to $.94 federal cigarette tax hike proposed in October 2014 by the Obama Administration. A table with confidence intervals on all the evaluation metrics can be found in Web Appendix W13.
The first scenario involves a 100% increase in taxes from the start of the observation window in 2006. This large increase in taxes results in a decrease in consumption volume of 36%, a decrease in spending of 14%, and a quit-rate increase of 30%. The consumption volume elasticity of −.36 is very close to the price elasticity of −.4 reported in previous meta-analyses ([17]). The simulation also suggests that a 100% tax increase results in a decrease in packs per day of nonquitting smokers of 32%. Despite the dramatic increase in tax rates, the reduced consumption results in just a 28% increase in tax revenues. In terms of the effects on brands, we find that Basic, Winston, and the lower-price-tier brands suffer greater losses in volume sales than do the high-price and mainstream tiers and Marlboro. These brand-level effects are driven by the pass-through policies described in the "Results" section.
The second scenario involves a shift toward complete smoke-free restrictions in all markets. In this case, category volume decreases by 7.5%, the quit rate increase is 9.1%, and packs per day for nonquitters drop by 5.5%. This reduction in usage combined with no increase in tax rates results in a loss of 6.2% in tax revenues. The effects on individual brands vary considerably. Marlboro, Winston, and Virginia Slims suffer losses of 14.0%, 16.3%, and 11.2%, respectively. Mainstream and low-price brands suffer approximately 8% losses in volume. In contrast, there is little impact on Basic and high-price-tier brands. Doubling antismoking advertising causes a 1% drop in category volume and a 5.2% increase in quit rate. The brand-level effects vary from negligible effects on Virginia Slims to a 5% decrease for Winston.
According to the [16] and the "Tax Burden on Tobacco" report ([43]), federal and state gross tax revenues have increased by approximately 20% from $273 to $330 million as combined federal and state taxes rates almost doubled from $1.34 in 2006 to $2.68 in 2016. Our policy simulation suggests that the tax revenues would increase by 28% for a 100% tax increase and decrease by 6% due to a complete smoke-free restriction. Combining the tax and smoke-free effects suggests a 22% change in revenues from a 100% tax increase and a complete nonsmoking policy. A comparison of this result with the CDC estimates suggests that our simulations are consistent with their findings.
Overall, the simulations reveal important elements of the category structure for consumers, regulators, and brands. For consumers, the tax increase results in a large decrease in usage and a substantial amount of quitting. However, nonquitting consumers are forced to pay considerably higher prices, which may have social welfare implications given the higher prevalence of smoking in low socioeconomic groups ([16]). Maximum smoke-free restrictions have a substantial but smaller impact on consumption and quitting. In contrast to the tax increase, the maximum smoke-free restrictions result in a loss of tax revenues. These two policies have very different effects across governmental and consumer stakeholders. Increased taxes and smoke-free restrictions both decrease consumption and increase quitting rates. However, taxes impose significant financial burdens on consumers while providing revenues to the government, whereas smoke-free restrictions lead to a loss in tax revenue but do not impose economic costs on consumers. In terms of brands, we find that stronger brands tend to gain share following tax hikes but lose share due to smoke-free policies. We do not find a readily identifiable pattern of the effects of antismoking advertising across brands.
Table 6 also includes results for simulations that involve smaller policy interventions. We examine a 37.6% tax increase, which corresponds to the $.94 federal cigarette tax hike proposed in October 2014 by the Obama Administration. A 37.6% tax increase results in a decrease in category consumption of approximately 16% and yields about a 11% increase in quit rate. Tax revenues increase by about 15%. This more moderate intervention yields 54% of the tax revenue from the draconian 100% tax increase. The quit rate for the 37.6% tax increase is about 37% of the rate of the 100% rate. The imposition of a single incremental smoke-free restriction reduces consumption by 1%, and the impact on tax revenues is 1%. As before, the critical policy decision is whether to place the burden on consumers through taxes or on the government through smoke-free restrictions. The pattern of brand effects is similar to the previous high-tax and high-restriction simulations. Again, taxes have relatively less effect on high-equity brands that tend to pass through less tax, whereas usage restrictions have a relatively greater effect on higher-market-share brands such as Marlboro, Virginia Slims, and Winston. The only relatively high-market-share brand that is not affected by restrictions is the budget brand Basic.
In Table 7, we compare the effects of the six countermarketing policies on heavy versus light smokers. For this comparison, we segment customers using a median split based on consumption volume in the initialization period. We find that across all policy simulations, antismoking activities have a larger impact on light compared with heavy smokers. Light smokers reduce volume of packs, decrease spending, and increase quit rates more than heavier smokers in response to countermarketing. This may reflect a higher level of addiction among heavier smokers.
Graph
Table 7. Counterfactual of the Three Antismoking Techniques by Heavy Versus Light Smokers.
| 100% Tax Increase | Max. Smoke-Free Restriction | Double Antismoking Ads | 37.6% Tax Increase | One-Level Smoke-Free Increase | 37.6% Tax + One-Level Smoke-Free Increase |
|---|
| Heavy Smokers | | | | | | |
| Consumption vol. | −34.73% | −6.59% | −1.27% | −15.73% | −1.08% | −16.72% |
| Spending ($) | −12.01% | −5.73% | −1.24% | −4.62% | −.95% | −5.60% |
| Packs a day | −33.58% | −6.36% | −1.00% | −15.42% | −1.06% | −16.38% |
| Quit rate ppt | 8.74 | 1.77 | 1.27 | 2.37 | .19 | 2.72 |
| Light Smokers | | | | | | |
| Consumption vol. | −39.62% | −10.27% | −1.94% | −18.24% | −1.58% | −19.63% |
| Spending ($) | −18.97% | −9.09% | −1.97% | −7.67% | −1.40% | −9.05% |
| Packs a day | −31.06% | −5.62% | n.s. | −14.35% | n.s. | −15.10% |
| Quit rate ppt | 12.23 | 4.47 | 2.35 | 5.25 | .72 | 5.94 |
110022242921994560 Notes: n.s. = nonsignificant at the 95% confidence interval. We group smokers into heavy and light smokers using the median split of cigarette purchase volume in the initialization period in 2005. The median consumption in 2005 is 220 packs a year. Table 7 reports only the mean counterfactual estimates for simplicity of interpretation. We provide a table with confidence intervals on all the evaluation metrics in Web Appendix W14.
Marketing and branding are powerful tools that can have both positive and negative effects on consumers and society. Marketing can provide value by connecting consumers to products that maximize individual utility, but marketing can also result in excessive or dangerous consumption. In the current research, we explore an underresearched but important aspect of marketing: the use of marketing tools to reduce the long-term consumption of dangerous branded products with negative health outcomes. There is an existing literature that has focused on techniques or "nudges" to change short-term behaviors ([53]). However, studies of short-term nudges often neglect important dynamic factors that may lessen one-time interventions' long-term impact.
Our research studies the effectiveness of marketing tools for disrupting consumption in a category that includes long-term consumption and high-equity brands. These aspects are important because purchase feedback effects, addiction, and brand loyalty may all have significant long-term consequences. Our research can be thought of as an application of attempting to reduce customer–brand bonding as we study interventions designed to disrupt and end consumer relationships with brands such as Marlboro. Disrupting brand relationships is an important area for researchers as there are numerous categories dominated by well-known and well-loved brands where excessive consumption may be injurious. For example, many consumers have life-long relationships with brands in fast food, soda, or snack foods ([41]; [49]). Beyond these health-oriented categories, antirelationship marketing might be employed to reduce consumption potentially viewed as having negative social consequences such as climate harming travel or gambling.
As demarketing or countermarketing designed to disrupt brand–consumer relationships becomes more prevalent, it is important to consider how different types of countermarketing affects various category stakeholders. In the remainder of the article, we discuss our research's implications for policy makers, consumers, and manufacturers. As part of this discussion, we also acknowledge several limitations to our research and future research directions.
Policy makers may be guided by a substantial academic literature. Our research further informs policy makers through a comparison of the relative effectiveness of pricing interventions (taxes), public usage restrictions (smoke-free policies), and antismoking communications. In general, we find that the pricing mechanism is the most powerful regulatory technique. The pricing mechanism also provides benefits to the regulatory state through incremental tax revenues, though at a cost to category users.
However, a gap in the countermarketing literature is a lack of research that focuses on how brand strength may impact countermarketing efficacy. This omission is a significant gap, given the importance of branding to consumers and its potential to attenuate social marketing initiatives. Because brands may serve a variety of purposes, such as guaranteeing quality or providing status, it is critical that policy makers consider branding issues. In this category, we find that the market leader is relatively less affected by increased taxes and more affected by policies that limit public consumption.
Our results could be interpreted as evidence that placing constraints on visual branding elements might be useful for reducing the consumption of unhealthy foods or beverages. These effects are likely to be especially pronounced for market leaders. The key implications for regulators and advocacy groups are that different instruments vary in terms of effectiveness and that the effects are heterogeneous across brand tiers. These heterogeneous effects and different levels of efficacy should be considered when designing or campaigning for regulatory interventions.
The effects of countermarketing on consumer welfare are complex. Different countermarketing tools change consumer behavior through different mechanisms. For example, taxes impose direct costs on consumption. Taxes, therefore, reduce consumer utility of consumption and increase costs. For example, we project that a 100% tax increase decreases consumption by 36% while lowering expenditures by only 14%. However, consumer utility is also indirectly affected by a reduction in health risks. From the consumer's perspective, tax-based countermarketing may be viewed as extremely negative. The consumer directly observes high costs and diminished consumption, whereas health benefits are more abstract or future-oriented.
In contrast, usage restrictions reduce consumption but do not impose additional costs. We find that a complete smoke-free policy results in a 7.5% drop in consumption and a 6.6% drop in spending. Smoke-free restrictions have a more moderate effect on usage and quitting but reduce tax collections. In the case of usage restrictions, the costs are mainly related to reduced convenience and the ability of consumers to use brands for social signaling. Again, the positive consequences for health are not directly observed by consumers.
Educational advertising to counter smoking appears to have little impact on consumers but also does not impose any direct costs on them. It is possible that antismoking advertising imposes a cost on smokers by stigmatizing the consumption of cigarettes. We should note that our observation window follows decades of educational messaging. Our results may indicate a diminishing return to antismoking advertising during the mature phase of countermarketing programs and may not be relevant to negative advertising in other vice categories. Furthermore, our findings related to the effects of antismoking advertising should also consider the realities of funding. While antismoking advertising does not directly impose costs on smokers, these campaigns are often funded by cigarette tax revenues.
The different countermarketing policies may also result in different changes to the category structure. If countermarketing operates differentially across segments, then countermarketing costs and benefits may not be equally shared across consumer groups. For example, taxes seem to favor stronger brands, whereas smoke-free restrictions have a greater negative impact on these brands. The implication is that the impact on consumer welfare is segment specific. If consumer segments vary in terms of socioeconomic factors, then the costs and benefits of alternative countermarketing approaches will vary across demographic groups.
Our research also has implications for manufacturers and brand managers. In particular, our results suggest that brand strength may differentially lessen or increase the impact of different countermarketing activities. As categories are subjected to countermarketing, brand managers may devise strategies or lobbying efforts based on their category position.
We find that brand strength moderates the impact of increased category taxes through several mechanisms. First, if stronger brands are able to charge price premiums ([ 1]; [ 4]), then per-pack tax increases result in a lower percentage-price increase. Second, stronger brands with larger market shares or distribution advantages may be better able to absorb some tax increases. For example, we find that Marlboro has the highest market share, the largest number of stockkeeping units, and some of the lowest tax pass-through rates. However, we also find evidence that strong cigarette brands are more susceptible to smoke-free policies. This finding is intuitive because smoke-free policies reduce consumers' ability to gain status from the consumption of strong brands.
While lessons learned from tobacco control efforts are also likely to be used when designing countermarketing tactics in categories such as fast food or soda, our findings suggest that efforts to tax soda or fast food might result in increased market shares for high-equity brands. However, this is speculation, and additional research is needed. It should also be noted that these categories are viewed differently from tobacco by most consumers. While tobacco is almost universally viewed as dangerous, opinions about fast food and soda are more diverse. In particular, many consumers believe that these categories are only harmful when they are consumed excessively.
Categories now targeted by antiobesity groups include powerful brands such as McDonald's and Coca-Cola. In response, Coca-Cola has launched antiobesity ads and argued that it is unfair to blame any single brand. Public relations responses are just one option for brands facing countermarketing. Our research suggests that different tactics are appropriate for different types of brands. Relationships between consumers and relatively weak brands may be disrupted using taxes, while for strong brands, the appropriate tactic seems to be usage restrictions that limit public consumption. Our results suggest that brand building is the correct response to taxes, whereas usage restrictions would call for other responses, such as lobbying.
While our results highlight the role of countermarketing in the cigarette category, we do note that this category has evolved in a unique manner. Specifically, product innovation in the form of e-cigarettes has had a significant impact on the category ([57]; [19]). We selected our data-collection window partly to minimize the impact of the diffusion of electronic cigarettes on traditional cigarettes. Interestingly, public health organizations have begun to use countermarketing tactics to limit electronic cigarettes and other vaping devices ([28]).
In terms of limitations, our unit of analysis is highly aggregated, as we focus on monthly quantity choices. It may be useful to analyze more granular weekly purchasing quantities to understand the role of stockpiling. Another caveat is that antismoking policies might also affect the propensity for consumers to scan products if countermarketing makes tobacco products less socially acceptable. Specifically, these policies might lead to more or fewer single-pack purchases that would be consumed before returning home. Another limitation of our analysis is that we focus only on the behavior of existing smokers. Countermarketing may also be used to prevent the adoption of vice products by nonusers. For example, the use of tactics that eliminate the symbolic benefits associated with brands might also reduce the entry of new smokers into the category. If antibranding efforts are enacted, then young nonsmokers may be less likely to develop the "positive associations" with brands such as Marlboro.
In addition, our selection and categorization of brands were based on pricing. It is possible that brands may exhibit asymmetric responses to countermarketing activities based on other branding elements. For example, researchers could conceivably examine how visual elements of branding such as color, font size, or the use of iconography influence consumer response to countermarketing. The success of the Australian government's ban on branding elements suggests that this type of research might yield valuable insights ([15]).
There are also additional opportunities for more dynamically oriented research. For instance, a dynamic structural model that considers both brand choice and purchase quantity decisions would allow researchers to examine how smokers trade off long-term financial costs and health concerns. We also acknowledge that without a dynamic structural model, our policy simulations are subject to the [37] critique.
This discussion of using a dynamic structural model of consumer choice also raises the issue of long-term dynamics of a category with significant countermarketing activity. As we have noted, consumers may act in a dynamically optimal manner and trade off health, money, and smoking utility. It is also possible that other category stakeholders may adopt strategic or forward-looking approaches. For instance, our results suggest that tax policies may lead to greater market power for the most dominant brands. This prediction is consistent with the observed growth in market share for Marlboro. If this is the case, then policy makers and brands may also wish to adopt dynamic forward looking strategies.
Policy makers may benefit from using a dynamic strategy that begins with tax hikes as the most effective tool for decreasing demand. At some point, it may then be useful to switch to policies that disrupt public consumption, such as smoke-free restrictions or plain packaging. Researchers should investigate the question of the optimal sequence of educational initiatives, taxes, and branding restrictions.
Brands may also wish to develop dynamic strategies when their categories are targeted by critics. In the case of tobacco, the industry faced a massive antimarketing campaign that resulted in a cultural shift against smoking. When a category is targeted, and public sentiment toward a category becomes negative, firms may benefit from knowing how to respond to different stages of a countermarketing lifecycle.
Finally, it should be noted that public health advocates and other regulatory agencies are not infallible. The use of countermarketing techniques to disrupt an industry is usually portrayed as a positive social influence. However, the rights and wrongs of social movements are often more complex. Countermarketing encompasses both activities that aim to persuade (e.g., educational advertising) and activities that aim to limit a brand's communications (e.g., advertising prohibitions, plain packaging mandates). Limitations on communications are a challenging issue, as prohibitions on branding are constraints on speech. There are those who would question the extent to which the state in general and regulators in particular should arbitrate what is good for individuals. Our goal in the current research was to measure the effectiveness of a variety of countermarketing tools. However, as countermarketing becomes more prevalent, legal and marketing research should also consider the boundary between marketing and protected speech.
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921994566 - Investigating the Effects of Excise Taxes, Public Usage Restrictions, and Antismoking Ads Across Cigarette Brands
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921994566 for Investigating the Effects of Excise Taxes, Public Usage Restrictions, and Antismoking Ads Across Cigarette Brands by Yanwen Wang, Michael Lewis and Vishal Singh in Journal of Marketing
Footnotes 1 Online supplement: https://doi.org/10.1177/0022242921994566
2 Researcher(s) own analyses calculated (or derived) based in part on data from Nielsen Consumer LLC and marketing databases provided through the NielsenIQ Datasets at the Kilts Center for Marketing Data Center at The University of Chicago Booth School of Business. The conclusions drawn from the NielsenIQ data are those of the researcher(s) and do not reflect the views of NielsenIQ. NielsenIQ is not responsible for, had no role in, and was not involved in analyzing and preparing the results reported herein.
3 Pradeep Chintagunta
4 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
5 The author(s) received no financial support for the research, authorship, and/or publication of this article.
6 Michael Lewis https://orcid.org/0000-0002-1599-5114
7 We chose the observation window 2005–2010 to include the significant federal tax hike in April 2009. It is also a period when electric cigarettes were yet to be a significant part of the nicotine delivery market.
8 The only difference is that our estimation sample has a larger proportion of senior panelists older than 65 years.
9 In Web Appendix W6 we further examine situations where smokers encountered changes in taxation rates and smoking restrictions due to migration across states. The migration sample is different from the estimation panel as we do not require a six-year panel. Instead, we only require the smokers to be in the Nielsen consumer panel one year before and after the move. The migration sample shows that when taxes increase, there is a shift toward high-price-tier brands, while an increase in smoke-free restrictions results in a decrease in the consumption of Marlboro.
The carryover parameters are determined by comparing log-likelihood from.1 to 1 at a.1 interval. We use the 2000–2005 period as the initialization period to create the antiadvertising stock.
Note that [12], [13]) refers to this parameter as the "translation parameter," whereas we refer to it as the "satiation parameter."
There is limited variation in advertising expenditures during our observation window. As such, our ability to precisely identify the effect of antismoking advertising is limited.
References Aaker David A. (1996), "Measuring Brand Equity Across Products and Markets," California Management Review, 38 (3), 102–20.
Aaker Jennifer L. (1997), "Dimensions of Brand Personality," Journal of Marketing Research, 34 (3) 347–56.
Adda Jerome, Cornaglia Francesca. (2010), "The Effect of Bans and Taxes on Passive Smoking," American Economic Journal: Applied Economics, 2 (1), 1–32.
Agarwal Manoj K., Rao Vithala. (1996), "An Empirical Comparison of Consumer-Based Measures of Brand Equity," Marketing Letters, 7 (3), 237–47.
Ailawadi Kusum L., Lehmann Donald R., Neslin Scott A. (2003), "Revenue Premium as an Outcome Measure of Brand Equity," Journal of Marketing, 67 (4), 1–17.
Andrews Craig J., Netemeyer Richard G., Burton Scot, Moberg Paul D., Christiansen Ann. (2004), "Understanding Adolescent Intentions to Smoke: An Examination of Relationships Among Social Influence, Prior Trial Behavior, and Antitobacco Campaign Advertising," Journal of Marketing, 68 (3), 110–23.
Bagwell Laurie Simon, Bernheim Douglas B. (1996), "Veblen Effects in a Theory of Conspicuous Consumption," American Economic Review, 86 (3), 349–73.
Balmford James, Borland Ron J., Yong Hua-Hie. (2016), "Impact of the Introduction of Standardised Packaging on Smokers' Brand Awareness and Identification in Australia," Drug and Alcohol Review, 35 (1), 102–09.
Becker Gary S., Murphy Kevin M. (1988), "A Theory of Rational Addiction," Journal of Political Economy, 96 (4), 675–700.
Becker Gary S., Grossman Michael, Murphy Kevin M. (1994), "An Empirical Analysis of Cigarette Addiction," American Economic Review, 84 (3), 396–418.
Bell David R., Lattin James M. (2000), "Looking for Loss Aversion in Scanner Panel Data: The Confounding Effect of Price Response Heterogeneity," Marketing Science, 19 (2), 185–200.
Bhat Chandra R. (2008), "The Multiple Discrete-Continuous Extreme Value (MDCEV) Model: Role of Utility Function Parameters, Identification Considerations, and Model Extensions," Transportation Research Part B, 42 (3), 274–303.
Bhat Chandra R. (2018), "A New Flexible Multiple Discrete-Continuous Extreme Value (MDCEV) Choice Model," Transportation Research Part B, 110, 261–79.
Bitler Marianne P., Carpenter Christopher S., Zavodny Madeline. (2010), "Effects of Venue-Specific State Clean Indoor Air Laws on Smoking-Related Outcomes," Health Economics, 19 (12), 1425–40.
Bonfrer Andre, Chintagunta Pradeep K., Corkindale David, Roberts John H. (2020), "Assessing the Sales Impact of Plain Packaging Regulation for Cigarettes: Evidence from Australia," Marketing Science, 39 (1), 234–52.
CDC (2014, 2017), "Fast Facts,"(accessed February 12, 2021), https://www.cdc.gov/tobacco/data%5fstatistics/fact%5fsheets/fast%5ffacts/index.htm.
Chaloupka Frank. (1991), "Rational Addictive Behavior and Cigarette Smoking," Journal of Political Economy, 99 (4), 722–42.
Chaloupka Frank, Warner Kenneth E. (2000), "The Economics of Smoking," in Handbook of Health Economics, Vol. 1b, Culyer Anthony J., Newhouse Joseph P., eds. Amsterdam: Elsevier Science, 1539–1726.
Chen Jiajie, Rao Vithala R. (2020), "A Dynamic Model of Rational Addiction with Stockpiling and Learning: An Empirical Examination of E-cigarettes," Management Science,66(12), 5886–905.
Chen Tao, Sun Baohong, Singh Vishal. (2009), "An Empirical Investigation of the Dynamic Effect of Marlboro's Permanent Pricing Shift," Marketing Science, 28 (4), 740–58.
Chetty Raj, Looney Adam, Kroft Kory. (2009), "Salience and Taxation: Theory and Evidence," American Economic Review, 99 (4), 1145–77.
Conlon Christopher T., Rao Nirupama I. (2020), "Discrete Prices and the Incidence and Efficiency of Excise Taxes," American Economic Journal: Economic Policy, 12 (4), 111–43.
Coppejans Mark, Gilleskie Donna, Sieg Holger, Strumpf Koleman. (2007), "Consumer Demand Under Price Uncertainty: Empirical Evidence from the Market for Cigarettes," Review of Economics and Statistics, 89 (3), 510–21.
Dunlop Sally M., Dobbins Timothy, Young Jane M., Perez Donna, Currow David C. (2014), "Impact of Australia's Introduction of Tobacco Plain Packs on Adult Smokers' Pack-Related Perceptions and Responses: Results from a Continuous Tracking Survey," BMJ Open, 4 (12), https://bmjopen.bmj.com/content/4/12/e005836.
Evans William N., Farrelly Matthew C., Montgomery Edward. (1999), "Do Workplace Smoking Bans Reduce Smoking?" American Economic Review, 89 (4), 728–47.
Farrelly Matthew C., Duke Jennifer C., Davis Kevin C., Nonnemaker James M., Kamyab Kian, Willett Jeffrey G., et al. (2012), "Promotion of Smoking Cessation with Emotional and/or Graphic Antismoking Advertising," American Journal of Preventive Medicine, 43 (5), 475–82.
Fehr Ernst, Falk Armin. (2002), "Psychological Foundations of Incentives," European Economic Review, 46 (4–5), 687–724.
Fox Maggie. (2018), "FDA Launches New Anti-Vaping Campaign Aimed at Teens," NBC(September 18), https://www.nbcnews.com/health/health-news/fda-launches-anti-vaping-campaign-aimed-teens-n910691.
Fullerton Don, Metcalf Gilbert E. (2002), "Tax Incidence," working paper, National Bureau of Economic Research.
Goli Ali, Chintagunta Pradeep K. (2018), "Selling Smokes or Smoking Sales: Investigating the Consequences of Ending Tobacco Sales," working paper, University of Chicago.
Gordon Brett R., Sun Baohong. (2015), "A Dynamic Model of Rational Addiction: Evaluating Cigarette Taxes," Marketing Science, 34 (3), 452–70.
Harding Matthew, Leibtag Ephraim, Lovenheim Michael F. (2012), "The Heterogeneous Geographic and Socioeconomic Incidence of Cigarette Taxes: Evidence from Nielsen Homescan Data," American Economic Journal: Economic Policy, 4 (4), 169–98.
Holt Douglas B. (2002), "Why Do Brands Cause Trouble? A Dialectical Theory of Consumer Culture and Branding," Journal of Consumer Research, 29 (1), 70–90.
Keller Kevin L. (1993), "Conceptualizing, Measuring, and Managing Customer-Based Brand Equity," Journal of Marketing, 57 (1), 1–22.
Keller Kevin L., Lehmann Donald R. (2006), "Brands and Branding: Research Findings and Future Priorities," Marketing Science, 25 (6), 740–59.
Khan Romana, Misra Kanishka, Singh Vishal. (2016), "Will a Fat Tax Work," Marketing Science, 35 (1), 10–26.
Lucas Robert E. (1976), "Econometric Policy Evaluation: A Critique," Carnegie-Rochester Conference Series on Public Policy, 1 (1), 19–46.
Ma Yu, Ailawadi Kusum L., Grewal Dhruv. (2013), "Soda Versus Cereal and Sugar Versus Fat: Drivers of Healthful Food Intake and the Impact of Diabetes Diagnosis," Journal of Marketing, 77 (3), 101–20.
McAlexander James H., Schouten John W., Koenig Harold F. (2002), "Building Brand Community," Journal of Marketing, 66 (1), 38–54.
Moore Sarah, Dahl Darren W., Gorn Gerald J., Weinberg Charles B. (2006), "Coping with Condom Embarrassment," Psychology, Health & Medicine, 11 (1), 70–79.
Moorman Christine, Ferraro Rosellina, Huber Joel. (2012), "Unintended Nutrition Consequences: Firm Responses to the Nutrition Labeling and Education Act," Marketing Science, 31 (5), 717–37.
Muñiz Albert M., O'Guinn Thomas C. (2001), "Brand Community," Journal of Consumer Research, 27 (4), 412–32.
Orzechowski and Walker (2018), "Tax Burden on Tobacco, 1970–2018," report, https://chronicdata.cdc.gov/Policy/The-Tax-Burden-on-Tobacco-1970-2018/7nwe-3aj9/data.
Pechmann Cornelia, Shih Chuan-Fong. (1999), "Smoking Scenes in Movies and Antismoking Advertisements Before Movies: Effects on Youth," Journal of Marketing, 63 (3), 1–13.
Pechmann Cornelia, Zhao Guangzhi, Goldberg Marvin E., Reibling Ellen T. (2003), "What to Convey in Antismoking Advertisements for Adolescents: The Use of Protection Motivation Theory to Identify Effective Message Themes," Journal of Marketing, 67 (2), 1–18.
Pinjari Abdul R., Bhat Chandra. (2011), "Computationally Efficient Forecasting Procedures for Kuhn-Tucker Consumer Demand Model Systems: Application to Residential Energy Consumption Analysis," working paper, University of Texas at Austin.
Pollay Richard W., Siddarth S., Siegel Michael, Haddix Anne, Merritt Robert K., Giovino Gary A., et al. (1996), "The Last Straw? Cigarette Advertising and Realized Market Shares Among Youths and Adults, 1979–1993," Journal of Marketing, 60 (2), 1–16.
Ryan Richard M., Deci Edward L. (2000), "Intrinsic and Extrinsic Motivations: Classic Definitions and New Directions," Contemporary Educational Psychology, 25 (1), 54–67.
Sekar Raju, Rajagopal Priyali, Gilbride Timothy J. (2010), "Marketing Healthful Eating to Children: The Effectiveness of Incentives, Pledges, and Competitions," Journal of Marketing, 74 (3), 93–106.
Shriver Scott, Bollinger Bryan. (2017), "Consideration and Cannibalization Effects from Retail Entry: A Structural Analysis of Multi-Channel Demand," working paper, New York University.
Seiler Stephan, Tuchman Anna, Yao Song. (2021), "The Impact of Soda Taxes: Pass-Through, Tax Avoidance, and Nutritional Effects," Journal of Marketing Research, 58 (1), 22–49.
Talukdar Debabrata, Lindsey Charles. (2013), "To Buy or Not to Buy: Consumers' Demand Response Patterns for Healthy Versus Unhealthy Food," Journal of Marketing, 77 (2), 124–38.
Thaler Richard H., Sustein Cass R. (2008), "Nudge: Improving Decisions About Health, Wealth, and Happiness," Constitutional Political Economy, 19 (4), 356–60.
Thomassen Øyvind, Smith Howard, Seiler Stephan, Schiraldi Pasquale. (2017), "Multi-Category Competition and Market Power: A Model of Supermarket Pricing," American Economic Review, 107 (8), 2308–51.
Tobacco Today (2012), http://www.tobaccotoday.info/2012/05/23/cigarettes-brand-evolution/.
Toubia Olivier, Stephen Andrew T. (2013), "Intrinsic vs. Image-related Utility in Social Media: Why Do People Contribute Content to Twitter?" Marketing Science, 32 (3), 368–92.
Tuchman Anna. (2019), "Advertising and Demand for Addictive Goods: The Effects of E-Cigarette Advertising," Marketing Science, 38 (6), 913–1084.
Wakefield Melanie A., Hayes Linda, Durkin Sarah, Borland Ron. (2013), "Introduction Effects of the Australian Plain Packaging Policy on Adult Smokers: A Cross-Sectional Study," BMJ Open, 3 (7), 1–9.
Wakefield Melanie, Coomber Kerri, Zacher Meghan, Durkin Sarah, Brennan Emily, Scollo Michelle. (2015), "Australian Adult Smokers' Responses to Plain Packaging with Larger Graphic Health Warnings 1 Year After Implementation: Results from a National Cross-Sectional Tracking Survey," Tobacco Control, 24 (Suppl 2), ii17–ii25.
Wakefield Melanie, Durkin Sarah, Spittal Matthew J., Siahpush Mohammad, Scollo Michelle, Simpson Julie A., et al. (2008), "Impact of Tobacco Control Policies and Mass Media Campaigns on Monthly Adult Smoking Prevalence," American Journal of Public Health, 98 (8), 1443–50.
Wang Yanwen, Lewis Michael, Singh Vishal. (2016), "The Unintended Consequences of Countermarketing Strategies: How Particular Antismoking Measures May Shift Consumers to More Dangerous Cigarettes," Marketing Science, 35 (1), 55–72.
~~~~~~~~
By Yanwen Wang; Michael Lewis and Vishal Singh
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 77- Is Distance Really Dead in the Online World? The Moderating Role of Geographical Distance on the Effectiveness of Electronic Word of Mouth. By: Todri, Vilma; Adamopoulos, Panagiotis; Andrews, Michelle. Journal of Marketing. Jul2022, Vol. 86 Issue 4, p118-140. 23p. 6 Charts, 1 Graph, 1 Map. DOI: 10.1177/00222429211034414.
- Database:
- Business Source Complete
Is Distance Really Dead in the Online World? The Moderating Role of Geographical Distance on the Effectiveness of Electronic Word of Mouth
The authors investigate how the geographical distance between online users is associated with electronic word-of-mouth (eWOM) effectiveness. Their research leverages variation in the visibility of eWOM messages on the social media platform of Twitter to address the issue of correlated user behaviors and preferences. The study shows that the likelihood that followers who are exposed to users' WOM subsequently make purchases increases with followers' geographic proximity to the users. The authors propose social identification as a potential mechanism for why geographical distance still matters online in eWOM: because consumers may form a sense of social identity based on their physical location, information regarding the spatial proximity of users could trigger online social identification with others. The findings are robust to alternative methods and specifications, such as further controlling for latent user homophily by incorporating user characteristics and embeddings based on advanced machine-learning and deep-learning models and a corpus of 140 million messages. The authors also rule out several alternative explanations. The findings have important implications for platform design, content curation, and seeding and targeting strategies.
Keywords: electronic word of mouth; social media; geographical distance; deep learning
Electronic word of mouth (eWOM) plays a key role in shaping consumer attitudes and behaviors. More consumers digitally share information, research what others say about products and services, and rely on eWOM to gain knowledge or make decisions than ever before ([25]; [81]). While eWOM is becoming one of the most influential sources of information among consumers, the effectiveness of traditional communication channels employed by marketers is declining ([93]; [102]; [104]). These contrasting trajectories have motivated marketers to harness the power of eWOM. To do so, they are paying more attention to their user base and the corresponding eWOM episodes ([ 3], [ 4]; [49]). At the same time, technological advancements and data opportunities in the online space enable marketers to access data such as user-generated content and user location ([ 6]; [112]). As the availability of location data in particular grows, marketers wonder whether location plays an important role in eWOM effectiveness. Some marketers believe that consumers may consider eWOM from farther distances as more widely accepted and universal and, thus, more valuable and informative ([42]). Others believe that eWOM from closer distances may be perceived as more relatable and relevant and, thus, more trustworthy ([19]; [95]). Yet other marketers suppose that distance does not matter online and, thus, does not influence how consumers value eWOM ([90]).
The literature on the role of geographical distance in eWOM effectiveness is not decisive either. On the one hand, a common belief for the effectiveness of eWOM is that technology bridges the geographical distance between consumers. Information technology reduces communication costs by decoupling the interaction process from geographic constraints. For instance, mobile devices enable shopping and communication between consumers independent from location, while the internet allows instant access to message exchanges and marketplaces. These reduced communication and access costs have led some scholars to pronounce the "death of distance" ([29]) and "end of geography" ([52]).
On the other hand, geographical distance may still play a role in online consumer behavior and WOM for several reasons ([51]). These include location-specific goods as well as economic costs related to shipping, contracting, monitoring, enforcement, travel, and inconvenience ([ 2]; [59]). Local user preferences and spatially correlated social ties as well as limited regional availability of alternative product options can also lead to commonalities in product adoption among nearby consumers (e.g., [21]; [72]; [79]). These reasons suggest that geography may still constrain the effectiveness of eWOM.
Our research investigates whether geographical distance is significantly associated with the effect of eWOM, beyond the aforementioned explanations. Specifically, we ask whether the geographical distance between pairs of familiar disseminators and receivers of online WOM messages in social media plays a role in driving recipients' purchase behaviors or whether distance does not matter for eWOM beyond utilitarian reasons—such as transaction costs— and proxying for correlated user behaviors and preferences. We therefore examine this research question in an online setting where economic costs, such as those relating to contracting and travel, are not of concern.
To investigate whether geographical distance is associated with the effectiveness of eWOM, we use rich data from the social media platform of Twitter, where users disseminated messages about their real-world product purchases ([103]). Our main identification strategy leverages variation in the visibility of these eWOM messages—enabled by a unique feature of the platform—causing some purchase messages to be visible and others to be invisible to followers ([ 1]). We compare the purchase behaviors of followers for whom users' purchase messages were included in their newsfeeds (i.e., the stream of tweets presented on the home screen of a user from accounts the user follows on Twitter) with the behaviors of followers for whom users' purchase messages were not included in their newsfeeds due to this design feature, while employing an extensive set of controls (discussed in the "Econometric Model Identification" subsection). We also use the salience of user locations on Twitter to provide evidence for the potential mechanism. In our empirical setting, users are familiar with each other and thus aware of the location of their peers. Their location is often salient, as users may highlight this information in their profiles; user profile information is observable in peers' timelines when hovering with the cursor over the profile picture, the username, or the person's name.
We find that the relationship between eWOM and the likelihood that message recipients make a purchase strengthens as the geographical distance between disseminators and receivers decreases. This relationship is economically significant, managerially relevant, and robust to alternative methods and identification strategies. Social identity may explain why geographic proximity could increase the effectiveness of eWOM, beyond utilitarian reasons and proxying for correlated user behaviors, as consumers may form their social identities based in part on their geographic location ([44]; [83]; [107]; [108]). To provide evidence for this potential mechanism, we investigate, for instance, how the salience of geographic cues as well as conditions that strengthen the role of geographical location in the social identification process ([101]; [109]) further enhance the effectiveness of eWOM.
Our findings of geographical distance variation in eWOM effectiveness have important implications for both theory and practice. We contribute to the online WOM influence studies by demonstrating that despite technology's promise, spatial distance continues to play an important role in a digital world ([ 9]; [71]). Specifically, we show how geographic proximity is significantly related to an increase in eWOM effectiveness. Geographic constraints thus tether the impact of electronic communications. This finding also contributes to the literature that shows how features of disseminators and receivers can drive eWOM outcomes ([20]; [49]) by identifying geographic proximity as a relational characteristic that can affect dyadic influence. Our work thus provides actionable strategies for boosting the effectiveness of online interpersonal communications. Our findings also imply that customizing seeding and targeting with more proximate connections or highlighting information and cues associated with social identity formation can increase the effectiveness of online advertising, product recommendations, social advertising, referral programs, and other marketing strategies and tools by leveraging local appeals and social identification.
Several studies demonstrate the importance of eWOM, documenting how it can be a major driver of consumer behaviors. For example, user opinions have been found to influence the consumption of movies ([34]; [70]), television shows ([47]; [71]), video games ([114]), and books ([33]; [69]). Whereas these studies examine the direct impact of eWOM on purchase decisions, the present research extends this line of inquiry by focusing on conditions under which the effect of online WOM may be attenuated or accentuated.
Research has begun to investigate the contextual factors that impact the effectiveness of eWOM. These factors include characteristics of the product ([22]), brand ([73]), or WOM message itself ([85]). Scholars have also examined how certain features of disseminators (senders) and recipients shape the effect of eWOM. For instance, a sender's brand loyalty ([49]), expertise ([11]), and identity disclosure ([44]) can each affect the influence of WOM messages. Recipient characteristics such as the number of ties to adopters, demographics ([65]), and product experience ([87]) can also affect the influence of online communications. In addition, characteristics of the sender–recipient dyad, such as the tie strength ([13]; [20]), similarity across personality traits ([ 1]), and sociodemographic similarity ([45]), can also affect online WOM performance. We add to this stream of research by shedding light on an important factor that characterizes the pairwise relationship between senders and receivers of eWOM: the geographical distance between them.
Geographical distance has been shown to affect certain outcomes in multiple online and offline commerce settings. For example, geographical distance can affect online trade flows and volume due to costs related to shipping, contracting, monitoring, enforcement, travel, and inconvenience, as well as in cases of location-specific goods ([59]). Such transaction costs can engender a "home bias" that leads consumers to prefer transacting with nearby others ([ 7]; [59]; [68]). Similarly, geographical distance has been shown to correlate with product diffusion in the offline world due to imitation or direct observation, as physically close neighbors are more likely to adopt the same product ([21]; [36]; [43]).[ 5] In this study, however, we investigate how the geographic proximity between eWOM message disseminators and receivers accentuates or attenuates the effectiveness of online WOM, whereas the extant explanations for the impact of geography—albeit in different online settings—are not applicable to the context of eWOM; similarly, other explanations that do not apply to our eWOM setting relate to the location specificity of products, distribution networks, or local network externalities.
Recent studies hint at the potential role geographical distance may play in facilitating persuasion. For instance, lab studies find that when consumers have no identifying information about online reviewers, they assume that the reviewers are similar to them and so are as persuaded by them as they are by reviewers who appear similar to them, and more persuaded than by reviewers who appear dissimilar to them ([82]). When ambiguous reviewers appear less similar, consumers are also less persuaded by their boastful reviews ([86]). One of the many ways these studies manipulate cues to indicate reviewer similarity is to show that the reviewer appears to live in the same or a nearby city as lab participants. In instances of familiar peers, rather than ambiguous reviewers, field studies find that consumers are more persuaded the more recent or intense their relationships are with their peers ([13]; [32]). One of the ways [13] measure relationship recency is to use as a proxy whether peers currently live in the same town. Whereas the aforementioned studies focus on how ambiguity about, trust in, or relationship with peers affects persuasion in settings without financial transactions or product purchases, our research focuses on whether the geographical distance between senders and receivers of online WOM messages is significantly related to eWOM effectiveness.
Most relevant to our work are two studies examining whether geographical distance affects information diffusion and adoption. Specifically, [45] examine how geographic proximity, measured as contiguous relationships between states, and sociodemographic proximity, measured as similarity in demographics between states, affect overall message propagation at the state level. They find that geographic proximity has no impact when sociodemographic similarity is accounted for. Because of data limitations, their operationalization of geographic proximity may mask the actual effect of distance. Importantly, they conduct an aggregate state-level analysis, noting that the "state-level nature of the data is a limitation" (p. 249) and "access to more disaggregated data would allow for a more granular analysis" (p. 264). Their analysis also prevents them from observing whether consumers are familiar with the message propagator or the social distance between peers ([75]); familiarity and similarity between users as well as relationship strength have been shown to impact peer influence ([13]; [82]). Besides, diffusion may rely on a different mechanism than eWOM ([98]).
[79] is also very relevant to our work. These authors first conduct a descriptive study in the offline world and find that living nearby adopters of a cellular service provider is associated with faster switching to that provider. Because they do not directly observe WOM episodes, they conduct their analysis at an aggregate level, where it is not possible to account for homophily (i.e., the tendency of individuals to choose friends with similar tastes and preferences) or attribute WOM influence among the ties of each user ([75], [76]). In addition, they measure geographical distance as the average distance among all possible ties of each consumer.[ 6] These limitations motivate them to conduct scenario-based experiments, in which they investigate the mediating role of perceived homophily and show that geographic proximity to an ambiguous online reviewer increases the likelihood of following the reviewer's opinions, as geographic proximity is used as a cue for perceived similarity. However, the effects of geography are likely to be different in such lab settings due to the lack of familiarity and social ties with online reviewers, which play an important role in persuasion, as the extant literature shows ([13]; [82]; [86]). In addition, while consumers may rely on any available cues to address these concerns of ambiguity and lack of familiarity, lab settings often provide few such cues beyond geography to inform their decisions ([82]). Thus, it remains unclear whether geographic proximity can still play a role in facilitating eWOM influence in real-world settings, where users are familiar with and socially connected to each other and have access to an abundance of available cues.
In summary, we empirically investigate whether the geographical distance between individual pairs of familiar disseminators and receivers of organic eWOM in social media plays an important role in driving recipients' subsequent purchase behaviors of physical goods. Table 1 outlines the related studies and additional important differences of our research.
Graph
Table 1. Overview of Related Studies on the Role of Geography in eWOM.
| Study (Chronological Order) | Research Focus | Context | Setting | WOM Type | Social Ties | Familiarity with Sender | Geographical Distance Measure | Analysis Level | Notes | Main Findings |
|---|
| Naylor, Lamberton, and Norton (2011) | How ambiguous reviewers affect persuasion | Scenario-based | Online reviews | Indirect; active; artificial | None | No | Binary (same vs. different city) | Choice instance | Same vs. different city domicile is one of many reviewer variables manipulated to match that of a lab participant. | Consumers assume that reviewers with no identifying information have similar tastes as them, so they are similarly persuaded as they are by reviewers who appear similar to them and more persuaded than by reviewers who appear dissimilar. |
| Aral and Walker (2014) | How tie strength and embeddedness affect app adoption | Free Facebook app adoption | Social media platform (Facebook) | Indirect; passive (automated notifications without variation) | Observed | Familiarity with the subject (not sender) | Binary (same vs. different town) | Notification episode | The authors note that they "estimate how relationship-level covariates [same town] are correlated with the extent or impact of influence" (p. 1363), and the "results may have limited generalizability to ... cases where there is a significant financial cost to adopting a product" (p. 1366). | Consumers will adopt a free Facebook (movie review) app more quickly the more recent their relationship is and the more embedded they are with an existing user. |
| Packard, Gershoff, and Wooten (2016) | How boastful WOM affects persuasion | Scenario-based | Online reviews | Indirect; active; artificial | None | No | Binary (nearby vs. distant location) | Choice instance | Nearby vs. distant location is one of several variables used to manipulate how similar the reviewer appears to a lab participant. | Consumers are less persuaded by boasters' WOM when trust cues are low. |
| Chen, Van der Lans, and Phan (2017) | How relationship characteristics in a social network affect diffusion | Microfinance program adoption | Potential offline WOM episode | Unobserved | Surveys | Yes | None | Aggregate | WOM occurs in geographically bounded networks (e.g., village, university), so geographical distance is not material. | Social (economical) relationships are the most (least) important drivers of adoption for a microfinance program. |
| Message propagation | University network platform | Direct; active | Observed | Yes | None | WOM episode | Relationship intensity (e.g., message volume) is the most important driver of information diffusion in a network. |
| Fossen, Andrews, and Schweidel (2017) | How social vs. geographic proximity affect diffusion | Message propagation | Social media platform (Twitter) | Direct; active | Unobserved | Unobserved | Binary (contiguous vs. noncontiguous state) | Aggregate | The authors note that "the state-level nature of the data is a limitation" (p. 249). The study does not control for advertising or other marketing activities and focuses on diffusion. | Sociodemographic similarity propagates online message spread. Geographic proximity has no effect when accounting for sociodemographic similarity. |
| Meyners et al. (2017) | How geographic proximity affects adoption | Cellular service provider adoption | Potential offline WOM episode | Unobserved | Observed (aggregate) | Yes (aggregate) | Average distance from all peer adopters | Aggregate | The authors note their field study "did not have information on either the valence of the signals from the social network or the location of nonadopters" (p. 63), averages distance across adopters, does not observe WOM, does not control for network externalities, and studies switch behavior. The lab studies have no homophily cues beyond age and gender, and nofamiliarity with or social ties to reviewers. | Geographic proximity to adopters of a cellular provider is associated with faster switching to the provider. |
| Scenario-based | Online reviews | Indirect; active; artificial | None | No | Continuous (miles), categorical (state) | Choice instance | Geographic proximity to an ambiguous reviewer increases the likelihood of following the reviewer's recommendation due to perceived homophily. |
| Present study | How geographical distance affects eWOM effectiveness | Product purchases of significant financial cost | Social media platform (Twitter) | Direct; active | Observed | Yes | Continuous (miles) | WOM episode | Our study has information on nonadopters and social ties; controls for homophily, advertising, and message content; and studies actual product purchases of significant financial cost in a social network where receivers are familiar with the WOM sender. | Geographical distance is negatively associated with the likelihood that eWOM influences purchases, above and beyond other effects. |
Our empirical setting concerns a large-scale venture of American Express (the service provider) on the microblogging social media platform Twitter. This collaboration introduced a new purchasing service that was seamlessly integrated for two months into the social media platform (Twitter) to leverage users' connections to stimulate eWOM. Specifically, the service enabled consumers to make purchases on the social media platform while simultaneously spreading the word about their purchases to their social media peers (receivers).
We further discuss this novel service by describing the data-generating process. In particular, the service provider first posts a short message (tweet) on the platform broadcasting the list of participating merchants and the corresponding products available for purchase. This announcement includes information about the product offerings (e.g., product, respective sale price) and the designated hashtags (i.e., a phrase or word preceded by a hash sign [#]) consumers must use to make a purchase. Consumers must have a microblogging account and sync their service provider account with their microblogging account. Once the service provider broadcasts the products available for purchase, users can purchase these products by posting a tweet that includes the designated hashtag. In addition to the necessary hashtag, consumers can choose to personalize the purchasing tweets that are (automatically) shared with their social media peers. Typically, such messages are posted on the users' social media profiles, and their followers (i.e., those subscribed to their timeline) automatically receive these messages on their own (home) newsfeeds (for an explanation of how we use the variation in message visibility in our main identification strategy, see the "Econometric Model Identification" subsection and Figure A1 in the Web Appendix). The social commerce provider tracks the tweets containing the designated hashtag and pairs them with the corresponding product. After the purchase confirmation, the service provider bills the users and ships the product.
Our data set includes all the confirmed transactions generated through this purchasing process. Specifically, each transaction in our data set contains the message ID of the purchasing message, the message content, the designated product hashtag, the date and time the message was posted, the corresponding user ID, and whether the message was rendered visible or invisible to each of the user's followers (i.e., included or not included in the follower's newsfeed).
For each user, our data set also contains the screen name of the user on the social media platform, when they joined the platform, the set of the user's followers and followees, all the messages users have posted on the platform since they joined, the geolocation of the user, and the self-reported description of the user's profile. Our data set also contains the same information for users who chose not to make a purchase.
In addition, our data set contains information about the product offerings. The service provider collaborated with known retailers and offered various products for purchase (e.g., see Figure A2 in the Web Appendix). In particular, the products involve video game consoles and related accessories, electronics and sports equipment (e.g., high-definition tablets, sports and action cameras), general-purpose gift cards, and fashion accessories (e.g., designer bracelets, handbags). These offerings from the social commerce service provider were available for purchase at a reduced price only through the platform (about a 25% discount, yielding an average retail price of $125).
Overall, our data set tracks the corresponding purchasing decisions of 132,995 social media users on Twitter with 1.4% of the users purchasing available products. The users are located across the continental United States, as illustrated in Figure 1, and on average follow 996 Twitter users and have 342 followers. Table 2 presents the summary statistics and description of the main variables and Figure 2 shows the corresponding correlations.
Graph: Figure 1. Geolocations of users.
Graph: Figure 2. Correlation levels among the main variables of interest.
Graph
Table 2. Descriptive Statistics
| Variable | Description | Mean/Median | SD | Min | Max |
|---|
| Purchase | Whether the recipient of the message made a purchase | .014 | .12 | 0 | 1 |
| Visible message | Whether the message was visible to the recipient | .77 | .42 | 0 | 1 |
| Geographical distance | Geographical distance between sender and recipient | 971.08a | 894.90 | 0 | 5,585 |
| Number of followers | Number of followers of user | 342a | 101,000 | 0 | 376,000 |
| Number of followees | Number of followees of user | 996a | 12,629 | 0 | 115,000 |
| Reciprocal relationship | Whether the relationship between the users is reciprocal | .08 | .27 | 0 | 1 |
| Number of interactions | Number of interactions between users | .26 | 6.10 | 0 | 1,612 |
| Sentiment of message | Intensity of message advocacy | .21 | .35 | −1 | 1 |
| Personalized message | Whether the message was personalized by the sender | .82 | .38 | 0 | 1 |
1 aWe report the median instead of the mean value.
2 Notes: The values of the variables Number of followers, Number of followees, Reciprocal relationship, and Number of interactions correspond to the time of posting the WOM message.
We enhanced our data set with a proprietary data set from the ad intelligence company Kantar Media, which includes the local (and national) advertising expenditures of each brand and for each product. We also supplemented our data set with additional information from the American Community Survey five-year estimates of the U.S. Census Bureau regarding local demographics.
We model users' purchase decisions as a function of eWOM message, sender, and recipient as well as relationship characteristics, including observed and latent homophily controls; homophily creates a natural correlation in behaviors that could be incorrectly interpreted as a causal effect (e.g., [10]; [76]). To further control for any unobserved confounds, we use in our research design the variation in the visibility of eWOM messages.
Consistent with prior literature (e.g., [79]), we use a continuous-time single-failure survival model. In particular, we model how quickly users purchase a product, if any, using a Cox ([39]) proportional hazard model and correcting for censoring of transactions that might have been intended to occur after the observation window ([64]). Specifically, our main estimation equation for the decision of peer i (eWOM recipient) is as follows:
Graph
( 1)
where is the hazard of peer (follower) of consumer to purchase the same product as consumer after[ 7] an eWOM message from represents the baseline hazard, captures whether the eWOM message of consumer j (sender) was rendered visible to (i.e., included in the home newsfeed of) peer (recipient), and measures the physical distance of i from consumer j in (log) miles. The coefficient of interest is and captures the relationship between geographical distance and the effectiveness of eWOM, while the coefficient accounts for spatially and nonspatially correlated user behaviors and preferences (e.g., homophily) that would have otherwise manifested as the association between geographical distance and the effectiveness of WOM; put simply, our research design leverages the variation in the visibility of eWOM messages, as influence can occur if and only if the eWOM message is rendered visible, whereas correlated user behaviors and preferences are extant even when the eWOM message is nonvisible. We also control for user-relationship (sender–recipient dyad) characteristics, X; message characteristics, M; disseminator characteristics, D; and recipient characteristics, R; as well as product fixed effects and geography ( ) and time (day of message) fixed effects.
To construct the aforementioned user-relationship, message, and disseminator and recipient controls, we employed machine-learning techniques to leverage the vast amount of unstructured and structured data.
More specifically, the user-relationship (sender–recipient dyad) characteristics, , include controls for observed and latent pairwise user similarity and tie strength. This set of variables—in addition to the research design—allows us to capture the correlation in latent tastes and preferences between WOM message disseminators and recipients to better distinguish the relationship of interest. In particular, the user similarity between the disseminator and the recipient of a WOM message is measured based on ( 1) the similarity of topics discussed in social media posts using the results of a machine-learning model for natural language processing (NLP), as well as the overlap of the local communities as captured by the ( 2) Jaccard similarity coefficient of followers ([40]) and ( 3) Jaccard similarity coefficient of followees ([40]).
The NLP model we employ for this task is the latent Dirichlet allocation (LDA) model ([21]); LDA is a probabilistic generative NLP model that we use to model the user-generated content of each user in our data set (a document in our corpus) as a distribution over topics and every topic as a distribution over words in the English dictionary. We build this model on the corpus of all the messages of the users in our data set as the topics users discuss online reflect their latent interests ([113]); using the complete corpus instead of only the user messages during the observation window improves the inference of the NLP model. In particular, for the implementation of the LDA model, we used 139,850,033 messages in total. We also use a part-of-speech tagger/tokenizer developed specifically for Twitter ([84]) for more accurate tokenization, the removal of stopwords, symbols, typos and uninterpretable words ([97]), and the creation of ngrams, while retaining online-specific textual features (i.e., hashtags, at-mentions, and emoticons) ([92]; [110]). We also use [58] online variational Bayes algorithm to efficiently estimate the LDA model on our corpus. In addition, instead of arbitrarily determining the value of the LDA parameter corresponding to the number of latent topics, we find the natural number of topics present in our corpus using the procedure and measure proposed by [15], evaluated in terms of the Kullback–Leibler ([66]) divergence measure; nonetheless, the findings are not sensitive to the number of topics. Finally, regarding the hyperparameters of our model, we learn an asymmetric prior directly from our data. Beyond capturing the disseminator–recipient similarity with these metrics, we alternatively measure the similarity as a single standardized factor, based on the principal factors method, to avoid any potential multicollinearity; we present both sets of results.
In addition, our model specifications include constructs capturing whether the user relationship is reciprocal and (the log of) the number of interactions between the two users ([32]). Finally, we also control for the sociodemographic distance of the users—based on the difference in average age and percentages of male, Black or African American, Hispanic, and Asian-origin residents in the locations of the disseminator and recipient of the message using Census data at the zip code level ([45])—as well as the time zone difference. Overall, the various metrics of user pair similarity capture both observed similarity (e.g., the number of user interactions) and latent similarity (e.g., common latent interests) to further control for potentially unobserved confounds and homophily.
The message characteristics, , capture the sentiment of the message (i.e., intensity of WOM advocacy) as well as whether the message was personalized (i.e., explicit rather than implicit advocacy). The sentiment of the message (measured on a continuous scale between −1 and +1) provides a richer metric of the advocacy intensity of the sender compared with other simple metrics, such as lexicon-based scores. The main method we employed uses a publicly available commercial sentiment analysis mechanism based on deep learning ([ 9]). Nonetheless, the results are robust to employing alternative machine learning methods for sentiment analysis.[ 8] Moreover, the message controls also include whether a user account handle is mentioned in the message and whether the sender started the message with a period as the first character; starting a message with a period as the first character affects the visibility of the message and is a common norm among Twitter users when they want to explicitly make a message visible to all users and not only the account mentioned in the message.[ 9] Finally, we also control for advertising expenditures of each brand during our observation period in the local region of the eWOM message recipient expressed in (logarithm of thousands of) U.S. dollars.
The disseminator and recipient characteristics, and , include the user (opinion) leadership and expertise measures following the extant literature. We capture these disseminator and recipient characteristics to further control for factors that could bias the true relationship between geographical distance and WOM effectiveness due to omitted individual characteristics and preferences. The user expertise levels are measured on the basis of the standardized similarity of the timeline of a user with the corresponding timelines of the participating vendor and product employing the probabilistic NLP machine-learning model we previously described ([23]; [80]). The motivation for this measure is, for instance, that users who frequently tweet about technological trends and topics similar to those in the social media accounts of the specific product and the corresponding vendor are more likely to be perceived by their social media peers as experts in the area of technological products ([61]). In addition, we further control for correlated user preferences and interests ([26]; [72]) by including in the econometric specifications the latent interests based on the aforementioned LDA model ([ 1]). The user (opinion) leadership is measured based on the additive smoothed ratio ([74]) of followers to followees of the user. The additive smoothed ratio is frequently used in empirical studies to prevent this metric from being oversensitive to small-scale changes in the numbers of followees or followers ([ 1]; [41]). Furthermore, we also control for the number of followers of each user, whether the user has a default profile on the platform, and how many months have elapsed since the user joined the social media platform. Finally, we also control for the age, gender, and income of the users based on the Census data ([45]).
We next discuss the research design we use to further distinguish the effect of geographical distance on eWOM effectiveness from unobserved confounds and correlated user behaviors and preferences, such as homophily. Our research design is enabled by a unique feature of the platform that causes some WOM messages to be visible and other messages to be invisible (not included) on the newsfeeds of other peers, as described next.
Typically, a message posted by a user on the Twitter platform appears in the newsfeeds of all her followers, as the newsfeeds were not algorithmically curated during this venture. Thus, in our context, whenever a user makes a purchase, her social media peers are exposed to her advocacy toward the product if the purchase is visible in their (home) newsfeeds and their purchasing decisions might, in turn, be affected through such WOM episodes; the specific offers were available for purchase only through Twitter. The research design framework we employ leverages information on whether a message broadcasted on the platform was indeed rendered visible in the newsfeeds of other users or not (see Web Appendix Figure A1).
More specifically, on Twitter, a publicly broadcasted message may not be included in the newsfeeds of some followers in instances when a Twitter account handle (if any) appears in a specific position of the message (i.e., if any Twitter account user name following the "@" symbol appeared at the very beginning of the message). For instance, if a consumer purchases a product by posting the message "#BuyActionCamPack @AmericanExpress" (see Web Appendix Figure A2, Panel A), the message is visible on the newsfeeds of all her followers. However, if the same consumer purchases the same product by posting the message "@AmericanExpress #BuyActionCamPack" (see Web Appendix Figure A2, Panel B), then the message is visible only to the social media users who follow both this consumer and the mentioned account.
Interestingly, due to the design of Twitter during our observation window, if a consumer initialized the purchase process by clicking on the announcements of the product offerings or a post of another consumer, the handle of the corresponding account (e.g., "@AmericanExpress") is automatically prepopulated in the purchasing message. Then, the consumer must click anywhere on the prepopulated field and manually write the purchasing message that includes the required designated hashtag. Notably, if the click happens to be recorded on the left of the prepopulated account handle, the cursor will be placed at the beginning of the field and the consumer can write the purchasing message starting at this position (see Web Appendix Figure A3, Panel A). Thus, if the consumer wants to complete the transaction, she is likely to write a message similar to the one in Web Appendix Figure A2, Panel A, starting immediately with her message, and the tweet will be rendered visible to all of her peers (recipients). Alternatively, if the click happens to be recorded on the right of the prepopulated account handle, the cursor will be placed at the end of the field and the consumer can write the purchasing message starting after the mentioned account handle (see Web Appendix Figure A3, Panel B). Then, if the consumer wants to complete the transaction, she is likely to write a message similar to the one in Web Appendix Figure A2, Panel B, starting immediately with an account handle, and the tweet will not be included in the newsfeeds of her peers (recipients) who do not follow the mentioned account.
Thus, the visibility of each WOM message to the peers of a consumer depends on ( 1) whether an account is mentioned (e.g., @AmericanExpress or another account), ( 2) the exact position of the account handle (whether it is at the very beginning of the message or not), ( 3) what social media accounts the consumer's peers follow, ( 4) whether a user click will be recorded more to the left or more to the right of the screen, and ( 5) whether an account handle will be prepopulated by the platform. Thus, the visibility of the messages is affected by platform design factors (e.g., prepopulating accounts) rather than certain user characteristics exploited by a curation algorithm. Importantly, with regard to the visibility of eWOM messages, the recipient does not control the visibility of these messages from the users she follows; this is important because the purchase decisions we study are the recipients' decisions.
Because our main identification strategy assumes that the visibility of eWOM messages and the geographical distance of consumers are not endogenous in this study, we also empirically investigate whether the visible and nonvisible message groups differ across the pretreatment variables in our model based on the normalized differences tests ([60]) that provide scale-invariant measures of the size of the differences. The normalized differences range from −.1768 to.1692, indicating that there are no significant differences between the two groups in observable characteristics, further enhancing the validity of the identification strategy; differences of.25 or less indicate a good balance between the two groups ([60]). Similarly, kernel distributions, quantile-quantile plots, and the orthogonality test ([60]) further confirm the validity. We also examine the geographical distance in the same way, reaching the same conclusion.
Overall, these unique features of Twitter induce an important variation in the visibility of messages, enabling us to examine the behaviors of peers in treatment (i.e., visible message) and comparison (i.e., nonvisible message) groups in a potential outcomes framework. Thus, differences in purchases between treatment and control groups can be attributed to the corresponding WOM messages and their characteristics, addressing the issue of correlated user behaviors and preferences. In this respect, our research is also relevant to the stream of work that has leveraged the variation in the visibility of advertising messages to estimate the effect of online ads ([ 1]; [48]; [46]).
We further enhance our identification strategy using only observations corresponding to dyadic relationships and social media peers who did not receive messages (either visible or invisible) from multiple disseminators ([ 1]; [12]). In addition to taking advantage of this nonintrusive research design, we also avoid any observer biases; the subjects are unaware of being part of the study and, thus, do not alter their behavior in anticipation of the study. Nevertheless, we also control for differences in the pairwise relationships between users by employing an extensive set of variables and fixed effects in our data-rich setting. Table A1 in the Web Appendix presents additional identification strategies.
Table 3 presents the results of the main econometric specifications of the eWOM effectiveness model. In particular, Model 1 constitutes our baseline specification as it models eWOM effectiveness based on the constructs of dyadic similarity and relationship strength between the disseminator and recipient of the eWOM message (i.e., pairwise user similarity, reciprocity of users' relationship, and number of user interactions), eWOM message advocacy (i.e., personalized message and sentiment of message), and user characteristics (i.e., expertise and leadership). Then, Model 2 introduces the notion of geographical distance, and Model 3 adds the information of eWOM message visibility and leverages the corresponding variation to further distinguish the relationship between geographical distance and eWOM effectiveness from correlated behaviors and homophily among users. That is, Model 3 disentangles the relationship between geographical distance and eWOM effectiveness from the correlational effect of geographical distance reported by Model 2; the eWOM influence is transmitted through visible messages only, whereas correlation in user behaviors and preferences (i.e., homophily) is present even with invisible messages. Then, Model 4 further controls for whether a user account was mentioned in the eWOM message and if the message was explicitly made visible while introducing additional disseminator and recipient controls (i.e., number of followers, leadership, default profile, time on the platform, interests, age, gender, and income) and other controls (i.e., the difference in average age and percentage of male, Black or African American, Hispanic, and Asian-origin residents in the locations of the disseminator and recipient of the message, the log of brand advertising expenditures in USD (in 1,000s) in the location of the recipient of the message, and the time zone difference between the locations of the disseminator and recipient of the message). Finally, Model 5 controls for the specific product mentioned in the eWOM message as well as state and day fixed effects.
Graph
Table 3. Estimation Results of eWOM Effectiveness Model.
| Model 1 | Model 2 | Model 3 | Model 4 | Model 5 |
|---|
| User similarity | 1.244*** | 1.240*** | 1.239*** | 1.257*** | 1.264*** |
| (.009) | (.009) | (.009) | (.010) | (.011) |
| Reciprocal relationship | 5.868*** | 5.378*** | 5.403*** | 4.829*** | 4.527*** |
| (.317) | (.290) | (.291) | (.269) | (.254) |
| Number of interactions | 1.015 | .979 | .977 | .987 | .980 |
| (.034) | (.033) | (.033) | (.034) | (.035) |
| Sentiment of message | 1.540*** | 1.614*** | 1.557*** | 1.728*** | 1.556*** |
| (.107) | (.112) | (.111) | (.133) | (.123) |
| Personalized message | 1.041*** | 1.046*** | 1.047*** | 1.053*** | 1.041*** |
| (.003) | (.003) | (.004) | (.005) | (.007) |
| User expertise | 1.204*** | 1.183*** | 1.184*** | 1.595*** | 1.191** |
| (.030) | (.029) | (.029) | (.130) | (.100) |
| User leadership | 1.006*** | 1.006*** | 1.005*** | 1.011*** | .998 |
| (.001) | (.001) | (.001) | (.002) | (.002) |
| Geographical distance | .891*** | .929*** | .946*** | .954** |
| (.007) | (.017) | (.020) | (.020) |
| Visible message | 1.451*** | 1.598*** | 1.608*** |
| (.171) | (.193) | (.188) |
| Visible message × Geographical distance | .951** | .954** | .945*** |
| (.019) | (.020) | (.020) |
| Additional user controls | No | No | No | Yes | Yes |
| Additional message controls | No | No | No | Yes | Yes |
| Additional controls | No | No | No | Yes | Yes |
| Product fixed effects | No | No | No | No | Yes |
| Geography fixed effects | No | No | No | No | Yes |
| Time fixed effects | No | No | No | No | Yes |
| Log-likelihood | −22,422.8 | −22,312.8 | −22,307.4 | −22,035.2 | −21,763.6 |
| 2,763.5 | −2,983.4 | 2,994.2 | 3,538.6 | 4,081.8 |
| 132,955 | 132,955 | 132,955 | 132,955 | 132,955 |
- 3 *p < .1.
- 4 **p < .05.
- 5 ***p < .01.
- 6 Notes: eWOM effectiveness analysis. The hazard ratios (HRs) represent the percent increase (HR > 1) or decrease (HR <1) in postpurchase hazard associated with each attribute.
According to the results presented in Table 3, we find that the coefficient of the variable capturing the relationship between geographical distance and WOM effectiveness is negative and statistically significant (Visible message × Geographical distance:.945, p < .01, Model 5); all reported coefficients correspond to hazard ratios (HRs) representing the increase (HR > 1) or decrease (HR < 1) in purchase likelihood associated with each attribute. (Table A2 in the Web Appendix also shows the coefficients of the control variables.) Beyond the effect of interest, we also find a negative and statistically significant spatial homophily–based effect of geographical distance (Geographical distance:.954) as well as a positive and statistically significant effect of similarity (homophily). Interestingly, these findings show that despite the "death of distance" postulated in the literature ([28]), geographical distance is negatively associated with the effectiveness of eWOM. Thus, our research is the first to establish that geographical distance has a negative relationship with eWOM outcomes even among familiar social media peers, in addition to the previously known homophily-based effect of geographical distance.
Moreover, the coefficients of all the other variables are in accordance with what one would expect as well as the extant literature on WOM (e.g., [13]; [20]; [45]). Specifically, we find that the increased user similarity and strength of relationship between users (User similarity: 1.264; Reciprocal relationship: 4.527) as well as more intense WOM advocacy (Sentiment of message: 1.556; Personalized message: 1.041) are associated with higher levels of purchase likelihood after exposure to WOM (Visible message: 1.608). Similarly, users with higher product expertise (User expertise: 1.191) seem more persuasive; thus, their followers are associated with a higher purchase likelihood after being exposed to their advocacy.
The relationship of interest is also of economic significance, as a decrease of 10 miles in the distance between users accentuates the effectiveness of eWOM by 12.78% based on the aforementioned coefficients; similarly, an increase of 100 miles in the distance corresponds to a decrease of 25.56%, and an increase of 1,000 miles corresponds to a decrease of 38.34%. For instance, for a recipient living in New York City, the relationship is reduced by 24.39% when the eWOM message originates from a sender in Philadelphia, and 38.82% from Miami, compared with the same message from a sender in New York City.
We further assess the economic significance of the findings by measuring the out-of-sample performance of the models. Specifically, we use a holdout evaluation scheme with an 80/20 random split of data and evaluate the models in terms of Harrell's C concordance coefficient, which measures the likelihood of correctly ordering survival times for pairs of senders and recipients of eWOM messages; the concordance measure is similar to the Mann–Whitney–Wilcoxon test statistic as well as the area under the receiver operating characteristic curve. The results show that our model achieves a predictive performance of.840. Thus, it outperforms the baseline by a large margin, as the baseline performance corresponds to a value of.5. This statistically significant difference further illustrates the managerial relevance of the findings, as they can enhance seeding and targeting strategies ([57]; [99]).
We further quantify the (dollar) value of this increase in out-of-sample performance ([89]). To conduct this calculation ([89]), we use estimates of the cost of targeting (e.g., promoting eWOM messages) and the average product price (Goldfarb and Tucker 2011); the cost of this type of targeting on Twitter is estimated to be $1.35 based on [111], while the average product price in our data set is $125. Combining these data reveals that our model suggests a profit of $.85 per targeted user, which corresponds to a 9% increase over the baseline of not using the information of geographical distance (i.e., $.78), while for random targeting the corresponding profit is only $.008.
Our findings are surprising, as in such an empirical setting the products are not location-specific, there are no transportation fees for consumers, and there is no contracting or potential conflict or ambiguity between senders and recipients of WOM messages ([59]; [79]). Thus, to fully understand our findings, we delve into a likely underlying mechanism of the identified effect and conduct additional analyses that allow us to assess the likelihood of this potential mechanism.
We hypothesize that the negative relationship between geographical distance and eWOM effectiveness could be due to the identification processes of social media users. Specifically, a user who resides near the sender of the message is likely to share a common social identity with the sender based on their geographic proximity ([44]; [62]; [83]; [107]) and thus might be more susceptible to WOM influence originating from this (local) sender.[10] Conversely, a recipient who resides farther away from the disseminator of the WOM message is not likely to share the same location-based social identity and thus is less likely to be persuaded ([37]; [38]; [96]) by mere exposure to eWOM advocacy.
To empirically assess this potential underlying mechanism, we first examine the moderating effect of the salience of the geographical distance from the source of WOM. Salience is activating common social identity identification ([44]; [55]; [56]; [68]; [77]; [106]), and thus, we empirically test the likely underlying mechanism by examining the moderating effect of the salience of the sender's location on the impact of geographical distance on the effectiveness of WOM. If the relationship is accentuated when the geographic proximity of the source of the WOM message is more salient, this would provide empirical support for the hypothesized mechanism of common social identity, as the salience of the location—and thus the salience of the geographic proximity—enhances the social identification processes of the recipient ([44]; [55]; [56]; [68]; [77]; [106]). This would also provide additional empirical evidence in favor of the main identification strategy. Alternatively, if more salient location cues attenuate the relationship, this would provide evidence against the hypothesized mechanism.
According to the results presented in Table 4, we find a negative and significant moderating effect of the salience of the sender's location on the impact of geographical distance on the effectiveness of WOM; the salience of location variable corresponds to whether the location of the WOM sender is explicitly mentioned in her profile. That is, the relationship between geographic distance and eWOM effectiveness is even more negative when the distance is more salient. This finding indicates that common social identity is a likely mechanism for the identified relationship. The results are robust to alternative econometric specifications.[11]
Graph
Table 4. Estimation Results of eWOM Effectiveness Model with the Moderating Effect of the Salience of Geographical Distance.
| Model 1 | Model 2 | Model 3 | Model 4 | Model 5 |
|---|
| User similarity | 1.244*** | 1.240*** | 1.238*** | 1.256*** | 1.260*** |
| (.009) | (.009) | (.009) | (.010) | (.010) |
| Reciprocal relationship | 5.868*** | 5.378*** | 5.525*** | 4.881*** | 4.606*** |
| (.317) | (.290) | (.298) | (.272) | (.258) |
| Number of interactions | 1.015 | .979 | .976 | .989 | .969 |
| (.034) | (.033) | (.033) | (.034) | (.035) |
| Sentiment of message | 1.540*** | 1.614*** | 1.642*** | 1.781*** | 1.568*** |
| (.107) | (.112) | (.118) | (.138) | (.124) |
| Personalized message | 1.041*** | 1.046*** | 1.043*** | 1.050*** | 1.035*** |
| (.003) | (.003) | (.004) | (.005) | (.008) |
| User expertise | 1.204*** | 1.183*** | 1.156*** | 1.581*** | 1.172* |
| (.030) | (.029) | (.028) | (.129) | (.099) |
| User leadership | 1.006*** | 1.006*** | 1.004*** | 1.010*** | .998 |
| (.001) | (.001) | (.001) | (.002) | (.002) |
| Geographical distance | .891*** | .947*** | .961* | .965* |
| (.007) | (.019) | (.021) | (.021) |
| Visible message | 1.430*** | 1.588*** | 1.783*** |
| (.168) | (.192) | (.214) |
| Visible message × Geographical distance | .949** | .950** | .918*** |
| (.021) | (.021) | (.020) |
| Visible message × Geographical distance × Salience of location | .920*** | .948*** | .974* |
| (.011) | (.012) | (.013) |
| Additional user controls | No | No | No | Yes | Yes |
| Additional message controls | No | No | No | Yes | Yes |
| Additional controls | No | No | No | Yes | Yes |
| Product fixed effects | No | No | No | No | Yes |
| Geography fixed effects | No | No | No | No | Yes |
| Time fixed effects | No | No | No | No | Yes |
| Log-likelihood | −22,422.8 | −22,312.8 | −22,280.2 | −22,024.7 | −21,751.1 |
| 2,763.5 | 2,983.4 | 3,048.5 | 3,559.6 | 4,106.8 |
| 132,955 | 132,955 | 132,955 | 132,955 | 132,955 |
- 7 *p < .1.
- 8 **p < .05.
- 9 ***p < .01.
- 10 Notes: eWOM effectiveness analysis with the moderating effect of salience of geographical distance. The salience of location variable corresponds to whether the location of the disseminator is explicitly mentioned in the profile of the disseminator. Additional table notes as in Table 3.
We also examine the likelihood of the hypothesized mechanism in additional ways. For instance, we examine the effectiveness of eWOM under conditions that strengthen the role of geographical location in the social identification process. Specifically, increased political homogeneity in the local area of the recipient of the eWOM message is likely to enhance the importance of the location-based social identity of the recipient as it increases the salience and significance of individuals' social identity due to political entities operating at geographic levels (e.g., precinct, county, state) and the characteristics of the local information environment (e.g., increased number of times an individual is reminded of the local identity, positive perceptions of the local community) ([62]; [91]; [101]). As a result, a pronounced location-based social identity of the recipient is likely to engender biases based on geographical distance, accentuating the relationship. Thus, if the relationship is accentuated when the political homogeneity in the local area of the recipient of the WOM increases, this would provide empirical support for the potential mechanism of social identification. Conversely, the opposite would provide empirical support against this potential mechanism.
Based on the results in Table A3 in the Web Appendix, we find a negative and significant moderating role of the political homogeneity in the local area of the recipient; we have collected data from the MIT Election Lab (https://electionlab.mit.edu) on political voting patterns at the precinct level for the 2016 elections and measure political homogeneity on the basis of the percentage of voters that would need to switch from the majority party to the minority party for the two parties to have equal votes. Put simply, the negative relationship between geographical distance and eWOM effectiveness is amplified when location-based social identity might be more pronounced due to increased political homogeneity. This finding lends support to the hypothesized mechanism of social identity. The results are also robust to alternative specifications.
Similarly, we also examine the moderating role of exogenous hardships in the local area of the recipient of the WOM message. If there have been significant local community hardships or natural disasters, then geography-based common social identity is likely to be more prominent for the residents of the affected area ([109]). Thus, if the relationship is accentuated when the geographic proximity of the source of the WOM message is combined with local community hardships for the recipient, this would provide additional support for the potential mechanism of common identity; we measure local community hardships using (exogenous) deaths related to extreme weather events in the location of the recipient of the message during the last five years prior to our observation window based on data from the National Oceanic and Atmospheric Administration. According to the results in Web Appendix Table A4, we find that the relationship between geographical distance and the effectiveness of eWOM is even more negative for users for whom location-based social identity is likely more pronounced due to local community hardships, lending empirical support to the hypothesized potential mechanism.[12]
We also examine whether the estimated moderating role of geographical distance is above and beyond other potential peer effects. Table A5 in the Web Appendix presents the corresponding results controlling for both the interaction between user similarity and visible message and the interaction between sociodemographic distance and visible message. Our findings remain highly robust, further illustrating that the estimated moderating effect of geographical distance is above and beyond other peer effects, including actual and perceived homophily. The results are also robust to alternative specifications, such as including additional interactions.
In addition to the aforementioned evidence and identification strategies, we assess various alternative explanations. Table 5 presents an overview of these, with the main ones discussed next and additional ones discussed in Web Appendix B.
Graph
Table 5. Alternative Explanations
| Alternative Explanation | Rationale | Identification Strategy | Table(s) |
|---|
| Location-Related |
| Location characteristics | Location-based characteristics may affect purchase likelihoods | Main identification strategy including geography fixed effects | All |
| Spatially correlated user preferences | User interests and brand preferences may be spatially correlated | Main identification strategy including latent user interests | All |
| Additional brand preference controls | A7 |
| Local time-difference effects | Time zone differences are correlated with geographical distance and may relate to differences in users' activities or moods | Main identification strategy including time zone differences | All |
| Local weather conditions | Local weather conditions may affect consumers' activities and moods | Additional weather controls | A9 |
| Small-city effects | Geographic distances are shorter in smaller and more remote locations, where the demand for products sold online might be higher due to potentially limited availability of other products | Main identification strategy including geography fixed effects | All |
| Exclusion of small and remote locations | A8 |
| Local marketing effects | Local marketing effects may affect purchase likelihoods | Main identification strategy including ad controls | All |
| Additional ad response controls | A6 |
| User-Related |
| Homophily | Correlated behaviors among similar (across observed characteristics) peers | Main identification strategy including user similarity controls | All |
| Additional brand preferences similarity | A7 |
| Additional user similarity controls | A11 |
| Propensity-score matching | A13 |
| Latent or unobserved homophily | Correlated behaviors among similar (across latent or unobserved characteristics) peers | Main identification strategy including user latent similarity controls | All |
| Additional latent homophily controls | A10 |
| Additional user similarity controls | A11 |
| Latent variable model | A14 |
| Propensity-score matching | A13 |
| User characteristics | User characteristics may affect purchase likelihoods | Main identification strategy including user controls | All |
| Local demographic effects | Sociodemographic distance between users may affect purchase likelihoods | Main identification strategy including sociodemographic controls | All |
| Income-level effects | Income levels may affect where users select to live; as such, geographical distance may be correlated with income levels | Main identification strategy including income controls | All |
| Message-Related |
| Message content | Message content characteristics may affect purchase likelihoods | Main identification strategy including message controls | All |
| Propensity-score matching | A13 |
| Nonrandom message visibility | Message visibility may not be random, despite provided evidence | Statistical tests | — |
| Main identification strategy including message visibility controls | All |
| Propensity-score matching | A13 |
| Covariate adjustment | A12 |
| Product-Related |
| Product characteristics | Product characteristics may affect purchase likelihoods | Main identification strategy including product fixed effects | All |
| Marketing promotions | Advertising or other marketing activities may affect purchase likelihoods | Main identification strategy including ad controls | All |
| Additional ad response controls | A6 |
| Other-Related |
| Unobserved effects | Unobserved effects correlated with geographical distance of visible eWOM messages | Propensity-score matching | A13 |
| Other unobserved time-varying effects | Any other unobserved effects that vary with time and are correlated with geographical distance of visible eWOM messages | Limited time-horizon | — |
| Main identification strategy including time fixed effects | All |
| Geographic distribution of ties | User might have more geographically proximate than geographically distant ties | Main identification strategy | All |
| Model idiosyncrasies | Model and model specification choices could potentially affect the results | Logistic regression | A15 |
| Alternative specifications | All |
| Spurious effects | Spurious effects or other statistical artifacts | "Placebo" studies | A16 |
One may be concerned that the results might be driven by unpaid or organic marketing effects in the local region of the eWOM message recipient. To evaluate this, we supplement our data set with local web search trends for each product from Google Trends ([14]). Table A6 in the Web Appendix presents the corresponding results controlling for both local marketing expenditures and ad response (via search behaviors) of the local audience. The results remain robust, alleviating concerns that local marketing promotion activities drive the results. The results are also robust to alternative specifications, such as using national Google Trends and advertising expenditures or estimating separate models for each potential confound.
We also examine the robustness of the findings to alternative specifications, such as controlling for homophily based on the overlap in brands that each social media user follows on the platform. Table A7 in the Web Appendix presents the corresponding results. The results remain robust, further corroborating our findings.
Another potential alternative explanation is that the results are driven by disseminator–recipient pairs located in small and remote locations as in such locations distances are in general shorter and demand for products sold online is higher due to the limited availability of other product alternatives ([21]; [79]). We assess this alternative mechanism by repeating the analysis excluding any observations that correspond to small and remote locations, as determined by the Census (i.e., locale assignments). Table A8 in the Web Appendix presents the results; the results remain robust.
We also assess the alternative explanation that the results are driven by the local weather conditions affecting the moods and activities of users ([46]; [67]). We assess this potential explanation by controlling for the temperature, precipitation, and sunshine levels in the location of the recipient using data from the National Oceanic and Atmospheric Administration. Web Appendix Table A9 presents the corresponding results; the results remain robust.
We also undertake an extensive set of tests to assess the robustness of the results and further strengthen the findings, as discussed next; see Table 5 and Web Appendix B for additional details.
First, to enhance the employed identification strategy and examine the robustness of the findings, we further control for latent user characteristics by tapping into the social network structure and recent deep-learning advances. Specifically, we use the method of DeepWalk, a deep-learning method for graphs ([88]), to learn the latent representations of the users and their similarity and further account for both network structure roles and latent user homophily. Table A10 in the Web Appendix presents the corresponding results. The results corroborate our findings. The results also remain robust to employing alternative deep-learning methods, such as the node2vec method ([53]).
We also repeat the analysis including multiple user similarity measures. In particular, the similarity measures correspond to the similarity levels between disseminators and recipients based on ( 1) the Jaccard coefficient of their followers, ( 2) the Jaccard coefficient of their followees, ( 3) the topics discussed in social media posts using the results of the LDA model, ( 4) the intrinsic brand and product preferences of the users based on the overlap in brands that each social media user follows on the platform, ( 5) the demographic information at the corresponding geographic locations (i.e., average age and percentages of male, Black or African American, Hispanic, and Asian-origin residents based on Census data), and ( 6) the latent characteristics of the users based on the deep-learning methods for representation learning; in addition to ( 7) the reciprocity of the relationship and ( 8) the number of interactions between the users. Table A11 in the Web Appendix presents the corresponding results; the results remain robust.
We also examine additional alternative identification strategies to control for any potentially remaining differences between the visible and nonvisible messages; Table A1 in the Web Appendix presents a summary of the different identification strategies. First, we enhance our identification strategy following the covariate adjustment method of [60]. Table A12 in the Web Appendix presents the corresponding results. The results remain robust; the results are also robust to including additional covariate interactions.
Moreover, as an alternative identification strategy, we combine propensity-score matching with the main research design. In particular, we model the propensity of each message to be rendered visible using all the variables that describe the users' relationship and the message characteristics as well as the geographical distance between the users.[13] We conduct the matching based on the propensity scores before estimating again the same econometric models (for additional details, refer to the corresponding table notes). For this robustness check, we use one-to-one matching with replacement and a caliper of.05, yielding a standardized mean (median) absolute difference of.009 (.007) across all the variables, which ensures that covariate balance has been successfully achieved ([18]); the density distributions of the propensity scores also indicate significant overlap and common support. As shown in Table A13 in the Web Appendix, the results remain robust. The results are also robust to nearest-neighbor matching with the generalized Mahalanobis distance.
Furthermore, as an additional alternative identification strategy, we build latent variable models where the sender–recipient similarity is latent and measured based on the various similarity features. Web Appendix Table A14 shows the corresponding results. Model 1 corresponds to the aforementioned latent variable model, while Model 2 combines the latent variable model with propensity-score matching estimating the model over the matched sample. The results of all the aforementioned alternative models are highly consistent and further corroborate our findings.
Finally, as an alternative strategy, to estimate the relationship between geographical distance and eWOM effectiveness, we also use a logit model ([105]) examining whether—rather than how quickly—a user purchases a product. As Web Appendix Table A15 shows, the results remain robust.
We supplement these robustness checks with falsification tests to further assess whether the previous models are picking up spurious effects. As shown in Web Appendix Table A16, the results indicate our findings are not a statistical artifact of the specifications.
Overall, the findings remain highly robust to various alternative identification strategies, econometric specifications, robustness checks, and falsification tests. Figure 3 illustrates the corresponding estimated effects across the specifications.
Graph: Figure 3. Hazard ratios (HRs) with 95% confidence intervals (whiskers) representing the percentage increase (HR > 1) or decrease (HR < 1) in postpurchase hazard across estimated models.
In this study, we investigate the relationship between geographical distance and the effectiveness of eWOM. Specifically, we examine whether the geographical distance between familiar disseminators and receivers of eWOM messages plays an important role—beyond utilitarian reasons and proxying for consumer tastes—in driving recipients' subsequent purchase behaviors. Our results show that the relationship between eWOM and the likelihood that message recipients subsequently also make product purchases significantly strengthens as the spatial proximity between disseminators and receivers grows.
Our findings help advance understanding of conditions that affect online WOM performance. Many of the characteristics previously shown to impact eWOM outcomes relate to the product, brand, or message ([73]; [85]). We contribute by illustrating the role of the important but often overlooked construct of geographical distance in eWOM effectiveness. In showing how geographical distance is still associated with the effectiveness of online WOM in the absence of geography-specific transaction costs between unambiguous users, we demonstrate how the social force field of geography can tether the potential of eWOM. That is, despite the promise of technology to reduce communication barriers and the proclaimed "death of distance" ([29]; [52]), we find that geographic constraints persist online in unexpected ways. Therefore, our results also help address the debate on whether and how geographical distance still matters online ([51]) by showing that it can shape the influence of eWOM.
We also contribute to the theory of eWOM examining why geographical distance is associated with eWOM effectiveness. Specifically, we find evidence that social identification may explain why the influence of online WOM is negatively related to the distance between WOM message disseminators and receivers. That is, our results suggest consumers are susceptible to online information and cues related to social identification as they can, in turn, enhance message persuasiveness. Thus, information and cues relating to social identity can be agents of eWOM influence. Whereas much of the literature on the role of geographic distance in e-commerce and other online settings offers economically driven explanations for the impact of geography, our study proposes behavioral bias relating to social identification may be an underlying mechanism that drives the relationship between geographic distance and eWOM. This finding highlights the need for future research to study additional non–economically driven explanations that can induce such biases.
Our findings have important implications for managers as well. For instance, a controversial argument in the industry is that solely characteristics of the disseminators catalyze the adoption of behaviors and products and thus much of marketing efforts to engineer WOM in social media focus on identifying such characteristics. However, our findings indicate that marketers should expand their focus to take into account the disseminator–recipient pairings and understand that factors pertaining to these pairs can be significantly related to the effectiveness of eWOM. In particular, our results suggest geographical distance matters in online WOM and,, thus marketers can readily take advantage of how geographical distance is associated with eWOM persuasion. Marketers may thus adopt data-driven strategies to selectively promote eWOM episodes according to the proximity of such episodes to each consumer, or to strategically engineer such episodes based on geolocation information. Interestingly, although research has begun to identify pairwise characteristics between senders and receivers that shape the influence of eWOM, such as tie strength and similarity across personality traits ([ 1]; [20]), many of these factors are not readily observable to managers who wish to capitalize on them. The distance between social media users, though, is more easily observable to managers.
Beyond promoting or engineering geographically proximate eWOM episodes, marketers may also benefit from promoting and/or engineering episodes containing other social identity cues. The likely connection between eWOM outcomes and social identity suggests that firms may also consider other cues relevant to social identity formation to further boost the success of interpersonal communications and WOM messages, as enhancing social identification may significantly increase message persuasion and user engagement in the online world.
Our findings also have important implications for the effective design of viral marketing campaigns and ad content. Specifically, brands may boost the persuasiveness of their marketing campaigns by infusing into their content local cues or other identification triggers to induce consumers' social identification processes. Relatedly, marketers are beginning to leverage users' connections on social networks to develop and deliver marketing communications as part of their social advertising efforts. Our research suggests that they could further improve the effectiveness of these strategies by selecting geographically proximate connections to their targets.
Furthermore, going beyond advertising strategies, the implications of our work also provide actionable guidelines for optimizing the delivery of digital content. In particular, our findings can help platforms increase the effectiveness of their content curation and ranking algorithms by incorporating information on content location or source origin and by factoring geolocation into their determination of which user-generated content to disseminate. For instance, content generated by spatially proximate consumers may draw more attention due to identification processes and thereby increase the effectiveness of content provision. In a similar vein, social media platforms may also consider incorporating location information in other functions. For instance, social media platforms may incorporate such information into various other machine-learning algorithms, such as their whom-to-follow recommendations. In addition, our findings could also be used by marketers and platforms to better predict the diffusion of information, products, and user behaviors in social media ([ 5]).
Lastly, deepening our understanding of the factors that can attenuate or accentuate the effectiveness of eWOM has important implications that extend to public policies. For instance, revealing how geographic proximity is positively associated with eWOM effectiveness is critical for the development of effective public policies to induce positive behavioral changes, such as voter turnout, civic engagement, and public health actions.
While our work makes important strides in understanding how geographic proximity is related to eWOM effectiveness, we acknowledge certain limitations, which mostly stem from data availability issues. For instance, we examine the relationship between geographical distance and eWOM in a single social media platform because the service provider launched this venture on only one platform. Future research could examine whether the observed relationship manifests differently on other platforms. Moreover, we did not manipulate the visibility of the messages on Twitter because the venture did not alter the functionality of the platform in any way; future research could consider directly manipulating the visibility of the messages. Similarly, we did not manipulate the geographical distance of users from their followers. In addition, while we capture actual purchases in our data, we do not capture other consumer behaviors that could indicate interest in the products, such as online searches, as this type of information was not available to us. It would be interesting for future research to further examine such potential effects. Future research could also further examine and validate the underlying mechanisms. While we provide evidence that social identification may account for the relationship, future work may conduct experiments to verify this. Lastly, we do not observe in our data private communications between individuals due to privacy reasons and ethical concerns. Nevertheless, we hope these limitations provide avenues for future research that can deepen understanding of the critical role geographic proximity plays in eWOM and other online settings.
sj-pdf-1-jmx-10.1177_00222429211034414 - Supplemental material for Is Distance Really Dead in the Online World? The Moderating Role of Geographical Distance on the Effectiveness of Electronic Word of Mouth
Supplemental material, sj-pdf-1-jmx-10.1177_00222429211034414 for Is Distance Really Dead in the Online World? The Moderating Role of Geographical Distance on the Effectiveness of Electronic Word of Mouth by Vilma Todri, Panagiotis (Panos) Adamopoulos and Michelle Andrews in Journal of Marketing
Footnotes 1 Jacob Goldenberg
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship and/or publication of this article.
4 Online supplement: https://doi.org/10.1177/00222437211034414
5 Local consumer tastes as well as spatially correlated user preferences can also lead to commonalities in product adoption among nearby consumers ([13]; [26]; [72]; [78]). Online user ties might also be spatially correlated, leading to more frequent exposure to the online advocacy of spatially proximate consumers ([8]; [21]; [54]). Similarly, limited regional availability of alternative product options in the offline world may also indirectly lead users to adopt the same products as other geographically proximate consumers ([35]).
6 In cellular provider settings, other explanations may also be at play, such as local network externalities. For instance, consumer utility increases with the local user base size in terms of help or information from local adopters. Local variations in service quality may also explain the correlation between geographic proximity and adoption.
7 By "after," we mean at any time point between the purchase by followee and the end of the product offering, which is also the end of our observation window, listed in the offering announcement (no purchases could have taken place after the end of the offering and the corresponding observation window). An observation in the model corresponds to a user who follows one (and only one) user who had made a purchase, regardless of whether the corresponding eWOM message was visible (or invisible) in the user newsfeed.
8 For instance, the results are robust to supervised methods, according to which we first assign sentiment scores to a small number of messages and then build machine-learning models for predicting the scores for the rest of the messages. The scores estimated through different methods are similar, and the findings remain robust.
9 See https://about.twitter.com/content/dam/about-twitter/en/tfg/download/campaigning-on-twitter-handbook-2019.pdf (accessed September 27, 2021).
The social identity theory holds that an individual's social identity is formed by the perception of belongingness to a community ([16]; [100]). Common social identity (between the message sender and recipient) can be used as a heuristic to make judgments and guide actions ([30]; [31]). Thus, social identity (group identification) has a powerful influence on human behavior (e.g., [17]; [63]; [94]).
These alternative specifications include examining the moderating role of the salience of common location (e.g., same city, same county) or the moderating role of the salience of common location on the association between geographic proximity (instead of geographical distance) and eWOM effectiveness; the geographic proximity variable is estimated as the inverse of the geographical distance variable.
The results are highly robust to alternative econometric specifications such as measuring local hardships over different time horizons or on the basis of weather-related property damages instead of deaths. Similarly, the results are robust to including in the moderating effect of local hardships whether the disseminator and the recipient of the eWOM message reside in the same local area or to examining the effect of geographic proximity instead of distance.
Following extant literature, we allow the matching to be affected by both senders' and recipients' characteristics because they can decrease the potential bias and variance of the estimates ([27]), but the results remain robust if only characteristics of the sender or recipient are used. The results also remain robust when we extend the propensity-matching model variables to also match consumers based on the corresponding region.
References Adamopoulos Panagiotis , Ghose Anindya , Todri Vilma. (2018), " The Impact of User Personality Traits on Word-of-Mouth: Text-Mining Social Media Platforms ," Information Systems Research , 29 (3), 612 – 40.
Adamopoulos Panagiotis , Ghose Anindya , Tuzhilin Alexander. (2021), " Heterogeneous Demand Effects of Recommendation Strategies in a Mobile Application: Evidence from Econometric Models and Machine-Learning Instruments ," MIS Quarterly , forthcoming , https://misq.org/heterogeneous-demand-effects-of-recommendation-strategies-in-a-mobile-application-evidence-from-econometric-models-and-machine-learning-instruments.html.
Adamopoulos Panagiotis , Todri Vilma. (2014), " Social Media Analytics: The Effectiveness of Promotional Events on Brand User Base in Social Media ," in Proceedings of the 35th International Conference on Information Systems (ICIS '14). Auckland, New Zealand : AIS , https://aisel.aisnet.org/icis2014/proceedings/SocialMedia/8/.
Adamopoulos Panagiotis , Todri Vilma (2015a), " The Effectiveness of Marketing Strategies in Social Media: Evidence from Promotional Events ," in Proceedings of the 21st ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD '15). New York: Association for Computing Machinery, 1641–50.
Adamopoulos Panagiotis , Todri Vilma (2015b), " Personality-Based Recommendations: Evidence from Amazon.com ," in Poster Proceedings of the 9th ACM Conference on Recommender systems (RecSys '15). Vienna, Austria : Association for Computing Machinery , https://people.stern.nyu.edu/padamopo/Personality-Based%20Recommendations%20Evidence%20from%20Amazon.pdf.
Adamopoulos Panagiotis , Todri Vilma , Ghose Anindya. (2020), " Demand Effects of the Internet-of-Things Sales Channel: Evidence from Automating the Purchase Process ," Information Systems Research , 32 (1), 238 – 67.
Agrawal Ajay , Catalini Christian , Goldfarb Avi. (2015), " Crowdfunding: Geography, Social Networks, and the Timing of Investment Decisions ," Journal of Economics & Management Strategy , 24 (2), 253 – 74.
Agrawal Ajay , Goldfarb Avi. (2008), " Restructuring Research: Communication Costs and the Democratization of University Innovation ," American Economic Review , 98 (4), 1578 – 90.
Ameri Mina , Honka Elisabeth , Xie Ying. (2019), " Word of Mouth, Observed Adoptions, and Anime-Watching Decisions: The Role of the Personal vs. the Community Network ," Marketing Science , 38 (4), 567 – 83.
API Harmony (2021), "IBM Alchemy Data News," (accessed September 27, 2021) , https://apiharmony-open.mybluemix.net/public/apis/ibm_watson_alchemy_data_news_api.
Aral Sinan , Muchnik Lev , Sundararajan Arun. (2009), " Distinguishing Influence-Based Contagion from Homophily-Driven Diffusion in Dynamic Networks ," Proceedings of the National Academy of Sciences of the United States of America , 106 (51), 21544 –4 9.
Aral Sinan , Nicolaides Christos. (2017), " Exercise Contagion in a Global Social Network ," Nature Communications , 8 , 14753.
Aral Sinan , Walker Dylan. (2012), " Identifying Influential and Susceptible Members of Social Networks ," Science , 337 (6092), 337 – 41.
Aral Sinan , Walker Dylan. (2014), " Tie Strength, Embeddedness, and Social Influence: A Large-Scale Networked Experiment ," Management Science , 60 (6), 1352 – 70.
Archak Nikolay , Ghose Anindya , Ipeirotis Panagiotis G.. (2011), " Deriving the Pricing Power of Product Features by Mining Consumer Reviews ," Management Science , 57 (8), 1485 –1 509.
Arun R. , Suresh Venkatasubramaniyan , Veni Madhavan C.E. , Narasimha Murthy M.N.. (2010), " On Finding the Natural Number of Topics with Latent Dirichlet Allocation: Some Observations ," in Advances in Knowledge Discovery and Data Mining , Mohammed J. Zaki, Jeffrey Xu Yu, B. Ravindran, and Vikram Pudi, eds. Springer , 391 – 402.
Ashmore Richard D. , Deaux Kay , McLaughlin-Volpe Tracy. (2004), " An Organizing Framework for Collective Identity: Articulation and Significance of Multidimensionality ," Psychological Bulletin , 130 (1), 80–114.
Atefi Yashar , Ahearne Michael , Maxham James G. III , Donavan D. Todd , Carlson Brad D.. (2018), " Does Selective Sales Force Training Work? " Journal of Marketing Research , 55 (5), 722 – 37.
Austin Peter C.. (2011), " An Introduction to Propensity Score Methods for Reducing the Effects of Confounding in Observational Studies ," Multivariate Behavioral Research , 46 (3), 399 – 424.
Backaler Joel. (2018), " To Grow Your Business Abroad, Partner with Local Influencers ," Harvard Business Review (September 25), https://hbr.org/2018/09/to-grow-your-business-abroad-partner-with-local-influencers.
Baker Andrew M. , Donthu Naveen , Kumar V.. (2016), " Investigating How Word-of-Mouth Conversations About Brands Influence Purchase and Retransmission Intentions ," Journal of Marketing Research , 53 (2), 225 – 39.
Bell David R. , Song Sangyoung. (2007), " Neighborhood Effects and Trial on the Internet: Evidence from Online Grocery Retailing ," Quantitative Marketing and Economics , 5 (4), 361 – 400.
Berger Jonah , Schwartz Eric M.. (2011), " What Drives Immediate and Ongoing Word of Mouth? " Journal of Marketing Research , 48 (5), 869 – 80.
Bhattacharya Parantapa , Zafar Muhammad Bilal , Ganguly Niloy , Ghosh Saptarshi , Gummadi Krishna P.. (2014), " Inferring User Interests in the Twitter Social Network ," in Proceedings of the 8th ACM Conference on Recommender Systems (RecSys '14) , https://people.mpi-sws.org/∼gummadi/papers/recsys14-userinterests.pdf.
Blei David M. , Ng Andrew Y. , Jordan Michael I.. (2003), " Latent Dirichlet Allocation ," Journal of Machine Learning Research , 3 (4–5), 993 – 1022.
Bloem Craig. (2017), " 84 Percent of People Trust Online Reviews as Much as Friends. Here's How to Manage What They See ," Inc. (July 31), https://www.inc.com/craig-bloem/84-percent-of-people-trust-online-reviews-as-much-.html.
Blum Bernardo S. , Goldfarb Avi. (2006), " Does the Internet Defy the Law of Gravity? " Journal of International Economics , 70 (2), 384 – 405.
Brookhart M. Alan , Schneeweiss Sebastian , Rothman Kenneth J. , Glynn Robert J. , Avorn Jerry , Stürmer Til. (2006), " Variable Selection for Propensity Score Models ," American Journal of Epidemiology , 163 (12), 1149 – 56.
Cairncross Frances. (1997), The Death of Distance: How the Communications Revolution Will Change Our Lives. Cambridge, MA : Harvard Business School.
Cairncross Frances. (2001), The Death of Distance 2.0. London : Texere.
Chaiken Shelly. (1987), " The Heuristic Model of Persuasion ," in Social Influence: The Ontario Symposium , Vol. 5. Hillsdale, NJ : Lawrence Erlbaum Associates.
Chaiken Shelly , Maheswaran Durairaj. (1994), " Heuristic Processing Can Bias Systematic Processing: Effects of Source Credibility, Argument Ambiguity, and Task Importance on Attitude Judgment ," Journal of Personality and Social Psychology , 66 (3), 460 – 73.
Chen Xi , Van Der Lans Ralf , Phan Tuan Q.. (2017), " Uncovering the Importance of Relationship Characteristics in Social Networks: Implications for Seeding Strategies ," Journal of Marketing Research , 54 (2), 187 – 201.
Chevalier Judith A. , Mayzlin Dina. (2006), " The Effect of Word of Mouth on Sales: Online Book Reviews ," Journal of Marketing Research , 43 (3), 345 – 54.
Chintagunta Pradeep K. , Gopinath Shyam , Venkataraman Sriram. (2010), " The Effects of Online User Reviews on Movie Box Office Performance: Accounting for Sequential Rollout and Aggregation Across Local Markets ," Marketing Science , 29 (5), 944 – 57.
Choi Jeonghye , Bell David R.. (2011), " Preference Minorities and the Internet ," Journal of Marketing Research , 48 (4), 670 – 82.
Choi Jeonghye , Hui Sam K. , Bell David R.. (2010), " Spatiotemporal Analysis of Imitation Behavior Across New Buyers at an Online Grocery Retailer ," Journal of Marketing Research , 47 (1), 75 – 89.
Cialdini Robert. (2016), Pre-suasion: A Revolutionary Way to Influence and Persuade. New York : Simon and Schuster.
Coleman Peter T. , Deutsch Morton , Marcus Eric C.. (2014), The Handbook of Conflict Resolution: Theory and Practice. Hoboken, NJ : John Wiley & Sons.
Cox D.R. (1972), " Regression Models and Life-Tables ," Journal of the Royal Statistical Society. Series B (Methodological) , 34 (2), 187 – 220.
Culotta Aron , Cutler Jennifer. (2016), " Mining Brand Perceptions from Twitter Social Networks ," Marketing Science , 35 (3), 343 – 62.
De Veirman Marijke , Cauberghe Veroline , Hudders Liselot. (2017), " Marketing Through Instagram Influencers: The Impact of Number of Followers and Product Divergence on Brand Attitude ," International Journal of Advertising , 36 (5), 798 – 828.
eMarketer (2019), " What Makes US Gen Z and Millennial Internet Users More Likely to Buy Products/Services that an Influencer Recommends? (% of respondents, Sep 2019) ," (November 5), https://chart-na1.emarketer.com/233274/what-makes-us-gen-z-millennial-internet-users-more-likely-buy-productsservices-that-influencer-recommends-of-respondents-sep-2019.
Forman Chris , Ghose Anindya , Goldfarb Avi. (2009), " Competition Between Local and Electronic Markets: How the Benefit of Buying Online Depends on Where You Live ," Management Science , 55 (1), 47 – 57.
Forman Chris , Ghose Anindya , Wiesenfeld Batia. (2008), " Examining the Relationship Between Reviews and Sales: The Role of Reviewer Identity Disclosure in Electronic Markets ," Information Systems Research , 19 (3), 291 – 313.
Fossen Beth L. , Andrews Michelle , Schweidel David A.. (2017), " Sociodemographic Versus Geographic Proximity in the Diffusion of Online Conversations ," Journal of the Association for Consumer Research , 2 (2), 246 – 66.
Ghose Anindya , Singh Param Vir , Todri Vilma. (2017), " Got Annoyed? Examining the Advertising Effectiveness and Annoyance Dynamics ," in Proceedings of the 38th International Conference on Information Systems (ICIS '17). Seoul, Korea : AIS , https://aisel.aisnet.org/icis2017/DataScience/Presentations/21/.
Ghose Anindya , Todri-Adamopoulos Vilma. (2016), " Towards a Digital Attribution Model: Measuring the Impact of Display Advertising on Online Consumer Behavior ," MIS Quarterly , 40 (4), 889 – 910.
Godes David , Mayzlin Dina. (2004), " Using Online Conversations to Study Word-of-Mouth Communication ," Marketing Science , 23 (4), 545 – 60.
Godes David , Mayzlin Dina. (2009), " Firm-Created Word-of-Mouth Communication: Evidence from a Field Test ," Marketing Science , 28 (4), 721 – 39.
Goldfarb Avi , Tucker Catherine. (2011), " Online Display Advertising: Targeting and Obtrusiveness ," Marketing Science , 30 (3), 389 – 404.
Goldfarb Avi , Tucker Catherine. (2019), " Digital Economics ," Journal of Economic Literature , 57 (1), 3 – 43.
Graham Stephen. (1998), " The End of Geography or the Explosion of Place? Conceptualizing Space, Place and Information Technology ," Progress in Human Geography , 22 (2), 165 – 85.
Grover Aditya , Leskovec Jure. (2016), " node2vec: Scalable Feature Learning for Networks ," in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD '16). New York : Association for Computing Machinery , 855 – 64.
Hampton Keith , Wellman Barry. (2003), " Neighboring in Netville: How the Internet Supports Community and Social Capital in a Wired Suburb ," City & Community , 2 (4), 277 – 311.
Heere Bob , Walker Matthew , Yoshida Masayuki , Ko Yong Jae , Jordan Jeremy S. , James Jeffrey D.. (2011), " Brand Community Development Through Associated Communities: Grounding Community Measurement Within Social Identity Theory ," Journal of Marketing Theory and Practice , 19 (4), 407 – 22.
Hinds Pamela J. , Mortensen Mark. (2005), " Understanding Conflict in Geographically Distributed Teams: The Moderating Effects of Shared Identity, Shared Context, and Spontaneous Communication ," Organization Science , 16 (3), 290 – 307.
Hinz Oliver , Skiera Bernd , Barrot Christian , Becker Jan U.. (2011), " Seeding Strategies for Viral Marketing: An Empirical Comparison ," Journal of Marketing , 75 (6), 55 – 71.
Hoffman Matthew , Bach Francis R. , Blei David M.. (2010), "Online Learning for Latent Dirichlet Allocation," in Advances in Neural Information Processing Systems , https://papers.nips.cc/paper/2010/hash/71f6278d140af599e06ad9bf1ba03cb0-Abstract.html.
Hortaçsu Ali , Martínez-Jerez F. Asís , Douglas Jason. (2009), " The Geography of Trade in Online Transactions: Evidence from eBay and Mercadolibre ," American Economic Journal: Microeconomics , 1 (1), 53 – 74.
Imbens Guido W. , Rubin Donald B.. (2015), Causal Inference in Statistics, Social, and Biomedical Sciences. Cambridge, UK: Cambridge University Press.
Ito Jun , Song Jing , Toda Hiroyuki , Koike Yoshimasa , Oyama Satoshi. (2015), " Assessment of Tweet Credibility with LDA Features ," in Proceedings of the 24th International Conference on World Wide Web. New York: Association for Computing Machinery, 953–58.
Jacobs Nicholas F. , Munis B. Kal. (2020), " Staying in Place: Federalism and the Political Economy of Place Attachment ," Publius: The Journal of Federalism , 50 (4), 544 – 65.
Jenkins Richard. (2014), Social Identity , 4th ed. London: Routledge.
Kalbfleisch John D. , Prentice Ross L.. (2011), The Statistical Analysis of Failure Time Data. Hoboken, NJ: John Wiley & Sons.
Katona Zsolt , Zubcsek Peter Pal , Sarvary Miklos. (2011), " Network Effects and Personal Influences: The Diffusion of an Online Social Network ," Journal of Marketing Research , 48 (3), 425 – 43.
Kullback Solomon , Leibler Richard A.. (1951), " On Information and Sufficiency ," Annals of Mathematical Statistics , 22 (1), 79 – 86.
Li Xinxin , Hitt Lorin M.. (2008), " Self-Selection and Information Role of Online Product Reviews ," Information Systems Research , 19 (4), 456 – 74.
Li Chenxi , Luo Xueming , Zhang Cheng , Wang Xiaoyi. (2017), " Sunny, Rainy, and Cloudy with a Chance of Mobile Promotion Effectiveness ," Marketing Science , 36 (5), 762 – 79.
Lin Mingfeng , Viswanathan Siva. (2015), " Home Bias in Online Investments: An Empirical Study of an Online Crowdfunding Market ," Management Science , 62 (5), 1393 – 1414.
Liu Yong. (2006), " Word of Mouth for Movies: Its Dynamics and Impact on Box Office Revenue ," Journal of Marketing , 70 (3), 74 – 89.
Lovett Mitchell J. , Peres Renana , Shachar Ron. (2013), " On Brands and Word of Mouth ," Journal of Marketing Research , 50 (4), 427 – 44.
Lovett Mitchell J. , Staelin Richard. (2016), " The Role of Paid, Earned, and Owned Media in Building Entertainment Brands: Reminding, Informing, and Enhancing Enjoyment ," Marketing Science , 35 (1), 142 – 57.
Ma Liye , Krishnan Ramayya , Montgomery Alan L.. (2014), " Latent Homophily or Social Influence? An Empirical Analysis of Purchase Within a Social Network ," Management Science , 61 (2), 454 – 73.
Manning Christopher , Raghavan Prabhakar , Schütze Hinrich. (2008), Introduction to Information Retrieval. Cambridge, UK: Cambridge University Press.
Manski Charles F. (1993), " Identification of Endogenous Social Effects: The Reflection Problem ," Review of Economic Studies , 60 (3), 531 – 42.
Manski Charles F.. (1999), Identification Problems in the Social Sciences. Cambridge, MA : Harvard University Press.
McGarty Craig , Haslam S. Alexander , Hutchinson Karen J. , Turner John C.. (1994), " The Effects of Salient Group Memberships on Persuasion ," Small Group Research , 25 (2), 267 – 93.
McPherson Miller , Smith-Lovin Lynn , Cook James M.. (2001), " Birds of a Feather: Homophily in Social Networks ," Annual Review of Sociology , 27 (1), 415 – 44.
Meyners Jannik , Barrot Christian , Becker Jan U. , Goldenberg Jacob. (2017), " The Role of Mere Closeness: How Geographic Proximity Affects Social Influence ," Journal of Marketing , 81 (5), 49 – 66.
Momtazi Saeedeh , Naumann Felix. (2013), " Topic Modeling for Expert Finding Using Latent Dirichlet Allocation ," Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery , 3 (5), 346 – 53.
Murphy Rosie. (2018), "Local Consumer Review Survey," in Bright Ideas/Research, https://www.brightlocal.com/research/local-consumer-review-survey/.
Naylor Rebecca Walker , Lamberton Cait Poynor , Norton David A.. (2011), " Seeing Ourselves in Others: Reviewer Ambiguity, Egocentric Anchoring, and Persuasion ," Journal of Marketing Research , 48 (3), 617 – 31.
Newman O. (1972), Defensible Space: Crime Prevention Through Urban Design. New York : Macmillan.
Owoputi Olutobi , O'Connor Brendan , Dyer Chris , Gimpel Kevin , Schneider Nathan , Smith Noah A.. (2013), "Improved Part-of-Speech Tagging for Online Conversational Text with Word Clusters," in HLT-NAACL , https://aclanthology.org/N13-1039.pdf.
Packard Grant , Berger Jonah. (2017), " How Language Shapes Word of Mouth's Impact ," Journal of Marketing Research , 54 (4), 572 – 88.
Packard Grant , Gershoff Andrew D. , Wooten David B.. (2016), " When Boastful Word of Mouth Helps Versus Hurts Social Perceptions and Persuasion ," Journal of Consumer Research , 43 (1), 26 – 43.
Park Eunho , Rishika Rishika , Janakiraman Ramkumar , Houston Mark B. , Yoo Byungjoon. (2018), " Social Dollars in Online Communities: The Effect of Product, User, and Network Characteristics ," Journal of Marketing , 82 (1), 93 – 114.
Perozzi Bryan , Al-Rfou Rami , Skiena Steven. (2014), " Deepwalk: Online Learning of Social Representations ," in Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD '14). New York : Association for Computing Machinery.
Provost Foster , Fawcett Tom. (2013), Data Science for Business: What You Need to Know About Data Mining and Data-Analytic Thinking. Sebastopol, CA : O'Reilly Media.
Quinn Annalisa. (2018), "Thought Leaders," The New York Times Magazine (November 20), https://www.nytimes.com/2018/11/20/magazine/everyone-wants-to-influence-you.html.
Roccas Sonia , Brewer Marilynn B.. (2002), " Social Identity Complexity ," Personality and Social Psychology Review , 6 (2), 88 – 106.
Sammut Claude , Webb, eds. Geoffrey I.. (2017), Encyclopedia of Machine Learning and Data Mining , 2nd ed. New York: Springer.
Shaban Hamza. (2019), "Digital Advertising to Surpass Print and TV for the First Time, Report Says," The Washington Post (February 20), https://www.washingtonpost.com/technology/2019/02/20/digital-advertising-surpass-print-tv-first-time-report-says/.
Shang Jen , Reed Americus , Croson Rachel. (2008), " Identity Congruency Effects on Donations ," Journal of Marketing Research , 45 (3), 351 – 61.
Sharma Gaurav. (2018), "Local and Micro-Influencers: How to Find, Approach and Engage Them for More Human Marketing," BrightLocal (November 6), https://www.brightlocal.com/blog/local-influencers-micro-influencers-more-human-marketing/.
Smith Joanne R. , Hogg Michael A.. (2008), " Social Identity and Attitudes ," in Attitudes and Attitude Change , William D. Crano and Radmila Prislin, eds. New York: Psychology Press, 337 – 60.
Son Jaebong , Lee Jintae , Larsen Kai R. , Woo Jiyoung. (2019), " Understanding the Uncertainty of Disaster Tweets and Its Effect on Retweeting: The Perspectives of Uncertainty Reduction Theory and Information Entropy ," Journal of the Association for Information Science and Technology , 71 (10), 1145 – 61.
Stephen Andrew T. , Lehmann Donald R.. (2016), " How Word-of-Mouth Transmission Encouragement Affects Consumers' Transmission Decisions, Receiver Selection, and Diffusion Speed ," International Journal of Research in Marketing , 33 (4), 755 – 66.
Sun Chenshuo , Adamopoulos Panagiotis , Ghose Anindya , Luo Xueming. (2021), " Predicting Stages in the Consumer Path-Purchase Journey: An Omnichannel Deep-Learning Model ," Information Systems Research , https://ssrn.com/abstract=3814630 or http://dx.doi.org/10.2139/ssrn.3814630.
Tajfel Henri. (1974), " Social Identity and Intergroup Behaviour ," Information (International Social Science Council) , 13 (2), 65 – 93.
Taylor Ralph B. , Gottfredson Stephen D. , Brower Sidney. (1985), " Attachment to Place: Discriminant Validity, and Impacts of Disorder and Diversity ," American Journal of Community Psychology , 13 (5), 525 – 42.
Todri Vilma. (2021), " Frontiers: The Impact of Ad-Blockers on Online Consumer Behavior ," Marketing Science , https://doi.org/10.1287/mksc.2021.1309.
Todri Vilma , Adamopoulos Panagiotis. (2014), " Social Commerce: An Empirical Examination of the Antecedents and Consequences of Commerce in Social Network Platforms ," in Proceedings of the 35th International Conference on Information Systems (ICIS '14). Auckland, New Zealand : AIS, https://aisel.aisnet.org/icis2014/proceedings/EBusiness/54/.
Todri Vilma , Ghose Anindya , Singh Param Vir. (2020), " Trade-Offs in Online Advertising: Advertising Effectiveness and Annoyance Dynamics Across the Purchase Funnel ," Information Systems Research , 31 (1), 102 – 25.
Train Kenneth E. (2003), Discrete Choice Methods with Simulation. Cambridge, UK : Cambridge University Press.
Turner John C.. (1982), " Towards a Cognitive Redefinition of the Social Group ," in Social Identity and Intergroup Relations , Henri Tajfel, ed. Cambridge, UK : Cambridge University Press , 15 – 40.
Twigger-Ross Clare , Bonaiuto Marino , Breakwell Glynis. (2003), " Identity Theories and Environmental Psychology ," in Psychological Theories for Environmental Issues , Mirilia Bonnes, Terence Lee, and Marino Bonaiuto, eds. Farnham, UK: Ashgate, 203–34.
Uzzell David , Pol Enric , Badenas David. (2002), " Place Identification, Social Cohesion, and Environmental Sustainability ," Environment and Behavior , 34 (1), 26 – 53.
Vezzali Loris , Cadamuro Alessia , Versari Annalisa , Giovannini Dino , Trifiletti Elena. (2015), " Feeling Like a Group After a Natural Disaster: Common Ingroup Identity and Relations with Outgroup Victims Among Majority and Minority Young Children ," British Journal of Social Psychology , 54 (3), 519 – 38.
Wang Xuerui , McCallum Andrew , Wei Xing. (2007), " Topical N-Grams: Phrase and Topic Discovery, with an Application to Information Retrieval ," in Seventh IEEE International Conference on Data Mining (ICDM '07) , IEEE , https://ieeexplore.ieee.org/document/4470313.
WebFX (2020), " How Much Does It Cost to Advertise on Twitter?" (accessed October 1, 2020), https://www.webfx.com/social-media/how-much-does-it-cost-to-advertise-on-twitter.html.
Wedel Michel , Kannan P.K.. (2016), " Marketing Analytics for Data-Rich Environments ," Journal of Marketing , 80 (6), 97 – 121.
Weng Jianshu , Lim Ee-Peng , Jiang Jing , He Qi. (2010), " Twitterrank: Finding Topic-Sensitive Influential Twitterers ," in Proceedings of the Third ACM International Conference on Web Search and Data Mining , https://ink.library.smu.edu.sg/cgi/viewcontent.cgi?article=1503&context=sis_research.
Zhu Feng , Zhang Xiaoquan. (2010), " Impact of Online Consumer Reviews on Sales: The Moderating Role of Product and Consumer Characteristics ," Journal of Marketing , 74 (2), 133 – 48.
~~~~~~~~
By Vilma Todri; Panagiotis Adamopoulos and Michelle Andrews
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 78- Is Nestlé a Lady? The Feminine Brand Name Advantage. By: Pogacar, Ruth; Angle, Justin; Lowrey, Tina M.; Shrum, L. J.; Kardes, Frank R. Journal of Marketing. Nov2021, Vol. 85 Issue 6, p101-117. 17p. 4 Diagrams, 1 Chart, 1 Graph. DOI: 10.1177/0022242921993060.
- Database:
- Business Source Complete
Record: 79- <italic>JM</italic>: Promoting Catalysis in Marketing Scholarship. By: Sridhar, Shrihari; Lamberton, Cait; Marinova, Detelina; Swaminathan, Vanitha. Journal of Marketing. Oct2022, p1. DOI: 10.1177/00222429221131517.
Ahead of Print- Database:
- Business Source Complete
Record: 80- Leapfrogging, Cannibalization, and Survival During Disruptive Technological Change: The Critical Role of Rate of Disengagement. By: Chandrasekaran, Deepa; Tellis, Gerard J.; James, Gareth M. Journal of Marketing. Jan2022, Vol. 86 Issue 1, p149-166. 18p. 5 Charts, 12 Graphs. DOI: 10.1177/0022242920967912.
- Database:
- Business Source Complete
Leapfrogging, Cannibalization, and Survival During Disruptive Technological Change: The Critical Role of Rate of Disengagement
When faced with new technologies, the incumbents' dilemma is whether to embrace the new technology, stick with their old technology, or invest in both. The entrants' dilemma is whether to target a niche and avoid incumbent reaction or target the mass market and incur the incumbent's wrath. The solution is knowing to what extent the new technology cannibalizes the old one or whether both technologies may exist in tandem. The authors develop a generalized model of the diffusion of successive technologies, which allows for the rate of disengagement from the old technology to differ from the rate of adoption of the new. A low rate of disengagement indicates people hold both technologies (coexistence), whereas a high rate of disengagement indicates they let go of the old technology in favor of the new (cannibalization). The authors test the validity of the model using a simulation of individual-level data. They apply the model to 660 technology pairs and triplets–country combinations from 108 countries spanning 70 years. Data include both penetration and sales plus important case studies. The model helps managers estimate evolving proportions of segments that play different roles in the competition between technologies and predict technological leapfrogging, cannibalization, and coexistence.
Keywords: cannibalization; disengagement; disruption; leapfrogging; new technologies; switching
In July 2020, Tesla became the world's most valuable automaker, surpassing Toyota in market value for the first time ([30]). But it was Toyota that in 1997 released the Prius, the world's first mass-produced hybrid electric vehicle. In 2006, Tesla Motors, an upstart entrant, bet that the future of the automotive industry would be fully electric cars. They announced they would produce luxury electric sports cars that could go more than 200 miles on a single charge. Incumbents dismissed the effort as futile because of the high entry barriers for auto production, the high cost of producing in California, and the challenges of establishing charging stations. But Martin Eberhard, Tesla's cofounder, noted in a blog in 2006, "a world of 100% hybrids is still 100% addicted to oil...Tesla Motors will remain focused on building the best electric cars for the foreseeable future. With each passing year, our driving range will get longer and the argument for plug-in hybrids will get weaker. To hell with gasoline" ([ 8]).
In contrast, Toyota bet that hybrids would be the future. "The current capabilities of electric vehicles do not meet society's needs, whether it may be the distance the cars can run, or the costs, or how it takes a long time to charge," said Takeshi Uchiyamada, Toyota's vice chairman, who had spearheaded the Prius hybrid in the 1990s ([18]). Toyota faced a hard choice: invest in hybrids, all-electrics, or both?
Globally, during times of potentially disruptive technological change, both industry incumbents and new entrants face difficult choices. For incumbents, the critical dilemma is whether to cannibalize their own successful offerings and introduce the new (successive) technology, survive with their old offerings, or invest in both. To address this dilemma, they need to know whether disruption is inevitable; that is, will the old technology sales decline due to the growth of the new technology and, if so, how much of their existing sales will be cannibalized over time? Or can both old and new technologies, in fact, coexist in tandem? The entrant's dilemma is whether to target a niche and avoid incumbent reaction or target the mass market and incur the wrath of the incumbent ([38]). To address these dilemmas, both incumbents and new entrants need to know how segments of consumers respond to the successive technology. Examples of technological change abound: electric cars versus hybrid cars versus gasoline cars, OLED TVs versus LCD TVs, streaming versus cable, music downloads versus CDs, laptops versus tablets, and app-enabled ridesharing versus taxicabs. Several incumbent firms have also stumbled or failed during disruptive change: Toyota, GM, HP, Nikon, Canon, Kodak, Sony, Nokia, Yellow Cabs, Comcast, and Sears.
Our central thesis in this article is that to effectively manage disruption, we must answer the following substantive research questions: First, when does an old technology coexist with a new, successive technology versus going into an immediate decline? If coexistence occurs, how can one account for the coexistence of two technologies in an empirical model? Second, how can one estimate the extent of cannibalization and leapfrogging of an old technology by a new technology over time? Third, can consumer segments explain coexistence, cannibalization, and leapfrogging in successive technologies, and if so, which segments?
These questions represent pressing concerns for senior managers ([21]). To address these questions, we first outline the theory of disruption, discuss research gaps, and define important constructs that are central to the new model and typology. Then, we develop a generalized model of the diffusion of successive technologies. A key feature of the generalized model is the rate of disengagement from the old technology, which is not forced to equal the rate of adoption of the successive technology, allowing both technologies to coexist. Next, we estimate four latent adopter segments from aggregate data, which correlate with the growth of the new technology, the cannibalization of the old, and/or the coexistence of both: leapfroggers, switchers, opportunists, and dual users (defined shortly).
We apply our model to three different types of aggregate data to ascertain model fit: ( 1) penetration of seven successive technology pairs across 105 countries (441 technology pair–country combinations) spanning multiple years, ( 2) sales of three contemporaneous technology pairs across 40 countries (92 technology pair–country combinations), and ( 3) case analyses of real disruption of large incumbents in the United States. The major benefit of using aggregate penetration and/or sales data is that such data are available abundantly compared to individual-level data. Indeed, much research uses this type of aggregate data to generate rich insights on adoption, diffusion, and generational competition (see [ 4]; [ 5]; [17]; [37]). In addition, we present a test validating the model using a simulation analysis on individual consumer-level data.
Our model and analysis provide both substantive and modeling innovations. Our research provides a better strategic understanding of how, in many situations, old technologies may not necessarily die but survive when new, successive technologies are introduced. The major contributions and implications are the following: First, disruption, though frequent, is not inevitable even when the successive technology grows rapidly, as old technologies can coexist as partial substitutes of the new. Second, the generalized model of diffusion of successive technologies helps strategists and marketers account for this coexistence by allowing the rate of disengagement from the old technology to differ from the rate of adoption of the new. Third, the separately estimated rate of disengagement enables a superior fit to data on technological succession. Fourth, the model helps estimate cannibalization by new, successive technologies, as well as sizes of four critical segments, providing key signals about disruption. The coexistence of both technologies occurs when there is a large segment of dual users. In contrast, the size of the leapfroggers segment correlates with the growth of the new technology, and the size of switchers and opportunists correlates with cannibalization of the old technology. Fifth, the profit implications of leapfrogging and cannibalization may vary greatly depending on which firms market which technology. Major incumbents may fail during the takeoff of new technologies due to underestimating the size of leapfroggers (opportunity cost) and switchers (real cost). Sixth, the generalized model can capture variations in segment sizes across technologies and global markets. The next sections present the theory, new typology, model, empirical applications, and strategic implications.
The theory of disruptive change ([ 2]; [ 6]) suggests that a new technology enters a market, improves in performance, and then surpasses the performance of the existing technology. During times of such technological change, leading incumbent firms fail, not because they were technologically incapable of producing and marketing these innovations themselves, but because they focus on their existing (mainstream) customers, who were satisfied with the existing technology because it met their needs on the primary dimension of performance ([ 6]).
Christensen and his coauthors suggest that the new technology enters, survives, and grows because it offers benefits on a secondary dimension of performance that appeals to niche segment consumers. Over time, the new technology improves in performance and at some point meets the standards of the mainstream segment on the primary dimension of performance. These customers then switch to the new technology. Disruption occurs if the incumbent focuses on the old technology to the exclusion of the new one.
Several authors have criticized the theory of disruption because of circular definitions, lack of large empirical evidence or a predictive model, and a failure to examine whether consumer behavior changes (e.g., [25]; [35]; [36]; [34]). However, no study has refuted the essential features of the theory of disruption: that successive technologies do compete, the competing technologies appeal to different segments, the new technology grows in performance over time, and the niche it serves grows in response to this improvement.
A major limitation of prior work on disruption is that it does not provide recommendations on some critical issues that concern both incumbents and new entrants: How can they estimate the extent of cannibalization over time and who are the customers most susceptible to the new technology? Could the two technologies coexist, and which segments drive the coexistence of both technologies and the growth of the new technology? This research seeks to address these issues.
To answer the previous questions using the theory of disruption, we define the concepts of successive technology, substitution, and segments.
A new successive technology (which can include both a technology and a product) addresses similar underlying consumer needs as the old technology (e.g., DVR vs. VCR) or may tap simultaneously into multiple needs (e.g., PC, laptop, tablet). Successive technologies do not include new generations of the same product. Note that in this article, we use the term "successive technology" synonymously with "new technology" and the term "old technology" synonymously with "prior technology," given the context of technological succession. "Cannibalization" is the extent to which the successive technology "eats" into real or potential sales (or penetration) of the old technology due to substitution.
Much research in marketing (e.g., [ 7]; [13]; [24]; [26]; Table 1) addresses the related issue of the diffusion of perfectly substitutable successive generations of the same technology (e.g., iPhone 8 vs. iPhone 7), in which the consumer always prefers the new generation to the old at the same price (e.g., iPhone 9 and 10). Thus, successive generations of the same technology exhibit perfect substitution. Here, consumers completely disengage from the old generation (of the same product) when they adopt the new one.
Graph
Table 1. A Comparison with Related Literature on Generational Substitution.
| Article | Key Question | Data | Partial Disengagement? | Leapfrogging Considered? |
|---|
| This article | To examine the diffusion of successive technologies while accounting for coexistence, cannibalization, and leapfrogging. | Multicountry penetration and sales data across several countries for technology pairs and triplets; case studies; simulation | Yes | Yes (four adopter segments considered) |
| Koh et al. (2019) | To quantify generational substitution, unbundling, and piracy effects. | Downloadable music; CDs; streaming | No | No |
| Guo and Chen (2018) | How consumers strategic behavior affects sales and profits for multigeneration products. | Numerical optimization | No | Yes |
| Shi et al. (2014) | To incorporate consumers' forward-looking behavior in multigenerational models. | Eight products across four firms | No | Yes |
| Lam and Shankar (2014) | What drives mobile device brand loyalty? | Survey data on attitudes toward mobile phone brands spanning two generations: 2.5 G versus 3G | No | No |
| Jiang and Jain (2012) | To develop an extension of the Norton–Bass model to separate adopters who substitute an old product generation with a new generation into those who adopted the earlier generation and those who did not. | Two generations of one category in one country; Three generations of one category in one country | No | Yes |
| Stremersch et al. (2010) | To test whether growth acceleration occurs across multiple product generations. | 39 technology generations in 12 product markets | No | Assumes no leapfrogging |
| Goldenberg and Oreg (2007) | To redefine the laggards concept and link it to the leapfrogging effect. | 54 products (not specifically successive generations) | N/A | Yes |
| Danaher et al. (2001) | To incorporate marketing mix variables in the diffusion of multigeneration products. | Two generations of one category in one country | No | Yes |
| Kim et al. (2000) | To develop a model able to incorporate both interproduct category and technological substitution effects simultaneously. | One technology market in two countries | No | No |
| Jun and Park (1999) | To propose a model that incorporates diffusion and choice effects to capture diffusion and substitution for multiple generations of products. | Successive generations of two technology categories, not multicountry | No | Not specifically |
| Mahajan and Muller (1996) | To develop a model that accounts for diffusion and substitution for successive generations of technological innovations. | Successive generations of one technology category | No | Yes |
| Norton and Bass (1987) | To develop a model that accounts for both diffusion and substitution for successive generations of high-tech products. | Successive generations of one technology category | No | N/A |
Technological competition is more complex than intergenerational competition because successive technologies may be only partial substitutes. That is, whereas some consumers prefer the successive technology over the old technology (e.g., teens), other consumers may find value in and prefer to hold both (e.g., homeowners who have PCs, laptops, and tablets or keep both mobile phones and landlines). For example, while the two technologies may differ in terms of the scientific principle, the old technology may still serve a need that the successive technology cannot fulfill. In such a case, a group of adopters could choose to hold both technologies, triggering the need for a model that does not force complete substitution. In this case, consumers do not fully disengage from the old technology and may co-own successive technologies.
For example, consider Figure 1a, which plots the penetration of VCRs and the successive technology of DVD players. Here we observe a fast adoption of DVD players, but over this same period, the decline in VCRs (Technology 1) is relatively small. In other words, a number of customers initially held on to both technologies before switching entirely to DVD players. Figure 1a also shows other such examples of the coexistence of successive technologies. Figure 1b shows a similar initial coexistence in sales of technology pairs. Therefore, to model the diffusion of successive technologies, one needs to allow for a rate of disengagement from the preceding technology that is not exactly equal to the rate of adoption of the new technology (i.e., one must allow for partial substitution). This inclusion of a separate rate of disengagement (F12 in this article) is one of the innovations we propose in this research. A low rate of disengagement indicates consumers hold on to both technologies, whereas a high rate indicates they discard the old technology in favor of the new. Thus, the greater the rate of disengagement, the greater the cannibalization of the old technology by the new technology.[ 5]
Graph: Figure 1a. Market penetration of select technology pairs.
Graph: Figure 1b. Sales of select technology pairs.
We define and derive mathematically a typology of four adopter segments for successive technologies: ( 1) "Leapfroggers" adopt the successive technology but would never have adopted the old technology and thus present a new consumer segment for the new technology. This is the niche in Christensen's theory of disruption that provides initial sales for the new technology. ( 2) "Switchers" are consumers who had already adopted the older technology but who choose to replace it with the successive technology after the latter technology is introduced. In Christensen's theory of disruption, this is the mainstream consumer segment that switches to the successive technology after it improves. The refinement in our empirics is that this segment switches continuously to the successive technology as it improves. Each year, customers switch as the successive technology matches their needs better than the old technology. ( 3) "Opportunists" are those who would have adopted the old technology but delayed the decision and instead end up adopting the successive technology. ( 4) "Dual users" are those who had already adopted the older technology but who elect to adopt/use both technologies once the successive technology is introduced. This segment also includes those who would have adopted the old technology but had delayed the decision and ended up adopting and using both technologies.
Many situations exist in which one technology substitutes for another but the substitution is only partial, either due to incomplete compatibility or because the old technology still has its uses. Thus, it makes sense to hang on to the old technology because it is still useful (e.g., VHS vs. DVD), even in the presence of the new. Currently, no model allows for this coexistence of successive technologies. However, multigenerational models such as [26] and [14] model the diffusion of successive generations of the same technology. Although the [26] model is not right for multitechnology substitution, a modification of the Norton–Bass model is well-suited for this context.
Our proposed model uses the multigenerational model of [26] as a starting point and extends this model to consider the context of the adoption of successive technologies that do not fully cannibalize each other (partial substitution). The major difference in our model is that we include a rate of disengagement from the old technology that does not equal the rate of adoption of the successive technology, which accounts for partial substitution in the case of successive technologies versus complete substitution in the case of successive generations of the same technology.
Herein, we ( 1) specify our intuition that motivates the derivation of adopter segments for successive technologies, ( 2) outline our model for the diffusion of two successive technologies (the Web Appendix provides an extension to multiple technologies), ( 3) discuss our critical departure from the basic model of multigenerational diffusion (i.e., we provide a more flexible model in which we do not force the rate of disengagement from Technology 1 [this term is used in this section to concisely reflect the old technology] to exactly match the rate of adoption of Technology 2 [we use this term for the successive technology]), and ( 4) illustrate the equations we used to decompose adoption into four adopter segments.
We specify the proposed model for the simplest case of the diffusion over time of two successive technologies as follows. Let and respectively represent the penetration of Technologies 1 and 2 at each time period . Then we model and as follows:
Graph
1
Graph
2
Note we have added the 1 in Equations 1 and 2 to account for the fact that we are only considering whole years. corresponds to the introduction year for Technology 2, and
Graph
3
refers to the fraction of all potential Technologyg consumers for each technology at time t. Here, g refers to a technology (rather than a generation of a technology as is typically considered in the literature on multigenerational diffusion). Our model contains eight parameters: and . The parameter represents the long-run penetration for Technology 1 if Technology 2 had never been introduced. Put another way, prior to the introduction of Technology 2, the penetration for Technology 1 will converge toward but will never reach because for Technology 2 will start to reduce the market share of Technology 1. Thus, Technology 2 begins to take market share from Technology 1 upon its introduction. Similarly, represents the additional market share for Technology 2 above that of Technology 1, so our model assumes that the long-run penetration for Technology 2 will equal . The parameters and are the coefficients of innovation for Technologies 1 and 2, respectively, and and are the coefficients of imitation for Technologies 1 and 2, respectively. and can then be thought of as the coefficients of disengagement. Thus, describes the rate at which customers adopt Technology 1 prior to the introduction of Technology 2, and models the rate of adoption of Technology 2 after its introduction. Finally, models the rate at which Technology 1 customers disengage upon the introduction of Technology 2.
Note that we make two critical departures in this specification from what is typical of multigenerational diffusion models. Typically, multigenerational diffusion models restrict . The proposed model removes such a restriction for the context of successive technologies. The potential advantage of modeling and separately is as follows: when the rate of disengagement by current Technology 1 customers exactly matches the rate of adoption by Technology 2 customers. However, in the case of successive technologies, across categories and countries, consumers may in fact hold both technologies simultaneously. For example, many families with older members have both a landline and a mobile phone. In addition, both technologies may grow simultaneously in different customer segments. Therefore, one of our innovations in developing a corresponding model to fit the context of successive technologies is to allow , which corresponds to people adopting Technology 2 at a faster rate than they leave Technology 1. If , then there is no substitution effect and people are holding on to both technologies. When is large, there is a large substitution effect. This is a strength of the model because we can directly measure the substitution effect rather than forcing .
Second, an important distinction from prior models is that we also do not constrain to equal or to equal , a constraint that is suitable when the changes between the two generations are incremental, as in multigenerational diffusion, but not when the technology is discontinuous ([24]), as in our more general case of successive technologies. Given that each successive new technology provides a substantial improvement in benefits, we expect the diffusion parameters p and q to vary for each new technology in a pair or triplet. Thus, our model does not constrain to equal or to equal .
Note that, similar to previous models, we make certain assumptions. First, we assume a pure Bass model formulation for the first technology ([ 1]). However, we acknowledge that the first technology may have been affected by a previous technology. Second, we model using the same functional form as and for two reasons. Empirically, we find that the model with this form fits our data well. In addition, by modeling using the same functional form as , our approach reduces to the standard [26] and [14] formulations whenever . Thus, we provide a strict generalization of previous models. Overall, however, our model is a generalized model that can apply to both generational diffusion and technology diffusion.
Let represent the observed yearly penetration of Technology at time . Then, estimating the eight parameters in Equations 1, 2, and 3 can be achieved using nonlinear least squares. In particular, we select and as the values that minimize
Graph
4
where represents the number of years of observation. We minimize Equation 4 using the NLS function in the statistical software package R. Once the parameters have been estimated, it is a simple matter to plug the estimates back into Equations 1 and 2 to predict future penetration for Technologies 1 and 2.
Next, we decompose penetration of Technology 2 into the four major segments defined earlier. Switchers (SW) and opportunists (O) represent a lost market for Technology 1 and thus its cannibalization (CAN), whereas leapfroggers (L) and dual users (DU) represent market growth (MG). Therefore, comprises the sum of these segments as such:
Graph
5
Similarly, comprises the initial market for this technology ( ) less cannibalization from Technology 2 as such:
Graph
6
We derive the various consumer segments as follows:
Graph
7
Graph
8
Graph
9
Graph
10
where .
It is not hard to verify that the four quantities in Equations 7– 10 satisfy Equations 5 and 6. Let us first consider . Recall that represents the total potential additional market for Technology 2 beyond that of Technology 1 and provides the fraction of potential customers who have actually adopted the new technology. Thus, corresponds to the total number of additional Technology 2 adopters who would never have adopted Technology 1. Next, consider . Note that represents the number of customers who would be expected to adopt Technology 1 in time period . However, of these customers switch directly to Technology 2, while customers adopt both technologies. Therefore, summing from up to gives the total number of opportunists (Equation 9). corresponds to dual users who adopt both technologies. Here, represents the number of people who have adopted Technology 1, and represents the fraction of these people who have adopted both technologies.
Finally, the switchers correspond to the remaining adopters of Technology 2, which can be shown to correspond to Equation 8. At , this equation is fairly intuitive because represents the current number of Technology 1 adopters and represents the fraction of potential customers who drop Technology 1 to adopt Technology 2 in period . Thus, Equation 8 assumes that current customers of Technology 1 switch to Technology 2 at the same rate as noncustomers of Technology 1. However, for , the intuition becomes more complicated because the number of Technology 1 customers will be less than as a result of prior switching.
Note that we have chosen to focus on identifying the adopters of the new technology. While we consider the role of dual users, who continue to find value in the old technology, we do not distinguish, for the sake of simplicity, between other types of old technology adopters—for example, those who may never adopt either technology, those who are yet to adopt the old technology but will not adopt the newer technology, and those who will stay loyal to the old technology.
We can extend this model to more than two technologies. In markets characterized by excessive turbulence, a third technology is often introduced in quick succession to the second technology. We can extend our model to account for different technologies: . Here, successive technologies cannibalize the market of earlier technologies. In the interest of brevity, we detail the model extension to three technologies and its application for data on technology triplets in Web Appendix W1.
The proposed model allows us to extract the sizes of the four adopter segments for each year and technology pair in each country using the defined equations. Our model has several additional desirable characteristics. First, the model parameters have natural interpretations. For example, corresponds to the rate that individuals would adopt technology in the absence of any competing technologies, and represents the rate that individuals disengage from Technology − to adopt Technology . Second, by setting , our model reduces to that of [26] and [14], so their model can be seen as a special but more restrictive version of our approach for this context. Our empirical results suggest that our model provides a significantly more accurate fit to the data on successive technologies. Third, market growth generated by a particular technology can be easily computed as the sum of leapfroggers and dual users, and cannibalization can be computed as the sum of switchers and opportunists. Fourth, we do not place any restrictions on the size of adopter segments. Thus, market growth can be positive or negative. The latter case occurs when the total market size actually declines with the introduction of a new technology, possibly indicating disruption by yet another technology. While not the norm, our empirical results suggest that market growth can at times be negative when a still newer technology emerges for which we do not have data.
One may ask what evidence we have that our model can correctly recover individual consumer segments given that we have only aggregate data. To validate our model for this purpose, we ran a series of simulation analyses following precedents in model simulation ([28]; [39]). For our data generation process, we simulated the adoption of two technologies by a large group of individual customers. The simulation demonstrates a good fit with only ten years of data for Technology 1 (i.e., the model yields a reasonably good fit with only five years after Technology 2 enters the market) (Simulation Exercise 1). With more years of simulated data, the fits become even more accurate. Next, we show the robustness of the simulation analysis to the inclusion of a continuous heterogeneity distribution (Simulation Exercise 2) and the absence of some of the segments altogether (Simulation Exercise 3). These exercises provide more confidence that our model can uncover meaningful structure from the aggregate data even when the model assumptions do not hold exactly. Details are in Web Appendix W2.
This section covers applications of the model using data from different contexts.
We examined the fit of the model using the market penetration[ 6] of seven technology pairs (telephone–mobile phone, dial-up internet–broadband, black-and-white TV–color TV, VCR–DVD player/recorder, DVD player–Blu-ray player, personal computer–laptop, and laptop–tablet) spanning 105 countries (441 technology pair–country combinations). The data were compiled from several sources (Passport Euromonitor, Fast Facts Database, and the telecommunications database of the International Telecommunications Union).
Overall, the proposed model fits the data well. Table 2 presents comparisons of the penetration data for four technology pairs using both mean-squared and median-squared errors of our proposed model with the separately estimated disengagement rate compared to the reduced form model using the simplifying assumption Our proposed model gets much smaller error rates than the latter model.
Graph
Table 2. In- and Out-of-Sample Fit Statistics for Technology Pairs Using Penetration Data.
| Training Errors on Model where |
|---|
| Tech 1 | Tech 2 | Tech 1 Mean | Tech 2 Mean | Overall Mean | Tech 1 Median | Tech 2 Median | Overall Median |
|---|
| Laptop | Tablet | .0043 | .0009 | .0026 | .0006 | .0001 | .0003 |
| Personal computer | Laptop | .0123 | .0016 | .0070 | .0018 | .0003 | .0010 |
| DVD player | Blu-ray | .0015 | .0001 | .0008 | .0004 | .0000 | .0002 |
| VCR | DVD player | .0032 | .0082 | .0057 | .0012 | .0056 | .0018 |
| Test Errors on Model where |
| Tech 1 | Tech 2 | Tech 1 Mean | Tech 2 Mean | Overall Mean | Tech 1 Median | Tech 2 Median | Overall Median |
| Laptop | Tablet | .0324 | .0134 | .0229 | .0030 | .0012 | .0023 |
| Personal computer | Laptop | .0390 | .0131 | .0260 | .0031 | .0017 | .0025 |
| DVD player | Blu-ray | .0491 | .0073 | .0282 | .0013 | .0034 | .0020 |
| VCR | DVD player | .0096 | .1223 | .0659 | .0025 | .0567 | .0089 |
| Training Errors on Our Method with F2 ≠ F12 |
| Tech 1 | Tech 2 | Tech 1 Mean | Tech 2 Mean | Overall Mean | Tech 1 Median | Tech 2 Median | Overall Median |
| Laptop | Tablet | .0014 | .0002 | .0008 | .0003 | .0000 | .0001 |
| Personal computer | Laptop | .0024 | .0004 | .0014 | .0013 | .0000 | .0005 |
| DVD player | Blu-ray | .0011 | .0000 | .0006 | .0004 | .0000 | .0001 |
| VCR | DVD player | .0008 | .0014 | .0011 | .0004 | .0005 | .0005 |
| Test Errors on Our Method with F2 ≠ F12 |
| Tech 1 | Tech 2 | Tech 1 Mean | Tech 2 Mean | Overall Mean | Tech 1 Median | Tech 2 Median | Overall Median |
| Laptop | Tablet | .0072 | .0017 | .0045 | .0012 | .0001 | .0003 |
| Personal computer | Laptop | .0084 | .0035 | .0059 | .0012 | .0003 | .0007 |
| DVD player | Blu-ray | .0530 | .0033 | .0281 | .0023 | .0009 | .0014 |
| VCR | DVD player | .0027 | .0622 | .0325 | .0006 | .0053 | .0015 |
Table 2 presents the results by old and new technology as well as the average error across both technologies for the four pairs (the subsample is displayed for brevity). We derived the mean errors in the "training," or in-sample data, by excluding the last time point for each curve, fitting each of the two competing models to the remaining time points, and calculating the mean of squared errors between the observed and predicted points for each technology pair across countries. In contrast, we derived the "test," or out-of-sample results, by excluding the last time point from each curve and fitting the models to the remaining time points (K = 1). However, in this case, the mean squared error is calculated using the squared difference between the final year's observed and predicted points and calculating the overall average error across countries for each technology pair. Overall, our model fits much better out of sample as well as in sample, which is the true test for better performance of our model. The median error rate refers to the in-sample and out-of-sample error rate across the different countries—using the median instead of the mean—to account for the fact that some countries may greatly influence the averages.[ 7] See Figure 2 for some illustrative fit plots. Web Appendix W3 presents an analysis for K = 3 and 5 years. Overall, this analysis indicates that our model, which allows , still outperforms a model that allows . Table 3 provides the mean parameter estimates for these technology pairs.
Graph: Figure 2. Sample fit plots from application of model with penetration data.Notes: Displayed are the fit plots for sample technology pairs. The black lines are the real data. The red line is plotted using our model (F2 ≠ F12) and the green dashed line is for the model with F2 = F12. The vertical lines represent the year of introduction of the new technology into the market.
Graph
Table 3. Parameter Definitions and Estimates.
| Parameter | Interpretation | Laptop–Tablet | PC–Laptop | DVD–Blu-ray Player | VCR–DVD Player |
|---|
| M | SD | M | SD | M | SD | M | SD |
| m1 | Long-run penetration potential for Technology 1 if Technology 2 had never been introduced | 76.65 | 54.49 | 81.69 | 30.62 | 73.08 | 41.35 | 66.65 | 35.63 |
| m2 | Additional market share for Technology 2 above that of Technology 1 | 20.60 | 24.83 | 11.06 | 17.61 | 16.93 | 46.81 | 9.15 | 15.16 |
| p1 | Coefficient of innovation for Technology 1 | .004 | .010 | .002 | .004 | .005 | .014 | .039 | .048 |
| q1 | Coefficient of imitation for Technology 1 | .250 | .056 | .221 | .061 | .544 | .164 | .182 | .130 |
| p2 | Coefficient of innovation for Technology 2 | .006 | .011 | .007 | .008 | .012 | .015 | .007 | .010 |
| q2 | Coefficient of imitation for Technology 2 | .222 | .109 | .172 | .071 | .162 | .140 | .521 | .262 |
| p12 | Coefficient of disengagement 1 | .003 | .013 | .001 | .002 | .014 | .025 | .011 | .021 |
| q12 | Coefficient of disengagement 2 | .022 | .056 | .025 | .049 | .157 | .295 | .193 | .118 |
| N | Count | 85 | | 85 | | 41 | | 41 | |
Our model allows us to decompose penetration for technology pairs into adopter segments. We provide an illustrative example for telephone–mobile phones in India. In Figure 3a, L1 is the projected penetration of Technology 1 (telephone) if the successive technology (mobile phone) were absent. S1 is the estimated penetration for Technology 1, indicating the effect of cannibalization (L1 − Cannibalization) due to switchers (SW) and opportunists (O). In Figure 3b, S2 (penetration for Technology 2 (mobile phone) is decomposed into leapfroggers (L2), total cannibalization (switchers (SW) + opportunists (O)), and dual users (DU). Here, the penetration of mobile phones is initially dominated by leapfroggers, followed by growth from cannibalization. In Figure 3c, S1 + S2 represents the evolution of the overall market due to market growth from Technology 2 (leapfroggers + dual users) compared to the presence of only Technology 1 (L1). Overall, the introduction of mobile phones in India created market expansion.
Graph: Figure 3a. Decomposition of penetration of telephone (old technology) in India.
Graph: Figure 3b. Decomposition of penetration of mobile phone (new technology) in India.
Graph: Figure 3c. Evolution of the market (India telephone and mobile phone).Notes:Figure 3a shows the projected penetration L1 of Technology 1 if the successive technology were absent and the effect after cannibalization from Technology 2, represented by S1, the estimated penetration. Figure 3b shows the breakdown of the penetration curve (S2) for Technology 2 (mobile phone in India) into leapfroggers (L2), cannibalization (switchers [SW] + opportunists [O]), and dual users (DU). Figure 3c shows the evolution of the overall market (S1 + S2) due to market growth (MG) from Technology 2 (leapfroggers + dual users) compared to the market in the presence of only Technology 1 (L1). The figures are plotted over the lifetime of available data for Technology 1.
We next present some key results derived from decomposition of the data across the 441 technology pair–country combinations ten years from the commercialization of the new technology, using our model. Figure 4a presents the average size of the adopter segments across categories. Notice that for the transition from dial-up to broadband, on average across countries, switchers form the dominant category in terms of market penetration (8%), followed by leapfroggers (6%), rather than dual users. In terms of validity, these results make sense because most adopters are unlikely to hold both dial-up and broadband. In contrast, for landline telephones–mobile phone, dual users (24%) dominate on average across countries; in other words, most adopters were keen on holding both technologies ten years from the commercialization of the new technology.
Graph: Figure 4a. Decomposition by adopter segments across technology pairs.
Furthermore, on average, growth of Technology 2 derived from cannibalization of Technology 1 due to switchers and opportunists is greater than from market growth due to leapfroggers and dual users for the Blu-ray and broadband markets. In contrast, market growth is greater than cannibalization for the other technology pairs. Overall, the results indicate the size of adopter segments and the effects of leapfrogging and cannibalization vary across categories.
Following marketing research discussing cross-country effects with multiple data sets (e.g., [19]; [29]), we examine if adopter segments vary across countries. We classify countries in our data set into developing and developed countries. Specifically, we use the analytical classification provided by the World Bank and gathered from various historical reports, as income classifications are rigorous and contemporaneous.[ 8] We term low and low-middle income countries as developing and middle and high-income countries as developed. We present the following results using data from 323 technology pair–country combinations in which we were able to identify the country income classification as of Year 10 from new technology commercialization. We identify 131 cases of high-income countries, 88 of upper-middle income, and 104 of low-income (includes low and low-middle income) countries.
The mean estimated penetration of Technology 2 ten years after the new successive technology commercialization is 18% for low-income countries and 23% for high-income countries. The mean estimated penetration of Technology 1 ten years after new technology commercialization is 24% for low-income countries and 49% for high-income countries. These estimates were very close to the actual penetration data for that year.
Overall, the mean for leapfroggers is significantly higher for low-income countries compared to both high-income countries (MeanLlowinc = 7.04, MeanLhighinc = 2.61, t = 4.10, p =.0001, using a two-sample T-test with unequal variances) and upper-middle-income countries (MeanLuminc = 4.21, t = 2.09, p =.038). The mean for dual users is significantly higher for high-income compared with low-income countries (MeanDUhighinc = 16.29, MeanDUlowinc = 6.23, t = 4.97, p <.0001) and upper-middle-income countries (MeanDUuminc = 9.10, t = 3.12, p =.002).
Thus, a key empirical generalization from our analysis is that developing countries exhibit a higher level of leapfrogging adoption than developed countries in the early life cycle of the successive technology, whereas developed countries exhibit a higher level of adoption by dual users than developing countries in the early life cycle of the successive technology (Figure 4b).
Graph: Figure 4b. Decomposition of adopter segments across income classifications of countries.
Overall, we find that adopter segments of successive technologies have some context-dependent variations, validating the need for a generalizable model that managers can use to understand the extent of cannibalization and/or market growth.
Next, we examine whether the model fits aggregate sales data. We use historical sales data (units in thousands) on three contemporary technology pairs (laptops–tablets, DVD players–Blu-ray players, and digital cameras–smartphones) from 40 countries, with 92 product–country combinations in total for the years 1990–2017 from the Euromonitor Passport database[ 9].
Table 4 shows the fit statistics. Results indicate that our model with a separately estimated disengagement also fits sales data well. The mean parameter estimates across the 92 product–country combinations are p1 =.02 (SD =.09), q1 =.54 (SD =.34), p2 =.02 (SD =.03), q2 =.29 (SD =.32), p12 =.09 (SD =.12) and q12 =.34 (SD =.33).
Graph
Table 4. Comparison of Fit Statistics for Sales Data of Technology Pairs.
| 1. Laptop Versus Tablet Across Countries |
|---|
| Mean Errors | Laptop | Tablet | Overall |
|---|
| Model with F2 = F12 | Our Model with F2 ≠ F12 | Model with F2 = F12 | Our Model with F2 ≠ F12 | Model with F2 = F12 | Our Model with F2 ≠ F12 |
|---|
| Training | .0067 | .0073 | .0114 | .0106 | .0090 | .0090 |
| Test | .0196 | .0119 | .1491 | .0996 | .0843 | .0557 |
| Median Errors | Laptop | Tablet | Overall |
| Model with F2 = F12 | Our Model with F2 ≠ F12 | Model with F2 = F12 | Our Model with F2 ≠ F12 | Model with F2 = F12 | Our Model with F2 ≠ F12 |
| Training | .0036 | .0036 | .0116 | .0108 | .0068 | .0072 |
| Test | .0114 | .0046 | .1509 | .0900 | .0354 | .0163 |
| 2. DVD Versus BD players Across Countries |
| Mean Errors | DVD player | BD player | Overall |
| Model with F2 = F12 | Our Model with F2 ≠ F12 | Model with F2 = F12 | Our Model with F2 ≠ F12 | Model with F2 = F12 | Our Model with F2 ≠ F12 |
| Training | .0219 | .0045 | .0084 | .0017 | .0152 | .0031 |
| Test | .0294 | .0028 | .1165 | .0224 | .0730 | .0126 |
| Median Errors | DVD player | BD player | Overall |
| Model with F2 = F12 | Our Model with F2 ≠ F12 | Model with F2 = F12 | Our Model with F2 ≠ F12 | Model with F2 = F12 | Our Model with F2 ≠ F12 |
| Training | .0199 | .0034 | .0070 | .0008 | .0115 | .0022 |
| Test | .0231 | .0006 | .1019 | .0108 | .0505 | .0019 |
| 3. Digital Cameras Versus Smartphones Across Countries |
| Mean Errors | Digital Cameras | Smartphones | Overall |
| Model with F2 = F12 | Our Model with F2 ≠ F12 | Model with F2 = F12 | Our Model with F2 ≠ F12 | Model with F2 = F12 | Our Model with F2 ≠ F12 |
| Training | .0010 | .0003 | .0063 | .0022 | .0036 | .0013 |
| Test | .0008 | .0002 | .0214 | .0124 | .0111 | .0063 |
| Median Errors | Digital Cameras | Smartphones | Overall |
| Model with F2 = F12 | Our Model with F2 ≠ F12 | Model with F2 = F12 | Our Model with F2 ≠ F12 | Model with F2 = F12 | Our Model with F2 ≠ F12 |
| Training | .0008 | .0002 | .0018 | .0009 | .0012 | .0004 |
| Test | .0001 | .0001 | .0050 | .0028 | .0014 | .0002 |
1 Notes: This table represents the in-sample (training) and out-of-sample (test) error rates for sales data. The explanations are similar to those provided for Table 2. All the raw numbers for this analysis were standardized by the largest observed sales level by each country to provide for a valid comparison by countries. The median error rate refers to the in-sample and out-of-sample error rate across the different countries—using the median instead of the mean—to account for the fact that some countries may greatly influence the averages.
We next apply our model to the competition within contemporary, emerging technology pairs in the United States. The application leads to some preliminary generalizations: First, an increase in switchers over time is associated with the cannibalization of sales of Technology 1. Especially when switchers dominate dual users, this increase in switchers is associated with a sustained decline of sales of Technology 1, disrupting incumbents (Cases 1, 4, Web Appendix W4 Case WA1 on digital camera–smartphones). Second, an increase in dual users over time compared with switchers buys time for older technologies and enables them to grow despite the growth of new technologies (Case 2, Web Appendix W4 Case WA2 on VCRs–DVD players). Third, an increase in and dominance of leapfroggers over time is associated with the growth of Technology 2 (Cases 2, 3, Web Appendix W4 Case WA1). Incumbents underestimate or ignore these entirely new consumer segments. Christensen mentioned this, but we show how to estimate its size and evolution.
CDs were the dominant music format in 2004, and Apple iTunes' music store had been offering legal digital music downloads since 2003. Although most music executives then believed that people would pay for legal online music, big record labels were slow in adopting digital downloads. Some industry analysts predicted that digital music would not replace CDs because either potential buyers would use it only to sample music before buying CDs or it would only be the terrain of teenagers using iPods ([ 9]). According to analyst expectations, digital downloads and CDs could be expected to grow in tandem. A pertinent question in 2004 was whether digital downloads would eventually cannibalize and disrupt music CDs or if both would in fact grow in tandem.
We analyzed data on sales (in millions of units) of music CDs (CDs and CD singles from 1983 to 2018) and digital downloads (including singles, music albums, and music videos from 2004 to 2018) from the Recording Industry Association of America. The analysis from our model (Figure 5a) suggests that switchers (red line) dominated other segments right from the beginning, and this segment grew over the years. Both dual users (orange line) and leapfroggers (green line) tapered off by Year 5. Thus, contrary to the analysts' early expectations, our model indicates that the technologies did not coexist. The immediate high cannibalization by switchers was associated with and probably responsible for the relatively rapid decline of music CDs.
Graph: Figure 5a. Decomposition of music CDs and digital downloads in the United States.
The decline of music CDs from 2005 caused both record labels and music retailers to suffer. About 800 music stores closed in 2006 alone ([33]).
While PCs and laptops were the dominant older technologies, the tablet, which was in the works for many years, took off with the introduction of the Apple iPad. At the D8 conference in 2010, when Walt Mossberg asked Steve Jobs whether he thought the tablet will replace the laptop, Jobs replied "PCs are going to be like trucks. They are still going to be around, they are still going to have a lot of value, but they are going to be used by one out of X people...Is the next step the iPad? Who knows? Will it happen next year or five years from now or seven years from now? Who knows? But I think we're headed in that direction" ([27]). HP dominated the market for the older technologies, but in 2011, CEO Leo Apotheker wanted to get HP out of the PC business ([12]). "The tablet effect is real," Apotheker is reported to have said on the call with analysts, "consumers are changing how they use PCs." Apotheker was soon ousted, and the decision was reversed. A pertinent question at this time was whether tablets would eventually cannibalize and disrupt sales of laptops (and PCs).
We analyzed U.S. sales data of laptops and tablets from Passport Euromonitor. Figure 5b shows that while leapfroggers (green line) were the dominant segment, switchers (red line) dominated dual users (orange line) in the first ten years, vindicating HP's initial bleak assessment. However, soon after, dual users (using both technologies) dominated switchers. Our analysis indicates why tablets would not immediately disrupt the market for laptops. Apple gained by attracting dual users while also capturing an entirely new adopter segment base: leapfroggers.
Graph: Figure 5b. Decomposition of laptop and tablet sales in the United States.
Next, we examine the case of hybrid cars versus all-electric cars.[10] When Tesla first commercialized the electric vehicle, senior managers and analysts scoffed at the idea for three reasons: ( 1) no domestic firm had successfully introduced a new automobile for a hundred years; ( 2) automobile manufacturing is asset-intensive, making the break-even point unacceptably high; and ( 3) California was a state with very high labor costs, especially in comparison to Japan, Korea, and China. To resolve these issues, the critical question for the entrant and the incumbent was whether to invest in hybrid cars, all-electric cars, or both.
To answer this question, we use our model to decompose U.S. retail car sales (in thousands of units) of hybrids (including plug-in hybrids) and all-electric cars, obtained from the Transportation Energy Data Book in the time interval 2000–2018. Results in Figure 5c indicate that the growth of all-electric car sales is driven by a predominance of leapfroggers (green line), while switchers (red line) also grow, albeit slowly. Because all-electric cars represent an emergent technology, we have only eight years of new technology data up to 2018. We use data until 2018 and predict two years ahead. Our model predicts that sales of electric cars would cross sales of hybrids in 2020 (two years ahead), driven predominantly by leapfroggers.
Graph: Figure 5c. Prediction in the hybrid and electric car market in the United States.
Investors may be anticipating Tesla to dominate this race. Before the COVID-19 crisis overtook global markets, Tesla reached a market valuation of $102 billion in January 2020, trailing only Toyota ([31]). In July 2020, Tesla was worth more than Toyota ([30]). Investors are putting pressure on leading incumbents in gasoline and hybrids to invest in all-electric ([10]).
We next examine the emergent technology of ride-sharing services such as Uber and Lyft. Because the data for this case were available only for New York City, we limit our analysis to only this city. In many American cities, including New York, drivers need a medallion to operate a taxi, and the city issues a fixed number of them. The ride-sharing service Uber arrived in New York in 2011. Ride-sharing services match passengers with drivers typically through smartphone apps and provide estimated time of arrival, driver tracking, prepayment, and driver and passenger rating. Under pressure from taxi service providers, regulators and politicians sought to regulate or limit Uber's service. The question of relevance in 2012 was whether ride sharing would disrupt taxi services or if they would coexist.
We analyze data on trips (in thousands) per day from 2010 on yellow taxis and from 2015 on ride-sharing apps.[11] Our analysis (Figure 5d) reveals an increase in cannibalization over time on the rides for yellow taxis due to switchers to ride-sharing services (red line). However, leapfroggers (green line) and dual users (orange line) also contributed to the rise of ride sharing. Thus, ride-sharing services grew by also attracting a whole new segment of consumers. Anecdotally, it seems ride-sharing services have responded to the needs of customers that previously had difficulty availing themselves of taxi services, including low-income consumers and those in remote locations, as well as individuals who are comfortable with app-based technologies. Over time, switchers ended up dominating the other two segments for ride-sharing apps, contributing to the decline of yellow cabs.
Graph: Figure 5d. Decomposition of trips by yellow taxis and ride-sharing services.
The cannibalization of taxicabs by Uber, Lyft, and other such ride-sharing services led to a crisis for taxi services. Medallion prices plunged, and the stock of Medallion Financial (a publicly traded company that manages loans used to purchase taxi medallions in several large U.S. urban markets, including New York) had gone down nearly 49% since Uber raised its Series C funding, according to an analysis done by [ 3].
First, technological disruption is frequent, with dominant incumbents failing in the face of takeoff and growth of a new technologies. However, disruption is neither always quick nor universal because new technologies sometimes coexist as partial substitutes of the old technology. Our generalized model of diffusion of successive technologies can help marketers capture disruption or coexistence due to the presence of a rate of disengagement from the old technology (0–1), which can vary from the rate of adoption of the new technology (F12 ≠ F2).
Second, the model enables a superior fit to aggregate penetration and sales data over prior multigenerational models that do not include such flexibility (i.e., they force F12 to equal F2). Furthermore, an added benefit of the generalized model is that when the rate of disengagement from the old technology equals the rate of adoption of the new, it reduces to a model of multigenerational diffusion.
Third, we identify four adopter segments that account for competition between successive technologies from aggregate data: "leapfroggers" correlate with the growth of the new technology, "switchers" and "opportunists" account for the cannibalization of the old technology, and "dual users" account for the coexistence of both technologies.
Fourth, the generalized model can capture variations in segment sizes across technologies and markets. Leapfroggers form a dominant component of adopters in the early life cycle of a new technology in developing markets compared with other segments. Dual users form a dominant component of adopters in the early life cycle of a new technology in developed markets compared with other segments.
The major strategic implications of our findings are as follows: First, many established incumbents stumble or fail due to a takeoff of a new technology. Our model can provide important signals about disruption and survival by estimating cannibalization versus coexistence and forecasting the evolution of four critical consumer segments from aggregate data. Incumbents often wait until the market for the new technology is large enough to be profitable ([ 6]) before committing resources to its development. Our analysis suggests that senior managers of strategy and managers of new products should be careful not to underestimate cannibalization by switchers, especially when they dominate dual users, or growth of new technologies due to leapfroggers (especially in developing countries).
Second, despite its frequent occurrence, disruption is not a given when a new successive technology enters the market. Thus, managers do not have to make a stark choice between the two technologies. Disruption may be averted by effectively targeting dual users and by carefully examining factors driving the prolonged (co)existence of the old technology.
Third, the profit implications of leapfrogging and cannibalization vary depending on which firms market which technology. All segments represent a real gain for entrants, as the takeoff of the new technology is always a win. For the incumbent not introducing the successive technology (e.g., HP), the takeoff of that technology is always a loss. Particularly, if the incumbent firm markets the old technology and a new entrant markets the successive technology, then leapfrogging and switching represent a net loss to the incumbent and a net gain to the entrant. For the incumbent introducing the successive technology (e.g., Sony in DVD players), the takeoff of the successive technology is a win if competitors would have introduced it or if the successive technology has a higher margin than the old technology. Leapfroggers are an opportunity loss for incumbents, but switchers are a real loss to incumbents. If the incumbent firm markets both technologies and if the margin on the new exceeds the margin on the old, then switching and leapfrogging represent a net gain to the incumbent. However, if multiple firms market each technology or if margins vary, then the rate of leapfrogging and cannibalization becomes critical to ascertain profitability given the costs.
Fourth, marketers may be able to develop forecasts on the basis of early sales or penetration data of the successive technologies, or from similar contexts, to understand how these various segments may grow (or shrink) over time. Such an understanding can help guide a firm's managerial and economic resource allocation strategies across both technologies over time.
Table 5 summarizes the major strategic implications of this research.
Graph
Table 5. Adopter Segments, Firm Type, and Market Outcomes in the Presence of Multiple Technologies.
| AdopterSegments | Market Outcome on the Introduction of Technology 2 | Firm Type |
|---|
| Incumbent Marketing Technology 1a | Entrant Marketing Technology 2a | Incumbent Marketing Technology 2b |
|---|
| Leapfroggers | Market growth of successive technology | Neutral | Win | Win |
| Opportunists | Cannibalization of old technology | Lose | Win | Neutral |
| Switchers | Cannibalization of old technology | Neutral/losec | Win | Neutral |
| Dual users | Market growth of both technologies | Neutral | Win | Win |
- 2 Notes:
- 3 a Assumes that the incumbent (or incumbents) dominated the market for the old technology and entrants pioneered the new technology.
- 4 b Assumes that the incumbent chooses to enter the new technology market rather than wait on the sidelines.
- 5 c Neutral for adoption/lose if sales is considered.
This study suffers from several limitations. First, we used aggregate data to test the model because they were abundantly available. As managers and researchers get access to richer, individual customer-level data, they may be able to provide better support to our modeling insights. Moreover, disaggregate choice models can be utilized to address issues such as cannibalization. However, macro diffusion models still have the ability to produce useful macro-level conclusions in ways that micro approaches sometimes cannot. Second, we consider a demand-based view of disruption in proposing the typology of adopter segments. Future research could complement these typologies and data sets with surveys to determine the characteristics of adopters of the new technology versus those who stay with the old technology, as well as what factors influence the size of adopter types. Third, an incumbent may respond to the new technology by making changes in variables such as price, and the omission of such control variables may violate some of the assumptions of the model. All these remain fruitful areas for future research.
Supplemental Material, Web_Appendix_R2 - Leapfrogging, Cannibalization, and Survival During Disruptive Technological Change: The Critical Role of Rate of Disengagement
Supplemental Material, Web_Appendix_R2 for Leapfrogging, Cannibalization, and Survival During Disruptive Technological Change: The Critical Role of Rate of Disengagement by Deepa Chandrasekaran, Gerard J. Tellis and Gareth M. James in Journal of Marketing
Footnotes 1 Peter Danaher
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study benefits from the Christian and Mary Lindback grant for minority and junior faculty, a grant from Don Murray to the Center for Global Innovation, Marshall School of Business, University of Southern California, and a research grant from the Institute on Asian Consumer Insight.
4 Online supplement: https://doi.org/10.1177/0022242920967912
5 Disengagement relates to technological substitution and can be distinguished from churn, which refers to brand switching (e.g., [23]), and from disadoption, wherein the consumer leaves the category entirely for various product and nonproduct reasons ([22]).
6 The measurement unit is market penetration or the percentage of households owning a technology. Penetration refers to the number of adopters divided by the number of households or inhabitants (depending on the data available for each technology pair).
7 All the raw numbers for this analysis were standardized using the largest observed penetration level within each country to provide for a valid comparison across countries.
8 Each year, the World Bank revises the analytical classification of the world's economies on the basis of estimates of gross national income per capita for the previous year and classifies countries into low-income, lower-middle-income, upper-middle-income, and high-income countries.
9 To determine early sales data more accurately in each country, we compared the earliest year of sales data with the corresponding penetration data from Euromonitor. Whenever penetration data started earlier, we used a simple proportion formula to calculate sales for earlier years.
Hybrid electric vehicles are powered by an internal combustion engine in combination with one or more electric motors that use energy stored in batteries, combining the benefits of high fuel economy and low tailpipe emissions with the power and range of conventional vehicles. All-electric vehicles use a battery pack to store the electrical energy that powers the motor. All-electric vehicles are zero-emission vehicles because they produce no direct exhaust or emissions.
https://toddwschneider.com/dashboards/nyc-taxi-ridehailing-uber-lyft-data/
References Bass Frank M.. (1969), " A New Product Growth Model for Consumer Durables ," Management Science , 15 (5), 215 – 27.
Bower Joseph L. , Christensen Clayton M.. (1995), " Disruptive Technologies: Catching the Wave ," Harvard Business Review , 73 (1), 43 – 53.
CBInsights (2015), " The 'Uber Effect' Is Crushing Taxi Medallion Prices and Spilling over into Public Markets ," (October 1), https://www.cbinsights.com/research/public-stock-driven-uber/.
Chandrasekaran Deepa , Tellis Gerard J.. (2007), " A Critical Review of Marketing Research on Diffusion of New Products ," Review of Marketing Research , 3 (1) 39 – 80.
Chandrasekaran Deepa , Tellis Gerard J.. (2008), " Global Takeoff of New Products: Culture, Wealth, or Vanishing Differences? " Marketing Science , 27 (5), 844 – 60.
Christensen Clayton M.. (2013), The Innovator's Dilemma: When New Technologies Cause Great Firms to Fail. Boston : Harvard Business Review Press.
Danaher Peter J. , Hardie Bruce G. S. , Putsis William P. Jr. (2001), " Marketing-Mix Variables and the Diffusion of Successive Generations of a Technological Innovation ," Journal of Marketing Research , 38 (4), 501 – 14.
Eberhard Martin. (2006), " Hybrids, Plug-in or Otherwise ," Tesla Blog (September 27), https://www.tesla.com/blog/hybrids-plug-or-otherwise.
Emigh Jacqueline. (2008), " Analyst: Music CDs Will Give Way to Digital Downloads by 2012 ," BetaNews (accessed July 20, 2020), https://betanews.com/2008/02/19/analyst-music-cds-will-give-way-to-digital-downloads-by-2012/.
Foldy Ben. (2020), " Auto Makers Charge Ahead with Electric-Vehicle Plans " The Wall Street Journal (July 19, 2020), https://www.wsj.com/articles/auto-makers-charge-ahead-with-electric-vehicle-plans-11595156400.
Goldenberg Jacob , Oreg Shaul. (2007), " Laggards in Disguise: Resistance to Adopt and the Leapfrogging Effect ," Technological Forecasting and Social Change , 74 (8), 1272 – 81.
Goldman David. (2011), " HP Kills TouchPad, Looks to Exit PC Business ," CNN Money (August 18), https://money.cnn.com/2011/08/18/technology/hp_pc_spinoff/index.htm.
Guo Zhiling , Chen Jianqing. (2018), " Multigeneration Product Diffusion in the Presence of Strategic Consumers ," Information Systems Research , 29 (1), 206 – 24.
Jiang Zhengrui , Jain Dipak C.. (2012), " A Generalized Norton–Bass Model for Multigeneration Diffusion ," Management Science , 58 (10), 1887 – 97.
Jun Duk Bin , Park Yoon S.. (1999), " A Choice-Based Diffusion Model for Multiple Generations of Products ," Technological Forecasting and Social Change , 61 (1), 45 – 58.
Kim Namwoon , Chang Dae Ryun , Shocker Allan D.. (2000), " Modeling Intercategory and Generational Dynamics for a Growing Information Technology Industry ," Management Science , 46 (4), 496 – 512.
Koh Byungwan , Il-Horn Hann , Srinivasan Raghunathan. (2019), " Digitization of Music: Consumer Adoption Amidst Piracy, Unbundling, and Rebundling ," MIS Quarterly , 43 (1), 23 – 45.
Kubota Yoko. (2012), " Toyota Drops Plan for Widespread Sales of Electric Car ," Reuters (September 24), https://www.reuters.com/article/us-toyota-electric/toyota-drops-plan-for-widespread-sales-of-electric-car-idUSBRE88N0CT20120924.
Ladron-de-Guevara Antonio , Putsis William. (2015), " Multimarket, Multiproduct New Product Diffusion: Decomposing Local, Foreign, and Indirect (Cross-Product) Effects ," Customer Needs and Solutions , 2 (1), 57 – 70.
Lam Shun Yin , Shankar Venkatesh. (2014), " Asymmetries in the Effects of Drivers of Brand Loyalty Between Early and Late Adopters and Across Technology Generations ," Journal of Interactive Marketing , 28 (1), 26 – 42.
Lehmann Donald R. , McAlister Leigh , Staelin Richard. (2011), " Sophistication in Research in Marketing ," Journal of Marketing , 75 (4), 155 – 65.
Lehmann Donald R. , Parker Jeffrey R.. (2017), " Disadoption ," AMS Review , 7 (1/2), 36 – 51.
Libai Barak , Muller Eitan , Peres Renana. (2009), " The Diffusion of Services ," Journal of Marketing Research , 46 (2), 163 – 75.
Mahajan Vijay , Muller Eitan. (1996), " Timing, Diffusion, and Substitution of Successive Generations of Technological Innovations: The IBM Mainframe Case ," Technological Forecasting and Social Change , 51 (2), 109 – 32.
Muller Eitan. (2020), " Delimiting Disruption: Why Uber Is Disruptive but Airbnb Is Not ," International Journal of Research in Marketing , 37 (1), 43 – 55.
Norton John A. , Bass Frank M.. (1987), " A Diffusion Theory Model of Adoption and Substitution for Successive Generations of High-Technology Products ," Management Science , 33 (9), 1069 – 86.
Paczkowski John. (2010), " Apple CEO Steve Jobs Live at D8: All We Want to Do Is Make Better Products ," All Things Digital (June 1), http://allthingsd.com/20100601/steve-jobs-session/.
Paulson Courtney , Luo Lan , James Gareth M.. (2018), " Efficient Large-Scale Internet Media Selection Optimization for Online Display Advertising ," Journal of Marketing Research , 55 (4), 489 – 506.
Putsis William P. Jr , Balasubramanian Sridhar , Kaplan Edward H. , Sen Subrata K.. (1997), " Mixing Behavior in Cross-Country Diffusion ," Marketing Science , 16 (4), 354 – 69.
Roberson Bill. (2020), " Tesla Takes over Top Spot from Toyota as World's Most Valuable Carmaker ," Forbes (July 1), https://www.forbes.com/sites/billroberson/2020/07/01/tesla-takes-over-top-spot-from-toyota-as-worlds-most-valuable-carmaker/.
Roper Willem. (2020), " Tesla's Electric Rise in Value ," Statista (August 7), https://www.statista.com/chart/20606/tesla-market-valuation-record/.
Shi Xiaohui , Fernandes Kiran , Chumnumpan Pattarin. (2014), " Diffusion of Multigenerational High-Technology Products ," Technovation , 34 (3), 162 – 76.
Smith Ethan. (2007), " Sales of Music, Long in Decline, Plunge Sharply ," The Wall Street Journal (March 21), https://www.wsj.com/articles/SB117444575607043728.
Sood Ashish , James Gareth , Tellis Gerard J. , Zhu Ji. (2012), " Predicting the Path of Technological Evolution: Testing SAW Versus Moore, Bass, Gompertz, and Kryder ," Marketing Science , 31 (6), 964 – 79.
Sood Ashish , Tellis Gerard J.. (2005), " Technological Evolution and Radical Innovations ," Journal of Marketing , 69 (3), 152 – 68.
Sood Ashish , Tellis Gerard J.. (2011), " Demystifying Disruption: A New Model for Understanding and Predicting Disruptive Technologies ," Marketing Science , 30 (2), 339 – 54.
Stremersch Stefan , Muller Eitan , Peres Renana. (2010), " Does New Product Growth Accelerate Across Technology Generations? " Marketing Letters , 21 (2), 103 – 20.
Tellis Gerard J.. (2013), Unrelenting Innovation: How to Create a Culture of Market Dominance. San Francisco : Jossey-Bass.
Tellis Gerard J. , Franses Philip Hans. (2006), " Optimal Data Interval for Estimating Advertising Response ," Marketing Science , 25 (3), 217 – 29.
~~~~~~~~
By Deepa Chandrasekaran; Gerard J. Tellis and Gareth M. James
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 81- Leveraging Cofollowership Patterns on Social Media to Identify Brand Alliance Opportunities. By: Malhotra, Pankhuri; Bhattacharyya, Siddhartha. Journal of Marketing. Jul2022, Vol. 86 Issue 4, p17-36. 20p. 1 Color Photograph, 5 Diagrams, 4 Charts, 6 Graphs, 1 Map. DOI: 10.1177/00222429221083668.
- Database:
- Business Source Complete
Leveraging Cofollowership Patterns on Social Media to Identify Brand Alliance Opportunities
The use of cobranding and brand extension strategies to access new markets has been a topic of significant interest. However, surprisingly few studies have examined cross-category connections of brands using publicly available digital footprints. In this study, the authors introduce a new, scalable automated approach for identifying potential cobranding and brand extension opportunities using brand networks derived from publicly available Twitter followership data. The digital user–brand relationship, established through followership activity, is regarded as an expression of interest toward the brand. Common followership patterns between brands are then extracted to capture cointerest between those brands' audience. By utilizing the cointerest patterns, the approach aims to derive cross-category brand–brand and brand–category connections, which can serve as important measures for assessing cobranding and extensions opportunities. This article introduces a new construct, transcendence, which measures the extent to which a brand's followers overlap with those of other brands in a new category. The analysis is conducted at different points in time to help managers track shifts in brand transcendence.
Keywords: cross-category; brand networks; asymmetry; cobranding; brand extensions; social media; social networks analysis; Twitter
Cobranding is a brand alliance strategy to bolster reach, awareness, and sales potential by tapping the prospective customers of partnering brands. Many types of cobranding schemes exist in the marketplace, including joint advertising campaigns (e.g., ads depicting the joint consumption of Coca-Cola and McDonald's), cause–brand alliances (e.g., UNICEF and Target), bundling (e.g., streaming deals that include joint Hulu and Spotify subscriptions), and cobranded products (e.g., Louis Vuitton launching an exclusive luggage line for BMW). Cobranding strategies enable brand extensions, with managers leveraging the existing brand names of their partners to enter new markets and categories ([14]). Cobranding is increasingly viewed as a valuable marketing strategy and has been shown to increase awareness, quality, market value, and brand equity ([ 8]; [46]). Although marketers have been leveraging the synergistic benefits of cobranding for decades, surprisingly little empirical research has tried to identify potential cobranding alliances using modern digital approaches. Most of the existing empirical research either uses observations from fast-moving consumer goods categories ([ 8]; [14]; [19]) or conducts analyses within a single category, such as camcorders in [24], car brands in [34], and LED TVs in [42]. Similarly, [36] and [35] use recommendation hyperlinks between Amazon web pages to create a large-scale network of books and demonstrate the value of shared purchasing patterns.
Obtaining broader insights into the identification of cobranding opportunities across diverse categories would generate relevant and meaningful information for brand owners. As [43] notes, "By mashing up two bona fide brands, especially in diverse industries, the impact can be exponential." For instance, a well-known cobranding deal between Starbucks and Spotify—two seemingly unrelated brands—enabled both brands to cross-promote their products and grow their customer base. By providing premium coffee-shop music, Starbucks incentivized Spotify users to join its loyalty program. In return, Spotify grew its user base through Starbucks' offer of a free coffee upon joining. Having knowledge about relevant cross-category brand connections is crucial to brand owners ([12]); however, there is little or no research on identifying these broader cross-category effects using current digital approaches.
This article introduces a new, scalable approach for generating cross-category branding insights using implicit brand networks on social media. The cross-category branding insights are revealed in the form of brand–brand and brand–category connections, which can serve as important measures for assessing cobranding and extensions opportunities. Unlike traditional social networks, which involve explicit interaction between the participating entities,[ 5] edges within a brand network are implicit (or tacit) and arise due to common followership between brands. [45] note the relevance of these tacit connections to decision making. This idea has been studied previously within the domain of collaborative filtering ([45]). Implicit networks, which condense the vast digital interest space of millions of users into a parsimonious form, provide direct insight into the digital ecosystem and are the subject of increasing research attention across domains ([45]). In this study, the cross-category connections of a focal brand in the implicit network are leveraged to help brand managers identify cobranding and extension opportunities.
The article introduces a new construct, called "brand transcendence," which is defined in the context of a large ecosystem of brands belonging to different categories. The transcendence of a brand into a new category is the extent to which its followers overlap with those of other brands in the new category. From a managerial perspective, this study provides an automated approach for identifying cross-category cobranding opportunities based on user cointerest, which is measured through overlap of brands' followers on Twitter. Importantly, the cointerest patterns captured through common followership do not necessarily reflect overlapping brand associations or guarantee brand fit, which is traditionally measured using the similarity of brand personality dimensions ([48]). However, such patterns are indicative of common tastes or interests among social media users. Following [39], we consider that the composition of a brand's follower base represents the tastes (and likes) of its audience. Thus, the greater the network overlap between two brands, the greater the similarity in tastes and interests between those brands' audiences. Taking these principles together, we study the transcendence of a brand into a new category based on the extent to which its followers overlap with those of other brands in the new category. Our approach also identifies central brands that have strong and consistent connections within their own category ([ 9]), with "centrality" being defined as the extent to which a brand's followers overlap with other brands in its own category.
By incorporating directionality into the network edges, we also capture the asymmetric relationships between brand pairs, which help identify brands that may potentially benefit more from a cobranding alliance. We outline how cross-category connections can provide both brand–category and brand–brand insights, depending on a brand's marketing goals (i.e., extension vs. cobranding). For instance, brand–category connections capture the transcendence of brands into new categories and show that certain categories are more viable for extensions than others. Brand–brand connections, in contrast, provide a more granular view of transcendence by revealing the individual brands that are suitable for cobranding. As user–brand relationships on social media may change over time, this article analyzes the brand network in both 2017 and 2020. This helps visualize the fluctuations of brand connections over time and investigate the impact of such fluctuations on cobranding alliances. Understanding whether critical connections with certain brands or prospective categories have waned helps managers promptly identify the problem and take appropriate action. Similarly, identifying new connections that have formed over time illustrates how past marketing actions can impact a brand's transcendence in users' minds.
Cross-category connections revealed through the network can be used to both assess the effectiveness of previous marketing campaigns and discover new alliance opportunities. For example, Bud Light's connection to Pepsi reflects the cointerest patterns between the two brands and, thus, affirms the effectiveness of joint marketing campaign led by the two brands previously. Similarly, Sierra Nevada's strong connections with travel and technology brands (e.g., Southwest Airlines, Discovery, SpaceX, Microsoft) highlight strong cointerest with these brands and present new cobranding opportunities that may not yet be known to its owners. We provide examples of both scenarios using information from external industry sources. Another practical application of our method is competitor analysis, which can help managers identify the differentiating connections of brands with respect to their competitors and gauge the type of users their competitors attract.
Finally, we validate the findings of our model against external survey ratings and conduct extensive robustness checks, including network simulations, to ensure that our final network estimates are not biased by fake users or bots. Consistently high correlation between our automated approach and external survey ratings affirms the validity of our methodology for identifying cross-category brand–brand and brand–category connections. Overall, the core contribution of this study is a new digital approach to analyzing audiences' interests across a broad brand ecosystem. The cross-category insights generated by this approach can help researchers and practitioners identify nontraditional branding opportunities that are difficult to infer from traditional survey-based approaches. From a managerial perspective, our brand network can efficiently and cost-effectively generate cross-category insights, given that most of the data collection and network analyses are automated. Furthermore, as our approach uses information that is publicly available on social media, it is easily scalable to a large number of brands, with the resulting network structures reflecting the preferences of a diverse set of users. In the next section, we discuss relevant studies in the marketing literature and describe how our work contributes to the field.
Researchers regard cobranding as a source of competitive advantage that helps brands differentiate themselves, gain consumer trust, acquire new channels of distribution, and enter new markets ([47]). Brand extensions (i.e., leveraging the existing brand's name to enter into new categories) are another widely adopted strategy for firms entering new markets ([ 1]). Brand extensions provide greater quality assurance to customers who are familiar with the original brand, reduce the costs of distribution, and increase the efficiency of promotional expenditure ([ 1]). Both brand extensions and cobranding strengthen the focal brand and reinforce customers' value perceptions of the new product ([20]).
Naturally, identifying underlying brand-to-brand connections on the basis of users' cointerests may be a key that enables brand managers to discover potential cobranding and extension opportunities. Most studies in this domain have used surveys ([ 5]). Although collecting input from prescreened participants is desirable, recruiting and maintaining a pool of such participants may be unfeasible due to cost or other constraints ([11]). Recent advances in social network analysis have enabled a wide range of scalable solutions that go beyond conventional market research methods. Although previous studies have considered the identification of brand-to-brand connections based on digital user traces, such research has been restricted to brand (or product) relationships within a single category. For example, [34] focus mainly on intracategory connections to create competitive market structures for car brands. Using survey approaches, [13] obtain intracategory maps for centrality and distinctiveness. Finally, [42] develop mapping methods to visualize large market structures within a single category (i.e., LED TVs).
The marketing literature acknowledges the importance of cross-category brand connections for generating extensions, licensing, and cobranding deals ([22]; [41]). However, only limited empirical work has been done in this area. This article introduces an automated, scalable approach for identifying cross-category brand cointerest patterns by leveraging the cofollowership data on Twitter. The use of common followership data on Twitter in our analysis follows the recent work of Culotta and Cutler (2016). However, while [11] aimed to derive perceptual attribute ratings from Twitter followership data, the goal of this work is to investigate asymmetric cross-category brand transcendence over time. Further, unlike Culotta and Cutler, our large-scale network approach does not require any supervised knowledge on exemplars and uses categorical affiliations of brands to infer brand perceptions on transcendence and centrality. Specifically, the inclusion of network-derived measures enables us to study both within-category competition and across-category complementarity between brands. Table 1 presents the unique contribution of this research compared with previous network studies.
Graph
Table 1. Comparison of This Study with Previous Network Studies in Marketing.
| Study | Focal Contribution | Data Source | Output | Asymmetry | Survey Validation | Addresses Bots/Fake Users Online | Competitor Analysis | Analysis Presented over Time |
|---|
| Netzer et al. (2012) | Create competitive structure maps using text mining and network analysis | Comentions on online discussion forum | Market structures within a category | No | Yes | No | No | No |
| Culotta and Cutler (2016) | Propose a methodology for inferring attribute-specific brand perceptions | Cofollowership data on Twitter | Perceptual maps for a set of predefined exemplars | No | Yes | No | No | No |
| Ringel and Skiera (2016) | Develop mapping methods for visualizing complex market structures | Consideration sets from online search | Mapping solutions for large complex market structures within a category of >1,000 products | Yes, using conditional probability | No | Yes, eliminating implausible clickstreams | Yes | No |
| Current study | Use implicit brand networks to infer asymmetric cross-category brand connections over time. | Cofollowership data on Twitter over time | Cross-category connections maps at two levels: brand–category brand–brand | Yes, using conditional probability | Yes | Yes, using network rewiring | Yes | Yes |
Our approach to identifying cobranding and brand extension opportunities harnesses the digital cofollowership patterns between brands. Survey research has shown that users follow brands on social media with the intention of purchasing a product or learning more about their favorite brands ([31]; [40]). Aspirations can also motivate consumers to follow brands ([ 3]). This digital user–brand relationship, which is established through followership activity, can be interpreted as an expression of affinity for the brand ([27]; [33]). Alternatively, this relationship can be viewed through the lens of homophily, meaning that people tend to associate with those who are similar to them in socially significant ways ([32]). This is further supported by consumer research studies ([ 6]; [10]), which show a strong relationship between a brand's image and characteristics and the identities of its followers.
Similarly, [39] find that the composition of one's follower base represents the tastes (and likes) of their audience. Thus, the more network overlap (i.e., common followers) between two entities, the greater the similarity of tastes and interests among those entities' audiences ([39]). [ 2] find that common friends (i.e., common mutual followers) have a positive effect on the adoption of an application on Facebook. Similarly, we expect that brands that share a high number of followers on Twitter have a user composition that represents similar tastes or interests (e.g., the partnership between GoPro and Red Bull, which leveraged the shared affinities of their common followers: action, adventure, and fearlessness). Given that individuals primarily follow a brand because they like its products and that most followers are customers ([40]), brands having more common followers implies that their customers may have complementary consumption patterns (e.g., the Starbucks–Spotify partnership, which facilitated the complementary consumption of Spotify music and Starbucks coffee). Building on these theoretical principles, we study the cobranding candidates of a focal brand based on the extent to which its followers overlap with the followers of another given brand (and/or category) of interest.
Some of the greatest brands in the world have defied category norms and transcended their initial market boundaries ([23]). For example, in 1995, Amazon positioned itself as "Earth's biggest bookstore," and its success with books enabled it to transcend its origins to become a leader in e-commerce. Although all brands theoretically operate within their categorical boundaries, such boundaries are often considered malleable ([ 7]). Important cobranding and extension opportunities can be missed if managers are not aware of connections that are relevant to brands in other categories ([ 5]). The brand network provides a solution to this problem by relying on a brand's social connections on Twitter to infer category-specific brand connections.
At a high level, our proposed algorithm extracts the category-specific connections of a brand by exploiting the overlap in brand followers on Twitter. Whereas some brands may possess strong connections within their own category, others may have diverse connections across new categories. This article introduces a new construct, transcendence, which measures the extent to which a brand's followers overlap with those of other brands in a new category. The transcendence of a nonsports brand along any given category—for example, say sports—is based on the extent to which its followers overlap with those of other brands in the sports category. Further, to measure the connections not shared by the overall brand category, we calculate "net transcendence" as the deviation of a brand's own idiosyncratic connections from its category average. Net transcendence is more informative than raw transcendence because it ignores the cross-category connections that are generic to the category and identifies those that are intrinsic to the brand itself.
Lastly, in addition to transcendence, there are brands that possess strong connections within their own category. These brands can be viewed as central. The concept of centrality (or typicality) bears direct relation to a brand's probability of recall, consideration, and choice among consumers' minds ([28]). Such central brands are those that come first to consumers' mind and serve as reference points in their categories ([13]). In the next section, we describe how the brand network is generated from followership patterns on Twitter and lay out important network details.
The key contribution of this article is the introduction of an automated framework for inferring cross-category branding insights using implicit brand networks derived from social media. With their ability to provide a direct digital window into the interests of millions of social media users, implicit brand networks can help mangers identify nontraditional branding opportunities that would otherwise be hard to perceive. In this section, we generate implicit brand networks using brand communities on Twitter and outline important network details.
Drawing from the notion that the social signal of "who follows a brand" provides a strong reflection of brand image ([11]), we use a set of 507 brands' Twitter accounts as the basis for our analysis. We select the most active Twitter brand accounts based on followership data from the social media directory FanPageList.com. We use Twitter's public application programming interface to collect the brands' lists of followers for 2017 and 2020. We manually verify that all Twitter handles correspond to the official brand account. Overall, the data set consists of brands from many major categories: airlines, luxury goods, retail, automotive, sports, technology, dining, food and beverages, lodging, media, travel, cruises, and beer. Each brand is assigned to a specific category based on the basic or superordinate category-level analyses ([28]).[ 6] To prevent bots or spam accounts from influencing the network analysis, all Twitter brand accounts included in the analysis are manually audited using the audience intelligence website SparkToro.[ 7] Furthermore, as we discuss in the section on robustness checks, we conduct network simulations to ensure that our final network estimates are not biased by such bots.
The next step is to extract the common followers between all brand pairs. The raw brand network is a weighted edge list, defined as 〈bi, bj, wij〉, where bi and bj are individual brands or nodes and wij is the common followers between those brands. If Fi and Fj represent the list of Twitter accounts following brands bi and bj, then an edge between two nodes is created if Fi ∩ Fj > 0. Alternatively, the weighted edge list can be represented as a weighted adjacency matrix Aij where
Graph
Overall, we extract two brand networks: one for 2017 and one for 2020. The original brand networks are highly dense, with common followers between almost all pairs of brands. The numbers of common followers vary from a few hundred to more than a million users. Although it is possible to work with such dense networks, valuable information may be lost due to the redundancy generated by the large number of connections ([44]). Further, connections based on too few followers may not indicate significant connectivity. Given the wide heterogeneity in raw edge weights (i.e., the numbers of common followers), we next aim to extract the truly relevant brand–brand connections.
A common way to extract a relevant network structure is by applying a global threshold to remove the edges with weights below a particular cutoff. This, however, can destroy the multiscale properties of the brand network. Instead, we use a disparity filter ([44]), which is a filtering algorithm for multiscale networks, to obtain a reduced but more meaningful representation of the network. This method preserves the important edges present at all scales by locally identifying the statistically relevant weights at the node level. The statistically relevant edges (at a given significance level; e.g., α) represent a significant deviation from a null model of uniform randomness. Thus, smaller brands with fewer common followers are not ignored during the network reduction process. Although the current analysis focuses on Twitter brand accounts with between a few thousand and more than a million followers, this method could also be applied to smaller brands with fewer than 1,000 followers. Following [44], we use the commonly specified significance level α = .05 to extract the important connections in the brand network. The filtered networks for 2017 and 2020 consist of roughly 14,000 edges between brands. Although we use the disparity filter to obtain a filtered representation of the original network, there are alternative information-filtering algorithms available in the network science literature, including the global threshold and global statistical significance filters. In Web Appendix A, we revisit these alternative methods and discuss our rationale for choosing the disparity filter.
Brand community sizes can vary both within and across categories. Brands with large brand communities (e.g., Chanel, Microsoft, Starbucks) tend to have more common followers than those with smaller communities. The normalization of edge weights is required to account for this variance. Thus, we use the conditional probability measure ([42]) to compute new network weights that not only normalize the effects of brand size but also account for asymmetry between brand pairs. Asymmetry between brand pairs may occur when the degree of connection between any two brands is unequal (i.e., the connection from A to B is not equal to the connection from B to A) ([15]). Ignoring the directionality of brand connections can lead to inaccurate estimates of consumer brand knowledge ([16]). We observe many cases of associative asymmetry in our brand network and use conditional probability to account for such scenarios. For instance, Figure 1 shows the cross-category connection between Starbucks and Stella Artois. A large percentage of Stella Artois fans are interested in Starbucks, and the outgoing directional strength is almost.20. However, fewer Starbucks fans are interested in Stella Artois, and the outgoing directional strength is comparatively much lower, at.0009. Incorporating directionality in the network reveals this crucial information, which is not visible in a simple, undirected, weighted network. Mathematically, the conditional probability measure calculates the normalized edges between any two brands A and B as
Graph
where the numerator is the number of common followers between brands A and B and the denominator is the number of followers of the focal brand.
Graph: Figure 1. Calculating asymmetry between brand pairs.
Figure 2 shows the entire brand network structure for 2020 using the dimensional reduction algorithm, t-distributed stochastic neighbor embedding (t-SNE). The t-SNE algorithm yields a two-dimensional undirected representation of the brand network, with the distance between brands in the t-SNE space being proportional to the mean conditional probabilities between brands. The colors of the brands correspond to their category affiliation. Interestingly, while most automotive brands in Figure 2 are well-distanced from other nonautomotive brands, Tesla is positioned in the technology category. A similar pattern is observed for certain retail brands such as Adidas and Reebok, which are closer to the sports group than to other brands in their category. Individual brand constructs on transcendence, as described in the next section, help reveal specific cross-category connections for a given brand.
MAP: Figure 2. t-SNE map of the undirected brand network.[ 8]
This section outlines the process of identifying centrality and transcendence by exploiting the connections of a brand within the network. For any given brand, the first step is to disentangle its connections across the main categories: airlines, luxury goods, retail, automotive, sports, technology, dining, food and beverages, lodging, media, travel, cruises, and beer. The algorithm then computes the weighted out-degree centrality of a brand across these categories. In weighted networks, out-degree centrality or node strength is commonly calculated as the sum of weights emanating from a focal node to all its connections ([ 4]). However, to account for the strength of edge weights and the number of connections of a focal node, [37] propose a new measure of weighted degree centrality:
Graph
( 1)
where is the degree of the focal node (i.e., the number of connections), is the node strength (i.e., the sum of the weighted connections), and is the tuning parameter from 0 to 1. For example, following [37], the transcendence of a given nonsports brand into the sports category is calculated as the function of its number and strength of outgoing connections to all other brands in the sports category.
More formally, the set of brands in the network can be represented as B, where any individual brand . Brand categories, G, are subsets of B, such as G ⊆ B.[ 9] The transcendence of focal brand b onto a new category G is evaluated as
Graph
( 2)
where is the number of outgoing edges from brand b to all k brands in category G, is the sum of weighted edges from brand b to all k brands in category G, and gives the number of brands in category G. Dividing by the total number of brands in a category, , helps ensure that large categories with many brands do not dominate the analysis. Following [37], we set to.5 to place equal importance on a brand's number of connections and the weight of those connections.
Furthermore, considering that brands whose followers tend to follow many other brands may inflate the network constructs, we divide our transcendence construct by the total degree of a brand in the network (i.e., its number of connections to other brands). Intuitively, brands whose followers tend to follow many brands have a higher degree than brands whose followers follow fewer brands. Thus, in the transcendence construct, the sum of the numerator increases with each new connection of a brand in the network. With the objective of normalizing for brands that inherently have higher aggregate transcendence due to higher degree in the network, we divide our final transcendence construct by the total degree of a brand. Thus, the final transcendence construct becomes
Graph
( 3)
where the additional variable in the denominator, is the total number of connections of a brand in the network.
Furthermore, in our transcendence construct, brand b's connections within its own category relate to its centrality ( ): the higher the strength of these connections, the more central the brand is in its own category. As noted previously, the notion of centrality is directly related to a brand's probability of being recalled, considered, and chosen by consumers ([28]). We multiply the centrality construct by the size of the focal brand's community (i.e., its number of followers) to account for the brand's popularity among users. Thus, the final centrality construct is
Graph
( 4)
where is brand b's connections within its own category and is its number of followers. Given that the values of vary from a few hundred to more than a million users, we use its scaled value. Given a set of nonoverlapping categories G1, G2, ..., Gp, the transcendence of brand b across p categories is a 1 × p-dimensional vector:
Graph
( 5)
The transcendence vector of a brand can also be analyzed with respect to its competitors in the category. The 1 × p-dimensional vector can be further extended into an n × p matrix, where n rows represent brands (i.e., ) and p columns represent transcendence across the p categories, as shown in Figure 3.
Graph: Figure 3. The transcendence matrix, tbG for n brands across p categories.
The average category connections (i.e., connections emanating from one category to another category ) are then calculated as follows:
Graph
( 6)
where . This formula measures the average transcendence of brands in one category (for example, ) into another category To separate a brand's unique connections ( ) from its category average ( ), we calculate the net transcendence of brand b into category as follows:
Graph
( 7)
where and . A positive value for indicates that the brand's transcendence is above the category average, while a negative value indicates that its transcendence is below average. As in the raw transcendence vector , the net transcendence vector of a brand b across p categories is
Graph
( 8)
The 1 × p-dimensional vector can be further extended into an n × p matrix, where rows represent n brands in a category and p columns represent the net transcendence of brands across the p categories, as shown in Figure 4.
Graph: Figure 4. Net transcendence matrix of n brands across p categories.
Figure 4 provides a more comprehensive view of the competitive landscape of a particular brand by highlighting its cross-category connections as well as those of its competitors. In the next section, we discuss the results of these analyses and identify their key managerial implications.
Depending on the business objective, the category-specific connections generated by the brand network can be visualized on two different levels: brand–category and brand–brand. Using examples from the automotive and beer categories, in this section we present our results on these two levels and discuss the managerial implications of our findings. First, we note the brand–category connections of automotive brands and identify the categories suitable for brand extensions. Second, we focus on brand–brand connections and discuss how asymmetry can be leveraged to attain more nuanced insights into the expected benefits of cobranding. Third, we highlight the network's ability to capture changes that occurred between 2017 and 2020, given that individual brand–brand connections may change over time. Finally, we discuss how the brand network described in this article can help managers identify the differentiating category-specific connections of brands with respect to their competitors and gauge the type of users their competitors attract.
To identify the brand–category connections of automotive brands in 2020, we study the net transcendence matrix, , as shown in Figure 5. All column values have been scaled, with positive values shown in red coloring and negative values shown in blue on the heatmap. For every brand (n)–category (p) relationship in the heatmap, values closer to dark red indicate a stronger perceived relationship between a brand and category. The stronger the relationship between the brand and category, the greater the user cointerest between that brand and category. As discussed previously, the cointerest patterns captured through the analysis of common followership do not necessarily guarantee brand fit, which is typically based on the similarity of brand personality dimensions ([48]). Instead, values in the transcendence matrix reveal cointerest patterns that enable managers to explore potential extension and cobranding opportunities that are difficult to infer from traditional survey-based approaches. For instance, the audience of the car brand Mercedes has strong cointerest with the luxury, technology, retail, and sports categories, suggesting that extensions may be possible in these categories.
Graph: Figure 5. Net transcendence matrix, t_netbG, reflecting brand–category connections of the automotive brands (2020).
Similarly, there may be brands that, despite having low net transcendence into different categories, have strong connections with their group, making them central in their own category. For example, Toyota and Dodge are highly central to the automotive category, although their net transcendence across categories is low. Tesla, in contrast, has high net transcendence into the technology category, despite having low centrality in the automotive group. We also observe that some car brands with high net transcendence across multiple categories have moderately low centrality in their group (e.g., Audi, Mercedes, Tesla, and Lamborghini). However, brands such as Chevrolet and Ford share strong cointerest in the automotive category and also have moderate net transcendence into beer, dining, and sports. Thus, centrality and transcendence are not mutually exclusive, and a brand may be perceived as both central and transcendent, depending on its connections in the network.
The brand network developed in this study can also help managers obtain a more granular view of transcendence by identifying brand–brand connections across categories. The different levels of analysis (i.e., brand–category and brand–brand) offered by the network can help managers understand why certain cobranding opportunities are more promising than others. Further, the network's analysis of the asymmetry between brand pairs can help identify brands which may potentially benefit more from a cobranding alliance. To identify strong, relevant cobranding candidates for a focal brand, we only consider brand–brand connections in categories where the net transcendence of the brand is positive. This ensures that the identified brand connection is not generic to the category but rather is intrinsic to the brand itself. For instance, the net transcendence of Mercedes into the luxury goods and retail categories is positive, meaning that, on average, its connections with luxury and retail goods are relatively higher than those of other car brands. Thus, for Mercedes, brands in the luxury and retail categories are considered suitable candidates for cobranding.
Figure 6 shows the brand–brand connections of Mercedes with brands in the luxury and retail categories. Some of Mercedes' strongest connections are with Louis Vuitton, Nike, Tissot, and Chanel, which highlights these brands' potential for alliances with Mercedes. However, given the asymmetrical nature of these relationships, the benefits gained through such alliances may not always be equal. For instance, the asymmetrical connection between Mercedes and Tissot reflects that a greater proportion of Tissot's audience is interested in Mercedes than vice versa. This indicates that there may be a greater potential benefit for Tissot from such an alliance. These results on asymmetry can provide additional insights to brand managers of both Mercedes and Tissot when evaluating potential cobranding candidates.
Graph: Figure 6. Top 20 brand–brand connections of Mercedes with brands in the luxury and retail categories using the Fruchterman–Reingold (1991) layout.
Indeed, prior strategic alliance literature suggests that unequal spillover benefits can be expected from asymmetrical brand alliances ([21]). However, the main findings of [21] suggest that even though the magnitude of financial gains in asymmetrical alliances is not equal, it is not a win–lose partnership but rather a win–win or a shareholder value-adding alliance for both the larger and smaller partner firms.[10] Similarly, in the case of Mercedes, even though the expected benefits may not be equal for the asymmetrical relationships (e.g., Mercedes–Chanel, Mercedes–Tissot, Mercedes–Rolex), future deals can be still beneficial to both brands. Although the smaller brand, Tissot, may achieve greater gains from this asymmetrical alliance (e.g., by having a greater proportion of its audience interested in Mercedes), the larger brand, Mercedes, may still gain access to a niche audience that may not be a part of its current demographic.
A brand's transcendence in the network, which arises from common consumer interest, may not be static. Brands may, for various reasons, wish to shift their transcendence to new categories in search of new alliances or cobranding opportunities. Our network can track such shifts in transcendence, allowing brand managers to better assess the effectiveness of their marketing actions and identify emerging or waning categories for future brand alliances.
Figure 7 shows the change in net transcendence of car brands into the technology category. The dynamic plots for other categories can be analyzed similarly. Interestingly, between 2017 and 2020, the technology connections of many car brands, including Honda, Jeep, Chrysler, Acura, and Chevrolet, decrease. However, some brands, including Tesla, Lamborghini, and Infiniti, show a steep rise in their connections with technology. This may be related to, for example, Infiniti's plans to go all-electric in 2021 and use intelligent technologies to reflect its new ethos ([30]). Thus, our brand network-based methodology can be used to assess the effectiveness of a brand's marketing campaign and showcase how marketing actions can impact the brand's transcendence in users' minds.
Graph: Figure 7. Change in net transcendence over time.
The network's ability to highlight shifts in brand transcendence over time can be of vital use to managers. The emergence of new connections with specific categories over time (e.g., Infiniti's increasing transcendence into the technology category) provides insight into the effectiveness of brands' marketing campaigns and affirms the possibility of future extensions in those categories. Similarly, the waning of a brand's connections with specific categories (e.g., Jeep's decreasing transcendence into the technology category) allows that company to identify potential problems and take appropriate action to address them. The brand network can also help managers investigate issues in more detail by uncovering the specific cross-category brand–brand connections that have diminished over time. Change in a brand's net transcendence into a category can be caused by several factors, including joint ads, new alliances, embedded promotions, or other external events. Although the method in this study does not examine the causes for such shifts, it provides managers with timely intelligence on the subject. Future marketing studies could build on this work to further investigate the causes for changes in brand transcendence over time.
This subsection discusses how the category-specific brand connections revealed through the transcendence matrix may not only allow managers to understand the position of their brands in consumers' minds but also help distinguish them from their competitors. Figure 8 shows the net transcendence, , of two beer brands, Bud Light and Sierra Nevada, into different categories. Whereas Bud Light has high transcendence into food and dining, Sierra Nevada has high transcendence into travel, airlines, and technology. Regarding centrality, Bud Light outperforms Sierra Nevada, with stronger connections within the beer category. Thus, whereas the former brand is positioned strongly among beer and food enthusiasts, the latter brand resonates more with technology and travel enthusiasts. This type of analysis can help brand managers identify the differentiating connections of their brands with respect to their competitors and also gauge the type of users their competitors attract.
Graph: Figure 8. Net transcendence matrix, t_netbG, of Bud Light and Sierra Nevada.
Brand managers can obtain richer insights into the cointerest patterns of their competing brands by examining the individual brand–brand connections in the different categories. For example, Bud Light is connected to more food, beverage, and dining brands (e.g., Pepsi, Coca-Cola, McDonald's, Subway, Taco Bell), while Sierra Nevada is mostly connected to airline, travel, and technology brands (e.g., Southwest Airlines, Discovery, SpaceX, Amazon, Netflix). As one might expect, some brand–brand connections can reflect previous marketing activities (e.g., joint advertisements or promotions, collaborations, licensing deals). In such cases, our brand network–based methodology enables managers to measure the effectiveness of a marketing campaign and showcases how marketing actions can impact a brand's transcendence. For example, Bud Light's connection with Pepsi reflects strong cofollowership patterns between the two brands, affirming the effectiveness of their earlier joint marketing campaign.[11] Alternatively, brand–brand connections can highlight potential new cobranding or alliance opportunities that were previously unknown to brand managers. For instance, Bud Light's strong connections with McDonald's and Taco Bell highlight strong cointerest between these brands, suggesting untapped cobranding opportunities. Similarly, Sierra Nevada could leverage the technology and travel interests of its fans, as revealed by the network, to partner with relevant travel brands such as Southwest Airlines, SpaceX, Discovery, Amazon, and Netflix. In the next section, we validate our results against external survey ratings and test the reliability of our findings.
To validate the effectiveness of our methodology, we compare the network ratings from our automated approach with directly elicited survey ratings. The survey was conducted through Amazon Mechanical Turk, which is a reliable source for conducting social sciences research ([11]). The survey respondents were asked to report their income, age, and gender to account for any demographic influence in the sample. The participants were required to be located in United States and be over 18 years old. To ensure high-quality responses, a prior task approval rate of 95% was required for all survey respondents. The brands were grouped by sector, and four separate surveys, consisting of 250 participants each, were conducted to validate the brand–category and brand–brand connections of beer and automotive brands. Next, we discuss our survey findings along with several robustness checks.
In this subsection, we examine whether the cobranding candidates identified by the network are also perceived by consumers to be such candidates. The network edge weights between brands are intended to reflect consumers' perceptions of the brands that could be paired for cobranding; thus, such a relationship should be reflected in the survey responses. For this validation, we select the five most-followed brands in the beer and automotive categories. Then, for each brand, we select nine cobranding candidates: ( 1) the top three cross-category cobranding opportunities (i.e., brands), as identified by the network, ( 2) the top three most-followed brands in the sample that are not included in part 1, and ( 3) three randomly drawn brands that are not included in parts 1 and 2.
For each focal brand, we ask the respondents to rate its cobranding candidates on a scale of 1 ("less likely to go together") to 10 ("highly likely to go together") according to how strongly they can be paired with the focal brand. The survey scores were then correlated with the outgoing edge weights from the focal brands to their cobranding candidates in the brand network. For every survey question, the brand order was randomized, and attention filters were included to identify invalid responses. To identify loyal fans, participants were separately requested to select their favorite auto and beer brands from the list. Details of the survey and the corresponding descriptive statistics are included in Web Appendix B.
Table 2 shows the Pearson correlation coefficients between the survey and network scores. Overall, the survey measures correlate well with the network estimates, with the survey's top-three-box score[12] achieving an average correlation of.67 with the network estimates. The overall correlation between average survey ratings and network constructs is.65.
Graph
Table 2. Pearson Correlation Coefficients of the Survey Estimates with the Network Constructs.
| Category | Brands | r (Mean) | r (Top Three Box) |
|---|
| Automotive | Tesla | .89 | .88 |
| Mercedes-Benz | .63 | .54 |
| BMW | .76 | .76 |
| Audi | .93 | .93 |
| Ford | .62 | .52 |
| Beer | Miller Lite | .64 | .66 |
| Sierra Nevada | .53 | .55 |
| Bud Light | .56 | .51 |
| Budweiser | .53 | .55 |
| Coors Light | .57 | .57 |
| Average correlation coefficient (r) | .67 | .65 |
In addition, we compute the correlations between network scores and the survey ratings of users who rate a specific brand as their favorite. When using only data from fans, the overall correlation coefficient increases to.70 for average survey ratings and.71 for the top-three-box survey ratings. Finally, we examine the scatter plots more closely to better understand the circumstances in which the network cobranding candidates align well with the survey responses. For example, Figure 9, Panel A, shows the cobranding candidates for Audi as suggested by the network, together with the corresponding survey ratings. The top three cobranding candidates suggested by the network (i.e., Microsoft, Nike, and Intel) also receive very high ratings from the survey respondents. Brands with low network connectivity with Audi (i.e., Lays, Forever21, and ABC) also receive lower ratings from the survey respondents. These results reaffirm the previous findings that network connectivity patterns between brands are useful marketing metrics for consumers' perceptions of which brands are likely to pair well together.
Graph: Figure 9. Network versus survey estimates for brand–brand connections of Audi and Budweiser.
Figure 9, Panel B, shows the cobranding candidates for Budweiser, as suggested by the network and survey responses. Some top cobranding candidates suggested by the network, including the NFL and Pepsi, also receive high ratings from the survey respondents. However, despite its strong connection to Budweiser in the network, Starbucks receives low ratings from the survey respondents. Although the strong network connectivity between Starbucks and Budweiser is not directly perceived by survey responders, it may indirectly reflect the complementary taste interests of coffee and beer drinkers, which Starbucks previously leveraged to launch a line of beer-like coffee drinks ([26]; [38]). Similarly, Budweiser has a lower network connectivity score with the soccer brand FIFA than the corresponding survey rating. On further investigation, whereas Budweiser's U.S. Twitter account, which was used in our survey validation, has low network connectivity with FIFA, the brand's global Twitter account is very strongly connected to FIFA. This suggests that brand managers should apply domain knowledge and managerial judgment to explore alternative cobranding candidates based on their market of interest. Using domain customization, brand managers can query the brand network to include brand accounts that best suit their target market and conduct a more tailored network analysis to identify where the proposed cobranding opportunity may work well.
The preceding analysis also highlights some of the limitations of survey-based validation. First, consumer responses to direct questions on brand perceptions are based on the respondents' existing notions of brand extendibility and may be confounded by prior user experiences ([25]). Second, asking consumers about their overall perceptions using direct rating scales may not reveal novel or unique brand extensions or cobranding ideas; rather, it may simply facilitate the testing of known concepts ([ 5]). The brand network, in contrast, leverages the cofollowership patterns of millions of Twitter users across a broad brand ecosystem to reveal cross-category cobranding ideas that may not be intuitive to consumers but, in hindsight, are effective. Overall, with an average correlation of.71 between network scores and fans' survey ratings, the validation results suggest that the automated network-based approach enables managers to quickly and inexpensively identify cobranding and brand extension ideas that would otherwise be difficult to anticipate.
Next, we validate whether the transcendence measures derived from the brand network align with consumers' perceptions of the brands. For both beer and automotive brands, consumer ratings along the luxury and technology categories were elicited. On a scale of 1 ("least likely") to 5 ("most likely"), participants were asked to rate the focal brands (e.g., Heineken) according to how strongly they associated them with a new category (i.e., luxury goods and technology). Further, to identify brands with strong centrality within their own group, participants were asked to rate the focal brands, on a scale of 1 ("least likely") to 5 ("most likely"), according to how strongly they believed them to be central in their own category. Finally, the average survey ratings for each brand across the luxury and technology categories are compared with the brand transcendence constructs obtained using the network measures. Similarly, the average survey rating for centrality is compared with the centrality construct obtained using the network measures. We also calculate the top-two-box score for each brand to assess the proportion of people who rate a brand very highly (i.e., a score of 4 or 5). Details of the survey and the corresponding descriptive statistics are included in Web Appendix B.
The Pearson correlation coefficients between the survey results and network constructs are listed in Table 3. Overall, the survey measures correlate well with the network estimates, with the top-two-box survey scores and mean survey scores achieving an average correlation of.63 with the network estimates.
Graph
Table 3. Pearson Correlation Coefficients of the Survey Estimates with the Network Constructs.
| Category | Constructs | r (Mean) | r (Top Two Box) |
|---|
| Automotive | Transcendence (luxury) | .59 | .61 |
| Transcendence (technology) | .66 | .67 |
| Centrality | .58 | .61 |
| Beer | Transcendence (luxury) | .52 | .53 |
| Transcendence (technology) | .72 | .69 |
| Centrality | .71 | .70 |
| Average correlation coefficient (r) | .63 | .63 |
Figure 10 includes the scatter plots for survey versus network measures. Given that the network and survey estimates are measured in different units, the plots have been scaled to 0–1 for easier interpretation. Overall, the scatter points are well distributed along the best fit line, with few outliers. As we discuss next, we then conduct a series of tests to ensure that the network accurately captures the shifts in connections over time. Overall, our general findings pass these tests, supporting the future use of implicit brand networks in marketing research.
Graph: Figure 10. Scatter plots of network versus survey estimates for automotive and beer brands.
The "Results" subsection discusses the brand network's ability to capture shifts in brand transcendence over time. We now test whether the waning of certain connections between 2017 and 2020 in the network is supported by the survey responses. To do this, we first identify the connections between brands and categories that exist in the 2017 network but decline in the 2020 network. Figure 11 illustrates the filtered cases for automotive brands and the corresponding survey results. Panel A shows that brands such as Mazda, Mini, Buick, Chrysler, and GMC all have connections with the luxury category in 2017. According to [17] in Forbes, at the time, the Mazda 2017 CX-9 Signature model was considered the most luxurious vehicle produced by Mazda to date. The author mentions that "Mazda has never been considered a luxury brand, but maybe it's time to reconsider that classification" ([17]). However, the results from the brand network in 2020 show that Mazda does not retain its connection with luxury. This is further validated by the survey participants, who also rate Mazda as very weakly associated with luxury. There is a similar pattern in Panel B, in which brands such as Dodge, Chevrolet, Jeep, Honda, and Chrysler show a significant drop in their connections with technology category between 2017 and 2020. This change is also reflected in the survey responses.
Graph: Figure 11. Shift in transcendence of car brands from 2017 to 2020.
Next, we test whether survey respondents also reject random connections that do not exist in either 2017 or 2020. To do so, we filter the cases where connections between brands and categories are absent in both the 2017 and 2020 networks and compare them with the survey responses.[13] We find that the average survey ratings are below 2 (out of 5) for most brand–category connections not existing in either network (i.e., 2017 and 2020).
Studies that mine brand perceptions from social media sources must consider the extent to which brand followers on social media represent the general population. It is also important to consider whether certain Twitter brands accounts are more appealing to a specific audience (e.g., young people, men). Recent studies have reported that Twitter followership data successfully captures attribute-specific consumer perceptions beyond demographic similarities ([11]). We investigate this issue further by comparing the survey ratings, which were provided by users of different demographics, with the transcendence values obtained from the brand network. Table 4 lists the correlation values between the network estimates and income-specific survey ratings for the transcendence of automotive brands into the technology category. For most income groups in the survey, there are adequately high correlations with the network transcendence constructs. Results for all the remaining survey demographic groups (i.e., age and gender) are included in Web Appendix D. We observe adequately high correlations between the demographic-specific survey ratings and the brand network constructs. This affirms that the overall brand network estimates are not heavily influenced by the demographics of Twitter users.
Graph
Table 4. Income-Specific Survey Correlations with the Network Constructs.
| Automotive Brands' Transcendence to Technology |
|---|
| Income | Survey-Based Measure (≤$29,999) | Survey-Based Measure ($30,000–$59,999) | Survey-Based Measure ($60,000–$99,999) | Survey-Based Measure ($100,000–$149,999) | Survey-Based Measure (≥$150,000) |
|---|
| Survey-based measure (≤$29,999) | 1.00 | | | | |
| Survey-based measure ($30,000–$59,999) | .96 | 1.00 | | | |
| Survey-based measure ($60,000–$99,999) | .97 | .98 | 1.00 | | |
| Survey-based measure ($100,000–$149,999) | .93 | .96 | .94 | 1.00 | |
| Survey-based measure (≥$150,000) | .93 | .91 | .94 | .88 | 1.00 |
| Network-based measure | .67 | .62 | .72 | .59 | .74 |
The presence of Twitter bots may inflate the number of common followers between brands, which can, in turn, lead to inaccurate network estimates of transcendence and centrality. In this section, we conduct multiple network simulations by repeatedly rewiring the edges to test whether the original brand network structure remains reasonably stable. We incrementally rewire the cofollowers from any random pair of edges and reperform the entire analysis. As Figure 12 shows, the rewiring stage involves the addition and removal of 5% of the cofollowers of any random pair of edges in the network, continuing until 50% of the cofollowership patterns have been altered. In each iteration, once the network rewiring is complete, the algorithm reruns the entire analysis (i.e., it applies the disparity filter to identify statistically significant edges, normalizes the edge weights, and calculates the transcendence across categories).
Graph: Figure 12. Process flow for each iteration.
Figure 13 illustrates the results of the simulations for the automotive brands and compares the transcendence values obtained after rewiring the network with the original values. The purpose of the test is to ensure that the original network estimates hold for small rewiring changes (i.e., that significant rewiring is needed to yield completely different network estimates). For all plots, the rewired network estimates correlate highly with the original estimates until a large percentage of the network (>30%) is rewired. This demonstrates that the brand network structure is not sensitive to small underlying changes that may occur due to bots.
Graph: Figure 13. Correlation of post rewiring transcendence values with original transcendence values for automotive brands.
Despite its relevance to various marketing decisions (such as cobranding and brand extensions), the identification of cross-category insights across a broad brand ecosystem is currently understudied in the marketing literature. This article uses implicit brand networks to identify the category-specific connections of brands and their competitors by exploiting the overlap in brand followers on Twitter. We introduce a new construct, transcendence, that measures the extent to which a brand shares cointerest with other brands in different categories. Depending on a firm's marketing objectives (i.e., their focus on extensions vs. cobranding), the transcendence of a brand can be studied at different levels: brand–category or brand–brand. These different levels of analysis can help managers identify viable cobranding opportunities.
Furthermore, we leverage the concept of asymmetry between brand pairs to provide more nuanced insights into possible cobranding opportunities and determine which brand can potentially benefit more from a cobranding alliance. We conducted the analysis over time to track shifts in brand transcendence, allowing brand managers to both assess the effectiveness of existing marketing strategies and identify new alliance opportunities. To ensure the reliability of our proposed methodology, we validate our findings against external survey ratings and conduct extensive robustness checks, including network simulations, to ensure that our final network estimates are not biased by Twitter bots.
From a methodological standpoint, the implicit brand networks utilized in this article condense the high-dimensional interest space of millions of brand followers into a parsimonious form that is more amenable to research and business applications. The readily accessible artifact, which is obtained with little human intervention in the processing of the underlying data, allows managers to efficiently infer cross-category branding insights in a scalable way. Compared with extant digital approaches that rely on extensive preprocessing, this straightforward automated approach enables practitioners to readily obtain the cointerest patterns of brands with respect to their competitors and gauge the types of users that their competitors attract. More specifically, given its automated data collection and network analyses, the brand network can act as an effective business intelligence tool for the identification of cobranding and extension opportunities across a broad ecosystem of brands.
Overall, our approach offers several benefits to marketers. It also highlights avenues for future research. First, although our analyses use Twitter brand communities, it would be interesting to compare similar communities on Facebook and Instagram. Brand networks on different social media platforms may vary based on factors such as user demographics, category, platform characteristics, or a brand's marketing strategy. Although consistent brand connections across different platforms can provide additional validity to findings of this study, meaningful insights may also be gleaned if substantial differences are observed. Such differences may, for example, stem from a brand's tailored marketing efforts on a specific platform. Using brand networks to track the effectiveness of such efforts can be beneficial to brand owners. Differing user demographics across platforms may also have an impact on brand network structures. Though this study did not identify substantial differences between the demographic-specific survey ratings and the transcendence values obtained from the brand network, future research could examine platform-specific brand networks to obtain richer insights. Second, future research could consider how to distinguish the content on brand pages that may affect consumers' decisions to follow brands, including promoted content on a brand's page, multichannel advertising across platforms (e.g., email, Facebook), and the use of trending topics or sponsored tweets.
Third, the analysis in this article relies on a brand's followers at a given point in time. Twitter does not provide data on when a user starts or stops following an account. The article's analysis of two different periods highlights the potential for our method to examine how transcendence changes over time. Because most aspects of the data collection and network analyses in this approach can be automated, brand managers could collect followership information at more regular intervals to examine changes in transcendence more frequently. Fourth, while our study relies on validation from two categories, future studies can consider expanding the survey-based validation for broader set of brands across multiple categories. However, for such validation, it is important to consider that certain cross-category cobranding candidates, revealed through the cofollowership patterns on social media, may not always be intuitive to survey respondents, though in hindsight they make sense and work. In conjunction with the brand network results, marketers should apply domain knowledge and managerial judgment to explore different extension and cobranding opportunities that may work best for the brand.
Fifth, it is also important for brand managers to consider that Twitter users from around the globe are free to follow any account(s) of a brand (which can include global or country-specific accounts). As the data collection and network analyses can be largely automated, marketers can create custom brand networks to include Twitter brand accounts (country-specific and/or global) that best suit their market of interest. Domain customization can help managers conduct a more targeted network analysis of where the proposed cobranding opportunity may work best. Lastly, although our approach relies on cofollowership patterns to identify cobranding opportunities, we do not investigate the drivers of common followership on Twitter and the extent to which these drivers lead to network overlap between brands. The reasons that users cofollow brands on Twitter are varied and complex, with many unobservable factors possibly at play. Industry research by Nielsen ([29]) indicates that 55% of Twitter users say they follow a brand because they like it, followed by 52% of users who want to keep up-to-date on the latest promotions and offers posted by the brand. There are various other reasons that users cofollow multiple brands on social media; thus, future research could use the brand network described in this article to investigate and better understand the drivers of cofollowership between brands on social media.
Overall, this work offers a new approach for researchers and practitioners interested in automatically monitoring cross-category brand connections over time. Network-based methods for brand management are relatively new and present many opportunities for future research. The methods introduced in this article provide a foundation for marketing researchers interested in leveraging implicit brand networks to gain richer insights into consumers and brands.
sj-pdf-1-jmx-10.1177_00222429221083668 - Supplemental material for Leveraging Cofollowership Patterns on Social Media to Identify Brand Alliance Opportunities
Supplemental material, sj-pdf-1-jmx-10.1177_00222429221083668 for Leveraging Cofollowership Patterns on Social Media to Identify Brand Alliance Opportunities by Pankhuri Malhotra and Siddhartha Bhattacharyya in Journal of Marketing
Footnotes 1 Jacob Goldenberg
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Pankhuri Malhotra https://orcid.org/0000-0003-4463-0656
5 There are two kinds of interaction networks: explicit and implicit. Explicit networks involve direct voluntary interaction between entities (e.g., a friendship network on Facebook). Implicit networks, in contrast, arise indirectly due to shared user preferences. For example, an implicit network may be established between products when people tend to buy them together ([49]).
6 The superordinate category is the highest umbrella category containing diverse exemplars with low degrees of class inclusion (e.g., beer, dining). The subordinate level is one level deeper and contains exemplars that are comparable across specific attributes (e.g., fruity beers, American dining). In this study, we use the basic or superordinate category levels, rather than specific product types, for comparisons between brands.
7 SparkToro is a software company that provides intelligence reports on Twitter brand accounts. Its algorithm is designed to identify spam accounts, bot accounts, propaganda accounts, and inactive accounts.
8 For color versions of the figures, see the online article.
9 Although this analysis relies on superordinate categories, depending on the marketing objectives, intercategory transcendence constructs can be readily obtained for subordinate categories. The assignment of brands to categories is flexible and can be changed simply by relabeling the attributes in the network.
While firm size generally relates to financial metrics ([21]), brand size on social media may not correspond to financial metrics, especially for brands that belong to different categories.
In this study, we manually examine external industry sources such as Beverage Industry (https://www.bevindustry.com) to discover joint marketing campaigns (e.g., advertising, licensing, cobranding deals) previously conducted by Bud Light and Sierra Nevada. The online portal publishes trends, innovations, product launches, marketing campaigns, and news regarding beer brands.
The survey top-three-box score is used to capture the survey responses of individuals who rate a brand very highly (i.e., a score of 8, 9, or 10).
Web Appendix C lists the filtered cases in which there are no (or close to zero) connections between brands and categories in either the 2017 or the 2020 network. The survey ratings are listed next to the network ratings.
References Aaker David A. , Keller Kevin Lane. (1990), " Consumer Evaluations of Brand Extensions ," Journal of Marketing , 54 (1), 27 – 41.
Aral Sinan , Walker Dylan. (2014), " Tie Strength, Embeddedness, and Social Influence: A Large-Scale Networked Experiment ," Management Science , 60 (6), 1352 – 70.
Ashley Christy , Tuten Tracy. (2015), " Creative Strategies in Social Media Marketing: An Exploratory Study of Branded Social Content and Consumer Engagement ," Psychology & Marketing , 32 (1), 15 – 27.
Barrat A. , Barthelemy M. , Pastor-Satorras R. , Vespignani A.. (2004), " The Architecture of Complex Weighted Networks ," Proceedings of the National Academy of Sciences , 101 (11), https://doi.org/10.1073/pnas.0400087101.
Batra Rajeev , Lenk Peter , Wedel Michel. (2010), " Brand Extension Strategy Planning: Empirical Estimation of Brand–Category Personality fit and Atypicality ," Journal of Marketing Research , 47 (2), 335 – 47.
Berger Jonah , Heath Chip. (2007), " Where Consumers Diverge from Others: Identity Signaling and Product Domains ," Journal of Consumer Research , 34 (2), 121 – 34.
Brexendorf Tim Oliver , Keller Kevin Lane. (2017), " Leveraging the Corporate Brand: The Importance of Corporate Brand Innovativeness and Brand Architecture ," European Journal of Marketing , 51 (9/10), 1530 – 51.
Cao Zixia , Sorescu Alina. (2013), " Wedded Bliss or Tainted Love? Stock Market Reactions to the Introduction of Cobranded Products ," Marketing Science , 32 (6), 939 – 59.
Carpenter Gregory S. , Nakamoto Kent. (1989), " Consumer Preference Formation and Pioneering Advantage ," Journal of Marketing Research , 26 (3), 285 – 98.
Childers Terry L. , Rao Akshay R.. (1992), " The Influence of Familial and Peer-Based Reference Groups on Consumer Decisions ," Journal of Consumer Research , 19 (2), 198 – 211.
Culotta Aron , Cutler Jennifer. (2016), " Mining Brand Perceptions from Twitter Social Networks ," Marketing Science , 35 (3).
Cutright Keisha M. , Bettman James R. , Fitzsimons Gavan J.. (2013), " Putting Brands in Their Place: How a Lack of Control Keeps Brands Contained ," Journal of Marketing Research , 50 (3), 365 – 77.
Dawar Niraj , Bagga Charan K.. (2015), " A Better Way to Map Brand Strategy ," Harvard Business Review , 93 (6), 90 –9 7.
Desai Kalpesh Kaushik , Keller Kevin Lane. (2002), " The Effects of Ingredient Branding Strategies on Host Brand Extendibility ," Journal of Marketing , 66 (1), 73 – 93.
DeSarbo Wayne S. , Grewal Rajdeep. (2007), " An Alternative Efficient Representation of Demand–Based Competitive Asymmetry ," Strategic Management Journal , 28 (7), 755 – 66.
Farquhar Peter H. , Herr Paul M.. (1993), The Dual Structure of Brand Associations. Hillsdale , NJ : Lawrence Erlbaum Associates.
Fogelson Jason. (2017), " 2017 Mazda CX-9 Signature Test Drive and Review: Luxury Without the Brand ," Forbes (January 23), https://www.forbes.com/sites/jasonfogelson/2017/01/23/2017-mazda-cx-9-signature-test-drive-and-review-luxury-without-the-brand/#768ae1927d6c.
Fruchterman Thomas M.J. , Reingold Edward M.. (1991), " Graph Drawing by Force-Directed Placement ," Software: Practice and Experience , 21 (11), 1129 – 64.
Hansen Karsten , Singh Vishal , Chintagunta Pradeep. (2006), " Understanding Store-Brand Purchase Behavior Across Categories ," Marketing Science , 25 (1), 75 – 90.
Helmig Bernd , Huber Jan-Alexander , Leeflang Peter S.H.. (2008), " Co-Branding: The State of the Art ," Schmalenbach Business Review , 60 (4), 359 – 77.
Kalaignanam Kartik , Shankar Venkatesh , Varadarajan Rajan. (2007), " Asymmetric New Product Development Alliances: Win-Win or Win-Lose Partnerships? " Management Science , 53 (3), 357 – 74.
Keller Kevin Lane. (2003), " Brand Synthesis: The Multidimensionality of Brand Knowledge ," Journal of Consumer Research , 29 (4), 595 – 600.
Keller Kevin Lane. (2014), " Designing and Implementing Brand Architecture Strategies ," Journal of Brand Management , 21 (9), 702 – 15.
Kim Jun B. , Albuquerque Paulo , Bronnenberg Bart J.. (2011), " Mapping Online Consumer Search ," Journal of Marketing Research , 48 (1), 13 – 27.
Klink Richard R. , Smith Daniel C.. (2001), " Threats to the External Validity of Brand Extension Research ," Journal of Marketing Research , 38 (3), 326 – 35.
Knowles Kitty. (2017), " What the Heck Is... Nitro Coffee? " Forbes (May 9), https://www.forbes.com/sites/kittyknowles/2017/05/09/nitro-coffee-what-is-nitro-cold-brew-coffee-starbucks-uk-costa-coffee/?sh=54cc748027ff.
Kuksov Dmitri , Shachar Ron , Wang Kangkang. (2013), " Advertising and Consumers' Communications ," Marketing Science , 32 (2), 294 – 309.
Loken Barbara , Ward James. (1990), " Alternative Approaches to Understanding the Determinants of Typicality ," Journal of Consumer Research , 17 (2), 111 – 26.
Macmillan Gordon. (2014), " 10 Reasons Why People Follow Brands on Twitter ," blog entry, Twitter (March 27), https://blog.twitter.com/en_gb/a/en-gb/2014/10-reasons-why-people-follow-brands-on-twitter.
Marketwire. (2018), " INFINITI to Go Electric from 2021 ," press release, Business Insider (January 16), https://markets.businessinsider.com/news/stocks/infiniti-to-go-electric-from-2021-1013133127.
McCue T.J.. (2018), " Social Media Is Increasing Brand Engagement and Sales ," Forbes (June 26), https://www.forbes.com/sites/tjmccue/2018/06/26/social-media-is-increasing-brand-engagement-and-sales.
McPherson Miller , Smith-Lovin Lynn , Cook James M.. (2001), " Birds of a Feather: Homophily in Social Networks ," Annual Review of Sociology , 27 (1), 415 – 44.
Naylor Rebecca Walker , Lamberton Cait Poynor , West Patricia M.. (2012), " Beyond the 'Like' Button: The Impact of Mere Virtual Presence on Brand Evaluations and Purchase Intentions in Social Media Settings ," Journal of Marketing , 76 (6), 105 – 20.
Netzer Oded , Feldman Ronen , Goldenberg Jacob , Fresko Moshe. (2012), " Mine Your Own Business: Market-Structure Surveillance Through Text Mining ," Marketing Science , 31 (3), 521 – 43.
Oestreicher-Singer Gal , Libai Barak , Sivan Liron , Carmi Eyal , Yassin Ohad. (2013), " The Network Value of Products ," Journal of Marketing , 77 (3), 1 – 14.
Oestreicher-Singer Gal , Sundararajan Arun. (2012), " The Visible Hand? Demand Effects of Recommendation Networks in Electronic Markets ," Management Science , 58 (11), 1963 – 81.
Opsahl Tore , Agneessens Filip , Skvoretz John. (2010), " Node Centrality in Weighted Networks: Generalizing Degree and Shortest Paths ," Social Networks , 32 (3), 245 – 51.
Pearson Bryan. (2016), " Coffee on the Rocks? Starbucks Reserve Might Shake Up Industry ," Forbes (December 16), https://www.forbes.com/sites/bryanpearson/2016/12/16/coffee-on-the-rocks-starbucks-reserve-might-shake-up-industry/?sh=13108b212297.
Peng Jing , Agarwal Ashish , Hosanagar Kartik , Iyengar Raghuram. (2018), " Network Overlap and Content Sharing on Social Media Platforms ," Journal of Marketing Research , 55 (4), 571 – 85.
Pereira Hélia Gonçalves , de Fátima Salgueiro Maria , Mateus Inês. (2014), " Say Yes to Facebook and Get Your Customers Involved! Relationships in a World of Social Networks ," Business Horizons , 57 (6), 695 – 702.
Ratneshwar Srinivasan , Pechmann Cornelia , Shocker Allan D.. (1996), " Goal-Derived Categories and the Antecedents of Across-Category Consideration ," Journal of Consumer Research , 23 (3), 240 – 50.
Ringel Daniel M. , Skiera Bernd. (2016), " Visualizing Asymmetric Competition Among More Than 1,000 Products Using Big Search Data ," Marketing Science , 35 (3), 511 – 34.
Rosnowski Suzanne. (2020), " 2020: The Year of the Brand Mashup ," Forbes (January 15), https://www.forbes.com/sites/forbesagencycouncil/2020/01/15/2020-the-year-of-the-brand-mashup.
Serrano M. Ángeles , Boguná Marián , Vespignani Alessandro. (2009), " Extracting the Multiscale Backbone of complex Weighted Networks ," Proceedings of the National Academy of Sciences , 106 (16), 6483 – 88.
Sundararajan Arun , Provost Foster , Oestreicher-Singer Gal , Aral Sinan. (2013), " Research Commentary—Information in Digital, Economic, and Social Networks ," Information Systems Research , 24 (4), 883 – 905.
Swaminathan Vanitha , Moorman Christine. (2009), " Marketing Alliances, Firm Networks, and Firm Value Creation ," Journal of Marketing , 73 (5), 52 – 69.
Thompson Stephanie. (1998), " Brand Buddies—Co-Branding Meal Solutions ," Brandweek , 39 (8), 22 – 30.
Van der Lans Ralf , van den Bergh Bram , Dieleman Evelien. (2014), " Partner Selection in Brand Alliances: An Empirical Investigation of the Drivers of Brand Fit ," Marketing Science , 33 (4), 551 – 66.
Zhang Kunpeng , Bhattacharyya Siddhartha , Ram Sudha. (2016), " Large-Scale Network Analysis for Online Social Brand Advertising ," MIS Quarterly , 40 (4), 849 – 68.
~~~~~~~~
By Pankhuri Malhotra and Siddhartha Bhattacharyya
Reported by Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 82- Leveraging Creativity in Charity Marketing: The Impact of Engaging in Creative Activities on Subsequent Donation Behavior. By: Xu, Lidan; Mehta, Ravi; Dahl, Darren W. Journal of Marketing. Sep2022, Vol. 86 Issue 5, p79-94. 16p. 1 Diagram, 2 Charts. DOI: 10.1177/00222429211037587.
- Database:
- Business Source Complete
Leveraging Creativity in Charity Marketing: The Impact of Engaging in Creative Activities on Subsequent Donation Behavior
Charities are constantly looking for new and more effective ways to engage potential donors in order to secure the resources needed to deliver services. The current work demonstrates that creative activities are one way for marketers to meet this challenge. Field and lab studies find that engaging potential donors in creative activities positively influences their donation behaviors (i.e., the likelihood of donation and the monetary amount donated). Importantly, the observed effects are shown to be context independent: they hold even when potential donors engage in creative activities unrelated to the focal cause of the charity (or the charitable organization itself). The findings suggest that engaging in a creative activity enhances the felt autonomy of the participant, thus inducing a positive affective state, which in turn leads to higher donation behaviors. Positive affect is demonstrated to enhance donation behaviors due to perceptions of donation impact and a desire for mood maintenance. However, the identified effects emerge only when one engages in a creative activity—not when the activity is noncreative, or when only the concept of creativity itself is made salient.
Keywords: autonomy; creativity; donation; positive affect
Charitable organizations exist to support a wide variety of causes, such as helping malnourished children, caring for the homeless, supporting animal welfare, and meeting environmental concerns, to name a few. The success of these organizations in supporting their causes largely depends on the donations they secure. According to the [58], approximately 1.56 million registered nonprofit organizations exist in the United States, together raising an estimated $390 billion in donations annually. Despite these large numbers, fundraising remains a major challenge for such organizations, with approximately 45% of charities unable to secure the required level of resources needed to deliver their services ([59]).
In light of this, it is not surprising that marketers at these organizations seek more effective ways to solicit donations, often utilizing nontraditional approaches and fundraising events (e.g., ice-cream socials, silent auctions, trivia nights) to engage potential donors ([11]). For example, in 2014, the ALS Association invited people around the globe to participate in its "Ice Bucket Challenge" to increase awareness of ALS, raising approximately $220 million from individual donors in the process (Holan 2014). Bloodwater.org devised the "Real Game of Thrones" campaign, which called on people to participate through Twitter and used a combination of pop culture, humor, and bathroom puns to raise money to build latrines throughout Africa. This creative campaign successfully raised enough money in 24 hours to build 21 latrines in Rwanda. Cookies for Kids, another charitable organization, sponsors creative charity events each year such as cookie swap parties, where participants decorate cookies and swap recipes to raise donations. These fundraising anecdotes suggest that charities are defining new ways of engaging potential donors, while raising questions about which types of activities most effectively enhance donation behaviors.
The current work meets this challenge by examining how engaging potential donors in creative activities can positively influence their propensity to donate money to a charitable cause. We argue that engaging in a creative activity induces a positive affective state, which in turn increases both the likelihood and amount of monetary donation made to the charitable organization. While prior work has independently examined links between creativity and positive affect (e.g., [ 9]; [47]), as well as between positive affect and helpfulness (e.g., [ 2]; [ 3]), we provide a deeper understanding of why and how creativity leads to enhanced donation behaviors. Specifically, we show that the link between creativity and positive affect is driven by the sense of autonomy that is induced by engaging in a creative activity (i.e., an attempt to create something novel). Further, by identifying the roles that desire for mood maintenance and perceived donation impact play, we provide insight into why the positive affect resulting from a creative activity leads to enhanced donation behavior.
The current research makes several important contributions. From a practical perspective, this research offers a simple and effective way for marketers to improve their donation appeals; it suggests that engaging potential donors in a creative activity enhances subsequent donation behavior. This recommended approach provides a real opportunity for charity marketers to increase the efficacy of their fundraising campaigns. At the theoretical level, the present work advances the marketing and charity literature streams in several ways. First, we demonstrate the positive effect of creative engagement on donation behavior. To our knowledge, no research thus far has examined whether and how engaging in creative activities can impact an individual's subsequent donation behavior toward a charitable organization. Second, we explicate the reasons and conditions that drive the relationship between creativity and donation behavior. We demonstrate that it is the act of actually engaging in a creative activity—rather than simple priming or making the concept of creativity salient—that drives the effect. Further, while prior work has consistently shown that a sense of autonomy can facilitate creativity (e.g., [18]), we demonstrate that engaging in a creative activity also heightens one's sense of autonomy, which in turn induces positive affect. As noted previously, we also highlight that the positive affect experienced during a creative activity bolsters desire for mood maintenance and perceived donation impact, thereby enhancing the likelihood of donation and the monetary amount donated. Finally, we find evidence that the positive effect of engaging in a creative activity on monetary donation is context independent. That is, engaging potential donors in a creative activity not directly related to the charitable cause or organization still has a positive influence on their donation behavior. Thus, the current work not only offers marketers a way to build effective donation campaigns but also provides a deeper theoretical understanding of the relationship between creativity and donation behavior.
Researchers have explored many facets of donation behavior, from the demographic and socioeconomic determinants of donation ([ 6]; [10]; [40]) to the extent to which other factors—such as motivation, psychological characteristics, and social cognition—can affect donation ([34]). In addition, prior research has proposed and examined various marketing strategies and tactics used to increase donations. For example, using public recognition ([73]), taking advantage of price promotions ([78]), designing more attractive appeals ([51]), expressing one's identity ([61]), and using positioning to enhance the effectiveness of the charity ([74]) have all been investigated.
More relevant to the current research, recent work has also started to examine the merits of engaging potential donors in different types of activities and tasks before soliciting donations. For example, [62] examined the influence of a storytelling event in the crowdfunding context, finding that direct (vs. indirect) storytelling positively affects customer engagement and donation likelihood. In contrast with more traditional donation requests ([69]), some charities are utilizing physical activities (e.g., walks and runs [[37]], sporting events [[36]], silent auctions [[41]], ice cream socials, trivia nights [[11]]) as precursors to the donation solicitation. Despite the initial academic interest in these tactics, the effectiveness of such approaches has been understudied in the literature, and reporting has shown mixed results. For example, while [37] have argued that positive fundraising outcomes result from physical activity events (e.g., running activities, golf tournaments), [75] did not find a positive relationship between sports activities and charitable event outcomes. The current work aims to add to the literature in this regard by validating the use of activities to increase the likelihood and amount of donation contributions. Specifically, we examine the impact of engaging potential donors in creative activities.
Activities involving creation of an output span a continuum ranging from routine tasks, such as simply copying a given design, to highly creative activities, such as creating an original work of art ([18]). Within this context, we argue that the inherent characteristic of creative engagement, which is differentiated from priming or simple salience of creativity and/or creativity-related concepts, is that an individual must engage, physically or mentally, in an activity requiring the production of something novel (i.e., the activity leans to one side of the continuum referenced previously). For example, actively generating an original cookie design would lead to creative engagement, but copying a cookie design or simply being primed by the concept of creativity (e.g., through exposure to creative stimuli) would not.
Importantly, we propose that engaging in a creative activity induces positive affect for the creator. In support of this notion, liberal arts literature finds that engaging in creative activities to generate novel outputs (e.g., music composition, visual arts, creative writing) can bring about positive thoughts and feelings ([66]). Relatedly, [ 9] show that engaging in a divergent creative task induces higher levels of positive affect. Results reported in the psychology literature also support these findings. [16], while explicating the construct of flow, interviewed people who engage in creative work on a regular basis (i.e., artists and musicians) and found that these individuals often experience positive affect and happiness when creating something original. [15] confirm these findings in an experimental lab setting and show that engagement in a creative activity induces a state of flow, leading to higher positive affect.
Why does engaging in creative activities lead to positive affect? One potential driver of this positive relationship is a heightened sense of autonomy (i.e., having a sense of choice and freedom from external control; [24]; [52]; [63]), attained by engaging in a creative activity.[ 5] By definition, an attempt to generate a creative output requires one to actively recognize remote associations between broad and distant concepts and then combine these loosely connected ideas and concepts in a novel fashion ([20]; [31], [32]; [69]). Such a process requires and encourages one to think freely and make different combinations and choices without being constrained by norms and rules ([ 4]; [30]; [43]). Thus, the process associated with creative generation should manifest a sense of choice and freedom (i.e., autonomy), which we contend induces positive affect.
Prior work offers initial support for this proposition. As we have discussed, engaging in a creative activity induces a state of flow, which then leads to higher positive affect ([15]); notably, empirical work has shown the state of flow to be associated with a sense of autonomy ([49]). Similarly, [18] found that being involved in a creative activity can enhance experienced enjoyment, but only when the activity imparts a sense of autonomy. [47] conducted a daily-diary study following the routine of 1,042 hobby musicians and found that the participants reported higher positive affect on the days they engaged in music composition and performance. Importantly, the authors found that this relationship was driven by satisfaction of one's needs for autonomy. Finally, [46] found that creative generation, such as the production of visual art, significantly reduced cortisol levels (a biomarker and proxy measure of stress in humans) and increased feelings of relaxation, pleasantness, and enjoyment. Their work shows that such art making is associated with the experience of being free from constraints (i.e., the sense of autonomy).
Given this discussion, we argue that engagement in creative activities heightens one's sense of autonomy, which in turn leads to positive affect. We further propose that the positive affect induced by participation in a creatively engaging activity will lead to enhanced donation behavior. We elaborate on this prediction in the following subsection.
Findings reported in the extant literature offer compelling evidence that being in a positive affective state enhances donation behavior ([ 1]; [17]; [26]; [42]; [44]; [60]). While prior work has consistently demonstrated a positive relationship between positive affect and donation behavior, it offers disparate explanations for this relationship ([ 7]). Indeed, we recognize that the relationship between positive affect and donation behavior is likely to be multiply determined, and we therefore identify three mechanisms that are most relevant to the context of creative engagement in question.
Perhaps the most common explanation for the clear link between positive affect and helping behavior derives from the mood maintenance model. This line of reasoning proposes that people tend to maintain positive mood states ([ 7]; [33]). Thus, individuals tend to help more when in a positive affective state, because doing so enables them to prolong said state ([12]; [45]). In the context of our work, this suggests that the positive affect realized by participating in a creative activity can best be maintained when a subsequent behavior, such as helping others through enhanced donation behavior, also fosters positive feelings ([ 7]).
A second potential mechanism derives from the social aspects associated with positive affect. Indeed, it has been argued that being in a positive affective state can directly influence one's perceived social connectedness ([38]; [39]). As such, the positive affect defined by participation in a creative activity is likely to enhance the value of creating and maintaining social connections ([13]; [25]). Further, valuing social connections has been shown to enhance feelings of care and concern toward others ([ 8]), which should subsequently enhance the donation behaviors of the individual.
Finally, prior work reports that positive affect can also boost self-efficacy and/or the perceived impact of one's actions ([ 5]; [64]). That is, experiencing positive affect can lead to the belief that one's actions are more efficacious, thus creating a heightened expectancy of positive outcomes ([53]). Thus, we contend that positive affect can enhance the perceived impact of one's potential donation and, in turn, raise the likelihood and amount of one's donation behavior ([14]). In our empirical work, we explicitly test the proposed chain of effects (i.e., engagement in a creative activity → autonomy → positive affect → enhanced donation behaviors), and the three aforementioned potential mechanisms underlying the impact of positive affect on donation behavior. Summarizing our arguments, we hypothesize the following:
- H1: Engaging in a creative activity (vs. an activity that does not provide an opportunity for novel creation) leads to enhanced donation behaviors (i.e., the likelihood of donation and the monetary amount donated).
- H2a: The relationship between engaging in a creative activity and donation behavior is serially mediated by autonomy and positive affect.
- H2b: The influence of positive affect on donation behavior is driven by (a) mood maintenance, (b) social connection, and/or (c) perceived donation impact.
We utilized a combination of field studies and controlled lab experiments to test our hypotheses. First, we conducted a pilot study in collaboration with a nonprofit organization as an initial test of our focal hypothesis and found support for the prediction that engaging in a creative activity enhances donation behavior (H1). Study 1, conducted in a controlled lab setting, replicated the initial pilot study findings, thus reconfirming support for H1. Study 2, a quasi-field experiment, shows that engaging in a creative activity increases both the likelihood and amount of monetary donations, whereas simply priming the concept of creativity does not (H1). Study 3 tested the full serial mediation prediction (H2a) by demonstrating that creative engagement induces a sense of autonomy, which in turn heightens positive affect, leading to higher donation behavior. Study 4 tested H2b, showing that the path from positive affect to donation behavior is indeed multiply determined. In every study, we report all experimental conditions and measures as collected and disclose any eliminated data points when applicable. The sample size was predetermined for each study based on current experimental norms but varied within an acceptable range depending on actual participant sign-ups. Study 3, a supplementary study (follow-up to Study 3 reported in the Web Appendix), and Study 4 were preregistered on aspredicted.org (see respective studies for details).
To gain an initial understanding of whether engaging in a creative (as compared with neutral) activity enhances monetary donation, we collaborated with a registered U.S. nonprofit organization operating an animal shelter in a small city in the Southwestern United States (population 46,000 according to 2010 U.S. Census data). Every year, employees of this charity produce T-shirt designs that are printed and used as giveaways or sold in fundraising activities. To test our focal hypothesis, the charity agreed to open the T-shirt design activity to the public and use it as a fundraising event. The T-shirt design campaign was launched by the charity via its social media platform, inviting members of the public (i.e., potential donors) to create T-shirt designs as part of a donation appeal. The charity had set two overarching guidelines for this T-shirt design campaign: ( 1) the submitted designs were to follow the theme of "Rescue 2020" and ( 2) the charity's logo had to be part of the design. The charity managed the entire event.
Relevant to our prediction (H1), we manipulated the opportunity to be creative (vs. not) within the T-shirt design campaign. While participants in both conditions had to develop a T-shirt design reflecting the charity's yearly theme and including its logo, those in the creativity condition were invited to develop an innovative T-shirt design and were explicitly instructed to be creative while doing so. Those in the neutral condition were not specifically instructed to be creative. Once participants submitted their designs, the charity presented them with a donation appeal that included a link to the donation page. At the end of the campaign period, the charity forwarded us the designs along with the corresponding donation amounts, having removed donors' identifying information (for additional methodological details, see Web Appendix A.1).
To examine the relationship between creativity and donation behavior, we first coded the participants who did not donate as 0 and those who donated (any amount greater than zero) as 1. We then conducted a binary logistic regression analysis testing the effect of engaging in a creative T-shirt design activity on donation rate (i.e., the percentage of participants who donated to the charity). We found that a significantly higher percentage of people in the creativity condition (34.48%) donated, as compared with the percentage of people in the neutral condition (12.20%; χ2 = 4.97, p = .026). Next, we examined the effect of creative engagement on the amount of money that was donated. A Shapiro–Wilk test ([65]) indicated that the donation amount data was not normally distributed (p <.001). Thus, in accordance with prior research ([57]; [61]), we used a nonparametric Mann–Whitney U test for the analysis. We found that the average donation amount made by participants in the creativity condition (M = 7.07, SD = 13.06) was significantly higher than that of participants in the neutral condition (M = 1.10, SD = 3.26; U = 739.50, p = .016; for additional results, see Web Appendix A.2).
By collaborating with a registered charity and assessing actual donation behaviors, we found initial evidence showing that including a creative activity as part of a donation appeal can be an effective approach to enhance donation behaviors (i.e., both higher donation rates and amounts). Interestingly, one could argue that the creativity of the generated output (i.e., the T-shirt design) may also have impacted the donation behavior. To test this alternative explanation, we asked two trained research assistants (employed within the domains of creativity and advertising, respectively) to rate each T-shirt design on its creativity (1 = "not at all creative," and 7 = "very creative"). Both raters were blind to the conditions and hypothesis. Validating our manipulation, a one-way analysis of variance (ANOVA) showed that the designs produced in the creative condition (M = 3.62, SD = .80) were rated as significantly more creative than those in the neutral condition (M = 3.05, SD = .98; F( 1, 68) = 6.72, p = .01). However, we did not observe a significant relationship between rated creativity of the generated designs and the donation rate (B = .04, t < 1) or the donation amount (B = .83, t < 1). We conducted a similar analysis using only the creative condition, where natural variability in output creativity may occur (even though everyone was instructed to be creative). Again, we did not find a significant relationship between the creativity of the generated designs and the donation rate (B = −.15, t = −1.34, p = .19) or the donation amount (B = −3.07, t < 1). Importantly, our conceptualization argues that it is the act of engaging in a creativity activity that leads to enhanced donation behavior, not the level of creative output achieved. Subsequent studies replicate this finding, consistently demonstrating that creative engagement itself, rather than the creativity of the generated outcome, positively impacts donation behavior (for brevity, these findings are reported in the respective Web Appendices).
This pilot study provided initial evidence of the proposed effect, but as a real-life field study conducted in collaboration with a third party, it is not without limitations (dependence on the charity's social media platform for sampling, the messaging guidelines set by the charity, etc.; for discussion, see Web Appendix A.3). As such, our first study aimed to replicate these initial findings observed in the field in a more controlled lab setting (i.e., provide a more robust test of H1).
Study 1 used a donation context inspired and adapted from a real-life social enterprise known as Elephant Parade. This organization invites everyday consumers to create/paint their own elephant toy using an "Artbox Kit" (containing a small white clay elephant and a variety of colors) in return for a monetary donation. The proceeds are subsequently used for elephant welfare and conservation projects worldwide.
Eighty-nine undergraduate students (49 women; Mage = 20.04 years, SD = 1.27 years) at a large North American university completed this study in exchange for course credit. To begin, participants were checked in and assigned to a designated computer desk, each of which was equipped with a small donation box (see Web Appendix B.1) and a white envelope containing $2 in quarters (i.e., eight quarters). The donation box was labeled with an Elephant Parade sticker and had a slit on the top. Four quarters were left in each donation box, creating the impression that the study administrator would not be able to tell if the participant donated or not, thus reducing any demand effects and obligation to donate. Participants were told that, in addition to the course credit, they would receive $2 (in an envelope on their desks) as a token of appreciation for their participation.
The experiment adopted a one-way design in which participants were assigned to complete either a creative or neutral activity, randomized by session (i.e., we ran only one condition per session). A drawing activity (inspired by Elephant Parade's clay elephant painting) was used to induce the focal manipulation. Participants were told that the researchers wanted to put their minds at ease before the study commenced and would therefore like them to engage in a coloring activity. All participants were given a sheet of paper with a picture of an elephant (see Web Appendix B.2) and asked to color it. Those in the creative condition received a box of Crayola markers in ten different colors and were told to be as creative as possible while coloring and decorating their elephant pictures. They were also told to use any number and variety of colors they liked for the task. In contrast, those in the neutral condition received gray crayon markers only, and were told to simply color the elephant picture. In both conditions, participants were asked to spend no more than five minutes on the coloring activity. The elephant coloring task, though based on a real-life activity (i.e., Elephant Parade donation protocol), mimics a widely used creativity task in the literature: the alien task ([72]). In these types of activities, creative thinking encourages people to violate standard characteristics of a stereotypical object (e.g., an elephant, as in our study; [48]; [50]). A stereotypical elephant picture would be colored gray, whereas a nonstereotypical elephant picture would be multicolored.
Next, participants were presented with the donation opportunity and informed that the researchers were helping raise money for the nonprofit organization Elephant Parade. Participants read a short description and donation appeal from Elephant Parade (see Web Appendix B.3), and were asked if they would like to contribute; they could donate any amount of the participation money (eight quarters) they wished, and put it in the donation box. The number of quarters each participant donated served as the key dependent variable. Finally, all participants provided their demographic information (age and gender) and were debriefed before being dismissed. (Gender and age were captured in all studies. However, no effects were observed for these variables in either this or any other study. For the sake of brevity, we do not discuss them further.) After each session, the research assistants removed the quarters from the donation boxes and recorded the number of donated quarters (i.e., the total number of quarters in the box minus the four quarters initially placed in each donation box).
The elephant designs were rated on creativity (1 = "not at all creative," and 7 = "very creative") by a research assistant who was blind to the conditions and hypothesis (for sample designs, see Web Appendix B.4). A one-way ANOVA confirmed that the designs produced in the creative condition (M = 4.24, SD = 1.61) were significantly more creative than those produced in the neutral condition (M = 1.51, SD = .95; F( 1, 87) = 97.41, p <.001).
First, we explored whether there was any difference between the two conditions on the donation rate (i.e., the percentage of participants who donated to the Elephant Parade foundation) and found that a significantly higher percentage of participants (80.95%) in the creative condition—as compared with those in the neutral condition (55.32%)—donated (χ2 = 6.83, p = .009). Next, we analyzed the effect of activity type on donation amount, which was assessed by the number of quarters donated to the Elephant Parade after completing either the creative or neutral activity. The donation data did not meet the normal distribution criteria (Shapiro–Wilk test: p <.001; [65]); therefore, a nonparametric Mann–Whitney U test was again used for the analysis. We found that those who completed the creative activity (M = 4.50 quarters, SD = 3.19) donated a significantly higher number of quarters than those who completed the neutral activity (M = 2.98 quarters, SD = 3.54; U = 720, p = .022).
The obtained results provide support for our focal hypothesis (H1), namely, that engaging in a creative activity enhances donation behavior, in terms of both the likelihood of donation (i.e., donation rate) and the donation amount. The study utilized a creative activity adapted from a real-life charity and assessed donation behavior through real monetary donations. It demonstrated that engaging potential donors in creative activities, before soliciting them for donations, can be an effective way to enhance donation behavior.
One potential criticism of this study could be the different number of colors provided in the creative versus neutral conditions. However, this procedure was necessary to manipulate creativity within the context of the study. Offering a variety of colors provides participants with an opportunity for creativity, that is, to think outside the box and beyond the stereotypical characteristics of an elephant (i.e., all gray). The sole use of gray crayon markers in the neutral condition, in contrast, conforms to the stereotypical characteristics of an elephant and curtails creative opportunity. To address this potential limitation, in future studies we adopt contexts in which we can provide the same materials to participants in both conditions.
In both the pilot study and Study 1, the creative activity was directly related to the charitable cause, thereby raising a question about the generalizability of the effect—that is, whether the observed effect is domain specific or whether it holds when the creative activity is independent of the donation context. We explore this possibility in Study 2. In addition, our studies do not delineate whether the obtained results were observed because participants engaged in a creative activity (as hypothesized), or simply because the concept of creativity was salient for the participants in the creativity condition. In other words, is it necessary to actually engage in a creative activity, or can mere exposure to the concept of creativity also enhance donation behaviors? Prior research has shown that priming creativity (making creativity salient without engagement) can influence cognitive processing, thereby affecting people's propensity to engage in dishonest behaviors ([27]). We examine this question in the next study.
Study 2 was aimed to discern whether engagement in a creative activity is needed to obtain the identified effects, or rather, if exposure to a simple creative prime would suffice. To this end, we added another focal condition to the experimental design used in Study 1: this time participants were exposed to creative stimuli only, with no opportunity to participate in a creative activity. Interestingly, this condition mimics the default strategy of many charitable organizations, in which potential donors are presented solely with a donation appeal (without an opportunity for active engagement). In addition, to test the context-independent nature of the effect, the creativity activity was kept independent from the donation context (i.e., the charitable cause). Finally, we conducted the study in a real-life setting; we followed a format used by baked goods company C. Krueger's, which hosts a holiday charity event wherein customers are invited to decorate cookies and make purchases. For our study, two booths were set up in the lobby of a university building with high foot traffic, featuring large signs advertising a cookie decoration event sponsored by the charity ChildHelp. Passersby were invited to participate in the event and decorate a cookie before being solicited for a monetary donation.
One hundred seventy adults (82 women; Mage = 21.09 years, SD = 2.47 years) agreed to participate in the event and were assigned to one of the three treatment conditions: creative engagement, creative exposure without engagement, or neutral engagement. The conditions were randomized and rotated by the hour. Once passersby agreed to participate in the event, they were told they would receive $2 as a token of appreciation for participating in the event. They were given a white envelope containing eight quarters and asked to sign a form indicating receipt of the money. The signing process was necessary for participants to feel ownership of the money they had earned, before being solicited for donations later in the study. Prior research has shown that signing one's name increases this sense of ownership ([22]; [68]).
Next, one of the "staff members" guided individual participants to a table bearing a plain cookie on a paper plate, four different icing colors, and a spatula. They were also handed an event participation instruction sheet, which served as our key manipulation. Each instruction sheet had the ChildHelp Foundation logo at the top, with "ChildHelp Foundation Annual Charity Event" printed underneath (see Web Appendix C.1). The task manipulation for the two engagement conditions (i.e., cookie decoration) was adapted from [18]. In the creative engagement condition, participants were told that this was an annual charity event hosted by the ChildHelp Foundation and, as part of the event, we wanted them to decorate their cookie in the most creative manner possible using the provided materials. Those in the neutral engagement condition were given a picture of a routinely (i.e., noncreatively) decorated cookie (see Web Appendix C.2) and asked to ice the cookie as shown in the picture, using the provided materials. Those in the creative condition had the freedom to use their imagination and creativity to come up with a novel cookie design, thereby promoting creative engagement. However, in the neutral condition, participants were simply asked to copy the noncreative cookie as depicted, negating any potential creative engagement process. In the creative exposure (without engagement) condition, in keeping with prior research showing that exposure to creative images can make the concept of creativity accessible ([76]), we simply showed participants three creative cookie designs (see Web Appendix C.3) and asked them to choose the most creative one.
To assess whether our manipulation made the concept of creativity salient, we conducted a separate online study. In this study, participants were randomly assigned to complete one of the three treatment condition tasks used in the main study (creative engagement, neutral engagement, or creative exposure without engagement). They were then presented with two types of measures that captured the salience of creativity implicitly and explicitly. The obtained results showed that, as anticipated, the concept of creativity was equally salient for both the creative engagement and creative exposure conditions, and both were significantly higher than the neutral condition (for study details and complete results, see Web Appendix C.4).
Next, in the main study, all participants were given a manila envelope with a survey featuring a donation appeal and some questions about the cookie event. Each envelope was marked with a unique identification number to enable us to match participants' survey responses, donation amounts, and their assigned condition. All participants were presented with a donation appeal from the ChildHelp Foundation: a nonsectarian, nonpolitical, registered charity dedicated to helping children living in distress in North America and overseas. Furthermore, participants were told that if they decided to contribute, they could put the quarters they wanted to donate in the manila envelope and leave it in the box beside their table. Lastly, to gain initial insights into the underlying process, we captured exploratory measures of participants' positive affective state and their perceived donation impact (for details, see Web Appendix C.5). At the end of the study, the participants in the two engagement conditions were invited to take their cookie with them, while those in the creative exposure without engagement condition were given a cookie at the end of the study (for consistency with the other two conditions).
We first examined the donation rate by calculating the percentage of participants who donated in each condition. A logistic regression revealed a significant difference in the donation rates across the three conditions (χ2( 2) = 12.26, p = .002). A significantly higher percentage of participants in the creative engagement condition (81.03%) donated, compared with both those in the creative exposure (without engagement) condition (50.88%; χ2( 1) = 11.01, p = .001) and those in the neutral condition (61.82%; χ2( 1) = 4.98, p = .026). We observed no difference between the latter two conditions (p = .244). Next, we assessed the donation amount, with the number of donated quarters serving as our key dependent variable. A Shapiro–Wilk test ([65]) indicated that our data were not normally distributed (p <.001); therefore, we used the nonparametric Kruskal–Wallis test for the analysis. The obtained results revealed a significant overall main effect of the activity type on monetary donation (H( 2) = 9.8, p = .007). Pairwise comparisons showed that those in the creative engagement condition (M = 6.10 quarters, SD = 3.30) donated significantly more quarters than both those who were in the creative exposure (without engagement) (M = 4.04 quarters, SD = 3.97, p = .003) and neutral engagement (M = 4.49 quarters, SD = 3.85, p = .021) conditions. There was no difference between the latter two conditions (p = .52).
In this study, we created a charity event in which individuals participated in different activities before receiving a donation solicitation. In line with our predictions, we found that participating in an activity that enabled creative engagement (i.e., creatively decorating a cookie) enhanced donation behavior (as compared with those who either reproduced a routine cookie design or were merely exposed to creative cookie designs). Importantly, the obtained results showed that donation behavior is only enhanced when participants actually engage in a creative activity, not simply when the notion of creativity is made salient. Further, the creative activity utilized in this study was independent of the charity cause, thereby demonstrating the context-independent nature of the effect.
Our findings, so far, provide consistent evidence for the hypothesized relationship between creative engagement and donation behavior. In the following studies, we extend our examination to understand the underlying process through which creative activity and higher donation behavior are connected. In particular, we examine the mediating role of positive affect in this relationship. In addition, in Study 3 we also test the role of autonomy as a driver of creative engagement's impact on positive affect, which consequently influences donation behavior. Study 4 then explores why positive affect has such a significant impact on donation behavior. Finally, given the null findings in the creative exposure condition (on donation behavior), we dropped this condition from subsequent studies.
Study 3 was conducted to fully test H2a and identify the underlying process through which creative engagement impacts donation behavior. In particular, we tested our prediction that engaging in a creative activity heightens one's sense of autonomy, which in turn induces positive affect, leading to higher donation behavior. This study was preregistered on aspredicted.org (https://aspredicted.org/blind.php?x=ag5ci3).
Two hundred adults (117 women; Mage = 32.76 years, SD = 11.51 years) recruited from the online platform Prolific completed this study in exchange for a small monetary compensation. At the outset, participants were told that in addition to their regular compensation, they would also receive $1 as a thank-you bonus for completing the study, with an opportunity to spend this money later if they chose to. They were further informed that the study was being conducted in collaboration with the charitable organization ChildHelp Foundation and were provided with information about the organization (see Web Appendix D.1). Next, all participants were told that the ChildHelp Foundation runs an annual charitable event, wherein individuals are invited to participate in various tasks. However, given the COVID-19 pandemic, this year's event would be virtual, and the organization needed help planning the function. Thus, the organization was inviting them to participate in this study as if they were actual donors participating in the event.
The activity type manipulation used an idea generation task, mimicking the "Think outside the cereal box" campaign Kellogg launched several years ago. In the creative condition, participants were told that as part of its annual charity event, the ChildHelp Foundation was inviting them to "think outside the cereal box" and generate a fun and creative way to use Froot Loops cereal, besides eating it for breakfast. Further, participants were told to be as creative as possible and use their imagination to generate an innovative way to use Froot Loops cereal. In the neutral condition, participants were asked to "think about the cereal" and share a traditional way of how they eat the Froot Loops cereal (for detailed instructions, see Web Appendix D.2).
Once participants completed the Froot Loops task, we assessed their donation behavior, sense of autonomy, and affective state. To capture participants' donation behavior, they were told that the ChildHelp Foundation was seeking donations, and they could help by donating part or all of their $1 bonus (in multiples of $.10) to the charity. All participants were then provided with a scale from $0 to $1 in increments of $.10 to indicate their donation amounts.
Sense of autonomy was measured by adapting established measures defined in the literature (i.e., [18]; [56]). Specifically, participants were asked to indicate how they felt while generating their ideas during the Froot Loops task: ( 1) "To what extent did you feel you had autonomy in generating your ideas during the Froot Loops task?," ( 2) "To what extent did you feel you had freedom in coming up with your ideas for the Froot Loops task?," ( 3) "How free did you feel in generating your ideas for the Froot Loops task?," and ( 4) "How much did you feel you were able to express yourself when generating your ideas for the Froot Loops task?" (1 = "not at all," and 7 = "very much"). Next, to measure positive affect, they were asked to think back to the Froot Loops activity and indicate how they felt during this activity on 11 items adapted from [19] and [18]. Specifically, all participants reported how they felt while completing the Froot Loops activity: ( 1) 1 = "very negative," and 7 = "very positive"; ( 2) 1 = "very unpleasant," and 7 = "very pleasant"; ( 3) 1 = "not at all nice," and 7 = "very nice"; and ( 4) 1 = "very bad," and 7 = "very good." This was followed by seven items anchored with 1 = "not at all," and 7 = "very much," asking ( 5) how positive they felt during the Froot Loops activity, ( 6) the extent to which they enjoyed the Froot Loops activity, ( 7) the extent to which they had a good time during the Froot Loops activity, ( 8) how much fun the Froot Loops activity was, ( 9) how satisfied they felt during the Froot Loops activity, (10) how pleasurable the Froot Loops activity was, and (11) how exciting the Froot Loops activity was.
A trained research assistant blind to hypothesis and condition rated the creativity of the generated Froot Loops ideas. As we expected, a main effect of creativity emerged such that those in the creative condition (M = 3.89, SD = 1.29) generated more creative ideas than those in the neutral condition (M = 1.24, SD = .81; F( 1, 198) = 301.96, p <.001, Cohen's d = 2.46).
We first examined the effect of engaging in the creative (vs. neutral) activity on the donation rate. A binary logistic regression revealed that a significantly higher percentage of participants in the creative condition (63.37%) donated money, compared with the percentage of participants in the neutral condition (45.45%; χ2( 1) = 6.50, p = .01). Next, we examined the difference between the creative and neutral conditions' donation amounts. As in previous studies, a Shapiro–Wilk test ([65]) indicated that the data were not normally distributed (p <.001); thus, we used the nonparametric Mann–Whitney U test for the analysis. Replicating the results from the previous studies, those who engaged in the creative (M = 41.49¢, SD = 41.60¢) versus the neutral (M = 27.07¢, SD = 37.04¢) activity donated significantly more money (U = 6,034.50, p = .007).
Factor analysis showed that all four items used to capture participants' sense of autonomy loaded onto the same factor; therefore, we averaged them to create a sense of autonomy index (α = .92). A one-way ANOVA revealed that those in the creative condition (M = 6.05, SD = 1.02) reported a significantly higher sense of autonomy than those in the neutral condition (M = 4.94, SD = 1.73; F( 1, 198) = 30.61, p <.001, Cohen's d = .78). In addition, factor analysis showed that the 11 items used to capture participants' positive affective state loaded onto the same factor, and we therefore averaged them to create a positive affect index (α = .96). A one-way ANOVA revealed that those in the creative condition (M = 5.24, SD = 1.24) reported a significantly higher positive affective state than those in the neutral condition (M = 4.52, SD = 1.31; F( 1, 198) = 16.15, p <.001, Cohen's d = .56).
To test the potential underlying process paths, we first examined the mediation effect of positive affect on the creative engagement/donation rate relationship. A bias-corrected bootstrap confidence interval obtained by resampling the data 10,000 times (with activity type as the independent variable, positive affect as the mediator, and the donation rate as the dependent variable) did not include zero (β = .31, SE = .12, bias-corrected 95% confidence interval [CI] = [.11,.59]), indicating a significant indirect (i.e., mediation) effect. Next, we conducted a serial mediation analysis to understand the role of autonomy in this relationship. Serial mediation (Model 6, [35]) conducted with activity type as the independent variable, sense of autonomy and positive affect as the serial mediators (in that order), and the donation rate as the dependent variable together revealed the presence of a significant indirect effect (β = .21, SE = .09, bias-corrected 95% CI = [.07,.41]).
We also conducted the same mediation analyses for the donation amount. A bias-corrected bootstrap confidence interval obtained by resampling the data 10,000 times (with activity type as the independent variable, positive affect as the mediator, and the donation amount as the dependent variable) did not include zero (β = 4.71, SE = 1.99, bias-corrected 95% CI = [1.32, 9.02]), indicating a significant indirect (i.e., mediation) effect. Next, we conducted a serial mediation analyses to understand the role of sense of autonomy. A serial mediation (Model 6, [35]) conducted with activity type as the independent variable, sense of autonomy and positive affect as the serial mediators (in that order), and the donation amount as the dependent variable together revealed the presence of a significant indirect effect (β = 2.99, SE = 1.48, bias-corrected 95% CI = [.29, 6.10]).
Study 3 results replicated the findings from the previous studies and showed that engaging in a creative (as compared with neutral) activity enhances donation behaviors (H1). Further, the findings from this study highlighted the underlying processes through which creative engagement affects monetary donation. As we hypothesized, creative (vs. neutral) engagement induces a higher positive affect, which in turn leads to enhanced donation behavior. Importantly, we found that sense of autonomy drives the relationship between creative engagement and positive affect (H2a is fully supported). To further confirm the role of autonomy in this relationship, we conducted a supplementary study in which we directly manipulated the sense of autonomy felt by the participant. Here, we showed that when felt autonomy is mitigated, the positive effect of creative engagement on positive affect (and, in turn, the donation behavior) is attenuated (for the details of this supplementary study, see Web Appendix E).
In the next study, we further explicate the underlying process through which creative engagement impacts donation behavior by examining the pathways through which positive affect leads to higher donation behavior.
We conducted Study 4 to test H2b and provide additional insight into the underlying process through which creative engagement impacts donation behavior. In particular, we assessed the possible role of mood maintenance, social connection, and perceived donation impact in driving positive affect's influence on donation behavior outcomes. This study was also preregistered on aspredicted.org (https://aspredicted.org/blind.php?x=j8rj2w).
Two hundred adults (109 women; Mage = 34.84 years, SD = 12.91 years) recruited from the online platform Prolific completed this study in exchange for a small monetary compensation. At the outset, participants were told that in addition to their regular compensation, they would also receive $1 as a thank-you bonus for completing the study, with an opportunity to spend this money later if they chose to. They were further informed that the study was being conducted in collaboration with the charitable organization Healthier Tomorrow and were provided with information about the organization (see Web Appendix F.1). Next, all participants were told that Healthier Tomorrow runs an annual charitable event, wherein individuals are invited to participate in various tasks. However, given the COVID-19 pandemic, this year's event was going to be virtual, and the organization needed help planning the function. Thus, the organization wanted to invite them to participate in this study as if they were actual donors participating in the event.
Next, participants were randomly presented with either the creative or neutral version of the event and asked to create (reproduce) a T-shirt design. Those in the creative condition were specifically asked to design an innovative T-shirt and be as creative as possible (for detailed instructions and the sample designs produced by participants, see Web Appendix F.2). In the neutral condition, participants were simply provided with a generic T-shirt design and asked to reproduce it, thus negating any potential creative engagement (for detailed instruction and the T-shirt design provided to the participants, see Web Appendix F.3). Participants were then directed to the T-shirt customization website (customink.com) to complete the design activity. Once participants had finished creating (reproducing) their designs they were asked to save them with a unique ID provided in the survey and then use the save/share function in the T-shirt customization website to email their design to a designated email address created for the study.
Once participants completed and submitted information about their designs, we assessed their donation behavior, affective state, mood maintenance, social connection, and perceived donation impact. To capture participants' donation behavior, we told them that Healthier Tomorrow was seeking donations, and they could contribute by donating part or all of their $1 bonus (in multiples of $.10) to the charity. All participants were then provided with a scale from $0 to $1 in increments of $.10 to indicate their donation amounts.
To assess participants' positive affective state, we asked them to think back to the T-shirt design task and indicate how they felt during this activity, on the same 11 items used in Study 3 (adapted from [19]] and [18]]). Participants responded to a mood-maintenance measure adapted from [23], where they were asked to think about their donation decision and indicate their agreement with the following statements on seven-point scales (1 = "not at all," and 7 = "very much"): "I thought ..." ( 1) "I would feel good about myself if I donate," ( 2) "donating will make me feel good," and ( 3) "if I donate it would be a personally rewarding experience." To measure social connection, we adapted items from [ 8] to suit the context of our study. Participants responded to the following items on seven-point scales (1 = "not at all," and 7 = "very much"): "To what extent did you feel ..." ( 1) "closer to Healthier Tomorrow," ( 2) "connected to Healthier Tomorrow," and ( 3) "that completing the design task affected the way you think about the relationship with Healthier Tomorrow." We measured participants' perceived donation impact by means of three items adapted from [14] and [67]. These items specifically asked the participants how much they thought their donation could ( 1) make a positive difference, ( 2) be valuable, and ( 3) do a lot of good (1 = "not at all," and 7 = "very much").
Ten participants did not complete the T-shirt design activity and were excluded from the analysis (including these participants in the analysis does not change the significance or pattern of results; see Web Appendix F.4).
We first examined the effect of engaging in the creative (vs. neutral) activity on the donation rate. A binary logistic regression revealed that a significantly higher percentage of participants in the creative condition (47.87%) donated money, compared with the percentage of participants in the neutral condition (27.08%; χ2( 1) = 8.85, p = .003). Next, we examined the difference in donation amounts between the creative and neutral conditions. As in previous studies, a Shapiro–Wilk test ([65]) indicated that the data were not normally distributed (p <.001); thus, we used the nonparametric Mann–Whitney U test for the analysis. In a replication of the results from the previous studies, those who engaged in the creative (M = 22.34¢, SD = 31.81¢) versus the neutral (M = 11.25¢, SD = 24.63¢) activity donated significantly more money (U = 5,517, p = .002).
Factor analysis showed that all 11 items used to capture participants' positive affective state loaded onto the same factor, and we therefore averaged them to create a positive affect index (α = .97). A one-way ANOVA revealed that those in the creative condition (M = 5.32, SD = 1.32) reported a significantly higher positive affective state than those in the neutral condition (M = 4.72, SD = 1.40; F( 1, 188) = 9.26, p = .003, Cohen's d = .45).
As an initial step, we examined the mediation effect of positive affect on the creative engagement/donation rate relationship. A bias-corrected bootstrap confidence interval obtained by resampling the data 10,000 times (with activity type as the independent variable, positive affect as the mediator, and donation rate as the dependent variable) did not include zero (β = .19, SE = .11, bias-corrected 95% CI = [.03,.48]), indicating a presence of a significant indirect (i.e., mediation) effect.
Next, we examined the pathways through which positive affect, as induced by engaging in creative activity, impacts donation rate. We tested a sequential-parallel mediation model with creative engagement as the independent variable, positive affect as the first mediator, three factors (mood maintenance, perceived donation impact, and social connection) as a second set of mediators in parallel, and donation rate as the dependent variable (see Figure 1) using structural equation modeling (for statistics for each path in the model, see Table 1). This model is very similar to a serial mediation model; however, no order is assumed among the second set of mediators (i.e., mood maintenance, perceived donation impact, and social connection) ([21]). We used bootstrapping procedures to compute 95% CIs by generating 10,000 resamples. The results indicated significant serial indirect effects through positive affect and mood maintenance (β = .031, SE = .014, bias-corrected 95% CI = [.011,.068], p = .001) and positive affect and perceived donation impact (β = .031, SE = .014, bias-corrected 95% CI = [.010,.067], p = .001). However, the serial indirect effect of positive affect and social connection on the creative engagement and donation rate relationship was not significant (β = .005, SE = .012, bias-corrected 95% CI = [−.015,.035], p = .57). Interestingly, with positive affect in the model, creative engagement did not directly impact any of the second set of three mediators, thereby demonstrating the importance of positive affect in the conceptualization. Positive affect positively influenced all three potential mediators, but only mood maintenance and perceived donation impact significantly impacted donation rate. Thus, the positive affect experienced during a creative activity bolstered the desire for mood maintenance and the perceived donation impact, which in turn enhanced donation behaviors.
Graph: Figure 1. Sequential-Parallel Mediation Model (Study 4).
Graph
Table 1. Sequential-Parallel Mediation Model (Study 4).
| Path | Predictor | Outcome | Path Estimates (Standardized) | SE | 95% CI |
|---|
| a1 | Activity type | Positive affect | .22 | .20 | [.08,.35] |
| a2 | Activity type | Mood maintenance | −.14 | .21 | [−.27, −. 01] |
| a3 | Activity type | Perceived donation impact | .09 | .23 | [−.05,.23] |
| a4 | Activity type | Social connection | .09 | .19 | [−.03,.21] |
| b′1 | Positive affect | Donation rate | −.13 | .03 | [−.29,.02] |
| Donation amount | −.14 | 1.98 | [−.28,.002] |
| b2 | Mood maintenance | Donation rate | .31 | .02 | [.14,.46] |
| Donation amount | .25 | 1.32 | [.06,.42] |
| b3 | Perceived donation impact | Donation rate | .37 | .02 | [.20,.52] |
| Donation amount | .32 | 1.16 | [.17,.46] |
| b4 | Social connection | Donation rate | .04 | .02 | [−.13,.21] |
| Donation amount | .04 | 1.43 | [−.16,.23] |
| c′ | Activity type | Donation rate | .19 | .06 | [.05,.32] |
| Donation amount | .17 | 3.81 | [.04,.29] |
| d1 | Positive affect | Mood maintenance | .51 | .07 | [.38,.61] |
| d2 | Positive affect | Perceived donation impact | .41 | .08 | [.28,.53] |
| d3 | Positive affect | Social connection | .63 | .07 | [.52,.72] |
1 Notes: Path analysis assessing the effect of creative (vs. noncreative) engagement on donation rate/amount through positive affect, mood maintenance, perceived donation impact, and social connection (estimates and 95% CIs for individual paths).
A similar analysis was conducted for donation amount. First, a bias-corrected bootstrap confidence interval obtained by resampling the data 10,000 times (with activity type as the independent variable, positive affect as the mediator, and the donation amount as the dependent variable) did not include zero (β = 1.80, SE = 1.13, bias-corrected 95% CI = [.04, 4.44]), indicating a significant indirect (i.e., mediation) effect of positive affect. Next, to examine the pathways through which positive affect impacts donation, we used structural equation modeling to test the hypothesized process model, with donation amount as the dependent variable (for statistics for each path in the model, see Table 1). Ninety-five percent confidence intervals obtained by generating 10,000 bootstrap resamples indicated significant serial indirect effects through positive affect and mood maintenance (β = 1.51, SE = .78, bias-corrected 95% CI = [.40, 3.72], p = .004) and positive affect and perceived donation impact (β = 1.60, SE = .76, bias-corrected 95% CI = [.53, 3.65], p = .001). However, similar to what we observed for the donation rate, the serial indirect effect of creative engagement through positive affect and social connection on donation amount was not significant (β = .33, SE = .81, bias-corrected 95% CI = [−1.06, 2.26], p = .58).
The Study 4 results replicated the findings from the previous studies and showed that engaging in a creative (vs. neutral) activity induces positive affect, which in turn enhances donation behaviors. Importantly, this study further examined the underlying process, providing understanding of how positive affect influences donation behavior. We found that the path from positive affect to donation behavior was multiply determined. Indeed, perceived donation impact and mood maintenance were both shown to be drivers of the affect/donation relationship. Interestingly, we did not find evidence for social connection as a mechanism that triggered the identified effects. Thus, it appears that attributions flowing from the positive experience of creative engagement are more at play in defining enhanced donation behavior, and efforts to maintain positivity also spill over into the donation outcomes. Future research should continue to explore these identified mechanisms and discern within which charitable contexts they are most applicable and effective.
The current research examined the relationship between creativity and donation behavior. A pilot study and four subsequent experiments demonstrated that engaging in a creative activity induces a sense of autonomy, which leads to a positive affective state, which in turn results in enhanced donation behaviors (i.e., the likelihood of donation and the monetary amount donated; for a summary of all study results, see Table 2). We further showed that the positive affect experienced by the creator leads to enhanced donation behavior, due to perceptions of increased donation impact and an effort to maintain the resulting positive mood.
Graph
Table 2. Summary of Study Results.
| Donation Amount | Donation Rate |
|---|
| Units Donated | | Task Engagement | Task Engagement |
|---|
| Study | Creative | Neutral | None | Creative | Neutral | None |
|---|
| Pilot Study | USD ($) | 7.07a(13.06) | 1.10a(3.26) | — | 34.48%a | 12.20%a | — |
| Study 1 | Number of quarters (maximum 8) | 4.50a(3.19) | 2.98a(3.54) | — | 80.95%b | 55.32%b | — |
| Study 2 | Number of quarters (maximum 8) | 6.10a b(3.30) | 4.49a c(3.85) | 4.04b c(3.97) | 81.03%a b | 61.82%a c | 50.88%b c |
| Study 3 | USD (¢) | 41.49b(41.60) | 27.07b(37.04) | | 63.37%b | 45.45%b | |
| Supplementary study (follow-up to Study 3) | USD ($) | Control | 12.52ab(7.58) | 8.43ab(7.10) | | 87.88%ad | 75.41%cd | |
| Autonomy inhibited | 9.60ac(8.30) | 11.84ac(7.72) | | 70.77%ac | 80.33%c | |
| Study 4 | USD (¢) | 22.34b(31.81) | 11.25b(24.63) | | 47.87%b | 27.08%b | |
2 Notes: Standard deviations are in parentheses. The contrasts are identified with superscript notation: ap ≤.05, bp ≤.01, cp >.1, dp ≤.1.
Our work was motivated by the documented fact that charitable organizations often struggle to find effective ways to engage donors and solicit donations ([59]). Thus, a central contribution of our research is in confirming that engaging potential donors through creativity can meet this challenge by increasing engagement and enhancing donation behaviors. Substantively, we can recommend incorporating creative activities into fundraising campaigns and charity events as a viable marketing strategy. Indeed, creative activities can be implemented through social media platforms (as exemplified in our pilot study) or in person during charity events and solicitations (as exemplified in Study 2). Current industry practices suggest that some charities have begun testing this approach (i.e., engaging potential donors through creative activities before soliciting them for donations). For example, Roots and Shoots (a charitable organization that supports environmental, conservation, and humanitarian issues) regularly posts a variety of gamified challenges on its website and invites potential donors to participate. Many of these challenges encourage people to incorporate creativity in defining their solutions (e.g., for a "World Chimpanzee Day Challenge," people were asked to design and submit a creative communication graphic to spread awareness about chimpanzee protection).
To gain additional insights on practitioners' points of view (concerning our proposed strategy), we sent an email to 220 charities nationwide, inviting them to participate in a short survey. The survey asked three questions that measured the usefulness and applicability of this donation strategy. The first question assessed whether the charity had previously used a creative activity as a preface to a donation request. The second question asked whether, if presented with evidence that engaging donors through a creative activity increases monetary donation, they would implement this strategy in their donation campaigns (1 = "not very likely," and 7 = "very likely"). The final question assessed how feasible they thought it would be to implement such a strategy (1 = "not very feasible," and 7 = "very feasible"). We obtained 29 responses from the surveyed national charities (13% response rate). Interestingly, 45% of the charities mentioned that they have previously used a creative activity as a preface to a donation request—showing that our research validates a tactic already in use by some charities today. Most importantly, charities indicated they would definitely be willing to implement this strategy in their donation campaigns (M = 6.55, SD = .69; t(28) = 20.04, p <.001, compared with the midpoint) and thought it would be feasible to implement such a strategy (M = 5.38, SD = 1.68; t(28) = 4.43, p <.001, compared with the midpoint). Though a small sample, these results are encouraging and affirm that utilizing creative activities in charity campaigns is both highly relevant and feasible in the marketplace.
The current work also provides several theoretical contributions to the field. First, we advance the marketing and charity literature streams by identifying that positive affect experienced during a creative activity is a key mechanism that bolsters subsequent donation behaviors. Second, we offer a deeper understanding of why engaging in a creative activity leads to higher donation behavior through positive affect. Specifically, we show that creative engagement enhances a sense of autonomy, which in turn induces positive affect, which then positively impacts donation behavior. In addition, the relationship between positive affect and enhanced donation behavior is shown to be multiply determined. We identify two specific mechanisms that link affect and behavior: namely, the positive attributions of the impact of one's donation and the mood maintenance tendency of the participant. Third, we establish that the act of creativity itself (not just being primed with creativity as a construct) is a necessary condition to achieve beneficial donation outcomes. Finally, we confirm that the creative activity employed need not relate directly to the organization and/or charitable cause underlying the sought-after donation behavior. This is important both theoretically and practically, as it establishes generalizability in our findings and provides more freedom to charities in defining the type of creativity activity appropriate for their donation campaigns.
More broadly, the current research adds to prior work demonstrating the consequences of engaging in creative thinking tasks. While a significant amount of research has been devoted to studying various factors and cognitive processes that impact creativity ([28]; [54]; [55]), much less attention has been paid to the implications and outcomes of being creative ([27]; [71]). Our research shows that there is value in understanding what implications creativity may have for subsequent consumption behaviors. Building up our understanding of the importance of creativity is especially significant in today's consumption environment, where customers are increasingly provided with opportunities to engage in creative activities, from participating in crowdsourcing platforms (e.g., MyStarbucksIdea.com, ideas.lego.com) to engaging in customization processes (e.g., NikeID, Casetify customized phone cases).
Limitations inherent to our research approach open up several avenues for additional investigation. First, research should be directed toward developing a better understanding of the generalizability of the effects we identify. Although we demonstrated that a creative activity does not have to be specific to the charity in question to provide a positive outcome, we did not assess a broad range of charities and donation appeals. To this end, we conducted a preliminary study examining the impact of the inherent history of the charity (i.e., whether the charity was well-established; for study details; see Web Appendix G) on donation behavior. Here, we found that creative engagement indeed led to enhanced donation behavior, but only when the charity was newly established. When the charity was well-established, the donation behavior was enhanced irrespective of the type of activity utilized in the appeal. Additional research is needed to better explore this potential boundary condition and, more generally, to define other contextual factors that might moderate the effects we have documented here.
Second, most of the creative activities tested in this research involved artistry and design (e.g., cookie decorating, T-shirt design, coloring). It remains to be seen whether other forms of creativity could produce similar effects. Indeed, we believe that the effects identified in this research are likely to be observed for any enjoyable creative activities that encourage people to explore and think freely. However, we conjecture that the positive effects defined herein may be attenuated if the creative activity is more convergent in nature (e.g., identifying the one right solution). While we did not directly test the effect of a convergent creative activity on donation behavior, prior research has found that engaging in convergent creative tasks may not lead to a positive affective state ([ 9]). Future research should explore this possibility and outline the breadth of creative activities that are effective precursors to enhanced donation behaviors.
Another interesting research question arising from our work concerns the identified difference between creative engagement and creative priming (on subsequent behaviors). We found that engaging in creative activities leads to higher donation behavior, but exposure to creative stimuli does not have a parallel effect. Previous research ([27]), however, has shown that creative priming can influence cognitive processes. Thus, it is important to further distinguish between creative engagement and exposure to creative materials and understand how they differentially impact subsequent behaviors. While both creative engagement and creative priming may influence cognition, perhaps only creative engagement can induce a positive affective state. Future research could further clarify the differences between these stimuli.
Finally, future research should continue to build understanding as to when creativity leads to positive (vs. negative) outcomes. Indeed, prior research has shown both positive and negative implications for creative thinking. For example, creativity has been shown to help overcome the burden of secrecy ([29]) and to enhance one's tendency to take a target's perspective ([77]). On the negative side, previous research has found that creativity can lead to dishonesty (e.g., [27]) and enhance unhealthy choices ([31]). We find an opportunity for future work in building on these initial studies to better understand where creativity can influence downstream consumption behaviors. Indeed, we hope that future research will expand on our findings and further investigate the outcomes of creativity for individuals, charities, nonprofit organizations, and the broader marketplace at hand.
sj-appendix-10.1177_00222429211037587 - Supplemental material for Leveraging Creativity in Charity Marketing: The Impact of Engaging in Creative Activities on Subsequent Donation Behavior
Supplemental material, sj-appendix-10.1177_00222429211037587 for Leveraging Creativity in Charity Marketing: The Impact of Engaging in Creative Activities on Subsequent Donation Behavior by Lidan Xu, Ravi Mehta and Darren W. Dahl in Journal of Marketing
Footnotes 1 Karen Winterich
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship and/or publication of this article.
4 Lidan Xu https://orcid.org/0000-0001-6568-5803
5 Prior work has proposed and examined autonomy and competence as two components of creative generation (e.g., [18]; [46]). We center our conceptualization on the importance of autonomy in defining the benefits of engaging in a creative activity (on donation behavior). In a supplementary study (follow-up to Study 3), we empirically test the role of competence in our research context.
References Aderman David. (1972), " Elation, Depression, and Helping Behavior ," Journal of Personality and Social Psychology , 24 (1), 91 – 101.
Aknin Lara B. , Dunn Elizabeth W. , Norton Michael I.. (2012), " Happiness Runs in a Circular Motion: Evidence for a Positive Feedback Loop Between Prosocial Spending and Happiness ," Journal of Happiness Studies , 13 (2), 347 – 55.
Andreoni James. (1990), " Impure Altruism and Donations to Public Goods: A Theory of Warm-Glow Giving ," Economic Journal , 100 (401), 464 – 77.
Ashton-James Claire E. , Chartrand Tanya L.. (2009), " Social Cues for Creativity: The Impact of Behavioral Mimicry on Convergent and Divergent Thinking ," Journal of Experimental Social Psychology , 45 (4), 1036 – 40.
Baron Robert A. (1990), " Environmentally Induced Positive Affect: Its Impact on Self-Efficacy, Task Performance, Negotiation, and Conflict ," Journal of Applied Social Psychology , 20 (5), 368 – 84.
6 Bussell Helen , Forbes Deborah. (2002), " Understanding the Volunteer Market: The What, Where, Who and Why of Volunteering ," International Journal of Nonprofit and Voluntary Sector Marketing , 7 (3), 244 – 57.
7 Carlson Michael , Charlin Ventura , Miller Norman. (1988), " Positive Mood and Helping Behavior: A Test of Six Hypotheses ," Journal of Personality and Social Psychology , 55 (2), 211 – 29.
8 Cavanaugh Lisa A. , Bettman James R. , Luce Mary Frances. (2015), " Feeling Love and Doing More for Distant Others: Specific Positive Emotions Differentially Affect Prosocial Consumption ," Journal of Marketing Research , 52 (5), 657 – 73.
9 Chermahini Soghra Akbari , Hommel Bernhard. (2012), " Creative Mood Swings: Divergent and Convergent Thinking Affect Mood in Opposite Ways ," Psychological Research , 76 (5), 634 – 40.
Chrenka Jason , Gutter Michael S. , Jasper Cynthia. (2003), " Gender Differences in the Decision to Give Time or Money ," Consumer Interests Annual , 40 , 1 – 4.
Chung Elizabeth. (2020), " 44 Fundraising Event Ideas for Non-Profits and Charities ," Classy (September 14), www.classy.org/blog/fundraising-event-ideas-raise-money-cause/.
Clark Margaret S. , Isen Alice M.. (1982), " Toward Understanding the Relationship Between Feeling States and Social Behavior ," in Cognitive Social Psychology , Hastorf Albert H. , Isen Alice M. , eds. New York : Elsevier , 73 – 108.
Crimston Daniel , Bain Paul G. , Hornsey Matthew J. , Bastian Brock. (2016), " Moral Expansiveness: Examining Variability in the Extension of the Moral World ," Journal of Personality and Social Psychology , 111 (4), 636 – 53.
Cryder Cynthia E. , Loewenstein George , Scheines Richard. (2013), " The Donor Is in the Details ," Organizational Behavior and Human Decision Processes , 120 (1), 15 – 23.
Cseh Genevieve M. , Phillips Louise H. , Pearson David G.. (2015), " Flow, Affect and Visual Creativity ," Cognition and Emotion , 29 (2), 281 – 91.
Csikszentmihalyi Mihaly. (1997), Flow and the Psychology of Discovery and Invention. New York : Harper Perennial.
Cunningham Michael R. , Steinberg Jeff , Grev Rita. (1980), " Wanting to and Having to Help: Separate Motivations for Positive Mood and Guilt-Induced Helping ," Journal of Personality and Social Psychology , 38 (2), 181 – 92.
Dahl Darren W. , Moreau Page. (2007), " Thinking Inside the Box: Why Consumers Enjoy Constrained Creative Experiences ," Journal of Marketing Research , 44 (4), 357 – 69.
De Dreu Carsten K.W. , Baas Matthijs , Nijstad Bernard A.. (2008), " Hedonic Tone and Activation Level in the Mood-Creativity Link: Toward a Dual Pathway to Creativity Model ," Journal of Personality and Social Psychology , 94 (5), 739 – 56.
Dietrich Arne. (2004), " The Cognitive Neuroscience of Creativity ," Psychonomic Bulletin & Review , 11 (6), 1011 – 26.
Dohle Simone , Montoya Amanda K.. (2017), " The Dark Side of Fluency: Fluent Names Increase Drug Dosing ," Journal of Experimental Psychology: Applied , 23 (3), 231 – 9.
Dunn Lea , White Katherine , Dahl Darren W.. (2020), " A Little Piece of Me: When Mortality Reminders Lead to Giving to Others ," Journal of Consumer Research , 47 (3), 431 – 53.
Ferguson Eamonn , Farrell Kathleen , Lawrence Claire. (2008), " Blood Donation Is an Act of Benevolence Rather Than Altruism ," Health Psychology , 27 (3), 327 – 36.
Forman David. R. (2007), " Autonomy, Compliance, and Internalization ," in Socioemotional Development in the Toddler Years: Transitions and Transformations, Brownell and C.A. , Kopp, C.B. eds. New York: The Guilford Press , 285 – 319.
Fredrickson Barbara L. (2001), " The Role of Positive Emotions in Positive Psychology: The Broaden-and-Build Theory of Positive Emotions ," American Psychologist , 56 (3), 218 – 26.
George Jennifer M. (1991), " State or Trait: Effects of Positive Mood on Prosocial Behaviors at Work ," Journal of Applied Psychology , 76 (2), 299 – 307.
Gino Francesca , Ariely Dan. (2012), " The Dark Side of Creativity: Original Thinkers Can Be More Dishonest ," Journal of Personality and Social Psychology , 102 (3), 445 – 59.
Goldenberg Jacob , Mazursky David , Solomon Sorin. (1999), " Toward Identifying the Inventive Templates of New Products: A Channeled Ideation Approach ," Journal of Marketing Research , 36 (5), 200 –2 10.
Goncalo Jack A. , Vincent Lynne C. , Krause Verena. (2015), " The Liberating Consequences of Creative Work: How a Creative Outlet Lifts the Physical Burden of Secrecy ," Journal of Experimental Social Psychology , 59 , 32 – 9.
Guilford Joy Paul. (1959), " Three Faces of Intellect ," American Psychologist , 14 (8), 469–79.
Guilford Joy Paul. (1968), Intelligence, Creativity, and Their Educational Implications. San Diego: Robert R. Knapp.
Guilford Joy Paul. (1982), " Is Some Creative Thinking Irrational? " Journal of Creative Behavior , 16 (3), 151–54.
Handley Ian M. , Lassiter G. Daniel , Nickell Elizabeth F. , Herchenroeder Lisa M.. (2004), " Affect and Automatic Mood Maintenance ," Journal of Experimental Social Psychology , 40 (1), 106 – 12.
Harvey James W. , McCrohan Kevin F.. (1988), " Strategic Issues for Charities and Philanthropies ," Long Range Planning , 21 (6), 44 – 55.
Hayes Andrew F. (2013), An Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach. New York : Guilford.
Hendriks Marieke , Peelen Ed. (2013), " Personas in Action: Linking Event Participation Motivation to Charitable Giving and Sports ," International Journal of Nonprofit and Voluntary Sector Marketing , 18 (1), 60 – 72.
Higgins Joan Wharf , Lauzon Lara. (2003), " Finding the Funds in Fun Runs: Exploring Physical Activity Events as Fundraising Tools in the Nonprofit Sector ," International Journal of Nonprofit and Voluntary Sector Marketing , 8 (4), 363 – 77.
Holan Mark. (2014), " Ice Bucket Challenge Has Raised $220 Million Worldwide ," Washington Business Journal (December 12), https://www.bizjournals.com/washington/news/2014/12/12/ice-bucket-challenge-has-raised-220-million.html.
Hollway Stephen , Tucker Lyle , Hornstein Harvey A.. (1977), " The Effects of Social and Nonsocial Information on Interpersonal Behavior of Males: The News Makes News ," Journal of Personality and Social Psychology , 35 (7), 514 –2 2.
Hornstein Harvey A. , LaKind Elizabeth , Frankel Gladys , Manne Stella. (1975), " Effects of Knowledge About Remote Social Events on Prosocial Behavior, Social Conception, and Mood ," Journal of Personality and Social Psychology , 32 (6), 1038 – 46.
Hudson John , Jones Philip. (1994), " Testing for Self-Interest: 'The Economic Person in Sociological Context' Revisited ," Journal of Socio-Economics , 23 (1), 101 – 12.
Isaac Mark R. , Schnier Kurt. (2005), " Silent Auctions in the Field and in the Laboratory ," Economic Inquiry , 43 (4), 715 – 33.
Isen Alice M. (1970), " Success, Failure, Attention, and Reaction to Others: The Warm Glow of Success ," Journal of Personality and Social Psychology , 15 (4), 294 – 301.
Isen Alice M. , Daubman Kimberly A. , Nowicki Gary P.. (1987), " Positive Affect Facilitates Creative Problem Solving ," Journal of Personality and Social Psychology , 52 (6), 1122 – 31.
Isen Alice M. , Levin Paula F.. (1972), " Effect of Feeling Good on Helping: Cookies and Kindness ," Journal of Personality and Social Psychology , 21 (3), 384 –8 8.
Isen Alice M. , Simmonds Stanley F.. (1978), " The Effect of Feeling Good on a Helping Task That Is Incompatible with Good Mood ," Social Psychology , 41 (4), 346 –4 9.
Kaimal Girija , Ray Kendra , Muniz Juan. (2016), " Reduction of Cortisol Levels and Participants' Responses Following art Making ," Art Therapy , 33 (2), 74 – 80.
Kim Keunyeong , Schmierbach Michael G. , Chung Mun-Young , Fraustino Julia Daisy , Dardis Frank , Ahern Lee. (2015), " Is It a Sense of Autonomy, Control, or Attachment? Exploring the Effects of In-Game Customization on Game Enjoyment ," Computers in Human Behavior , 48 (July) , 695 – 705.
Koehler Friederike , Neubauer Andreas B.. (2020), " From Music Making to Affective Well-Being in Everyday Life: The Mediating Role of Need Satisfaction ," Psychology of Aesthetics, Creativity, and the Arts , 14 (4), 493 – 505.
Kowal John , Fortier Michelle S.. (1999), " Motivational Determinants of Flow: Contributions from Self-Determination Theory ," Journal of Social Psychology , 139 (3), 355 – 68.
Kozbelt Aaron , Durmysheva Yana. (2007), " Understanding Creativity Judgments of Invented Alien Creatures: The Roles of Invariants and Other Predictors ," Journal of Creative Behavior , 41 (4), 223 – 48.
Krause Vincent , Goncalo Jack A. (2019), " Fat, Drunk, and Lazy: How Engaging in Creative Tasks Can Cause Unhealthy Choices ," in Academy of Management Proceedings , 1 , 10562.
Liang Jianping , Chen Zengxiang , Lei Jing. (2016), " Inspire Me to Donate: The Use of Strength Emotion in Donation Appeals ," Journal of Consumer Psychology , 26 (2), 283 – 88.
Marsh Richard L. , Landau Joshua D. , Hicks Jason L.. (1996), " How Examples May (and May Not) Constrain Creativity ," Memory & Cognition , 24 (5), 669 – 80.
Masters John C. , Furman Wyndol. (1976), " Effects of Affective States on Noncontingent Outcome Expectancies and Beliefs in Internal or External Control ," Developmental Psychology , 12 (5), 481 – 82.
Mehta Ravi , Zhu Rui (Juliet). (2009), " Blue or Red? Exploring the Effect of Color on Cognitive Task Performances ," Science , 323 (5918), 1226 – 29.
Moreau C. Page , Dahl Darren W.. (2005), " Designing the Solution: The Impact of Constraints on Consumers' Creativity ," Journal of Consumer Research , 32 (6), 13 – 22.
Moreau C. Page , Herd Kelly B.. (2010), " To Each His Own? How Comparisons with Others Influence Consumers' Evaluations of Their Self-Designed Products ," Journal of Consumer Research , 36 (5), 806 – 19.
Munz Kurt P. , Jung Minah H. , Alter Adam L.. (2020), " Name Similarity Encourages Generosity: A Field Experiment in Email Personalization ," Marketing Science , 39 (6), 1071 – 91.
National Center for Charitable Statistics (2020), " The Nonprofit Sector in Brief ," (June 18), https://nccs.urban.org/project/nonprofit-sector-brief.
Nonprofit Finance Fund (2018), "Survey," (accessed September 30, 2021), https://nff.org/learn/survey.
O'Malley Michael N. , Andrews Lester. (1983), " The Effect of Mood and Incentives on Helping: Are There Some Things Money Can't Buy? " Motivation and Emotion, 7 (2), 179 – 89.
Rifkin Jacqueline R. , Du Katherine M. , Berger Jonah. (2021), " Penny for Your Preferences: Leveraging Self-Expression to Encourage Small Prosocial Gifts ," Journal of Marketing , 85 (3), 204 – 19.
Robiady Nurlita Devian , Windasari Nila Armelia , Nita Arfenia. (2021), " Customer Engagement in Online Social Crowdfunding: The Influence of Storytelling Technique on Donation Performance ," International Journal of Research in Marketing , 38 (2), 492–500.
Runco Mark A.. (1991), Divergent Thinking. Norwood, NJ : Ablex.
Ryan Richard M. , Rigby C. Scott , Przybylski Andrew. (2006), " The Motivational Pull of Video Games: A Self-Determination Theory Approach ," Motivation and Emotion , 30 (4), 344 – 60.
Schutte Nicola S. (2014), " The Broaden and Build Process: Positive Affect, Ratio of Positive to Negative Affect and General Self-Efficacy ," Journal of Positive Psychology , 9 (1), 66 – 74.
Shapiro Samuel Sanford , Wilk Martin B.. (1965), " An Analysis of Variance Test for Normality (Complete Samples) ," Biometrika , 52 (3/4), 591 – 611.
Stuckey Heather L. , Nobel Jeremy. (2010), " The Connection http:Between Art, Healing, and Public Health: A Review of Current Literature ," American Journal of Public Health , 100 (2), 254 – 63.
Taylor Shelley E. , Gollwitzer Peter M.. (1995), " Effects of Mindset on Positive Illusions ," Journal of Personality and Social Psychology , 69 (2), 213 – 26.
Trudel Remi , Argo Jennifer J. , Meng Matthew D.. (2016), " The Recycled Self: Consumers' Disposal Decisions of Identity-Linked Products ," Journal of Consumer Research , 43 (2), 246 – 64.
Tysiac Ken. (2016), " Getting Creative in Fundraising ," Journal of Accountancy , 222 (1), 34 –3 8.
U.S. Census Bureau (2010), " U.S. Census Bureau QuickFacts ," https://censusreporter.org/profiles/.
Vincent Lynn C. (2013), " Creative and Entitled: How the Creative Identity Entitles Dishonest Behaviors ," unpublished doctoral dissertation , Cornell University.
Ward Thomas B. (1994), " Structured Imagination: The Role of Category Structure in Exemplar Generation ," Cognitive Psychology , 27 (1), 1 – 40.
Winterich Karen Page , Mittal Vikas , Aquino Karl. (2013), " When Does Recognition Increase Charitable Behavior? Toward a Moral Identity-Based Model ," Journal of Marketing , 77 (3), 121 – 34.
Winterich Karen Page , Zhang Yinlong , Mittal Vikas. (2012), " How Political Identity and Charity Positioning Increase Donations: Insights From Moral Foundations Theory." International Journal of Research in Marketing , 29 (4), 346 – 54.
Woolf Jules , Heere Bob , Walker Matthew. (2013), "Do Charity Sport Events Function as 'Brandfests' in the Development of Brand Community? " Journal of Sport Management , 27 (2), 95 – 107.
Yang Xiaojing , Ringberg Torsten , Mao Huifang , Peracchio Laura A.. (2011), " The Construal (in)Compatibility Effect: The Moderating Role of a Creative Mind-Set ," Journal of Consumer Research , 38 (4), 681 – 96.
Yang Zheshuai , Hung Iris W.. (2021), " Creative Thinking Facilitates Perspective Taking ," Journal of Personality and Social Psychology , 120 (2), 278 – 99.
Zhang Kuangjie , Cai Fengyan , Shi Zhengyu. (2021), " Do Promotions Make Consumers More Generous? The Impact of Price Promotions on Consumers' Donation Behavior ," Journal of Marketing , 85 (3), 240 – 55.
~~~~~~~~
By Lidan Xu; Ravi Mehta and Darren W. Dahl
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 83- Machiavellianism in Alliance Partnerships. By: Musarra, Giuseppe; Robson, Matthew J.; Katsikeas, Constantine S. Journal of Marketing. Jun2022, p1. DOI: 10.1177/00222429221100186.
Ahead of Print- Database:
- Business Source Complete
Record: 84- Machine Learning for Creativity: Using Similarity Networks to Design Better Crowdfunding Projects. By: Wei, Yanhao "Max"; Hong, Jihoon; Tellis, Gerard J. Journal of Marketing. Mar2022, Vol. 86 Issue 2, p87-104. 18p. 1 Diagram, 10 Charts, 2 Graphs. DOI: 10.1177/00222429211005481.
- Database:
- Business Source Complete
Machine Learning for Creativity: Using Similarity Networks to Design Better Crowdfunding Projects
A fundamental tension exists in creativity between novelty and similarity. This research exploits this tension to help creators craft successful projects in crowdfunding. To do so, the authors apply the concept of combinatorial creativity, analyzing each new project in connection to prior similar projects. By using machine learning techniques (Word2vec and Word Mover's Distance), they measure the degrees of similarity between crowdfunding projects on Kickstarter. They analyze how this similarity pattern relates to a project's funding performance and find that ( 1) the prior level of success of similar projects strongly predicts a new project's funding performance, ( 2) the funding performance increases with a balance between being novel and imitative, ( 3) the optimal funding goal is close to the funds raised by prior similar projects, and ( 4) the funding performance increases with a balance between atypical and conventional imitation. The authors use these findings to generate actionable recommendations for project creators and crowdfunding platforms.
Keywords: crowdfunding; combinatorial creativity; funding goal; imitation; networks; novelty; Word2vec; Word Mover's Distance
Crowdfunding has grown rapidly and become an important source of capital. In 2014, start-up investment generated through crowdfunding was almost half of investment from venture capital ($16 billion vs. $30 billion; see [ 4]). In addition to its fundraising capability, crowdfunding provides useful marketing opportunities. First, project creators can use crowdfunding platforms to advertise ideas and build a reputation ([ 9]). Second, firms can use crowdfunding sites to test the market reaction to a new project ([31]; [40]).
Despite the importance of crowdfunding in investment and marketing, creating successful crowdfunding projects remains a major challenge. The top crowdfunding website, Kickstarter, applies an all-or-nothing policy, whereby a project creator collects funds only if the project's funding is successful (i.e., the raised fund pass the funding goal). On Kickstarter, only about 30% of submitted projects end up being successfully funded.[ 7] Further, about 67% of these successful projects raised no more than $10,000 (Kickstarter [20]). To help project creators, previous studies have explored various drivers of success in crowdfunding (see Table 1).
Graph
Table 1. Current Study and Literature on Crowdfunding.
| Study | Number of Observations | Main Outcome Variables | Main Source of Explanation | Content Analysis | Similarity-Based Criteria |
|---|
| This study | 98,058 ideas | Final success and funds raised | Similarity between ideas | Yes | Yes |
| Mollick (2014) | 48,526 ideas | Final success | Social networks and quality signals | Yes | No |
| Van de Rijt et al. (2014) | 200 ideas | Final success | Early success in funding period | No | No |
| Younkin and Kuppuswamy (2017) | 7,617 ideas | Final success | Creator ethnicity and gender | No | No |
| Steigenberger and Wilhelm (2018) | 197 ideas | Funding percentage | Rhetorical signals in product pitches | Yes | No |
| Kuppuswamy and Bayus (2017) | 10,000 ideas | Funding dynamics | Proximity to goal | No | No |
| Dai and Zhang (2019) | 28,591 ideas | Funding dynamics | Investor prosocial motives | No | No |
| Burtch, Ghose, and Wattal (2015) | 128,701 investors | Investor's decision | Privacy policy of platform | No | No |
| Zvilichovsky, Danziger, and Steinhart (2018) | 892 investors | Investor's decision | Make-it-happen motives | No | No |
| Agrawal, Catalini, and Goldfarb (2015) | 8,149 investors | Investor's decision | Geographical distance between investors and creators | No | No |
| Bapna (2017) | 519 investors | Investor's decision | Product certifications | No | No |
1 Notes: In terms of content analysis, [27] uses the number of spelling errors in project description; [32] use human coders to label signals in idea pitches. For studies on peer-to-peer lending, which is sometimes recognized as a form of crowdfunding, see [43], [23], and [28].
One promising yet unexplored area is the similarity pattern among projects. The prior similar projects of a new project can hold important clues for the new project's funding outcome. This thought has its root in the theory of combinatorial creativity, which views every new idea as some recombination of existing ideas (e.g., [29]; [39]). For predicting success, this theory provides a novel perspective to the current literature on crowdfunding. We may evaluate a new project directly using the level of success of its prior similar projects (given that we have a way to measure the similarity between projects). It requires us to neither specify the exact factors underlying the success of projects nor quantify the effects of these factors on success. Although this approach is intuitively appealing from both a theoretical and method point of view, it has never been applied in crowdfunding (or many other contexts in marketing).
In addition to predicting success, measuring the similarity pattern among projects enables us to characterize and examine projects in ways novel to the crowdfunding literature. First, we can measure the degree of novelty of a project and then investigate whether novelty is rewarded or penalized; whether repeated imitation of an idea devalues the idea; and, if so, at what point this devaluation starts. Second, we can measure how much a project's funding goal "overshoots" the amount of funds that were raised by prior similar projects and examine whether this overshooting (or undershooting) benefits fundraising. Third, we can measure styles of imitation (e.g., does the new project strictly follow a stereotype or also reach out for atypical elements from other types of projects?). Overall, the examination of these similarity-based characteristics can deliver useful insights for designing new projects.
Specifically, this study aims to answer the following research questions:
- How can we measure the degrees of similarity between all the projects on a crowdfunding site in an objective and automated way?
- How does the similarity pattern relate to funding performance? Specifically, to what extent…
- … does the success of prior similar projects predict a new project's success?
- … is novelty rewarded or penalized?
- … is it better to let the funding goal overshoot or undershoot the funds raised by prior projects?
- … does atypicality benefit or hurt funding performance?
- How can the platform use the similarity pattern to provide creators concrete guidance to design better projects?
To answer these questions, we collect data on 98,058 Kickstarter projects from 2009 to 2017 in the three largest categories: Film & Video, Music, and Publishing. We measure the semantic similarity between the descriptions of any two projects by applying two recently developed machine learning techniques, Word2vec ([25]) and Word Mover's Distance (WMD; [22]). We calculate the "effort" that one must incur to move the words of one document to the words of the other document. The smaller this effort is, the more similar the two documents are.
To operationalize the similarity pattern between projects, we represent it with a similarity network. The nodes represent projects. The strength of a link ( 1) increases with the degree of similarity and ( 2) decreases with the time lapse between two projects. When predicting project j, we focus on all the projects prior to j, with each prior project weighted by its link strength with j. Conceptually, the funding outcome of a project reveals the investor preference for this project. However, because investor preferences change over both time and projects, not every prior project is equally relevant for evaluating the investor preference for a given new project. In this regard, the similarity network offers a way to select the most relevant prior projects and thus provides useful information for predicting success.
We examine funding performance from two aspects: whether the funding is successful and how much money is raised. There are several novel findings. First, the average level of success by prior projects, weighted by their links to the focal project, is a significant predictor of the focal project's funding performance. This result holds after we control for the project creator's prior success, the project's funding goal, description length, and the presence of images and videos in the description. Overall, the similarity network is an information source to significantly improve the out-of-sample prediction for funding success.
Second, a project's funding performance exhibits an inverted U-shaped relation with the novelty of the project. Here, we measure the novelty of a project via the sum of its links with all prior projects. A larger sum means a greater amount of total similarity between the project and its prior projects, indicating a lower level of novelty (or a higher level of imitativeness). This inverted U-shaped relation can be surprising yet is intuitive. It suggests that successful projects tend to strike a balance between ( 1) being novel and ( 2) appearing familiar to investors.
Third, it is optimal to set the funding goal close to the amount of funds raised by prior similar projects. Specifically, we define goal overshoot as the focal project's log funding goal minus the average log funds raised by prior projects, weighted by their links to the focal project. In other words, goal overshoot compares a project's funding goal against a benchmark set by prior similar projects. We find that setting a goal either too low or too high compared with the benchmark decreases the funds to be raised (i.e., an inverted U-shaped effect). Setting a goal lower than the benchmark has a limited effect on the probability of success; setting a goal too much higher than the benchmark decreases the probability of success.
Fourth, a project is more likely to succeed when it grounds itself in a stream of closely linked projects yet simultaneously borrows from some projects outside this stream. Under combinatorial creativity, this outside-the-stream imitation constitutes a nonconventional or "atypical" use of prior ideas (an example of atypical imitation outside the crowdfunding context is when an article in marketing cites research from a largely unrelated discipline such as physics or biology). We find an inverted U-shaped relation between atypicality and funding performance; neither too little nor too much atypicality benefits fundraising.
Drawing on these findings, we devise two recommendation tools that the platform may use to help creators improve projects. Our first recommendation tool helps to set funding goals. Choosing the goal is an important decision for project creators, which matters greatly for funding outcomes ([27]). However, research has provided little concrete guidance on setting the goal optimally. Our results enable us to benchmark a project's goal against the funds raised by prior similar projects. For many projects, we find that the goals were far from optimal, in which case we recommend a ±10% goal adjustment to improve expected funding outcomes. Our second recommendation tool helps creators improve project content. Specifically, we recommend a prior project for the creator to imitate. The recommendations are customized for individual projects to increase each project's chance of success.
The crowdfunding literature has focused on several aspects of crowdfunding as endpoints, including the final success of funding campaign (which is also a focus of this research), the dynamics during fundraising period, and investor decisions. The literature also has focused on a variety of sources of explanatory variables for these endpoints, including the project content (as in this article), creator characteristics, and investor behaviors. Table 1 provides a summary. Our study contributes to the literature in several important aspects. First, we apply combinatorial creativity in crowdfunding, using machine learning to construct a similarity network among projects. Second, we significantly improve the out-of-sample prediction of funding success. Third, we derive novel insights on how the similarity pattern with prior projects affects a project's funding performance. From a methodological point of view, the machine-learned similarity network provides important advantages. Previous studies look for explanatory criteria based on specific predefined factors, such as quality signals ([ 3]; [27]) and rhetorical signals ([32]). In contrast, our similarity-based criteria do not restrict attention to specific factors. In addition, we base the similarity measures on the contents of project descriptions, which are particularly important in crowdfunding because investors make decisions on the basis of these descriptions. As a result, our models gain insight, explanation, and prediction.
This study also contributes to the broader literature on ideation examined under various contexts aside from crowdfunding, such as crowdsourcing, consumer product development, and the music market. Table 2 provides a summary. These studies have analyzed several sources of explanation for idea success, including the templates of innovation, problem decomposition, creator's social network, idea prototypicality, or genre divergence. Compared with these studies, the current study is unique in its focus on crowdfunding, a source of explanation based on the similarity pattern between ideas, novel findings, and the massive number of ideas examined.
Graph
Table 2. Current Study and Literature on Ideation.
| Study | Context | Number of Ideas | Main Outcome Variables | Main Source of Explanation | Methodology of Analysis |
|---|
| This study | Crowdfunding | 98,058 | Final success; funds raised | Similarity between ideas | Empirical |
| Goldenberg, Mazursky, and Solomon (1999) | Crowdsourcing | 359 | Idea quality | Idea templates | Experiment |
| Luo and Toubia (2015) | Crowdsourcing | 4,298 | Idea quality | Idea cues | Experiment |
| Stephen, Zubcsek, and Goldenberg (2016) | Crowdsourcing | 786 | Idea quality | Social networks | Experiment |
| Toubia and Netzer (2017) | Crowdsourcing | 4,129 | Idea quality | Prototypicality | Experiment |
| Goldenberg, Lehmann, and Mazursky (2001) | Consumer products | 197 | Product success | Idea templates | Empirical |
| Berger and Packard (2018) | Music popularity | 1,879 | Song popularity | Genre divergence | Empirical |
The next sections describe the concept of combinatorial creativity, the research design, the results, and the managerial implications. We conclude with a discussion of the findings and possible extensions of the research.
Before we describe the details of our study, we briefly overview the concept of combinatorial creativity, which forms the conceptual basis for our analysis. Previous studies on idea generation have addressed how creativity is generated from a cognitive standpoint. A general descriptive framework is the "Geneplore" model proposed by [14]. Geneplore is a portmanteau of the words "generate" and "explore," signifying that the development of creative ideas is an iterative interaction of two processes: the generation process and exploration process. In the generation process, people retrieve various pieces of information based on prior knowledge. Then, they create seeds of ideas, called preinventive forms, by recombining those retrieved components. In the exploration process, the preinvented forms can be focused, expanded, or evaluated in further depth. After going through these two processes iteratively, people finally come up with creative ideas.
Underlying the Geneplore model is the combinatorial nature of creativity, which sees a new idea as a recombination of existing knowledge. [29], p. 66) conceptualizes economic development as "the carrying out of new combinations." [39], p. 331) theorizes that "knowledge can build upon itself in a combinatoric feedback process." Although the concept of combinatorial creativity has a long history in the literature of innovation and growth, researchers have only recently begun to gather empirical evidence and applications on this topic. A most representative work is [36]. The authors examine the impact of a piece of scientific research in relation to its bibliography. They show that scientific research tends to have a higher impact when it balances the uses between atypical knowledge and conventional knowledge. Later studies attempt to expand the application of combinatorial creativity in different contexts such as patents ([41]), idea competition ([33]; [35]), and motion pictures ([38]).
Compared with the aforementioned studies, ours is unique in several ways. First, this research is the first to apply the concept of combinatorial creativity in the context of crowdfunding. Second, we construct the pairwise connections between ideas on the basis of direct comparison between the content of ideas. In contrast, [36] use the citation network between papers, [41] use the classification by the U.S. Patent and Trademark Office, [33] use the communication network among ideators, [35] examine how an idea deviates from the average, and [38] infers the similarity between movies indirectly from consumer revealed preferences. Third, previous studies mostly focus on the similarity or connection pattern itself. We devote attention to the interactions between the similarity pattern and idea attributes. For example, we examine the extent to which a project's funding goal "overshoots" the typical level of funds raised by prior similar projects.
This section describes the data, construction of the similarity network, network-based metrics for predicting funding outcomes, control variables, and models.
Before we give details on these components of our analysis, it is useful to discuss the conceptual role of the similarity network, as it is the core of our research design. When we try to predict the funding outcomes for a new project, the key determinant is investors' preferences—what types of projects they find worth supporting, what elements of ideas they find appealing, and so on. In this regard, the funding outcomes of historical projects provide a database containing many instances of revelation of investors' preferences (where each historical project constitutes one instance). However, the crucial question is which part of this database we should use when predicting the funding performance of a given new project. An intuitive answer is not difficult. For example, if the new project aims to develop a video about motorcycle ride trips, then the success or failure of a prior film project on a family drama is unlikely to offer relevant information. However, the funding outcomes of prior projects on traveling or motorcycles should seem relevant. In addition, we probably should give more weight to more recent projects. Projects completed ten years ago, even if their content is similar to the new project's, may no longer be relevant because investor preferences may have significantly changed since then.
We use the similarity network to operationalize the aforementioned intuition. We let each link in the network measure the degree of similarity (computed using machine learning methods) as well as the time proximity between a project pair. Then, when predicting the funding outcome for a focal project, we weigh each prior project by its link strength with the focal project. In addition to selecting relevant prior projects for prediction, the network allows us to characterize a new project in relation to prior projects (e.g., novelty of the new project). These characteristics may affect the investors' preferences and thus funding outcomes. We define and discuss the metrics for these characteristics subsequently in this section.
We collect data from Kickstarter, one of the largest reward-based crowdfunding platforms. We acquire the information of each project from May 2009 to the end of 2017.
We focus on English-language projects in the United States. We also focus on the projects belonging to the top three largest project categories: Film & Video, Music, and Publishing. To adjust for inflation over time, we normalize all the monetary values (e.g., funding goal and funds raised by projects) to 2017 dollars using the Consumer Price Index. Following [27] procedure, we eliminate the projects with outlying project goals (i.e., goals smaller than $100 or larger than $1,000,000 [1.24% of data]). Furthermore, we do not consider the projects with fewer than 50 words in the description text (2.5% of data), in order to correctly measure the content similarity between projects. In the end, our data include 98,058 projects on Kickstarter (Film & Video: 37,641 projects, Music: 35,943 projects, and Publishing: 24,474 projects).
We focus on two measures for the outcome of a project: ( 1) funding success and ( 2) funds raised. Funding success is a binary variable. A project is classified as a success if it reaches the preset funding goal before the project campaign ends (project creators set the campaign lengths of their projects, the vast majority of which are either 30 or 60 days [the maximum length allowed is 60 days]). "Funds raised" denotes the total amount of money that the project collects at the end of the project campaign, whether it exceeds the funding goal or not. Note that under Kickstarter's all-or-nothing policy, a project cannot collect the funds unless the funding goal is successfully reached. Nevertheless, both funding success and funds raised are important outcomes reflecting the investors' interests and confidence in the project.
The average rate of funding success over our entire data from 2009 to 2017 is about.46. The average success rate is higher in the Music category than the other two categories (Film & Video:.43, Music:.56, Publishing:.34) and has persisted over time. This difference in the success rate is partly explained by Music projects tending to ask for lower funding goals (median goal in each category: Film & Video: $7,500, Music: $4,359, Publishing: $5,107).
To measure the similarity between projects semantically, we apply machine learning techniques on the descriptive texts of crowdfunding projects. Note that our network is constructed to represent the similarity relations between projects (i.e., nodes are projects). This network is different from, for example, the semantic network between words in [35]. The method we adopt to measure similarity between projects is called WMD, which is based on Word2vec. We describe this method next. Alternatively, one could measure similarity with latent Dirichlet allocation (LDA; [ 7]) that factors text documents into topics. However, WMD provides a better prediction performance than LDA in our application (see Web Appendix A). WMD also has some practical advantages for our application, as we discuss subsequently.
We apply Word2vec ([25]; [26]) to measure the similarity between the words in project descriptions. Word2vec is a machine learning algorithm designed to learn the semantic relations between words. Specifically, it applies a two-layer neural network to convert words into high-dimensional vectors. Taking a large corpus of text as an input, the model generates a vector to represent each word in the corpus. Word2vec positions each word in the vector space such that words that share common contexts in the corpus are located close by. Word2vec is also capable of capturing many semantic relations with vector operations. For example, the vector representing "King" minus the vector for "Kings" is equal to the vector for "Queen" minus the vector for "Queens."
One implementation of Word2vec is known as the skip-gram model. It uses the two-layer neural network to predict the surrounding words when a central word is given. Google adopts this implementation to provide a pretrained Word2vec using the Google News corpus. The vocabulary contains more than three million unique words or phrases, each of which is presented by a 300-dimensional vector ([25]). We use this Word2vec trained by Google in this study. Alternatively, one can train context-specific Word2vec on the crowdfunding project descriptions. However, using Google's Word2vec here allows for easier implementation, especially for the platform, as we discuss next.
A simple way to measure the similarity between two documents is to count the overlapping words that appear in both documents. However, two similar documents do not need to share even a single word (consider, e.g., "President speaks on immigration" and "Trump talks about borders"). Therefore, we want to account for the similarities between words.
We apply a recently developed method of measuring document similarity called WMD ([22]), which is built on Word2vec. Specifically, the method regards a document as the collection of word vectors as prescribed by Word2vec and minimizes the total travel distance of moving all word vectors in one document to all word vectors in another document. The minimized travel distance is used as a measure of dissimilarity between the two documents. WMD has been shown to perform better than previous methods in measuring document similarity (see Web Appendix A).
The application of WMD with Google's pretrained Word2vec makes the computation of similarity between two crowdfunding projects a truly pairwise operation independent of other crowdfunding projects. This feature is different from the LDA and related text analysis models (e.g., latent semantic indexing; [12]), which require training first on the entire corpus to extract latent topics. As new projects are posted to the crowdfunding website every day, the corpus changes over time, which would require retraining the text model every once in a while. The WMD, coupled with Google's Word2vec, does not require such retraining and thus should be easier to implement for the platform.
Before moving on, we briefly describe some summary statistics to help explain our implementation. Take the Publishing category as an example. There are 24,474 projects in this category. We take the text in the description (including the blurb) of each project. After removing stop words (e.g., "the," "a"), rare words, and words not in Google's Word2vec vocabulary, there are 19,032 unique words that we use in computing the WMDs between projects.[ 8] An average project has 209.64 unique words (SD = 161.73). There are 299,476,101 unique pairs ( = 24,474 × (24,474–1)/2) of projects in this category. A WMD is computed for each pair. We do not consider similarities between projects of different categories (e.g., a Publishing project and a Music project). Accounting for cross-category similarities entails a much larger computational cost, but it may improve predictions and insights. This is an interesting topic for future research.
For each category, the distribution of the WMDs is bell-shaped and mostly symmetrical. The distributions of the three categories are very similar (Film & Video: M = 3.10, SD = .18; Music: M = 3.01, SD = .20; Publishing: M = 3.12, SD = .18). We note that the mean WMD in the Music category is slightly smaller than the other two categories. We think this difference may be because films and books tend to involve lengthier and more explicit storytelling, which allows for more ways to differentiate projects.
Networks are generally designed to keep track of pairwise relations between individuals. We use a network to present the similarity relations between projects in a given crowdfunding category. Each node represents a project. An unweighted network can be constructed by placing a link between two projects if and only if their similarity passes a threshold. Such an unweighted network is actually sufficient for deriving the main qualitative results in this article. However, we can achieve better prediction performance by constructing a weighted network, where every pairwise relation takes a continuous value . We call this value the link strength between node i and j.
As discussed, we consider two properties when assigning the value for : ( 1) the value of should increase with the similarity between the contents of i and j, and ( 2) the value of should decrease with the time gap between i and j. As discussed, the second property is to account for the fact that investor tastes may be time-varying. Together, the two properties enable us to focus on the projects that are more similar and more recent to the focal project when predicting the funding outcomes for the focal project.
Specifically, let denote the WMD between any two projects i and j. Let denote the starting date of any project i. Let be a decay factor. We specify that, for any i and j in the same category such that ,
Graph
( 1)
In Equation 1, L is the logistic function. We choose the exact values of , , and using a 10-fold cross-validation (5-fold and 15-fold cross-validation give very similar values). The logistic function leads to a significantly better prediction performance for funding success than some of the other functions of , such as , , or . Also note the logistic function becomes a step function when and are sufficiently large, so the specification in Equation 1 contains the unweighted network as a special case. Finally, we note that only the relative link strengths carry a meaning; scaling by a common factor for all (i, j) pairs does not affect our subsequent analyses.
We focus on several metrics based on the similarity network to help predict new projects' funding performance. Before we detail each metric, it is useful to set up an illustrative example, which we use throughout the article (see Figure 1). For illustration purposes, we restrict attention to only six projects—the actual similarity network contains tens of thousands of projects for each category. The thickness of a link represents the link strength . The horizontal position of a node reflects its starting date . Each project also has a completion date . In this example, we treat for simplicity.[ 9]
Graph: Figure 1. An illustration of similarity network.
Graph
Table 3. Selected Details of the Example Projects.
| Project | Details |
|---|
| 1 | Title: World Run—One Step at a Time. |
| Creator: Alex H. ("Alex is a young filmmaker from a small town …") |
| Goal: $15,000 |
| Blurb: "A documentary exploring what it is like to travel miles on foot. Watch Jesper Olsen break the world record for ultra-running." |
| Excerpt from Description: "Jesper Kenn Olsen is approaching the finale of a four-year-long ambitious goal to run around the world. He has gone places that very few people have gone before, and seen the world at a runner's pace—roughly 26 miles a day. This documentary chronicles primarily the second half of Jesper's journey: from the southernmost tip of South America to the northern shores of Newfoundland … Jesper is an ultra-runner from Denmark … Part travelogue, part sports documentary, and part commentary on human identity, it will examine human achievement and investigate how defined personal and politcal boundaries really are." |
| 2 | Title: Low Life |
| Creator: Luke E. ("Luke began his career as a freelance photographer …") |
| Goal: $10,000 |
| Blurb: "A comedy about depression." |
| Excerpt from Description: "THE GOAL. We are looking to raise money to shoot this 6 × 5 minute web series set in Los Angeles … Whilst all the cast and crew are working for free or deferred payments, the money raised will help us hire gear, and to pay for locations, catering, and insurances … THE STORY. The modern world is a scary place—suffocating, fractured, and alienating. On the surface, Jef is a regular guy. He' s good looking, charming, and funny. He also has a good job and a beautiful girlfriend. But he just doesn't feel right. And he' s not sure what to do about it. He is bombarded by advertising messages, annoying campaigners, and friends who all have solutions." |
| 3 | Title: Atlanta to Alaska and Back, a Vintage Motorcycle Documentary. |
| Creator: Aurorah Y. ("I'm an animator and astrologer from Atlanta …") |
| Goal: $25,000 |
| Blurb: "I'm making a film about restoring a vintage motorcycle and riding it from Atlanta to Alaska and back." |
| Excerpt from Description: "I'm the creator behind a feature length documentary film about restoring a 1972 Honda CB 350, and yes; I am also the girl who is doing the restoration (with help), and yes, riding the restored vintage bike from Atlanta to Alaska and back. It's an 11,000 mile round trip. This campaign is to cover the first 1/3 of production on the film which will encompass the bike's restoration. I will be doing the majority of the work on my bike with the help and guidance of skilled experienced mechanics … I'll be camping out a lot, and my camera man and his van will going to follow me and help film the trip." |
| 4 | Title: Burning Memories |
| Creator: Amelia B. ("Amelia B. is a 21-year-old filmmaker …") |
| Goal: $1,500 |
| Blurb: "A family drama about a woman and her younger half-brother dealing with their father's abandonment issues to find happiness." |
| Excerpt from Description: "Burning Memories is a senior thesis student short film about a family drama taking place in current times. The story is about a 30-year-old woman who seeks control over all the men in her life because of her father's abandonment issues; but, it is the arrival of her 12-year-old half-brother that makes her reflect on her life choices and question whether they are what make her happy. We are planning to raise more than $1500." |
| 5 | Title: Head First Diaries: A Pan-American Adventure Travel Series |
| Creator: Miles & Aaron ("Miles is 22 years old, and a recent grad from UCLA …") |
| Goal: $15,000 |
| Blurb: "Head First is a visual journey across the Americas through the eyes of two young explorers pursuing freedom understanding & adventure." |
| Excerpt from Description: "Our goal is to put you in the front seat with us over the course of the next year on this adventure of a lifetime. What sets Head First apart is our passion for telling a story about travel that captures the beauty of not just the destination but also the journey … The show will include 18 episodes that document the outstanding nature, people, and culture life on the road brings … Two young explorers. One dog. A quarter century old VW van." |
| 6 | Title: US Highway 98, The Florida I Never Knew - Motorcycle Journey |
| Creator: Drew P. ("Drew is a fourth year student at UCF …") |
| Goal: $1,000 |
| Blurb: "A five day solo motorcycle camping trip throughout the state of Florida exclusively along U. S. Highway." |
| Excerpt from Description: "Hey Kickstarter! My name is Drew Perlmutter and I'm a documentary filmmaker … U.S. Highway 98 was established in 1933 and runs for 964 miles. 671 of those miles run through Florida. The highway coincidentally starts in my hometown of Palm Beach, and runs all the way to the Alabama / Florida state line … I will be spending 5 days traveling exclusively on this highway for its entirety throughout Florida … So, why a motorcycle? Besides the incredible sense of adventure from being on the open road, it' s about being absorbed in the scene, rather than being a passive observer." |
The six projects are taken from the Film & Video category in our data (for details, see Table 3). Project 1 is a documentary about a world runner's journey. Project 2 is a comedic portrait of depression issues in modern lives. Project 3 is a documentary of restoring and riding a motorcycle for an Atlanta-Alaska trip. Project 4 is a family drama of dealing with abandonment issues and finding happiness. Project 5 is a pan-American travel journey in an old van. Project 6 is a motorcycle journey throughout Florida.
Research on creativity has long been interested in the role of novelty in new product performance (see, e.g., [15]; [30]). Intuitively, a new idea will be perceived as less novel (or more imitative) when it is very similar to many previous ideas, especially the more recent ideas. The similarity network offers us a unique opportunity to objectively measure an idea's degree of novelty using machine learning techniques.
Let and denote the starting and completion dates of any project i. For any focal project i, we define the amount of prior similarity as . We take the logarithm because the sum has a highly skewed distribution. This metric resembles the definition of "degree" commonly used in social network analysis, with the notable exception that it restricts attention to the nodes that arrive before the focal node. Take the network in Figure 1 as an example. For project 5, we sum across the four links between it and projects 1–4. For project 6, we sum across all five links connected to it.
The amount of prior similarity (inversely) measures the novelty of the focal project against the projects prior to it. Ex ante, the sign of the effect of novelty on funding performance is ambiguous. On the one hand, a very novel idea may be exciting but also difficult to understand and thus appreciate. On the other hand, an imitative idea is easy to understand but may appear unexciting and mundane to investors.
As discussed, one purpose of the similarity network is to help us select the most relevant prior projects when predicting the success of a project. For any focal project, we define the prior success rate as the average level of success among the projects prior to this project, with each prior project weighted by its link with the focal project. Formally, let be a dummy indicating whether a project j was successfully funded. The prior success rate for a focal project i equals . By construction, takes into account both the similarity and recency of a prior project j with respect to the focal project. Conceptually, the prior success rate tries to capture a new project's quality (i.e., how appealing it is to investors) through the quality of historical projects (revealed by funding outcomes).
Take Figure 1 as an example. The prior success rate for project 6 depends mostly on project 3 (failed) and project 5 (successful). It also depends on the other three projects, but to a lesser extent. The exact prior success rate for project 6 is.416. As another example, the prior success rate for project 5 is.099, which comes mostly from project 3 (failed) and, to a lesser extent, projects 1, 2, and 4. For those familiar with the social network analysis, this metric should resemble the definition of average peer behavior, except that it is restricted to the peers that arrive before the focal node.
In the calculation of prior success rate, all successes (or failures) of prior projects are treated equally regardless of how the successes (or failures) had been affected by the observed attributes of these projects. This equal treatment is convenient but omits some useful information. Particularly, consider two previous projects that were both successfully funded, but the first had a high funding goal while the second had a very conservative goal. Statistically, it should seem reasonable to regard the success of the former project to be "bigger" than the latter.
Along this line of thought, we construct a refined version of the prior success rate. Specifically, we first estimate a logistic model regressing funding success onto a set of creator and project characteristics including the goal, description length, time trend, and so on (for the list, see the "Control Variables" subsection). Use to denote the residual of this regression for project i. We define as the prior success residual for project i. In other words, we replace with in the definition of the prior success rate. As we show subsequently, this replacement leads to a better prediction performance.
Funding goal is an important instrument in the hands of project creators, which affects not only whether a project will be successfully funded but also the amount of funds it will collect ([27]). However, academic research has provided little guidance on how to set the goal optimally. One reason that makes offering guidance difficult is that there is likely no one-size-fits-all solution; the optimal goal should be highly dependent on the nature and content of the project. In this regard, the similarity network offers us a unique opportunity to use the historical funding achievement by similar projects as a project-specific reference point for setting the goal.
Along this line, we measure the difference between the focal project's log funding goal and the average of the log funds raised by prior projects, with each prior project weighted by its link length with the focal project. Formally, the goal overshoot for i is defined as . Here, denotes the funding goal of project i and denotes funds raised by project j. They are put in the log scale because their distributions are highly skewed.
Our goal overshoot measures how much more funds the focal project asks than what the prior projects could raise. Under the social network analogy, it compares a focal node's goal with the average achievement of its preceding peers. For example, in Figure 1, the goal overshoot of project 6 mostly compares the project's goal of $1,000 with the funds raised by project 3 ($2,100) and project 5 ($15,673). It also takes into account the funds raised by the other three projects, but to a lesser extent. The exact goal overshoot is negative, at −1.78. This negative overshoot indicates a relatively conservative goal by project 6.
A handful of more recent studies have found that the so-called atypical combination of prior ideas has an impact on a new idea's success ([35]; [36]; [38]). The general concept of atypical combination is as follows. Usually, a new idea imitates multiple prior ideas. The imitation of a prior idea is atypical if it looks uncommon or "out of place" compared with the other prior ideas imitated by this new idea. An example can be made in the context of scientific papers. Consider the bibliography of a marketing paper. In this case, citing marketing papers can be regarded as conventional, but citing a physics paper would be regarded as atypical. From a network perspective, a salient feature of this physics paper is that it is likely isolated from (i.e., neither citing nor cited by) the other papers in this bibliography.
Unfortunately, it is difficult to define atypicality with a weighted network. So, for the definition of atypicality, we work with an unweighted network (like a citation network). We choose a cutoff and regard two projects as similar and thus linked in the unweighted network if their WMD is below the cutoff. Then, for each project i, there is a subnetwork consisting of i's prior similar projects (which corresponds to a paper's bibliography in the analogy to citation networks). We define the atypicality for i as the proportion of the isolated nodes in this subnetwork. In the special cases where i has no or only one prior similar project, the concept of atypicality does not apply, and we set the atypicality to zero. This is also the definition used in [38]. Specifically, we set the cutoff for the WMD at the.5 percentile of the WMDs across all pairs in a category. Higher cutoffs make isolated nodes too rare. Lower cutoffs result in too many projects having no prior similar projects.
Our control variables fall into two categories: ( 1) project-related features or characteristics and ( 2) creator-related features or characteristics.
We control for the features of the focal project that may affect the project's funding performance. Specifically, we include the log funding goal of the project, the log number of images in the project description, whether the project description features a video (dummy), the log length of the project description text, the project category (dummies), and controls for the time of the project (a time trend and quarter dummies). These features were included in [27], [42], and [ 9].[10]
Creators may learn how to make a successful project through their experience on the platform. Accordingly, we include ( 1) a dummy indicating whether the creator had any prior projects before the focal project and ( 2) the average success rate of the creator's prior projects. The second feature is set to zero if the creator had no prior project. To control for creator heterogeneity more extensively, one could use creator fixed effects. However, this approach is infeasible given the nature of our data: close to 80% of the projects were made by one-time creators.
Our regression models focus on the effects of the network-based metrics on the funding performance of crowdfunding projects. We use a logistic model to analyze the funding success:
Graph
( 2)
In Equation 2, denotes whether project i is successfully funded (i.e., the total funds raised exceeded the funding goal). Vector collects the project-related and creator-related control variables. Note that includes a time trend, quarter dummies, and category dummies (Film & Video, Publishing, and Music). Vector collects metrics based on the similarity network (some of which are information-weighted, as we discuss subsequently).
In addition to , we examine how and affect the funds raised by projects. We specify the model as
Graph
( 3)
where denotes the total amount of funds that the project i raised, regardless of whether it exceeded the funding goal.
We hold out the last quarter of 2017 at the end of our data for the out-of-sample prediction test. We estimate Models 2 and 3 on the projects from January 2016 to September 2017, which we call the main sample (Web Appendix B shows that our results are not sensitive to the choice of January 2016 as the start date of the main sample). The projects between 2009 and December 2015 are included in the computation of the network metrics for all the projects in the main sample as well as holdout sample. For example, when calculating the network metrics for a project that started on May 1, 2016, we include all the projects completed between 2009 and May 1, 2016 (in the same crowdfunding category). Table 4 provides the size of each sample as well as the summary statistics of variables in the main sample.
Graph
Table 4. Summary Statistics.
| Sample Size | Film & Video | Music | Publishing | All |
|---|
| From May 2009 to Dec. 2015 | 31,465 | 30,303 | 18,647 | 80,415 |
| Main sample (Jan. 2016 to Sep. 2017) | 5,558 | 5,043 | 5,139 | 15,740 |
| Holdout (last quarter of 2017) | 618 | 597 | 688 | 1,903 |
| Variable | Average | SD | Min | Max |
| Success | .43 | .49 | 0 | 1 |
| Funds raised (log) | 5.78 | 3.42 | 0 | 14.15 |
| Prior success residual | .04 | .35 | –1.30 | 1.28 |
| Amount of prior similarity | 4.94 | 2.34 | .00 | 9.43 |
| Goal overshoot | 8.46 | 8.11 | –25.12 | 55.47 |
| Atypicality | .04 | .17 | 0 | 1 |
| Log funding goal | 8.56 | 1.42 | 4.61 | 13.82 |
| Log number of images | .87 | 1.09 | 0 | 4.70 |
| Featuring a video | .73 | .44 | 0 | 1 |
| Log description length | 5.77 | .89 | 3.22 | 8.68 |
| Creator with past projects | .19 | .39 | 0 | 1 |
| Creator past success rate | .11 | .30 | 0 | 1 |
2 Notes: The reported variable statistics in the lower panel are for the main sample. We add 1 to the variable before taking log for the following variables: funds raised, funding goal, and number of images. The prior success residual and goal overshoot are information-weighted.
For parameters , , and , which determine link strengths (see Equation 1), we choose their values by cross-validation on the main sample. Specifically, we maximize the area under the receiver operating characteristic (ROC) curve of the success model. For the computation of the prior success residual, the logistic model used to extract the residual is estimated using all the projects before the holdout sample. These choices ensure that no data from the holdout sample are used to estimate any part of our models.
Beyond prediction, we may make causal interpretations on the coefficients in Models 2 and 3, if there is no omitted variable that affects funding outcomes and correlates with the regressors. In our context, the probable omitted variable is the project quality that is unobserved to researchers but taken into account by creators when posting the project. This omitted variable is largely unaddressed by previous studies (e.g., [27]; [32]). In this article, we control for this quality using the prior success residual. It is possible that the unobserved quality known to creators goes beyond what historical projects can tell; they may use knowledge from other crowdfunding sites or other creative industries. Controlling for such outside knowledge is beyond the scope of this research. In addition, note that while potential endogeneity may affect the interpretation of our coefficient estimates, it does not invalidate the prediction performance of our models ([13]).
Some of our network metrics are based on averages across prior projects. We multiply these metrics by an "information weight" before they enter in the models. Next, we use the prior success residual to explain the rationale behind this weighting. The prior success residual is an average weighted by . Intuitively, measures the relevant information supplied by prior project j for predicting about i. Thus, the information supplied by the prior success residual, which aggregates across all the j such that , increases with .
To capture the amount of relevant information, we multiply the prior success residual with an information weight . We make this specification for for two reasons. First, the prior success residual should have little relevance when there is little similarity between project i and prior projects. Thus, the information weight should be zero when . Second, we know that in Bayesian updating, the value of an additional signal tends to diminish as one obtains more signals. Similarly, the marginal gain in information weight should decrease as increases. Our specification for satisfies both of these two properties. Note the definition of coincides with the amount of prior similarity, but the two have very different roles in our models. The amount of prior similarity characterizes project novelty, whereas measures information.
There is also empirical support for the specification of the information weight. In either Model 2 or 3, if we allow the coefficient in front of the prior success residual to vary flexibly with the sum (by using dummies for different levels of this sum), the coefficient estimate will turn out to increase approximately linearly in . This result justifies our specification of . We show this result in Web Appendix C.
In addition to the prior success residual, we also weight the prior success rate and goal overshoot with , because these two metrics are also based on averages across prior projects weighted by . We do not weight the amount of prior similarity or atypicality.
In this section, we first present the estimates of the regression models outlined in the "Models" section. Then, we present the prediction performance of the models in the holdout sample.
Table 5 presents the estimation results for funding success. Table 6 presents the results for funds raised. In each table, we gradually build up the model by adding network metrics, with the last column representing the full model.[11]
Graph
Table 5. Model Estimates for Funding Success.
| (1) | (2) | (3) | (4) |
|---|
| Coef. | SE | Coef. | SE | Coef. | SE | Coef. | SE |
|---|
| Prior success residual | 2.8*** | (.077) | 2.87*** | (.081) | 2.69*** | (.100) | 2.7*** | (.100) |
| Amount of prior similarity | | | .189*** | (.038) | .219*** | (.040) | .207*** | (.040) |
| Amount of prior similarity2 | | | −.0199*** | (.004) | −.0214*** | (.004) | −.0203*** | (.004) |
| Goal overshoot | | | | | .031*** | (.008) | .0309*** | (.008) |
| Goal overshoot2 | | | | | −.0159*** | (.002) | −.0159*** | (.002) |
| Atypicality | | | | | | | 1.12** | (.531) |
| Atypicality2 | | | | | | | −1.18** | (.556) |
| Log funding goal | −.126 | (.139) | −.108 | (.140) | −.66*** | (.157) | −.658*** | (.157) |
| Log funding goal2 | −.0296*** | (.008) | −.0306*** | (.008) | .00667 | (.010) | .00656 | (.010) |
| Log number of images | .375*** | (.024) | .37*** | (.024) | .357*** | (.025) | .357*** | (.025) |
| Featuring a video | 1.2*** | (.057) | 1.2*** | (.057) | 1.18*** | (.057) | 1.18*** | (.057) |
| Log description length | 1.47*** | (.272) | 1.31*** | (.274) | 1.26*** | (.281) | 1.28*** | (.282) |
| Log description length2 | −.0783*** | (.023) | −.0652*** | (.024) | −.0663*** | (.024) | −.0675*** | (.024) |
| Creator with past projects | −.881*** | (.085) | −.877*** | (.085) | −.88*** | (.085) | −.879*** | (.085) |
| Creator past success rate | 2.38*** | (.117) | 2.37*** | (.117) | 2.36*** | (.117) | 2.36*** | (.117) |
| Trend and quarter dummies | Yes | Yes | Yes | Yes |
| Category dummies | Yes | Yes | Yes | Yes |
| R-squared | .335 | .336 | .339 | .339 |
| N | 15,740 | 15,740 | 15,740 | 15,740 |
- 3 *p < .1.
- 4 **p < .05.
- 5 ***p < .01.
- 6 Notes: The dependent variable is whether the project was successfully funded. See the subsections "Network-Based Metrics" and "Control Variables" for detailed definitions of regressors. The prior success residual, goal overshoot, and goal overshoot squared are all information-weighted. Pseudo R-squared is used. The cross-validated parameters for network construction are , , (see Equation 1).
Graph
Table 6. Model Estimates for Funds Raised.
| (1) | (2) | (3) | (4) |
|---|
| Coef. | SE | Coef. | SE | Coef. | SE | Coef. | SE |
|---|
| Prior success residual | 3.31*** | (.060) | 3.44*** | (.062) | 3.42*** | (.083) | 3.43*** | (.083) |
| Amount of prior similarity | | | .29*** | (.035) | .295*** | (.038) | .278*** | (.039) |
| Amount of prior similarity2 | | | −.0295*** | (.004) | −.0336*** | (.004) | −.0318*** | (.004) |
| Goal overshoot | | | | | .0793*** | (.006) | .0792*** | (.006) |
| Goal overshoot2 | | | | | −.0198*** | (.001) | −.0199*** | (.001) |
| Atypicality | | | | | | | 1.12** | (.472) |
| Atypicality2 | | | | | | | −1.02** | (.494) |
| Log funding goal | .925*** | (.116) | .941*** | (.116) | −.103 | (.131) | −.102 | (.131) |
| Log funding goal2 | −.053*** | (.007) | −.0534*** | (.007) | .00674 | (.008) | .00677 | (.008) |
| Log number of images | .65*** | (.024) | .638*** | (.024) | .62*** | (.024) | .62*** | (.023) |
| Featuring a video | 1.55*** | (.052) | 1.54*** | (.052) | 1.5*** | (.052) | 1.5*** | (.052) |
| Log description length | 2.23*** | (.232) | 1.94*** | (.235) | 1.87*** | (.238) | 1.89*** | (.238) |
| Log description length2 | −.131*** | (.020) | −.109*** | (.020) | −.106*** | (.020) | −.107*** | (.020) |
| Creator with past projects | −.875*** | (.080) | −.851*** | (.080) | −.864*** | (.081) | −.863*** | (.081) |
| Creator past success rate | 2.1*** | (.092) | 2.06*** | (.093) | 2.08*** | (.092) | 2.08*** | (.092) |
| Trend and quarter dummies | Yes | Yes | Yes | Yes |
| Category dummies | Yes | Yes | Yes | Yes |
| R-squared | .501 | .503 | .512 | .512 |
| N | 15,740 | 15,740 | 15,740 | 15,740 |
- 7 *p < .1.
- 8 **p < .05.
- 9 ***p < .01.
- 10 Notes: Dependent variable is the log of the funds raised. The notes for Table 5 apply.
We first examine the coefficient estimates for the control variables. Consistent across the columns in both tables, accompanying project descriptions with images and videos has positive impacts on funding outcomes.[12] Providing a longer description initially has a positive effect on funding outcomes but starts to impose a negative effect when it becomes too lengthy. The past success of the project creator is indicative of their future success. A creator with past projects is less likely to succeed than a new creator if these past projects all have failed. The effect of the funding goal is more complex, which we discuss subsequently (together with the effect of goal overshoot).
Before examining the effects of each of the network metrics, we discuss the parameters used to assign the link strength (see Equation 1). The cross-validated values for these parameters are , , and . Parameter is an annual discounting factor, so decreases by a factor of.9309 if the time gap between project i and j increases by a year. Without considering this time factor (i.e., setting ), about 20% of the project pairs have larger than.01, 8% larger than.1, and 3% larger than.5. In other words, the values for the vast majority of links are small. As a result, when predicting the funding performance of a focal project, our models rely mainly on those prior projects that are most similar to the focal project.
All columns of Tables 5 and 6 include the prior success residual. By definition, this metric captures the average level of success of the projects before the focal project, giving more weight to the projects similar to and more recent relative to the focal project (the annual discounting factor is ). We see that the prior success residual has a significantly positive effect on both the funding success (coef. = 2.70, p < .01) and funds raised (coef. = 3.43, p < .01). The estimates are consistent across all columns. So, a higher prior success residual is a significant indicator of better funding outcomes, even after we control for various project and creator features.
As we have discussed, the prior success residual is a refinement of the prior success rate, in the sense that it adjusts the success or failure of each prior project against the control characteristics of that project. For example, it gives the success of a prior project more importance if that project had a more ambitious goal. For comparison, we estimate the model using the prior success rate (instead of residual). The prediction performance turns out to be somewhat lower, which motivates us to use the prior success residual. Please see Web Appendix A.[13]
Columns 2–4 of Tables 5 and 6 include the amount of prior similarity. Given the way we have defined this metric, a zero value for the amount of prior similarity means that the focal project does not share any similarity with any of the projects prior to it (i.e., a completely original idea). A large positive value means the focal project has a great amount of similarities with past projects.
Our prior belief is that the effect of the amount of prior similarity is likely nonlinear. Therefore, we include a quadratic term. Interestingly, we see an inverted U-shaped effect on funding outcomes. In Table 5, the coefficient for the amount of prior similarity is positive (coef. = .207, p < .01), but the coefficient for its squared term is negative (coef. = −.0203, p < .01). The same shape holds in Table 6, with the linear coefficient being positive (coef. = .278, p < .01) and the quadratic coefficient being negative (coef. = −.0318, p < .01). In either table, the estimated quadratic relation reaches its peak around the median of the amount of prior similarity.[14]
This result suggests that a project benefits most from a balanced level of novelty. Featuring ideas that are either too novel or too formulaic hurts a project in raising funds. In the literature on new product development, some studies suggest that novelty has a positive effect on success (e.g., [10]; [30]), some suggest a negative effect (e.g., [18]), and others suggest less clear relations (e.g., [19]). We find a clear inverted U-shaped effect of novelty on funding performance in crowdfunding. This insight is useful for creators when selecting or fine-tuning their crowdfunding ideas, and we take it into account when devising recommendation tools for creators.
Columns 3 and 4 of Tables 5 and 6 include the funding goal overshoot. Given the way we have defined it, this network metric compares the focal project's log funding goal against the average log funds raised by the prior projects, especially those projects similar to the focal project. In other words, this metric captures the focal project's goal relative to prior similar projects. In contrast, the log funding goal included in the control variables captures the absolute level of the funding goal.
We include linear as well as quadratic terms for the goal overshoot (each term is information-weighted). In Table 5, we see that the coefficient for the linear term is positive (coef. = .0309, p < .01), while the coefficient for the squared term is negative (coef. = −.0159, p < .01). We plot this estimated quadratic curve in Figure 2, Panel A. We see an inverted U-shape. Note that this curve does not capture the complete effect of funding goal on success. When a creator increases the goal, it affects two variables in our model: the goal overshoot and the log funding goal. In Panel B of Figure 2, we plot the quadratic effect estimated for the log funding goal. We see the curve is always downward sloping. Taking the two curves together, we see that an increase in funding goal has a relatively small impact on the success probability when the goal overshoot is negative or around zero, but the impact becomes negative and exacerbates quickly when the goal overshoot becomes positive and large.
Graph: Figure 2. Effect of funding goal (overshoot) on success.
In Figure 3, we plot the same curves but for Table 6, which predicts funds raised instead of funding success. Panel A shows an inverted U-shape, but Panel B is almost flat. Taking the two effects together, we see that the funding goal has an inverted U-shaped effect on the expected funds to be raised, with the peak achieved when there is a positive but close-to-zero level of goal overshoot.[15]
Graph: Figure 3. Effect of funding goal (overshoot) on funds raised.
These results are interesting and not straightforward; yet they are also intuitive. When the goal is set too high compared with prior similar projects, investors find the creator requesting an unnecessarily large amount of funds. As a result, investors may question the creator's intention or the profitability of the project. When the goal is set too low compared with prior similar projects, investors may start to question whether the project is viable with the asked amount of funds. In this sense, the zero overshoot point sets up a benchmark for how much funds investors typically consider reasonable for a new project.
One interesting question is why creators set suboptimal goals. In our main sample, more projects have goal overshoots than undershoots. More interestingly, we find that the extent of goal overshoot tends to decrease with creator experience, measured by the number of creator's past Kickstarter projects. This relation suggests that, on average, experience corrects creators' overconfidence and allows them to set more optimal goals. On Kickstarter, raised funds cannot be claimed by the creator unless they surpass the goal. Thus, setting a higher goal increases the risk of a funding failure. However, as our results show, lowering the goal too much decreases the amount of raised funds. Therefore, setting the optimal goal means a careful trade-off between increasing the chance of success and raising more funds. We take this trade-off into account when designing a recommendation tool for creators to adjust their funding goals.
The last columns of Tables 5 and 6 include the measure of atypicality. This network metric relies on the observation that a new idea is typically grounded in a stream of highly related prior ideas. It then measures the extent to which the new idea also imitates prior ideas that are outside this stream. Recall that our measure of atypicality ranges between 0 and 1.
In Table 5, the linear coefficient for atypicality is positive (coef. = 1.12, p = .03), and the quadratic coefficient is negative (coef. = −1.18, p = .03). Table 6 shows a similar result, with the linear coefficient being positive (coef. = 1.12, p = .02) and the quadratic coefficient being negative (coef. = −1.02, p = .04). These estimates indicate that the effect of atypicality has an inverted U-shape. That is, neither too little nor too much atypicality benefits funding performance. Thus, the creator should keep a balance between conventional and atypical combinations of prior ideas.[16]
Among our network metrics, atypicality actually brings the smallest improvement for predicting success in our holdout sample. However, this small improvement is likely due to the fact that the vast majority of the projects in data have near-zero levels of atypicality. Thus, the lack of predictive power does not necessarily imply a lack of effect. In fact, most projects can benefit from an increase in atypical imitation. With everything else equal, if an average project changes its atypicality from zero to the optimal level, the probability of success will increase by almost 20% of the original level by our estimates.
Predicting success is a central research topic in the context of crowdfunding. We test the performance of our model in predicting funding success. Table 7 reports the prediction performance measured by accuracy, F1-score, and ROC-area under the curve (AUC), on the projects in the last quarter of 2017. The models are estimated using the data before this quarter.
Graph
Table 7. Prediction Results for Holdout Sample.
| Accuracy | F1-Score | ROC-AUC |
|---|
| Network metrics, creator controls, and project controls | .799 | .774 | .874 |
| Creator and project controls | .749 | .714 | .830 |
| Project controls | .725 | .688 | .798 |
| Project controls—only time and category | .582 | .442 | .552 |
11 Notes: All numbers are computed on the holdout sample from October 1 to December 31, 2017. The models are trained using data before October 1, 2017. Both the accuracy and F1-score use.5 as the probability threshold when predicting success.
The table compares four different model specifications. The top row shows our full model. The second row drops all network metrics. The third row further drops creator-related controls. The last row presents the leanest model that keeps only a time trend, quarter dummies, and category dummies.
Across all three prediction measures (accuracy, F1-score, and ROC-AUC), the full model performs substantially better than the models without network metrics, demonstrating the value of the information added by the similarity network. These improvements compare very well with [35] and [34]. In fact, if we use the leanest model in the last row of Table 7 as a baseline, then adding the network metrics improves the out-of-sample prediction more than adding all other project and creator controls (i.e., creator past success rate, creator with past projects, funding goal, text description length, number of images, and video dummy).
This section derives actionable implications for project creators and platforms, taking advantage of the unique information in our similarity network. We focus on two major decisions of project creators: ( 1) how to set the appropriate funding goal and ( 2) how to improve the content of the project. We devise the two recommendation tools accordingly. The recommendations can be offered by the platform to a creator after the creator submits a new project to the platform (but before the project is posted on the site).
The goal recommendation suggests an adjustment (e.g., a 5% increase) to the funding goal set by the creator. The adjustment aims to let the creator collect more funds, and it is based on our models presented in the "Results" section. Thus, a unique feature of our recommendation is that we account for the effect of goal overshoot, which benchmarks the new project's goal against the funds historically raised by similar projects. Consequently, our recommendation is customized for individual projects.
The content recommendation suggests a prior project for the creator to consider imitating (by revising the submitted project). The objective is to improve the probability of success. The recommendation is based on our models presented in the "Results" section, so again, it is customized for individual projects.
Recall that Kickstarter applies an all-or-nothing policy, whereby project creators collect the raised funds only if the funding is successful (i.e., the raised funds surpass the goal). As a result, setting the correct goal is crucial. Setting the goal too high may cause the creator to end up collecting nothing. Setting the goal too low may lead to insufficient funds collected for the project.
Currently, Kickstarter does not advise creators to set goals in a way to maximize their funds collection. Figure 4 shows the current goal-setting page when one submits a project on Kickstarter. The creator first inputs an estimated project budget. Kickstarter then suggests a funding goal, but the suggestion is aimed to simply satisfy the creator-inputted budget after mechanically deducting taxes and fees.
Graph: Figure 4. Goal-setting page on Kickstarter.
We use our estimated models (the last columns of Table 5 and 6) to suggest a goal adjustment to improve the project's funding outcomes. For each project in our holdout sample, we consider adjustments of the goal ranging from −10% to +10%. There are two reasons why we restrict attention to such conservative adjustments. First, doing so ensures that, in the event of successful funding, the creators will receive funds in the ballpark of what they initially decided was adequate for the project. Second, it alleviates the concern of funding competition among projects. At an aggregate level, there is likely a limit on the total funds available from investors for crowdfunding. A large goal adjustment may benefit an individual project, but if many projects make large and upward goal adjustments, these projects likely will have to compete heavily for the aggregate funds available. As a result, analyzing fundraising under large goal adjustments requires accounting for competitive effects between projects, which is out of the scope of our article.
Changing the goal is likely to simultaneously affect the success probability and expected funds to be raised. Improving one of these two aspects may hurt the other aspect. To balance them, we select the goal adjustment for recommendation in the following way. We check two possibilities: ( 1) increasing the success probability by at least 5% with the constraint of not hurting the expected funds by more than 1% and ( 2) increasing the expected funds by at least 5% with the constraint of not hurting the success probability by more than 1%. We first try possibility ( 1) then ( 2). We do not make a recommendation if neither possibility exists.
Table 8 presents the results. Out of the 1,903 projects in the holdout sample, we end up recommending goal increases for 155 projects (column 1) and goal decreases for 697 projects (column 2). For the remaining 1,051 projects, we do not recommend a goal adjustment. What drives the direction of our recommendation is the network metric of goal overshoot. To illustrate this driving force, we can compare a project's original goal with the average funds raised by its prior similar projects. Table 8 shows how this comparison turns out, on average. We see that the projects in column 1, on average, had a goal that was too conservative ($2,306 vs. $10,025), whereas the projects in column 2, on average, had a goal that was too ambitious ($28,293 vs. $4,588).
Graph
Table 8. Recommending Funding Goal Adjustments.
| Recommend a Goal Increase | Recommend a Goal Decrease |
|---|
| Funding goal (original) | $2,306 | $28,293 |
| Average funds raised by prior similar projects | $10,025 | $4,588 |
| Success probability (original) | .825 | .118 |
| Change in success probability | −.4% | +7.9% |
| Change in funds raised | +6.6% | +2.1% |
| Number of projects | 155 | 697 |
12 Notes: Goal adjustments are constrained to be within ±10%. The numbers in each column are averaged across the projects belonging to that column. For each project i, the "average funds raised by prior similar projects" weigh each prior project j by .
On average, the goal-increasing adjustments bring a 6.6% increase to the expected funds to be raised at the expense of only a.4% loss in success probability. The goal-decreasing adjustments bring a 7.9% increase in success probability as well as a 2.1% increase in the expected funds to be raised—overly high goals hurt both the success probability and expected funds to be raised. These improvements are substantial, especially considering that ( 1) adjusting the funding goal is costless and ( 2) the adjustments are not large—within ±10% of the original goals.
Given a new project i, we aim to make a suggestion to help the creator to improve the project's content. The way we implement this suggestion is to recommend a prior project j for the creator to imitate. This imitation amounts to an increase in the similarity between i and j, which will then change our predictions for the funding outcomes for i. We want to recommend a prior project j so that the expected funding outcomes of project i can be improved.
The key question here is exactly which prior project to recommend. We choose the project such that an increase in the similarity between it and project i will most substantially improve project i's success probability (it gives very similar results if we instead focus on improving the funds raised). Specifically, for each prior project j, we calculate the change in project i's success probability if the WMD between j and i is reduced by 10%. This calculation is made using our model for funding success (last column of Table 5). Note this reduction in WMD generally changes all the network metrics for project i. We recommend a prior project if it leads to the largest increase in success probability among all prior projects and if the relative increase is larger than 5%.
Take Figure 1 as an illustration. Suppose that project 6 is the new project. We first try recommending project 1 and calculate how the model-predicted success probability for project 6 will change if the WMD between project 1 and 6 is reduced by 10% (which will result in a stronger link between the two projects). We try each of projects 1–5 in turn and select the one that results in the highest increase in success probability for project 6. Suppose we select and recommend project 5. Then project 6's creator may bring project 6 closer to project 5 by, for example, enriching their motorcycle journey with more interactions with the cultures and people along the journey (for details of each project, see Table 3).
Table 9 presents the results. We make recommendations for 887 projects out of the 1,903 projects in the holdout sample. For the remaining projects, we do not make recommendations because the improvement in success probability would be smaller than the 5% threshold. Column 1 displays the statistics of the projects for which we make recommendations, whereas column 2 displays the statistics of the projects for which we do not make recommendations. There are several observations about the original attributes of the projects for which we make recommendations. First, the funds raised by these projects are noticeably lower compared with the other projects ($2,167 vs. $8,309). In addition, they have a substantially lower success probability (.209 vs..633). They also tend to have a far smaller amount of prior similarity (3.3 vs. 6.9).
Graph
Table 9. Recommending a Prior Project for Imitation.
| With Recommendation | No Recommendation |
|---|
| Funds raised (original) | $2,167 | $8,309 |
| Success probability (original) | .209 | .633 |
| Amount of prior similarity (original) | 3.3 | 6.9 |
| Improvement in success probability | .055 | — |
| Improvement in funds raised | $240 | — |
| Number of projects | 887 | 1,016 |
13 Notes: The numbers in each column are averaged across the projects belonging to that column.
These stark differences give the following intuition. Recall that the success probability has an inverted U-shaped relation with the novelty of the project. The projects with a very small amount of prior similarity are too novel for investors to understand or appreciate. So, by asking the creators to bring these highly novel projects slightly closer to some successful prior projects, we can significantly increase their appeal to investors and thus their probability of success (from.209 to.264).
Another reason that our recommendation does not target projects with a large amount of prior similarity is that for these projects, imitating a single prior project can cause only a small marginal effect on the success probability. This observation suggests that we recommend multiple prior projects. However, such recommendations are more complex to design and demand a larger effort from creators. In addition, when such recommendations are adopted by creators, there will be larger changes in the similarity networks. Thus, aggregate effects at the platform level will likely be substantial. These effects are beyond the scope of our models and typically require a structural model to capture (which is similar to the reason why we have restricted ourselves to small adjustments in the goal recommendation).
How can we improve crowdfunding project proposals to design winning projects? Few studies have empirically addressed this important and exciting question. This study aims to help answer this question using the concept of combinatorial creativity. First, we extend and apply the concept of combinatorial creativity in crowdfunding, taking advantage of modern machine learning methods to measure a complete similarity pattern between crowdfunding projects on Kickstarter. Second, using this similarity pattern, we significantly improve the out-of-sample prediction of funding success. Third, we derive novel insights on how a project's funding performance is affected by its similarity pattern with prior projects. Our results allow us to provide concrete recommendations for creators to improve their project proposals.
Several results emerge from our analysis. First, the similarity network provides an important information source for predicting success. In particular, the historical success of similar projects is a significant predictor for new project success, after we control various project and creator characteristics. Second, there is an inverted U-shaped relationship between novelty and funding performance—being either too novel or too imitative hurts the project. Third, when setting the funding goal, creators should benchmark the goal against the funds raised historically by similar projects. A goal that is either too high or too low compared with this benchmark lowers the funding performance of the project. Fourth, there is an inverted U-shaped relation between the extent of atypical imitation and funding performance. Either too little or too much atypicality harms the funding performance of a project.
We take advantage of the similarity network to devise recommendation tools to assist project creators in two critical decisions: ( 1) setting the funding goal and ( 2) developing the project content. Currently, Kickstarter reviews new project proposals (and rejects a portion of them) but does not offer customized help to creators to improve these proposals. Our recommendations can be offered by the platform at this review stage to assist creators, especially less experienced ones, in developing successful projects. Our first recommendation suggests a small adjustment to the funding goal. Our second recommendation suggests a prior project for the creator to imitate. Both suggestions are customized for individual projects, taking into account the effects of prior success, novelty, goal overshoot, and atypicality. For individual projects in our holdout sample, we show that these recommendations can lead to meaningful increases in funding performance. With the use of the external Word2vec by Google, the recommendation tools are easy to implement and update over time on the platform.
Although this article focuses on crowdfunding as the empirical context, the developed approach can be applied to other text-based creative settings, such as mobile apps, venture capital, news stories, and blog websites. The first task in these settings is to obtain a repository of text descriptions of the ideas, whether they are apps, start-ups, or articles. Once the text descriptions are obtained, one can adopt the WMD to automate the measurement of similarity between ideas. Next, one can construct a similarity network between these ideas, accounting for both the content similarity and time proximity, as we have done in this article. Then, the network-based metrics we developed, including prior success residual, novelty, and atypicality, can be applied to improve predictions of idea success. The goal overshoot may not be directly applicable in contexts outside crowdfunding, but variants may be used. For example, in the case of mobile apps, one could compute a "price overshoot" as the difference between a focal app's price and the average price of prior similar apps.
The aforementioned applications of similarity networks allow us to explore interesting questions specific to the domain of application. For example, for user-generated content such as blog articles, does novel or familiar content appeal to consumers, or is there an optimal level of balance between novelty and familiarity? How should writers determine the appropriate length of an article—can prior similar articles provide a useful benchmark? For new app development, can prior prices of similar apps help in setting the price of a new app? How should the developer choose the optimal degree of imitation? Should the app focus on developing a core function or integrating multiple functions delivered by different past apps? In the age of the long tail ([ 2]), it is more exciting than ever to study the similarity pattern among the great number of products as well as the relationship between this similarity pattern and product success.
Footnotes 1 Yanhao "Max" Wei and Jihoon Hong contributed equally to this work.
2 Debanjan Mitra
3 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
4 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the Lloyd Greif Center for Entrepreneurial Studies at the University of Southern California.
5 Jihoon Hong https://orcid.org/0000-0001-9499-1957
6 Online supplement:https://doi.org/10.1177/00222429211005481
7 Kickstarter goes through a review process for submitted projects. The rejection rate for the submitted projects is around 20% ([5]), and the average success rate of the accepted projects is about 37% (Kickstarter [20]). 30% 80% × 37%.
8 After removing stop words and rare words, we drop 3.8% of the remaining words because they are not included in Google's Word2vec vocabulary. This percentage is 6.2% for the Film & Video category and 4.6% for Music.
9 In the data, is typically around either 30 or 60 days after . The project creator sets the fundraising duration . The maximum allowed duration is 60 days.
A project's funding success can be affected by how it rewards investors. However, neither [27] nor [32] control for rewards. We have tried including the number of reward options as a control; our main findings hold.
To check for multicollinearity, we compute the variance inflation factors. After removing the squared terms and centering all nondummy regressors, the maximum variance inflation factor is 5.07.
Web Appendix B explores how the effect of featuring a video varies across project categories.
We can also define prior funds residual. Specifically, we compute the residual using a regression of log funds raised (instead of a logit regression of funding success). We examine the model based on the prior funds residual in Web Appendix A.
The inverted U-shaped result holds if we replace the quadratic specification with dummies for different levels of the amount of prior similarity. The peak is still around the median amount of prior similarity.
The results in Figures 2 and 3 hold qualitatively if we add cubic terms for both goal overshoot and log funding goal.
The same qualitative conclusion holds when we replace the quadratic specification with dummies for different brackets of atypicality. However, the dummy regression suggests a skewed inverted U-shape, with the optimal atypicality in.1 ∼.3. See Web Appendix B.
References Agrawal Ajay , Catalini Christian , Goldfarb Avi. (2015), " Crowdfunding: Geography, Social Networks, and the Timing of Investment Decisions ," Journal of Economics & Management Strategy , 24 (2), 253 – 74.
Anderson Chris. (2006), The Long Tail: Why the Future of Business Is Selling Less of More. New York : Hyperion Books.
Bapna Sofia. (2017), " Complementarity of Signals in Early-Stage Equity Investment Decisions: Evidence from a Randomized Field Experiment ," Management Science , 65 (2), 933 – 52.
Barnett Chance. (2015), " Trends Show Crowdfunding to Surpass VC in 2016 ," Forbes (June 9), https://www.forbes.com/sites/chancebarnett/2015/06/09/trends-show-crowdfunding-to-surpass-vc-in-2016.
Benovic Carol. (2016), " The Project Review Process: From Submitting to Getting Approved ," Kickstarter (July 14) , https://www.kickstarter.com/blog/everything-you-need-to-know-about-the-project-review-process.
Berger Jonah , Packard Grant. (2018), " Are Atypical Things More Popular? " Psychological Science , 29 (7), 1178 – 84.
Blei David M. , Ng Andrew Y. , Jordan Michael I.. (2003), " Latent Dirichlet Allocation ," Journal of Machine Learning Research , 3 (January), 993 – 1022.
Burtch Gordon , Ghose Anindya , Wattal Sunil. (2015), " The Hidden Cost of Accommodating Crowdfunder Privacy Preferences: A Randomized Field Experiment ," Management Science , 61 (5), 949 – 62.
Butticè Vincenzo , Colombo Massimo G. , Wright Mike. (2017), " Serial Crowdfunding, Social Capital, and Project Success ," Entrepreneurship Theory and Practice , 41 (2), 183 – 207.
Dahl Darren W. , Moreau Page. (2002), " The Influence and Value of Analogical Thinking During New Product Ideation ," Journal of Marketing Research , 39 (1), 47 – 60.
Dai Hengchen , Zhang Dennis J.. (2019), " Prosocial Goal Pursuit in Crowdfunding: Evidence from Kickstarter ," Journal of Marketing Research , 56 (3), 498 – 517.
Deerwester Scott , Dumais Susan T. , Furnas George W. , Landauer Thomas K. , Harshman Richard. (1990), " Indexing by Latent Semantic Analysis ," Journal of the American Society for Information Science , 41 (6), 391 – 407.
Ebbes Peter , Papies Dominik , van Heerde Harald J.. (2011), " The Sense and Non-Sense of Holdout Sample Validation in the Presence of Endogeneity ," Marketing Science , 30 (6), 1115 – 22.
Finke Ronald A. , Ward Thomas B. , Smith Steven M.. (1992), Creative Cognition: Theory, Research, and Applications. Cambridge : MIT Press.
Goldenberg Jacob , Mazursky David. (2002), Creativity in Product Innovation. Cambridge, UK : Cambridge University Press.
Goldenberg Jacob , Lehmann Donald R. , Mazursky David. (2001), " The Idea Itself and the Circumstances of Its Emergence as Predictors of New Product Success ," Management Science , 47 (1), 69 – 84.
Goldenberg Jacob , Mazursky David , Solomon Sorin. (1999), " Toward Identifying the Inventive Templates of New Products: A Channeled Ideation Approach ," Journal of Marketing Research , 36 (2), 200 – 210.
Hyytinen Ari , Pajarinen Mika , Rouvinen Petri. (2015), " Does Innovativeness Reduce Startup Survival Rates? " Journal of Business Venturing , 30 (4), 564 – 81.
Im Subin , Workman John P. Jr. (2004), " Market Orientation, Creativity, and New Product Performance in High-Technology Firms ," Journal of Marketing , 68 (2), 114 – 32.
Kickstarter (2019), " Stats ," (accessed October 26, 2019), https://www.kickstarter.com/help/stats.
Kuppuswamy Venkat , Bayus Barry L.. (2017), " Does My Contribution to Your Crowdfunding Project Matter? " Journal of Business Venturing , 32 (1), 72 – 89.
Kusner Matt , Sun Yu , Kolkin Nicholas I. , Weinberger Kilian Q.. (2015), " From Word Embeddings to Document Distances ," in Proceedings of the 32nd International Conference on Machine Learning. New York : Association for Computing Machinery , 957 – 66.
Lin Mingfeng , Prabhala Nagpurnanand R. , Viswanathan Siva. (2013), " Judging Borrowers by the Company They Keep: Friendship Networks and Information Asymmetry in Online Peer-to-Peer Lending ," Management Science , 59 (1), 17 – 35.
Luo Lan , Toubia Olivier. (2015), " Improving Online Idea Generation Platforms and Customizing the Task Structure on The Basis of Consumers' Domain-Specific Knowledge ," Journal of Marketing , 79 (5), 100 – 114.
Mikolov Tomas , Chen Kai , Corrado Greg , Dean Jeffrey. (2013), " Efficient Estimation of Word Representations in Vector Space ," in Proceedings of Workshop at International Conference on Learning Representations , https://arxiv.org/abs/1301.3781.
Mikolov Tomas , Sutskever Ilya , Chen Kai , Corrado Greg , Dean Jeffrey. (2013), " Distributed Representations of Words and Phrases and Their Compositionality ," in Proceedings of Advances in Neural Information Processing Systems , 3111 – 19 , https://arxiv.org/abs/1310.4546.
Mollick Ethan. (2014), " The Dynamics of Crowdfunding: An Exploratory Study ," Journal of Business Venturing , 29 (1), 1 – 16.
Netzer Oded , Lemaire Alain , Herzenstein Michal. (2019), " When Words Sweat: Identifying Signals for Loan Default in the Text of Loan Applications ," Journal of Marketing Research , 56 (6), 960 – 80.
Schumpeter Joseph A. (1934), The Theory of Economic Development: An Inquiry into Profits, Capital, Credit, Interest, and the Business Cycle. Cambridge, MA : Harvard University Press.
Sethi Rajesh , Smith Daniel C. , Whan Park C.. (2001), " Cross-Functional Product Development Teams, Creativity, and the Innovativeness of New Consumer Products ," Journal of Marketing Research , 38 (1), 73 – 85.
Stanko Michael A. , Henard David H.. (2016), " How Crowdfunding Influences Innovation ," MIT Sloan Management Review , 57 (3), 15.
Steigenberger Norbert , Wilhelm Hendrik. (2018), " Extending Signaling Theory to Rhetorical Signals: Evidence from Crowdfunding ," Organization Science , 29 (3), 529 – 46.
Stephen Andrew T. , Zubcsek Peter Pal , Goldenberg Jacob. (2016), " Lower Connectivity Is Better: The Effects of Network Structure on Redundancy of Ideas and Customer Innovativeness in Interdependent Ideation Tasks ," Journal of Marketing Research , 53 (2), 263 – 79.
Tellis Gerard J. , MacInnis Deborah J. , Tirunillai Seshadri , Zhang Yanwei. (2019), " What Drives Virality (Sharing) of Online Digital Content? The Critical Role of Information, Emotion, and Brand Prominence ," Journal of Marketing , 83 (4), 1 – 20.
Toubia Olivier , Netzer Oded. (2017), " Idea Generation, Creativity, and Prototypicality ," Marketing Science , 36 (1), 1 – 20.
Uzzi Brian , Mukherjee Satyam , Stringer Michael , Jones Ben. (2013), " Atypical Combinations and Scientific Impact ," Science , 342 (6157), 468 – 72.
Van de Rijt Arnout , Kang Soong Moon , Restivod Michael , Patile Akshay. (2014), " Field Experiments of Success-Breeds-Success Dynamics ," Proceedings of the National Academy of Sciences , 111 (19), 6934 – 39.
Wei Yanhao Max. (2020), " The Similarity Network of Motion Pictures ," Management Science , 66 (4), 1647 – 71.
Weitzman Martin L. (1998), " Recombinant Growth ," Quarterly Journal of Economics , 113 (2), 331 – 60.
Xu Yan , Ni Jian. (2019), " Asymmetric Information and Entrepreneurial Disincentives in Crowdfunding Markets ," working paper, SSRN (June 22) , https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3480888.
Youn Hyejin , Strumsky Deborah , Bettencourt Luis M.A. , Lobo Jose. (2015), " Invention as a Combinatorial Process: Evidence from US Patents ," Journal of the Royal Society Interface , 12 (106), 20150272.
Younkin Peter , Kuppuswamy Venkat. (2017), " The Colorblind Crowd? Founder Race and Performance in Crowdfunding ," Management Science , 64 (7), 3269 – 87.
Zhang Juanjuan , Liu Peng. (2012), " Rational Herding in Microloan Markets ," Management Science , 58 (5), 892 – 912.
Zvilichovsky David , Danziger Shai , Steinhart Yael. (2018), " Making-the-Product-Happen: A Driver of Crowdfunding Participation ," Journal of Interactive Marketing , 41 , 81 – 93.
~~~~~~~~
By Yanhao "Max" Wei; Jihoon Hong and Gerard J. Tellis
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 85- Managing Members, Donors, and Member-Donors for Effective Nonprofit Fundraising. By: Kim, Sungjin; Gupta, Sachin; Lee, Clarence. Journal of Marketing. May2021, Vol. 85 Issue 3, p220-239. 20p. 3 Charts, 8 Graphs. DOI: 10.1177/0022242921994587.
- Database:
- Business Source Complete
Managing Members, Donors, and Member-Donors for Effective Nonprofit Fundraising
Nonprofit organizations (NPOs) play a central role in many economies in the form of private entities serving a public purpose. Strengthening the fundraising capabilities of NPOs can have a large impact on their survival and effective functioning. NPOs typically elicit financial contributions through multiple forms of giving, such as donation and membership. These options enable individuals to express their altruism by giving in one or multiple forms. The authors develop a utility-based multiple discrete-continuous model that provides insights into potentially large differences in individuals' giving behaviors. Through Bayesian Gaussian processes, the model also incorporates changes in givers' preferences for forms of giving. The authors apply their model to five years of individual giving data. They find that the effects of lifetime, recency, seasonality, and appeals on donation and membership options change nonmonotonically over time and in distinctive ways. The authors demonstrate that the model estimates help predict who will give in more than one form in the future as well as build appeal targeting strategies. The model also shows that fundraising attempts should emphasize participation rather than amount, and that long-lapsed members are still worth pursuing for renewal, whereas long-lapsed donors are less productive for repeat giving.
Keywords: Bayesian estimation; charitable giving; choice models; fundraising capabilities; Gaussian process; nonprofit
In 2019, U.S. households donated $309.7 billion, nearly 2% of their annual income, to nonprofit organizations (NPOs) ([30]). Yet many NPOs are under considerable financial pressure to close the gap between their resources and the social missions they serve. In the midst of the COVID-19 pandemic of 2020, charitable giving in the United States increased during the first half of the year relative to the previous year; however, fears remained that the sluggish economy would depress long-term giving ([60]). A major challenge that many NPOs face is the volatility of individual giving: the literature reports that approximately half of newly acquired donors churn after they give once ([37]; [56]). In response to such instability, NPOs strive to increase repeat giving by individuals as well as to identify and retain givers who are more committed. The purpose of this article is to help NPOs to succeed in their endeavor by developing a model that provides insights into givers' preferences, differences between givers, and the dynamics of giving behavior.
How does our research contribute to better marketing for a better world? In this article, we view nonprofits broadly as belonging to the "third sector" of the political economy, in addition to the for-profit business and government sectors, which are well-defined ([35]). The nonprofit sector includes a broad array of private organizations serving public purposes. While a literature has debated alternative theories about the economic role of NPOs ([43]), there is little doubt that, in practice, NPOs play a substantial role in sectors such as religion, health care, education, arts and culture, and the environment.
Individual charitable giving is a key source of funding for many NPOs, especially for small and midsize organizations in the United States ([50]). Not surprisingly, therefore, the average ratio of fundraising cost to donation is quite high, at about 12% ([ 4]; [42]). To improve the effectiveness and efficiency of fundraising from individuals, our research develops and applies marketing science tools, thereby strengthening the ability of NPOs to fulfill their social missions.
We propose a framework of individual giving behavior that incorporates three important factors that have previously remained understudied in the literature. First, NPOs' revenue depends on both the decision to give (i.e., choice to participate) and how much to give (i.e., the dollar amount).[ 6] Previous studies often do not differentiate between givers' participation and amount decisions ([24]; [25]). Distinguishing between the two decisions enables the NPO to learn whether its marketing efforts should focus on encouraging more participation, larger donation amounts, or both. This allows the NPO to set priorities to maximize return on its efforts.
Second, many NPOs structure fundraising programs for individuals around two major forms of giving: donation and membership. For instance, the list of the 100 largest nonprofits in the United States based on 2019 revenues includes eight organizations that focus on the environment and animals ([34]), which is the domain of operation of the focal NPO in our empirical application. Of those eight, five nonprofits allow individuals to give through both donation and membership programs. Donors support an NPO's mission financially without any expectation of reciprocal tangible benefits. Instead, donors receive a "warm glow" that makes their own financial contributions to the NPO more valuable to them than the contributions of others ([ 2]). Furthermore, the U.S. government incentivizes donations by providing tax deductions ([ 9]). Membership, in contrast, confers a defined set of membership benefits that expand with the level of contribution in addition to the warm glow and tax deductions. Membership programs can thus be viewed as "commercial" activities in the sense that NPOs exchange mission-related goods or services for membership fees ([64]). For instance, members of an art museum may receive benefits such as invitations to exclusive tours.
Multiple forms of giving allow individuals to choose to be pure donors, pure members, or member-donors. Pure donors and pure members give exclusively through donation programs and membership programs, respectively, whereas member-donors support an NPO through both donation and membership programs. These choices are expressions of individual motivations. The literature in charitable giving has categorized these motivations into two kinds: intrinsic and extrinsic (e.g., [12]; [ 6]). Intrinsic motivation is the value of giving per se, represented by preference for helping others, while extrinsic motivation is the desire to receive tangible benefits associated with giving. Donors' choice of donation as the form of giving reveals their greater intrinsic motivation since there is no tangible benefit received in return. By contrast, members' choice of membership reveals greater extrinsic motivation (than donors) because members choose to receive membership benefits. Membership is also intended to induce a sense of belonging and identity that increases the likelihood of repeat giving, and thereby reduces the volatility of NPOs' cash inflows ([16]; [15]). Although multiple forms of giving allow greater choices for individuals, they involve additional administrative costs for the NPO. Therefore, from the NPOs' perspective, managing two streams of giving makes sense when the incremental revenue from multiple forms of giving is expected to outweigh the incremental costs.
Third, recognizing that repeat giving is highly desired by NPOs, we focus on the dynamics of individual giving behavior. By contrast, extant research has primarily focused on single observations of individual giving (i.e., cross-sectional data). We expect that previous giving decisions are likely to influence subsequent decisions to give ([41]; [65]). This is evident with membership that remains valid for a specific duration; an individual having paid an annual membership fee may not consider giving until the membership is about to expire, 12 months later. Donations, unlike membership, have no such associated term, making it less obvious how the probability of repeat giving should change over time. We conjecture that an individual may not actively consider giving for some time after donating due to a "licensing effect" ([36]). This effect postulates that the choice probability of a more hedonic option increases after a prior, virtuous, license-generating decision. [36] found that subjects in the license condition exhibited less altruism and reduced intention to donate. In our context, this implies that recent donors may feel justified not giving for some time, after which the desire to donate may reemerge and grow. Conversely, if an individual has not donated or renewed membership for a long time, this might be a sign of churn. Therefore, it is important for NPOs to understand how to think about and act on individuals who vary in the time since their last giving occasion.
In addition to repeat giving, member-donors' decision to give in the two forms typically evolves over time with some individuals who start out as pure members or pure donors subsequently adding a second form of giving. We conceptualize this to be the result of two phenomena. First, the individual's intrinsic desire to give may change as the relationship with the NPO deepens. Second, the individual experiences diminishing marginal utility ([ 3]; [21]; [48]) from giving larger amounts in a single form. For instance, some members may not value the additional benefits conferred by a higher tier of membership and instead add a donation to their giving portfolio. Similarly, some donors seek a sense of belonging and tangible benefits in addition to the warm glow from larger donations and thus add membership to their giving portfolio. Importantly, the availability of multiple forms of giving allows individuals to express their idiosyncratic giving preferences to start their relationship with the NPO and to modify it over time.
To model giving behavior, we propose a utility-based multiple discrete-continuous model with Gaussian process (GP) priors. We apply the model to the five-year giving behavior of a cohort of 2,171 individuals who began their relationship with an NPO in 2011 to investigate three research questions: ( 1) What factors influence the choices of multiple forms of giving and amount, and how do their effects change over time?, ( 2) How can our framework be used to increase participation rates and contribution amounts?, and ( 3) Can our model identify contributors who are more committed to the NPO?
Our empirical analyses yield several insights that we summarize briefly here. First, the propensity to donate increases over the lifetime of the relationship, whereas we do not see such a positive trend for membership. Instead, propensity to participate in membership remains cyclical, peaking every 12 months, even as the lifetime grows. This finding is consistent with membership being more reciprocal in nature, in the sense of being more extrinsically motivated.
Second, we find evidence that the propensity to give again remains low for seven months following a donation, providing support for a licensing effect. The propensity peaks at 11 and 24 months after giving, making those opportune times to ask for a new donation. Thereafter, however, there is no remaining positive predisposition, implying that these donors are likely to be similar to new prospects. By contrast, lapsed members remain positively predisposed for a duration longer than 24 months and may be worth continuing to pursue. Therefore, our research supports previous findings that providing tangible benefits increases giving, possibly by encouraging feelings of reciprocity or gratitude.
Third, the effects of donation appeals for our focal NPO were inconsistent, indicating room for improvement through targeting. However, membership appeals were effective at encouraging renewals, perhaps because they were naturally targeted to reach individuals just before their membership was about to expire.
Fourth, not surprisingly, we find substantial heterogeneity in the giving preferences of individuals; the heterogeneity is partially explained by observable characteristics such as demographics, but a significant part is not.
Finally, we find that our models can help the NPO predictively identify member-donors, who are more committed contributors in terms of both amount and frequency of giving, and also develop targeting strategies for more effective appeal campaigns. In the "Discussion and Conclusions" section, we use these findings to present a list of recommendations for NPOs.
The literature on charitable giving is vast. We provide a brief review, with the goal of highlighting differences between our proposed framework and the extant literature. With regard to the separation of givers' decisions to participate versus the amount given, as noted, this distinction has typically not been made in the literature. [25], p. 142) emphasize that "previous studies consider only one decision, leaving readers to assume, implicitly, that results apply to both (choice and amount) or to consider the two dimensions interchangeably." However, [24] demonstrate that different mechanisms govern the two decisions. For instance, they find that mood management (i.e., how one feels about oneself) primarily governs the participation decision, while empathetic feelings (i.e., how one feels about the victim) predicate the amount decision. By separating the two decisions, we are able to not only obtain insights into givers' behavior but also derive implications for more effective fundraising.
Previous research has recognized that benefits provided by charitable organizations can encourage feelings of reciprocity or gratitude that may lead to increased giving ([26]; [10]). However, benefits can have the undesirable effect of "crowding out" altruistic motives by diluting intrinsic motivation or the signaling value of prosocial behavior ([12]; [ 6]). Moreover, external benefits may shift one's mindset from an altruistic to a more monetary perspective ([31]) and potentially decrease the amount of charitable giving ([53]). Previous research thus makes ambiguous predictions regarding the effects of tangible benefits on giving, some studies reporting positive effects, others negative effects.
Despite the prevalence of multiple forms of giving in practice, and the recognition that benefits can have an effect on subsequent giving, previous marketing studies typically treat giving via membership and donation as a composite ([52]; [63]; [37]; [39]). Such aggregation across forms of giving masks potentially large differences in individuals' motivations and giving behaviors. By contrast, in our framework we recognize the different options facing givers. Consequently, we are able to offer prescriptions about how an NPO should manage available forms of giving to strengthen revenues. As we have noted, while multiple forms of giving may increase the top line, they also typically involve separate administrative structures and thus additional costs for the NPO. Finally, by studying individuals' choices in the presence of giving options with and without external benefits—namely, membership and donation—we are able to shed some light on the opposing viewpoints in the literature about the effects of external benefits.
Although there is a large body of research on charitable giving and its drivers, much of it is based on cross-sectional analysis and "no research has investigated the longitudinal dynamics of individual donation decisions" ([41], p. 1). More recently, [65], p. 11) note that there are relatively few studies on repeated prosocial actions over time and propose that future research should examine the factors that lead to increases or decreases in repeated donations. We refer the reader to excellent review papers available in the literature ([ 5]; [56]; [65]) and provide selected examples that use cross-sectional designs here. [31] find that subjects who were given performance incentives performed more poorly than those who were offered no compensation in a door-to-door fundraising context. [ 6] confirm that the effects of extrinsic incentives on prosocial behavior crucially depend on visibility; monetary rewards facilitate private, rather than public, prosocial activity. That is, people want to be seen by others as generous; therefore, receiving visible extrinsic incentives dilutes the signals of their prosocial acts. [21] find evidence that both warm glow and social pressure affect charitable giving. They argue that high social pressure solicitation leads to decreased welfare of the givers.
Related to the management of charitable donations is the literature on crowdfunding. This literature has identified three forms—investment-based, reward-based, and donation-based crowdfunding—the latter two bearing similarities to fundraising by NPOs ([11]). [40] provide examples of reward-based crowdfunding, in which individuals receive tangible but not monetary benefits from their donations to the project. This type is most similar to the benefits one would receive in the membership model of NPOs. Furthermore, [28] offer examples showing that the reward itself is important for the participant to fund.
As we have discussed, our research builds on the previous literature to deepen understanding of individual giving behavior by focusing on three important but understudied aspects: ( 1) separating choice to participate from amount of giving, ( 2) giving in multiple forms, and ( 3) the dynamics of individual giving. The remainder of the article is organized as follows. In the next section, we describe our data. We then discuss the modeling framework and estimation approach and report our results. A discussion of managerial implications follows, and we conclude with a summary of findings and suggestions for future research.
We collaborated with a large nonprofit scientific research center (SRC; the organization has asked to remain anonymous) that studies an animal species. The mission of the SRC is to promote environmental and natural causes, which it does by conducting scientific research and by a large public outreach initiative. An important source of funding for the SRC is financial gifts from over 100,000 individuals via donation and membership programs.
We consider the cohort of all 2,542 individuals in the United States who made their first financial gift to the SRC during 2011, and we follow them until February 2016. The giving data of this cohort was obtained in late 2016. Individuals who gave more than $10,000 in any year are excluded from the cohort we study because they are managed by a designated fundraiser via a separate and different process. To be able to estimate the dynamics of giving, we retain the subset of 2,171 individuals who gave at least twice during our time window.[ 7] For each individual in the cohort, we have monthly data on giving: the amount given, when given, and to which giving options. We also obtained data on the following demographic: age, gender, physical distance of residence from the SRC, and estimated annual income (all assumed to be time invariant).
Our data also include marketing activities of the SRC, namely, the number and kinds of appeals sent to each individual. The SRC conducts two major appeal campaigns for donations: at the end of the calendar year, for which appeals are sent in November, and in the spring in the United States; it also conducts several minor campaigns throughout the year. For membership, the SRC sends annual renewal appeals to current members just before their membership term is about to expire. The organization did not conduct any individually targeted marketing activities during the period of our data. Although we do not have access to the content of appeals, the SRC told us that for each kind of appeal (i.e., donation or membership) and campaign, messages were identical across individuals.
To test whether appeals were targeted to givers who were more generous or more responsive to appeals, leading to an endogenous relationship between giving and appeals, we conducted an endogeneity test motivated by [44]. Separately for donation and membership, we estimated a system of two equations. In the first equation, the number of giving occasions of a giver in a year is modeled as a random coefficient Poisson regression whose rate parameter is a function of the number of appeals and year fixed effects. In the second equation, the number of appeals sent to a giver in a year is modeled as a Poisson regression whose rate parameter is a function of the appeals coefficient and intercept of the previous random coefficient Poisson regression, and year fixed effects. We find that in the estimated second equation the 95% credible intervals of coefficients of both the intercept and the appeals coefficient include zero; this holds for both donation and membership. Therefore, we conclude that appeals are not endogenous in our data, corroborating what the SRC told us: namely, that givers who give more often and/or are more appeal-responsive do not receive more appeals.
Table 1 presents summary statistics of the data. We classify the 2,171 individuals in our sample into three giving groups. "Pure donors" are the 232 individuals who gave only donations during the five-year period, "pure members" are the 720 individuals who gave only through the membership program, and "member-donors" are the 1,219 individuals who gave through both the donation and membership options. We identified members of each group on the basis of the full five-year observation period. Table 1 highlights that giving behaviors are considerably different across the three giving groups. On average, member-donors give 1.3 times and $80 per year, while non-member-donors give.8 times and $42 per year. Member-donors' giving frequency is thus larger by a factor of 1.6 times, and dollar amount by a factor of 1.9 times, than that of non-member-donors. Relative to pure donors or pure members, member-donors receive a larger number of appeals overall but receive similar numbers of appeals of each kind. This further supports the SRC's assertion that appeals are not targeted at the individual level.[ 8]
Graph
Table 1. Characteristics of Three Groups of Individuals in the Sample Data.
| | Pure Donors | Pure Members | Member-Donors | All Givers |
|---|
| Number of individuals (#) | 232 | 720 | 1,219 | 2,171 |
| Annual mean gift amount (US$) | 45.8 (85.6) | 40.6 (55.6) | 80.2 (141.7) | 63.4 (115.9) |
| Annual mean gift frequency (#) | .7 (.8) | .8 (.5) | 1.3 (.9) | 1.1 (.8) |
| Annual mean appeals (#) | 3.8 (2.4) | 6.1 (3.5) | 6.5 (3.8) | 6.1 (3.6) |
| Annual mean donation appeals (#) | 3.1 (2.1) | 2.8 (1.8) | 3.6 (2.0) | 3.3 (2.0) |
| Annual mean membership appeals (#) | .7 (.9) | 3.3 (2.6) | 2.9 (2.6) | 2.8 (2.6) |
| Mean age (years) | 57.2 (11.5) | 59.9 (11.8) | 61.9 (11.0) | 60.7 (11.4) |
| Female (%) | 71.6 | 53.3 | 61.6 | 60.0 |
| Mean distance from SRC (miles) | 881.1 (769.3) | 853.7 (817.8) | 876.5 (833.8) | 869.4 (821.6) |
| Monthly median income (US$) | 5,458 (89,038) | 5,467 (39,599) | 6,250 (14,014) | 5,935 (38,411) |
60022242921994584 Notes: Figures in parentheses are standard deviations. The three groups shown are mutually exclusive, and an individual's membership in a group is determined based on the entire window of five years of data.
Members pay an annual fee in one of nine tiers. However, the SRC adjusts membership fees as well as the number of tiers in most years. Moreover, the SRC encourages members to make an additional contribution by giving more than the minimum membership fee. These two phenomena are apparent in the histogram in Web Appendix A, which shows the membership fees paid in the five most popular tiers. These constitute 98% of all membership payment transactions. Because there are many more than five payment amounts in the data, we prefer to model individual membership decisions not as the selection of a tier, but as a continuous amount decision.
All members receive quarterly magazines related to the SRC's research, as well as token gifts such as a mug, blanket, and so on after joining and after each renewal. Moreover, members in higher tiers are invited to attend exclusive scientific tours. Under U.S. tax laws, membership fees are tax deductible to the extent of the fee amount less $20, which is considered to be the annual private value of the free quarterly magazines.
Figure 1 shows the mean giving amount and frequency by group over time. We see declines in giving amount and frequency for pure donor and pure member groups, whereas the member-donor group displays an increase in the amount and frequency of giving over the five years. As a consequence, the SRC's management especially values member-donors. Therefore, predictively identifying potential member-donors in the early stage of the contributor–NPO relationship and nurturing them will be beneficial for the SRC as well as for contributors who find the SRC to be a good match for their philanthropic goals.
Graph: Figure 1. Mean annual giving amount and frequency by group.Notes: The three groups shown are mutually exclusive and an individual's membership in a group is determined based on the entire window of five years of data.
Figure 2 shows the number of appeals sent by the SRC during the five-year period. As we have discussed, there is conspicuous seasonality in donation appeals, but less so in membership appeals. This is because the SRC sends donation appeals mostly in the spring in the United States, which is a migratory season for the animal species (and thus interest is high), and it is during "tax season." By contrast, the SRC sends membership renewal appeals throughout the year to individuals one or two months ahead of their membership expiration date.
Graph: Figure 2. Number of appeals sent by month for donation (left) and membership (right).
The previous data description sheds light on the differences in giving patterns across the three groups. However, we would like to quantify the role of factors that influence givers' decisions over time. For this, we require a model that accounts for the roles of both observed factors, such as appeals, demographics, givers' lifetime, and time since the last giving occasion, and unobserved factors, such as individual differences. Moreover, a model can be used to predict the effectiveness of future marketing actions, such as targeted appeals by the NPO.
As we have noted, a model of individual giving behavior needs to have separate mechanisms to deal with discrete (i.e., participation) and continuous (i.e., amount) decisions. Single discrete-continuous models have been widely used in the marketing literature due to their flexibility and utility theory–based primitives ([ 7]; [18]; [19]; [32]; [46]; [49]). However, this framework assumes that individuals choose only one option at a time, which is not the case in many real-life situations. To overcome this limitation, [38] proposed a multiple discrete-continuous utility maximizing model. However, [13], [14]) points out that Kim, Allenby, and Rossi's model is difficult to estimate due to its computational complexity. Instead, he proposes a tractable, closed-form utility maximizing model that allows individuals to also choose multiple alternatives simultaneously. This model is appropriate for our data, wherein 14% of givers gave in multiple forms and amounts in the same month at least once. Our model of giving behavior, which is further described in the "Givers' Decision Making" subsection, is in line with this stream of literature.
Accommodating parameter evolution is important in our context, and in many marketing contexts in which consumers' preferences and responsiveness to marketing activities change over time. In addition, as previously discussed, we expect that an individual's probability of giving will change with the passage of time since the last giving occasion, due to the fact that membership has an expiration date and the licensing effect ([36]). Moreover, Figure 1 also suggests that individuals' intrinsic preferences and responsiveness to solicitation appeals for giving options are likely to change over the lifetime of their relationship with the NPO ([52]). The considerable seasonality in giving observed in our data is yet another dynamic factor. To accommodate these dynamics, we extend [14]'s multiple discrete-continuous (extreme value) choice model by allowing the structural parameters to change over time through GP priors ([22]; [54]). We model these dynamics through three different GP components: lifetime, recency, and calendar time ([22]). We explain details of these dynamics in the "Capturing Time-Varying Parameters in the Giving Decision" subsection.
Previous literature has found that individuals are very heterogeneous in their giving behavior; some are more altruistic than others ([59]). Further, women ([66]), individuals who have less real or perceived distance to the NPO ([61]), and older individuals ([47]) have been found to be more generous. Thus, including individual differences in the model is useful for targeting. Controlling for individual heterogeneity also ensures we are not biasing the dynamic components of the model by introducing spurious state dependence ([33]). Therefore, we incorporate observed individual heterogeneity via demographics and unobserved heterogeneity via a random effects specification ([ 1]).
We model an individual's decision to give to an NPO that allows multiple giving options (e.g., donation and membership). We consider the individual's choices among the giving options and the amount of money given to each option as utility-maximizing behavior. Individual is assumed to maximize latent utility by choosing the amounts of money allocated to the outside good , and to giving option after tax deduction, , at time :
Graph
s.t∑j=0Jqijt=Qit, qi0t>0, qijt≥0.1
The utility in Equation 1 is quasiconcave, continuously differentiable, and an increasing function of expenditure .[ 9] To be a valid utility function, we need restrictions on Equation 1 such that , and . The expenditure on the outside good is defined as a numeraire that includes the aggregate expenditure on all outside goods (e.g., rent, grocery, gas), but excludes charitable giving to the focal NPO. Individuals consume outside goods in each time period, and the price of the outside good is normalized to unity ( ). The "price" of giving to option , , is related to an individual's marginal tax rate. [ 8] reports that in two-thirds of countries, including the United States, taxpayers can claim charitable giving as a deduction from taxable income. If the marginal tax rate is , as determined by an individual's income and the characteristics of giving option , we define .[10] When an individual gives to option , the net expenditure on the gift is because of the tax deduction associated with giving. We assume that the tax exemption (i.e., ) is spent on the outside option and for simplification ignore the gap in time between the giving occasion and when the tax deduction is received. The sum of all expenditures , which includes both spending on the outside good and giving to options, is the income of individual in time .
is the baseline marginal utility associated with option at the point of zero giving for individual in period . This can be seen by calculating the marginal utility of spending with respect to option (i.e., ). For two choices and with the same prices, if the baseline marginal utility of choosing charitable option is higher than for option , then an individual will choose option at the point of no giving to all options. Therefore, determines the set of options which the individual chooses. In the subsequent discussion, we refer to as the "baseline utility parameter," which we interpret as the individual's preference to choose option . Similarly, captures the baseline utility of the outside option.
Parameter serves to capture satiation effects in that it determines the change in marginal utility for inside option as the amount given to option by individual increases. This point is illustrated in Web Appendix B. In general, it can be shown that the higher the value of , the lower the satiation due to expenditure on option . Therefore, influences the amount given to option and allows multiple options to be chosen together (e.g., nonzero giving to both membership and donation options) through its impact on satiation ([14]). We refer to as the "satiation parameter" and interpret all parameters related to as affecting the levels of satiation.
The GP prior is a flexible and parsimonious way of representing time-varying parameters. It is flexible in that it can accommodate various data patterns such as periodicity and short-term and long-term effects via different covariance function specifications and the additive property of the GP (for details, see [22]]). It is parsimonious because a GP structure is determined by a small number of hyperparameters.
When compared with another common dynamic model—the hidden Markov model (HMM)—GPs require neither an assumption of discrete hidden states nor the Markov assumption. To illustrate the substantive importance of this feature, consider recent research which has shown that reciprocal motives to give can decay over time ([20]). Potential dynamic patterns of decay could be a gradual decline, a sudden drop, or all variations between these two extremes. A small number of discrete states would be able to capture the sudden drop pattern, while a larger number of discrete states would be needed to capture the gradual decline. Rather than making ad hoc assumptions, a good model should be able to detect the number of discrete states based on the data. Unlike HMMs, GPs do not assume a discrete, fixed number of hidden Markov states. Instead, they estimate a continuous state and automatically infer the type of decay from the data.
Furthermore, because the duration over which reciprocity can decay is unknown, the Markov assumption of an HMM requires the modeler to also conduct model selection on the correct number of time periods, which adds additional complexity to the model specification process. Again, the GP overcomes this problem by estimating the relevant duration as a feature of the model, based on the data.
Here, we briefly introduce the concept of GP. Readers are referred to [54] for a comprehensive review, and [22] and [23] for other GP applications in the marketing literature. GP is defined by a mean function and a covariance function, just as a multivariate Gaussian distribution is characterized by a mean vector and a covariance matrix. We denote this using notation , where is the mean function and is a covariance function. In the GP literature, the mean function is commonly set to be a constant, reflecting the lack of prior knowledge about the trends of the function ([54]; [55]). In contrast, a covariance function, also termed a kernel, plays a major role in determining the structure of a GP by governing its smoothness, amplitude, and differentiability.
As mentioned, we want to incorporate dynamic effects of lifetime, recency, and seasonality into our model. For this, we use lifetime ( ), defined as the time (in months) elapsed since individual gave for the first time; recency ( ), defined as the time elapsed since 's last giving occasion to option ; and calendar time ( ) as GP inputs. Among various available options of covariance functional forms ([54]; [55]), we use the squared exponential (SE) covariance function to model dynamic lifetime, recency, and appeal effects,[11] and adopt periodic covariance function (Per) to model the option specific seasonality. Details of covariance function specifications are available in Web Appendix C.
We incorporate factors such as lifetime, recency, seasonality, appeals, and individual heterogeneity into the model by parameterizing the baseline marginal utility ( ) as follows:
Graph
ψijt=exp βLj(li)+βRj(rij)+βSj(t)+βAj(li)+βAij×appealijt+βDj×demoi+β0ij+εijt,2
As the relationship between the giver and the NPO evolves, individuals' intrinsic preferences for giving option and responsiveness to appeals can change over time. Moreover, we expect recency effects on baseline utility to be nonmonotonic. To capture such effects, we define , , to represent time-varying influences of lifetime, recency, and seasonality for giving option , with input values in parentheses, respectively. The dynamic appeal effects for giving option are captured by where represents the number of option-specific appeals that individual receives at time .[12] Note that we use lifetime as a GP input that varies along with the individual-level time scales at any point in calendar time, thereby effectively capturing heterogeneity in the appeal effects. We model such dynamics in baseline utility via GP priors that are common across individuals as follows:
Graph
Graph
Graph
βAj(li)∼GPμβAj, KSEl,l';ηβAj,ρβAj.3
We incorporate observed individual heterogeneity by including demographics and incorporate unobserved heterogeneity via a random effects specification ([ 1]). is a vector that includes age, gender, and distance of residence from the focal NPO, and is the vector of corresponding parameters. and account for unobserved heterogeneity in intrinsic preference and responsiveness to appeals for option , respectively.
The error term represents unobserved factors that affect the attractiveness of option , and represents unobserved factors associated with the outside option. Following [14], we assume that both and follow i.i.d. type I extreme-value distributions.
To incorporate the factors that affect the level of satiation in giving, we parameterize similar to as follows:
γijt=exp δLj(li)+δRj(rij)+δSj(t)+δAj(li)+δAij×appealijt+δDj×demoi+δ0ij.4
represents time-varying average satiation level for option over time since the first giving occasion, and is the dynamic effect of time elapsed since the last giving occasion on satiation. and respectively represent the seasonality and time-varying effects of appeals on satiation for option . Similar to modeling dynamics in baseline utility, we incorporate dynamics in satiation via GP priors that are common across individuals (i.e., the specification is the same as Equation 3 except that we change to ). is the effect of observed individual characteristics on satiation, and and account for unobserved heterogeneity in satiation that are associated with the intrinsic preferences and responsiveness to appeals, respectively.
The utility-maximizing allocations of the giver's budget across options (including the outside option) can be derived by solving the Kuhn–Tucker conditions. (For a detailed derivation, see [13]]). From the Kuhn–Tucker conditions and i.i.d. type I extreme value distribution assumption on error terms, we obtain the following closed-form expression for the probability of optimal expenditure allocation on chosen giving options:
Graph
M!∏m=0M1qimt+γimtpimt∑m=0Mqimt+γimtpimt∏m=0Mexp(Vimt)∑j=0Jexp(Vijt)M+1,5
where and are defined as
Graph
Graph
lnqijtγijtpijt+1−lnpijt.6
[14] discusses empirical identification issues of the general multiple discrete-continuous choice model and discusses the specification termed the "gamma-profile," which we adopt here. A unique feature of our model is its incorporation of dynamics over time using GP priors in the multiple discrete-continuous choice model. We need restrictions on GP priors due to the additive structure of baseline utility and satiation specifications. Specifically, sums of two latent functions, such as , are equivalent to , where and for any , as both sums imply the same probability of giving. We address this problem by following [22] identification strategy, which is to set the initial function values of recency and seasonality to zero in both baseline and satiation, but we impose no restriction on lifetime terms. With these restrictions, and capture the average baseline utility and satiation level trends over time since the first giving occasion, respectively, and , , , and capture deviations from these mean trends. We discuss prior specifications and estimation procedures in Web Appendices D and E, respectively.
We compare the full model estimated using Equation 5, which incorporates both random effects (unobserved heterogeneity) and GP (dynamics), which we denote by M1, with more restrictive benchmark models, M2–M5. As we have discussed, a distinctive feature of our framework is that we allow for multiple forms of giving, whereas previous research combines different forms of financial giving as a composite amount ([37]; [39]; [52]; [63]). We investigate the effect of not distinguishing between forms of giving by estimating a "single giving option" model (M2), which shares all features of M1 (such as GP and random effects), but the parameters are restricted to be common across the two giving options. We discuss the results of M2 subsequently. To assess the importance of controlling for individual heterogeneity and time varying preferences, we estimate three restricted model specifications, M3–M5. M3 includes random effects, but omits the GP component. M4 incorporates dynamics, but omits random effects. M5 omits both random effects and GP components.
We find that accounting for multiple forms of giving and controlling for unobserved heterogeneity and dynamics in parameters significantly improves the predictive accuracy of the model. M1 exhibits the best model fit, significantly better than the second-best-fitting model, M2. This implies that aggregation across forms of giving indeed disguises differences in individual giving behavior, thus explicit modeling of both forms is necessary. The superiority of M1 over M3 indicates that including dynamics is desirable, while the superiority of M1 over M4 suggests that including unobserved heterogeneity is desirable. Hereinafter, we focus on the results of the full model (M1) because it shows the best performance (model comparison results are available in Web Appendix F).
Figures 3 through 6 plot the evolution of time-varying parameters estimated using the GP priors. In all figures, the solid lines represent posterior means and the gray areas are 95% credible intervals. Subscripts 1 and 2 stand for donation and membership, respectively.
Graph: Figure 3. Lifetime effects on baseline utility (βL1(li), βL2(li)).Notes: The solid lines are posterior means and the dotted lines are local polynomial regression fit to the means. The shaded areas are 95% posterior credible intervals.
Table 2 shows posterior means and standard deviations related to the GP and the statistic. As explained in Web Appendix C, these parameter estimates summarize the characteristics of the GP structures. Table 3 provides observed and unobserved heterogeneity parameter estimates and the statistic. The statistic is a measure of convergence of the model ([27]). values less than 1.1 indicate successful convergence. Next, we discuss the estimated parameters with an emphasis on their implications for the SRC.
Graph
Table 2. Covariates of Baseline Utility and Satiation Parameters.
| Baseline Utility | Satiation |
|---|
| Component | Parameter | Mean | SD | | Component | Parameter | Mean | SD | |
|---|
| Lifetime | | −14.36 | (.23) | 1.01 | Lifetime | | 3.89 | (.16) | 1.01 |
| | 1.09 | (.09) | 1.00 | | | 2.35 | (2.02) | 1.00 |
| | .05 | (.02) | 1.00 | | | .02 | (.01) | 1.00 |
| | −15.66 | (.25) | 1.01 | | | 4.11 | (.13) | 1.05 |
| | 1.59 | (.52) | 1.01 | | | 1.96 | (1.99) | 1.01 |
| | .26 | (.09) | 1.00 | | | .01 | (.00) | 1.00 |
| Recency | | 2.28 | (1.18) | 1.01 | Recency | | 1.26 | (.73) | 1.00 |
| | .13 | (.06) | 1.00 | | | .00 | (.00) | 1.01 |
| | 11.37 | (4.28) | 1.01 | | | 1.55 | (.52) | 1.01 |
| | 2.35 | (.49) | 1.00 | | | .01 | (.01) | 1.01 |
| Seasonality | | .41 | (.08) | 1.00 | Seasonality | | 1.35 | (.74) | 1.00 |
| | 1.14 | (.49) | 1.00 | | | .13 | (.18) | 1.00 |
| | .43 | (.11) | 1.00 | | | 1.37 | (.84) | 1.00 |
| | .33 | (.30) | 1.00 | | | .01 | (.02) | 1.00 |
| Appeals | | .48 | (.09) | 1.00 | Appeals | | -.06 | (.05) | 1.00 |
| | .71 | (.20) | 1.00 | | | 1.30 | (.75) | 1.00 |
| | .14 | (.05) | 1.00 | | | .00 | (.00) | 1.01 |
| | .84 | (.10) | 1.00 | | | −.20 | (.04) | 1.02 |
| | 1.20 | (.45) | 1.01 | | | 1.55 | (1.22) | 1.01 |
| | .15 | (.05) | 1.00 | | | .00 | (.00) | 1.00 |
70022242921994584 Notes: Subscripts 1 and 2 represent donation and membership, respectively. Mean and SD refer to posterior means and posterior standard deviations, respectively. The statistic, called potential scale reduction factor, measures Hamiltonian Monte Carlo model convergence. values less than 1.1 indicate successful convergence.
Graph
Table 3. Observed and Unobserved Heterogeneity in Baseline Utility and Satiation Parameters.
| Baseline Utility | Satiation |
|---|
| Parameter | Mean | SD | | Parameter | Mean | SD | |
|---|
| .36 | (.07) | 1.00 | | −.01 | (.05) | 1.00 |
| −.09 | (.06) | 1.00 | | −.02 | (.04) | 1.00 |
| .01 | (.00) | 1.01 | | −.01 | (.00) | 1.01 |
| .01 | (.00) | 1.01 | | −.00 | (.00) | 1.03 |
| −.07 | (.04) | 1.00 | | .06 | (.03) | 1.00 |
| −.03 | (.03) | 1.01 | | −.00 | (.02) | 1.00 |
| 1.28 | (.04) | 1.00 | | .05 | (.03) | 1.00 |
| 1.05 | (.04) | 1.01 | | .02 | (.01) | 1.00 |
| .27 | (.06) | 1.03 | | .03 | (.02) | 1.00 |
| .15 | (.05) | 1.01 | | .02 | (.01) | 1.00 |
80022242921994600 Notes: Subscripts 1 and 2 represent donation and membership options, respectively. Mean and SD refer to posterior means and posterior standard deviations, respectively. Values in bold are estimates whose 95% credible intervals do not include zero. The statistic is a measure of convergence of the model. values less than 1.1 indicate successful convergence.
In Figure 3, we show how the baseline utility of donation ( ) and membership ( ) evolves with lifetime, which is defined as the time in months since an individual's first giving occasion. All the individuals in our cohort first gave in 2011, but at different dates, as a result their lifetimes are different from calendar times. The time-varying baseline utilities reflect changes in the propensity to donate or to renew a membership as the length of an individual's relationship with the SRC grows.
In the left panel of Figure 3, we find that the propensity to donate increases with lifetime, implying that individuals tend to add donation to their giving portfolio, or donate repeatedly over their lifetime, even in the presence of a membership option that provides tangible benefits. The increasing pattern is consistent with previous findings that prosocial behaviors can become habitual over time ([29]; [65]).
The propensity to participate in membership (right panel of Figure 3) is cyclical, exhibiting peaks every 12 months. We note that the cyclicality pertains not to the utility derived from consuming membership benefits, which our data cannot identify, but to the utility from the act of renewing membership. Further, unlike donation, the utility from membership does not increase over the five-year period. [15] reported that members with longer lifetimes have lower hazards of lapsing. However, we do not find evidence consistent with this finding, perhaps because of the availability of donation as an option in our setting.
We show how satiation levels for donation ( ) and membership ( ) vary with lifetime in Web Appendix G. We see that both and tend to increase over time, which means givers feel less satiated with their giving as lifetime increases. In other words, the amount they give increases as their relationship with the SRC continues.
Individuals may not actively consider giving every month, due to the aforementioned licensing effect of donation and the duration of a valid membership. The estimated recency effects help us think about how people's giving preferences change with the passage of time since the last giving occasion.
In Figure 4, we show the estimated changes in the baseline utility of donation ( ) with recency. The utility is low immediately after giving for about seven months, which is consistent with a licensing effect because a recent donation gives the donor "license" to not give for some time ([36]). However, the 95% credible interval of contains zero for most of the seven-month period. This finding is in line with a recent meta-analysis that reported relatively small licensing effects in the literature ([17]). The estimated in each of the first two years is cyclical, with peaks at 11 and 24 months, which can be interpreted as an intrinsic cyclicality of donating since we have controlled for the effects of appeals and seasonality in the model. After 24 months, the 95% credible interval contains zero throughout. This suggests that the SRC can reasonably assume that if more than 24 months have elapsed since the last donation, an individual is likely to have churned.
Graph: Figure 4. Recency effects on baseline utility (βR1(rij), βR2(rij)).Notes: The solid lines are posterior means and the shaded areas are 95% posterior credible intervals.
Estimated changes in baseline utility of membership ( ) with recency are shown in the right panel of Figure 4, and they exhibit a different pattern. The utility decreases for a brief period, then peaks at about 12 months, and then declines slowly till around month 56. As expected, the initial drop and 12-month peak capture the annual membership renewals. The subsequent trend indicates that even if an individual does not renew membership after one year, there is a reasonable possibility of renewal in future years.
The contrast in the effect of recency on baseline utilities of donation versus membership is informative. The longer-lasting positive effect in the case of membership suggests that the twin rewards of tangible benefits (e.g., mugs, T-shirts) and intangible benefits (e.g., sense of affiliation, and warm glow) play a role, and therefore renewal efforts should be continued. To some extent, our finding is consistent with the reported positive effects of extrinsic benefits on prosocial behavior ([10]; [15]; [26]).
Unlike baseline parameter estimates, the effects of recency on satiation for both options are small, and the 95% credible intervals contain zero except for a limited time for membership. The results are available in Web Appendix G.
As discussed previously, there is considerable seasonality in giving to the SRC. Figure 5 displays option-specific seasonal changes in baseline utility over calendar time ( ). Seasonality is salient for donations and peaks every spring and calendar year-end; spring is the migration season of the animal species that the SRC studies, and year-end increases are due to the tax-year definition. GP priors with periodic kernel capture this pattern well. The seasonality of membership shows only a calendar year-end peak. This is because more members have joined, and consequently renewed, at the end of the calendar year, presumably to capitalize on tax benefits. Again, seasonal effects on satiation are smaller in magnitude compared with those on baseline utility (see Web Appendix G).
Graph: Figure 5. Seasonal effects on baseline utility of donation (βS1(t), βS2(t)).Notes: The solid lines are posterior means and the shaded areas are 95% posterior credible intervals.
In Figure 6, we show effects of appeals on baseline utility over time. Both pure donors and pure members received both kinds of appeals: pure donors received 3,686 donation appeals and 827 membership appeals, while pure members received 10,331 donation appeals and 12,035 membership appeals. Moreover, on approximately 20% of occasions, members did not receive renewal notices when their membership was about to expire.[13] These variations in the data make it possible to estimate the effectiveness of each type of appeal separately from inherent unobserved differences between donors and members.
Graph: Figure 6. Appeal effects on baseline utility (βA1(li), βA2(li)).Notes: The solid lines are posterior means and the dotted lines are local polynomial regression fit to the means. The shaded areas are 95% posterior credible intervals.
Time-varying appeal effects on baseline utility differ between the giving options. In the left panel of Figure 6, the effects of donation appeals on baseline utility ( ) over givers' lifetime exhibit a notable decrease in the early stages followed by a gradual increase over time. Although donation appeals generally have positive effects on donation participation, their effectiveness is not consistent; the 95% posterior intervals contain zero in many periods. This implies that some donors would have given regardless of receiving donation appeals, while others do not give despite receiving appeals. We conjecture that this is partly because the SRC sends standardized appeals to all potential donors rather than targeted and personalized donation appeals. This suggests room for improvement in the efficiency and effectiveness of donation appeals by targeting. One interesting anomaly is that the effectiveness of donation appeals shows a conspicuous peak around month 48. This is because in 2015 the SRC celebrated a special event based on the time since its founding (we do not provide details to protect the anonymity of the SRC). To encourage participation in the celebration, the SRC sent more appeals at the year-end of 2015 (see Figure 2) perhaps with a special message of celebration. Therefore, individuals were especially responsive to donation appeals at that time.
Membership appeals ( ) have a consistent positive effect on membership renewal, although these appeals are more effective in the first three years and substantially less so later. We conjecture that membership appeals are effective because membership programs provide a natural structure for sending targeted renewal notices to individuals whose membership is about to expire. Interestingly, compared with lifetime effects, the membership appeal effect exhibits a delay relative to the 12-month cycle: it peaks around month 15, month 31, and month 43. This is related to the fact that many members do not renew their membership in exactly 12 months. Among individuals who renewed at least three times, the mean interrenewal time is about 15 months between the first two renewals, and about 14 months between the second and third renewals. This delayed renewal behavior leads to loss of revenue for the SRC because an annual membership renewed in month 15 instead of month 12 next expires in month 27, instead of month 24. Thus, in a five-year span, three renewals each delayed by three months lead to a loss of nine months of membership fee.
Web Appendix G shows the time-varying effects of appeals on satiation for each option. We see that the 95% credible intervals of donation appeals ( ) contain zero in all periods. By contrast, the average effect of membership appeals on satiation ( ) is consistently negative, suggesting that members may feel satiated with the large number of membership appeals. Previous literature has found mixed effects of solicitation appeals on the amount decision; [45] reported that solicitation appeals are effective for encouraging participation in donations, but have no effect on the amount given, whereas [63] found that too many appeals cause annoyance and reduce giving. We find evidence in support of the former result for the donation option and the latter result for the membership option.
We notice that appeal effects on baseline show greater variability over time than appeal effects on satiation. To compute the magnitude of the effect of an incremental appeal on the probability of participation separately from the amount given, we proceed as follows, separately for donation and membership appeals. For each individual we add one hypothetical appeal at a random point in time during the five-year period and use the estimated model to compute the percentage change in the participation probability and expected amount given, relative to the baseline number of appeals in the observed data. We choose the random point in time at which the additional appeal is inserted 500 times and repeat the exercise, thereby obtaining the average percentage changes. For donations, we find that on average the additional appeal increases the probability of participation by 1.21% (95% credible interval = [1.18, 1.24]), but the additional appeal does not change the expected amount given (conditional on participation). Similarly, for membership, we find an increase in participation probability of 1.5% (95% credible interval = [1.35, 1.66]) but no change in the expected amount given. Therefore, we conclude that individuals are more responsive in the likelihood to participate than in the amount given. The detailed analysis procedure is described in Web Appendix H.
The upper part of Table 3 shows the estimated effects of demographic characteristics and their convergence diagnostic . Our results add nuance to previous studies that have found that women give more frequently than men ([56]). We find that women have higher baseline utility for the more intrinsically motivated option, donation, whereas men have higher intrinsic preference for the more extrinsically motivated option, membership. Older individuals have higher baseline utility than younger individuals for both options, which corresponds to [47]'s findings. We also find that individuals who live closer to the SRC exhibit a lower satiation level with the donation option and make larger donations. This corresponds to [61] finding that donations increase as real or perceived distance to the recipient decreases. The SRC could use this finding to guide its future targeting efforts.
The lower part of Table 3 presents posterior estimates of the standard deviations of the random intercept and the appeal effect. We can judge the statistical significance of an estimated standard deviation by whether the posterior mean is no less than twice the posterior standard deviation. This is because we have restricted all standard deviation estimates to be positive, and therefore the posterior distributions of these parameters do not contain zero. Using this criterion, the standard deviations of baseline utility parameters are all greater than zero, whereas the standard deviations of satiation parameters are in all cases not different from zero.
Based on the baseline utility, we show individuals whose 95% credible intervals of intrinsic preferences for the two giving options do not contain zero in Web Appendix I. There are 165, 130, and 83 individuals whose estimated intrinsic preferences for donation only, membership only, and both, respectively, meet the criteria (i.e., the 95% credible intervals do not contain zero). We use these results to identify potential member-donors in the "Managerial Applications" section.
One of the key contributions of this article is to shed light on factors that drive different forms of giving. As discussed previously, the effects of lifetime, recency, seasonality, appeals, and individual heterogeneity on donation and membership are distinctive. To further assess the importance of allowing multiple forms of giving in the individual model, we estimate a "single giving option" model (see M2).
We found that the "single giving option" model blurs the dynamic effects of lifetime, recency, seasonality, and appeals on baseline utility. Moreover, it misses key insights about appeal effects on satiation. Therefore, in the absence of insights about differences in giving behavior, the SRC can mistakenly focus its marketing efforts on one option, when a focus on the other option would be more productive. The results and discussion are available in Web Appendix J.
While the primary purpose of the proposed model is to describe how specific factors drive individual giving behavior, the model estimates can also usefully inform decision making aimed at augmenting the SRC's revenues. We discuss here two applications of our model for fundraising by the SRC. The first application demonstrates how our model can help the SRC to predictively identify potential member-donors, the second explores how the SRC can increase revenue from each giving option by using targeted appeals.
To enable these predictions, we have derived the conditions in our model for a giving option to be chosen by a utility-maximizing individual, as well as the optimal expenditure on the chosen option (see Web Appendix K). The condition for option to be chosen can be written as follows:
λit<ψiltpilt,7
where is a parameter stemming from the giver's budget constraint, expressed mathematically in Web Appendix K. The optimal expenditure on the chosen option is
qilt=γiltψiltλit−pilt.8
Because and are functions of error terms, we draw 3,000 random samples of and from the type I extreme value distribution. We use the conditions in Equations 7 and 8 in the following analyses.
Member-donors are particularly valuable to the SRC compared with pure donors and pure members because they give more frequently, they give larger amounts, and their donation amounts increase over time (see Figure 1). It is thus important to predict which pure donors and pure members are likely to become member-donors in the future, prioritize these individuals, and nurture relationships with them.
Here, we show how our model estimates can be used to predictively identify member-donors. For this exercise, we divide the 62 months of our data into the first 42 months (January 2011 to June 2014) as an estimation sample, and the remaining 20 months (July 2014 to February 2016) as a holdout sample. In the estimation sample, the cohort of 2,171 individuals includes 284 "pure donors" and 963 "pure members"; these are individuals who give only through donation programs or membership programs, respectively, during the estimation sample period. In the holdout sample, 52 of the 284 pure donors (18.3%) and 243 of the 963 pure members (25.2%) became member-donors.
We use the model estimates based on the estimation sample to calculate an ordinal metric of the potential to become a member-donor in the future, which we term the "donor susceptibility to membership" for pure donors, and "member susceptibility to donations" for pure members. The exercise is conducted separately for pure members and pure donors. We describe first the exercise for pure donors. For each individual and each month in the estimation sample, we calculate the expected probability of choosing membership (i.e., the option not chosen by a pure donor) by using Equation 7.[14] The expected probability depends on lifetime, recency, seasonality, and individual heterogeneity, because the numerator of the right-hand side of Equation 7—the baseline utility —is a function of these factors. Because the expected probability of choosing membership is time-varying, we compute the average probability of membership across months for each pure donor. We then rank the 284 pure donors in decreasing order of this average probability, the donor susceptibility to membership. For example, if pure donor A has a higher donor susceptibility to membership than another pure donor B, we expect that A has a higher probability of becoming a member-donor in the future than B. An analogous exercise is conducted for pure members, whereby we rank-order them based on a similar ordinal metric, the member susceptibility to donations. To compare the predictive accuracy of our model, we conducted the same analyses with the aforementioned benchmark models M2–M5.
Figure 7 plots the gains charts computed in the holdout sample based on the exercises described previously. The gains chart on the left depicts on the horizontal axis pure donors arranged in decreasing order of the expected probability of membership computed in the estimation sample. The vertical axis depicts the percentage of the 52 member-donors in the holdout sample who are "captured." The dotted (black) 45-degree line indicates baseline performance, which is random targeting without a model. In the left panel, for example, randomly choosing 50% of 284 pure donors in the estimation period would yield half of 52 (i.e., 26) member-donors in the holdout sample period. The solid (red) line indicates the % captured using the proposed model. For instance, our model helps to identify 90% of the member-donors (i.e., 47 of 52 member-donors) if we choose 50% of pure donors. Similarly, in the right panel, if we randomly choose 50% of 963 pure members in the estimation period, we would expect to identify 122 of the 243 member-donors in the holdout sample, whereas using our model the SRC can identify 75% of member-donors (i.e., 183 of 243 member-donors).
Graph: Figure 7. Gains charts for targeting member-donors in holdout sample.
Comparison of the left and right panels of Figure 7 shows that model M1 does better in predicting which pure donors will become member-donors than in predicting which pure members will become member-donors. One reason for this difference is that at least 25.2% (243 out of 963) of pure members need to be targeted to reach 100% of those who converted, while the corresponding number is only 18.3% (52 out of 284) for pure donors. This is indicated by the solid black lines labeled "maximum" in the left and right panels. Thus, there is less opportunity for any model in the case on the right. A second reason is indicated by an analysis of the estimated probabilities of membership (data underlying left panel) and donation (data underlying right panel). The coefficient of variation (i.e., standard deviation/mean) of the probabilities of membership estimated via model M1 is.52, and of the estimated probabilities of donation is.39. We conjecture that the lower variability in the estimated probabilities across pure members translates into a lower ability of the model to bring to the top of the data those who became member-donors.
The other dashed lines indicate the gains from using the benchmark models (M2–M5). Gains from the proposed model are greater than those from the benchmark models at all levels of cutoffs (i.e., 50% and other levels) in both pure donor and pure member cases. The proposed model thus provides a useful predictive tool that enables the SRC to identify pure donors and pure members who will become member-donors.
Here, we illustrate the impact of targeting appeals on the basis of the revenue gains for each giving option, thereby improving the effectiveness of appeals. Specifically, we consider a counterfactual situation in which the SRC reallocates 1% of appeals sent during the sample period, holding constant the total number of appeals sent. To achieve this, we simulate two cases: counterfactual and baseline. In the baseline case, appeals sent to individuals in the data are left unchanged. In the counterfactual case, 1% of appeals sent are randomly chosen and reassigned according to the targeting strategies explained next.
As Web Appendix I shows, we can identify individuals with a high intrinsic preference for each giving option. We have further found that (a) recency effects show that repeat giving is more likely when recency is between 11 and 13 months for both options, (b) lifetime effects on baseline utility are high every 12 months for membership, and (c) both options exhibit considerable seasonality. These findings suggest that the SRC may benefit by sending appeals in months when the likelihood of giving is high. Using these results, we consider three targeting strategies for the SRC: ( 1) one that targets individuals with higher intrinsic preference for a specific giving option (i.e., individuals who are identified in Web Appendix I) but does not target timing; ( 2) one that targets timing when the likelihood of repeat giving is high due to lifetime, recency, and seasonal effects on baseline utility (i.e., conditions a, b, and c explained previously) but does not target individuals; and ( 3) one that targets both individuals and timing.[15] In each strategy, we compare outcomes with the baseline case, in which the SRC does not reallocate appeals. To obtain the expected revenues in both counterfactual and baseline cases, we calculate the optimal expenditure using Equation 8. To compute the standard error of the revenue increase in each counterfactual scenario, we randomly draw 500 posterior samples of every parameter estimate and repeat the calculations.
In Figure 8, we show the "appeal elasticity of expected revenue" across different targeting strategies. We define appeal elasticity as the percentage gain in revenues relative to baseline for each strategy. The left panel indicates that from the "individual," "timing," and "both" strategies, the SRC can expect, on average, 1.12%, 1.37%, and 1.53% increases in the donation revenue, respectively. Targeting "when" is slightly more effective than targeting "who" in our empirical context. Further, as we expected, the gains from targeting both individuals and timing are the largest. Currently, the SRC conducts two untargeted appeal campaigns for donation in the spring and at the year-end, and our result indicates that the SRC has a lot of room for improving. The right panel indicates that individual targeting (i.e., strategy 1) results in a.32% decline in membership revenue. This is unsurprising because, currently, the SRC sends an appeal to individuals when their memberships are about to expire. Reallocating appeals to individuals who have high preference for membership is suboptimal because targeted timing is critical. However, we do find that sending membership renewal appeals at the right time, or targeting both the right individuals and the right time, can increase membership revenue by 1.30% and 1.94%, respectively.[16] Again, targeting "both" is significantly more effective than targeting "when" for the membership case as well. These analyses show that the SRC can be better off in terms of revenue with more targeted marketing efforts based on the model.
Graph: Figure 8. Appeal elasticity of expected revenue across different targeting strategies.Notes: Error bars indicate 95% credible intervals.
Nonprofit organizations play a central role in many economies in the form of private entities discharging a public purpose. Because individual philanthropy is the primary funding source for many such organizations, strengthening their fundraising capabilities can have a large impact on their survival, growth, and effectiveness. Our article is motivated by this goal. We propose an empirical model of giving behavior to an NPO that offers multiple giving options. To our knowledge, extant research does not consider this case. Our utility-based framework accounts for separate mechanisms that determine baseline utility and satiation for each giving option by modeling factors that affect individual giving decisions. A unique feature of our multiple discrete-continuous choice model is that it allows the structural parameters to dynamically change over time via Bayesian GPs. Incorporating the dynamics of structural parameters is important in modeling giving behavior because we expect the effects of lifetime, recency, seasonality, and responsiveness to appeals on baseline utility and satiation to change over time as the relationship between the giver and the NPO evolves. With our proposed model, managers of NPOs can manage more effectively.
Our analysis of the five-year giving data of the SRC leads to several insights about how to better manage fundraising activities of NPOs. We summarize these insights next.
Some nonprofit organizations have questioned the need for maintaining two ways of individual giving (i.e., donation and membership), because they involve separate cost structures. These costs include administrative and managerial overheads, ongoing fundraising campaigns for donations, and costs of benefits provided to members. Our data indicate that a strategy of having multiple giving programs has two benefits. First, it allows differently motivated individuals to choose a form of giving that is best for them. Second, it creates a path for more committed givers to broaden and deepen their engagement over time by contributing in multiple forms (see Figures 1, 3, and 4). Further, data on individual characteristics and past giving that are easily available to the NPO can be helpful to identify those givers who are likely to adopt a second giving option in the future and thus enable the organization to focus its resources on cultivating such individuals (see Figure 7).
Our results show that givers exhibit greater responsiveness with respect to participation in a giving program than with the amount of donation or the membership tier (see the "How Effective Are Donation and Membership Appeals?" subsection). Although our data do not reveal the reasons for this, there is a clear implication for NPOs, which is to focus their efforts on encouraging repeat donations, membership renewals, and adding on a second form of giving. In our case, it is less useful to emphasize giving larger amounts or upselling to higher membership tiers.
Given the low rates of repeat giving experienced by many NPOs, an important question facing the SRC and most NPOs in general is how to manage individuals who have not given for some time. In the case of donors, we learned that after two years since the last giving occasion, there is little positive predisposition to give again (see Figure 4). Therefore, the SRC should pursue donors for two years after their first giving occasion as potential repeat givers. Subsequently, these individuals may need the same level of effort as new prospects to be cultivated all over again.
In contrast to lapsed donors, members who have not renewed for several years are still positively predisposed to return to the NPO (see Figure 4). Thus, renewal efforts should be continued and may yield greater success than pursuing a new "cold" prospect for membership. Our recommendation is similar to one that follows from the "recency trap" in the context of customer relationship management ([51]); because customers with higher recency (i.e., bought a long time ago) tend to have lower purchase likelihood now, the firm ignores them in its marketing, thereby making them even less likely to buy in the current and subsequent periods. Moreover, relative to new prospects, the NPO knows more about lapsed members' interests and motivations given data on their past usage of the NPO's products and services, thereby allowing customized marketing.
In the case of the SRC, appeals for donations have been inconsistently effective in the data (see Figure 6). This suggests room to improve, perhaps through reassignment of appeals via targeting. Our analysis indicates directions to guide such targeting, in particular, that targeting "when" an appeal is sent is likely to be more fruitful than "to whom" an appeal is sent (see Figure 8). Of course, there may also be room to improve through personalization of the content of appeals, but our data are unable to speak to this question. By contrast, membership appeals, whose timing is already targeted, are effective, with one caveat. The SRC may be sending too many renewal appeals (the average "pure member" currently gets 3.3 appeals each year), creating a negative effect (see Web Appendix G). Understanding this issue better may require focus groups or surveys of current members.
To what extent do our findings based on the organization we studied extend to the context of nonprofits more broadly? Interested givers are very diverse in terms of their level of engagement with the inherent subject (i.e., the animal species), the utility they derive from the content, and their capacity to give. In this respect, our SRC is likely to be similar to many NPOs that are engaged with arts, culture and entertainment (e.g., theaters, galleries, museums, historic preservation societies), and the environment (e.g., botanical gardens, parks, conservation societies), and we expect that many of our findings about giving behaviors may readily translate. For instance, in these settings, it is possible to construct multitiered membership programs with well-defined benefit packages that are tied to the central mission of the NPO. By contrast, our findings are likely to be less applicable to NPOs that are engaged in contexts such as religion (churches), health care (hospitals), or education (universities), because the concept of membership is less natural in these settings.
As noted in the introduction, nonprofit organizations often play key roles in domains such as health, education, the arts, and conservation of the environment and animals. Successful fundraising is critical to the survival and health of most nonprofits; however, fundraising is costly, resulting in a dilution of their full potential impact on the world. An important reason for the high cost of fundraising is the inordinate focus on acquiring new givers because of low repeat giving rates. Our research highlights the need and the opportunity to use data and marketing science tools to understand how individuals' motivations for giving evolve over time and thereby develop strategies to increase giving rates and amounts. In particular, we examine the strategy of using multiple forms of giving that provides a path for committed givers to grow their giving over time. The net effect of employing these tools should be that a larger percentage of funds raised can be deployed in the service of the mission of the nonprofit.
Our study has the following limitations and leaves open several directions for future research. First, we did not model nonfinancial contributions such as volunteering or donations in kind, although these are important aspects of giving. In future research, it will be important to accommodate the value of nonfinancial contributions within the model.
Second, our model and data account for only one focal NPO. In reality, individuals often give to more than one organization and NPOs compete with one another for philanthropic dollars. The NPO we study is involved in a very specific scientific research area (an animal species), and therefore does not face much direct competition. However, in other contexts, competition can be intense between NPOs or between NPOs and for-profits. Therefore, modeling competitive interactions in giving can be a fruitful avenue for future studies.
Finally, the sparsity of our data (e.g., annual frequency of giving is 1.1) hinders modeling the evolution of parameters at the individual level; thus, we only allow for population-level evolution of parameters. In other applications, disaggregating the population-level evolution of parameters to the individual level may reveal additional layers of important insights.
In summary, our work examines the who, when, and why of individual-level giving to an NPO. We hope that our framework enables NPOs' fundraising strategies to be more effective, thereby empowering them to contribute more to a better world.
Supplemental Material, sj-lyx-1-jmx-10.1177_0022242921994587 - Managing Members, Donors, and Member-Donors for Effective Nonprofit Fundraising
Supplemental Material, sj-lyx-1-jmx-10.1177_0022242921994587 for Managing Members, Donors, and Member-Donors for Effective Nonprofit Fundraising by Sungjin Kim, Sachin Gupta and Clarence Lee in Journal of Marketing
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921994587 - Managing Members, Donors, and Member-Donors for Effective Nonprofit Fundraising
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921994587 for Managing Members, Donors, and Member-Donors for Effective Nonprofit Fundraising by Sungjin Kim, Sachin Gupta and Clarence Lee in Journal of Marketing
Footnotes 1 Scott Neslin
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Sungjin Kim https://orcid.org/0000-0001-8657-1506 Sachin Gupta https://orcid.org/0000-0001-9459-5233 Clarence Lee https://orcid.org/0000-0001-6912-4800
5 Online supplement: https://doi.org/10.1177/0022242921994587
6 NPOs also depend on nonfinancial resources such as volunteering or donations in kind. We only consider financial contributions.
7 To check robustness, we also estimate the models by retaining individuals who gave at least three times, four times, and five times. Across all these samples, the results remain qualitatively the same. Additional details of the results are available from the authors on request.
8 A possible concern is that member-donors receive 3.6 donation appeals each year on average, compared with only 3.1 appeals received by pure donors, leading to an upward bias in the estimated effect of donation appeals. We conducted a robustness check by truncating the sample by removing member-donors who received a "high" number of donation appeals. To determine the threshold for "high," we visually inspected the histograms of number of appeals for pure donors and member-donors. Removing 137 member-donors who received 23 or more donation appeals during the period of our data resulted in the right tails of the two histograms becoming similar. The estimated donation appeal effect of the proposed model in this truncated sample remained almost unchanged relative to the full sample. Additional details of the results are available from the authors on request.
9 This specification is a variant of the "Gamma profile" in the literature. In the initial stage of our research, we tried other specifications such as the "Alpha profile," but the current specification showed a better fit. For other possible specifications, see [14], [57], and [62].
Some forms of charitable giving are not fully tax deductible in the United States because of the private value of goods and services received as benefits of giving. For example, in our empirical context the membership fee is not fully tax deductible.
We tested other covariance function specifications, such as rational quadratic kernel ([55]) and Matern kernel ([23]), and the empirical results remained qualitatively the same.
Our empirical application uses appeals lagged by one month as predictors because the SRC sends appeals one month prior to the month in which the contribution is expected to be received. For example, year-end donation appeals are sent in November, and membership renewal appeals are sent one month before the membership expiration date.
This appears to be due to operational issues, details of which are not known to us.
To compute the expected probability, we calculate the condition in Equation 7 for each of 3,000 random draws of the error term and then take the proportion that meets the condition.
We recognize that there may be other bases to target as well, such as targeting based on individuals' estimated responsiveness to appeals. However, for brevity, we limit ourselves to these three strategies.
To interpret the magnitude of these estimated elasticities, it is useful to note that the intervention is a reassignment of 1% of appeals and not an increase in appeals. Another point of comparison could be advertising elasticities; drawing on a meta-analysis, [58] report the average short-term and long-term advertising elasticities to be.12 and.24, respectively.
References Allenby Greg M., Rossi Peter E. (1998), "Marketing Models of Consumer Heterogeneity," Journal of Econometrics, 89 (1/2), 57–78.
Andreoni James. (1989), "Giving with Impure Altruism: Applications to Charity and Ricardian Equivalence," Journal of Political Economy, 97 (6), 1447–58.
Andreoni James. (1990), "Impure Altruism and Donations to Public Goods: A Theory of Warm-Glow Giving," Economic Journal, 100 (401), 464–77.
Andreoni James, Payne A. Abigail. (2011), "Is Crowding Out Due Entirely to Fundraising? Evidence from a Panel of Charities," Journal of Public Economics, 95 (5/6), 334–43.
Andreoni James, Payne A. Abigail. (2013), "Charitable Giving," Handbook of Public Economics, 5, 1–50.
Ariely Dan, Bracha Anat, Meier Stephan. (2009), "Doing Good or Doing Well? Image Motivation and Monetary Incentives in Behaving Prosocially," American Economic Review, 99 (1), 544–55.
Arora Neeraj, Allenby Greg M., Ginter James L. (1998), "A Hierarchical Bayes Model of Primary and Secondary Demand," Marketing Science, 17 (1), 29–44.
Austin Amy. (2016), "Two Thirds of Countries Offer Individual Tax Incentives on Charity Donations," Accountancy Daily(May 17), https://www.accountancydaily.co/two-thirds-countries-offer-individual-tax-incentives-charity-donations.
Auten Gerald E., Sieg Holger, Clotfelter Charles T. (2002), "Charitable Giving, Income, and Taxes: An Analysis of Panel Data," American Economic Review, 92 (1), 371–82.
Bartlett Monica Y., DeSteno David. (2006), "Gratitude and Prosocial Behavior: Helping When It Costs You," Psychological Science, 17 (4), 319–25.
Belleflamme Paul, Omrani Nessrine, Peitz Martin. (2015), "The Economics of Crowdfunding Platforms," Information Economics and Policy, 33, 11–28.
Bénabou Roland, Tirole Jean. (2006), "Incentives and Prosocial Behavior," American Economic Review, 96 (5), 1652–78.
Bhat Chandra R. (2005), "A Multiple Discrete-Continuous Extreme Value Model: Formulation and Application to Discretionary Time-Use Decisions," Transportation Research Part B: Methodological, 39 (8), 679–707.
Bhat Chandra R. (2008), "The Multiple Discrete-Continuous Extreme Value (MDCEV) Model: Role of Utility Function Parameters, Identification Considerations, and Model Extensions," Transportation Research Part B: Methodological, 42 (3), 274–303.
Bhattacharya C.B. (1998), "When Customers Are Members: Customer Retention in Paid Membership Contexts," Journal of the Academy of Marketing Science, 26 (1), 31–44.
Bhattacharya C.B., Rao Hayagreeva, Glynn Mary Ann. (1995), "Understanding the Bond of Identification: An Investigation of Its Correlates Among Art Museum Members," Journal of Marketing, 59 (4), 46–57.
Blanken Irene, van de Ven Niels, Zeelenberg Marcus. (2015), "A Meta-Analytic Review of Moral Licensing," Personality and Social Psychology Bulletin, 41 (4), 540–58.
Chiang Jeongwen. (1991), "A Simultaneous Approach to the Whether, What and How Much to Buy Questions," Marketing Science, 10 (4), 297–315.
Chintagunta Pradeep K. (1993), "Investigating Purchase Incidence, Brand Choice and Purchase Quantity Decisions of Households," Marketing Science, 12 (2), 184–208.
Chuan Amanda, Kessler Judd B., Milkman Katherine L. (2018), "Field Study of Charitable Giving Reveals That Reciprocity Decays over Time," Proceedings of the National Academy of Sciences, 115 (8), 1766–71.
DellaVigna Stefano, List John A., Malmendier Ulrike. (2012), "Testing for Altruism and Social Pressure in Charitable Giving," Quarterly Journal of Economics, 127 (1), 1–56.
Dew Ryan, Ansari Asim. (2018), "Bayesian Nonparametric Customer Base Analysis with Model-Based Visualizations," Marketing Science, 37 (2), 216–35.
Dew Ryan, Ansari Asim, Li Yang. (2020), "Modeling Dynamic Heterogeneity Using Gaussian Processes," Journal of Marketing Research, 57 (1), 55–77.
Dickert Stephan, Sagara Namika, Slovic Paul. (2011), "Affective Motivations to Help Others: A Two-Stage Model of Donation Decisions," Journal of Behavioral Decision Making, 24 (4), 361–76.
Fajardo Tatiana M., Townsend Claudia, Bolander Willy. (2018), "Toward an Optimal Donation Solicitation: Evidence from the Field of the Differential Influence of Donor-Related and Organization-Related Information on Donation Choice and Amount," Journal of Marketing, 82 (2), 142–52.
Falk Armin. (2007), "Gift Exchange in the Field," Econometrica, 75 (5), 1501–11.
Gelman Andrew, Rubin Donald B. (1992), "Inference from Iterative Simulation Using Multiple Sequences," Statistical Science, 7 (4), 457–72.
Gerber Elizabeth M., Hui Julie. (2013), "Crowdfunding: Motivations and Deterrents for Participation," ACM Transactions on Computer-Human Interaction, 20 (6), 1–32.
Gęsiarz Filip, Crockett Molly J. (2015), "Goal-Directed, Habitual and Pavlovian Prosocial Behavior," Frontiers in Behavioral Neuroscience, 9, 135.
Giving USA (2020), "Giving USA 2020: The Annual Report on Philanthropy for the Year 2019," technical report.
Gneezy Uri, Rustichini Aldo. (2000), "Pay Enough or Don't Pay at All," Quarterly Journal of Economics, 115 (3), 791–810.
Hanemann W. Michael. (1984), "Discrete/Continuous Models of Consumer Demand," Econometrica, 52 (3), 541–61.
Heckman James J. (1981), "Heterogeneity and State Dependence," in Studies in Labor Markets, Rosen Sherwin, ed. Chicago: University of Chicago Press, 91–140.
Hrywna Mark. (2019), "NPT Top 100 (2019): An In-Depth Study of America's Largest Nonprofits," The Nonprofit Times(November 4), https://www.thenonprofittimes.com/report/npt-top-100-2019-an-in-depth-study-of-americas-largest-nonprofits/.
Kallman Megan E., Clark Terry N. (2016), The Third Sector: Community Organizations, NGOs, and Nonprofits. Champaign: University of Illinois Press.
Khan Uzma, Dhar Ravi. (2006), "Licensing Effect in Consumer Choice," Journal of Marketing Research, 43 (2), 259–66.
Khodakarami Farnoosh, Petersen J. Andrew, Venkatesan Rajkumar. (2015), "Developing Donor Relationships: The Role of the Breadth of Giving," Journal of Marketing, 79 (4), 77–93.
Kim Jaewhan, Allenby Greg M., Rossi Peter E. (2002), "Modeling Consumer Demand for Variety," Marketing Science, 21 (3), 229–50.
Kumar V., Sharma Amalesh, Donthu Naveen, Rountree Carey. (2015), "Practice Prize Paper—Implementing Integrated Marketing Science Modeling at a Non-Profit Organization: Balancing Multiple Business Objectives at Georgia Aquarium," Marketing Science, 34 (6), 804–14.
Kuppuswamy Venkat, Bayus Barry L. (2018), "A Review of Crowdfunding Research and Findings," in Handbook of Research on New Product Development, Golder Peter N., Mitra Debanjan, eds. Cheltenham, UK: Edward Elgar Publishing Section, 361–73.
Leliveld Marijke C., Risselada Hans. (2017), "Dynamics in Charity Donation Decisions: Insights from a Large Longitudinal Data Set," Science Advances, 3 (9), e1700077.
List John A. (2011), "The Market for Charitable Giving," Journal of Economic Perspectives, 25 (2), 157–80.
Malani Anup, Philipson Tomas, David Guy. (2003), "Theories of Firm Behavior in the Nonprofit Sector. A Synthesis and Empirical Evaluation," in NBER Chapters. Cambridge, UK: National Bureau of Economic Research, Inc, 181–216.
Manchanda Puneet, Rossi Peter E., Chintagunta Pradeep K. (2004), "Response Modeling with Nonrandom Marketing-Mix Variables," Journal of Marketing Research, 41 (4), 467–78.
Meer Jonathan, Rosen Harvey S. (2011), "The ABCs of Charitable Solicitation," Journal of Public Economics, 95 (5), 363–71.
Mehta Nitin, Chen Xinlei J., Narasimhan Om. (2010), "Examining Demand Elasticities in Hanemann's Framework: A Theoretical and Empirical Analysis," Marketing Science, 29 (3), 422–37.
Midlarsky E., Hannah M. E. (1989), "The Generous Elderly: Naturalistic Studies of Donations Across the Life Span," Psychology and Aging, 4 (3), 346–51.
Morgan John. (2000), "Financing Public Goods by Means of Lotteries," Review of Economic Studies, 67 (4), 761–84.
Nair Harikesh, Dubé Jean-Pierre, Chintagunta Pradeep. (2005), "Accounting for Primary and Secondary Demand Effects with Aggregate Data," Marketing Science, 24 (3), 444–60.
National Council of Nonprofits (2019), "Nonprofit Impact Matters: How America's Charitable Nonprofits Strengthen Communities and Improve Lives," technical report (accessed February 24, 2021), https://www.nonprofitimpactmatters.org/site/assets/files/1/nonprofit-impact-matters-sept-2019-1.pdf.
Neslin Scott A., Taylor Gail A., Grantham Kimberly D., McNeil Kimberly R. (2013), "Overcoming the 'Recency Trap' in Customer Relationship Management," Journal of the Academy of Marketing Science, 41 (3), 320–37.
Netzer Oded, Lattin James M., Srinivasan V. (2008), "A Hidden Markov Model of Customer Relationship Dynamics," Marketing Science, 27 (2), 185–204.
Newman George E., Shen Y. Jeremy. (2012), "The Counterintuitive Effects of Thank-You Gifts on Charitable Giving," Journal of Economic Psychology, 33 (5), 973–83.
Rasmussen Carl E., Williams Christopher K.I. (2005), Gaussian Processes for Machine Learning. Cambridge, MA: MIT Press.
Roberts S., Osborne M., Ebden M., Reece S., Gibson N., Aigrain S. (2013), "Gaussian Processes for Time-Series Modelling," Philosophical Transactions: Mathematical, Physical and Engineering Sciences, 371 (1984), 1–25.
Sargeant Adrian, Woodliffe Lucy. (2007), "Gift Giving: an Interdisciplinary Review," International Journal of Nonprofit & Voluntary Sector Marketing, 12 (4), 275–307.
Satomura Takuya, Kim Jaewhan, Allenby Greg M. (2011), "Multiple-Constraint Choice Models with Corner and Interior Solutions," Marketing Science, 30 (3), 481–90.
Sethuraman Raj, Tellis Gerard J., Briesch Richard A. (2011), "How Well Does Advertising Work? Generalizations from Meta-Analysis of Brand Advertising Elasticities," Journal of Marketing Research, 48 (3), 457–71.
Simpson Brent, Willer Robb. (2008), "Altruism and Indirect Reciprocity: The Interaction of Person and Situation in Prosocial Behavior," Social Psychology Quarterly, 71 (1), 37–52.
Stiffman Eden. (2020), "Giving was up 7.5% in the First Half of 2020," The Chronicle of Philanthropy(October 6), https://www.philanthropy.com/article/giving-was-up-7-5-in-the-first-half-of-2020-new-report-says.
Touré-Tillery Maferima, Fishbach Ayelet. (2017), "Too Far to Help: The Effect of Perceived Distance on the Expected Impact and Likelihood of Charitable Action," Journal of Personality and Social Psychology, 112 (6), 860–76.
Tuchman Anna E., Nair Harikesh S., Gardete Pedro M. (2018), "Television Ad-Skipping, Consumption Complementarities and the Consumer Demand for Advertising," Quantitative Marketing and Economics, 16 (2), 111–74.
Van Diepen Merel, Donkers Bas, Franses Philip H. (2009), "Dynamic and Competitive Effects of Direct Mailings: A Charitable Giving Application," Journal of Marketing Research, 46 (1), 120–33.
Weisbrod Burton A. (1998), To Profit or Not to Profit: The Commercial Transformation of the Nonprofit Sector. Cambridge, UK: Cambridge University Press.
White Katherine, Habib Rishad, Dahl Darren W. (2020), "A Review and Framework for Thinking about the Drivers of Prosocial Consumer Behavior," Journal of the Association for Consumer Research, 5 (1), 2–18.
Winterich Karen, Mittal Vikas, Ross William T.Jr. (2009), "Donation Behavior Toward In-Groups and Out-Groups: The Role of Gender and Moral Identity," Journal of Consumer Research, 36 (2), 199–214.
~~~~~~~~
By Sungjin Kim; Sachin Gupta and Clarence Lee
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 86- Managing the Versioning Decision over an App's Lifetime. By: Lee, Seoungwoo; Zhang, Jie; Wedel, Michel. Journal of Marketing. Nov2021, Vol. 85 Issue 6, p44-62. 19p. 2 Diagrams, 8 Charts, 2 Graphs. DOI: 10.1177/00222429211000068.
- Database:
- Business Source Complete
Record: 87- Marketing Agility: The Concept, Antecedents, and a Research Agenda. By: Kalaignanam, Kartik; Tuli, Kapil R.; Kushwaha, Tarun; Lee, Leonard; Gal, David. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p35-58. 24p. 1 Color Photograph, 5 Charts. DOI: 10.1177/0022242920952760.
- Database:
- Business Source Complete
Marketing Agility: The Concept, Antecedents, and a Research Agenda
Changes in the way customers shop, accompanied by an explosion of customer touchpoints and fast-changing competitive and technological dynamics, have led to an increased emphasis on agile marketing. The objective of this article is to conceptualize and investigate the emerging concept of marketing agility. The authors synthesize the literature from marketing and allied disciplines and insights from in-depth interviews with 22 senior managers. Marketing agility is defined as the extent to which an entity rapidly iterates between making sense of the market and executing marketing decisions to adapt to the market. It is conceptualized as occurring across different organizational levels and shown to be distinct from related concepts in marketing and allied fields. The authors highlight the firm challenges in executing marketing agility, including ensuring brand consistency, scaling agility across the marketing ecosystem, managing data privacy concerns, pursuing marketing agility as a fad, and hiring marketing leaders. The authors identify the antecedents of marketing agility at the organizational, team, marketing leadership, and employee levels and provide a roadmap for future research. The authors caution that marketing agility may not be well-suited for all firms and all marketing activities.
Keywords: agile marketing; experimentation; iteration; marketing agility; sensemaking; speed
The digital transformation of enterprises, emergence of new channels (e.g., social media, mobile devices), and deluge of customer data are altering the practice of marketing.[ 6] The way customers shop and interact with brands has changed considerably in recent years ([105]), and researchers increasingly view shopping as a customer journey rather than a linear path to purchase ([58]). The challenges faced by marketing managers are highlighted by the COVID-19 pandemic, with some analysts calling for a fundamental rethinking of marketing models ([13]).
In response, scholarly research has advanced the need for new and flexible organizational models and recognized that marketing needs to be "agile" ([59]; [77]). Marketing agility (MA) is, in fact, viewed as a key priority for achieving marketing excellence ([45]). Chief marketing officers (CMOs) also emphasize the importance of MA. For example, Theresa McLaughlin, CMO of TD Bank, notes,
Our focus as a marketing organization is on creating experiences that add value for customers throughout their entire journey. Based on what we've learned so far, agile could play a significant role in helping us deliver on that. ([83])
Similarly, James Lyski, CMO at CarMax, the leading automotive retailer for used cars, underscores the importance of an agile approach to marketing:
If you wait till you perfect your product, you're behind the innovation curve. We use an agile development model where teams are constantly iterating. Most of the experiments they do are designed to fail. You learn a lot more from failure than success. ([ 9])
Despite the increasing importance ascribed to MA, it is not clear what it actually is. Is it the application of agile principles to marketing from other areas, such as manufacturing and software development? Is it a key priority of marketing excellence that relies on simplified structures to pursue organic growth through a test-and-learn approach ([45])? Is MA a fundamentally new idea? The first objective of this article is to review prior research and synthesize it with in-depth managerial interviews to propose a definition of MA and compare it with extant constructs.
In addition to the lack of a common understanding of MA, there are also growing concerns about its efficacy. For instance, a recent report recently surmised that agility is not a silver bullet, and its benefits are not realized unless "applied for the right reasons, in the right places, and in the right way" ([ 4]). Other analysts warn that MA is not the right fit for certain situations such as long sales cycles ([ 1]). The second objective of this article is to identify the potential downsides of and execution challenges associated with MA.
There is also a paucity of insights about the factors that enable firms to pursue MA. For example, a survey by the Boston Consulting Group finds that even though nine of ten marketing executives feel that agility is important for the marketing function, only one of five considers their firm to be agile ([108]). Reflecting the urgency of this challenge, the Marketing Science Institute (MSI) has identified "Organizing for Marketing Agility" as a key research priority for 2018–2020. The third objective of this article is to draw on insights from in-depth interviews and extant literature to identify what is known about the key organizational-, team-, marketing leadership–, and marketing employee–related antecedents of MA, and to propose directions for future research on MA.
[45] identify MA as a key dimension of marketing excellence and conceptualize it "as a firm's strategic means for executing growth activities by the marketing organization and its members through simplified structures and processes, fast decision making, and trial and error learning." (p. 10). Building on this definition, we first review extant literature to understand how agility is conceptualized both at the organizational and the functional levels in related disciplines.[ 7] In addition, we evaluate constructs related to adaptability, speed, and iteration that are conceptually similar to MA (see Table 1).
Graph
Table 1. Comparing Marketing Agility with Related Constructs.
| | Emphasis On... |
|---|
| Construct | Definition | Marketing Decisions | Sensemaking | Speed | Iteration |
|---|
| Marketing Concepts Related to Marketing Agility |
| Adaptive marketing capabilities | "Vigilant market learning, adaptive experimentation, and 'open' marketing that mobilizes dispersed and flexible partner resources" (Day 2011, p. 188) | Yes | Yes | No | Yes |
| Market-focused strategic flexibility | "The firm's capabilities and intent to generate firm-specific real options for the configuration and reconfiguration of appreciably superior customer value propositions" (Johnson et al. 2003, p. 77) | Yes | Yes | No | No |
| Market orientation | "Organization-wide generation of market intelligence pertaining to current and future customer needs, dissemination of the intelligence across departments, and organization-wide responsiveness to it" (Kohli and Jaworksi 1990, p. 6) | Yes | Yes | No | No |
| Market-based organization learning | "A core competency pertaining to external foci...[that] is less visible than most internally focused organizational learning competencies" (Sinkula 1994, p. 37) | Yes | Yes | No | No |
| Agility in Other Organizational Domains |
| Strategic agility | "The ability to exploit, or create to one's advantage changing patterns of resource deployment in a thoughtful and purposeful but also fast and nimble way rather than remain hostage to preset plans and existing business models" (Doz 2020, p. 1) | No | Yes | Yes | Yes |
| Organizational agility | "A firm's ability to cope with rapid, relentless, and uncertain changes and thrive in a competitive environment of continually and unpredictably changing opportunities" (Lu and Ramamurthy 2011, p. 932) | No | Yes | Yes | Yes |
| Agile manufacturing | "A manufacturing paradigm that focuses on smaller scale, modular production facilities, and agile operations capable of dealing with turbulent and changing environments" (Cao and Dowlatshahi 2005, p. 531) | No | Yes | Yes | Yes |
| Supply chain agility | "Supply chain's capability to adapt or respond in a speedy manner to a changing marketplace environment" (Swafford, Ghosh, and Murthy 2006, p. 172) | No | Yes | Yes | Yes |
| Software development agility | "A software team's ability to efficiently and effectively respond to user requirement changes" (Lee and Xia 2010, p. 88) | No | Yes | Yes | Yes |
| Organizational Concepts Related to Marketing Agility |
| Dynamic capabilities | "The firm's processes that use resources—specifically the processes to integrate, reconfigure, gain and release resources—to match and even create market change. Dynamic capabilities thus are the organizational and strategic routines by which firms achieve new resource configurations as markets emerge, collide, split, evolve, and die" (Eisenhardt and Martin 2000, p. 1107) | No | Yes | Yes | No |
| Ambidexterity | "An organization's ability to be aligned and efficient in its management of today's business demands while simultaneously being adaptive to changes in the environment" (Raisch and Birkinshaw 2008, p. 375) | No | Yes | No | No |
| Improvisation | "The degree to which composition and execution converge in time" (Moorman and Miner 1998, p. 698) | No | Yes | Yes | Yes |
| Design thinking | "A creative and strategic process characterized by the following hallmarks: abductive reasoning, iterative thinking and experimentation, holistic perspective, and human-centeredness." (Beverland, Wilner, and Micheli 2015, p. 593) | Yes | Yes | No | Yes |
We complement the received view on agility with in-depth interviews of 22 managers involved in marketing, brand and product management, analytics, and consulting roles with experience in agile approaches to marketing (see [117]). The field perspective allows us to tap into insights related to the practice of MA. The interviews lasted between 30 and 90 minutes each, and the managers had an average of 20 years of work experience and represented a wide distribution of roles and industries (see the Appendix). Synthesizing the received view with managerial interviews, we propose the following definition of MA:
Marketing agility refers to the extent to which an entity rapidly iterates between making sense of the market and executing marketing decisions to adapt to the market.
The proposed definition of MA complements the existing view by offering a more granular and process-based perspective. The starting point of this process is sensemaking of market developments to quickly assess the need for a marketing decision, receive feedback, and iterate between sensemaking and marketing decisions (see Figure 1). We propose that MA is a unique combination of four key concepts: sensemaking, iteration, speed, and marketing decisions. We next elaborate on each of these four key concepts.
Graph: Figure 1. Marketing agility: The construct, antecedents, and execution challenges.
At its core, sensemaking is "built out of vague questions, muddy answers, and negotiated agreement that attempt to reduce confusion" ([112], p. 636). It is triggered when "members confront events, issues, and actions that are somehow surprising or confusing" ([66], p. 21)—a characteristic of the contemporary marketing manager's operating environment. As such, we propose that sensemaking is a critical conceptual pillar of MA. Consider the following example:
When New York's taxi drivers went on strike for an hour at John F. Kennedy Airport from 6:00 to 7:00 p.m. in protest of the "Muslim travel ban," Uber responded at 6:30 p.m. by turning off surge pricing at the airport, with the stated intention of informing customers that they have travel options at normal prices ([23]). However, this change in pricing quickly escalated into a crisis. Customers voiced their criticism of Uber's behavior using messages such as "Congrats to @Uber_NYC on breaking a strike to profit off of refugees being consigned to Hell. Eat shit and die." Soon, the hashtag #DeleteUber turned into a vast protest marked by thousands of angry, emotional tweets ([20]), resulting in 400,000 users deleting their Uber accounts ([12]). In response to the customer backlash, Uber dedicated $3 million for the legal defense of drivers that were affected by the government policy ([49]).
At its heart, sensemaking is an entity's response to an unexpected or ambiguous development that involves noticing and bracketing the development, establishing a shared understanding of the development, and attempting to create a more ordered environment to draw further cues ([67]). In this sense, Uber's conduct exemplifies sensemaking as an integral part of MA by highlighting its key elements. Uber noticed an unexpected development (the strike by New York City taxi drivers), formed a potential understanding of the development (risk of being perceived as opportunistic due to higher surge prices), rapidly (within 30 minutes from the start of the strike) made a marketing decision (switched off surge pricing), iterated (by learning from customer backlash), and made another decision (allocating $3 million in legal defense for people affected by the travel ban). It is important to note that sensemaking, per se, does not imply that the resulting marketing decision is appropriate.
Consistent with extant views (see Table 1), the field interviews identified iteration as a conceptual pillar of MA. As noted by the vice president of marketing at a medical equipment manufacturer, MA is
the ability of any firm to be able to really quickly identify any initiatives, be able and nimble to execute them, get the feedback, and refine the initiative....The rapid evolution and iterative process to perfection (if there is such a thing!) of marketing agility provides the company all the ammunition it needs to tackle the ever-changing market landscape.
Iteration implies repeatedly refining marketing decisions before relaunching or scaling them. Iteration, therefore, is quite different from being guided by a deliberate plan and implementing "preorchestrated" marketing decisions—similar to organizational improvisation ([79]). In this sense, executing iteratively or the "small-bets" approach characterizes MA. Iterations, therefore, enable marketing managers to better match the changing needs of the market. By minimizing up-front risk and recognizing that change is an ongoing phenomenon, agile marketing entities can pivot and pursue a new task if feedback suggests the need for further adaptation. Importantly, MA iterations can also reveal that pursuing a new direction is a bad idea. As noted by the digital lead at a Fortune 100 firm,
In one of our recent campaigns, we decided to go completely against the grain in our marketing approach and do something new, something more social and integrated with seed marketing. The first week of the campaign went very well, but the second week saw a drop, as sales were not coming. There was no precedence of this pattern so we had to quickly decide whether we stick to the plan or go back to the old approach. We were getting a lot of awareness, but our competitors had a heavily discounted offering, and our efforts for awareness in the category actually ended up helping our competitor. It was a very hard decision, but we had to respond and change from the cool new spending to the old-fashioned approach of providing discounts through the channel.... The point was that we tried something new, it worked well initially, then it did not, we learnt from it quickly, and responded to it quickly too.
Both extant literature and field interviews identify speed as a key facet of agility (see Table 1). As observed by the director of marketing at an Asian retailer,
Marketing agility is about being able to adapt quickly to one's environment and to be sufficiently flexible in one's marketing strategy. It is especially important in today's world, as technology has changed a lot how we do things. The world is revolving so rapidly that things that used to take ten years to change are now taking only ten months.
In the case of MA, speed refers to the time taken by firms to sense market changes, initiate actions, gather feedback, and adjust marketing decisions. As stated by the vice president of marketing at a medical equipment firm, "The basic tenet of marketing agility is based on fast decision making on the best available information at the time." Elaborating on the importance of speed, a senior product manager contrasted MA with more traditional approaches:
Gone are the days when marketing strategy and associated marketing spend were decided at the beginning of a year and never discussed till the end of the year. We live in a very agile world now where we have to make rapid and relentless adjustments to our strategy. The rapid pace with which we move to alter our strategy more toward customer needs is what we call marketing agility.
Although more obvious, the final distinguishing feature is the centrality of marketing decisions in MA. Our interviews suggested that MA occurs across marketing areas. For example, a senior product manager underscored the importance of agility in advertising spending:
We see lots of companies are making quick adjustments to their ad strategy and spend to win highly competitive ad space. The firm that sticks with the traditional mode of marketing is not going to get maximum benefit out of their marketing spends, whereas an agile shop is going to leave everyone behind when it comes to [return on investment] of their marketing spend.
Marketing agility also occurs in product development. As noted by a senior product manager at an automobile retailer,
Agile marketing is essentially a way in which an agile product team work through concepts and requirements to figure out the fastest and most efficient way to release a feature or functionality to the customer that helps either meet a need, test out a concept, and collect learnings that can be iterated and improved upon. It is a way to quickly gauge if an idea has potential without getting too deep into the weeds or devoting too much development resources.
In addition to the range of marketing decisions, field interviews revealed two nuances. First, participants observed that while MA nearly always involves learning, it does not always prompt action. In the words of a country manager at a financial services firm,
There are times when you have to be agile in terms of listening and sensing, but not to respond. For us, it's brand management and social marketing; you have to be more nuanced and careful, as speed is not your ally because sometimes you have limited visibility on the background developments. As such, in these instances where you have limited visibility or there are too many factors at play and it is difficult to anticipate, then it might be better to be agile in terms of listening, sensing, and learning, as opposed to responding or taking any specific action.
Similarly, the vice president of corporate communications at a public infrastructure firm highlighted the importance of balancing actions/activities with a wait-and-watch approach:
More traditional marketing (vs. agile marketing) involves preplanned campaigns for products/services guided by a monthly calendar. The contrast is agile marketing, which involves constant monitoring of what's happening around the firm (e.g., actions of one's competitors), and seizing the opportunity to tweak one's own campaigns.... Rather than just acting (and reacting to changing circumstances), it is important to watch and see, to understand one's brand, and to be cognizant of the entire ecosystem so as to know where is the sweet spot to intervene, or simply to wait.
Second, marketing decisions are not only about responding to market developments; iterative probing can also lead to proactive decisions. The digital lead of a Fortune 100 firm specifically noted the need to be proactive:
It has to be a balance of reactive and proactive. It can't be either only. If you are not responding, then you are not being customer centric. If you are not being proactive, then you are only following.
Agile firms thus pursue both reactive and proactive marketing decisions and also recognize that in certain situations the best decision could well be a quick decision to not act.
The proposed definition of MA uses the term "entity" to imply that this construct can occur at the level of an organization, a marketing team, a marketing leader, and an marketing employee. It is nonetheless true that agility, in its basic form, is often observed at an individual level. In common parlance, "agility" is used to describe the nimble movements of athletes, warriors, and animals. A common criticism of organizational research is that by ascribing individual attributes to collective entities (e.g., teams, business units, organizations), one commits fallacies of anthropomorphism or personification ([110]).
In the abstract, however, constructs are indifferent to level-of-analysis because they serve as a "shorthand for a variety of phenomena that can be posited at any hierarchical level" ([82], p. 251). As such, we submit that MA as a collective construct is theoretically meaningful. A multilevel conceptualization of MA allows us to develop a richer understanding of the micro, meso, and macro behaviors associated with it.
Sensemaking at the individual, team, and organizational levels poses different challenges. Individuals make sense of "what is going on" using frames and schemas that often vary based on their functional backgrounds and experience ([42]). Sensemaking in teams implies developing shared frames or jointly ascribing meaning. For agile marketing teams, sensemaking implies bridging the different "thought worlds" of employees from varying functional backgrounds. For example, [74] note that, despite using similar terms and sharing new product development goals, marketers and designers may attribute very different meanings to the same words. Thus, agility as a collective and multilevel construct is not simply the aggregation of individual or team agile actions.
Marketing agility is conceptually related to several constructs in marketing and related disciplines. Table 1 provides the definitions of the related constructs and outlines their similarities and differences with MA. Agility in marketing differs from agility in other functions (e.g., software development, supply chain) because of its focus on marketing decisions (see Table 1). We next briefly outline how MA differs from related concepts in marketing and other disciplines.
We posit that MA is related to four key marketing constructs: adaptive marketing capabilities, market-focused strategic flexibility (MFSF), market orientation, and market-based organizational learning. Table 1 highlights the conceptual overlaps between MA and these marketing constructs. We note that the unique combination of four conceptual pillars differentiates MA from these constructs. For example, whereas MA explicitly emphasizes speed of marketing decisions, adaptive marketing capabilities emphasize vigilant learning but not speed. In addition, while adaptive marketing capabilities emphasize mobilization of dispersed and flexible partner resources, these aspects are not defining features of MA. It is also distinct from MFSF, as MA is focused on rapidly making sense of market developments and executing marketing decisions in an iterative manner; conceptualizations of MFSF, however, do not focus on these aspects. Furthermore, MA also differs from market orientation due to its explicit focus on speed and iteration in executing marketing decisions. Finally, while MA emphasizes iterative and rapid sensemaking and speedy execution of marketing decisions, market-based organizational learning does not emphasize these aspects.
Marketing agility is also related to higher-level constructs such as dynamic capabilities, improvisation, ambidexterity, and design thinking in other disciplines. For example, by emphasizing marketing decisions and iteration, MA differs from dynamic capabilities, even though both emphasize sensemaking and speed. Similarly, while both MA and design thinking emphasize iterations and experimentation, MA differs from design thinking due to its emphasis on speed (see Table 1).
In summary, while the MA construct is conceptually related to extant marketing and organizational constructs, it also possesses distinctive characteristics that make it theoretically rich. Arguing that there are no differences between organizational constructs that have some overlapping content would be akin to indulging in the "shades of gray" fallacy ([115]).
Although MA facilitates rapid adaptation to the market, pursuing and sustaining MA also involves potential pitfalls. In this section, we elaborate on some of the challenges associated with executing MA, including those related to brand building, the marketing ecosystem, data governance issues, MA as a fad, and marketing leaders. We also outline the criteria for evaluating which marketing activities are likely to benefit from an agile approach and point to future research opportunities.
Consistency of brand image is critical for maintaining strong brand associations in customers' minds. [85] note that the pressure to emphasize agile marketing over core brand values is a balancing act for brand managers as it is essential to ensure consistency of a brand's core identity. Across the interviews, several participants alluded to MA hurting customer attitudes and perceptions toward the brand. The vice president of corporate communications at a public infrastructure company cautioned,
It is important to be transparent and authentic to the world. Let the firm's core be the Northern Star instead of just jumping on the bandwagon. The core values of the brand are what's most endearing to consumers. Brands have to think long-term, sensing and listening, and brands have to take the driver seat in steering sales and customer management, rather than letting external forces dictate what they do.
The possibility that agile marketing actions could hurt brand meaning points to a paradox faced by marketing leaders and teams. From a sales perspective, agile marketing helps drive growth. However, from a broader, long-term perspective, fuzzy brand associations—a potential by-product of frequent experimentation—is potentially detrimental in product markets. These countervailing forces raise important future research questions for brand management, such as, What activities relating to brand management are amenable for MA? When do the costs of managing brand image and consistency exceed the benefits of MA? How can these costs be effectively managed?
Marketing agility might be difficult to pursue in mature industries such as consumer packaged goods (CPG) due to dependence on channel partners. For example, a senior marketing consultant we interviewed observed that it is challenging for Procter & Gamble to pursue MA in product development because of an 18-month advance notification that Walmart requires from all suppliers planning product launches. Similarly, product innovation in many industries requires close collaboration with suppliers. Thus, MA in product development is challenging unless the entire supply chain is on board.
The MA of firms that rely on the services of third-party vendors (e.g., advertising agencies, market intelligence providers) is also constrained by the speed with which the extended enterprise operates. Firms should account for this constraint when building their agency relationships. Who is willing and able to move as fast as they are? It is worth noting that this challenge may worsen if early adopters of agile marketing have the advantage of grabbing the partners that are committed to agility and leaving later entrants with partners unable or unwilling to transition to an agile environment ([25]).
In extending agility to the marketing ecosystem, there are formidable control and coordination problems to overcome, including monitoring, managing conflicts of interest, and ensuring accountability. These challenges raise important questions: What are the mechanisms available to extend MA to the marketing ecosystem? How should contract durations with partners be designed for agile marketing? While longer contracts with partners might help in developing shared interpretive frames that aid sensemaking, it could also prevent firms from flexing and changing rapidly as market conditions change. It seems that relational norms might play a crucial role in extending MA across partners. Future research needs to investigate the roles of formal versus informal mechanisms in extending MA to the marketing ecosystem as well as whether the efficacy of the mechanism is contingent on the nature of the marketing activity at hand (see the "Which Marketing Activities Are Likely to Benefit More from Agile Execution?" subsection).
As organizations rely increasingly on using customer data to make agile decisions, the trade-off between achieving speed and navigating data governance issues will become increasingly salient ([54]). Better decisions can be made with more comprehensive data that provide a 360-degree view of customers. However, in an era when data privacy concerns are assuming greater importance, the use of data for MA is likely to result in regulatory challenges. In 2018, U.S. firms experienced more than 1,200 data breaches in which 400 million customer records were compromised ([103]). The financial impact of data breaches is severe ([70]). Regulatory actions similar to the European Union's General Data Protection Regulation, the California Consumer Privacy Act, and Health Insurance Portability and Accountability Act laws could impede the easy access and use of data. It is conceivable that future privacy regulations could force organizations to rely on minimized and anonymized customer data. In the face of stricter regulations pertaining to data privacy and security, pursuing MA could be challenging.
At the same time, increasing pressure to innovate or launch marketing campaigns at a faster rate could tempt agile teams to overlook ethical concerns. A recent study by Deloitte Consulting found that less than half of chief executive officers (CEOs) are spending enough time and resources to manage privacy and ethical issues ([53]). To ensure that the execution of MA does not violate privacy concerns, ethical guardrails are needed. This could reduce MA, but not all agree that this will occur. The chief digital officer of MetLife likens ethical guardrails to car brakes ([53]): "People think brakes are to make a car slow down when, in fact, the purpose of brakes is that they enable cars to go fast. The same is true with organizational ethics." More research is needed to understand whether stricter regulations pertaining to data privacy and security make it difficult to pursue MA.
As reflected by the high level of interest in the business press, MA is quickly becoming corporate jargon. Yet, as with other organizational fads (e.g., quality circles, total quality management), there is a risk that firms will adopt the trappings of MA, but not its essence. For example, in an attempt to become agile, many corporations have tried to mimic Spotify's organization structure by adopting the easier, more superficial parts such as naming their teams "squads" and "tribes" ([102]).
"Cargo cult"[ 8] MA, or the pursuit of MA as an organizational fad, is likely to waste resources and yield poor outcomes. The superficial pursuit of MA is also a risk because the adoption of some of the cultural artifacts of MA may create an illusory sense of progress. It may also preclude a deeper and more difficult transformation. Future research on MA will need to develop training programs and organizational change approaches as well as metrics to clearly identify whether MA is operating and how it is contributing to firm performance.
A survey by the Association of National Advertisers finds that marketers and advertising agencies are growing increasingly frustrated by the difficulty of finding the right talent ([87]). The acute leadership talent shortage that firms are likely to face as they transition to MA reflects a confluence of many factors. The emergence of MA requires roles within organizations that did not previously exist. For example, roles such as social media and digital analytics managers require personnel that have not only the "hard skills" in data management and advanced analytics but also conceptual foundations of marketing such as brand management. Indeed, with the increased role of technology in marketing, agile marketing organizations find themselves competing for leadership talent with technology giants. The hiring of senior executives with technology backgrounds by firms such as Nike, Starbucks, and Chipotle to drive their consumer-direct digital strategies are cases in point ([29]). The paucity of managers with the mix of skills required by an agile marketing organization presents a unique challenge for smaller firms as they have limited resources. As such, a fruitful avenue for future research is to examine how small and medium-sized enterprises could overcome the challenge of recruiting marketing executives with the technology skills and marketing background to run an agile marketing organization.
In this subsection, we provide insights on which marketing activities might stand to benefit the most from MA. Building on the proposed definition, the challenges in executing MA and insights from the agility stream of research, we identify four criteria to assess whether a particular marketing activity is likely to benefit from agility: market response unpredictability, activity decomposability, customer validation, and ecosystem dependence (for definitions, see Table 2). We propose that a marketing activity is best suited for agility when the market response is highly unpredictable, activity can be broken down into smaller components, it is plausible to get customer validation, and there is less need to involve external partners.
Graph
Table 2. Assessing the Fit Between Marketing Activities and Marketing Agility.
| Marketing Activities | Evaluation Criteria | Benefits of MA |
|---|
| Market Response Unpredictability | Activity Decomposability | Customer Validation | Ecosystem Dependence |
|---|
| Definition: The extent to which the market is unpredictable with respect to customer preferences and competitor responses. The higher the unpredictability, the greater the benefits of agility (experimenting and iterating could ensure that the solution delivered meets the market needs). | Definition: The extent to which an activity can be broken down into smaller components. The greater the decomposability, the greater the benefits of agility (each component can be tested and iterated on independently). | Definition: The extent to which it is possible to get the concept at each iteration validated from customers. The greater the ability to validate, the greater the benefits of agility (validate mock-ups or prototypes from customer groups before investing valuable resources). | Definition: The extent to which the execution of an activity requires the involvement of external partners/agencies. The greater the degree of dependence on the marketing ecosystem, the lower the benefits of agility (greater dependence increases coordination costs). | |
| Content creation (e.g., advertising campaign, branding content, website design, social media campaign) | High. Response of customers to creative content is uncertain. | High. Ability to break down epic campaigns into smaller stories. | High. Ability to test and get feedback on user stories. | High. Typically requires the involvement of ad agencies. | The inherent risk of an improper message is high because of large up-front investments, so the benefits of MA are significant. |
| New product development (e.g., customer research for idea generation, concept development, concept testing) | High. Response of customers to product ideas is uncertain. | High. The product development process can be broken down into opportunity identification, prototype design, and testing before delivery. | High. Ability to test and get customer feedback on different concepts is high. | Moderate. Might require the involvement of suppliers to ensure feasibility. | Because consumer feedback is critical for success, and the up-front investments are large, the benefits of MA are significant. |
| Media buying (e.g., omnichannel marketing, ad word bidding, search engine optimization, YouTube and Facebook ad placement) | High. Response of customers to different media vehicles or platforms is uncertain. | High. Media can be purchased in small chunks. | High. Ability to test media's conversion effectiveness by examining customer response. | High. Requires the involvement of ad agencies. | The inherent risk of a wrong media mix is high because of significant up-front media costs, so the benefits of MA are significant. |
| Marketing strategy making (e.g., product life cycle planning, planned obsolescence, multibrand strategy, market entry strategies) | High. Response of markets to marketing strategy making is uncertain. | Low. It is difficult to break down strategic choices such as product life cycle planning, brand strategies, and market entry strategies. | Low. These are strategic issues that are guided by corporate objectives, and experimenting may not be appropriate. | Moderate. Might require the involvement of suppliers or external agencies in long-term planning. | Given the nature of these decisions in setting strategic direction, the need for control outweighs the benefits of agility (experimentation and validation). |
The rationale for selecting these criteria is as follows. The risk of large up-front investments is significantly more when the market response is likely to be unpredictable. Agile principles help reduce the risk by iterating through "smaller bets." Likewise, the ability to iterate and experiment is higher when an activity can be broken down into smaller components and validated independently. Similarly, agility is beneficial if it is possible to use customers to validate preliminary ideas before getting too "deep into the weeds." In some situations, however, customers may not be able to validate or provide feedback. Finally, if an activity requires the involvement and participation of the marketing ecosystem, the benefits of agility needs to be balanced with the increased need for coordination. In some situations, external partners might be unable to match the clock speed (i.e., short iterative cycles) of firms, making it difficult to execute an activity in an agile manner.
Using the aforementioned criteria, we assess the benefits of agile approaches for four categories of marketing activities: content creation, product development, media buying, and marketing strategy making. As outlined in Table 2, marketing activities such as content creation, product development, and media buying are likely to benefit the most from agile principles. This assessment is based on the observation that market response unpredictability increases the risk of large up-front investments in development (i.e., marketing campaign– or product-related). Furthermore, the ability to modularize these activities and test repeatedly with customer groups increases the likelihood of delivering the "right" messages or products through the "right" media platforms ([62]). In contrast, marketing strategy making entails a comprehensive set of activities such as situational analysis and considerations of long-term strategic direction ([73]). These activities are less suited for market experimentation. As such, we propose that decisions related to product life cycle planning, brand management, and market growth strategies (i.e., organic or mergers and acquisitions) are better suited for traditional planning and control techniques.
We next elaborate on the antecedents of MA. The antecedents of MA at each hierarchical level are likely to be distinct; as such, explicating the key issues and challenges at different hierarchical levels is necessary to advance theory. At the organizational level, MA is enabled by marketing technology (MarTech) factors, organization structure, organizational capabilities, the organization budgeting process, and organizational culture. The factors that drive MA at the leadership level are the CMO's background characteristics, CMO power, and the CMO–chief information officer (CIO) interface factors. Similarly, at the team level, MA is contingent on the autonomy available to teams, the diversity of teams in terms of their functional backgrounds and skills as well as more psychological factors such as superordinate identity and social cohesion. Finally, at the marketing employee level, MA depends on both the traits of employees as well as the training imparted to adapt to changing information. Table 3 outlines future research opportunities at the organizational and other hierarchical levels.
Graph
Table 3. A Roadmap for Future Research on Marketing Agility.
| Focus | Conceptual | Empirical |
|---|
| Organization | ✓ What are the organizational factors that drive or impede data virtualization for a firm? ✓ What factors drive firms to adopt tools (e.g., machine learning, AI) that are critical to drive MA? ✓ What kinds of formal and informal mechanisms are needed to drive coordination across teams and drive MA at an organizational level? ✓ What are the lower-level capabilities required to drive MA at an organizational level? ✓ How should organizational and marketing budgets account for the unique challenges presented by MA? ✓ How do firms balance the culture that facilitates MA with other facets of an organization that require more traditional business practices?
| ✓ What incentives can be used to facilitate data sharing and transparency across business units and functions? ✓ To what extent do marketing managers within and across organizations rely on automation tools such as machine learning and AI to make marketing decisions? What is the corresponding impact of these tools on a firm's MA? ✓ How should the organization structure be designed to coordinate between marketing teams? ✓ What are the impediments in executing the discovery and delivery tracks of MA in parallel? ✓ What are the metrics to measure the effectiveness of agile marketing efforts? ✓ What is the impact of MA on stock market returns and firm risk?
|
| Execution Challenges | |
| ✓ What are the implications of MA for brand management? ✓ Does the marketing ecosystem impede the pursuit of MA? ✓ Does the pursuit of MA exacerbate concerns over data privacy and security? ✓ Is there a risk of MA being pursued as an organizational fad?
| ✓ Does the pursuit of MA impair brand meaning? ✓ Do the challenges of coordinating with external partners outweigh the benefits of MA? ✓ What are the ethical guardrails needed to navigate data governance issues and make speedy marketing decisions? ✓ How should MA be measured to distinguish the true essence of MA from the superficial trappings?
|
| Team | ✓ What are the effects of diversity in the marketing team on MA? ✓ How should incentive systems balance performance appraisal of individuals within a MA team and across multiple MA teams? ✓ To what extent should incentives be based on outcomes versus behavior for agile marketing teams? ✓ How do superordinate identity and social cohesion interact in marketing teams, and what is the combined impact on MA? ✓ How can superordinate identity and social cohesion be cultivated in situations where physical proximity is difficult to achieve (e.g., telecommuting)?
| ✓ Does diversity in marketing teams aid sensemaking? ✓ How should teams be incentivized to encourage experimentation and avoid indirectly encouraging failure? ✓ Under what conditions should firms use permanent versus temporary teams to drive MA?
|
| Leadership | ✓ What attributes and characteristics are required for CMOs to drive MA? ✓ Are CMOs with sales and/or technology background likely to be better suited for MA? ✓ What enables CMOs to distinguish between true and false brand stories?
| ✓ Are CMOs with broader responsibilities able to make faster marketing decisions? ✓ What kind of power do CMOs need for MA? Is structural power more important than expert power for MA? ✓ What kind of skill-sets do CMOs and CIOs need to have to complement each other and pursue MA?
|
| Execution Challenges | |
| ✓ Is hiring of the "right" marketing leaders suited for MA difficult?
| ✓ How can marketing leaders with the right mix of marketing and technological skills be identified, hired and retained?
|
| Employee | ✓ What are the personality traits that are likely to drive the fit of an employee in a firm that embraces MA? ✓ What mechanisms can promote functional, as opposed to dysfunctional coping strategies for employees in agile marketing organizations? ✓ How should marketing managers be trained to thrive in an MA environment? ✓ Does MA pose challenges to business schools and their training methods?
| ✓ Are the big five personality dimensions likely to predict a marketing employee's performance in an agile marketing organization? ✓ Is marketing employee conscientiousness undesirable for MA? ✓ How should marketing employees navigate multiple identities such as belonging to a team versus belonging to a functional area? ✓ How should HR develop training programs to update the skills of marketing employees? Should HR iterative cycles be synchronized with marketing iterative cycles? ✓ How should the education curriculum of business schools be designed to train entry-level marketing employees for MA
|
While we conceptualize MA as a multilevel construct, it is likely to reside at the organizational level in the form of routines, processes, flexible structures, and cultural norms/values. Without organizational capabilities and structure, MA is likely to be fortuitous and not sustainable. For firms such as CarMax, MA is driven by the firm's superior ability to tightly integrate two distinct aspects of MA: discovering market opportunities and developing or delivering marketing campaigns or solutions. Likewise, for firms such as Spotify, MA is supported by an organization structure that enables knowledge sharing and integration. The structure also facilitates lateral communication between multiple teams, reduces conflicts, and enables sensemaking and speed at the organizational level. At the organizational level, MA implies developing shared interpretive frames across units, divisions, and teams ([42]; [67]). Moreover, cultural values and norms play an important role in creating shared interpretive frames at this level, although, in reality, sensemaking might entail order or meaning negotiated through compromise ([42]; [67]). We next discuss the role of each organizational antecedent, and for each antecedent, we offer a brief overview of relevant prior research and propose directions for future research.
The technological infrastructure and processes deployed for gathering and analyzing market information are broadly referred to as MarTech. Investments in MarTech aid CMOs, marketing employees, and teams to spot opportunities and trends, experiment, and respond to or drive changes in the market. Accordingly, we explore the MarTech characteristics that are relevant and critical to developing and sustaining MA.
"Data virtualization" refers to the ability of organizations to integrate data from disparate sources and bring structured and unstructured data from multiple sources into a unified, logically virtualized data layer for decision making ([26]). Much of the time and resources in a data-rich marketing environment are invested in data identification and "ETL" tasks (i.e., extraction, transformation, and loading of the required data). Inaccessibility of relevant and timely data is a significant barrier to pursuing MA. The best value from MarTech can be realized when the heterogeneous data sets are integrated and underlying relational patterns uncovered in a timely manner. However, as we have noted, these data sets typically exist in silos controlled by different functional groups. The data hoarding tendency creates data quality issues, as each functional group may rely on its own data sets for decision making. Data virtualization not only enhances the speed at which the data can be accessed, it also makes data democratic—that is, easily accessible to all employees.
With the availability of high-dimensional data and the added complexity of nonnumeric data such as images, text, video, and audio, the selection of the appropriate tool for driving decisions assumes greater importance ([106]; [111]). The choice of the right analytics tools for the type of data at hand is important in enabling firms to exploit the different types of data available to them. Data analytics tools that are geared toward analyzing structured data may not be suited for analyzing unstructured data ([36]). Firms are increasingly using machine learning and deep learning tools to address these limitations. According to a recent study, three out of four firms that have adopted these tools have, on average, experienced a 10% improvement in customer satisfaction owing to their ability to act rapidly to changing customer needs ([21]). These tools are often superior in their ability to handle a larger number of numeric and nonnumeric features as well as to process real-time data at significant speed. At the same time, a simple tool that can run a needed classification model or produce a specific report with a few clicks might be better suited for some marketing tasks than a complex tool. More generally, the fit between the data, tool, and task at hand is critical in improving the effectiveness of analytics at achieving agility ([36]).
There are several challenges related to MarTech that firms must confront. In a traditional marketing organization, the intelligence gathered through market research often resembles the "waterfall approach," in which a series of tasks such as objective definition, hypothesis formation, research design, data collection, data analysis, and report generation are performed sequentially ([ 7]). This approach can hinder the achievement of MA. The compartmentalized approach to intelligence generation creates silos: users of business intelligence (e.g., brand managers, marketing managers) on the one side and entities that generate the insights (e.g., the data science teams, data warehousing teams, data owners) on the other. Research has found that market intelligence usage is greater in informal organizations than in large formally organized firms ([27]). The siloed process limits feedback of marketers (end users), making the intelligence-gathering process less adaptable to rapid changes. Furthermore, the intelligence generated is often not viewed as a "shareable" resource, as groups or individuals hoard data.
There are two areas for future research. First, we need to understand the organizational impediments to achieving data virtualization and therefore MA. Previous research has shown that trust is the most significant predictor of market research utilization ([80]). Additional research is needed to understand the factors that engender trust, facilitate democratization of data, and discourage data hoarding. Research on how the incentives/rewards and metrics used to assess the performance of different functional units influence their willingness to share data is likely to be valuable. Second, while tools that leverage machine learning and deep learning principles are valuable for generating marketing micro campaigns, its actual use by marketers for decisions is not without challenges. Marketing employees are often averse to using algorithms that they do not understand ([28]). This reluctance raises important questions such as what are the managerial and organizational factors that impede the adoption and use of MarTech tools? In this context, increased transparency of the algorithms and increased feedback from marketing teams could increase trust and greater use of MarTech tools for rapid marketing decisions.
What are the organization structures that are well-suited for enabling MA? A key requirement of MA is that the autonomy of teams should be preserved and knowledge integration across the organization should be exploited. The organization design adopted by Spotify is a case in point. Spotify relies on a distinct organizational form to achieve MA in product development. Instead of a hierarchical or multidivisional organization structure that clusters employees based on functional expertise, Spotify uses small cross-functional teams (see Figure WA1 in the Web Appendix). The core organizational unit is an autonomous "squad" that is responsible for a discrete aspect of the product ([68]). For example, a squad might be responsible for a particular feature of the product, such as the display cover of an album, and comprises employees with skills needed to design, develop, test, and release. Each squad has a product owner who is responsible for the vision of the feature, prioritizing the product backlog and setting goals for each iteration.
At a higher level are "tribes," which are a collection of multiple squads working in related areas (e.g., music player, or backend infrastructure) ([68]). The members of squads within a tribe are often colocated, and each tribe has a lead who is responsible for creating the environment for its squads and extracting the best value from them. The organization structure of Spotify also facilitates lateral formal and informal communication (e.g., through alliances, chapters, and guilds; see the Web Appendix).
Fundamentally, flexible structures that promote cross-functional collaboration and cross-pollination of knowledge are crucial for enabling MA at an organizational level. While marketing is better equipped to probe and detect trends because of its boundary-spanning role, the actions required for adaptation might straddle several functional areas (e.g., marketing, manufacturing, supply chain, sales). As a result, a cross-functional team is the most disaggregate unit for executing MA. The precise composition of a cross-functional team depends on the nature of the task at hand. For example, a team tasked with redesigning the online customer experience of returning products could comprise employees with experience in marketing, sales, logistics/supply chain, and user experience designers. Similarly, a team tasked with price optimization could comprise employees with experience in analytics/artificial intelligence (AI)/machine learning tools, sales, logistics/supply chain, and marketing.
At an organizational level, because it is important to develop shared interpretive frames (i.e., sensemaking) of how the activities of different teams impact financial metrics, it is necessary that the organization structure enables frequent interactions and coordination. To this end, firms need to use both formal and informal coordination mechanisms to manage dependencies across teams, thereby allowing for MA at the organizational level. Formal coordination mechanisms refer to periodic meetings of multiple teams (i.e., "scrum of scrum meetings") to ensure that teams, while pursuing different proximal goals, share at least one common distal goal (i.e., consistency in marketing strategy). Informal coordination mechanisms refer to organic, casual, and personal communication between members across teams in the organization. The informal coordination mechanisms could foster communities of interest and leverage best practices. A natural question for future research, therefore, is what is the right balance between formal and informal coordination mechanisms to foster MA?
Marketing agility can be viewed as both improvisational and a dynamic- or higher-order capability that facilitates learning—what is sometimes called "learning to learn." What are the specific processes or routines that enable some organizations to be better at MA than their competitors? To offer ideas, we profile CarMax, a used car retailer that is well known for bringing an agile approach to marketing in automotive retail ([89]). CarMax uses self-directed product teams comprising seven to nine marketing, operations, and information technology (IT) employees.[ 9] While teams are presented with a problem, they are not instructed on how to solve the problem. The teams use a "test-and-learn" approach to ideate and deliver the best possible result.
However, CarMax's advantage appears to lie in its execution of MA—this is where it shows superior capabilities and learning ability. The company uses a dual-track approach for executing MA (see Figure 2). The first track, referred to as the "discovery track," focuses on ideating, prototyping, and validating ideas for the product or campaign, whereas the "delivery track" (or "development track") focuses on turning those ideas into an actual product or campaign. Although MA emphasizes iterations and speed, time gaps between iterations are possible if the ideas are poorly defined or not validated. The uniqueness of MA at CarMax is that the firm allows the discovery and delivery tracks to operate in parallel. A key benefit of the discovery track is that ideas are validated with prototypes or mockups before they are developed or built. As a result, expensive failures are avoided. In particular, CarMax validates ideas by gathering feedback on whether customers find the ideas to be usable or valuable. CarMax involves the development or delivery employees in discovery sprints to ensure that the delivery personnel can actually build the product or execute the campaign.
Graph: Figure 2. An illustration of dual-track marketing agility at CarMax.
The validated ideas from the discovery track are prioritized and essentially become the product backlog or the storyboard for the delivery track. The delivery track works on the validated ideas or stories from the backlog and performs its sprints on testing and learning about usability and getting the features right. The ability to integrate discovery and delivery/development has enabled CarMax to increase the speed of delivering products and messages validated by the market. As a result, the risk of the campaign or product failing is minimized to a great extent. Importantly, CarMax pursues agility in both product development (e.g., remote appraisal tool, online home delivery tool) and advertising or marketing campaigns (e.g., with external agencies for marketing messages).
We encourage future research to further investigate the types of lower-level capabilities that aid (or impede) the discovery and delivery processes of MA. In a digital, social media, and mobile marketing environment, the ability to capture and apply data from a variety of venues to sustain the ideas storyboard in the discovery track is potentially a specialized capability. Without a steady flow of stories and ideas in the discovery track, the delivery storyboard or backlog could dry up and the speed of marketing decisions could diminish considerably. Similarly, the ability to correctly prioritize ideas in both discovery and delivery backlogs could be a specialized process capability that helps in further increasing the speed of executing the "right" marketing ideas. Finally, a crucial aspect of MA is validating ideas before expensive resources are committed in the delivery track. Thus, the ability to identify the "right" users or customers to validate ideas generated in the discovery track could also be a distinct lower-level capability.
An important area of inquiry is understanding the importance of funds and resources to sustain MA. If marketing budgets continue to be developed using a traditional marketing budgeting process where resources are often tied to channel-, product-, and market-specific objectives, MA is not feasible. We propose that an agile marketing budget, for example, should try to tie resources to goals (e.g., increase brand awareness by 1%, increase retention by 5%) and the business value they generate. The specific channel, product, and market to accomplish the goal should be tactical and should emerge on the fly. This change should enable marketing leaders to move resources fluidly across channels, products, and markets. However, we caution that tracking and measuring the performance of agile marketing efforts is likely to be challenging. For example, with the customer's path to purchase becoming nonlinear, "last click" is an unreliable metric for evaluating the success of individual digital marketing campaigns. It is imperative, therefore, that the measurement of agile marketing reflect a clear understanding of the role of different campaigns at different stages of the customer's purchase journey and how they interact with each other. Against this background, the following research questions are promising and worthy of future research: ( 1) How should agile marketing budgets be developed? ( 2) What are the measures for evaluating the performance of MA campaigns? and ( 3) How are these measures related to business value?
The marketing literature has identified organizational culture as being manifest, inter alia, through shared values, beliefs, and norms (for a review, see [78]]). For example, [35] identify six values reflected in behavioral norms that are critical to a market-oriented culture: the market as the raison d'etre of the organization, collaboration, respect (and empathy), keeping promises, openness, and trust. Other research, examining learning cultures, has found that cultures that emphasize learning and development are associated with higher innovativeness ([47]). It is likely that the importance of these values would extend to an agile marketing culture as well given that MA can be viewed to subsume market orientation and to emphasize adaptation to the market through learning. However, the dimensions that distinguish MA, namely an emphasis on speed and the iterative nature of learning, imply that other cultural values will also be central to an agile marketing culture. In particular, we posit that these key values are likely to include speed and the embrace of uncertainty and discovery. These latter values map onto the dimensions of iteration and sensemaking of MA.
In addition to studying the values associated with market-oriented and learning cultures, the marketing literature has also examined how different cultures emerge at the intersection of two dimensions along which cultural values can be mapped: an external versus internal orientation and a favoring of formal versus informal processes (the "competing values framework"; [76]). Given the focus of MA on information acquisition we would expect an agile marketing culture to be relatively externally oriented; likewise, given the focus of MA on information flow, we would expect an agile marketing culture to depend on relatively informal processes.
Consistent with the posited importance of the aforementioned values to an agile marketing culture, our interviewees expressed the belief that MA is dependent on a culture that is not beset by established rules, procedures, or extensive up-front planning and control, but where executing tactics before they are fully fleshed out and in the face of incomplete information is embraced and seen as a basis for learning. For instance, a senior vice president of analytics at a B2B services firm stated that "agile requires a culture of building an experimental mindset." Other executives evoked the importance of a culture where acting quickly in the absence of complete information is prized by contrasting it with the culture associated with a more traditional marketing organization. For instance, when contrasting between the culture of a traditional CPG marketing organization with the culture required for MA, a senior marketing consultant remarked,
Most mature companies in the CPG space like to have consistent repeatable processes to grow the business. For example, [a CPG firm] has a lot of systems in place to repeat innovation. This is often slow. It builds rigidity and makes it hard to deviate from the process. Agility is tough for mature CPG businesses.
Reflecting on the importance of values and norms establishing the embrace of uncertainty and learning, the digital lead at an IT product and services firm commented,
If you have to become agile, you have to empower people that are capable, unafraid, and be willing to learn that you will make mistakes....For example, in our organization one of the key things we are working on is the language we use. Specifically, we don't call adverse outcomes as "failures," we call them "learnings."
In addition to being viewed in terms of values and norms, culture also takes the form of cultural artifacts ([44]; for review, see [78]]). Mirroring this view, agile practices have often been accompanied by artifacts such as daily, short (typically 15-minute) stand-up meetings (i.e., where participants actually stand during the meeting) and open work environments ([ 3]). Likewise, the terminology used to describe agile processes, such as "sprints" to refer to iterations and "scrums" to refer to meetings ([19]), reflect the value placed on speed and iteration. Accompanying titles, such as "scrum master" and "product owner," can also be viewed as artifacts. Furthermore, although not specific to agility, to encourage the embrace of acting under uncertainty, large organizations such as Procter & Gamble and Tata have instituted "heroic failure awards" and "dare to fail" awards ([81]).
A key future research priority is to investigate how an agile marketing culture can be created and sustained. [35] identify a four-stage organizational change process in the development of a market-oriented culture that starts with a mass mobilization led by senior management. Although the establishment of an agile marketing culture can likely arise via a similar process, it is plausible that an agile culture could emerge in a more piecemeal fashion. This question is important to consider because ( 1) agile practices can be manifest at lower levels than the whole organization, such as at the level of the individual project team, and ( 2) MA is likely to be more limited than market orientation in the breadth of its relevance to the organization or marketing function; to wit, some marketing processes (e.g., product life cycle planning, brand building, highly regulated activities) are not amenable to or would gain little value from rapid iteration and might even be harmed by it. Therefore, it would be interesting for future research to consider how an agile culture might initially arise at the level of individual teams or projects and subsequently spread outwards to other (relevant) parts of the organization. Such a process would not negate the importance of top management in fostering culture, but it might suggest a different process through which organizational transformation might occur than that identified by [35].
A related question for future research is how agile parts of the organization might interact with nonagile parts of the organization. To elaborate, because agility is not well-suited to some marketing activities (e.g., brand building, product life cycle planning) as well as to many nonmarketing activities (e.g., accounting, human resources [HR]), to the extent that elements of an agile marketing culture (e.g., valuing speed) bleed into these activities it might be harmful. At the same time, activities suited to agile marketing necessarily impinge on activities, such as brand building, that are less suited to agile marketing. Thus, organizations need to ensure that activities and processes that are not built to be agile are guided by individuals that understand agile marketing culture and have the ability to support and interface with agile marketing teams. In summary, how to manage the interactions between the agile and nonagile parts of the organization is an important question for future research.
In this section, we examine the team-level factors that are effective in driving and supporting MA. In particular, we discuss both organizational characteristics of teams (i.e., team composition and diversity, team empowerment, incentive structures) and team-related psychological factors (i.e., superordinate identity and social cohesion) that may play a role in fostering MA.
Cross-functional teams are a mainstay in marketing—from sales and advertising (e.g., [65]) to new product development (e.g., [41]) and customer relationship management (e.g., [90]). Previous research has suggested that diversity in teams facilitates rapid, real-time information exchange particularly when the tasks and technologies involved are complex ([16]; [22]). The availability of diverse viewpoints and enriched "schemas" may also enable teams to engage in improved sensemaking ([10]; [116]). Furthermore, the exchange and cross-fertilization of diverse knowledge and perspectives can spark creative ideas and processes, allowing teams to uncover and test novel marketing ideas ([37]).
As such, diversity is a critical facet for the ability of marketing teams to drive MA. However, too much diversity could lead to information overload, resulting in the reliance on simplifying heuristics (e.g., status quo bias) or falling back on more familiar decision-making processes ([ 5]; [107]). Although such reliance may allow decisions to be made more quickly, it could also result in less iterative learning, as well as less effective sensemaking. This is because, teams, driven by their desire for certainty and avoidance of complexity and ambiguity, suppress further (albeit necessary) enquiry and become tunnel-visioned ([116]).
The importance of team autonomy in enhancing performance may be traced back to the work of sociotechnical systems theorists, whose interventions frequently involved the creation of autonomous or self-managing work groups ([18]; [69]). From this perspective, increased autonomy for teams acts to reduce bureaucratic constraints, enabling team members to more effectively identify and respond to new situations. In marketing and sales, the empowerment of teams and employees is particularly advantageous when the issues at hand are not highly structured ([91])—conditions that may be expected in complex and uncertain situations where MA is vital. Indeed, field interviews indicate that for teams to adopt MA, they need distributed empowerment (i.e., the autonomy for accessing resources and making decisions is decentralized and assigned to team members as opposed to a designated leader within a team). Such empowerment of resource access and decision-making authority allows teams to respond to market forces and make marketing decisions promptly. As noted by a senior product manager in retailing,
They (i.e., agile firms) are wired to think in an agile fashion and they help teams achieve this by empowering them to come up with the solution that best addresses the pain point or fixes a problem. The "empowering" aspect helps a team act like a start-up and not be shackled in phase-gate type processes that act more as road blockers.
Notably, while broad organizational guidelines are essential to ensure that different teams follow the same organization-wide strategy consistently, it is also necessary to avoid the stipulation of overly specific task-level guidelines.
In the pursuit of MA, sensemaking and iterative learning are inevitably accompanied by their share of setbacks. As such, incentive structures that account for the high probability of failures are needed. Prior research on the role of risk-encouraging incentives in product innovativeness and the creativity of marketing programs (e.g., pricing) in product development teams suggests that incentives should encourage teams to take appropriate risks without penalizing them for minor failures ([48]; [98]). An incentive system that only rewards positive performance without protecting teams from taking risks and learning from failures is likely to inhibit MA. In the absence of an appropriate incentive system, teams may be overly fastidious in ensuring that they "dot all the i's and cross all the t's" before developing or designing minimally viable campaign or product ideas. Thus, incentives that promote risk taking are an important form of external stimulus that promotes an experimentation mindset and rapid iteration.
In addition to potentially exacerbating intrateam conflict, diverse teams (vs. individuals) are also prone to making polarizing decisions ([50]; [84]), rendering it difficult for them to respond swiftly to changing marketing situations. Information overload resulting from too much (perceived) functional diversity may lead teams to fall back on the use of heuristics and other familiar yet suboptimal decision models, which could have divergent effects on speed, iterations, and sensemaking. Thus, future work could examine the potentially opposing effects of team diversity on different dimensions of MA as well as brainstorm ways to help marketers better manage the resultant trade-offs that arise from these opposing effects.
Additional research is also needed on the type of incentive structures that organizations would have to design and implement to foster MA. In particular, how should rewards be distributed among team members, and on which criteria should these team rewards be based? To what extent should incentives depend on outcomes (to spur teams to make the best marketing decisions possible), and to what extent should they be process-based instead to encourage MA actions? Critically, how can firms encourage risk taking and experimentation within teams without also indirectly encouraging failure? More research is also needed on how incentives should balance short-term goals versus long-term goals and individual performance versus team performance.
Superordinate identity refers to the degree to which members identify with the team (and not only with their functional areas) and perceive an individual stake in the team's success ([100]). While teams with a low superordinate identity tend to retain their entrenched functional identities and biases, those with a high superordinate identity are more likely to perceive intrateam similarities and are more willing to accept the divergent attitudes and perspectives of members from other functional areas ([92]). Consequently, a high degree of superordinate identity in a team promotes greater project ownership and thus greater cooperation and more effective sensemaking, such that members can more openly and effectively integrate and construct from diverse functional perspectives within the team and make more cogent marketing decisions. At the same time, marketing decisions can be made more rapidly due to a greater sense of shared responsibility and higher motivation for greater responsiveness to changing marketplace circumstances. Therefore, the more the members of a team share a superordinate identity, the greater their MA is likely to be.
Social cohesion refers to the strength of interpersonal ties among team members, or how closely members bond with one another at a social level ([14]). [99] found an inverted U-shaped relationship between social cohesion and the innovativeness of new consumer products. In the same vein, [14] document mixed effects of interpersonal cohesion within teams on both external new product performance (profitability and market success) and internal team sentiments (team member satisfaction). While a high degree of social cohesion can foster superordinate identity, reduce conflict, and help teams create shared mental models and engage in tacit knowledge transfer ([46]), too much cohesion could lead to "groupthink" and consensus and conformance seeking ([51]). In the context of MA, although the latter effects could lead to faster marketing decisions, they could also impede sensemaking and inhibit improvisations.
As teams become more global, interactions more computer-mediated, and telecommuting more prevalent (especially in the post-COVID-19 "new normal"), physical proximity—an important antecedent to building superordinate identity and social cohesion—may be lacking ([92]). How can superordinate identity and social cohesion be effectively cultivated in cases where physical proximity is difficult to achieve, especially given that the longevity of teams is significantly compressed today ([39])? Relatedly, under what conditions would it be beneficial to have temporary versus more permanent cross-functional teams? On the one hand, permanent cross-functional teams should, in general, aid MA because stability may facilitate a team's learning and developing shared interpretive frames. On the other hand, temporary teams may be preferred for one-off tasks, especially if there are concerns that permanent teams could demonstrate a tendency for "groupthink" ([51]). Furthermore, it is also important to examine how teams can balance superordinate identity and the potential negative effects of social cohesion, given that the latter can limit the expression of dissenting views and the challenging of preexisting assumptions, thus inhibiting the discovery of novel linkages that are essential to drive MA.
In cross-functional teams, difficulties in coordination and learning are magnified because there is a need to bridge the thought worlds of different functions. Building on sensemaking, [10] propose a three-step "resource sensemaking" process—exposing, co-opting, and repurposing—to facilitate cross-functional perspective taking and help teams overcome functional barriers (e.g., between marketing and design) in new product development. Future research could investigate whether these processes help strengthen team cohesion and enable sensemaking within agile marketing teams.
In this section, we elaborate on factors that enable marketing leaders and employees to pursue MA. Specifically, we discuss the attributes of marketing leaders, the role of CMO power and the CMO–CIO interface issues that enable or inhibit MA. In addition, we highlight the role of marketing employee personality traits, training, and coping mechanisms to drive MA.
The role of leaders in driving MA cannot be overemphasized. Extant research suggests that marketing leaders are likely to need three sets of characteristics to drive MA ([31]). First, senior leaders need to have strategic sensitivity: the sharpness of perception of, and the intensity of awareness and attention to, strategic development. Second, there must be unity among senior leaders to make bold, fast decisions, without being bogged down in "win-lose" politics. Finally, senior leaders should have the ability to reconfigure capabilities and redeploy resources rapidly ([31]). In addition to these capabilities, marketing leaders face unique challenges. The role of a CMO is often ambiguous in organizations. For example, a "test-and-learn" mindset requires CMOs to be accommodative of far greater ambiguity in decision making ([113]).
The preceding discussion suggests that it is critical to identify the characteristics and attributes of a CMO that are likely to drive MA. Given that MA requires constant experimentation and refinement, it is plausible that CMOs with prior experience in sales are better suited to drive MA because sales personnel are frequently required to engage in adaptive selling to respond to the differing characteristics of prospects ([72]). At the same time, CMOs are also expected to make sense of the high volume of market and customer data and distinguish between true and false brand stories. Marketing responses to false brand stories or rumors need to be swift, as false news tends to propagate faster in a digital world due to its greater perceived novelty ([109]). It is plausible that some CMOs are more adept at understanding and using AI tools (e.g., scoring web pages, predicting the reputation of the source, using Heat AI for predictive sensing) to assess the plausibility of online brand stories. As such, CMOs with prior technology and analytics experience are better suited to drive MA.
The second area that warrants additional research is whether marketing leaders or departments have the requisite structural power to mobilize or redirect resources and therefore drive MA in their organizations ([33]). One indicator of a CMO's structural power is their level of compensation. Chief marketing officers are rarely among the organization's highest-paid executives ([94]), and this lack of CMO structural power might impede the process of attracting resources for MA. In addition to lower structural power, marketing executives also often have narrow lines of responsibility (i.e., lower expert power), increasing the need to coordinate with other executives and slowing down marketing decisions. Recent evidence indicates that corporate activities that once belonged to marketing are being taken over by more able, better-trained parts of the organization, and marketers are reduced to a narrow communications role instead of facing research, strategy, product, and pricing decisions ([95]). In this sense, the CMO's ability to get buy-in from the rest of the top management team leaders is instrumental in leading agile marketing to be embraced across the organization.
For example, it is noteworthy that the CMO of CarMax has a broad range of responsibilities and is also part of the executive officer team (i.e., holds the executive vice president title). The CMO currently manages the company's marketing functions, including customer insights and strategic direction; advertising; CarMax.com; branding; creative; digital; store marketing; targeted marketing; media, public, and community relations; and internal communications ([ 6]). As such, one conjecture is that CMOs with an enterprise-wide role are better able to make speedier marketing decisions compared with CMOs with more tactical roles (e.g., communications and advertising). Therefore, we encourage future research to examine the following questions: Do CMOs need structural power (e.g., executive title) for organizational buy-in and to sustain MA at the organizational level? Are CMOs with broader lines of responsibility (vs. CMOs with narrow lines of responsibility) better able to pursue MA?
The third area that warrants more research is the CMO–CIO interface and its implications for MA. The interdependencies between CMOs and CIOs are heightened in an agile marketing environment. While the CIO manages the technology that enables the collection, integration, security, and access to the firm's data, the CMO typically manages the marketing-related data analysis, interpretation, and program development ([114]). Often, customer-related digital activities are divided between IT and marketing, effectively splitting decision making between the two functional leaders.
The experience of CarMax reveals that a collaborative relationship between marketing and IT has contributed to its transformation to a company that delivers customer-centric innovation at greater speed. CarMax adopted agile marketing in 2014 with the appointment of a new CMO and CIO ([89]). Although the marketing assets of CarMax before 2014 were superior customer service and an incredible amount of customer data, the transformation to a technology firm focused on delivering a customer experience happened subsequently. Prior to 2014, the relationship between marketing and IT was a traditional customer–supplier relationship. The marketing group felt constrained by the speed with which IT operated.
The complementary backgrounds of the CMO and CIO—the former with a passion for data/analytics and the CIO with a master's of business administration in Marketing and prior experience as the founder of a MarTech company—helped in aligning the two functions to foster technology-enabled experimentation and implementation. The IT group reviewed its architectural choices and adopted a public cloud approach to host the development of customer-facing systems ([89]). The complementary backgrounds allowed the different thought worlds of marketing and IT to be bridged and enabled sensemaking at the CMO–CIO level. In addition, the two leaders colocated marketing and technology employees in a shared space and involved them in all product teams to realize these benefits at a team level.
Drawing on the CarMax example, we encourage future research to address the following research questions to understand the implications of the marketing–IT interface for MA: ( 1) What are the skills and experiences of CMOs and CIOs that are complementary? ( 2) Do complementary background skills and experience between CMOs and CIOs aid MA? and ( 3) Does cross-functional involvement (i.e., CMOs involved in technology procurement decisions and CIOs involved in marketing campaigns) promote MA?
What are the attributes of employees who are likely to thrive in an agile marketing organization? Although there is little specific research on the topic, given the importance of MarTech for MA, we can surmise that knowledge and comfort with technology tools will be essential. Likewise, given the importance of making decisions in the face of uncertainty, we can surmise that individuals high in the Big Five personality trait of "openness to experience" would be a natural fit for MA given that openness is related to curiosity and appreciation for novelty ([71]). It is also important to understand how employees respond to and cope with ambiguous and potentially stressful situations that are inherent in an agile environment. Prior research has delineated a variety of ways in which people respond to stress, including problem-focused coping, targeted at problem solving, and emotion-focused coping, aimed more at alleviating the negative feelings associated with the situation ([56]). Some of these strategies may be functional, such as planning or active coping (i.e., "taking active steps to try to remove or circumvent the stressor"; [17], p. 268). Others, such as the venting of emotions, denial, and mental disengagement, are more dysfunctional.
Although, as discussed, openness is a well-established trait associated with embrace of the unknown, future research should consider how other personality traits would suit employees to an agile marketing environment. Counterintuitively, we posit that high conscientiousness might actually be detrimental to performance in an agile marketing environment. Previous research has noted that organizations generally prize conscientiousness in employees because they tend to be disciplined, organized, and task-focused ([ 8]). In turn, conscientiousness is the personality factor most associated with professional success ([ 8]).
Paradoxically, however, high conscientiousness might be a bad fit for an agile marketing organization in which uncertainty, speed, and iteration (task-switching) are the norm. This is because conscientious people thrive on structure, order, thoroughness, and attention to detail. Indeed, [88] found, to their surprise, that higher conscientiousness was associated with lower adaptability; they attributed this finding to the need of conscientious individuals to be thorough and methodical and, therefore, to perceive situations that demand multitasking and adaptation as "threatening and stressful" (p. 82). Subsequent research has found conscientiousness to be negatively associated with both the preference and ability to adapt and be flexible ([96]). Likewise, [97] found that individuals with high conscientiousness performed poorly in a multitasking assignment; they surmised, "A conscientious individual is likely to have difficulty switching tasks without adequate time to ensure a thorough and detail-oriented approach to task completion" (p. 53) As such, we expect that rapidly iterating between sensemaking and decision making is likely to be difficult for highly conscientious individuals.
At the same time, given that conscientious individuals tend to be dutiful, one might wonder whether instructing them to be agile could, in fact, lead them to become highly agile? Although this prospect is superficially appealing, it seems that the uncertainty, swiftness, and task-switching key to agility may be fundamentally at odds with their nature. Thus, it seems unlikely that instructing highly conscientious employees to be agile will be sufficient to make them highly agile (though, with effort, they might become agile to some degree; see the subsequent discussion on adopting an agile mindset). Regardless, this question is an important one for future research to resolve.
In addition, more research is needed on the tools or training that can enhance employees' performance in an agile marketing environment. Previous research has found that training individuals to adopt a mindset relative to the task at hand can positively influence their performance on the task ([24]). Thus, the ability of training to inculcate different mindsets in employees, particularly mindsets related to speed and iteration, could be examined in relation to effects on employee performance in agile environments.
With respect to skills training for agile marketing that involves new technologies, at least a core group of marketing employees should be trained on various AI tools, such as machine learning (e.g., Heat AI for social listening, natural language processing and scoring social media content for credibility). In terms of process, HR personnel need to abandon their annual or quarterly planning cycles and adopt shorter retrospectives to assess training needs. A retrospective is a regularly cadenced (e.g., biweekly) meeting with employees or teams involved in a particular project or initiative to review how things have gone since the last retrospective. The time between "retros" needs to be short to allow new training ideas to be tested and reviewed ([38]). In addition to training marketing employees for technological skills, HR also needs to expose marketing employees to tools, concepts, language, and artefacts of other disciplines and expand their thought worlds for better sensemaking.
Furthermore, entry-level marketing employees may also not have the requisite skills to operate in an agile marketing organization. This is in part because business school curricula do not appear to emphasize either the basic MA features we have outlined or the technical tools that tend to be accompany agile marketing in many organizations. Among other efforts, the marketing curriculum can be improved to ensure that students are trained to use these technology tools (e.g., Campaign Monitor for email campaigns, Hootsuite for social media and marketing campaigns, HubSpot for customer relationship marketing, and Google pay-per-click ad campaigns for search engine marketing). Beyond tools, marketing curricula might increase the adaptability of students to agile marketing organization by introducing exercises and simulations that train students to make sense, iterate, and work fast in learning about and responding to marketplace events. Such activities could highlight the risks and rewards of MA. At the most extreme, marketing curricula could provide less structured and less well-defined syllabi and assignments with evolving objectives so that students are trained to iterate and adapt. That said, critical questions remain. How frequently should the curriculum for marketing courses be refined, and how can faculty be incentivized to do so? What are the training requirements to enable frequent and rapid adjustments to marketing course curricula? More research is needed to understand how business schools can more effectively educate students for jobs in an agile marketing organization.
Furthermore, employees need to be trained to cope with stress that may accompany working in an agile marketing organization. Frequent cycles of failed experimentation may lead marketing employees to resort to more dysfunctional coping strategies, such as emotional venting and denial ([75]). As such, it is critical to identify specific mechanisms (e.g., training, staff appraisal, incentives) that could shift marketing managers toward more problem-focused coping (i.e., coping strategy aimed at problem solving) and effective sensemaking, rather than dysfunctional coping strategies. Relatedly, several additional questions arise for future research. For example, what are the ways to instill an elevated sense of control in marketing employees given that they are likely to experience failure more often than success? What are the ways to promote a monitoring (i.e., "seeking out information about one's situation and its potential impact") instead of a blunting (i.e., "dealing with an impending stressor by attempting to distract oneself from it") coping strategy ([17], p. 275)? More importantly, how can organizations train marketing employees to be ambidextrous and switch between coping strategies when the need arises, such as from active coping (i.e., taking active steps to respond to a stressor) to restraint coping (i.e., holding back and waiting for the right opportunity to respond)?
Finally, it is important to consider both the positive and negative impacts of MA on employees. Among MA's likely benefits to employees include quick implementation of their ideas, less bureaucracy, and validation through customer testing rather than on the basis of a high-ranking manager's opinion, often dubbed "HIPPO" ("Highest Paid Person's Opinion"; [34]). These factors might increase employee engagement. A potential downside of MA to employees, particularly those that have thrived in a more traditional marketing organization, is that it might threaten their identity. For most employees, their role and position at work is not just a job but also a central aspect of their identity. Notably, an emphasis on data, analytics, and "test and learn" is likely to be threatening to employees whose skills are suited for a more traditional marketing culture ([60]; [75]). Likewise, a potential threat to their sense of identity, in an agile marketing organization, employee roles are likely to be less well-defined and to shift depending on the team to which they are assigned. Future research should examine the impact on employees of working in an agile marketing organization.
As noted previously, the business impact of MA is not a given, and the benefits realized are likely to be contingent on several factors. Given our process-based conceptualization of MA, we discuss the potential impact of MA on both product-market outcomes and stock market metrics and offer directions for future research.
What are the likely market performance benefits of MA, and under what conditions are these benefits likely to be muted? While academic research on the performance effects of MA (and agility in general) is currently lacking, anecdotal evidence suggests that the time to market is shorter for agile marketing firms ([ 2]). For example, Canada's TD Bank has embraced agility in its digital marketing function. Using two-week design sprints over three months, the bank was able to cut costs by 30% and reduce campaign turnaround times from four months to two weeks ([86]). While such anecdotes illuminate the promise of MA in accelerating time to market, such benefits are likely to be limited by several factors, depending on context. For instance, significant reduction in time to market may not be feasible in industries where purchase cycles are longer and more complex. Likewise, the extent to which time to market is reduced would vary depending on fit of the marketing activity for MA (see Table 2). Finally, the benefits of shorter time to market may be negligible if firms and/or the marketing ecosystem do not possess the competencies to consistently identify, prioritize and test the "right" ideas.
Some anecdotal evidence also suggests that MA could have a positive impact on customer satisfaction ([ 2]). We expect the impact of MA on customer satisfaction to have both positive and negative aspects. On the one hand, because marketing campaigns and products are developed through iteration and validation with customer groups, the campaign or product is likely to be in sync with market needs. As such, customer satisfaction is likely to be higher. On the other hand, the impact of MA on customer satisfaction could be muted if employee satisfaction turns out to be lower. As noted in the preceding section, the impact of MA on employee satisfaction could be either positive (e.g., better engagement because of autonomy in decision making) or negative (e.g., identity and/or coping concerns). Similarly, the impact of MA on the brand meaning for consumers needs to be managed carefully. This is because while frequent experimentation with marketing messages could increase brand differentiation, it could also dilute brand meaning or diminish brand relevance over time. Future empirical research, therefore, could investigate the impact of MA on a multitude of product-market outcomes and the contingencies associated with these relationships.
What is the value relevance of agile marketing for stock market performance? It is worth noting that publicly listed firms frequently announce their adoption of MA or disclose the marketing experiments they pursued. For example, Ford Motors announced in 2018 that it is adopting a "more agile marketing model" that could potentially save more than $150 million in its marketing spending ([99]). Similarly, Google announced in 2014 that it ran several A/B tests to decide on the precise shade of blue to be used in the advertising links for Google search and Gmail. Furthermore, it was reported that this agile marketing initiative netted Google incremental revenues of $200 million ([43]). Future research, therefore, could use the event-study method to test whether such announcements have value relevance for stock markets. There is some support for the expectation that investors and analysts are likely to respond positively to specific elements of MA such as speed of marketing decisions. For example, [61] find that firms that are slower to preannounce price increases in response to their competitors are likely to experience lower abnormal returns.
Given the increasing importance of considering risk implications of marketing decisions ([40]), future research could also assess the impact of MA on the risk of stock returns as reflected in investor uncertainty. For example, our in-depth interview with CarMax revealed that the process of iterations and validation in both the discovery and delivery aspects of campaigns results in a bad concept or poor delivery getting weeded out before the campaign or product is launched. If so, firms with higher MA are likely to have lower investor uncertainty. However, if firms tweak their marketing decisions frequently, the volatility of stock returns could increase. Clearly, a systematic and detailed investigation of the impact of MA on investor uncertainty is warranted.
The 2018–2020 MSI Research Priorities highlight the need to study how organizations should change internally and interact with their environment to cope with rapid changes. The ongoing COVID-19 crisis brings into sharp focus the need for marketers to be ready to change and flex quickly as things shift ([77]). Against this backdrop, our study explicates the concept of MA—a theoretically rich, multidimensional construct which is sufficiently distinct from that of agility in other functions and other conceptually related organizational constructs. Our study develops a research agenda pertaining to the organizational, team, marketing leadership, and employee antecedents of MA. We, however, also caution that there are numerous challenges in executing MA and as such it may not be well-suited for all firms and marketing activities.
Supplemental Material, JM.18.0481_Web_Appendix_PDF - Marketing Agility: The Concept, Antecedents, and a Research Agenda
Supplemental Material, JM.18.0481_Web_Appendix_PDF for Marketing Agility: The Concept, Antecedents, and a Research Agenda by Kartik Kalaignanam, Kapil R. Tuli, Tarun Kushwaha, Leonard Lee and David Gal in Journal of Marketing
Graph
| No. | Title | Industry | Exp | Function | Duration |
|---|
| 1 | VP Learning and Talent Management | Tourism and Hospitality | 25 | Marketing & Human Resource Management | 90 |
| 2 | VP Corporate Communications | Public Infrastructure | 17 | Marketing & Corporate Communications | 90 |
| 3 | CEO | Retailing | 40 | General Management | 60 |
| 4 | Director of Marketing | Retailing | 20 | Marketing | 60 |
| 5 | Senior Marketing Consultant | Consulting | 25 | Consulting | 60 |
| 6 | Managing Director | Consulting | 25 | General Management | 30 |
| 7 | Senior Product Manager | Retailing | 12 | Marketing | 30 |
| 8 | Senior Product Manager | Retailing | 12 | Marketing | 60 |
| 9 | VP Marketing | Medical Equipment | 20 | Marketing | 45 |
| 10 | VP Marketing | Chemicals | 20 | Marketing | 60 |
| 11 | Senior Product Manager | Retailing | 20 | Marketing | 45 |
| 12 | Chief Brand Officer, VP Global Marketing | IT Products & Services | 24 | Sales & Marketing | 35 |
| 13 | Global Integration Lead | Chemicals | 21 | Marketing | 30 |
| 14 | Manager Content Marketing | IT Products & Services | 13 | Marketing | 35 |
| 15 | Director Market Access | Healthcare | 15 | Marketing | 30 |
| 16 | Director Business Development | Food | 25 | Sales | 30 |
| 17 | Senior VP Analytics | B2B Services | 20 | Marketing | 30 |
| 18 | Industry Solutions Lead | IT Products & Services | 13 | Strategy & Business Development | 35 |
| 19 | Technical Sales Professional | IT Products & Services | 5.5 | Sales | 45 |
| 20 | Regional Business Leader for Retail and Consumers | IT Products & Services | 25 | General Management | 30 |
| 21 | Digital Lead for E-Commerce and Digital | IT Products & Services | 12 | Marketing | 30 |
| 22 | Country Manager | Financial Services | 18 | General Management | 45 |
| 23 | Senior Director Regional Client Management | Financial Services | 26 | Business Development | 30 |
1 Notes: VP = Vice President, CEO = Chief Executive Officer; Exp = number of years of experience; Duration = time for each interview in minutes.
Footnotes 1 Author Contributions The authors are listed in the order of their contributions.
2 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 ORCID iDs Kartik Kalaignanam https://orcid.org/0000-0003-0821-3687 Tarun Kushwaha https://orcid.org/0000-0002-2809-2467 David Gal https://orcid.org/0000-0002-3446-9304
5 Online supplement https://doi.org/10.1177/0022242920952760
6 We use "customers" to denote both business-to-customer (B2C) and business-to-business (B2B) customers.
7 The review covered both academic journals (e.g., Journal of Marketing, Journal of Marketing Research, Journal of Consumer Research, Administrative Science Quarterly, Organization Science, Academy of Management Journal, Academy of Management Review, Strategic Management Journal, Information Systems Research, and Management Science) and practitioner outlets (e.g., Harvard Business Review, California Management Review, and MIT Sloan Management Review).
8 An analogy can be drawn to the well-known "cargo cult" phenomenon, in which technologically primitive societies attempt to gain the benefits of a technology through imitating the symbols and rituals of technologically more advanced societies—such as by building airstrips, mock airplanes, and mock radios in the anticipation that planes will arrive and deliver cargo ([63]).
9 We conducted an in-depth interview with a senior product executive at CarMax to understand the company's unique processes and routines.
References Abramovich Giselle. (2018), "Agile Is the Mindset of the Modern Marketing Organization," CMO by Adobe (June), https://cmo.adobe.com/articles/2018/6/agile-is-the-mindset-of-the-modern-marketing-organization.html#gs.ao4xm5.
Aghina Wouter, Handscomb Christopher, Ludolph Jesper, Rona Daniel, West Dave. (2020), "Enterprise Agility: Buzz or Business Impact," McKinsey (March 20), https://www.mckinsey.com/business-functions/organization/our-insights/enterprise-agility-buzz-or-business-impact.
Aghina Wouter, Ahlback Karin, De Smet Aaron, Lackey Gerald, Lurie Michael, Murarka Monica, et al. (2018), "The Five Trademarks of Agile Organizations," McKinsey (January 22), https://www.mckinsey.com/business-functions/organization/our-insights/the-five-trademarks-of-agile-organizations.
Ahlback Karin, Comella-Dorda Santiego, Mahadevan Deepak. (2018), "The Drawbacks of Agility," McKinsey (May 7), https://www.mckinsey.com/business-functions/organization/our-insights/the-organization-blog/the-drawbacks-of-agility.
Andrews Jonlee, Smith Daniel C. (1996), "In Search of Marketing Imagination: Factors Affecting the Creativity of Marketing Programs for Mature Products," Journal of Marketing Research, 33 (2), 174–87.
Auto Remarketing (2014), "CarMax Picks New Chief Marketing Officer," (August 8), https://www.autoremarketing.com/retail/carmax-picks-new-chief-marketing-officer.
Barabba Vincent P., Zaltman Gerald. (1990), Hearing the Voice of the Market: Competitive Advantage Through Creative Use of Market Information. Boston : Harvard Business School Publishing.
Barrick Murray R., Mount Michael K., Judge Timothy A. (2001), "Personality and Performance at the Beginning of the New Millennium: What Do We Know and Where Do We Go Next?" International Journal of Selection and Assessment, 9 (1/2), 9–30.
Berthiaume Dan. (2019), "CarMax Disrupts Operations for Omnichannel Success," Chain Store Age (August 22), https://chainstoreage.com/technology/csa-live-from-etail-boston-carmax-disrupts-operations-for-omnichannel-success.
Beverland Michael B., Micheli Pietro, Farrelly Francis J. (2016), "Resourceful Sensemaking: Overcoming Barriers Between Marketing and Design in NPD," Journal of Product Innovation Management, 33 (5), 628–48.
Beverland Michael B., Wilner Sarah, Micheli Pietro. (2015), "Reconciling the Tension Between Consistency and Relevance: Design Thinking as a Mechanism for Brand Ambidexterity," Journal of the Academy of Marketing Science, 43 (5), 589–609.
Bhuiyan Johana. (2017), "Everything You Need to Know About Uber's Turbulent 2017," Vox (August 20), https://www.vox.com/2017/8/20/16164176/uber-2017-timeline-scandal.
Boudet Julien, Gordon Jonathan, Gregg Brian, Perrey Jesko, Robinson Kelsey. (2020), "How Marketing Leaders Can Both Manage the Coronavirus Crisis and Plan for the Future," McKinsey (April 8), https://www.mckinsey.com/business-functions/marketing-and-sales/our-insights/how-marketing-leaders-can-both-manage-the-coronavirus-crisis-and-plan-for-the-future.
Brockman Beverly K., Rawlston Melissa E., Jones Michael A., Halstead Diane. (2010), "An Exploratory Model of Interpersonal Cohesiveness in New Product Development Teams," Journal of Product Innovation Management, 27 (2), 201–19.
Cao Qing, Dowlatshahi Shad. (2005), "The Impact of Alignment Between Virtual Enterprise and Information Technology on Business Performance in an Agile Manufacturing Environment," Journal of Operations Management, 23 (5), 531–50
Carbonell Pilar, Rodriguez Ana I. (2006), "Designing Teams for Speedy Product Development: The Moderating Effect of Technological Complexity," Journal of Business Research, 59 (2), 225–32.
Carver Charles S., Scheier Michael F., Weintraub Jagdish K. (1989), "Assessing Coping Strategies: A Theoretically Based Approach," Journal of Personality and Social Psychology, 56 (2), 267–83
Clegg Chris W. (2000), "Sociotechnical Principles for System Design," Applied Ergonomics, 31 (5), 463–77.
Cohn M.L., Sim Susan E., Lee Charlotte P. (2009), "What Counts as Software Process? Negotiating the Boundary of Software Work Through Artifacts and Conversation," Computer Supported Cooperative Work (CSCW), 18 (5/6), 401–33.
Collins B. (2017), "#DeleteUber's Creator: Resist Trump or 'Pay a Price'," The Daily Beast (April 11), https://www.thedailybeast.com/deleteubers-creator-resist-trump-or-pay-a-price.
Columbus Louis. (2018), "10 Ways Machine Learning is Revolutionizing Marketing" Forbes (February 25), https://www.forbes.com/sites/louiscolumbus/2018/02/25/10-ways-machine-learning-is-revolutionizing-marketing/#781ce105bb64.
Conboy Kieran. (2009), "Agility from First Principles: Reconstructing the Concept of Agility in Information Systems Development," Information Systems Research, 20 (3), 329–54.
Cresci E. (2017), "#DeleteUber: How Social Media Turned on Uber," The Guardian (January 30), https://www.theguardian.com/technology/2017/jan/30/deleteuber-how-social-media-turned-on-uber.
Cutts Quintin, Cutts Emily, Draper Stephen, O'Donnell Patrick, Saffrey Peter. (2010), " Manipulating Mindset to Positively Influence Introductory Programming Performance," in Proceedings of the 41st ACM Technical Symposium on Computer Science Education. New York : Association for Computing Machinery, 431–35.
Day George. (2011), "Closing the Marketing Capabilities Gap," Journal of Marketing, 75 (4), 183 –95.
Demirkan Haluk, Delen Dursun. (2013) "Leveraging the Capabilities of Service-Oriented Decision Support Systems: Putting Analytics and Big Data in Cloud," Decision Support Systems, 55 (1), 412–21.
Deshpandé Rohit. (1982), "The Organizational Context of Market Research Use," Journal of Marketing, 46 (4), 91–101.
Dietvorst Berkeley J., Simmons Joseph P., Massey Cade. (2015). "Algorithm Aversion: People Erroneously Avoid Algorithms After Seeing Them Err," Journal of Experimental Psychology: General, 144 (1), 114–26.
Dignan Larry. (2019), "Nike Bets on Tech CEO Donahoe to Accelerate Digital transformation: Will It Work?" ZDNet (October 27), https://www.zdnet.com/article/nike-bets-on-tech-ceo-donahoe-to-accelerate-digital-transformation-will-it-work/.
Doz Yves. (2020), "Fostering Strategic Agility: How Individual Executives and Human Resource Practices Contribute," Human Resource Management Review, 30 (1), 1–14.
Doz Yves, Kosonen Mikko. (2008), "The Dynamics of Strategic Agility: Nokia's Rollercoaster Experience," California Management Review, 50 (3), 95–118.
Eisenhardt Kathleen M., Martin Jeffrey A. (2000), "Dynamic Capabilities: What Are They?" Strategic Management Journal, 21 (10), 1105–21
Feng Hui, Morgan Neil A., Rego Lopo L. (2015), "Marketing Department Power and Firm Performance," Journal of Marketing, 79 (5), 1–20
Gallagher Deb. (2012), "The Decline of the HPPO (Highest Paid Person's Opinion)," MIT Sloan Management Review (April 1), https://sloanreview.mit.edu/article/the-decline-of-the-hppo-highest-paid-persons-opinion /.
Gebhardt Gary F, Carpenter Gregory S., Sherry John F. Jr. (2006), "Creating a Market Orientation: A Longitudinal, Multifirm, Grounded Analysis of Cultural Transformation," Journal of Marketing, 70 (4), 37–55.
Ghasemaghaei Maryam, Hassanein Khaled, Turel Ofir. (2017), "Increasing Firm Agility Through the Use of Data Analytics: The Role of Fit," Decision Support Systems, 101 (September), 95–105.
Gilson Lucy L., Shalley Christina E. (2004), "A Little Creativity Goes a Long Way: An Examination of Teams' Engagement in Creative Processes," Journal of Management, 30 (4), 453–70.
Gothelf Jeff. (2017), Lean vs. Agile vs. Design Thinking: What You Really Need to Know to Build High-Performing Digital Product Teams. Sense and Respond Press.
Hadida Allègre L., Heide Jan B., Bell Simon J. (2019), "The Temporary Marketing Organization," Journal of Marketing, 83 (2), 1–18.
Han Kyuhon, Mittal Vikas, Zhang Yan. (2017), "Relative Strategic Emphasis and Firm-Idiosyncratic Risk: The Moderating Role of Relative Performance and Demand Instability," Journal of Marketing, 81 (4), 25–44.
Haon Christophe, Gotteland David, Fornerino Marianela. (2009), "Familiarity and Competence Diversity in New Product Development Teams: Effects on New Product Performance," Marketing Letters, 20 (1), 75–89.
Harris Stanley G. (1994), "Organizational Culture and Individual Sensemaking: A Schema-Based Perspective," Organization Science, 5 (3), 309–21.
Hern Alex. (2014), "Why Google Has 200 m Reasons to Put Engineers Over Designers," The Guardian (February 5), https://www.theguardian.com/technology/2014/feb/05/why-google-engineers-designers.
Homburg Christian, Pflesser Christian. (2000), "A Multiple-Layer Model of Market-Oriented Organizational Culture: Measurement Issues and Performance Outcomes," Journal of Marketing Research, 37 (4), 449–62.
Homburg Christian, Theel Marcus, Hohenburg Sebestian. (2020), "Marketing Excellence: Nature, Measurement, and Investor Valuations," Journal of Marketing, 84 (4), 1–22.
Huckman R., Staats Bradley. (2013), "The Hidden Benefits of Keeping Teams Intact," Harvard Business Review, 91 (12), 27–29.
Hurley Robert F., Hult Thomas M. (1998), "Innovation, Market Orientation, and Organizational Learning: An Integration and Empirical Examination," Journal of Marketing, 62 (3), 42–54.
Im Subin, Montoya Mitzi M., Workman John P. Jr. (2013), "Antecedents and Consequences of Creativity in Product Innovation Teams," Journal of Product Innovation Management, 30 (1), 170–85.
Isaac Mike. (2017), "What You Need to Know About #DeleteUber," The New York Times (January 31), https://www.nytimes.com/2017/01/31/business/delete-uber.html.
Isenberg Daniel J. (1986), "Group Polarization: A Critical Review and Meta-Analysis," Journal of Personality and Social Psychology, 50 (6), 1141–51.
Janis Irving L. (1972), Victims of Groupthink: A Psychological Study of Foreign-Policy Decisions and Fiascoes. Boston : Houghton Mifflin.
Johnson Jean L., Lee Ruby Pui-Wan, Saini Amit, Grohmann Bianca. (2003), "Market-Focused Strategic Flexibility: Conceptual Advances and an Integrative Model," Journal of the Academy of Marketing Science, 31 (1), 74–89.
Kane Gerald C. (2019), "Establish Ethical Guardrails to Guide Digital Growth," The Wall Street Journal (October 21), https://deloitte.wsj.com/cmo/2019/10/21/establish-ethical-guardrails-to-guide-digital-growth/.
Kane Gerald C., Palmer Doug, Phillips Anh Nguyen, Kiron David, Buckley Natasha. (2019), "Accelerating Digital Innovation Inside and Out: Agile Teams, Ecosystems, and Ethics," MIT Sloan Management Review (June 4), https://sloanreview.mit.edu/projects/accelerating-digital-innovation-inside-and-out /.
Kohli Ajay K., Jaworski Bernard J. (1990), "Market Orientation: The Construct, Research, Propositions, and Managerial Implications," Journal of Marketing, 54 (2), 1–18.
Lazarus Richard S., Folkman Susan. (1984), Stress, Appraisal, and Coping. New York : Springer Publishing Company.
Lee Gwanhoo, Xia Weidong. (2010), "Toward Agile: An Integrated Analysis of Quantitative and Qualitative Field Data on Software Development Agility," MIS Quarterly, 34 (1), 87–114.
Lee Leonard J., Inman Jeffrey, Argo Jennifer J., Böttger Tim, Dholakia Utpal, Gilbride Timothy. et al. (2018), "From Browsing to Buying and Beyond: The Needs-Adaptive Shopper Journey Model," Journal of the Association for Consumer Research, 3 (3), 277–93.
Lemon Katherine N., Verhoef Peter C. (2016), "Understanding Customer Experience and the Customer Journey," Journal of Marketing, 80 (6), 1 –62.
Leung Eugenia, Paolacci Gabriele, Puntoni Stefano. (2018), "Man Versus Machine: Resisting Automation in Identity-Based Consumer Behavior," Journal of Marketing Research, 55 (6), 818–31.
Lim Leon Gim, Tuli Kapil R., Dekimpe Marnik. (2018), "Investors' Evaluation of Price-Increase Preannouncements," International Journal of Research in Marketing, 35 (3), 359–77.
Lin Ching-Torng, Chu Hero, Tseng Yi-Hong. (2006), "Agility Evaluation Using Fuzzy Logic," International Journal of Production Economics, 101 (2), 353–68.
Lindstrom Lamont. (1993), Cargo Cult: Strange Stories of Desire from Melanesia and Beyond. Honolulu: University of Hawaii Press.
Lu Ying, Ramamurthy K. (Ram). (2011), "Understanding the Link Between Information Technology Capability and Organizational Agility: An Empirical Examination," MIS Quarterly, 35 (4), 931–54.
Lynch Jacqueline, West Douglas C. (2017), "Agency Creativity: Teams and Performance: A Conceptual Model Links Agency Teams' Knowledge Utilization, Agency Creativity, and Performance," Journal of Advertising Research, 57 (1), 67–81.
Maitlis Sally. (2005), "The Social Processes of Organizational Sensemaking," Academy of Management Journal, 48 (1), 21–49.
Maitlis Sally, Christianson Marlys. (2014), "Sensemaking in Organizations: Taking Stock and Moving Forward," Academy of Management Annals, 8 (1), 57–125.
Mankins Michael, Garton Eric. (2017), "How Spotify Balances Employee Autonomy and Accountability," Harvard Business Review (February 9), https://hbr.org/2017/02/how-spotify-balances-employee-autonomy-and-accountability.
Manz Charles C., Stewart Greg L. (1997), "Attaining Flexible Stability by Integrating Total Quality Management and Socio-Technical Systems Theory," Organization Science, 8 (1), 59–70.
Martin Kelly D., Borah Abhishek, Palmatier Robert W. (2017), "Data Privacy: Effects on Customer and Firm Performance," Journal of Marketing, 81 (1), 36–58.
McCrae Robert R., Costa Paul T. (1987), "Validation of the Five-Factor Model of Personality Across Instruments and Observers," Journal of Personality and Social Psychology, 52 (1), 81–90.
McFarland Richard G., Challagalla Goutam N., Shervani Tasadduq A. (2006), "Influence Tactics for Effective Adaptive Selling," Journal of Marketing, 70 (4), 103–17.
Menon Anil, Bharadwaj Sundar G., Adidam Phani Tej, Edison Steven W. (1999), "Antecedents and Consequences of Marketing Strategy Making: A Model and a Test," Journal of Marketing, 63 (2), 18–40.
Micheli Pietro, Jaina Joe, Goffin Keith, Lemke Fred, Verganti Roberto. (2012), "Perceptions of Industrial Design: The 'Means' and the 'Ends'," Journal of Product Innovation Management, 29 (5), 687–704.
Mick David Glen, Fournier Susan. (1998), "Paradoxes of Technology: Consumer Cognizance, Emotions, and Coping Strategies," Journal of Consumer Research, 25 (2), 123–43.
Moorman Christine. (1995), "Organizational Market Information Processes: Cultural Antecedents and New Product Outcomes," Journal of Marketing Research, 32 (3), 318–35.
Moorman Christine. (2020), "Making the Most of Your Marketing Team During COVID-19," Forbes (March 30), https://www.forbes.com/sites/christinemoorman/2020/03/30/making-the-most-of-your-marketing-team-during-covid-19/#7225f1a02321.
Moorman Christine, Day George S. (2016), "Organizing for Marketing Excellence," Journal of Marketing, 80 (6), 6–35.
Moorman Christine, Miner Anne S. (1998), "Organizational Improvisation and Organizational Memory," Academy of Management Review, 23 (4), 698–723.
Moorman Christine, Zaltman Gerald, Deshpandé Rohit. (1992), "Relationships Between Providers and Users of Market Research: The Dynamics of Trust within and Between Organization," Journal of Marketing Research, 29 (3), 314–28.
Morgan Jacob. (2015), "Are You Embracing Failure or Encouraging Failure?" Forbes (June 2), https://www.forbes.com/sites/jacobmorgan/2015/06/02/are-you-embracing-failure-or-encouraging-failure/#710e431132ad.
Morgeson Fredrick P., Hofmann David A. (1999), "The Structure and Function of Collective Constructs: Implications of Multilevel Research and Theory Development," Academy of Management Review, 24 (2), 249–65.
Morrison Mary E. (2019), "TD's Agile Approach to Always-On Marketing," The Wall Street Journal (January 27), https://deloitte.wsj.com/cmo/2019/01/27/tds-agile-approach-to-always-on-marketing/.
Myers David G., Lamm Helmut. (1976), "The Group Polarization Phenomenon," Psychological Bulletin, 83 (40), 602–27.
Niessing Joerg, Aaker David. (2015), "Being Too Agile Could Kill Your Brand," INSEAD Knowledge Blog (October 8), https://knowledge.insead.edu/blog/insead-blog/being-too-agile-could-kill-your-brand-4299#p7sOzU6i5VGrcUFP.99.
O'Brien Diana, Main Andy, Kounkel Suzanne, Stephan Anthony R. (2019), "Diffusing Agility Across the Organization," Deloitte Insights (October 15), https://www2.deloitte.com/us/en/insights/topics/marketing-and-sales-operations/global-marketing-trends/2020/agile-marketing.html#.
Odell Patty. (2017), "Marketing and Advertising Talent Crisis Looms: ANA study," Chief Marketer (September 21), https://www.chiefmarketer.com/marketing-and-advertising-talent-crisis-looms-ana-study/.
Oswald Frederick L., Hambrick David Z., Jones Andrew L. (2017), " Keeping All the Plates Spinning: Understanding and Predicting Multitasking Performance," in Learning to Solve Complex Scientific Problems. New York : Routledge, 77–96.
Overby Stephanie. (2017), "Together, CarMax CMO And CIO Steer Popular Brand Toward Transformation," CMO by Adobe (accessed on June 25, 2020), https://cmo.adobe.com/articles/2017/3/carmax-cmocio-partnership-a-road-map-for-brands.html#gs.azc7jb.
Peltier James W., Zahay Debra, Lehmann Donald R. (2013), "Organizational Learning and CRM Success: A Model for Linking Organizational Practices, Customer Data Quality, and Performance," Journal of Interactive Marketing, 27 (1), 1–13.
Perry Monica L., Pearce Craig L., Sims Henry P. Jr. (1999), "Empowered Selling Teams: How Shared Leadership Can Contribute to Selling Team Outcomes." Journal of Personal Selling & Sales Management, 19 (3), 35–51.
Pinto Mary Beth, Pinto Jeffrey K., Prescott John E. (1993), "Antecedents and Consequences of Project Team Cross-Functional Cooperation," Management Science, 39 (10), 1281–97
Raisch Sebastian, Birkinshaw Julian. (2008), "Organizational Ambidexterity: Antecedents, Outcomes, and Moderators," Journal of Management, 34 (3), 375–409.
Rajgopal Shivram, Srivastava Anup. (2020), "Is Technology Subsuming Marketing?" Harvard Business Review (February 25), https://hbr.org/2020/02/is-technology-subsuming-marketing.
Ritson Mark. (2020), "Marketers' Strategic Responsibilities Are Eroding Away to Nothing," Marketing Week (April 29), https://www.marketingweek.com/mark-ritson-marketers-strategic-responsibilities-eroding /.
Robert Christopher Yu, Cheung Ha. (2010), "An Examination of the Relationship Between Conscientiousness and Group Performance on a Creative Task," Journal of Research in Personality, 44 (2), 222–31.
Sanderson Kristin R., Bruk-Lee Valentina, Viswesvaran Chockalingam, Gutierrez Sara, Kantrowitz Tracy. (2016), "Investigating the Nomological Network of Multitasking Ability in a Field Sample," Personality and Individual Differences, 91 (March), 52–57.
Sarin Shikhar, Mahajan Vijay. (2001), "The Effect of Reward Structures on the Performance of Cross-Functional Product Development Teams," Journal of Marketing, 65 (2), 35–53.
Schultz E.J. (2018), "Ford Taps Bryan Cranston for Swagger-Filled U.S. Campaign," AdAge (October 19), https://adage.com/article/cmo-strategy/ford-taps-bryan-cranston-swagger-filed-u-s-campaign/315326.
Sethi Rajesh, Smith Daniel C., Park Whan C. (2001), "Cross-Functional Product Development Teams, Creativity, and the Innovativeness of New Consumer Products," Journal of Marketing Research, 38 (1), 73–85.
Sinkula James M. (1994), "Market Information Processing and Organization Learning," Journal of Marketing, 58 (1), 35–45.
Stanier Curtis. (2019), "Why the Spotify Model Won't Solve All Your Problems in Product Delivery," Medium (November 3), https://medium.com/@crstanier/why-the-spotify-model-won-t-solve-all-your-problems-4c31640c719a.
Statista (2019), "Annual Number of Data Breaches and Exposed Records in the United States from 2005 to 2019," (accessed on July 10, 2020), https://www.statista.com/statistics/273550/data-breaches-recorded-in-the-united-states-by-number-of-breaches-and-records-exposed/.
Swafford Patricia M., Ghosh Soumen, Murthy Nagesh. (2006), "The Antecedents of Supply Chain Agility of a Firm Scale Development and Model Testing," Journal of Operations Management, 24 (2), 170–88.
Swaminathan Vanitha, Sorescu Alina, Steenkamp Jan-Benedict E.M., O'Guinn Thomas Clayton Gibson, Schmitt Bernd. (2020), "Branding in a Hyperconnected World: Refocusing Theories and Rethinking Boundaries," Journal of Marketing, 84 (2), 24–46.
Urban Glen, Timoshenko Artem, Dhillon Paramveer, Hauser John R. (2020), "Is Deep Learning a Game Changer for Marketing Analytics?" MIT Sloan Management Review, 61 (2), 71–76.
Van de Ven Andrew H. (1986), "Central Problems in the Management of Innovation," Management Science, 32 (5), 590–607.
Visser Jody, Sheerin Alannah, Field Dominic, Ratajczak David. (2018), "How Agile Marketing Organizations Get That Way," Boston Consulting Group (September 6), https://www.bcg.com/en-us/publications/2018/how-agile-marketing-organizations-get-that-way.aspx.
Vosoughi Soroush, Roy Deb, Aral Sinan 2018, "The Spread of True and False News Online," Science, 359 (6380), 1146–51.
Walsh James P. (1995), "Managerial and Organizational Cognition: Notes from a Trip Down Memory Lane," Organization Science, 6 (3), 280–321.
Wedel Michel, Kannan P.K. (2016), "Marketing Analytics for Data-Rich Environments," Journal of Marketing, 80 (6), 97–121.
Weick Karl E. (1993), "The Collapse of Sensemaking in Organizations: The Mann Gulch Disaster," Administrative Science Quarterly, 38 (4), 628–52.
Whitler Kimberly, Morgan Neil A. (2017), "Why CMOs Never Last," Harvard Business Review, 95 (4), 47–54.
Whitler Kimberly, Boyd Eric D., Morgan Neil A. (2017), "The Criticality of CMO-CIO Alignment," Business Horizons, 60 (3), 313–24.
Winter Sidney G. (2003), "Understanding Dynamic Capabilities," Strategic Management Journal, 24 (10), 991–95.
Wright Charles R., Manning Michael R., Farmer Bruce, Gilbreath Brad. (2000), "Resourceful Sensemaking in Product Development Teams," Organization Studies, 21 (4), 807–25.
Zeithaml Valarie A, Jaworski Bernard J., Kohli Ajay K., Tuli Kapil R., Ulaga Wolfgang, Zaltman Gerald. (2020), "A Theories-in-Use Approach to Building Marketing Theory," Journal of Marketing, 84 (1), 32–51.
~~~~~~~~
By Kartik Kalaignanam; Kapil R. Tuli; Tarun Kushwaha; Leonard Lee and David Gal
Reported by Author; Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 88- Marketing Ideas: How to Write Research Articles that Readers Understand and Cite. By: Warren, Nooshin L.; Farmer, Matthew; Gu, Tianyu; Warren, Caleb. Journal of Marketing. Sep2021, Vol. 85 Issue 5, p42-57. 16p. 1 Diagram, 2 Charts, 1 Graph. DOI: 10.1177/00222429211003560.
- Database:
- Business Source Complete
Marketing Ideas: How to Write Research Articles that Readers Understand and Cite
Academia is a marketplace of ideas. Just as firms market their products with packaging and advertising, scholars market their ideas with writing. Even the best ideas will make an impact only if others understand and build on them. Why, then, is academic writing often difficult to understand? In two experiments and a text analysis of 1,640 articles in premier marketing journals, this research shows that scholars write unclearly in part because they forget that they know more about their research than readers, a phenomenon called "the curse of knowledge." Knowledge, or familiarity with one's own research, exacerbates three practices that make academic writing difficult to understand: abstraction, technical language, and passive writing. When marketing scholars know more about a research project, they use more abstract, technical, and passive writing to describe it. Articles with more abstract, technical, and passive writing are harder for readers to understand and are less likely to be cited. The authors call for scholars to overcome the curse of knowledge and provide two tools—a website (writingclaritycalculator.com) and a tutorial—to help them recognize and repair unclear writing so they can write articles that are more likely to make an impact.
Keywords: citations; methods; readability; relevance; text analysis; writing
Some articles are easy to understand. Professors, doctoral students, and practitioners alike effortlessly absorb and remember the ideas. Other articles befuddle anyone who is not already familiar with the research. Consider two hypothetical titles that would not be out of place in a premier marketing journal:
- Title 1: "The Interactive Effects of Ideological Orientation and Corporate Sociopolitical Activism on Owned Media Engagement"
- Title 2: "How Liberal and Conservative Consumers Respond When Brands Post Polarizing Messages on Social Media"
The titles describe the same research but use different writing styles. Consequently, most readers have an easier time understanding Title 2. Why is Title 1 less clear than Title 2? Why do scholars tend to write using the unclear style in Title 1? Which article is more likely to succeed in the academic marketplace?
Academic journals provide a marketplace of ideas ([47]). Successful ideas spread. Scholars cite them. Managers implement them. They change scholarly discourse, policy decisions, and industries ([18]). Yet editors (e.g., [22]; [27]; [47]), presidents (e.g., [13]; [46]; [53]), and fellows (e.g., [43]; [71]) worry that the research published in top marketing journals has little influence on marketers, policy makers, consumers, or even other scholars. A conservative measure of impact is whether an article gets cited. Even by this metric, few are succeeding. [53], p. 412) states, "The vast majority of the research that gets published, even in our top journals—perhaps 70% of it—hardly has any measurable scholarly impact in terms of citations."
Why do many articles make little impact? One reason is because their writing is unclear. Readers who are not already familiar with the research struggle to understand it. And when readers do not understand an article, they are unlikely to read it, much less cite it.
We argue that knowledge, although vital, makes researchers less likely to write clearly about their research. Conducting good research requires authors to know a lot about their work. It takes years to create research that meaningfully advances scientific knowledge. Consequently, academic articles are written by authors who are intimately familiar with their topic, methods, and results. Authors, however, often forget that potential readers (e.g., doctoral students, scholars in other subdisciplines) are less familiar with the research, a phenomenon called "the curse of knowledge" ([32]; [54]). The curse of knowledge prevents authors from recognizing when their writing is too abstract, technical, or passive. And, as we will show, abstraction, technical language, and passive writing make articles less clear and less likely to be cited. If unchecked, knowledge can thus prevent scholars from writing articles that make an impact.
We build our argument as follows. First, we review the research on how writing influences scholarly impact and suggest why this work has found inconsistent results. We next hypothesize that familiarity with their own research can lead scholars to overuse abstract, technical, and passive writing, practices that make research articles more difficult to understand and less likely to make an impact. We then describe three studies that document each step in our hypothesized process: ( 1) scholars are less likely to understand articles that use more abstract, technical, and passive writing; ( 2) scholars are less likely to cite articles that are more difficult to understand; and ( 3) scholars are less likely to recognize that their writing is unclear, and more likely to write unclearly, when they know more about the research project.
We contribute to research practice by revealing why researchers write unclearly, how they can write more clearly, and that clear writing can help them make a larger impact. Specifically, we show that knowledge increases the use of three practices (abstraction, technical language, and passive writing) that muddle scholarly writing and limit its impact. We provide scholars with two tools (Web Appendices A and B) to help them recognize and repair unclear writing. We also contribute to the literature on academic writing by highlighting the difference between writing clarity and readability. Previous research has attempted to assess the effect of an article's readability by measuring the average length of its sentences and words (e.g., [59]; [65]; [66]). Yet this work has produced mixed results, including a pair of studies that show that articles with higher readability scores are cited less often ([65]; [66]). In contrast, we show that clarity—which we conceptualize as a function of abstraction, technical language, and passive writing—is distinct from readability and better predicts which articles scholars understand and cite. Thus, if scholars want to make an impact, they should limit abstract, technical, and passive writing rather than worry about the length of their words and sentences.
Articles succeed in the marketplace of ideas when they influence other scholars or practitioners—ideally, both ([43]; [44]; [47]). It is difficult to measure an article's influence on practice, but researchers can measure scholarly impact by counting citations (e.g., [ 6]; [29]). On average, articles have a larger impact when they offer relevant, high-quality ideas and when they are published by an established author or in a prestigious journal ([30]; [65]; [66]). But the literature suggests that the impact of an article may also depend on its writing[ 6] ([35]; [38]; [52]; [66]).
To be influenced by an idea, scholars need to understand it. Yet academic writing is not easy to understand ([17]; [23]; [34]; [45]). For example, [59] note that the Journal of Marketing and Journal of Consumer Research are tougher to understand than PC World and the New York Times. However, academic journals and popular press outlets target different readers. Academic writing being unclear to the average New York Times reader may not be a problem, so long as scholars can understand it. Unfortunately, many do not: 87% of the marketing professors who completed our first study reported that they sometimes or often do not understand articles published within their research area. The fact that scholars struggle to understand articles within their own field suggests that unclear writing could be limiting the impact of academic research.
Researchers have attempted to study how writing influences impact by correlating the number of times an article is cited with its "readability," and they have measured readability by assuming that articles with longer words and sentences are more difficult to read (e.g., fog index, Flesch score; [ 2]; [17]; [28]; [45]). Do articles with higher readability scores have a larger impact? The answer is unclear. Three studies report that articles with more readable writing are less likely to be cited ([65]; [66]; [70]). Other studies, however, report no relationship between readability and citations ([19]; [42]) or that articles with a higher readability score are more likely to earn acclaim ([29]; [49]; [59]).
We believe that research on readability cannot tell us whether writing clarity reliably influences an article's scholarly impact because readability provides only a rough proxy of how easy an article is to understand ([ 4]; [41]). Sentence and word length may indicate whether an article is readable (i.e., easy to read) but not whether it is clear (i.e., easy to understand). Although long sentences are less readable, readers often better understand ideas that are connected in a single longer sentence rather than split into two shorter sentences ([14]). Conversely, writers can make sentences shorter—and thus more readable—without making them clearer. "A transaction transpired" is shorter than "Person A purchased product X from company Z," which in turn is shorter than "Homer bought a 4K Ultra High-Definition television from Best Buy." Readers can better understand Homer buying a high-def TV than a transaction transpiring, yet the shorter sentence is more readable.
We hypothesize that unclear writing is limiting the impact of scholarly research, yet readability measures are ill-equipped to identify this problem, let alone diagnose its underlying cause. If long words and sentences do not lead to unclear writing, then what does? We argue that scholars possess an asset that, when communicating their ideas, can become a curse: knowledge. As scholars become more knowledgeable about their own research, they have a harder time imagining what it is like to be a reader who is unfamiliar with it. This "curse of knowledge" prevents scholars from detecting when their writing is unclear.
Scholars spend years learning about their research area, honing their hypotheses, running their studies, and analyzing their results. They know a lot about their research. Once people know something, it is difficult for them to imagine what it is like to not know it ([20]; [56]). People use their own knowledge as a starting point when estimating what others know and often fail to account for the fact that others might not have access to the same information ([12]; [39]). For example, students who know that Napoleon was born in Corsica assume that a higher percentage of other students also know this fact ([48]).
The tendency to assume that others know what you know creates a curse of knowledge ([12]; [32]; [54]). Because researchers understand their theory, methods, and analyses, they tend to overestimate what their readers will understand. For example, after proposing that laughter is caused by "incongruity of knowledge from perception and abstract knowledge," [61], pp. 58–59) writes, "I shall not pause here to relate anecdotes as examples of this, for the purpose of illustrating my explanation; for this is so simple and easy to understand that it does not require them."
The curse of knowledge can prevent even all-star scholars like Schopenhauer from realizing when they are writing unclearly. When mixed with other motives, knowledge can create a cocktail of impenetrable prose. Knowledge, combined with length restrictions and the desire to articulate a theoretical contribution, leads to abstract writing ([39]; [68]). Knowledge, combined with a desire to signal competence, leads to technical writing ([ 2]; [10]). Finally, knowledge, combined with a desire to present research as being objective and general, leads to passive writing ([16]). Because scholars know a lot about their own work, it is difficult for them to recognize when abstraction, technical language, and passive writing are likely to confuse readers.
Abstraction refers to the process of thinking about tangible objects or activities as part of a broader, intangible category ([57]; [64]; [69]). Scholars write about abstract concepts, such as brand experience, satisfaction, postpurchase behavior, or, in Schopenhauer's case, the incongruity of knowledge from perception. Concrete concepts, in contrast, are things that we can see, feel, taste, smell, and hear, such as a brick building, a grease fire, or a puddle of melted ice cream. At a restaurant, customers can smell the caramelized onions, taste the flank steak, and feel the broken spring poking their leg through the seat cushion. They can feel a wet burning sensation if the waiter spills coffee on their lap and hear his half-hearted apology as they begin to thumb-type a one-star review into the Yelp app on their iPhone. This negative dining experience may have left the customers feeling dissatisfied, but they cannot hold a "negative experience" in their hand, nor can they pick up "dissatisfaction" and eat it, because these concepts are abstract.
As people learn more about something, they naturally begin to think about it more abstractly ([ 1]; [54]). Seasoned researchers see "service failures" rather than long lines and spilled coffee. They think "negative word of mouth" when a customer writes a one-star review or tweets about finding a fingernail in their chowder. Researchers need to write about abstract constructs to advance theory. Readers, however, will not understand the meaning of service failures, postpurchase behaviors, or other abstractions unless ( 1) they have previously mapped these abstract concepts onto concrete actions and sensations, or ( 2) the writing provides examples to help them do this ([54]). Not only are readers less able to understand abstract writing, they are also less likely to remember it ([ 3]; [58]). After reading excerpts as part of an experiment, participants were more likely to recall concrete phrases (e.g., "rusty engine") and sentences (e.g., "when an airplane blasts down the runway and passengers lurch backward in their seats") than abstract phrases (e.g., "subtle fault") and sentences (e.g., "moving air will push up against a surface placed at an angle to the airflow"; [ 3]).
Because of the curse of knowledge, however, it is easy for writers to forget that readers will struggle to connect abstract ideas to the actions and sensations that give them meaning. Consequently, as people become more knowledgeable about something, they often become worse at explaining it ([33]). For example, an experiment by [33] asked one group of participants (teachers) to explain how an electronic circuit works to another group of participants (students). Students made three times more mistakes and took 50% longer when guided by teachers with advanced training in electronics than when guided by teachers with less training. This occurred because the more knowledgeable teachers gave abstract instructions (e.g., "close the circuit"), but students better understood concrete instructions (e.g., "place the tip of the wire into the connector").
In summary, one reason why academic articles are difficult to understand is because they are abstract. Authors, who long ago made the journey from concrete (e.g., spilled coffee, one-star Yelp reviews) to abstract (e.g., customer service failure, postpurchase behaviors), forget that their readers were not along for the ride. As a result, authors often fail to ground their ideas with concrete examples, leaving readers stranded in the ether. We thus predict that scholars will be less likely to understand and cite articles that use more abstract writing.
Familiarity with a research topic not only causes scholars to think abstractly, it also unlocks a trove of technical vocabulary (i.e., jargon) that they can use to describe their research. Researchers instinctively use technical language, and they are especially likely to do so when they want to impress readers ([ 2]; [10]). However, technical language makes writing harder to understand. Technical language refers to words and phrases that are used by a particular profession or group but not by everyone else. Researchers develop technical terms so they do not need to repeat a longwinded phrase each time they refer to something ([54]). For example, "incongruity" is faster than writing "things that don't fit together," just as "marketization" uses fewer words than writing "when a country transitions from a planned economy to a market economy."
As researchers become more familiar with their research topic, they naturally begin to wield a more technical vocabulary. Instead of writing "People use what they see, hear, taste, smell, and feel to understand the world," they write sentences like "Abstract rational knowledge is the reflex of the representation from perception" ([61], p. 58). Researchers need to use technical language, but they also need to calibrate it to their target audience; otherwise, technical language will limit the number of readers who understand the research ([54]). The sentence "the ANOVA confirmed that we operationalized our manipulations sufficiently" will make sense to experimenters, but it might lose ethnographers, econometricians, and managers. Likewise, the sentence "Assemblage theory provides a bottom-up framework for analyzing social complexity by emphasizing fluidity, exchangeability, and multiple functionalities" (https://en.wikipedia.org/wiki/Assemblage%5ftheory) will confuse anyone who is not already familiar with assemblage theory.
Because of the curse of knowledge, researchers tend to overestimate the amount of technical language that their readers will understand. Even when readers understand technical language, they need to work harder to process it, which leaves them less able to comprehend, remember, and think about the researchers' focal idea ([ 1]; [50]). Furthermore, rather than making the researcher sound smarter, using technical language in place of simpler words can cause readers to doubt the researcher's intelligence. Stanford University students who read essays with technical words (e.g., "institutional," "development") understood less of the essay and thought the writer was less intelligent than students who read essays with simpler language (e.g., "social," "advance"; [50]).
In summary, a second reason why academic writing is unclear is because researchers overuse technical language. We predict that scholars will be less likely to understand and cite articles that use technical compared with colloquial writing.
Consider the following description of an experiment: "A press release was read about a new product with emphasis placed on quality. They were then given notification of the release date." Does "quality" refer to the quality of the product or press release? What about "notification of the release date"? Did the press release or the experimenters emphasize its quality? Who or what told participants about the release date? The authors of this fictional methods section know the answer to all of these questions. Yet, just as familiarity with their theory can cause authors to overuse abstract and technical language, familiarity with their methods can enable a third practice that makes academic writing unclear: passive writing.
We use the term "passive" to describe writing that obscures who is doing something or what is being done. Most sentences include some person, place, or thing (i.e., an actor) performing an action, but writers can hide these actors and actions through passive writing. For example, the sentence "An exclusion manipulation was administered" identifies neither who administered the manipulation nor whom the manipulation was administered to. Moreover, the words "exclusion" and "manipulation" obscure actions by disguising verbs, "exclude" and "manipulate," as nouns. Active writing, conversely, clearly names the actors and the actions they perform: "The experimenter manipulated whether the participants felt excluded."
One form of passive writing that is familiar to most academics is passive voice. Whereas active voice begins by naming the actor that performs the action (e.g., "the experimenter conducted research," or "people know things"), passive voice demotes the actor to a supporting role (e.g., "research was conducted by the experimenter," or "it is known by people") or eliminates the actor altogether (e.g., "research was conducted," or "it is known"; [ 7]). Writers can use passive voice to hide the actors in their sentences. "It was hypothesized," for example, obscures who did the hypothesizing.
Passive voice is not the only way that writers can hide the actors and actions in their sentences. The sentence "Abstract rational knowledge is the reflex of the representation from perception" ([61], p. 58) uses active voice but does not reveal who or what is knowing, representing, or perceiving. Similarly, an author who writes "We study the effects of corporate sociopolitical activism on owned media engagement" obscures that she is studying how consumers respond when businesses post political messages online.
Research suggests that readers are less likely to understand passive writing ([ 7]; [15]; [63]). [ 8], for example, created two versions of an essay. One version used active writing (e.g., "Researchers conclude that adding sulfur..."); the other used passive writing (e.g., "The conclusion of researchers is that the addition of sulfur..."). Participants read the active version faster and remembered it better.
In summary, a third reason why academic writing is difficult to understand is because it is too passive. We predict that scholars will be less likely to understand and cite articles that use more passive writing.
We conducted three studies designed to answer the following questions. ( 1) Do scholars understand less of academic articles that use more abstract, technical, and passive writing? We examined this question in Study 1 by asking marketing professors to read and evaluate an excerpt from an article with a high, average, or low amount of abstract, technical, and passive writing. ( 2) Do articles that are easier to understand have a larger impact than articles with unclear writing? We examined this question in Study 2 by analyzing the relationship between writing clarity, readability, and citations in articles published in the Journal of Marketing (JM), Journal of Marketing Research (JMR), and Journal of Consumer Research (JCR). ( 3) Does knowledge (i.e., familiarity with their own work) prevent researchers from realizing that their writing will be difficult for readers to understand? We examined this question in Study 3 by testing whether PhD students are less likely to detect unclear writing when they describe their own research compared with when they describe a colleague's research. Figure 1 illustrates our conceptual model and predictions. Note that we begin by testing the effect of writing practices on reader understanding (Study 1) and research impact (Study 2) before directly testing the hypothesis that knowledge enables unclear writing (Study 3).
Graph: Figure 1. Conceptual model and predictions.
To test whether articles with abstract, technical, and passive writing are more difficult to understand and less likely to make an impact, we analyzed the text of 1,640 articles published in JM, JMR, and JCR between 2000 and 2010.[ 7] We used 2010 as a cutoff to allow a minimum of ten years for the audience to read, learn from, and cite the articles. The sample included 428 articles from JM, 562 articles from JMR, and 650 articles from JCR. Similar to previous research, we measured the readability of each article. In addition, we attempted to measure three practices that readability indices do not capture but that we predict will make academic writing unclear: abstraction, technical language, and passive writing.
Recall that abstraction refers to thinking about tangible objects or activities as part of a broader, intangible category ([57]; [64]; [69]). Writing that uses concrete words and more examples is less abstract. We therefore operationalized abstraction by measuring ( 1) the extent to which the words in the article were concrete and ( 2) the number of examples the article uses per page.
To calculate the extent to which each word in the article was concrete or abstract, we used an established list of concreteness ratings by [11]. These authors measured the concreteness of 39,954 English words and two-word phrases (e.g., "zoom in," "pin up") by asking 4,000 Amazon Mechanical Turk respondents to rate a subset of words on a five-point scale from "abstract" (coded as 1) to "concrete" (coded as 5). For example, the word "logo" received a score of 4.41, whereas the word "equality" received a score of 1.41. This list provided a concreteness rating for most of the words in the articles; we then calculated an overall concreteness rating for each article by using the average concreteness rating for all of the words that had been rated in the Brysbaert, Warriner, and Kuperman study.
We measured examples by counting how many times the article used the following phrases: "for example," "for instance," "namely," "e.g.," "as in," and "such as." To control for article length, we divided the total number of example phrases by the number of pages.
By definition, technical words (e.g., "manipulation," "endogeneity") are less likely to appear in websites, blogs, and Facebook posts than in academic journals. We therefore operationalized technical language by measuring how frequently the words in the article are used in other writing, on average (i.e., frequency). We used the frequency with which a word appears as a sign that a word is not technical. We hypothesize that articles with frequently used words will be easier to understand and have a larger impact than articles with infrequently used words.
We assigned a frequency score to each word in each article using a database collected by Peter Norvig, Google's director of research ([62]). The database lists the number of times that the 50,000 most frequently used words in the English language appear in the Google Web Trillion Word Corpus ([ 9]). To make the word frequency measure easier to interpret, we normalized the raw count measure by ( 1) assigning the most common word ("the") a score of 1, ( 2) calculating a score for the other 49,999 words on the list by dividing the number of times each of these words appeared by the number of times "the" appeared, and ( 3) assigning a score of 0 to any words that did not crack the top 50,000. Thus, every word in the article had a frequency score ranging from 0 to 1. We calculated a score for each article by averaging the frequency score of all of its words.
Passive writing refers to a menagerie of styles that mask the actors and actions in a sentence. We operationalized passive writing by measuring the percentage of each article that used passive voice. We measured passive voice rather than attempting to capture all of the ways that authors write passively because passive voice is prevalent in academic writing, easy to measure, and almost always makes it harder for readers to identify who is doing what in a sentence ([63]). Other forms of passive writing (e.g., transforming a verb into a noun) are tougher to measure. We thus relied on passive voice as a proxy for all passive writing and hypothesized that articles that use more active voice will be easier to understand and have a larger impact.
We took three steps to calculate an active voice score for each article. First, we counted the total number of sentences in each article. Next, we used a pattern-matching package in Python to classify whether each sentence used a form of the verb "to be" followed by a past participle of another verb (e.g., "participants were given instructions"). We coded these sentences as being passive. Finally, we calculated the ratio of active voice in the paper as follows:
Graph
1
The first objective of Study 1 was to test whether scholars are less likely to understand articles that use more abstract, technical, and passive writing. To do this, we measured the concreteness, examples, word frequency, and active voice in the first page of each article in our sample. We selected the article excerpts that scored highest, closest to average, and lowest on these four measures in each of three subfields in marketing: consumer behavior, strategy, and quantitative modeling. We then asked marketing scholars to read one of the article excerpts and indicate the extent to which they understood what they read. We predicted that scholars would understand less of excerpts with more abstract, technical, and passive writing. The second objective was to explore how scholars respond when they understand less of an article, which we did by asking about their impression of both the article and its authors.
We recruited marketing academics to complete an online survey by emailing 2,771 tenured or tenure-track marketing faculty employed by the top 300 universities on University of Texas at Dallas's business school global ranking. We also posted a link to our survey on our LinkedIn accounts and on Facebook groups that target marketing academics. In total, 266 participants completed the focal dependent variable and 255 completed the survey.[ 8] We thanked participants by donating $510 to the United Food Bank, $2 on behalf of each who completed the study. We provide details about the participants in the Web Appendix.
The study used a 3 (clarity: low, average, high) × 3 (research area: strategy, modeling, consumer behavior) between-subjects design. Participants read an excerpt from one of nine academic articles. We included only the first page to page and a half of each article to keep the excerpts the same length and the survey short. To get three excerpts from each subfield, and to keep the articles as similar as possible within each journal, we considered only JM papers in the strategy subfield, JMR papers in the modeling subfield, and JCR papers in the consumer behavior subfield. Then, we used the algorithm described in Web Appendix C4 to select the articles that scored the highest, lowest, and closest to average on our four clarity measures.
Participants read an article excerpt and briefly summarized its research question and intended contribution. Participants then indicated the percentage of the excerpt that they understood (%Understood),[ 9] the extent to which they agreed that the writing was clear (Understandable), and the percentage of the article that they thought the average practitioner would understand (%Practitioner Understanding). Participants next rated their opinion of the article (Impression of the Article) and their inferences about the authors (Impression of the Authors). On the following page, participants read title and author of the article and indicated if they were familiar with it. The results are similar (significance does not change) if we include familiarity as a covariate. We next collected exploratory measures by asking participants to reread and evaluate the clarity of the first page of their most recently published article. We report the details about these exploratory measures and results in Web Appendix C9. Next, we measured whether the participants read an article from within their research area by asking if they were most familiar with consumer behavior (60.2%), quantitative/modeling (10.5%), strategy/managerial (24.6%), or another research area (4.6%).[10] Finally, we asked participants about their goals and beliefs related to scholarly writing, current academic position, years since completing their doctorate, gender, age, and English ability. We describe all of the measures in Web Appendix C3.
We sent a posttest survey to the same pool of marketing professors to check whether the clear, average, and unclear articles in the experiment differed along dimensions other than writing clarity. Posttest participants (N = 107) read the title, author, and abstract from the three high-clarity articles, the three average-clarity articles, or the three low-clarity articles. Participants rated the quality of the research and prestige of the authors based on the title, abstract, and author information. Neither ratings of research quality (Mclear = 4.58, Maverage = 4.22, Munclear = 4.40; F( 2, 104) = 1.40, p =.25) nor ratings of author prestige (Mclear = 4.94, Maverage = 4.81, Munclear = 4.90; F( 2, 104) =.131, p =.88) differed significantly across the clear, average, and unclear articles. We provide details about the posttest in Web Appendix C5.
We examined the extent to which participants understood the excerpt using a 3 (research area: strategy, modeling, consumer behavior) × 3 (clarity: high, average, low) × 2 (research match: match, mismatch) analysis of variance. As we predicted, participants understood a higher percentage of the articles with high clarity (82.0%) than articles with average (73.7%) or low (68.5%) clarity (F( 2, 248) = 6.18, p =.002, η2 =.047; see Table 1). Participants also understood a higher percentage of the article when it matched their research (82.4% vs. 71.2%; F( 1, 248) = 7.91, p =.005, η2 =.031), but clarity and research match did not interact (F( 2, 248) =.37, p =.69, η2 =.003). We observed similar results for the supplemental understanding measures (for details, see Table 1). Research area did not have a main effect (F( 2, 248) = 1.03, p =.36, η2 =.008), nor did it interact with the other factors (ps >.10). Because research area had neither a significant main effect nor an interaction, we do not discuss it further. Interested readers can find the means and standard deviations for all of the conditions in Web Appendix C6 and Web Appendix C7.
Graph
Table 1. The Effects of Writing Clarity and Research Match in Study 1.
| Writing Clarity | Research Match |
|---|
| Low Clarity | Average Clarity | High Clarity | Outside the Reader's Area | Inside the Reader's Area |
|---|
| % understood(0 to 100%) | 68.5a (25.3) | 73.7a (21.0) | 82.0b (15.9) | 71.2a (21.6) | 82.4b (20.2) |
| Understandable(1 to 7) | 3.86a (1.71) | 4.04a (1.63) | 4.80b (1.42) | 3.99a (1.60) | 4.79b (1.59) |
| % practitioner understanding(0 to 100%) | 37.3a (24.6) | 42.7a (25.1) | 54.6b (26.1) | 44.5a (26.3) | 46.1a (26.4) |
| Impression of the article(1 to 7) | 3.51a (1.67) | 3.21a (1.68) | 4.11b (1.68) | 3.26a (1.64) | 4.40b (1.60) |
| Impression of the authors(1 to 7) | 4.93a (1.18) | 5.43b (1.04) | 5.47b (1.02) | 5.21a (1.16) | 5.39a (1.00) |
1 Notes: This table shows the means (standard deviations in parentheses) for each measure in the low-clarity, average-clarity, and high-clarity conditions and the match and mismatch conditions.
2 Means with different superscript letters are significantly different from one another (p <.05).
Did writing clarity influence participants' impression of the article? Yes. Participants formed a better impression of the excerpts with clear writing than the excerpts with average or unclear writing (Mclear = 4.11, Maverage = 3.21, Munclear = 3.51; F( 2, 245) = 3.59, p =.029, η2 =.028). Consistent with our prediction that scholars do not like unclear articles because they do not understand them, the %understood measure mediated the effect of writing clarity on impression of the article (indirect effect:.30, 95% confidence interval = [.17,.45]; for details, see Web Appendix C8). Participants also had a more favorable impression of the articles within rather than outside their research area (M = 4.40 vs. M = 3.26; F( 1, 248) = 20.96, p <.001, η2 =.079), but participants enjoyed the clear articles more regardless of whether the research matched their interest (clarity × match interaction: F( 1, 248) =.11, p =.89, η2 =.001).
One reason why scholars write technically is to signal status and competence ([10]). Did technical writing help writers appear more competent? Not in our sample. Participants who read the excerpts with clear writing thought that the authors were more competent (Mclear = 5.47, Maverage = 5.43, Munclear = 4.93; F( 2, 245) = 3.12, p =.046, η2 =.025). Furthermore, the %understood measure mediated the effect of writing clarity on impression of the author (indirect effect:.11, 95% confidence interval = [.05,.19]), which indicates that scholars perceived the authors to be more competent because they better understood the writing (details in Web Appendix C8). Competence ratings did not depend on whether the article was in the participant's research area (main effect: F( 1, 245) =.51, p =.48, η2 =.002; interaction: F( 2, 245) =.44, p =.64, η2 =.004).
The Flesch score and fog index were only modestly correlated with our measures of clarity (see Web Appendix C6). Moreover, when we entered the readability scores as covariates in a 3 (clarity) × 2 (research match) analysis of covariance, neither the Flesch score (F( 2, 258) =.07, p =.80, η2 =.000) nor the fog index (F( 2, 258) =.39, p =.53, η2 =.002) predicted the amount that scholars understood, whereas both clarity (F( 2, 258) = 3.28, p =.039, η2 =.025) and research match (F( 1, 258) = 18.12, p <.001, η2 =.066) remained significant.
Study 1 shows that marketing scholars do not always understand the articles published in top marketing journals, especially when the articles use more abstract, technical, and passive writing. Importantly, the extent to which an article used concrete language, examples, common words, and active voice predicted how much of the article readers understood. This was true even when scholars read articles from their own subdiscipline, which shows that unclear writing prevents articles from effectively communicating even to a niche audience of like-minded scholars. Scholars who understood less of an article formed a worse impression of both the article and its authors. These results offer preliminary evidence for our hypothesis that academic articles with unclear writing make a smaller impact. We directly test this hypothesis in Study 2.
Our second study analyzed the full sample of 1,640 articles published in JM, JMR, and JCR between 2000 and 2010 to test whether articles that use more abstract, technical, and passive writing have a smaller impact than articles with clear writing.
We tested the extent to which our four writing clarity measures—concreteness, examples, word frequency, and active voice—predicted the number of times an article was cited after controlling for as many relevant variables as we could collect. We initially analyzed the entire article text but had two concerns with doing so: ( 1) the writing in the methods and results sections might necessarily be technical and ( 2) differences in the writing in these sections might depend more on the article's methodology than its writing clarity. Thus, we also tested whether our results were robust by analyzing only the text from the title to the beginning of the methods section. For brevity, we report the full-text analysis here and the pre–methods section analysis in Web Appendix D6.
We measured impact by collecting the number of times the article had been cited on Google Scholar (Google_Citation) and Web of Science (WoS_Citation) as of May 10, 2020.
We controlled for a variety of factors that could potentially influence how often an article is cited but that are not necessarily related to abstract, technical, or passive writing. We briefly describe the control variables here and provide details in Web Appendix D1.
To test whether abstract, technical, and passive writing influence citations over and above the effect of readability, we controlled for two popular readability measures: the Flesch reading ease score ([21]; [23]; [59]; [65]; [66]) and the fog index (e.g., [ 2]; [25]; [59]). Because the Flesch and fog measures are highly correlated, including both measures in the same model could lead to multicollinearity. We thus used the Flesch score in the primary analyses (Flesch), and ran supplementary analyses using the fog index instead. The results were similar in both analyses (see Web Appendix D7).
Articles with high-quality and relevant research have a larger impact than articles with lower quality and relevance ([66]; [67]). We thus controlled for five factors related to this "universalist perspective" ([66]). First, we coded whether the article won an award (Award). Second, we coded for whether the article used each of the following methods: econometrics models, survey data, experiments, qualitative research, and meta-analyses. Third, we coded the topic of the paper by creating a set of dummy variables to classify the article into one of 11 topics. Finally, we controlled for the length of the article (NumberofPages) and whether it was a lead article (LeadArticle).
Previous research suggests that the impact of an article depends on factors related to its presentation, aside from its readability ([37]; [38]; [66]; [70]). We followed this research by controlling for ( 1) the number of acronyms per page (Acronyms), ( 2) the length of the title (Title: Length), ( 3) whether the title of the article used words such as "marketing" and "new" (Title: Attention_Grabber), ( 4) whether the title included a colon (Title: Colon), ( 5) whether the title included a question mark (Title: Question), ( 6) the number of tables in the article (NumberofTables), ( 7) the number of figures in the article (NumberofFigures), and ( 8) whether the article included one or more appendices (Appendix).
The impact of an article also depends on "social constructivist factors," including the journal in which it is published and the prestige of the authors ([65]; [66]). We used dummy variables to control for the source of the article, including whether the article was published in JM, JMR, or JCR and the specific issue in which the article was published (Issue). We also controlled for four variables related to the authors and the citation network: ( 1) the rank of the authors' university (AuthorRank), ( 2) the number of authors (NumberofAuthors), ( 3) the number of references in the article (NumberofReferences), and ( 4) the recency of the references in the article (AgeofReferences).
Scholars have more time to cite articles that were published earlier. Thus, we controlled for both the linear (Quarters) and quadratic (Quarters2) amount of time that had passed since the article was published ([66]).
We manually coded a random subsample of 100 articles to ensure that the computer algorithms measured the independent and control variables accurately. The hit rate was 86%, a level that the literature considers acceptable ([ 5]).
We tested the effects of the independent and control variables on Google Scholar citations and Web of Science citations. To make it easier to compare the size of the regression coefficients, we standardized the independent and control variables by setting their means to equal 0 and their standard deviations to equal 1.
The articles in our sample were cited an average of 415 times on Google Scholar and 155 times on the Web of Science. The median citation counts, however, were 204 and 80, respectively, and the standard deviations were 740 and 277, which indicates that the distributions were overdispersed. The citation count measures were not zero inflated, as all of the articles were cited at least once on Google Scholar, and only one was not cited on Web of Science. The model that is most appropriate given these properties is negative binomial regression[11] ([26]; [40]), which is what we used to predict citations depending on the independent measures and control variables:
Graph
2
where i indicates the article, Citation is Google_Citation in Model 1 and WoS_Citation in Model 2, and is the robust error term.
We report descriptive statistics and the correlations between variables in Web Appendix D2. The variance inflation factor scores for all of the variables had an average of 1.24 and a maximum of 2.13, which was well below 10, indicating that multicollinearity was not a problem.
We assessed the effects of the independent and control variables on the number of times an article was cited on both Google Scholar (Model 1) and Web of Science (Model 2) (see Table 2). Articles were cited more when they used more concrete words (Google: β =.081, p =.025; WoS: β =.10, p =.005) and examples (Google: β =.083, p =.033; WoS: β =.080, p =.037), which is consistent with the hypothesis that abstraction limits the impact of an article. Articles that used common words were cited more often than articles that used uncommon words (Google: β =.13, p <.001; WoS: β =.10, p =.004), which is consistent with the hypothesis that technical language limits the impact of an article. In addition, articles that used more active voice were cited more often (Google: β =.11, p <.001; WoS: β =.11, p =.001), which is consistent with the hypothesis that writers who do not clearly state who is doing what will have less impact. In contrast, yet consistent with previous research ([65]; [66]; [70]), articles with higher readability scores were cited less often (Google: β = −.069, p =.061; WoS: β = −.073, p =.049). In summary, articles with concrete language, examples, active voice, and common words were cited more; articles with higher readability scores were not.
Graph
Table 2. The Relationship Between Writing Clarity, Readability, and the Control Measures on Citations in Study 2.
| Sample: 1,640 Articles Published in JM, JMR, and JCR between 2000 and 2010 | Model 1 | Model 2 |
|---|
| DV: Google Citation | DV: Web of Science Citation |
|---|
| β | SE | β | SE |
|---|
| Writing | Clarity: Concrete (Not abstract 1) | .081** | (.03) | .10*** | (.03) |
| Clarity: Examples (Not abstract 2) | .083** | (.03) | .080** | (.04) |
| Clarity: Frequency (Not technical) | .13*** | (.03) | .10*** | (.03) |
| Clarity: Active voice (Not passive) | .11*** | (.03) | .11*** | (.02) |
| Readability: Flesch Score | −.069* | (.03) | −.073** | (.03) |
| Controls related to the research | Award | .12*** | (.03) | .14*** | (.03) |
| Method: Econometrics | .025 | (.03) | .031 | (.04) |
| Method: Survey | −.058* | (.03) | −.046 | (.02) |
| Method: Experiment | −.0053 | (.03) | −.023 | (.03) |
| Method: Qualitative | .066 | (.04) | .081* | (.05) |
| Method: Meta-Analysis | −.023 | (.02) | −.022 | (.02) |
| Number of Pages | .12*** | (.04) | .11** | (.04) |
| Lead Article | .044 | (.03) | .070* | (.04) |
| Controls related to the presentation | Acronyms | .11** | (.03) | .12*** | (.03) |
| Title: Length | −.13*** | (.03) | −.10*** | (.03) |
| Title: Attention Grabbers | −.076** | (.03) | −.060* | (.03) |
| Title: Colon | .070** | (.03) | .059* | (.03) |
| Title: Question | .060* | (.02) | .055* | (.03) |
| Number of Tables | −.072** | (.03) | −.11*** | (.03) |
| Number of Figures | .041 | (.03) | .039 | (.03) |
| Appendix | −.024 | (.02) | −.037 | (.02) |
| Controls related to the authors and source | Author Rank | −.062** | (.03) | −.058** | (.02) |
| Number of Authors | .023 | (.02) | .022 | (.02) |
| Number of References | .18*** | (.04) | .18*** | (.04) |
| Age of References | −.029 | (.02) | −.028 | (.02) |
| Journal 1: JM vs. JMR | .40*** | (.09) | .37*** | (.08) |
| Journal 2: JCR vs. JMR | −.0038 | (.08) | −.00093 | (.07) |
| Linear Time: Quarters | .36*** | (.04) | .34*** | (.03) |
| Quadratic Time: Quarters2 | −.0061 | (.03) | −.031 | (.03) |
| Fit | Log-likelihood | –11,148.07 | –9,576.01 |
| χ2 | 722.51*** | 661.59*** |
- 3 *p <.1.
- 4 **p <.05.
- 5 ***p <.01.
- 6 Notes: For brevity, we omitted the effects of the dummy variables for topics and issues. We provide the complete table of results, and comparative models with goodness-of-fit statistics in the Web Appendix D4.
To what extent does abstract, technical, and passive writing influence the number of times an article is cited? The nature of negative binomial models makes it difficult to interpret the size of the coefficients; however, these models do let us approximate how, all else being equal, an article with average writing differs from an article with above- or below-average writing. For example, an article published in JM that is one standard deviation above average on all four measures of clear writing (e.g., from 88% to 95% active voice, from 2.33 to 3.66 examples per page) earns 157 more citations on Google Scholar and 60 more citations on Web of Science than an article with average clarity. Conversely, a JM article that is one standard deviation below average earns 56 fewer citations on Google Scholar and 16 fewer on Web of Science than an article with average clarity.
We tested whether our results were robust across different ways of analyzing the data. First, we examined if the results changed when we analyzed only the text from before the method section. The results were similar regardless of whether we analyzed the entire article or just the text before the method section (see Web Appendix D6). Second, we examined if the results changed when we measured readability using the fog index rather than the Flesch score. The results were similar regardless of whether we controlled for readability using the Flesch score or the fog index (see Web Appendix D7). Third, we examined if the results changed if we used a different method to account for the unobserved variance related to the journal, issue, and year in which the article was published. In the negative binomial models, we addressed the endogeneity caused by unobserved factors by calculating the time trend and including dummy variables for the journal and issue. A more common approach to control for the time-specific and journal-specific factors in a panel data set is the fixed-effect method ([24]; [31]). We did not use the fixed-effect method in our main analysis because the unconditional fixed effects estimators produce inconsistent and biased estimates in nonlinear models, including negative binomial models ([26]). To test the robustness of our model choice, we analyzed the data using linear fixed-effect panel regressions after transforming the citation variables from integers to natural continuous numbers by performing a logarithmic transformation on the dependent variables. The results were similar regardless of whether we controlled for unobservable variance with dummy variables or fixed effects (see Web Appendix D8). Finally, we investigated the effect of writing clarity on the likelihood that an article wins an award, as an alternative proxy for impact. The analysis with awards as the dependent variable produced similar results, although the effects were smaller compared with the analysis with citations as the dependent variable (see Web Appendix D9).
Academic articles with abstract, technical, and passive writing were cited less often than articles with concrete, nontechnical, and active writing. The size of these effects was not trivial, and the results were robust across different ways of analyzing the data. The results complement Study 1 by showing that articles have a larger impact when their writing is easier to understand.
Why do scholars write unclearly? We have argued that the curse of knowledge enables abstract, technical, and passive writing by preventing scholars from realizing when their writing is unclear. We test this hypothesis next.
Scholars need to be knowledgeable about their work to conduct effective research. Yet we hypothesize that this very knowledge can curse scholars by enabling abstract, technical, and passive writing. If we are correct, then knowing more about a research project will make scholars less likely to recognize when their writing about the project is unclear. In Study 3, we tested this prediction by asking marketing PhD students to summarize both their own research and research by a colleague. The students subsequently rated the clarity of their summaries. We compared the students' subjective ratings of writing clarity with ( 1) the clarity measures from Studies 1 and 2, and ( 2) clarity ratings from two scholars who read the summaries. We predicted that students would be more likely to overestimate the clarity of their writing when describing their own research than when describing their colleague's research.
We sent an email asking 688 students enrolled in marketing PhD programs in North America to participate in our study for a chance to win an Amazon gift card. Forty-eight students completed part of the study, and 47 finished it.
Participants wrote two summaries: one about their own research project (high-knowledge condition) and one about a colleague's research project (low-knowledge condition). We used a within-subject design and counterbalanced the order of the high- and low-knowledge conditions.
Before writing the high-knowledge summary, participants read, "Please write one paragraph (at least 3-5 sentences) summarizing the research project on which you have spent the most time." Before writing the low-knowledge summary, participants read, "Please write one paragraph (at least 3-5 sentences) summarizing a research project by a fellow PhD student (either in your program or at another school). Pick a research project that you are familiar with but that you are not a collaborator on." We told participants to summarize each project "so that other marketing scholars will understand" it. We asked them to use a third-person voice (e.g., "the authors found...") without naming themselves or their colleagues (for complete instructions, see Web Appendix E). A minority of the participants (18%) did not use an impersonal, third-person voice. To hide the condition of these summaries from the human raters, we changed the personal pronouns (e.g., I, we), names (e.g., Albert, Dr. Einstein), and relationship markers (e.g., my colleague, my classmate) to third-person pronouns and generic labels (e.g., they, the authors).
Participants answered four questions after writing each summary. Two were manipulation checks: "How knowledgeable are you about the research you described in this summary?" and "How familiar are you with this research?" The other two measured the extent to which participants thought that their writing was clear (i.e., writer-rated clarity): "How clear do you think this summary will be to other PhD students in marketing?" and "How well do you think other PhD students in marketing will understand this summary?" All measures used seven-point scales (e.g., 1 = "not at all clear" and 7 = "extremely clear"). Finally, we asked participants about their research area, year in the PhD program, geographic location, English fluency, age, and gender. The results did not depend on the order in which participants wrote the essay, their research area, year in the program, age, gender, or proficiency in English (i.e., none of the interactions were significant).
We assessed writing clarity using machines and human readers. As in Studies 1 and 2, we calculated a machine-rated clarity score by averaging the measures of concreteness, examples, frequency, and active voice. We also calculated a reader-rated clarity score by averaging the ratings of two advanced PhD students, who read the summaries and assigned each a score between 1 ("least clear") and 7 ("most clear) (See Figure 2). The ratings were correlated even though the readers had different areas of expertise (r =.47, p <.001): one was a consumer psychologist, the other was an empirical modeler. The readers were not aware of the purpose of the study or the condition of the summaries; that is, they did not know whether or not the participant was writing about their own research.
Graph: Figure 2. The mean clarity scores in Study 3.Notes: Error bars = ±1 SE. The columns on the left illustrate how PhD students rated their own writing about both their own research and their colleague's research on a 1–7 scale. The columns in the center illustrate how clear other PhD students thought the writing was on a 1–7 scale. The columns on the right illustrate how clear our clarity algorithm thought the writing was on a normalized scale.
The knowledge manipulation worked. Participants reported that they were more familiar with their own research than their colleague's research (M = 6.49 vs. 4.01; t(46) = 11.73, p <.001, Cohen's d = 1.71). Note that we used paired-samples t-tests to compare the high-knowledge and low-knowledge means in this study.
Consistent with the hypothesis that knowledge makes researchers believe their writing is clearer, writers thought that their description of their own research was clearer than their description of a colleague's research (Mhigh knowledge = 5.47, Mlow knowledge = 4.93; t(46) = 2.84, p =.007, Cohen's d =.41).
In contrast to writers' subjective clarity ratings, readers understood directionally less when writers summarized their own research than when they summarized a colleague's research (Mhigh knowledge = 4.24, Mlow knowledge = 4.51; t(46) = −1.21, p =.23, Cohen's d = −.18).
The curse of knowledge was even more apparent in the extent to which writers used abstract, technical, and passive writing, as measured by our clarity algorithm. Participants wrote significantly less clearly, according to our algorithm, when they summarized their own research than when they summarized a colleague's research (Mhigh knowledge = −.11, Mlow knowledge =.12; t(47) = −2.81, p =.007, Cohen's d = −.41).
PhD students were more likely to overestimate the clarity of their writing when they were more familiar with the research project that they summarized. To test whether this finding would generalize, we conceptually replicated the results in Study 4, which we report in Web Appendix F. We recruited college-educated workers on Prolific and taught half of the workers about research on social exclusion and the other half about research on advertiser credibility. We then asked all of the workers to proofread a report about social exclusion research. Prolific workers who had previously learned about social exclusion research were less likely to realize that abstract, technical, and passive sentences in the report would be unclear to other readers.
The results of both Study 3 and Study 4 confirm that as scholars become more knowledgeable about a research project, they become less likely to realize when their writing about it is unclear. How can scholars exorcise this curse of knowledge? How can they write so their research will be more likely to make waves in the marketplace of ideas? We begin to answer these questions next.
Why do many articles gain little share in the marketplace of ideas ([43]; [53]; [71])? One reason is the writing. Consider again the title "The Interactive Effects of Ideological Orientation and Corporate Sociopolitical Activism on Owned Media Engagement." This style permeates academia, yet it is difficult to understand because it is abstract, technical, and passive. "Corporate sociopolitical activism" is more abstract than "controversial political messages." "Ideological orientation" is more technical than "liberal or conservative." And "owned media engagement" obscures that the research investigates how consumers respond (actor/action 1) when firms post social media messages (actor/action 2).
We show that readers are less likely to understand articles that use more abstract, technical, and passive writing (Study 1). Consequently, articles that use abstract, technical, and passive writing are less likely to be cited (Study 2). The curse of knowledge enables unclear writing by preventing scholars from realizing when their writing is unclear. As they become familiar with a research project, scholars use more abstract, technical, and passive writing, and are less likely to realize when readers will struggle to understand (Study 3).
How can scholars overcome this curse? How can they write in a way that reaches experimentalists and econometricians, sociologists and neuroscientists, PhD students and professors emeriti, chief executive officers, brand managers, educators, journalists, and policy makers? First, they need to acknowledge that unclear writing is a problem. Second, they need to understand how to fix it.
Unclear writing is a problem ([34]; [55]). Marketing scholars could not understand 24% of the opening page of a typical article published in a premier marketing journal (see Study 1), and 87% reported that they sometimes, often, or always struggle to understand academic articles published within their research area (see Web Appendix C2). If even marketing scholars struggle to understand the articles published in marketing journals, what hope do scholars from neighboring disciplines, let alone practitioners, have?
The first step toward improving academic writing is admitting there is a problem. The curse of knowledge, however, prevents scholars from seeing this problem, at least in their own writing. Marketing professors who completed Study 1, for example, believed that their writing was clearer than the average writing in a premier marketing journal (see Web Appendix C9). Similarly, PhD students who described their own research in Study 3 overestimated the extent to which other PhD students would understand their writing.
In addition to the curse of knowledge, there are at least three other misguided beliefs, or myths, that prevent scholars from recognizing that unclear writing is a problem.
Some scholars believe that inaccessible writing is not a problem because they think that articles should only speak to a narrow audience of scholars. Data from Study 1, however, reveal that most marketing scholars want their research to reach a broad audience: 96% of the scholars who completed Study 1 indicated that they try to write so that most scholars can understand their ideas, 66% said their research targets scholars in different academic areas, and 72% said that their research targets practitioners (see Web Appendix C10). Journal editors, similarly, aim to publish research that reaches a broad audience. Their editorials have encouraged marketing scholars to publish boundary-breaking research that influences both other scholarly disciplines and marketing practice ([36]; [47]). We show that ideas are more likely to break out of their subdisciplinary niche if scholars describe them clearly.
Scholars could be writing unclearly because they believe that readers will respond more favorably to unclear writing. They might think that technical language will make them look smart ([ 2]; [10]; [55]) or that unclear writing engages readers by forcing them to work harder to understand it ([60]). Previous research appeared to support these beliefs by showing that ( 1) articles with lower readability scores are cited more often ([65]; [66]; [70]) and ( 2) journals with lower readability scores are considered more prestigious ([ 2]). Our research, however, suggests that these results are misleading because readability measures do not capture the problems that make academic writing unclear. We found only a weak relationship between readability measures, including the Flesch score and fog index, and measures of abstraction, technical language, and passive writing. Moreover, articles with less abstract, technical, and passive writing were more likely to be understood and cited; articles with higher readability scores were not.
JM, along with dozens of writing guides, advises scholars to use active voice and avoid technical language (https://www.ama.org/submission-guidelines-journal-of-marketing). Does the field need another reminder to avoid abstract, technical, and passive writing? We think so. The problem is that even if scholars believe that writing should be clear in theory, the curse of knowledge can prevent them from realizing when their writing is unclear in practice. Consequently, many continue to use abstract, technical, and passive writing despite repeated warnings. For example, most of the titles of the articles published in the May/June 2020 issues of JM, JMR, and JCR used these unclear writing practices. [51] offer a notable exception. Rather than settle for an abstract, technical, and passive title, such as "Option Assortments: The Effect of Search Volume on Information Acquisition," they wrote a concrete, nontechnical, and active title: "Product Lineups: The More You Search, The Less You Find."
How can scholars write more clearly? The most common answer is to try harder. The literature suggests that extra practice, effort, care, and revision will cure unclear writing ([55]; [59]; [60]). Effort is necessary, but our research suggests that effort alone will not lift the curse. In fact, effort will make scholars even more familiar with their writing, which may perversely make it harder for them to realize when it is too abstract, technical, passive, or otherwise unclear. Although effort alone may not lift the curse of knowledge, there are three steps that scholars can take to keep it at bay.
First, scholars should ask someone else to read their writing. It is better if this reader is not already familiar with the research, as knowledgeable readers will be more likely to glide through abstract, technical, and passive writing that will confuse less knowledgeable readers.
Second, scholars can check the extent to which their writing uses abstraction, technical language, and passive writing on a website that we created (writingclaritycalculator.com). Scholars can copy and paste part or all of their article into our writing clarity calculator and learn how their writing compares with the articles published in JM, JMR, and JCR. The website will not save or store the writing, but it will provide an objective measure of the extent to which it is abstract, technical, or passive. For example, the writing clarity calculator told us that a previous draft of the empirical section of our paper scored in the 75th percentile for concreteness, 95th on examples, 50th on frequency, and 70th on active voice. This informed us that we were doing a good job of including examples but could benefit from making the writing more active and concrete. (The writing in methods sections is often necessarily technical, so we did not worry as much about the frequency score for this section.) We thus revised this section by using words that were more concrete and by activating sentences that used passive voice unnecessarily. These steps made our writing more concrete (95th percentile) and active (75th percentile). We discuss the writing clarity calculator in greater detail in the Web Appendix.
Third, scholars need to better understand how to spot, and when to fix, abstract, technical, and passive writing. Scholars cannot and should not remove all abstraction, technical language, and passive writing. They need abstraction to describe theory, technical words to describe methods, and passive voice to focus readers on the important part of a sentence. The problem is that scholars overuse these practices without realizing it. We thus created a tutorial (see Web Appendix B) to help scholars learn ( 1) how to recognize abstract, technical, and passive writing; ( 2) why and when to use these practices; and ( 3) how to avoid overusing them.
We hope that future research will address several limitations of our work. The biggest limitation of our research stems from its greatest strength: the measures of abstraction, technical language, and passive writing. Our studies show that these measures assess the extent to which scholars understand academic writing better than traditional readability indices. The measures also give scholars a way to objectively calibrate the extent to which they are writing clearly. These measures, however, are crude. The concreteness measure is limited because phrases and sentences might be more or less concrete than the average of the words that comprise them. The examples measure is limited because skilled writers know how to use examples without explicitly saying "for example" (or one of the other markers we measured) and because unskilled writers often give examples that are just as abstract as the ideas they are trying to explain. The technical language measure is limited because scholars sometimes brew jargon by repurposing a word that is frequently used to mean something else (e.g., assemblage theorists use "body" to refer to a collection of people and objects rather than a collection of flesh and bone). The passive writing measure is limited because it assesses only passive voice without capturing the other tricks that writers use to obscure who is doing what, such as ambiguous pronouns and disguising verbs as nouns (e.g., [54]). We encourage researchers to improve these measures so the field can better assess what makes academic writing clear and influential.
Another limitation is that our measure of scholarly impact, the number of times an article has been cited, is imperfect. Citations are not an end in themselves, nor do they necessarily capture intellectual indebtedness. Articles can receive "perfunctory mentions," which increase their citation count but do not reflect genuine scholarly impact ([65]). Moreover, citations do not measure the extent to which an article influences nonacademic stakeholders, including managers, consumers, and policy makers. Future research could address these limitations by exploring different measures of impact, including the extent to which the research is mentioned in the press, discussed on social media, and changes the behavior of firms, consumers, or policy makers.
Scholars want their research to be relevant. JM's mission is to "develop and disseminate knowledge about real-world marketing questions useful to scholars, educators, managers, policy makers, consumers, and other societal stakeholders around the world" ([47], p. 1). Similarly, JMR aspires to be "the journal of first choice among authors who seek a broad audience" ([27], p. 1), and JCR attempts to take a "big tent approach" by publishing articles that "build bridges rather than silos" ([36]). The only way for scholars to build bridges and to disseminate knowledge to a broad audience is to cast aside the curse of knowledge and write so that readers outside of their narrow subject area understand what they have to say. By writing clearly, scholars will expand the market for their ideas. Clear writing will also help the marketing discipline take an important step toward realizing its potential to transform business, policy, and the lives of consumers ([46]).
Supplemental Material, sj-pdf-1-jmx-10.1177_00222429211003560 - Marketing Ideas: How to Write Research Articles that Readers Understand and Cite
Supplemental Material, sj-pdf-1-jmx-10.1177_00222429211003560 for Marketing Ideas: How to Write Research Articles that Readers Understand and Cite by Nooshin L. Warren, Matthew Farmer, Tianyu Gu and Caleb Warren in Journal of Marketing
Footnotes 1 The authors contributed equally to this work.
2 Stefan Stremersch
3 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
4 The author(s) received no financial support for the research, authorship, and/or publication of this article.
5 Online supplement https://doi.org/10.1177/00222429211003560
6 [65] and [66] identify three types of factors that potentially influence the number of times it is cited: (1) the importance and relevance of the research (universalist factors); (2) the prestige of the authors and their social connections (social constructivist factors); and (3) title length, attention-grabbers, equations, tables, figures, and writing (presentational factors). Because our research focuses on writing, but not other presentational factors, we discuss writing as its own category alongside universalist, social constructivist, and other presentational factors.
7 We did not collect data from Marketing Science because this journal tends to focus more on quantitative methods, which are not as easy to communicate using words or assess using text analysis.
8 Three hundred thirty-two participants opened the survey, but 66 dropped out after seeing the article but before completing any measures. The dropout rate was similar regardless of the article that participants viewed (omnibus effect: F(8, 323) =.78, p =.62).
9 The words in parentheses correspond to the variables presented in Table 2.
We coded research area as a match when participants familiar with consumer behavior read an article from JCR, participants familiar with quantitative research read an article from JMR, or participants familiar with managerial research read an article from JM. Otherwise, we coded the research area as a mismatch (e.g., if a quantitative researcher read an article from JCR).
Poisson is not an appropriate model for our sample because the test of equidispersion was rejected (p <.001).
References Alba Joseph W. , Hutchinson Wesley J.. (1987), " Dimensions of Consumer Expertise ," Journal of Consumer Research , 13 (4), 411 – 54.
Armstrong J. Scott. (1980), " Unintelligible Management Research and Academic Prestige ," Interfaces , 10 (2), 80 – 86.
Begg Ian. (1972), " Recall of Meaningful Phrases ," Journal of Verbal Learning and Verbal Behavior , 11 (4), 431 – 39.
Benjamin Rebekah George. (2012), " Reconstructing Readability: Recent Developments and Recommendations in the Analysis of Text Difficulty ," Educational Psychology Review , 24 (1), 63 – 88.
Berger Jonah , Humphreys Ashlee , Ludwig Stephan , Moe Wendy W. , Netzer Oded , Schweidel David A.. (2019), " Uniting the Tribes: Using Text for Marketing Insight ," Journal of Marketing , 84 (1), 1 – 25.
Bettencourt Lance A. , Houston Mark B.. (2001), " The Impact of Article Method Type and Subject Area on Article Citations and Reference Diversity in JM , JMR , and JCR ," Marketing Letters , 12 (4), 327 – 40.
Bostian Lloyd R.. (1983), " How Active, Passive and Nominal Styles Affect Readability of Science Writing ," Journalism Quarterly , 60 (4), 635 – 70.
Bostian Lloyd R. , Thering Ann C.. (1987), " Scientists: Can They Read What They Write? " Journal of Technical Writing and Communication , 17 (4), 417 – 27.
Brants Thorsten , Franz Alex. (2006), Web 1T 5-Gram Version 1 LDC2006T13. Philadelphia : Linguistic Data Consortium.
Brown Zachariah C. , Anicich Eric M. , Galinsky Adam D.. (2020), " Compensatory Conspicuous Communication: Low Status Increases Jargon Use ," Organizational Behavior and Human Decision Processes , 161 , 274 – 90.
Brysbaert Marc , Warriner Amy Beth , Kuperman Victor. (2014), " Concreteness Ratings for 40 Thousand Generally Known English Word Lemmas ," Behavior Research Methods , 46 (3), 904 – 11.
Camerer Colin , Loewenstein George , Weber Martin. (1989), " The Curse of Knowledge in Economic Settings: An Experimental Analysis ," Journal of Political Economy , 97 (5), 1232 – 54.
Campbell Margaret C. (2017), " Consumer Research Contribution: Love It or Leave It ," in Advances in Consumer Research , Gneezy Ayelet , Griskevicius Vladas , Williams Patti , eds. Duluth, MN : Association for Consumer Research , 1 – 5.
Clark Roy Peter. (2008), Writing Tools: 55 Essential Strategies for Every Writer , 1st ed. Boston : Little, Brown and Company.
Coleman E.B. , Blumenfeld J.P.. (1963), " Cloze Scores of Nominalizations and Their Grammatical Transformations Using Active Verbs ," Psychological Reports , 13 (3), 651 – 54.
Cornelis Louise H.. (1997), Passive and Perspective. Amsterdam : Rodopi.
Crosier Keith. (2004), " How Effectively Do Marketing Journals Transfer Useful Learning from Scholars to Practitioners? " Marketing Intelligence & Planning , 22 (5), 540 – 56.
Deighton John A. , Mela Carl F. , Moorman Christine. (2021), " Marketing Thinking and Doing ," Journal of Marketing , 85 (1), 1 – 6.
Didegah Fereshteh , Thelwall Mike. (2013), " Which Factors Help Authors Produce the Highest Impact Research? Collaboration, Journal and Document Properties ," Journal of Informetrics , 7 (4), 861 – 73.
Fischhoff Baruch , Beyth Ruth. (1975), " I Knew It Would Happen: Remembered Probabilities of Once-Future Things ," Organizational Behavior and Human Performance , 13 (1), 1 – 16.
Flesch Rudolph. (1948), " A New Readability Yardstick ," Journal of Applied Psychology , 32 (3), 221 – 33.
Frazier Gary L. (2011), " From the Incoming Editor ," Journal of Marketing , 75 (4), 1 – 2.
Gazni Ali. (2011), " Are the Abstracts of High Impact Articles More Readable? Investigating the Evidence from Top Research Institutions in the World ," Journal of Information Science , 37 (3), 273 – 81.
Germann Frank , Ebbes Peter , Grewal Rajdeep. (2015), " The Chief Marketing Officer Matters! " Journal of Marketing , 79 (3), 1 – 22.
Goes Paulo B. , Lin Mingfeng , Yeung Ching-man Au. (2014), " 'Popularity Effect' in User-Generated Content: Evidence from Online Product Reviews ," Information Systems Research , 25 (2), 222 – 38.
Greene William H.. (2003), Econometric Analysis , 5th ed. New York : Macmillan.
Grewal Rajdeep. (2017), " Journal of Marketing Research: Looking Forward ," Journal of Marketing Research , 54 (1), 1 – 4.
Gunning Robert. (1952), The Technique of Clear Writing. New York : McGraw-Hill.
Hartley James , Sotto Eric , Pennebaker James. (2002), " Style and Substance in Psychology: Are Influential Articles More Readable Than Less Influential Ones? " Social Studies of Science , 32 (2), 321 – 34.
Haslam Nick , Koval Peter. (2010), " Predicting Long-Term Citation Impact of Articles in Social and Personality Psychology ," Psychological Reports , 106 (3), 891 – 900.
Hausman Jerry A. , Hall Bronwyn , Griliches Zvi. (1981), " Econometric Models for Count Data with an Application to the Patents-R&D Relationship ," NBER Technical Paper Series.
Heath Chip , Heath Dan. (2007), Made to Stick. New York : Random House.
Hinds Pamela J. , Patterson Michael , Pfeffer Jeffrey. (2001), " Bothered by Abstraction: The Effect of Expertise on Knowledge Transfer and Subsequent Novice Performance ," Journal of Applied Psychology , 86 (6), 1232 – 43.
Holbrook Morris B.. (1986), " A Note on Sadomasochism in the Review Process: I Hate When That Happens ," Journal of Marketing , 50 (3), 104 – 08.
Huber Joel. (2008), " The Value of Sticky Articles ," Journal of Marketing Research , 45 (3), 257 – 60.
Inman J. Jeffrey , Campbell Margaret C. , Kirmani Amna , Price Linda L.. (2018), " Our Vision for the Journal of Consumer Research : It's All About the Consumer ," Journal of Consumer Research , 44 (5), 955 – 59.
Jamali Hamid R. , Nikzad Mahsa. (2011), " Article Title Type and Its Relation with the Number of Downloads and Citations ," Scientometrics , 88 (2), 653 – 61.
Judge Timothy A. , Cable Daniel M. , Colbert Amy E. , Rynes Sara L.. (2007), " What Causes a Management Article to Be Cited—Article, Author, or Journal? " Academy of Management Journal , 50 (3), 491 – 506.
Kelley Colleen M. , Jacoby Larry L.. (1996), " Adult Egocentrism: Subjective Experience Versus Analytic Bases for Judgment ," Journal of Memory and Language , 35 (2), 157 – 75.
Kennedy Peter. (2008), A Guide to Econometrics , 6th ed. Hoboken, NJ : Wiley Blackwell.
Klare George R.. (2000), " The Measurement of Readability: Useful Information for Communicators ," ACM Journal of Computer Documentation , 24 (3), 107 – 21.
Lei Lei , Yan Sheng. (2016), " Readability and Citations in Information Science: Evidence from Abstracts and Articles of Four Journals (2003–2012) ," Scientometrics , 108 (3), 1155 – 69.
Lutz Richard J.. (2018), " On Relevance ," 2018 ACR Fellows address , https://www.rrbm.network/wp-content/uploads/2018/12/Lutz-On-Relevance.pdf.
MacInnis Deborah J. , Morwitz Vicki G. , Botti Simona , Hoffman Donna L. , Kozinets Robert V. , Lehmann Donald R.. (2020), " Creating Boundary-Breaking, Marketing-Relevant Consumer Research ," Journal of Marketing , 84 (2), 1 – 23.
Metoyer-Duran Cheryl. (1993), " The Readability of Published, Accepted, and Rejected Papers Appearing in College & Research Libraries ," College & Research Libraries , 54 (6), 517 – 26.
Mick David Glen. (2006), " Meaning and Mattering Through Transformative Consumer Research ," in Advances in Consumer Research , Vol. 33 , Price Linda LaVonne , Pechmann Cornelia , eds. Duluth, MN : Association for Consumer Research , 1 – 4.
Moorman Christine , van Harald J. , Heerde C. , Moreau Page , Palmatier Robert W.. (2019), " JM as a Marketplace of Ideas ," Journal of Marketing , 83 (1), 1 – 7.
Nickerson Raymond S. , Baddeley Alan , Freeman Barbara. (1987), " Are People's Estimates of What Other People Know Influenced by What They Themselves Know? " Acta Psychologica , 64 (3), 245 – 59.
Oliver Barbara , Dallas Merry Jo , Eckman Molly. (1998), " Communication of Empirical Knowledge: An Investigation of Readability and Quality of Research in Textiles and Apparel ," Clothing and Textiles Research Journal , 16 (1), 27 – 35.
Oppenheimer Daniel M.. (2006), " Consequences of Erudite Vernacular Utilized Irrespective of Necessity: Problems with Using Long Words Needlessly ," Applied Cognitive Psychology , 20 (2), 139 – 56.
Park Sang Kyu , Sela Aner. (2020), " Product Lineups: The More You Search, The Less You Find ," Journal of Consumer Research , 47 (1), 40 – 55.
Peracchio Laura A. , Escalas Jennifer Edson. (2008), " Tell Me a Story: Crafting and Publishing Research in Consumer Psychology ," Journal of Consumer Psychology , 18 (3), 197 – 204.
Pham Michel Tuan. (2013), " The Seven Sins of Consumer Psychology ," Journal of Consumer Psychology , 23 (4), 411 – 23.
Pinker Steven. (2014), The Sense of Style: The Thinking Person's Guide to Writing in the 21st Century. New York : Penguin Books.
Ragins Belle Rose. (2012), " Editor's Comments: Reflections on the Craft of Clear Writing ," Academy of Management Review , 37 (4), 493 – 501.
Roese Neal J. , Vohs Kathleen D.. (2012), " Hindsight Bias ," Perspectives on Psychological Science , 7 (5), 411 – 26.
Rosch Eleanor. (1999), " Principles of Categorization ," in Concepts: Core Readings , Margolis Eric , Laurence Stephen , eds. Cambridge, MA : MIT Press , 189 – 205.
Sadoski Mark , Goetz Ernest T. , Rodriguez Maximo. (2000), " Engaging Texts: Effects of Concreteness on Comprehensibility, Interest, and Recall in Four Text Types ," Journal of Educational Psychology , 92 (1), 85 – 95.
Sawyer Alan G. , Laran Juliano , Jun Xu. (2008), " The Readability of Marketing Journals: Are Award-Winning Articles Better Written? " Journal of Marketing , 72 (1), 108 – 17.
Schimel Joshua. (2012), Writing Science: How to Write Papers That Get Cited and Proposals That Get Funded. Oxford, UK : Oxford University Press.
Schopenhauer Arthur. (1969), The World as Will and Representation , Vol. 1 , Payne E.F.J. , trans. Mineola, NY : Dover Publications.
Segaran Toby , Hammerbacher Jeff. (2009), Beautiful Data: The Stories Behind Elegant Data Solutions. Sebatopol, CA : O'Reilly Media Inc.
Slobin Dan I.. (1966), " Grammatical Transformations and Sentence Comprehension in Childhood and Adulthood ," Journal of Verbal Learning and Verbal Behavior , 5 (3), 219 – 27.
Spiggle Susan. (1994), " Analysis and Interpretation of Qualitative Data in Consumer Research ," Journal of Consumer Research , 21 (3), 491 – 503.
Stremersch Stefan , Camacho Nuno , Vanneste Sofie , Verniers Isabel. (2015), " Unraveling Scientific Impact: Citation Types in Marketing Journals ," International Journal of Research in Marketing , 32 (1), 64 – 77.
Stremersch Stefan , Verniers Isabel , Verhoef Peter C.. (2007), " The Quest for Citations: Drivers of Article Impact ," Journal of Marketing , 71 (3), 171 – 93
Tellis Gerard J. , Chandy Rajesh K. , Ackerman David S.. (1999), " In Search of Diversity: The Record of Major Marketing Journals ," Journal of Marketing Research , 36 (1), 120 – 31.
Trope Yaacov. (2004), " Theory in Social Psychology: Seeing the Forest and the Trees ," Personality and Social Psychology Review , 8 (2), 193 – 200.
Trope Yaacov , Liberman Nira. (2010), " Construal-Level Theory of Psychological Distance ," Psychological Review , 117 (2), 440 – 63.
Van Wesel Maarten , Wyatt Sally , ten Haaf Jeroen. (2014), " What a Difference a Colon Makes: How Superficial Factors Influence Subsequent Citation ," Scientometrics , 98 (3), 1601 – 15.
Wells William D.. (1993), " Discovery-Oriented Consumer Research ," Journal of Consumer Research , 19 (4), 489 – 504.
~~~~~~~~
By Nooshin L. Warren; Matthew Farmer; Tianyu Gu and Caleb Warren
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 89- Marketing Thinking and Doing. By: Deighton, John A.; Mela, Carl F.; Moorman, Christine. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p1-6. 6p. DOI: 10.1177/0022242920977093.
- Database:
- Business Source Complete
Marketing Thinking and Doing
The history of marketing reveals an uneasy relationship between marketers and their academic counterparts. At best, they support one another's endeavors and may even partner to develop ideas and technologies. At worst, they ignore one another and may even view their counterpart with some disdain. While the latter is not useful, this 100-year old ambivalence in marketing is in some ways quite natural and its foundational quality quite old. Aristotle, for example, distinguished thinking (theoria) from doing (praxis).
We think there is a strong case to be made for stronger interactions between the two for the betterment of marketing. Consider weaving as an analogy. Individual fibers have value separately; when combined, they can produce useful materials or beautiful tapestries. To apply the analogy to marketing, academics and practitioners operate in distinct worlds with their own styles and requirements. The result for each can be a limited view of marketing—one focused on the threads relevant to their worlds. However, when each weaves at least some of the other's thinking and doing with their own, the resulting fabric is likely to be more valuable to the field and to the world at large.
Why write about the opportunity for a thinking-doing weave to introduce the Special Issue on "From Marketing Priorities to Research Agendas"? The reason lies in our purpose, which is to publish a set of articles that offers insights regarding how the Marketing Science Institute's (MSI's) priorities—determined every two years by polling corporate members—might be understood and advanced from an academic perspective. To that end, MSI created the MSI Scholars program in 2018 for mid-career scholars interested in translational research and invited them to participate in this challenge. Our purpose was to support these scholars on their quest and to invite practitioners and academic perspectives to challenge and complement their work.
Observing the process of writing and reviewing the articles and commentaries over the past few years has shown us it is much harder than it should be to fit together two things that should go hand in glove. Reflecting on these experiences and our roles in the field more broadly, we observed challenges, inspiration, and important lessons that we want to record in this editorial.
On an optimistic note, we must insist that academics and practitioners of marketing are stronger together. This is why we selected weaving, which brings together different fibers into a masterful whole, as an analogy. We also believe, to quote [ 3], p. 5), that "we are more alike...than we are unalike." In fact, many marketing practitioners act like academics, using the marketplace as their laboratory, while many scholars know how to take a practitioner's perspective and seek to improve marketing practice. Many academics have moved their research activities into the field, and many practitioners have brought more rigor and scholarship into their organizations. The process of creating a better weave is already underway.
Although we believe that the two roles should mainly stay distinct, we focus on how the weave we envision offers important academic and practitioner[ 3] benefits. We then consider strategies for fostering these more productive interactions across the field.
Academics need footholds to move knowledge forward. When these footholds are sourced from the literature, the ideas are guaranteed to reflect other people's thinking and to be several years old. Observing marketing and consumer behavior in the marketplace, in contrast, increases the chance that academics will be exposed to more novel questions and puzzles. As an example, in response to Hamilton et al.'s (2021) discussion of the social journey, [ 9], chief marketing officer of Pernod Ricard, identifies the concept of "social toxicity" as a menacing spillover of that journey that occurs when others pollute the environment with hate speech. [10] document challenges experienced by marketers in measuring and creating value from digital advertising as a springboard for offering research directions.[ 4]
Stronger engagement with marketing exposes scholars to new sources of data and facilitates the trust required to enable its exchange. The revolution in choice modeling inspired by scanner data and the surge in recent research on digital media related to social influence and advertising exemplify this opportunity. An added benefit is that new data often enable better identification of empirical phenomena and the ability to rule out alternative explanations.
Practitioners, especially in digital technology–based enterprises, are breaking new ground in developing data science techniques that diffuse into academia ([25]). For example, many recent online advertising insights summarized in [16], such as efficient measurement design when user and algorithms affect selection into treatment, had their genesis at digital advertising enterprises.
Observing effective and ineffective marketing actions allows academics to draw important lessons. They may also encounter instances of thought leadership, such as the conception of customer transformation offered by [29], senior vice president of Innovation at Salesforce—a view that could be melded with market orientation research to guide teaching about large-scale digital transformation.
Given demands on time and attention and the need to grow their organizations, marketers may be hard-pressed to consider multiple perspectives to view a problem. Yet different lenses can provide different perspectives, as is evident in the academic articles and commentaries. [13] identify social forces acting on the customer journey, which [11] further dissect into social network structures and dynamics in business-to-business markets. In addition, in response to Cui et al.'s (2021) informational view of omnichannel marketing, [15] and [ 1] add important complementary governance and manufacturer perspectives that identify novel problems and solutions important to marketers. Marketers with decisions to make may benefit from trying on these different lenses to assess where the greatest insights lie.
By necessity, marketers making decisions deal with concrete variables (e.g., price, channel) and operate within a particular industry context. These features may limit their ability to see the more powerful abstractions associated with what they have learned that could apply to other parts of their business, other marketing instruments, or in other companies they may lead over time. Academic marketing knowledge, in contrast, is built across industries and can serve these needs.
The need for a rapid-fire stream of marketing decisions means marketers are often forced to be a mile wide and an inch deep in their marketing knowledge base. Yet for most of these decisions—from changing consumer behavior to pricing, marketing capabilities, and online advertising—there is a deep knowledge base available to tap if one knows where to look. Academics, whose individual research often leads them to be a mile deep and an inch wide, can help identify and translate this knowledge.
Academics historically have been the source of many tools used in industry, such as conjoint analysis, marketing mix modeling, and, more recently, attribution modeling. [ 5] extend our understanding of how to improve the last item, which should increase return on marketing investment. [ 8] draw a connection to the customer-based valuation models advanced by [21] by offering insight into managing new sources of data for customer acquisition, growth, and retention. For example, Du et al.'s (2021) insights about incorporating social network data into customer acquisition, using unstructured data, and harnessing causal data for proactive retention offer advances that we expect will impact practice in meaningful ways.
In the swirl of day-to-day operations, it is easy for managers to be swept into fads and new technologies. The detached stance of an academic with no stake in the game and nothing to sell can offer trusted insights. As examples from the Special Issue, [18] sound a set of warnings to marketers about how artificial intelligence (AI) may undermine their distinctive contributions and skills, and [17] caution that excessive shifts in pursuit of agility may threaten brand equity and partnerships.
While the payoffs from working together may be compelling in principle, practitioners and academics often do not work together smoothly in practice. While editing this Special Issue, we saw practitioner commentators chafe at academic abstraction and academics reluctant to translate elegant theorizing into marketing guidelines. More generally, time constraints and the costs of outreach keep each in their own world. Considering these hurdles, we suggest ways to broker these interactions. We start by offering recommendations that apply to both practitioners and academics and then offer suggestions specific to each group. We finish with broader ideas for the entire field of marketing. Across these areas, we seek to be as provocative and helpful as possible to move the field more toward the thinking–doing weave.
A shared focus on marketing problems—the marketing phenomenon—can help overcome many natural differences between academics and practitioners. For example, [23] raise the problem that digital offerings tend to weaken consumers' sense of psychological ownership. In response, [12], a digital music consultant, suggests a problem of interest to both music marketers and scholars that arises when a fan with a sense of psychological ownership behaves in a way that is at odds with the law (e.g., shares digital offerings that they do not own). Likewise, [20], a management consultant at KPMG, points to the shared problem of designing human experiences that merge digital with physical elements.
Our field has a limited history of academics and practitioners working on research together. This Special Issue involves two such commentary teams. In both, we observed the practitioners pointing to actions they have found useful and the academics conceptualizing them in ways that surfaced their broader meanings and implications for the field. For example, Nick Hughes' work at M-PESA and M-KOPA uncovered novel marketing agility mechanisms ([14]). Likewise, Bob Lurie's experience working on managing insights at Eastman Chemical surfaced a set of content factors (data, decisions, and decision makers) and post–data capture process steps critical to converting data to growth ([24]).
Academics and practitioners each have their own brand of theory. Practitioners tend to use informal if-then ideas that explicitly or implicitly drive actions. Academics use formal predictions and mathematical relationships. These different approaches can make interactions challenging. In fact, both have value. Academics often turn to practitioners in nascent areas to unearth their theories (in-use), as [17] do in their examination of marketing agility. The authors then bring their own theory lens to identify boundary conditions when agile marketing actions may harm companies. [19], the chief marketing officer of Adobe, in turn, challenges the authors with unique observations based on her experiences. And so the plaiting goes.
[ 4] tell managers to "get out of the building" to see if their business plans are market-ready. We think this admonition applies to scholars, too. Wading about in the studied milieu to observe firsthand what is happening sparks imagination. Once out, academics should note problems, inconsistencies, and even contradictions. Can these observations be explained by existing theories? If not, what new ideas, theories, constructs, and data are needed to offer insight? These excursions can be informal, such as sharing a conversation on a flight, or they can be more formal, such as connecting with alumni to learn about problems or attending practitioner or MSI conferences to get exposure. We are encouraged by our observations that more scholars are forging field connections.
Many academics teach executive MBA students or MBA students who have recently left jobs. These practitioners offer a gateway experience for faculty members to begin weaving. Ask them to discuss their problems and ideas; try your ideas out on them, and look for the challenges and fruitful extensions.
The MSI Scholar teams were tasked with developing novel research ideas about the MSI priorities—which are concrete problems facing marketers. This translation process is not easy; it requires knowledge of various concepts and theories and marketers' problems. More important, it requires moving between the two worlds to try out different ideas and explanations to determine whether they offer insight into the marketing problem. [27], for example, offer sociological and psychological theories to highlight how AI shapes the customer experience. [ 7], an ethics commentator, offers a provocative linguistic lens.
Engagement can result from informal activities, such as sharing research papers or having conversations on problems of interest, or formal activities, such as data sharing and running studies together. In any case, it is important for academics to think about how they can add value. If what you are doing will help your practitioner contact enhance her reputation in the organization, motivations to participate are strengthened. It is equally important to protect competitive secrets and to deliver value even as papers are working their way through the review process. This might involve, for example, sharing findings and ideas with the partner organization throughout the research process—not just at the end of it.
Getting close to practitioners does not mean accepting that they have accurately captured the problems or are even working on the right problems. As [28], p. 82) notes, "You have to be careful not to believe everything you hear—people in business usually know a set of rules that work well for running their own business, but they often have no idea of where these rules come from or why they work." An example of this type of healthy confrontation can be found in the marketing challenges leveled in several articles dealing with consumer privacy (see [ 5]; [10]; [27]). At the same time, academics should allow their ideas and results to be challenged by practitioners. [ 6], from The Economist, for example, challenges Puntoni et al.'s (2021) view that AI harms the customer experience by suggesting their examples are extreme and by pointing to AI efficiency gains that customers clearly value.
The author teams reflect people from across the marketing discipline—often with very different types of training. It was therefore important for members to translate their approaches to one another even as they moved from the marketers' priorities to future research. In addition, given the scope of MSI's priorities, several teams picked up additional members outside of the MSI Scholars to fill knowledge gaps. For example, [17] added several scholars who brought an organizational vantage point to work with two psychologically oriented consumer researchers. Weaving requires not only such unique content experts, but also translators who can move questions and ideas between worlds.
Marketing is enriched greatly by core disciplines, such as economics and psychology, because they bring new ways of thinking and unique tools to the study of marketing problems. However, we think maintaining focus on marketing phenomena will ensure payoffs for the field. We can practice this balance by taking ideas and tools from the core disciplines and assessing how they apply and might be extended in marketing contexts.
Published academic papers that are not translated for practice are like the proverbial tree falling in the woods. Many outlets can help translation, including practitioner conferences, professional journals, letters to the editor, book writing, and case writing; many academics are fortunate to have the support of professional staff members who facilitate this translation process. For those less fortunate, we encourage business schools to make this investment. The strategy pays off for academics because exposure improves collaboration opportunities, which, in turn, may lead to better data and research ideas. Journals have a role to play here, and the Journal of Marketing has invested in translating our papers for practitioners and for the classroom ([22]).
We asked the author teams to develop research agendas on the MSI priorities. This required them to challenge the very research frameworks and ideas they had constructed with piercing questions and tentative answers. These strong ideas can help propel the field forward to benefit both academics and practitioners. Our teaching also helps steward the future of practice by preparing future marketing leaders with ways of thinking and skills that can lead to better marketing.
It is not difficult to find academics experts on topics that keep practitioners up at night. Host an event on such a topic and invite these experts to a roundtable for discussion and debate. Some firms already invite academics to attend industry conferences or to participate in advisory boards. For example, the Media Rating Council could be further enriched by including advertising experts from the academy.
Read the academic expert's papers and invite him to share ideas. Most would consider it a great honor. Ask the expert to visit conference rooms and labs to observe a research study or an important decision that is unfolding. Academics tend to be curious, and many would find such a visit interesting. Embedding a PhD student, whose stipend is typically covered by a university, can go a long way toward developing tighter integrations between marketers and academics—and it is free. As the relationship improves, join a university marketing research center that brings companies together to discuss problems and share solutions. It is an easy way to gain access to academic knowledge and to learn from other industries.
Scholars are eager for the right kind of data, typically cause-and-effect data and before-and-after data, along with the right to publish a suitably anonymized version. If offered such data, academics will analyze it and share insights. Further, we encourage practitioners to write up the marketing adventures of day-to-day business life and share them with scholars to inspire new research programs.
Just as academics should consider getting out into the field, marketers should get out among academics. Business schools are often eager to let students hear the voice of practice, whether through a lecture, a live case, or a debate with an ostensible competitor or supplier. At best, classrooms can be low-risk labs to sift good ideas from bad with real-time feedback. Short visits to relevant PhD seminars—even for an hour to introduce or debate a challenging problem—may be an equally powerful venue for sharing your ideas with scholars of the future.
Impactful ideas in management practice have boundary conditions. Ask an academic for a point of view on the limits to your idea's applicability. While they might not be as able to address the contextual limitations as well as work colleagues, they often can bring in the lens of other contexts and a broad view of the literature to identify possible limits that might otherwise be hidden from your view. Understanding these boundary conditions is useful to scale knowledge.
Two final points about the theories-in-use we mentioned earlier. When wrong, theories can lead to problems because marketers may not even recognize they hold assumptions that are driving them in unproductive directions. However, when a theory adds value, it would be helpful to make it explicit, codify it, and share it more broadly in the organization. Academics have strong skills in unearthing these theories-in-use and could help marketers see the "bones" of beliefs that are often hidden from plain sight ([31]).
Pointing to ignorance can be galvanizing. [30] describes this as "usable ignorance" because it focuses attention and effort. MSI's priority process is one such mechanism. We encourage the field's associations and institutes to develop even stronger ways to focus attention on its most important problems. Consider, as a model product, the paper and commentaries produced by MSI's priority on the "evolving landscape of MarTech and advertising": [10] examine four types of inefficiencies challenging marketing practice on this topic. [26], chief brand officer at Procter & Gamble, issues a rallying cry by noting that "half my advertising is being wasted" and offers possible solutions on measurement, fraud, transparency, and harmful content. Likewise, [25], a director with the Competition and Markets Authority in the United Kingdom, adds to these concerns by considering why measurement matters for effective competition in digital advertising markets.
Journals and marketing associations have a role to play in hosting collaboration opportunities, whether through special issues or hosting discussions and conferences. This is MSI's raison d'être, and its events have fostered strong engagement over the years. The Journal of Marketing hosted an event at the Summer AMA conference to celebrate the 2020 JM/Sheth Foundation award to "The Chief Marketing Officer Matters." The authors presented their findings and then the CEO of Bajaj Allianz Life Insurance, India; the Global CMO of SAP; and the CMO of Chief Outsiders talked about when and how CMOs make an impact. The outcome was a much richer view of this question (see [ 2] for video coverage).
Improve the incentives for collaborations. In our view, universities should not devalue applied research. The relevant question is "Does it change thinking in the field?" If so, we believe it should be rewarded. Universities can also sponsor competitions for data, such as Wharton's Customer Analytics activities, which bring marketing data to academics. Associations can sponsor prizes to encourage collaborations and journals can reward authors who contribute to practice with awards, such as the Gary L. Lilien ISMS-MSI Practice Prize at INFORMS. We think there is room for a dissertation award built on this premise. Companies can set up contests to involve academics or spark joint academic–practitioner work.
This editorial has argued for a better weave between marketing practice and scholarship. Our goal is simple: to improve marketing, its underlying knowledge, its practice, and its impact on the world through the many ways we have detailed in this editorial. Of course, not everyone will agree with us. An argument might even be made for a looser weave. Indeed, in the 1960s reports from the Ford and Carnegie Foundations made just such an argument and set business schools down a path separating scholarship from practice that was crucial to producing the success business schools now enjoy. We are emphatically not interested in a return to the pre-1960s era when the weave was over-tight and business schools favored teaching institutional detail over theory. We are also not advocating that everyone be a weaver. There is a place in business schools for people who like to integrate thinking and doing as well as for people who are not so inclined. There is a place in companies for both types of people, too. But from our experience editing this Special Issue, we have concluded that for those who want to weave, there are benefits to doing so, and the ideas we have detailed in this editorial should help. In the end, as you learn the threads your counterparts handle, your understanding of marketing will be more complete and your work helping our profession—both in the quality of its ideas and its ability to create value in organizations and in the world—will be strengthened.
Footnotes 1 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
2 Funding The author(s) received no financial support for the research, authorship, and/or publication of this article.
3 We focus this editorial on "practitioners" within marketing organizations. However, our points also apply to marketing practitioners engaged in questions related to public policy and to other societal stakeholders engaged in marketing, even as critics.
4 We might add that empirical work can also benefit from starting in the marketplace and not in the lab or with archival data. Starting with interviews, a quick field experiment, or survey can be informative for getting the lay of the land.
References Ailawadi Kusum. (2021), "Commentary: Omnichannel from a Manufacturer's Perspective," Journal of Marketing, 85 (1), 121–25.
AMA (2020), "Does the CMO Matter? Definitely," https://www.ama.org/2020/09/01/does-the-cmo-matter-definitely/.
Angelou Maya. (1991), I Shall Not Be Moved. New York : Random House.
Blank Steve, Dorf Bob. (2020), The Start-Up Owner's Manual. Hoboken, NJ : John Wiley & Son.
5 Cui Tony Haitao, Ghose Anindya, Halaburda Hanna, Iyengar Raghuram, Pauwels Koen, Sriram S., Tucker Catherine, Venkataraman Sriraman. (2021), "Informational Challenges in Omnichannel Marketing: Remedies and Future Research," Journal of Marketing, 85 (1), 103–20.
6 Cukier Kenneth. (2021), "Commentary: How AI Shapes Consumer Experiences and Expectations," Journal of Marketing, 85 (1), 152–55.
7 Donath Judith. (2021), "Commentary: The Ethical Use of Powerful Words and Persuasive Machines," Journal of Marketing, 85 (1), 160–62.
8 Du Rex Yuxing, Netzer Oded, Schweidel David A., Mitra Debanjan. (2021), "Capturing Marketing Information to Fuel Growth," Journal of Marketing, 85 (1), 163–83.
9 Forbus Pamela. (2021), "Commentary: The Case for a Healthier Social Customer Journey," Journal of Marketing, 85 (1), 93–97.
Gordon Brett R., Jerath Kinshuk, Katona Zsolt, Narayanan Sridhar, Shin Jiwoong, Wilbur Kenneth C. (2021), "Inefficiencies in Digital Advertising Markets," Journal of Marketing, 85 (1), 7–25.
Grewal Rajdeep, Sridhar Shrihari. (2021), "Commentary: Toward Formalizing Social Influence Structures in Business-to-Business Customer Journeys," Journal of Marketing, 85 (1), 98–102.
Griffin Jim. (2021), "Commentary: Music's Digital Dance: Singing and Swinging from Product to Service," Journal of Marketing, 85 (1), 223–26.
Hamilton Ryan, Ferraro Rosellina, Haws Kelly L., Mukhopadhyay Anirban. (2021), "Traveling with Companions: The Social Customer Journey," Journal of Marketing, 85 (1), 68–92.
Hughes Nick, Chandy Rajesh C. (2021), "Commentary: Trajectories and Twists: Perspectives on Marketing Agility from Emerging Markets," Journal of Marketing, 85 (1), 59–63.
John George, Scheer Lisa K. (2021), "Commentary: Governing Technology-Enabled Omnichannel Transactions," Journal of Marketing, 85 (1), 126–30.
Johnson Garrett A. (2020), " Inferno: A Guide to Field Experiments in Online Display Advertising," Working paper, https://ssrn.com/abstract=3581396.
Kalaignanam Kartik, Tuli Kapil, Kushwaha Tarun, Lee Leonard, Gal David. (2021), "Marketing Agility: The Concept, Antecedents, and a Research Agenda," Journal of Marketing, 85 (1), 35–58.
Kozinets Robert V., Gretzel Ulrike. (2021), "Commentary: Artificial Intelligence : The Marketer's Dilemma," Journal of Marketing, 85 (1), 156–59.
Lewnes Ann. (2021), "Commentary: The Future of Marketing Is Agile," Journal of Marketing, 85 (1), 64–67.
Lieberman Scott. (2021), "Commentary: Managing Human Experience as a Core Marketing Capability," Journal of Marketing, 85 (1), 219–22.
McCarthy Daniel M., Fader Peter S., Hardie Bruce G.S. (2017), "Valuing Subscription-Based Businesses Using Publicly Disclosed Customer Data," Journal of Marketing, 81 (1), 17–35.
Moorman Christine, van Heerde Harald J., Page Moreau C., Palmatier Robert W. (2019), " JM as a Marketplace of Ideas," Journal of Marketing, 83 (1), 1–7.
Morewedge Carey, Monga Ashwani, Palmatier Robert, Shu Suzanne, Small Deborah. (2021), "Evolution of Consumption: A Psychological Ownership Framework," Journal of Marketing, 85 (1), 196–218.
Morgan Neil A., Lurie Robert S. (2021), "Commentary: A Strategic Perspective on Capturing Marketing Information to Fuel Growth: Challenges and Future Research," Journal of Marketing, 85 (1), 184–89.
Porter Jonathan. (2021), "Commentary: Inefficiencies in Digital Advertising Markets: Evidence from the Field," Journal of Marketing, 85 (1), 30–34.
Pritchard Marc. (2021), "Commentary: 'Half My Digital Advertising Is Wasted...,'" Journal of Marketing, 85 (1), 26–29.
Puntoni Stefano, Reczek Rebecca Walker, Giesler Markus, Botti Simona. (2021), "Consumers and Artificial Intelligence: An Experiential Perspective," Journal of Marketing, 85 (1), 131–51.
Varian Hal. (2016), "How to Build an Economic Model in Your Spare Time," The American Economist, 61 (1), 81–90.
Wild Jason. (2021), "Commentary: Beyond Data: The Mindset and Disciplines Needed to Fuel Growth," Journal of Marketing, 85 (1), 221–26.
Zaltman Gerald. (1991), " Assessing Progress on Meeting MSI Priorities," in Marketing Science Institute Report Number 91-107. Cambridge, MA : Marketing Science Institute.
Zeithaml Valarie A, Jaworski Bernard J., Kohli Ajay K., Tuli Kapil R., Ulaga Wolfgang, Zaltman Gerald. (2020), "A Theories-in-Use Approach to Building Marketing Theory," Journal of Marketing, 84 (1), 32–51.
~~~~~~~~
By John A. Deighton; Carl F. Mela and Christine Moorman
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 90- Marketplace Literacy as a Pathway to a Better World: Evidence from Field Experiments in Low-Access Subsistence Marketplaces. By: Viswanathan, Madhubalan; Umashankar, Nita; Sreekumar, Arun; Goreczny, Ashley. Journal of Marketing. May2021, Vol. 85 Issue 3, p113-129. 17p. 1 Diagram, 6 Charts. DOI: 10.1177/0022242921998385.
- Database:
- Business Source Complete
Marketplace Literacy as a Pathway to a Better World: Evidence from Field Experiments in Low-Access Subsistence Marketplaces
Multinational companies increasingly focus on subsistence marketplaces, given their enormous market potential. Nevertheless, their potential is untapped because subsistence consumers face extreme constraints. The authors contend that subsistence consumers need marketplace literacy to participate effectively and beneficially in marketplaces. Marketplace literacy entails the knowledge and skills that enable them to participate in a marketplace as both consumers and entrepreneurs. This is crucial for subsistence consumers, as they often must function in both roles to survive. Previous research, however, has not empirically examined the influence of marketplace literacy on well-being or marketing outcomes related to well-being. To address this gap, the authors implemented three large-scale field experiments with approximately 1,000 people in 34 remote villages in India and Tanzania. They find that marketplace literacy causes an increase in psychological well-being and consumer outcomes related to well-being (e.g., consumer confidence, decision-making ability), especially for subsistence consumers with lower marketplace access, and it causes an increase in entrepreneurial outcomes related to well-being (e.g., starting a microenterprise) for those with higher marketplace access. Overall, this research generates practical implications for the use of marketplace literacy as a pathway to a better world.
Keywords: consumer well-being; entrepreneurship; field experiment; marketplace access; marketplace literacy; randomized control trial; subsistence contexts
We demonstrate that marketing can serve as a pathway to a better world by improving the lives and livelihoods of subsistence consumers, many of whom live in extreme poverty and lack access to marketplaces (i.e., are among the world's most vulnerable consumers). The success of marketers who look to emerging markets for growth is inextricably linked to the well-being of hundreds of millions of subsistence consumers and microentrepreneurs, who face poor infrastructure, material resource constraints, and low literacy. Yet such markets are some of the fastest growing in the world ([47]), with nearly $5 trillion of consumption spending annually ([20]). As a result, many multinational companies (e.g., Procter & Gamble, Nokia, Unilever) market products to subsistence consumers.[ 6] A primary disconnect between marketers' engagement in such markets and subsistence consumers' demand for their products, however, is the latter's inability to participate in marketplaces effectively and beneficially. Effective participation entails the knowledge and skills of what to buy and sell, how to participate in the marketplace as both a consumer and an entrepreneur, and a deeper understanding of why marketplace activities occur. Unique to subsistence marketplaces is that microentrepreneurship serves as a primary source of livelihood to meet consumption needs ([ 4]). Beneficial marketplace participation involves making optimal consumer decisions and generating income by creating value for others through microentrepreneurship. With low marketplace participation and suboptimal decision making, firms may sell substandard products ([18]). Moreover, firms may not partner with local entrepreneurs, which is an important strategy for market entry in emerging markets ([44]).
We contend that subsistence consumers' effective and beneficial marketplace participation requires not only material resources (e.g., access to capital) but also marketplace literacy ([52]). We define marketplace literacy as the knowledge and skills that enable marketplace participation as both a consumer and an entrepreneur. This form of literacy is distinct from consumer literacy, as it encompasses marketing at a broader level to include the perspective of buyers and sellers. Marketplace literacy is crucial for people in subsistence contexts because they often need to function in both consumer and entrepreneurial roles to survive ([55]). In particular, such literacy helps consumers evaluate product quality, compare prices, and develop consumer confidence. It also helps entrepreneurs envision a business opportunity, market to new customers, distribute products efficiently, and run a microenterprise profitably ([52]). Despite the importance of marketplace literacy, we lack empirical research on its impact. We assess the causal impact of marketplace literacy on improvements in subsistence consumers' lives and livelihoods, which manifest in their psychological well-being and consumer and entrepreneurial outcomes related to well-being.
An important moderator of the effect of marketplace literacy on these outcomes is marketplace access,[ 7] a contextual variable that is taken for granted in resource-rich contexts but is a severe challenge in many subsistence marketplaces. In particular, many subsistence consumers lack access to marketplaces ([44]) due to distance and cost, both of which constrain their ability to obtain important marketing information ([46]). We examine how varying levels of marketplace access moderate the effect of marketplace literacy on several outcomes that relate to well-being.
To test our framework, we implemented three large-scale randomized control trials (RCTs) with approximately 1,000 individuals (who consisted of women farmers and isolated tribal members) across 34 villages in rural India and Tanzania. Across our field experiments, we implemented a marketplace literacy educational program for a treatment group and either no program or a placebo educational program for a control group and measured changes in several perceptions and self-reported behaviors. An analysis of the treatment effect showed that marketplace literacy (which was gained through the educational program) caused an increase in psychological well-being and consumer outcomes (e.g., consumer confidence, decision-making ability), especially for those with lower marketplace access. In addition, the treatment effect caused an increase in entrepreneurial outcomes (e.g., entrepreneurial intention, starting a microenterprise) for those with higher marketplace access.
We generate several theoretical implications. First, we advocate that subsistence consumers play a significant role in mainstream marketing research because they represent a sizable proportion of consumers globally and function in extreme conditions that challenge the theories developed for resource-rich consumers and markets. Second, although existing work has described marketplace literacy educational programs (e.g., [56]; [52]), our research is the first to demonstrate the causal effect of marketplace literacy on several outcomes related to well-being. We provide rigorous evidence that disentangles the causal effect of marketplace literacy from potential biases, such as omitted variables, that increase well-being and reverse causality ([ 5]). In doing so, we demonstrate that marketplace literacy offers a pathway to a better world for subsistence consumers and highlight its value as a central individual difference variable for future research. This pathway is fundamentally about literacy in a marketing domain, which encompasses buyers, sellers, and their interplay. Third, we introduce marketplace access as an important contextual factor that should be measured and incorporated when studying a range of marketing phenomena across income levels.
From a practical standpoint, we identify marketplace literacy as a critical form of marketing-related literacy (in addition to financial literacy) that should be cultivated to enable subsistence consumers' effective and beneficial participation in marketplaces. Both of these outcomes benefit marketers and society as a whole. We make recommendations on how firms, public policy makers, and development organizations can work in concert to cultivate marketplace literacy. Further, we describe how marketplace literacy interventions can be scaled and how governments can help subsistence consumers to overcome a lack of marketplace access to engage in entrepreneurship. Importantly, through conducting this research, hundreds of low-income and resource-constrained individuals in remote areas of India and Tanzania experienced improvements in their lives and livelihoods.
Subsistence, which entails the condition of existing with extremely limited financial and basic (e.g., food, water, housing) resources, is a harsh reality for much of the world's population, especially in developing countries ([59]). In such contexts, subsistence consumers face the persistent stress of material resource deprivation and cognitive and affective constraints ([53]). Low income co-occurs with low basic literacy, which exacerbates the limitations that subsistence consumers face in making decisions that require abstract thinking ([51]; [54]). Another distinctive feature of subsistence marketplaces is the intertwined nature of consumption and entrepreneurship ([55]). Given their chronic resource constraints, a lack of access to financial capital, limited technical skills, and periodic surges in household expenses, subsistence consumers become entrepreneurs out of necessity rather than by opportunity or choice ([ 4]). They operate multiple seasonal microenterprises to meet their short-term needs, including selling fruits and vegetables, making handicrafts, or managing an eatery ([55]). Indeed, their entrepreneurship helps drive their consumption, and vice versa ([55]). Finally, subsistence marketplaces are rich in social capital, characterized by strong interpersonal relationships with frequent one-on-one interactions that provide a platform for learning among and between buyers and sellers ([57]). In turn, this relational environment helps subsistence consumers overcome their constraints.
Marketplace literacy has been broadly described as a capability necessary to participate in the marketplace as a consumer and an entrepreneur ([56]). We build on this description to formally define marketplace literacy as the knowledge and skills that enable marketplace participation as both a consumer and an entrepreneur. This definition parallels definitions of functional literacy in other domains (e.g., financial literacy; [19]), which emphasize an understanding of information and having the skills to use this information to complete tasks in life. Thus, we isolate this core functional literacy, which centers on knowledge and skills, from the outcomes that it can lead to, such as confidence and well-being.
Marketplace literacy is distinct from previously examined forms of literacy in several ways. First, unlike consumer knowledge and consumer literacy, it incorporates entrepreneurial literacy. Marketplace literacy not only allows for more effective consumer decision making but also enables subsistence consumers to start and maintain income-generating activities. Second, it incorporates the "know-why" of marketing exchanges, which is essential to making better choices ([60]). Third, it is distinct from marketing-related entrepreneurial skills, which have only recently been explored (e.g., [ 5]), and does not incorporate consumer literacy and its synergy with entrepreneurship. Further, whereas prior research has described the usefulness of marketplace literacy educational programs (e.g., [56]; [52]), the causal outcomes of marketplace literacy have not been empirically assessed through field experiments. We address these gaps in this article (for a review of the literature, see Web Appendix B).
Marketplace literacy comprises three types of knowledge and skills: know-what, know-how, and know-why ([56]). Know-what is a person's objective knowledge, such as knowing what to buy and pay as a buyer ([38]) and what to sell and for what price as a seller. Know-how is a person's procedural skill set, such as an ability to compare product attributes to make a purchase decision ([ 1]; [33]) and to differentiate one's enterprise from that of competitors to attract consumers ([11]). Unique to marketplace literacy is the know-why of marketplace exchanges, which can address a difficulty with abstract thinking and cognitive constraints. This form of conceptual knowledge enables subsistence consumers to make decisions with a deeper understanding of cause-and-effect relationships. For example, the value assigned to a marketing exchange is up to a customer, but an understanding of why value is important results in improved consumer outcomes (e.g., being confident about a decision, choosing the right product from the right seller). Further, in terms of entrepreneurship, knowing why a certain business is better to pursue than another or why customer orientation matters yields better entrepreneurial outcomes. Thus, a person's know-why, over and above their know-what and know-how ([56]), can place that person on a path to search for and process marketing information more effectively ([35]).
Marketplace literacy can enable subsistence consumers to choose the right store, negotiate the right price, compare products, assess quality, and purchase the right product. It also fosters greater consumer confidence, which is the ability to assertively acquire information and protect oneself from deception ([ 8]). Further, the evaluation of marketing information also requires an understanding of sellers' motives and marketing tactics ([17]). Entrepreneurial literacy enables subsistence consumers to function as more effective consumers because it helps them make better decisions on the basis of their knowledge of a seller's motivations, capacities, and constraints. This generates immediate benefits, such as saving money, obtaining better quality products, and purchasing products that better fit one's needs. Further, entrepreneurial knowledge and skills help subsistence consumers pursue microentrepreneurship more effectively, including being able to identify a marketing opportunity, starting a microenterprise, and running it profitably ([31]).
Due to deprivation on multiple fronts, however, many subsistence consumers lack marketplace literacy. Given that low income is strongly associated with low literacy, subsistence consumers face several cognitive constraints ([54]) that impede their development of marketplace literacy ([56]). For example, they engage in concrete thinking and form product impressions by processing isolated pieces of information (e.g., buying the least expensive product) rather than abstracting information across product attributes to gauge overall value. Alternatively, they engage in pictographic thinking and rely on their visual sense in place of reading and processing textual and numerical information. For example, rather than use units of measurements, they visualize usage situations and estimate an amount to buy. Subsistence consumers may visualize dollar bills in place of actually calculating the total cost of a basket of products. They may pass up on a discount as they are unable to compute the final price ([54]). Overall, it is challenging for subsistence consumers to combine disparate pieces of information to draw abstract, higher-level judgments of value, such as combining the "give" with the "get." Difficulty with abstractions that is reflected in concrete and pictographic thinking also inhibits a deeper understanding of basic concepts (e.g., good health, being a consumer, what a business does) and causal relationships (i.e., knowing why gauging value can lead to money saved and better-quality products; [54]). Further, many subsistence consumers face affective constraints and consequently avoid unfamiliar products and settings as well as valuable marketing experiences to maintain their self-esteem and not have their low literacy exposed ([ 2]; [54]).
Entrepreneurs in subsistence contexts also may lack marketplace literacy. In terms of sellers' cognitive constraints, the very notion of an enterprise may be difficult to envision, and an understanding of why one should pursue a particular business over another can be limited. Indeed, given the cognitive constraints they face, many subsistence entrepreneurs may be unable to draw a causal connection between being customer oriented and being successful, even though such an orientation can benefit them. A majority of subsistence microenterprises are formed by replicating the business models of successful entrepreneurs in the marketplace ([36]). Many potential entrepreneurs, however, are excluded from institutions and experience social inequalities, which hinder their ability to participate in the marketplace, scan their environment for opportunities, and start their own enterprise ([36]). By gaining marketplace literacy, potential entrepreneurs can overcome such barriers to engage in entrepreneurship confidently and effectively ([49]). We present our conceptual framework in Figure 1).
Graph: Figure 1. Conceptual framework.
Well-being is a multidimensional construct ([15]) composed of two dimensions ([21]). The first (objective) dimension refers to individuals' satisfaction with making choices that will enhance their quality of life, which is reflected in their socioeconomic indicators and behavioral choices. The second (subjective) dimension captures individuals' psychological well-being, which includes autonomy, self-acceptance, positive relations, situational control, personal growth, and purpose in life ([41]). Psychological well-being is a broad concept; and thus, we focus on only some aspects that are relevant to our research context, such as autonomy, empowerment, and domestic stability. If people's choices are constrained by external forces beyond their control or they are compelled to behave in a way that prevents them from satisfying their needs and aspirations, then their autonomy is threatened ([40]). Low-income consumers sense a loss of control due to severely constrained options in the marketplace ([22]), and as a result, they are forced to make suboptimal choices. In addition to these material resource constraints, gender-based norms impede low-income women's autonomy and empowerment in general, and their ability to enforce decisions within their families in particular. However, when they have some degree of control over their environment, a voice in their family's decisions, and a stable home, women experience well-being ([ 6]).
Given the multiple sources of deprivation that subsistence consumers face, consumer and entrepreneurial knowledge and skills can have broad ripple effects that extend beyond these domains to both objective and subjective (i.e., psychological) dimensions of well-being. In subsistence marketplaces, consumption and entrepreneurship are necessary for not only a person's survival but also improving their quality of life ([49]). Yet, a variety of negative factors can arise for subsistence consumers in a marketplace, such as humiliation, unfamiliarity, a lack of confidence, and an inability to visualize future outcomes. As a result, subsistence consumers are unable to function effectively as consumers, which reduces their well-being ([32]). Alternatively, those with marketplace literacy can be more effective in acquiring and consuming products that meet their families' needs ([56]). More effective consumer behavior leads to saving money, obtaining better-quality products, avoiding being cheated, and processing marketing information to make the right decision. In turn, marketplace literacy can lead to a person feeling in control of their decisions and performing purchase-related tasks with confidence. Such a sense of autonomy and competence has a positive influence on well-being ([40]).
Further, marketplace literacy can lead to the confidence to start a business, source from the right supplier, innovate, and market to the right customer to generate income. Autonomy and empowerment can enable individuals to gain control of their physical environment, but they usually lack the knowledge or skills to find and make use of such opportunities ([22]). Entrepreneurship can offer a path for individuals to alleviate their material constraints and function with autonomy ([48]). For example, microentrepreneurs report greater life satisfaction due to higher financial security and a sense of achievement after starting a microenterprise ([10]). Therefore, we propose:
- H1: An increase in marketplace literacy causes an increase in psychological well-being.
We expect that marketplace literacy will lead to consumer-specific outcomes related to well-being, such as consumer confidence and decision-making ability. These outcomes influence both subjective and objective dimensions of well-being, such as individuals' satisfaction with their choices and improvements in their livelihoods ([21]), which include consumer savings and purchasing better-quality products. Consumer confidence is the extent to which a consumer feels capable of making decisions in the marketplace, can assertively acquire and use information to make decisions, and can protect themself from being misled by sellers ([ 8]). Confident consumers search for information more effectively and have more positive experiences in a marketplace ([30]). Acquiring market-related information, however, involves navigating the marketplace in search of sources of new information ([42]). Yet many subsistence consumers avoid unfamiliar marketplace environments and interactions with marketers due to low self-esteem and the stigma of being perceived as poor or low-literate ([ 2]; [51]; [54]). As a result, they face deceptive practices from sellers ([23]).
We build on prior work (e.g., [56]) to argue that marketplace literacy will engender consumer confidence in subsistence consumers. Previous research has linked various forms of literacy, such as basic ([58]), financial ([19]), and consumer literacy ([25]; [33]), to perceptions of self-efficacy in decision making. In addition, marketplace literacy offers subsistence consumers a deeper understanding of why marketing exchanges occur, and an ability to comprehend abstract notions, such as exchange value. It can enable people to assertively navigate the marketplace to gather information from stores and sellers and determine how and why to make specific judgments. Further, it enables individuals to understand marketers' persuasive tactics ([60]) and to have the autonomy and judgment to differentiate useful from redundant information, all of which should lead to confidence in the marketplace.
Central consumer decisions include what to buy and at what price ([ 9]). Whereas such decisions seem straightforward for most, especially in resource-rich contexts, for subsistence consumers who face cognitive and affective constraints, such decisions are difficult. When achieved, however, these decisions can increase well-being ([32]). Marketplace literacy enables individuals to function as more effective consumers by aiding them in knowing how and why to gauge product quality and negotiate better prices. These actions are rudimentary, yet central, aspects of decision making that capture exchange value, or "the get" (quality) and "the give" (price), which is a cornerstone of effective consumer decision making ([39]). Therefore, we propose:
- H2: An increase in marketplace literacy causes an increase in consumer confidence and decision-making ability.
Similar to consumer outcomes, entrepreneurial outcomes influence a broader notion of well-being ([21]). Entrepreneurship creates economic and social value for people who are living in poverty ([12]). "Entrepreneuring," or the process of removing constraints to identify entrepreneurial opportunities, however, requires skills and capabilities ([48]). As we have noted, consumer and entrepreneurial roles are two sides of the same coin in subsistence contexts, but becoming an entrepreneur requires additional resources. In addition to a lack of infrastructure, financial capital, and government support, potential entrepreneurs who live in poverty face other limitations. For example, they may not be willing to take risks, know how to deal with the formal aspects of running a business ([27]), or know why to market to a certain customer.
Beyond general business ([27]), financial ([13]), and marketing ([ 5]) education, we expect that marketplace literacy is central to engendering entrepreneurship, as it covers what microenterprise to start, how to start it, and why to start it over another option. Marketplace literacy also provides an understanding of customer orientation and, thus, sheds light on how to attract and retain customers and why being mindful of the competition can reduce the risks involved with operating a microenterprise. Indeed, starting a microenterprise can greatly improve a person's quality of life ([48]).
Marketplace literacy also equips potential entrepreneurs with consumer literacy, which helps them understand how and why consumers make decisions despite resource constraints ([56]). This is necessary for entrepreneurs in subsistence marketplaces because, here, marketing actions are likely to succeed when undertaken through a communal perspective ([11]) due to the frequent one-on-one interactions that occur among buyers and sellers ([57]). Specifically, knowing what to sell and how and why to sell it over an alternative is driven by what customers want to buy, how they will acquire it, and why they should buy it over a competitive product. Therefore, we propose:
- H3: An increase in marketplace literacy causes an increase in entrepreneurial intentions and microenterprise start-ups.
Next, we develop predictions about the moderating role of an important contextual variable: marketplace access. Marketplace access can be viewed in several ways, such as geographic proximity (e.g., physical distance by road; [46]) or low cost or duration of travel (e.g., transportation alternatives; [45]) to a marketplace. In addition to providing a platform for buying and selling, marketplaces provide a variety of marketing information and the opportunity for buyers and sellers to interact with one another and with marketing information ([60]). Given that many subsistence consumers lack marketplace access ([44]), they are unable to rely on the marketplace to provide them with marketing information and to learn from other consumers and entrepreneurs about how to make decisions confidently and effectively. For example, individuals in rural communities need to overcome a significant information divide to adopt beneficial products ([34]). We contend that overcoming such a divide requires marketplace literacy, which can be developed even with limited marketplace access.
Individuals who live in remote contexts can indeed develop marketplace literacy, although they need to overcome the barriers of scarce marketplace information and their own cognitive and affective constraints. As market information becomes less accessible to subsistence consumers who already face cognitive constraints, their search for products and evaluation of product choices becomes even more challenging. Nevertheless, individuals have been shown to overcome these challenges through a variety of coping strategies ([51]; [54]). One means is through one-on-one social interactions, which offer a way to develop knowledge from the few buyers and sellers in their community ([57]). In particular, they use local social connections to obtain and validate information ([34]). Thus, although individuals with low marketplace access are disadvantaged by sparse information from the marketplace, they develop marketplace literacy through alternative means. In low-access contexts, individuals face additional deprivation and, as a result, need to develop capabilities to overcome such constraints to survive ([22]). As such, marketplace literacy may develop organically out of necessity to address urgent needs. Thus, for subsistence consumers with lower (vs. higher) marketplace access, we expect that their marketplace literacy will be even more consequential because they require the knowledge and skills to analytically process sparse marketing information to make independent decisions without relying on the marketplace to provide such information. In other words, marketplace literacy will have greater influence on subsistence consumers' autonomous functioning when marketplaces are less accessible.
In contrast, subsistence consumers with relatively higher access to marketplaces can more easily leverage the external information provided by the marketplace—such as information from other consumers, sellers, and competitors, and marketing cues—to make decisions independently. When accessing information in a marketplace is less effortful, consumers develop familiarity through repeated exposure to marketing cues ([ 3]) and, thereafter, utilize their memory to make purchase decisions ([37]). We predict that when subsistence consumers can obtain marketing information in higher-access contexts, their marketplace literacy will play a smaller role in affecting their psychological well-being, consumer confidence, and decision-making ability. Therefore, we propose:
- H4: An increase in marketplace literacy causes an increase in psychological well-being for those with lower access to marketplaces.
- H5: An increase in marketplace literacy causes an increase in consumer confidence and decision-making ability for those with lower access to marketplaces.
In contrast to our arguments in H4 and H5, we argue that higher (vs. lower) marketplace access will lead to a higher impact of marketplace literacy on entrepreneurship. Starting a business is much more resource intensive than what is required to function as a consumer. For entrepreneurs, marketplace access provides the infrastructure, financial and otherwise, to build a business and to attract a volume of customers as well as a social network of entrepreneurs from whom to learn. Such relational-based learning and emulating is essential in subsistence marketplaces ([57]).
Because subsistence entrepreneurship is born out of necessity rather than opportunity, the creation of microenterprises is enabled with access to the market institutions and infrastructure to compensate for a lack of resources ([36]) and skills to act on such external resources ([48]). For example, access to finance through market institutions is critical for the success of entrepreneurs in poverty contexts ([26]). In resource-constrained contexts, entrepreneurs gather financial resources from social connections in the marketplace ([12]), which would not be possible unless a potential entrepreneur has access. In addition to material and financial resources, developing knowledge and skills to identify and act on entrepreneurial opportunities also requires marketplace participation, which is easier when access to marketplaces is convenient and less costly. Thus, unlike our prediction for psychological well-being and consumer outcomes, we expect that higher (vs. lower) marketplace access will lead to a bigger impact of marketplace literacy on entrepreneurship. Therefore, we propose:
- H6: An increase in marketplace literacy causes an increase in entrepreneurial intentions and microenterprise start-ups for those with higher access to marketplaces.
We tested our hypotheses with primary data obtained from three large-scale panel field experiments in India and Tanzania. For each field experiment, we implemented an RCT with individuals who live in remote subsistence contexts as our unit of analysis. Our experiments consisted of five parts: ( 1) sample recruitment, ( 2) pretreatment measurement of the outcomes for all participants, ( 3) random assignment of participants into either a treatment group or a control group, ( 4) implementation of an educational intervention to manipulate our focal construct, and ( 5) posttreatment measurement of outcomes. As a result, our randomly assigned interventions, which manipulate our theoretical variables exogenously, are orthogonal to other factors that could potentially drive changes in the outcomes of interest ([ 5]). We present a timeline of our field experiments in Web Appendix C.
In all three field experiments, we manipulated our core construct, marketplace literacy, with the use of an established educational program that has been used to develop marketplace literacy for subsistence consumers ([56]; [52]). The marketplace literacy program has been conducted in several countries (e.g., India, Tanzania, Argentina, Mexico, Honduras, Uganda, United States). It has ranged from four- to eight-hour educational sessions over one to six days and is taught using pictures, verbal discussions, hands-on exercises, and slides and videos shown on a laptop. The content focuses on knowing what, how, and why marketplace exchanges occur in terms of interactions between consumers and sellers/entrepreneurs.
Aspects of consumer literacy include the objective knowledge of what to buy, the procedural knowledge of how to buy products at the right store at the right price, and the conceptual knowledge of why to look for value in an exchange. Aspects of entrepreneurial literacy include the objective knowledge of what to sell, the procedural knowledge of how products move through the value chain, and the conceptual knowledge of why to choose a business over another and why being customer oriented is important. The program teaches consumer and entrepreneurial literacy in an iterative way, in which each is mutually reinforced. Participants are asked to role play as a buyer and a seller. They are then asked about what has occurred in the role plays (bargaining between a consumer and a seller), how it occurred (ways to negotiate a better price as a consumer or garner a higher price as a seller), and why (to save money as a consumer or generate profit as a seller). We present a list of topics that we covered in our marketplace literacy program and our method in Web Appendix D.
We created our control condition in different ways across our studies and detail this subsequently. Further, we examined how marketplace access moderates the impact of marketplace literacy on several outcomes related to well-being. We both manipulated and measured marketplace access to capture variability in its measurement, using various proxies.
We conducted a field experiment with women farmers across several villages in the rural parts of the state of Tamil Nadu, India, to test the main effect of marketplace literacy on psychological well-being (H1) and whether this relationship is moderated by marketplace access (H4). We worked with an established field research team that had two decades of experience in implementing educational programs in over 100 villages in rural India.
We manipulated marketplace literacy with the marketplace literacy program and manipulated marketplace access by identifying three clusters of villages. Each cluster varied in its road-based geographic distance to a large marketplace. Drawing on these distances, we implemented our treatment (marketplace literacy program vs. a control) at three levels of access within each cluster of villages to ensure variability in marketplace access in both the treatment and control (relatively low, medium, and high marketplace access). In terms of our analysis, we used the actual geographic distances in kilometers to a marketplace. We summarize our experiment next and provide more detail in Web Appendix E.
In each village, the field research team recruited 22 to 25 women farmers who agreed to participate in the longitudinal experiment in exchange for an incentive (400 Indian rupees, approximately US$6). We focused on women farmers due to their known positive impact on their families ([16]). Our total sample consisted of 392 women farmers (196 who participated in the marketplace literacy program and 196 who did not).[ 8] The participants in the treatment groups and control groups were comparable across multiple variables (i.e., the treatment was randomized; see Web Appendix E for details).
To begin, the research team administered a presurvey to the 392 women farmers across the 18 villages. The survey was translated into the local language of Tamil and was administered in person to each participant, one at a time. Then, the marketplace literacy program was delivered to the treatment group of 196 women farmers in nine villages over two half-day sessions. Three instructors, each assigned to a cluster of villages, delivered the program in three villages each (with one village at each of the three access levels). Then, nine weeks after the presurvey and eight weeks after the program, the research team individually administered the postsurvey with the same measures as the presurvey to the 392 women farmers. A marketing research firm in India entered the data.
Our dependent variable, psychological well-being, is an umbrella concept that encompasses a person's sense of control of their life choices and life satisfaction ([28]). However, for subsistence consumers, their role in their family's decision-making process is an important component of their autonomous functioning ([43]). Scholars who study vulnerable consumers argue that well-being should incorporate autonomy ([32]), empowerment ([ 7]), and domestic stability ([ 6]). We reflected these aspects in our measure and assessed the 392 women farmers' pre- and posttreatment psychological well-being with a five-item, five-point Likert scale (Table 1). We summed their responses to these items, computed a mean, and created a change-based dependent variable (for details on variable creation, see Table 2).[ 9]
Graph
Table 1. Variables and Measures.
| Construct | Description | Adapted From | Measure | Scale |
|---|
| Marketplace literacy program | Independent variable | Viswanathan et al. (2009) | Treatment: marketplace literacy program versus control | Treatment = 1,control = 0 |
| ΔPsychological well-being Cronbach's alpha: Field Experiment 1: pre (.51); post (.53) Field Experiment 2: pre (.78); post (.67) | Dependent variable | | My home life is stable | 1 ("Strongly disagree") to 5 ("Strongly agree"); mean of post − pre |
| I can stand up to family members if something is not right |
| I am in control of my life |
| I have a say in what happens within my family |
| I have the freedom to make my own decisions |
| Measured marketplace access | Moderator | Talukdar (2008) | Measured distance (kilometers) from participants' villages to marketplace (r) | Ratio data |
| ΔConsumer confidence Cronbach's alpha: Field Experiment 2: pre (.79); post (.83) | Dependent variable | Bearden, Hardesty, and Rose (2001) | Information Acquisition | 1 ("Strongly disagree") to 5 ("Strongly agree"); mean of post − pre |
| I know where to find the information I need prior to making a purchase |
| I know where to look for product information |
| I am confident in my ability to research important purchases |
| I ask the right questions to ask when shopping |
| I have the skills required to obtain needed info. before making important purchases |
| Consideration Set Formation |
| I am confident in my ability to recognize a brand |
| I can tell which brands meet my expectations |
| I trust my judgement when choosing brands to consider |
| I know which stores to shop at |
| I focus easily on a few good brands |
| Personal Outcomes Decision Making |
| I have doubts about my purchase decisions (r) |
| I agonize over what to buy (r) |
| I wonder if I've made the right purchase selection (r) |
| I never seem to buy the right thing for me (r) |
| The things I buy are not satisfying (r) |
| Marketplace literacy measure Cronbach's alpha: Field Experiment 2: pre (.70); post (.70) Field Experiment 3: pre (.87); post (.91) | Manipulation Check | None | I know why... | 1 ("Strongly disagree") to 5 ("Strongly agree"); mean of post − pre |
| ...buyers choose one product over another |
| ...buyers choose one shop over another |
| ...buyers gather information before buying |
| ...buyers evaluate products before buying |
| ...sellers choose to sell what they sell |
| ...sellers should understand the needs of customers |
| ...sellers gather information about marketplaces |
| ...sellers price a product in a certain way |
| Reported marketplace access | Moderator | Stifel and Minten (2017) | Self-reported distance to marketplace in minutes and cost, normalized, summed, reverse-coded | Ratio data |
| ΔQuality assessment | Dependent variable | Huang, Lurie, and Mitra (2009) | I check the quality of products before purchase | 3 = "Yes," 2 = "Maybe," and 1 = "No"; post − pre |
| ΔPrice negotiation | Dependent variable | Levy and Gvili (2020) | I negotiate to get a good price from a seller | 3 = "Yes," 2 = "Maybe," and 1 = "No"; post − pre |
| ΔEntrepreneurial intention | Dependent variable | Chen, Greene, and Crick (1998) | I intend to set up a business in the future | 1 ("Strongly disagree") to 5 ("Strongly agree"); post − pre |
| Started a microenterprise | Dependent variable | None | Did you start a NEW business AFTER attending the program? | 1 = "Yes," 0 = "No" |
10022242921998384 Notes: pre = pretreatment data; post = posttreatment data; N.A. = not present in the field experiment; ✓ = present in the field experiment; (r) = reverse-coded. For Field Experiment 1, control was the absence of the marketplace literacy program; for Experiments 2 and 3, the control was a sustainability literacy program.
Graph
Table 2. Variable Creation and Model Specification.
| Field Experiment 1 |
|---|
| Change dependent variable [(posttreatment)t+1 − (pretreatment)t] | ΔPsychological well-being(t+1)-t = (Mean Psychological well-beingt+1) – (Mean Psychological well-beingt) |
| First-difference OLS regression model | (1) ΔPsychological well-beingi,(t+1)−t = β0 + β1Marketplace Literacyi + β2Measured Marketplace Access + β3(Marketplace Literacy × Measured Marketplace Access)i + Δ∊i(t+1)−t |
| Field Experiment 2 |
| Change dependent variables [(posttreatment)t+1 – (pretreatment)t] | ΔPsychological well-being(t+1)−t = (Mean Psychological well-beingt+1) − (Mean Psychological well-beingt)ΔConsumer Confidence(t+1)-t = (Mean Consumer Confidencet+1) − (Mean Consumer Confidencet) |
| Two OLS first-difference regression models | (2) ΔPsychological well-beingi, (t+1)−t = β0 + β1Marketplace Literacyi + β2Reported Marketplace Accessi + β3(Marketplace Literacy × Reported Marketplace Access)i + Δ∊i,(t+1)−t(3) ΔConsumer Confidencei, (t+1)−t = β0 + β1Marketplace Literacyi + β2Reported Marketplace Accessi + β3(Marketplace Literacy × Reported Marketplace Access)i + Δ∊i, (t+1)−t |
| Field Experiment 3 |
| Change dependent variables [(posttreatment)t+1 − (pretreatment)t] | ΔQuality Assessment(t+1)−t = (Quality Assessmentt+1) − (Quality Assessmentt)ΔPrice Negotiation(t+1)−t = (Price Considerationt+1) − (Price Negotiationt)ΔEntrepreneurial Intention(t+1)−t = (Entrepreneurial Intentiont+1) − (Entrepreneurial Intentiont) |
| Three OLS first-difference regression models | (4) ΔQuality Assessmenti, (t+1)−t = β0 + β1Marketplace Literacyi + β2Reported Marketplace Accessi + β3(Marketplace Literacy × Reported Marketplace Access)i + Δ∊i, (t+1)−t(5) ΔPrice Negotiationi, (t+1)−t = β0 + β1Marketplace Literacyi + β2Reported Marketplace Accessi + β3(Marketplace Literacy × Reported Marketplace Access)i + Δ∊i, (t+1)−t(6) ΔEntrepreneurial Intentioni, (t+1)−t = β0 + β1Marketplace Literacyi + β2Reported Marketplace Accessi + β3(Marketplace Literacy × Reported Marketplace Access)i + Δ∊i,(t+1)−t |
| Logistic regression model | (7) P(Start a Microenterprise)t = 1/{1 + exp[β0 + β1Marketplace Literacyi + β2Reported Marketplace Accessi + β3(Marketplace Literacy × Reported Marketplace Access)i + ∊it]} |
20022242921998384 Notes: i = Indian woman farmer in Field Experiments 1 and 2 and Tanzanian tribal member in Experiment 3; OLS = ordinary least squares; t = Time, Δ = [(posttreatment)t+1 − (pretreatment)t].
We coded our focal variable, marketplace literacy, as 1 if the participant received the marketplace literacy program and 0 if she did not. Following prior research (e.g., [46]), we measured our moderator, measured marketplace access, by the road distance to the preidentified marketplace. Our field research team drove from the epicenter of each village to the hub marketplace along commonly used transportation routes and recorded the exact number of kilometers (range: 2.3 km to 19.7 km). We reverse-coded this measure to interpret our results in terms of increasing access (i.e., shorter distance in kilometers).[10]
We estimated a first-difference ordinary least squares (OLS) regression model with Δpsychological well-being as a function of marketplace literacy, measured marketplace access, and their interaction for woman farmer i with robust standard errors clustered at the village level to account for nonindependent observations within a given training session. We present our model specification in Table 2.
We present model-free evidence with a difference-in-differences analysis of pre- and posttreatment psychological well-being between the treatment (marketplace literacy program) and control (no program) groups in Table 3. We find initial support for H1. We present the model estimation results of Equation 1 in Table 4. Marketplace literacy (as manipulated through the marketplace literacy program) caused an increase in psychological well-being (b = 2.612, p <.01; H1 is supported). Measured marketplace access had no effect on a change in psychological well-being (b =.000, p =.75), which was expected, given a random assignment of our treatment across access levels. In terms of a heterogeneous treatment effect, for participants with relatively higher access to marketplaces as measured marketplace access increased, marketplace literacy caused a smaller increase in psychological well-being (b = −.402, p <.01). Thus, in support of H4, we find that marketplace literacy caused a larger increase in psychological well-being for the women farmers with lower access to a marketplace.
Graph
Table 3. Difference-in-Differences Model Free Evidence of Main Effect of Marketplace Literacy.
| Treatment Group | Control Group | Change from Pre to Post |
|---|
| Pre | Post | t | p-Value | Pre | Post | t | p-Value | Treatment | Control | t | p-Value |
|---|
| Field Experiment 1 | |
| Psychological well-being | 4.15 | 4.45 | 3.79 | <.01 | 4.14 | 4.16 | .29 | .81 | .31 | .02 | 2.07 | <.05 |
| Field Experiment 2 | |
| Psychological well-being | 3.11 | 3.62 | 1.79 | .08 | 3.17 | 3.19 | .02 | .98 | .51 | .02 | 1.72 | .09 |
| Consumer confidence | 3.18 | 3.82 | 2.10 | <.05 | 3.20 | 3.09 | −1.09 | .29 | .64 | −.11 | 3.74 | <.01 |
| Marketplace literacy | 3.20 | 4.36 | 2.11 | <.05 | 3.39 | 3.25 | −1.01 | .32 | 1.26 | −.14 | 2.23 | <.05 |
| Field Experiment 3 | |
| Quality assessment | 2.07 | 2.69 | 2.03 | <.05 | 2.13 | 2.06 | −1.54 | .56 | .62 | −.07 | 2.16 | <.05 |
| Price negotiation | 2.17 | 2.49 | 1.86 | <.10 | 2.24 | 2.26 | .34 | .72 | .32 | −.03 | 1.96 | <.05 |
| Entrepreneurial intention | 3.73 | 4.21 | 2.11 | <.05 | 3.81 | 4.17 | 1.97 | <.05 | .48 | .36 | 1.20 | .21 |
| Microenterprise start-up | 0 | 42 | N.A. | N.A. | 0 | 37 | N.A. | N.A. | 42 | 37 | N.A. | N.A. |
| Marketplace literacy | 2.56 | 3.90 | 4.56 | <.01 | 2.39 | 2.48 | .96 | .67 | 1.34 | .09 | 4.37 | <.01 |
30022242921998384 Notes: N.A. = not applicable. This table features participants' mean responses to each scale. The treatment was a marketplace literacy program versus a control (no program in Experiment 1 and a sustainability literacy program in Experiments 2 and 3). Differences between the pretreatment stage between treatment and control groups are nonsignificant in all cases (p >.25).
Graph
Table 4. Field Experiment 1 Estimation Results.
| Focal Variables | Change in Well-Being |
|---|
| β (SE) | p-Value |
|---|
| Marketplace literacy | 2.612 (.582) | <.01 |
| Measured marketplace access | .000 (.000) | .75 |
| Marketplace literacy × Measured marketplace access | −.402 (.100) | <.01 |
| Sample size | 380 |
| Adjusted R-square | .143 |
40022242921998380 Notes: The results feature beta coefficients with standard errors in parentheses.
In our second field experiment, we aimed to retest the effect of marketplace literacy on psychological well-being (H1) and the moderating effect of marketplace access on this relationship (H4). Further, we aimed to test the effect of marketplace literacy on consumer outcomes related to well-being, such as consumer confidence (H2) and the moderating effect of marketplace access on this relationship (H5). We improved on our first experiment in a few ways. First, although marketplace literacy programs have been shown to generate marketplace literacy (e.g., [56]), in our second experiment, we developed an independent measure of this construct to serve as a manipulation check. Second, because the control condition in our previous experiment was a nonintervention (i.e., no program), in this experiment, we used a different kind of educational program as an active control, one that was unrelated to marketplace literacy in content but was similar in duration and delivery. We ensured that the same instructor implemented treatment and control programs. Third, we selected villages with relatively lower access, even within the narrow range we studied, to create a strong test of our hypotheses in terms of the influence of marketplace access. Finally, we measured marketplace access on the basis of monetary cost and duration ([45]), rather than geographic distance to reach a marketplace, to test a more comprehensive measure of access.
We worked with the same field research team as in our first field experiment. We identified ten villages that had similar profiles in terms of population, square mileage, and number of working households. We randomly assigned five villages to a treatment group in which we administered a marketplace literacy program. We randomly assigned the other five villages to a control group in which we administered a program identical to the marketplace literacy program (in terms of contact hours, facilities, instructors, monetary incentive, and medium of instruction), except for topical content. The control program was a sustainability literacy program on environmental issues, such as air, land, and water pollution. Participants in the control and treatment groups were comparable across multiple variables (Web Appendix F).
In each village, the field research team recruited 25 to 28 women farmers who agreed to participate in the experiment in exchange for an incentive (410 Indian rupees, approximately US$6). Each participant completed a presurvey, an educational program (marketplace literacy program; treatment) or sustainability literacy program (control) and a postsurvey. The experiment was administered over approximately 50 days in a staggered fashion (for the timeline, see Web Appendix C). In the first phase, the field research team administered a presurvey to the 258 women farmers, in person and one at a time, in all ten villages to measure their baseline marketplace literacy, psychological well-being, and consumer confidence. Then, those in the treatment group received the marketplace literacy program, and those in the control group received the sustainability literacy program from the same instructor (for details, see Web Appendix F). The instructor traveled to the ten villages in succession to deliver the content in two-day sessions in each village. Two weeks after the 258 participants received either the treatment or the control, the research team administered a postsurvey with the same measures as the presurvey. A marketing research firm in India entered the data.
In both the pre- and postsurveys, we captured each participant's psychological well-being using the same scale as in our first field experiment (Table 1). Further, we measured consumer confidence using a multi-item Likert scale of three types of confidence ([ 8]): information acquisition, consideration set formation, and decision making. We added the three scales together and computed their mean. Then, we created two change-based dependent variables (for details on variable creation, see Table 2).
We coded marketplace literacy as 1 if the participant received the marketplace literacy program and 0 if she received the sustainability literacy program. In terms of marketplace access, for each village, the field research team identified a major hub marketplace that served the village and referenced this marketplace when they collected two measures of access in the presurvey. Participants reported the number of minutes it took and how much it cost (in Indian rupees) to reach the marketplace. We normalized the temporal distance (min = 1 minute, max = 180 minutes, mean = 48.12 minutes) and cost (min = 0 Indian rupees, max = 400 Indian rupees, mean = 73.24 Indian rupees) and then created a composite variable by adding the normalized measures together. We reverse-coded this variable, reported marketplace access, to interpret the results in terms of increasing access. Finally, we created a manipulation check of the marketplace literacy program by developing an eight-item scale of marketplace literacy and included it in the pre- and postsurveys (Table 1).
We estimated two OLS first-difference regression models, one for each outcome of Δpsychological well-being and Δconsumer confidence, as a function of marketplace literacy, reported marketplace literacy, and their interaction for woman farmer i with robust standard errors clustered at the village level to account for nonindependent observations (for our model specification, see Equations 3 and 4 in Table 2).
We present model-free evidence of our treatment effect, which provides initial support for H2, with a difference-in-differences analysis (Table 3). We present the estimation results of Equations 2 and 3 in Table 5. Marketplace literacy caused an increase in psychological well-being (marginal; b =.234, p =.08) and consumer confidence (b =.301, p <.01), which supports H1 and H2, respectively. Reported marketplace access is not associated with change in psychological well-being (b =.000, p =.87) and consumer confidence (b =.103, p =.12). For participants with relatively higher access to marketplaces, marketplace literacy caused a smaller increase in psychological well-being (b = −.848, p <.05) and confidence (b = −.168, p <.01). Thus, in support of H4 and H5, we found that marketplace literacy had a stronger impact on psychological well-being and consumer confidence for the women farmers with the lower (vs. higher) marketplace access.[11]
Graph
Table 5. Field Experiment 2 Estimation Results.
| Focal Variables | Change in Well-Being | Change in Consumer Confidence |
|---|
| β (SE) | p-Value | β (SE) | p-Value |
|---|
| Marketplace literacy | .234 (.142) | <.10 | .301 (.091) | <.01 |
| Reported marketplace access | .000 (.000) | .87 | .100 (.065) | .12 |
| Marketplace literacy × Reported marketplace access | −.848 (.431) | <.05 | −.168 (.045) | <.01 |
| Sample size | 239 | 239 |
| Adjusted R-square | .131 | .125 |
50022242921998380 Notes: The results feature beta coefficients with standard errors in parentheses.
We ran a third field experiment for a few reasons. First, we wanted to explore additional outcomes other than consumer confidence, such as consumer decision-making ability. Second, we wanted to test the effect of marketplace literacy and marketplace access on entrepreneurial outcomes to test H3 and H6. Finally, to study whether the benefits of marketplace literacy extend beyond rural Indian villages to even more remote settings, we focused on men and women in isolated tribal areas of Tanzania. We worked with a field research team with seven years of experience in the implementation of educational programs in rural Tanzania.
The research team identified four tribal villages that ranged from 0 to 13 kilometers from a weekly marketplace. Within each village, the team identified two clusters of dwellings that were geographically separated from each other, to arrive at a total of eight locations. In each location, the research team recruited approximately 25 to 30 women and men, which totaled 248 participants who agreed to participate for an incentive (20,000 Tanzanian shillings [TZS], approximately US$8).[12] Within each village, we randomly assigned one location to the treatment group, the marketplace literacy program, and the other location to the control group, the sustainability literacy program. The locations and participants were comparable across several factors (see Web Appendix G). The topics were similar to what we taught in our previous experiments.
We administered our experiment over 34 days in a staggered fashion (see Web Appendix C for the timeline). In the first phase, the field research team administered a presurvey to the 248 participants, in person and one at a time, in the eight locations. The survey was translated into the local language of Swahili. Then, the instructor traveled to the eight locations in succession to deliver the treatment in a four-hour, single-day session in each location. Those in the treatment group received the marketplace literacy program, and those in the control group received the sustainability literacy program from the same instructor. Three weeks after the treatment, the research team administered a postsurvey. Our field team in Tanzania entered the data.
In both the pre- and postsurveys, we captured participants' decision-making ability with two proxies: quality assessment (whether they check product quality before making a purchase; adapted from [24]) and price negotiation (whether they negotiate for a better price; adapted from items in engagement in price negotiation scale [[29]]) (Table 1).[13] In terms of their entrepreneurial behaviors, we measured their pre- and postentrepreneurial intention with a single-item Likert scale that captured their intention to start a business ([14]). Finally, in the postsurvey, we assessed whether they started a microenterprise three weeks after the educational intervention (1 = yes, 0 = no; 25% of the participants started a microenterprise). For the three measures that we captured in the pre- and postsurveys, we created change-based measures (Table 2).
We coded the variable marketplace literacy as 1 if the participant received the marketplace literacy program and 0 if they received the sustainability literacy program. In terms of marketplace access, we used the same measure of reported marketplace access as our second field experiment, in which we recorded the reported temporal distance (min = 0 minutes, max = 150 minutes, mean = 46.65 minutes) and monetary cost (min = 0 TZS, max = 2,500 TZS, mean = 803.46 TZS) to a weekly marketplace and created a composite measure of their normalized scores. We reverse coded this variable to interpret the results in terms of increasing access. Finally, we measured participants' marketplace literacy to subsequently serve as a manipulation check (Table 1).
We estimated three OLS first-difference regression models, one for each outcome of Δquality assessment, Δprice negotiation, and Δentrepreneurial intention, as a function of marketplace literacy, reported marketplace access, and their interaction for tribal participant i with robust standard errors clustered at the village level. We also estimated a logistic regression model for the outcome of started a microenterprise (see Table 2 for Equations 4–7).
We present model-free evidence with a difference-in-differences analysis in Table 3, which supports our manipulation and provides support for H2 but not for H3. The latter result, which allows for the possibility that we explore subsequently, is that the effect of marketplace literacy on entrepreneurial outcomes emerges at a specific level of marketplace access. We present the estimation results of Equations 4–7 in Table 6. Marketplace literacy caused an increase in quality assessment (b =.185, p <.05) and price negotiation (marginal; b =.090, p =.09), which provided support for H2. Reported marketplace access was not associated with quality assessment (b =.000, p =.23) or price negotiation (b =.000, p =.46). For those with greater reported marketplace access, marketplace literacy caused a smaller increase in quality assessment (marginal; b = −.001, p =.07) and price negotiation (b = −.002, p <.05). Thus, in support of H5, we found that marketplace literacy impacted consumer decision-making ability more for the tribal men and women with lower marketplace access.
Graph
Table 6. Field Experiment 3 Estimation Results.
| Consumer DVs | Entrepreneurial DVs |
|---|
| Focal Variables | Change in Quality Assessment | Change in Price Negotiation | Change in Entrepreneurial Intention | Started a Microenterprise |
|---|
| β (SE) | p-Value | β (SE) | p-Value | β (SE) | p-Value | β (SE) | p-Value |
|---|
| Marketplace literacy | .185 (.091) | <.05 | .090 (.540) | <.10 | .270 (.542) | .62 | .021 (.052) | .68 |
| Reported marketplace access | .000 (.000) | .23 | .000 (.000) | .46 | .109 (.055) | <.05 | .302 (.184) | <.10 |
| Marketplace literacy × Reported marketplace access | −.001 (.000) | <.10 | −.002 (.001) | <.05 | .658 (.395) | <.10 | .366 (.210) | <.10 |
| Sample size | 248 | 248 | 248 | 248 |
| Adjusted R-square | .077 | .084 | .075 | .062 |
60022242921998380 Notes: DVs = dependent variables. The results feature beta coefficients with standard errors in parentheses.
Marketplace literacy did not have a main effect on a change in entrepreneurial intention (b =.270, p =.62) or microenterprise start-ups (b =.021, p =.68). Thus, we do not find support for H3. Reported marketplace access was associated with entrepreneurial intention (b =.109, p <.05) and participants' starting a microenterprise (marginal; b =.302, p =.09). For the participants with greater marketplace access, their marketplace literacy caused an increase in entrepreneurial intention (marginal; b =.658, p =.07) and the start-up of a microenterprise (marginal; b =.366, p =.09). Although these results achieve only marginal significance, they are indicative of a major shift in entrepreneurial behavior in a context in which it is difficult to detect such changes in just three weeks. Thus, in support of H6, we find that marketplace literacy led to more entrepreneurship for the tribal women and men with higher access to marketplaces.[14]
A relevant question is whether marketplace literacy is needed to generate consumer outcomes related to well-being over and above consumer literacy alone. To assess this, we ran a field experiment in an isolated tribal region of Tanzania (Web Appendix H). We implemented the treatment (a marketplace literacy program) and an active control (a consumer literacy program that did not cover entrepreneurial literacy). We measured quality assessment and price negotiation, pre- and posttreatment. A difference-in-differences analysis showed that consumer decision making improved in both conditions but was greater for those in the marketplace literacy condition than for those in the consumer literacy condition. Thus, we demonstrate that whereas consumer literacy is important, subsistence consumers benefit to a greater extent from a marketplace-level understanding with buyer and seller perspectives to improve their consumer decision making. Although these initial results support our line of reasoning, we acknowledge that further research is needed.
We demonstrate that marketing can serve as a pathway to a better world by improving the lives and livelihoods of subsistence consumers. We now discuss the theoretical and practical implications of our findings.
We offer four takeaways from our research, which we subsequently describe in detail.
First, research on low-income and low-literate consumers should occupy a larger focus in marketing research because improvements in their well-being can have broad ripple effects. Second, whereas consumer knowledge/literacy have been tied to improved consumer decision making, in subsistence contexts, marketplace literacy, which comprises both entrepreneurial and consumer literacy, is required to improve people's lives and livelihoods. Third, whereas consumer knowledge and skills relate to knowing what to do and how to do it, knowing why marketplace exchanges between buyers and sellers occur can improve subsistence consumers' well-being. Fourth, marketplace access is an important variable that enables buying and selling and should not be taken for granted. Rather, it should be considered to assess when marketplace literacy will yield specific well-being outcomes.
A substantial body of work on subsistence marketplaces has described individuals' cognitive and affective constraints (e.g., [51]; [56]) that negatively impact their well-being. Subsistence consumers' well-being can be constrained by their inability to participate beneficially and effectively in the marketplace. We identify marketplace literacy as a pathway to address this challenge. Research on how to develop marketing interventions to improve well-being should place a greater emphasis on subsistence consumers.
Conventionally, marketing is consumer-focused, but subsistence consumers often depend on entrepreneurial ventures to meet their basic consumption needs ([ 4]), and thus, consumption and entrepreneurship are inextricably interlinked ([55]). We show that in addition to consumer-specific knowledge and skills, subsistence consumers need seller-based knowledge and skills to function effectively as consumers and entrepreneurs. This broadened view of marketing, which incorporates both buyers' and sellers' perspectives, is unique to marketplace literacy and has sizeable implications for helping subsistence consumers benefit from livelihood opportunities.
We extend previous descriptions of marketplace literacy to formally define and operationalize it. Our conceptualization and measurement of marketplace literacy captures its unique focus on a person's know-why, in addition to know-what and know-how, which gives them a deeper understanding of cause-and-effect relationships and leads to better decision making ([54]). We recommend that future work on consumer literacy incorporate not only what consumers do and how to be an effective consumer but also why to engage in effective consumer practices and why various marketing activities occur.
Despite the fact that several emerging economies suffer from chronic infrastructure constraints, work on the pervasive effects of contextual factors, such as marketplace access, is limited. We identify marketplace access as an important moderator of the relationship between marketplace literacy and well-being that leads to differing effects for psychological well-being and consumer outcomes versus entrepreneurial outcomes. Future research should take into account multiple physical deprivations to understand how they limit or enhance the effect of literacy (in its different forms) on outcomes related to well-being. A promising avenue is to examine the moderating effect of limited virtual access (due to a lack of access to the internet, bandwidth, or ability to navigate online shopping environments).
Methodologically, we address a gap in empirical work on marketplace literacy by testing its causal effect on consequential outcomes in field contexts with the use of RCTs. By recruiting a sample of individuals and implementing primary data collection methods pre- and posttreatment, we constructed a novel panel data set with participants who are extremely difficult to reach. As a result, hundreds of individuals in India and Tanzania benefited. Marketing researchers should implement RCTs with educational manipulations to identify causal effects to rule out bias that arises from omitted variables and reverse causality. Importantly, such educational RCTs benefit participants during the research process.
Marketplace literacy leads to significant improvements in psychological well-being and confidence, especially for subsistence consumers who were far from a marketplace. It also leads to tangible improvements in subsistence consumers' behaviors. For example, after the marketplace literacy program, many participants began to negotiate a product's price or check its quality. Such behaviors can result in greater cost savings and better quality products purchased. Further, just three weeks after our program, 11% of our participants far from a marketplace, and 25% of those close to a marketplace, started a microenterprise. Follow-up interviews revealed that such income-generating activities differed from their pretreatment life circumstances.
Even brief programs (4–12 hours) with rudimentary methods can aid in the development of marketplace literacy and generate substantive outcomes. Such programs can be customized for different audiences (e.g., women vs. men) and occupations (farmers vs. artists) and can have their causal impact assessed. Lessons can be reinforced through short clips on web-based smartphone applications (e.g., WhatsApp, which has widespread adoption, even in rural areas of developing economies) or through text messages on feature phones. Further, our measure of marketplace literacy can be calibrated to identify literacy levels in communities and individuals within communities. This provides a basis to create, scale, and assess physical and virtual programs that are tailored to specific audiences.
Partnerships with entrepreneurs are central for multinational companies to access underserved consumers in markets that are difficult to reach ([44]). We find that marketplace access is central in translating marketplace literacy into entrepreneurship, especially for potential entrepreneurs who live in remote areas. Marketplace literacy programs in remote areas can be supplemented with organized visits in which participants are exposed to marketplaces. We advocate for the provision of virtual marketplace access through feature phones and smartphones, both of which have rapidly increased in adoption in subsistence marketplaces ([50]). Virtual access also can enable the dissemination of marketplace literacy, such as through lesson summaries in WhatsApp clips combined with virtual learning forums.
We acknowledge that our work is restricted to subsistence contexts in emerging markets. Nevertheless, in marketing contexts in advanced economies, a certain level of literacy is assumed, even though many people in advanced economies lack marketplace literacy. Thus, marketplace literacy programs may need to be designed to address a mismatch between low literacy and low income and a literate shopping environment. Further, given that most in advanced economies operate as employees more so than entrepreneurs, marketplace literacy should include employee literacy in concert with consumer literacy. The benefits of marketplace literacy may also extend to resource-rich contexts in which people are confronted with radically new products and business models that place higher-income and more literate stakeholders in circumstances of situationally low to moderate literacy.
Policies designed to focus on broader marketplace exchanges with seller and buyer perspectives are an important alternative to the current practice of almost exclusively providing a consumer perspective (e.g., the Bureau of Consumer Protection). Further, given that multinational firms engage in corporate social responsibility initiatives to build stakeholder engagement, policy makers should hold marketers accountable for playing a central role in engendering marketplace literacy. This can be achieved through consumer and entrepreneur protection agencies, subsidies, and metrics that reward firms adhering to such standards. In terms of broader implications, we recommend that the United Nations include marketplace literacy explicitly in its sustainable development goals to build consumer and entrepreneurial knowledge and skills and improve well-being.
We conducted field experiments in India and Tanzania in rural contexts with moderately low to extremely low marketplace access. Similar research should be conducted in other low-income urban and rural contexts in other developing economies as well as advanced economies to capture the social and cultural heterogeneity that exists in how individuals obtain marketplace information. Future research also should explore individual-level heterogeneity in the effect of marketplace literacy on outcomes of well-being. For example, gender roles, income, and general education will likely influence the extent to which marketplace literacy influences well-being, especially at different access levels. We also acknowledge the potential for a Hawthorne effect, such that those who participated in our marketplace literacy program may have reported improvements in dependent variables simply because they received the education or because they were being observed, although, we addressed this issue by implementing an active control condition in Field Experiments 2 and 3 (control participants received sustainability literacy education). In light of the extraordinary challenge of obtaining such data in these remote contexts, we used self-reports to capture behavioral outcomes. Although this is typical of such research, it is nevertheless a limitation of our work. Finally, we demonstrated our effects within relatively short time spans. Future research should study longer time frames to determine the long-term effects of marketplace literacy on consumer and entrepreneurial outcomes.
Supplemental Material, sj-docx-1-jmx-10.1177_0022242921998385 - Marketplace Literacy as a Pathway to a Better World: Evidence from Field Experiments in Low-Access Subsistence Marketplaces
Supplemental Material, sj-docx-1-jmx-10.1177_0022242921998385 for Marketplace Literacy as a Pathway to a Better World: Evidence from Field Experiments in Low-Access Subsistence Marketplaces by Madhubalan Viswanathan, Nita Umashankar, Arun Sreekumar and Ashley Goreczny in Journal of Marketing
Footnotes 1 Vikas Mittal
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Madhubalan Viswanathan https://orcid.org/0000-0003-1529-0373 Arun Sreekumar https://orcid.org/0000-0003-1582-3774
5 Online supplement: https://doi.org/1177/0022242921998385
6 The term "subsistence" describes a state of having only enough resources to barely make ends meet. It captures conditions characterized by a range within the low income, from extreme poverty to the cusp of low and lower-middle income.
7 When we discuss marketplace access, we refer to access to specific marketplaces, which are concentrated and organized commercial areas that bring together a volume of buyers and sellers to allow for a variety of marketing exchanges. We illustrate the importance of marketplaces in subsistence contexts in Web Appendix A.
8 After the field experiment was completed, the field research team provided the marketplace literacy program to the control group participants so they could also derive benefits from the program.
9 We acknowledge the marginal to moderate internal consistency reliability of this five-item measure across our two studies. We captured aspects of an umbrella construct and traded internal consistency reliability for content validity from a sample in a broad and diverse domain. Further, the lower literacy and income levels of our participants played a role. Finally, we used fewer items to accommodate our participants with lower literacy in a very challenging setting. Details of psychometric analyses are available upon request.
Because we randomly assigned our treatment, the marketing literacy program, to our sample, we did not include demographics or other controls in our estimation because the experimental design rules them out as predictors of changes in our dependent variable. Further, we verify that the treatment and control groups were very similar (see Web Appendix E, Table E2).
A potential concern is that our treatment, the marketplace literacy program, might cause changes in reported marketplace access, which could then make this measure of access endogenous. To examine this, we remeasured reported marketplace access in the postsurvey and then tested whether the marketplace literacy program versus the sustainability literacy program caused a change in reported marketplace access. We did not find a change in reported marketplace access between conditions and across time (Mmarketplace lit prog pre-to-post =.06, Msustainability lit prog pre-to-post =.04, t =.22, p =.69).
The research team ensured that none of the participants were microentrepreneurs to test the effect of our treatment on starting a microenterprise. Previously, the participants did not work outside the home (i.e., were homemakers) or were farmers.
We focused on single-item, three-point scales (yes, maybe, no) as opposed to the multi-item scales used in our studies in India due to these participants' very low basic literacy. We strived to keep the data collection to a short duration.
The main effect of marketplace literacy is not significant, but its interaction with marketplace access is. This makes sense for a few reasons. First, this is in line with Simpson's paradox, which proves that a trend can appear in different groups of data but disappear when these groups are combined. For the effect of marketplace literacy to emerge, it is important to examine varying levels of access. Second, our results are consistent with other work that uses the RCT method that does not find a main effect but finds heterogeneous treatment effects (i.e., interaction effects).
References Adkins Natalie R., Ozanne Julie L. (2005a) "Critical Consumer Education: Empowering the Low-Literate Consumer," Journal of Macromarketing, 25 (2), 153–62.
Adkins Natalie R., Ozanne Julie L. (2005b), "The Low Literate Consumer," Journal of Consumer Research, 32 (1), 93–105.
Alba Joseph W., Hutchinson J. Wesley. (1987), "Dimensions of Consumer Expertise," Journal of Consumer Research, 13 (4), 411–54.
Alvarez Sharon A., Barney Jay B. (2014), "Entrepreneurial Opportunities and Poverty Alleviation," Entrepreneurship Theory and Practice, 38 (1), 159–84.
Anderson Stephen, Chandy Rajesh, Zia Bilal. (2018), "Pathways to Profits: The Impact of Marketing vs. Finance Skills on Business Performance," Management Science, 64 (12), 5559–83.
Annan Jeannie, Donald Aletheia, Goldstein Markus, Martinez Paula Gonzalez, Koolwal Gayatri. (2019), Taking Power: Women's Empowerment and Household Well-Being in Sub-Saharan Africa. Washington, DC: World Bank.
Baker Stacey M., Gentry James W., Rittenburg Terri L. (2005), "Building Understanding of the Domain of Consumer Vulnerability," Journal of Macromarketing, 25 (2), 128–39.
Bearden William O., Hardesty David M., Rose Randall L. (2001), "Consumer Self-Confidence: Refinements in Conceptualization and Measurement," Journal of Consumer Research, 28 (1), 121–34.
Bettman James R., Johnson Eric J., Payne John W. (1991), "Consumer Decision Making," in Handbook of Consumer Behaviour, Kassarjian Harold H., Robertson Thomas S., eds. Englewood Cliffs, NJ: Prentice Hall, 50–84.
Bhuiyan Muhammad F., Ivlevs Artjoms. (2019), "Micro-Entrepreneurship and Subjective Well-Being: Evidence from Rural Bangladesh," Journal of Business Venturing, 34 (4), 625–45.
Boso Nathaniel, Story Vicky M., Cadogan John W. (2013), "Entrepreneurial Orientation, Market Orientation, Network Ties, and Performance: Study of Entrepreneurial Firms in a Developing Economy," Journal of Business Venturing, 28 (6), 708–27.
Bruton Garry D., Ketchen David J.Jr, Duane Ireland R. (2013), "Entrepreneurship as a Solution to Poverty," Journal of Business Venturing, 28 (6), 683–89.
Bulte Erwin, Lensink Robert, Nhung Vu. (2017), "Do Gender and Business Trainings Affect Business Outcomes? Experimental Evidence from Vietnam," Management Science, 63 (9), 2885–902.
Chen Chao C., Greene Patricia G., Crick Ann. (1998), "Does Entrepreneurial Self-Efficacy Distinguish Entrepreneurs from Managers?" Journal of Business Venturing, 13 (4), 295–316.
Diener Ed. (1994), "Assessing Subjective Well-Being: Progress and Opportunities," Social Indicators Research, 31 (2), 103–57.
Duflo Esther. (2012), "Women Empowerment and Economic Development," Journal of Economic Literature, 50 (4), 1051–79.
Friestad Marian, Wright Peter. (1994), "The Persuasion Knowledge Model: How People Cope with Persuasion Attempts," Journal of Consumer Research, 21 (1), 1–31.
Garrette Bernard, Karnani Aneel. (2010), "Challenges in Marketing Socially Useful Goods to the Poor," California Management Review, 52 (4), 29–47.
Gaurav Sarthak, Cole Shawn, Tobacman Jeremy. (2011), "Marketing Complex Financial Products in Emerging Markets: Evidence from Rainfall Insurance in India," Journal of Marketing Research, 48, 150–62.
Global Consumption Database (2017), "The Developing World's 4.5 Billion Low-Income People Already a $5 Trillion Market" (accessed December 8, 2020), https://datatopics.worldbank.org/consumption/market.
Haq Rashida, Zia Uzma. (2013), "Multidimensional Wellbeing: An Index of Quality of Life in a Developing Economy," Social Indicators Research114 (3), 997–1012.
Hill Ronald P. (1991), "Homeless Women, Special Possessions, and the Meaning of 'Home': An Ethnographic Case Study," Journal of Consumer Research, 18 (3), 298–310.
Hill Ronald P. (2002), "Stalking the Poverty Consumer: A Retrospective Examination of Modern Ethical Dilemmas," Journal of Business Ethics, 37 (2), 209–29.
Huang Peng, Lurie Nicholas H., Mitra Sabyasachi. (2009), "Searching for Experience on the Web: An Empirical Examination of Consumer Behavior for Search and Experience Goods," Journal of Marketing, 73 (2), 55–69.
Jae Haeran, DelVecchio Devon. (2004), "Decision Making by Low-Literacy Consumers in the Presence of Point-of-Purchase Information," Journal of Consumer Affairs, 38 (2), 342–54.
Karlan Dean, Valdivia Martin. (2011), "Teaching Entrepreneurship: Impact of Business Training on Microfinance Clients and Institutions," Review of Economics and Statistics, 93 (2), 510–27.
Klinger Bailey, Schündeln Matthias. (2011), "Can Entrepreneurial Activity Be Taught? Quasi-Experimental Evidence from Central America," World Development, 39 (9), 1592–1610.
Lee Dong-Jin, Sirgy M. Joseph, Larsen Val, Wright Newell D. (2002), "Developing a Subjective Measure of Consumer Well-Being," Journal of Macromarketing, 22 (2), 158–69.
Levy Shalom, Gvili Yaniv. (2020), "Online Shopper Engagement in Price Negotiation: The Roles of Culture, Involvement and EWOM," International Journal of Advertising, 39 (2), 232–57.
Loibl Cäzilia, Cho Soo H., Diekmann Florian, Batte Marvin T. (2009), "Consumer Self-Confidence in Searching for Information," Journal of Consumer Affairs, 43 (1), 26–55.
Mano Yukichi, Iddrisu Alhassan, Yoshino Yutaka, Sonobe Tetsushi. (2012), "How Can Micro and Small Enterprises in Sub-Saharan Africa Become More Productive? The Impacts of Experimental Basic Managerial Training," World Development, 40 (3), 458–68.
Martin Kelly D., Hill Ronald P. (2012), "Life Satisfaction, Self-Determination, and Consumption Adequacy at the Bottom of the Pyramid," Journal of Consumer Research, 38 (6), 1155–68.
McGregor Sue. (2011), "Consumer Acumen: Augmenting Consumer Literacy," Journal of Consumer Affairs, 45 (2), 344–57.
Miller Grant, Mobarak A. Mushfiq. (2015), "Learning About New Technologies Through Social Networks: Experimental Evidence on Non-Traditional Stoves in Bangladesh," Marketing Science, 34 (4), 480–99.
Moorman Christine, Diehl Kristin, Brinberg David, Kidwell Blair. (2004), "Subjective Knowledge, Search Locations, and Consumer Choice," Journal of Consumer Research, 31 (3), 673–80.
Nikiforou Argyro, Dencker John C., Gruber Marc. (2019), "Necessity Entrepreneurship and Industry Choice in New Firm Creation," Strategic Management Journal, 40 (13), 2165–90.
Park C. Whan, Iyer Easwar S., Smith Daniel C. (1989), "The Effects of Situational Factors on In-Store Grocery Shopping Behavior: The Role of Store Environment and Time Available for Shopping," Journal of Consumer Research, 15 (4), 422–33.
Park C. Whan, Mothersbaugh David L., Feick Lawrence. (1994), "Consumer Knowledge Assessment," Journal of Consumer Research, 21 (1), 71–82.
Pels Jaqueline. (1999), "Exchange Relationships in Consumer Markets?" European Journal of Marketing, 33 (1), 19–37.
Ryan Richard M., Deci Edward L. (2006), "Self-Regulation and the Problem of Human Autonomy: Does Psychology Need Choice, Self-Determination, and Will?" Journal of Personality, 74 (6), 1557–86.
Ryff Carol D. (1989), "Happiness Is Everything, or Is It? Explorations on the Meaning of Psychological Well-Being," Journal of Personality and Social Psychology, 5 (6), 1069–80.
Schmidt Jeffrey B., Spreng Richard A. (1996), "A Proposed Model of External Consumer Information Search," Journal of the Academy of Marketing Science, 24 (3), 246–56.
Seymour Greg, Peterman Amber. (2018), "Context and Measurement: An Analysis of the Relationship Between Intrahousehold Decision Making and Autonomy," World Development, 111, 97–112.
Sheth Jagdish N. (2011), "Impact of Emerging Markets on Marketing: Rethinking Existing Perspectives and Practices," Journal of Marketing, 75 (4), 166–82.
Stifel David, Minten Bart. (2017), "Market Access, Well-Being, and Nutrition: Evidence from Ethiopia," World Development, 90, 229–41.
Talukdar Debabrata. (2008), "Cost of Being Poor: Retail Price and Consumer Price Search Differences Across Inner-City and Suburban Neighborhoods," Journal of Consumer Research, 35 (3), 457–71.
Tanchua Jennelyn, Shand Diane. (2016), "Emerging Markets May Offer the Most Potential for the World's Largest Consumer-Focused Companies" (accessed December 8, 2020), https://www.spglobal.com/en/research-insights/articles/emerging-markets-may-offer-the-most-potential-for-the-worlds-largest-consumer-focused-companies.
Tobias Jutta M., Mair Johanna, Barbosa-Leiker Celestina. (2013), "Toward a Theory of Transformative Entrepreneuring: Poverty Reduction and Conflict Resolution in Rwanda's Entrepreneurial Coffee Sector," Journal of Business Venturing, 28 (6), 728–42.
Venugopal Srinivas, Viswanathan Madhubalan, Jung Kiju. (2015), "Consumption Constraints and Entrepreneurial Intentions in Subsistence Marketplaces," Journal of Public Policy & Marketing, 34 (2), 235–51.
Vimalkumar M., Singh Jang B., Sharma Sujeet K. (2020), "Exploring the Multi-Level Digital Divide in Mobile Phone Adoption: A Comparison of Developing Nations," Information Systems Frontiers (published online June 16), https://doi.org/10.1007/s10796-020-10032-5.
Viswanathan Madhubalan. (2013), Subsistence Marketplaces. ebookpartnership, eText, and Stipes Publishing.
Viswanathan Madhubalan, Gajendiran Suyamprakasam, Venkatesan Raj. (2008), "Understanding and Enabling Marketplace Literacy in Subsistence Contexts: The Development of a Consumer and Entrepreneurial Literacy Educational Program in South India," International Journal of Educational Development, 28 (3), 300–19.
Viswanathan Madhubalan, Rosa Jose A. (2007), "Product and Market Development for Subsistence Marketplaces: Consumption and Entrepreneurship Beyond Literacy and Resource Barriers," Advances in International Management, 20 (1), 1–17.
Viswanathan Madhubalan, Rosa Jose A., Harris James E. (2005), "Decision Making and Coping of Functionally Illiterate Consumers and Some Implications for Marketing Management," Journal of Marketing, 69 (1), 15–31.
Viswanathan Madhubalan, Rosa Jose A., Ruth Julie A. (2010), "Exchanges in Marketing Systems: The Case of Subsistence Consumer–Merchants in Chennai, India," Journal of Marketing, 74 (3), 1–17.
Viswanathan Madhubalan, Sridharan Srinivas, Gau Roland, Ritchie Robin. (2009), "Designing Marketplace Literacy Education in Resource-Constrained Contexts: Implications for Public Policy and Marketing," Journal of Public Policy & Marketing, 28 (1), 85–94.
Viswanathan Madhubalan, Sridharan Srinivas, Ritchie Robin, Venugopal Srinivas, Jung Kiju. (2012), "Marketing Interactions in Subsistence Marketplaces: A Bottom-Up Approach to Designing Public Policy," Journal of Public Policy & Marketing, 31 (2), 159–77.
Wallendorf Melanie. (2001), "Literally Literacy," Journal of Consumer Research, 27 (4), 505–11.
World Bank (2020), Poverty and Shared Prosperity 2020: Reversals of Fortune. Washington, DC: World Bank.
Wright Peter. (2002), "Marketplace Metacognition and Social Intelligence," Journal of Consumer Research, 28 (4), 677–82.
~~~~~~~~
By Madhubalan Viswanathan; Nita Umashankar; Arun Sreekumar and Ashley Goreczny
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 91- Measuring the Real-Time Stock Market Impact of Firm-Generated Content. By: Lacka, Ewelina; Boyd, D. Eric; Ibikunle, Gbenga; Kannan, P.K. Journal of Marketing. Sep2022, Vol. 86 Issue 5, p58-78. 21p. 3 Color Photographs, 7 Charts. DOI: 10.1177/00222429211042848.
- Database:
- Business Source Complete
Measuring the Real-Time Stock Market Impact of Firm-Generated Content
Firms increasingly follow an "always on" philosophy, publishing multiple pieces of firm-generated content (FGC) every day. Current methodologies used in marketing are unfit to unbiasedly capture the impact of FGC disseminated intermittently throughout the day on stock markets characterized by ultra-high-frequency trading. They also neither distinguish between the permanent (i.e., long-term) and temporary (i.e., short-term) price impacts nor identify FGC attributes capable of generating these price impacts. In this study, the authors define price impact as the impact on the variance of stock price. Employing a market microstructure approach to exploit the variance of high-frequency changes in stock price, the authors estimate the permanent and temporary price impacts of the firm-generated Twitter content of S&P 500 information technology firms. The authors find that firm-generated tweets induce both permanent and temporary price impacts, which are linked to tweet attributes of valence and subject matter. Tweets reflecting only valence or subject matter concerning consumer or competitor orientation result in temporary price impacts, whereas those embodying both attributes generate permanent price impacts. Negative-valence tweets about competitors generate the largest permanent price impacts. Building on these findings, the authors offer suggestions to marketing managers regarding the design of intraday FGC.
Keywords: real-time marketing; microstructure; high-frequency data; firm-generated content; Twitter
With U.S. firms investing in excess of $37 billion in 2020 ([69]), social media is one of the most pervasive communication channels used by marketers ([ 5]; [36]). The result of this investment is the creation of corporate social media accounts that support firm-generated content (FGC), defined as a firm's communications disseminated through its own online communication tools ([45]). Many firms have adopted an "always on" approach in their social media marketing, disseminating multiple pieces of FGC throughout the day. Figure 1 illustrates the high-frequency approach to FGC dissemination using the example of information technology (IT) firms' activity on Twitter. Each piece of FGC is characterized by its attributes (e.g., Figure 1 shows FGC's valence and subject matter as key attributes),[ 5] and the timestamps show the dissemination time for each piece of FGC. Each piece of FGC and its timestamp can be accurately recorded to the second and mapped against the corresponding timestamp of trading activity that takes place at ultra-high frequency, that is, at subsecond intervals (Hasbrouck and Saar 2013). As a result of these high-frequency activities, large volumes of intraday data emerge. For example, an S&P 500 IT firm can issue in excess of 6,000 tweets in a given month, and trading in a single firm's stock often yields well over 10 million trading-related messages (e.g., quotes, cancellations, transactions) during the same interval (see Web Appendix A).
Graph: Figure 1. A high-frequency approach to FGC dissemination on the example of IT firms' activity on Twitter.
Current marketing methodologies are challenged when analyzing high-frequency data because the trading data, which is used in capturing the impact of FGC on firm value, is characterized by unequal time intervals. Low-frequency event studies using end-of-day price measures and time series analysis, such as standard vector autoregressive (VAR) models, are unable to effectively address these problems (see Web Appendix B). These methods rely on discrete and uniform time intervals that do not align with the time intervals associated with high-frequency trading data, which encapsulate the intraday evolution of firm stock price. Often, these methods aggregate trading data to regular intervals, which can lead to the elimination of upwards of 99% of intraday trading observations in studies employing end-of-day price data (see Web Appendix C). The frequency of FGC as an intraday variable and the effects of other non-FGC events potentially further bias low-frequency analyses that employ standard low-frequency analytical methods. Furthermore, the richness of such an assessment and marketing researchers' understanding of "always on" strategies are compromised unless the research identifies both the short- and long-term impacts of this form of marketing ([26]). Consequently, the findings derived from current examinations lack detailed ex post insights on the impact of FGC generated at intraday frequencies, which leaves marketing managers unable to effectively design future intraday marketing resource allocation strategies ([41]).
Employing the market microstructure approach, which relies on ultra-high-frequency trading data analysis, we investigate the stock price impact of FGC where price impact is defined as the impact on the variance of stock price. This approach of estimating price impact of FGC as impact on the variance of stock price rather than on changes in the level of stock price is driven by both methodological and theoretical necessity. Although estimating level changes in stock price as a result of an event such as FGC dissemination could be approached from the perspective of computing simple price impact measures, when working with high-frequency data, this approach is inadequate for at least two reasons. First, simple price impacts are misleading when trades in stock markets are serially correlated. Second, in the presence of transient impacts, as is the case in this study (e.g., we capture price impact at second-by-second intervals), simple price impacts rely on getting the timing just right, which is methodologically unfeasible when investigating large data sets, as is the case here. Our methodological approach is in line with market microstructure theory, addresses the previously outlined issues, and is consistent with the microstructure literature ([11]; [65]). Using this approach and assessing S&P 500 IT firms' Twitter activity, we contribute to the marketing literature at three levels.
First, by using high-frequency data, this is the first study to document the subminute impact of individual pieces of FGC disseminated during the day; therefore, the insights presented are unlikely to be affected by the confounding effects that the use of end-of-day data is susceptible to. We estimate the price impact of FGC at the second and minute levels by computing the variance of fast-paced (e.g., subsecond-by-subsecond) intraday changes in stock price as they occur in financial markets ([11]; [12]; [42]), thereby demonstrating the instantaneous impact of FGC. We obtain the variance of intraday changes in price through state-space modeling with Kalman filtering. By doing so, we contribute to the emerging stream of marketing research studying ( 1) the impact of FGC on firm financial outcomes ([ 8]; [15]) and ( 2) real-time marketing ([64]), and we respond to research priorities established by the Marketing Science Institute (2018), emphasizing the need to help marketers "get marketing right" by providing insights into the instantaneous impact of FGC.
Second, using the microstructure perspective, we reveal both the permanent and temporary price impact of FGC as new forms of FGC impact on firm-level performance. In the process, we address [26] call for research capable of identifying both the short- and long-term financial impact of a marketing activity, which has thus far remained difficult to quantify. By being able to distinguish between temporary and permanent price impacts at the fine-grained level of analysis, marketing managers can demonstrate both the short- and long-term impacts of FGC on firm financial performance ([53]). This, in turn, will allow them to overcome short-termism in marketing and improve long-term growth initiatives ([18]; [58]).
Finally, to support intraday actionable FGC design, we examine the extent to which key attributes of FGC, including content valence and subject matter, influence the occurrence of permanent and temporary impacts of FGC on price. Although FGC valence and subject matter have been examined by previous research (e.g., [20]; [27]; [36]) and are recognized as key components of marketing excellence ([39]), what constitutes the "right content" is largely unknown according to research priorities recently published by the Marketing Science Institute (2020). We show that FGC reflecting only one of the attributes of valence (positive or negative) or subject matter (consumer or competitor orientation) generates temporary price impact, whereas FGC that incorporates both valence and subject matter is associated with permanent price impacts on stock price and thus correlates with long-term firm performance.
Using a two-stage least squares (2SLS) estimation framework to examine S&P 500 IT firms' Twitter activity, we find that negative- or positive-valence tweets are consistently linked with a reduction in permanent price impact and an increase in temporary price impact. Similar findings are obtained when examining tweets that only reflect a consumer or competitor subject matter, although the reduction in permanent price impact and increase in temporary price impact they elicit are of smaller magnitudes. These findings indicate that tweets reflecting only valence (positive or negative) or subject matter (consumer or competitor orientation) result in temporary price impacts, which is commonly associated with the incorporation of noise into the price discovery process ([60]). This type of effect has not been studied previously in the marketing literature; however, given that it is a source of uncertainty in the value of firms that can increase a firm's cost of capital ([17]), it demands attention. Employing the market microstructure approach to exploit the variance of high-frequency changes in stock price, this is the first study that reveals tweets' temporary price impacts and identifies tweet attributes that elicit such short-term impacts on price.
We further find that tweets that embody both attributes—valence and subject matter—generate permanent price impacts; however, this impact varies according to the type of valence and subject matter. The results evidence the importance of interaction effects between tweet valence and subject matter in generating a higher permanent price impact. The average negative- and positive-valence tweet, when viewed through the lens of a consumer or competitor orientation, generates a permanent price impact, and a competitor-oriented tweet with a negative valence is likely to have the highest permanent price impact. This is a crucial finding from the perspectives of marketing practice and intraday social media marketing strategy design because valence, as a singular attribute, is associated with decreasing permanent price impact. Our research shows that information-rich tweets that include both variance and subject matter can result in permanent price impacts, and it demonstrates investors' ability to act on information contained in FGC at subsecond levels (Hendershott et al. 2011; [65]).
To illustrate the relevance and magnitude of these findings, in Figure 2, tweets A and B are characterized by negative and positive valence, respectively, and tweets C and D reflect only consumer and competitor orientation, respectively. In line with our research findings, the permanent price impacts associated with these tweets are more than three standard deviations lower than the average permanent price impact estimates for all the 153,041 tweets in our sample and are therefore below the 10th percentile of the estimates. In contrast, tweets E to H reflect varying combinations of both valence and subject matter (consumer or competitor orientation). Consistent with our findings, we show that these tweets generate permanent price impact estimates ranked above the 90th percentile in our sample of tweet trades' permanent price impact estimates. The temporary and permanent impact estimates for the average tweet are as large as 279 and 187 times, respectively, what we document for the average regular intraday transaction in our sample.
Graph: Figure 2. Examples of tweets characterized by valence and subject matter (consumer and competitor orientation).
Firms increasingly use social media because it provides greater reach and can be less costly than traditional channels for FGC dissemination ([45]). FGC is often posted several times a day ([41]) and serves as a valuable source of high-frequency marketing data that can offer insights into the growth potential of a firm ([18]). With the advancement of data collection tools ([79]), marketing researchers can now record each piece of FGC and create large data sets depicting FGC attributes and their dissemination time. Recorded with accuracy to the second, FGC can then be mapped against the corresponding trading activity that takes place at subsecond intervals and used to study its financial impact. However, measuring the impact of FGC sampled at intraday levels requires marketing researchers to be able to utilize high-frequency trading data with observations occurring at unequal time intervals.
The market microstructure approach to estimating price impact offers marketing researchers tools to, piece-by-piece, algorithmically link FGC to time-specific trading activity at a fine-grained level of analysis (e.g., subseconds, seconds, minutes). Unlike symmetrical asset pricing models, market microstructure recognizes that a firm's stock price is only informationally efficient to the extent that it reflects all available and relevant information ([21]). A firm's stock price, while reflecting information, is also distorted by noise generated by (temporary) non-information-based factors. Such factors can include trading frictions due to low levels of liquidity, defined as the ability to trade large quantities of a firm's stock quickly with little or no price impact ([ 2]; [28]), or the activity of traders who lack adequate information regarding the value of a stock, known as "uninformed traders" in the market microstructure literature ([24]; [47]). Estimating the proportion of stock price driven by information (relevant to the value of a firm) and the proportion driven by noise is a critical aspect of the analyses many studies conduct in the market microstructure literature (see Web Appendix D). However, this holistic view of both temporary and permanent price impacts is often missing from marketing research. The market microstructure approach allows marketing researchers to estimate the price impact of FGC at high frequencies and to identify both types of price impacts. A crucial element in such analysis is knowing the "event time" (i.e., timestamp), which refers to the time at which an event occurs, such as the FGC dissemination time. By deploying time series models to estimate changes in the components of price at high-frequency intervals (e.g., seconds) and then linking the FGC timestamp to the components, the instantaneous impact of FGC on firm stock price can be estimated.
Extant research has primarily focused on linking FGC with consumer behavior ([15]; [13]; [36]; [45]; [55]; [72]) and firm performance ([ 8]; [15]; [64]) (see Web Appendix E). With the focus on firm performance, [15], [ 8], and, most recently, [64] show that FGC affects firm value. [15] document an indirect effect of FGC volume on shareholder value measured using abnormal returns and idiosyncratic risks. [ 8] were the first to demonstrate the direct impact of humorous FGC on firm value as measured by abnormal stock market returns. To demonstrate these impacts, they employed an event study that estimated abnormal returns and VAR modeling. However, these methods, as deployed, focus on daily activity, which can result in aggregation bias and misevaluation of FGC's impact on firm value ([64]). Moreover, the richness of such an assessment is compromised because short- and long-term impacts of FGC are not estimated ([26]; [58]). Finally, current marketing methods do not examine FGC attributes at a fine level of granularity (i.e., intraday frequencies), preventing marketing managers from moving beyond a "throw it on the wall and see what sticks" strategy ([37], p. 47) in the design and dissemination of intraday FGC ([36]).
The market microstructure approach responds to calls for more powerful methodological approaches that allow marketing researchers to harness the potential of rich data sources and develop insights capable of advancing theory and informing contemporary marketing practice (e.g., [18]; [36]; [49]; [79]). The fine-grained level of analysis available using a market microstructure approach overcomes the limitations of low-frequency methodologies, such as VAR and daily event studies, to study FGC and its impact on firm value (i.e., it estimates the variance in firm stock price following FGC dissemination). Utilizing high-frequency intraday data, it adds richness to the assessment of FGC's financial impacts by distinguishing between permanent and temporary price impacts.
Temporary price impacts are short-term impacts that result in momentary changes in the price of a stock before it returns to its pre-event (e.g., pre-FGC) value, and they are often the result of uninformed trader activity (see Web Appendix D). Uninformed trader activity could be driven by several factors; for example, it could be linked to investor uncertainty about the relevance of information ([34]; [38]) or trading friction due to liquidity constraints ([ 2]; [16]). Ignoring temporary price impacts can lead to misunderstanding the total impact of FGC, with prior research suggesting that temporary price impacts result in larger transaction costs ([14]) and firm cost of capital ([17]). In contrast, an event (e.g., FGC) can generate a permanent price impact and result in the price attaining an enduring new value after the event. This occurs when the event provides information that updates informed investor/trader expectations related to a firm's long-term performance ([52]). Importantly, the microstructure approach also supports intraday actionability by assessing how the attributes of information signaled by these events influence temporary and permanent price impacts.
Consistent with the market microstructure literature, this study estimates the permanent and temporary price impact of FGC by first conducting a state-space decomposition of firm stock price into its efficient (permanent) and inefficient/noise (temporary) components and then linking the changes in these components to individual pieces of FGC. State-space modeling is a tool for modeling an observed variable as the sum of unobserved variables ([35]), and it is commonly used for the decomposition of price ([11]; [35]; [57]; [65]). Due to its efficiency when applied to ultra-high-frequency data like stock price movements, the state-space modeling approach for decomposing price has significant economic and methodological advantages over other commonly used methods ([31]) such as VAR models.
An assumption underlying a standard VAR model is that data are sampled at regular frequencies, as variables at time t are regressed on variables dated at t − 1, t − 2, and so on. However, FGC and intraday trading data are often sampled at unequal time intervals, which suggests there would be many instances of missing variables in a model calibrated on regular time intervals ([63]). The modeling of such data using VAR requires the alignment of variables misaligned in time either downward, by aggregating the data to a lower frequency, or upward, by interpolating the high-frequency data with heuristic rules such as polynomial fillings. Downward alignment eliminates potentially valuable information in the high-frequency data. Data aggregation is problematic ([68]), as it can alter the lag order of autoregressive moving average models ([ 1]), reduce the efficiency of the parameter estimation and forecast ([74]), affect Granger causality and cointegration among component variables ([54]), and induce spurious instantaneous causality ([10]). Upward alignment has also been deemed inefficient and dubious ([61]) because a VAR approach assumes that the model specifies the high-frequency data-generating process. However, interpolation is not based on the multivariate model that generates the data but instead on heuristic rules, which, at a minimum, inevitably incorporate noise into the data and distort it.
State-space modeling offers a solution to the irregular frequency challenge inherent in intraday transaction data ([62]). Specifically, the use of state-space modeling with a Kalman filter in maximum likelihood estimation of parameter estimates ensures maximum efficiency in dealing with unequal time intervals or irregular frequency in data. The use of a Kalman filter accounts for changes across periods of analysis with missing observations. This is a critical consideration in the use of state-space modeling for decomposing high-frequency time series because standard approaches do not deal with the "missing observations" caused by unequal data intervals. For example, estimating a standard autoregressive framework implies truncation of the lag structure and could potentially discount valuable information in high-frequency data. Using the Kalman filter facilitates the decomposition of any realized change in the time series (e.g., variance in the stock prices), such that the permanent or temporary component at any interval is estimated using all past, present, and future observations in the series. Thus, the purpose of filtering is to ensure that estimates are updated with the introduction of every additional observation ([19]).
With the estimation of FGC's temporary and permanent price impacts, marketing researchers can explore how FGC attributes influence the occurrence of these two types of price impact, which are driven by the existence of heterogeneously informed trading agents in financial markets ([29]; [60]). Thus, how information events, such as FGC, are observed and deciphered vary significantly between the two main groups of agents in financial markets: the informed and uninformed traders/investors (for a discussion on how the activities of informed and uninformed traders drive the asymmetric effects of information events in financial markets, see Web Appendix D). The valence and subject matter of FGC are attributes that should provide information signals to (informed) investors and thus generate a permanent price impact. This is because FGC subject matter (consumer and competitor orientation) relate to a firm's competitive advantage ([46]; [48]), which is not often public and can be difficult to observe because it is embedded in a firm's culture ([23]). The role of valence has also been documented in the literature ([71]; [76]), with the impact of negative valence appearing to be stronger than that of positive valence and thus more commonly associated with permanent price impacts ([75]). There is also reason to expect that FGC valence may interact with FGC subject matter and induce a permanent price impact. The basis for this expectation comes from a branch of signaling theory recognizing that signal recipients combine information signals to make more informed decisions ([ 6]; [73]).
Although ample evidence suggests that FGC valence and subject matter may generate a permanent price impact, note that the price impact of FGC cannot occur without trading in financial markets. Trading activity incorporates the information and/or noise content of an event (e.g., FGC) into price. Therefore, because trading agents in financial markets are heterogeneously informed due to how they observe and decipher the information content of events, their trading activities also generate varied price impacts. Specifically, a permanent price impact will arise as a result of trading activity by traders/investors who have been able to correctly decipher the information content of FGC—these are the informed traders. Conversely, the trading activity of those unable to decipher the information content of FGC (i.e., uninformed traders) will only induce temporary price impacts ([24]) because their trading activity is uncorrelated with firm value ([ 4]; [29]). Accordingly, FGC that incorporates valence and subject matter can be associated with both permanent and temporary price impacts simply because of heterogeneously informed trading agents. The trading activity of informed traders thus contributes to the efficient component of price (i.e., permanent price impact), which is driven by information, whereas the activity of uninformed traders incorporates noise (i.e., temporary price impact), which is uncorrelated with firm-relevant information.
We examine the instantaneous stock market impact of FGC by studying S&P 500 IT firms' activity on Twitter. Twitter is a social media communication channel characterized by "fast-paced and short-lived information flows" ([50], p. 177), from which deep insights can be derived if appropriate methods are developed ([79]). In addition, with 92% of firms tweeting multiple times a day ([10]), Twitter FGC is an example of high-frequency intraday marketing data. Finally, the Securities and Exchange Commission's Regulation Fair Disclosure recognizes Twitter FGC as potentially carrying "useful" information for investors. For these reasons, Twitter provides a suitable context to study. We study the IT sector because IT firms are often considered early adopters of trends ([ 7]). The IT sector provides a comprehensive sample of firms disseminating multiple pieces of FGC throughout the day (see Web Appendix F). It is a major driver of economic activity, with the leading five IT firms in the United States accounting for more than 22% of the S&P 500 ([30]). Globally, the IT sector is valued at $11.5 trillion, representing over 15.5% of the global gross domestic product ([14]). Finally, the diverse consumer base of IT firms is useful for characterizing the relevance of FGC subject matter (consumer and competitor orientation) and its interaction effects with valence. A review of 10-K filings for each firm shows that 7% of the sample consists of firms marketing solely in B2C markets, 72% solely in B2B markets, and 21% selling in both B2C and B2B markets.
We obtained a sample of tweets using an application programming interface (API) to access Twitter data. In line with previous research ([50]; [77]) and following [16], we employed the API to access tweets from corporate accounts for S&P 500 IT firms. In total, we obtained 153,041 firm-generated tweets from 64 firms, which we then used in our analysis. On average, this is 2,391.2 tweets per firm over our sample period spanning January 2013 to August 2018. It should be noted that these are tweets that fall within the limits of Twitter's API in terms of the maximum number of tweets that can be accessed over a given time period. Among the IT firms we initially selected, we eliminated seven because either they did not have established corporate Twitter accounts or Twitter's API limited access to their data. Firms included in the sample engaged in high-frequency intraday marketing activity. On average, they generated a minimum of 1.07 to a maximum of 37.03 tweets a day, with the total average equaling 4.42 tweets per firm per day (see Web Appendix F), which confirms the appropriateness of the selected sample. Table 1 shows the sample of ten S&P 500 IT firms generating the highest number of tweets per day.
Graph
Table 1. Twitter Data Sample.
| S&P 500 IT Firm | Number of Tweets | Number of Tweet Days | Minimum Number of Tweets per Day | Maximum Number of Tweets per Day | Average Number of Tweets per Day | Single Tweet Days (%) | Number of Tweets Excludeda | Number of Days Excludeda |
|---|
| Red Hat | 3,102 | 204 | 2 | 155 | 15.13 | 0 | 20 | 25 |
| DXC Technology | 2,239 | 155 | 3 | 30 | 14.35 | 0 | 5 | 8 |
| CA | 3,052 | 232 | 1 | 69 | 13.09 | .26 | 0 | 27 |
| Cognizant | 3,094 | 294 | 1 | 32 | 10.48 | .65 | 34 | 36 |
| Oracle | 3,187 | 331 | 1 | 105 | 9.59 | .85 | 20 | 24 |
| F5 Networks | 3,041 | 355 | 2 | 32 | 8.54 | 0 | 15 | 30 |
| Gartner | 3,229 | 393 | 1 | 85 | 8.19 | .77 | 16 | 24 |
| FLIR Systems | 3,228 | 413 | 1 | 30 | 7.79 | 1.33 | 29 | 27 |
| ANSYS | 3,141 | 425 | 1 | 42 | 7.37 | 1.97 | 22 | 20 |
| PAYCHEX | 3,072 | 428 | 1 | 68 | 7.16 | 1.46 | 10 | 30 |
- 2 a Excluded due to excessive return volatility.
- 3 Notes: Table 1 reports the frequency statistics for a sample of tweets generated between January 8, 2013, and August 17, 2018, for S&P 500 IT firms with stocks included in the S&P 500 Index. The table shows the ten firms with the highest average number of tweets per day. Web Appendix F provides the full sample of 64 S&P 500 IT firms.
We recorded each tweet with a timestamp to the nearest millisecond.[ 6] We then used these timestamps to obtain corresponding ultra-high-frequency stock trading activity data from the Thomson Reuters Tick History v2 database in Datascope for each tweet in the sample (see Table 2). This stock trading data supplemented the Twitter data set. Our data set included data for the trading days between January 2013 and August 2018. After performing data cleaning using the criteria consistent with that of [19] and [32], the stock trading data included 8,177,183,865 instances of trading activity or messages (i.e., quotes, cancellations, and transactions), which includes 520,356,393 transactions and 7,656,827,472 orders.[ 7]
Graph
Table 2. Trading Data Descriptive Statistics
| Messages | Transactions | Orders |
|---|
| Before Cleaning | 8,182,063,205 | 522,403,178 | 7,659,660,027 |
| After Cleaning | 8,177,183,865 | 520,356,393 | 7,656,827,472 |
| Percentage of Data Removed After Data Cleaning | .06% | .39% | .04% |
1 Notes: Table 2 reports precleaning and postcleaning statistics for the number of messages (quotes, cancellations, and transactions) generated for 64 S&P 500 IT firms between January 8, 2013, and August 17, 2018. Data cleaning is completed by following Chordia, Roll and Subrahmanyam (2001) and Ibikunle (2015), and each individual message is measured U.S. currency.
After excluding days (and tweets generated on these days) with comparatively high levels of price volatility, the descriptive statistics show that the average time between trades is 7.159 seconds and the mean number of tweets per firm during the sample period is 2,377.22. The mean number of tweets per day is 54.03, and the mean number of tweets per day per firm is.844 (see Table 3 for details).
Graph
Table 3. Descriptive Statistics of Tweet Activity.
| Mean Number of Tweets | Minimum Number of Tweets | Maximum Number of Tweets |
|---|
| Tweets per Day per Firm | .844 | .00 | 6.79 |
| Tweets per Firm | 2,377.22 | 30.00 | 3,040 |
| Tweets per Day | 54.03 | 0.00 | 353.00 |
4 Notes: Table 3 reports precleaning and postcleaning statistics for the number of messages (quotes, cancellations, and transactions) generated for 64 S&P 500 IT firms between January 8, 2013, and August 17, 2018. Each individual message is measured U.S. currency.
To investigate tweets' permanent and temporary price impacts respectively, we first used state-space modeling to estimate the permanent and temporary components of price at a given time interval using trading observations within that time interval.[ 8] The primary interval of interest was one second; however, we estimated for one minute as well for robustness. Next, we linked these estimates to firms' tweet activity using tweets' timestamps, which were labeled to the second. Thereafter, we estimated the impact of each tweet on the temporary and permanent components of price by estimating the corresponding absolute change in the components following each tweet as the respective temporary and permanent price impacts. The methodological steps are outlined next.
The first step involved modeling price as the sum of a nonstationary permanent (information-driven) component and a stationary temporary (noise) component.[ 9] In this step, the only relevant observations were the prices of the 520,356,393 transactions obtained from the Thomson Reuters Tick History v2 database. These prices were defined as the prices of stocks at intraday periods and intervals. In its simplest form, the structure of the state-space model for price, a multiple of S stock prices, T intraday periods, and N intervals, is expressed as follows:
Graph
( 1)
and
Graph
( 2)
where
Graph
( 3)
for s = 1,...,S, = 1,...,T, and t = 1,...,N; and t index event and clock times, respectively ([56]); and an event occurs when a transaction is recorded. Thus, T = 520,356,393 and N equals the number of one-second or one-minute intervals during a stock trading day. is the price of stock s at interval t and period , is a nonstationary permanent component of the price of stock s at interval t and period , is a stationary transitory component of the price of stock s at interval t and period , and is an idiosyncratic disturbance error in the permanent price component of stock s at interval t and period . and are assumed to be mutually uncorrelated and normally distributed.[10]
The model captured in Equations 1–3 is a special case of the general state-space representation. The standard state-space model is formulated for a vector of time series with a frequency/time interval , and this is given by (for simplicity, we temporarily ignore the stock notation and period ):
Graph
( 4)
where disturbances and are mutually and serially uncorrelated. The initial state vector is also uncorrelated with the disturbances. The mean vector and variance matrix are usually implied by the dynamic process for in Equation 4 ([57]). The remaining terms, and , are system matrices and are generally assumed to be fixed for . The elements of these system matrices are usually known; however, some elements that are functions of the fixed parameter vector need to be estimated. Equations 1 and 2 can be represented as the state-space Equation 4 by choosing as a single time series (this is the stock price series in this study), where , , , and . We note that and vary for each frequency for . Unlike standard variable decomposition approaches, this model naturally deals with irregular frequency/missing observation issues because the Kalman filter is used for its estimation, which is critical in a high-frequency analysis.[11]
The structure of the model shows that only changes in (now reinstating the stock notation and period ) affect price permanently, and is temporary because its effects are transient and hold no significance for long-term firm performance. This is because this model decomposes price into two parts. The first, , captures smoothed (constant) changes in price, which are driven by informed trading activity, while the second captures irregular changes in price, which deviate from the smoothed evolution and are therefore driven by uninformed trading activity (noise or friction in the pricing process). By using maximum likelihood (constructed using the Kalman filter), we estimated (i.e., permanent component) and (i.e., temporary component), where t is equal to either one second or one minute. Specifically, we first partitioned our sample into one-second and one-minute (clock) intervals, and then estimated and for these intervals by using the prices at different event periods ( ) during the intervals. This suggests that, as in [57], our permanent and temporary components ( and ), as estimated using the state-space model, were time variant (see Table 4 in Menkveld, Koopman, and Lucas [2007, p. 220]). We imposed the time-variant structure to be consistent with the time intervals studied in subsequent multivariate regressions, which is in line with [11], who also compute time-variant permanent and transitory components of price.
Graph
Table 4. Permanent and Temporary Components of Price: Descriptive Statistics.
| Price Component | Mean | Median | Standard Deviation | Minimum | Maximum |
|---|
| Temporary price component | .011 | .008 | .009 | .000 | .297 |
| Permanent price component | .055 | .010 | .048 | .000 | .644 |
5 Notes: Table 4 reports descriptive statistics for the permanent and temporary impact of tweeting. and are the respective estimates of the temporary and permanent components of the price for firm stock s at interval t, estimated by maximum likelihood (constructed using the Kalman filter).
We used the Kalman filter to evaluate the conditional mean and variances of the state vector (ignoring the stock notation and period ) given past observations : , , To initialize the Kalman filter, we also had and , where . This initialization only works if is a stationary process. However, as in our case, is often not a stationary process because it is obtained from stock price series, which are inherently nonstationary given the rational expectation of economic growth over time. Therefore, "diffuse initialization" (i.e., infinite variance distribution; see [44]) is used and estimated by numerically maximizing the log-likelihood. This is evaluated with the Kalman filter due to prediction error decomposition. According to the structure of the state-space model, our estimated outputs, and , were modeled as variances of permanent and temporary components of price, respectively. is a proxy for information reflected in the price (i.e., the permanent component of price), and is a proxy for noise reflected in the price (i.e., the temporary component of price). Stock prices should only experience permanent movements because of the arrival of new information; thus, we would expect to be higher than . The two estimated coefficients are variances; therefore, the coefficient encapsulating information , which is the primary driver of price from an efficient market perspective, should be larger. captures frictions/noise and should therefore have a lower value.[12]
Our empirical framework required linking an individual intraday tweet to a corresponding trade/transaction with price pt in our sample. We call each tweet-linked trade a "tweet-trade," and t, in this case, corresponds to both trade and time. Accordingly, we designated a trade in the stock of a firm as a "tweet-trade" if it was the first trade to occur immediately after a tweet in our sample and if it occurred within 60 seconds of the tweet. For robustness, we varied this threshold but find our inferences to be consistent if the threshold is reduced to 30 and 45 seconds, suggesting that the link between tweets and trading of firms' stock is not merely coincidental. The tweet-trade's time of occurrence allowed us to link a tweet to a corresponding pair of and , which we estimated for the one-second and one-minute intervals covered by our sample period, including those with no tweet-trades. Thus, each second and minute in our sample has a corresponding set of and . We could therefore determine the information reflected in the price (i.e., price efficiency) and noise contained in the price at every second or minute. This information allowed us to estimate whether the change in both components was occasioned by the posting of a tweet. Table 4 presents the descriptive statistics for the one-minute intervals, including tweet-trades. is higher than , which is consistent with our expectation that most of the observed tweets at time t reflect fundamental information rather than frictions or transitory components of price. This is also in line with microstructure literature ([11]; [35]; [57]; [65]).
The next step in our analysis was determining how a tweet/tweet-trade changes the composition of price with regard to and , which is required in further analysis when examining the impact of tweet valence and subject matter (i.e., consumer and competitor orientation). We linked each tweet-trade to a pair of and using the tweet-trade timestamps at the second level and then computing 30-second percentage absolute changes for both and . The changes in and following a tweet are designated as (permanent price impact) and (temporary price impact), respectively (45- and 60-second percentage changes are also computed for robustness)[13]:
Graph
( 5)
Graph
( 6)
Thereafter, we also constructed and for each non-tweet-trade in our sample. Using these measures, we constructed daily ratios of the impact of a non-tweet-trade relative to that of an average tweet-trade in stock s during day d. We then tested the null that the mean daily ratios in stock s equal 1 on average across our sample period by using their standard errors for statistical inference. We expected to reject the null if the tweet-trades, on average, generated a larger or lower price impact than all the trades on an average day.[14] We present the results of the hypothesis testing in Table 5. The ratios employed in the analysis were winsorized at.5 and 99.5 percentiles within each stock. This statistical approach was consistent with prior marketing research ([ 9]), and it allowed us to eliminate outliers or extreme values and improve the chance of obtaining statistically significant estimates. Winsorization was also necessary due to the inherent noisiness of high-frequency trading data used in estimating and .
Graph
Table 5. Ratios of the Price Impact of Tweet-Trades to the Price Impact of Other Trades.
| Price Impactt | 60-Second Threshold | 45-Second Threshold | 30-Second Threshold |
|---|
| Temporary price impact | 279.67***(7.51) | 230.12***(9.58) | 222.55***(10.23) |
| Permanent price impact | 146.83***(4.33) | 178.87***(3.21) | 189.04***(5.17) |
- 6 ***Statistical significance at the.01 level.
- 7 Notes: The t-statistics testing the null that the ratios equal 1 are presented in parentheses. and , which correspond to the temporary price impact and permanent price impact for stock s at time/interval t, are obtained from and (defined in Table 4). The ratios of and for each tweet-trade to the and for the other trades during an average trading day are then computed. The mean ratios are presented in Table 5 and their corresponding t-statistics are reported in parentheses.
The estimates in Table 5 show that, on average, tweet-trades generate larger permanent and temporary intraday price impacts than non-tweet-trades. All the estimates are statistically significant at the.01 level, thus refuting the null hypothesis that there is no difference between the impact generated by tweet-trades and other trades. The price impact of a tweet-trade is 150–300 times larger than that of the average trade. Using the 60-second threshold, , which corresponds to the temporary price impact generated by the average tweet-trade, we find that the average tweet-trade has 279.67 times the impact of the average trade, suggesting that tweets result in large but momentary movements of price. This finding suggests that FGC generates temporary effects that can induce increases in the cost of trading a firm's stock and the firm's cost of capital ([14]; [17]). , the permanent impact of the average tweet-trade, is about 146.83 times larger than the average trade's permanent impact, suggesting that FGC can cause investors to update their expectations about a firm's future performance, which in turn leads to price movement. Overall, the analysis indicates that, on average, tweet-trades occurring in the wake of a potentially information-laden tweet substantially impact stock prices both permanently and temporarily relative to non-tweet trading activity.
Estimating the effects of tweet-trades within subminute to minute windows addresses methodological issues associated with the occurrence of confounding events. Therefore, to a very high level of accuracy, we can attribute estimated temporary and permanent price impacts to the observed FGC. Given the fine-grained level of analysis, it is highly unlikely that any other relevant event could be driving the effects we capture. The sampling at high-frequency intervals also raises the question of whether investors and other trading agents could digest and act on the contents of tweets within the price impact windows we examine. Addressing this question requires an understanding of the nature of trading in financial markets today, especially in the case of highly traded stocks such as the S&P 500 stocks in our sample. Today's markets are dominated by algorithmic traders (commonly known as "algos") capable of digesting and acting on information in FGC (e.g., tweets) within the windows we examine in our analysis. The effects of this speed of activity are evidenced by the findings of [65], who, using S&P 500 stock data, show that information arriving in U.S. markets is exploited within seconds and that this activity is driven by algorithmic trading.
Although all the ratios are statistically significant and suggest that FGC influences the permanent and temporary components of price, the obvious question is how economically meaningful tweet-trades are compared with other events that impact stock price. To answer this question, we conducted further analysis to examine the corresponding ratios of other large-impact non-tweet-trades in the same period by computing ratios similar to the ones presented in Table 5. This involved substituting a permanent price impact measure for each tweet-trade with that of other trades generating price impacts corresponding to one standard deviation or more above the daily mean in each stock. The obtained average ratios for the three thresholds are 7.9, 5.2, and 1.3 for the 30-, 45-, and 60-second windows, respectively. The inference drawn from this analysis is that the information content of tweet-trades is several times higher than that of the average non-tweet, high-impact trade. In comparing the temporary price impacts associated with the same trades with those of the tweet-trades, we find that tweet-trade ratios are again several times higher. This suggests that tweet-trades tend to be noisier than other trades associated with a more permanent price impact, and this provides a basis for demonstrating to marketers the significance of the relatively high levels of both permanent and temporary price impacts that can be generated in financial markets with the use of tweets. A robustness comparative analysis based on [22] is consistent with the presented findings (see Web Appendix G).
To add intraday actionability, we used and , which encapsulate the permanent and temporary impacts of intraday tweets on firm value, as dependent variables to determine how tweet valence and subject matter (consumer and competitor orientation) influence the impact of tweets on stock price. To investigate whether tweet valence and subject matter drive the price impact of tweet-trades, we estimate Equation 7:
Graph
( 7)
where corresponds to or , respectively, for a tweet-trade t in stock s. and are stock and time fixed effects. We used the VADER rule-based algorithm ([40]) to determine the valence of the tweets. VADER outperforms other commonly used benchmark methodologies such as Linguistic Inquiry and Word Count, Affective Norms for English Words, and the machine learning algorithm support vector machine in the literature as well as in our robustness tests. We also utilized [66] library and followed their method for measuring the competitor ( ) and consumer orientation for each tweet, which is in line with [ 3] and [78]. Consumer and competitor subject matter are dummy variables that equal 1 when a tweet-trade's content is about consumer and/or competitors. We also studied the interaction effects of these attributes. and refer to positive-valence tweets related to competitors and negative-valence tweets related to competitors, respectively, for a tweet-trade t in stock s, and and refer to positive-valence tweets related to consumers and negative-valence tweets related to consumers, respectively, for a tweet-trade t in stock s.
To avoid omitted variable bias and to ensure completeness, the model also includes , which reflects a vector of known determinants of price impact based on past research in the market microstructure literature, as well as the natural logarithm of the number of an account's followers at the time of a tweet-trade t's tweet ( ). includes the natural logarithm of trading volume ( ), the natural logarithm of average trade size ( ), volatility ( ), effective spread ( ), the natural logarithm of a high-frequency trading proxy ( ), and order imbalance ( ). We measured trading volume as the dollar volume of transactions executed in stock s prior to a corresponding tweet-trade t. We computed average trade size as the trading volume prior to tweet-trade t divided by the number of transactions just prior to a corresponding tweet-trade t in stock s. is the standard deviation of midpoint dollar price returns from the start of the trading day up to the trade just before the corresponding tweet-trade t in stock s. (in basis points) was computed as twice the absolute value of the last trade price less the prevailing price midpoint prior to the corresponding tweet-trade t in stock s divided by the prevailing price midpoint. Price midpoint is the average of the prevailing best bid and ask prices. is the ratio of the number of messages (quotes, cancellations, and transactions) to actual transactions from the start of the trading day until prior to a corresponding tweet-trade t in stock s. Finally, is the ratio of the difference between the number of sell and buy orders and the average of both from the start of the trading day until prior to a corresponding tweet-trade t in stock s. To eliminate outliers in the data caused by the characteristic noisiness of high-frequency trading data, all variables are winsorized at.5 and 99.5 percentiles within each stock.
We estimated Equation 7 using both panel least squares and 2SLS instrumental variable (IV) estimation approaches. Panel-corrected standard errors were computed to obtain heteroskedasticity and autocorrelation robust standard errors. We performed the IV estimation to account for the likelihood of endogeneity due to selection bias caused by a firm's decision regarding whether to use Twitter ([25]). The IV approach we employed was based on approaches adopted by an increasing number of studies in the marketing literature ([80]). For a given firm in our sample of S&P 500 IT firms, our approach involved first identifying the firms in the same two-digit Standard Industrial Classification that had sent a corresponding tweet on the previous or same day as the firm and then estimating the mean value of the potentially endogenous variables (consumer and competitor orientation) for these firms. The mean estimates were employed as an instrument for the firm in question. This variable met the requirements for an instrument because price impacts observed in the other firms' stocks were unlikely to be driven by the focal firm's tweets and, at the same time, tweeting activity has been shown to be correlated for firms in similar industries. In each of the first-stage regressions, we regressed each of the consumer and competitor variables separately on the corresponding IVs and the control variables defined previously for each firm/stock and obtained the F-statistics as tests of the null of weak instruments. The fitted values for each of the measures from the first-stage regressions were then employed as the variables in place of the consumer and competitor orientation variables in the second-stage regressions.
The first-stage F-statistics, testing the null of weak instruments, show that our IV model does not suffer weak instrument issues. The test statistic is higher than the threshold of 10 needed for 2SLS inferences to be reliable when instrumenting for endogenous variables ([70]). We also conducted further tests to examine the instruments' relevance and the validity of the overidentifying restrictions in the IV regressions. The Cragg–Donald and Kleibergen–Paap Lagrange multiplier statistics we obtained reject the nulls of weak instruments and underidentification according to the [33] critical values, respectively. Essentially, these test the null hypothesis that the instruments we used have insufficient explanatory power to predict the endogenous variables in the model for identification of the parameters. All the p-values we obtained in the Sargan χ2 test also indicate that we cannot reject the null that the overidentifying restrictions are valid. All the 2SLS estimates for Equation 7 are presented in Table 6, and the results of the panel least squares estimations are presented in Web Appendix H.
Graph
Table 6. The Relationship Between Permanent and Temporary Price Impact and Tweet Valence and Orientation.
| Variables | Permanent Price Impact | Temporary Price Impact | Key Findings |
|---|
| −.007(1.06) | .009**(2.39) | Consumer-related tweets are, on average, associated with a larger temporary price impact relative to other tweets. |
| −.034**(2.03) | .026**(2.46) | Competitor-related tweets are, on average, associated with a larger temporary price impact and lower permanent price impact relative to other tweets. |
| −.063**(2.37) | .032**(2.46) | Tweets with only negative or only positive valence are associated with increasing temporary price impact and decreasing permanent price impact. |
| −.108***(3.11) | .033**(2.20) |
| .047**(1.97) | .072**(2.50) | Tweets reflecting both valence and subject matter are associated with increases in both permanent and temporary price impact. The increase in permanent price impact contrasts the decrease in permanent price impact that tweets with only valence or only subject matter are associated with.Except for tweets reflecting negative valence and consumer orientation , permanent price impact is more pronounced than temporary price impact. |
| .606***(3.69) | .011**(2.09) |
| .088**(2.10) | .019**(2.21) |
| .220***(4.88) | .077**(2.43) |
| −.039***(-4.51) | −.034***(-6.68) | Increases in firm stock trading activity are linked with reductions in both permanent and temporary price impacts. |
| .089***(6.64) | .061***(7.25) | Larger firm stock trade sizes induce larger permanent and temporary price impacts. |
| −.123***(−3.62) | .015**(2.13) | Firm stock volatility is linked with increases in temporary price impact and decreases in permanent price impact. |
| .241**(2.66) | .014**(2.06) | Deterioration in firm stock liquidity is associated with increases in permanent and temporary price impacts. |
| −.000(−.26) | −.021***(−3.83) | Algorithmic and high-frequency trading is linked with decreases in temporary price impact. Its effect on permanent price impact is benign. |
| −.371***(−6.89) | .046**(2.39) | Order imbalance is linked with reductions in permanent price impact and increases in temporary price impact. |
| −.082**(−2.43) | .037**(2.43) | The number of followers of a firm's twitter account amplifies the propensity for tweets to generate larger temporary price impact and reduce permanent price impact. |
| .35 | .49 | |
| Observations | 139,997 | 139,997 | |
| Kleibergen–Paap LM | 31.32*** | 110.24*** | Tests the null hypothesis that the employed instruments have insufficient explanatory power to predict the endogenous variables in the model for identification of the parameter. |
| Cragg–Donald | 79.08*** | 88.66*** | Tests the same null hypothesis as the Kleibergen–Paap LM test. |
| Sargan's χ2p-value | .37 | .46 | Tests the null hypothesis that the overidentifying restrictions are valid. |
- 8 **p <.05.
- 9 ***p <.01.
- 10 Notes: Table 6 reports 2SLS estimated coefficients for Equation 7. Standard errors are robust to heteroscedasticity and autocorrelation, coefficients are multiplied by 107, and t-statistics are reported in parentheses. LM = Lagrange multiplier.
The results presented in Table 6 show the importance of tweet valence and subject matter in determining the permanent and temporary impacts of tweets on firm value. The existence of permanent and temporary price impacts associated with tweet attributes supports the signal theory perspective ([43]) and shows that investors pay attention to the tweet attributes of valence and subject matter. The estimates of permanent price impact for and are negative and statistically significant (−.063, p < .05, and −.108, p < .01, respectively). This suggests that tweets displaying only positive or only negative valence are linked to less permanent impacts in stock price. The positive and statistically significant and estimates of the temporary price impact estimation also indicate that such tweets are linked to increasing temporary price impact (.032, p < .05, and <.033, p < .05, respectively), and they suggest that tweet valence generally contributes more noise to stock price than stock-relevant information. The findings reinforce the role of positive and negative valence FGCs and their impact on firm value ([75]; [76]).
With respect to tweet subject matter, only tweets conveying information about competitors generate statistically significant permanent price impacts (−.034, p < .05). Therefore, on average, tweets about a firm's competitors generate lower permanent price impact relative to other tweets. Conversely, the positive and statistically significant estimates for for temporary price impact (.026, p < .05) show that these types of tweets are more likely to contribute to the noise component of price; in other words, they generate a larger temporary price impact than other tweets, on average. Thus, tweets conveying competitor orientation appear to result in a lower permanent price impact, suggesting that this form of subject matter is comparatively less impactful and relevant to investors' expectations about a firm's future performance ([48]). Notably, tweets about consumers do not yield any permanent price impact that is statistically different from that of other tweets and, thus, by themselves do not appear to offer a signal capable of causing investors to permanently update their firm performance expectations. Consumer-related tweets, similar to those about competitors, also generate more temporary price impact than other tweets on average, which suggests that their potential for inducing noise in stock price is higher than that of the average tweet in our sample. The and estimates of the temporary price impact are positive and statistically significant (.009, p < .05, and.026, p < .05, respectively). This finding implies that, as is the case with valence, tweets reflecting only competitor or only consumer orientation generate noise in the price discovery or trading processes and lower permanent price impact.
Inferring from information-based market microstructure models ([24]; [47]), the more information about a firm investors observe, the more they become informed about the valuation of the firm. In line with this expectation, the interaction variables we include in Equation 7 should yield positive estimates for the estimations. As we expected, all the , , , and estimates of permanent price impact are positive and statistically significant (respectively,.047, p < .05;.606, p < .01;.088, p < .05; and.220, p < .01), even though, as already stated, , , , and are negative and statistically significant (except for ). Thus, increases in both negative and positive valence, when viewed through the lens of subject matter, are linked with increased permanent price impact. These estimates show that tweet valence, when contextualized by subject matter or vice versa, is seen by investors/traders as firm-relevant information. In the context of these findings, the incorporation of valence and subject matter into FGC can yield increases in permanent price impact.
Furthermore, the findings suggest that tweets about competitors with a negative valence are likely to have the highest permanent price impacts (.606, p < .001). This finding is crucial from the perspective of marketing practice and intraday social media marketing strategy design because valence and competitor subject matter as singular attributes of FGC are independently associated with decreasing permanent price impact. The findings underscore the view that investors seek additional information (i.e., information beyond what they already have) when making trading decisions ([ 6]; [73]) and that they operate according to classical market microstructure models. For example, [47] and [24] emphasize the crucial importance of information for price discovery in financial markets. This also confirms [51] findings that information from microblogging platforms, such as Twitter, impact investors' decisions.
To illustrate the relevance of these findings, Figure 3 presents tweets A and B as examples of FGC characterized by negative and positive valence, respectively, but not containing any subject matter related to competitor or consumer orientation. Consistent with our findings, the permanent price impact estimates for the tweet trades corresponding to both tweets are more than three standard deviations lower than the average permanent price impact estimate and are thus below the tenth percentile of the estimates. The estimates for the negative and positive tweets' tweet-trades are.0017% and.0035%, respectively. In contrast with A and B, tweets C, D, and E reflect varying combinations of both valence and subject matter. Our findings suggest that these tweets should generate significant permanent price impact, and indeed, the permanent price impact estimates for the tweet-trades corresponding to tweets C, D, and E are above the 90th percentile in our sample of tweet-trades' permanent price impact estimates. The estimates are 3.74%, 2.84%, and 1.32% for tweets C, D, and E, respectively.
Graph: Figure 3. Examples of tweets generating temporary and permanent price impacts.
The effects of the tweet attributes we studied on temporary price impact, , also deserve attention. The results suggest that the relationship between valence and temporary price impact is generally magnified when combined with subject matter. For example, the and estimates, which capture the relationship between on the one hand and and on the other, are positive and statistically significant (.032, p < .05, and <.033, p < .05, respectively), while the estimates for and , which capture the relationship between on the one hand and and on the other, are also positive and statistically significant (.072, p < .05, and <.077, p < .05, respectively). The latter set of estimates is at least two times larger than the former. The overall implication of these positive and statistically significant coefficient estimates related to temporary price impacts is that, although tweets reflecting both valence and subject matter are likely to generate permanent price impact, these attributes may also be associated with increased temporary price impact. Thus, on average, tweets inject noise (uncertainty) into the prices of stocks traded in financial markets.
In conclusion, the estimates presented in Table 6 highlight the relevance of tweet attributes for the price discovery process in financial markets and reinforce the importance of studying the multifaceted nature of FGC ([45]). We find that tweets, as with many events observed in relation to trading in financial markets, generate both permanent and temporary price impacts. However, whereas tweets containing singular attributes—either positive or negative valence or either consumer or competitor orientation—readily inject noise into the price discovery process and thus generate temporary price impact, those that include more than one attribute generate permanent price impact and thus generally enhance the efficiency of the price discovery process.
In this research, we examine the real-time impact of FGC on the variance of firms' stock price. In the current fast-paced online communication landscape, marketers must understand the financial impact of firms' "always on" marketing ([64]). The assessment of FGC's financial impacts, however, is in an early stage ([ 8]; [15]; [64]). This research contributes to this emerging stream of marketing research and addresses multiple calls for new methods that are able to develop real-time insights from online data ([ 5]; [49]; [59]; [79]). By employing the market microstructure approach to study S&P 500 IT firms' Twitter activity, this study contributes to marketing literature and practice.
This study offers several implications for marketing research. First, aligning with the work by [15], [ 8], and [64], it advances understanding of FGC's financial impact by providing an assessment of FGC's impact on the variance of stock price in real time (i.e., seconds). By employing a market microstructure approach, we show how to algorithmically link individual pieces of FGC to time-specific trading activity at a fine-grained level of analysis. In the process, we demonstrate the limitations of low-frequency methodologies such as daily event studies, which are subject to aggregation bias and may yield biased estimates of the impact of FGC on firms' financial outcomes, while offering an alternative and more robust method of analysis for studying intraday marketing activity. In our examination of the impact of FGC on variance, we fully utilized high-frequency transaction data characterized by unequal time intervals and demonstrate how to retain data that otherwise would have been eliminated in studies that use end-of-day stock price. By doing so, we provide marketing researchers with a new approach that allows them to harness the potential of online data.
Second, we distinguish between FGC's temporary and permanent price impacts. Specifically, we show that FGC impacts investor expectations related to a firm's future performance, thus generating permanent price impact, and it also injects uncertainty about a firm's value into the firm's stock price, thus inducing temporary price impact. Our research, therefore, adds a new perspective to the marketing literature stream on the financial impact of FGC. This assessment of FGC's temporary and permanent price impacts adds richness to the examination of marketing's financial impact and enables the quantifying of long- and short-term financial impacts of marketing activity.
Finally, this research has implications for the design of intraday marketing strategies. By examining FGC valence and subject matter (consumer and competitor orientation), we advance a growing body of research documenting the complex nature through which marketing signals impact financial markets and firm financial outcomes. We show that, by themselves, FGC valence and subject matter are more prone to injecting uncertainty about a firm's stock price into the market, and thus, they generate temporary price impacts rather than permanently changing investors' and traders' beliefs about firm value. Used together, FGC valence and subject matter both hold statistically significant and economically meaningful relevance for price discovery in financial markets. In other words, they can influence investors' expectations related to firms' future performance and thus result in permanent price impacts. Recent research by [ 6] provides evidence of interactions between marketing signals, and our research shows that the interaction between FGC valence and subject matter can also impact firm stock price.
Thus far, firms have struggled to demonstrate financial accountability regarding FGC's impact on firm value ([15]; [45]) or provide evidence of its immediate contribution to their financial outcomes ([53]; [58]). We provide marketing managers with evidence of FGC's impact on variance in firms' stock price. Specifically, we show that tweets can generate both permanent and temporary price impacts. By manipulating tweet attributes, such as valence and subject matter, marketing managers can design Twitter content to generate varying degrees of permanent or temporary impact. From a market quality perspective, firm managers should prefer tweets that generate a permanent price impact, and our research provides some useful indications about how to achieve this outcome. We show that tweets expressing degrees of positive or negative valence regarding either consumers or competitors generate a permanent price impact. We therefore encourage marketing managers to design information-rich tweets that both ( 1) focus on consumers or competitors and ( 2) communicate valence. Our results suggest that firms should utilize valence and subject matter in their tweets if they would like their stock to be more informative with respect to their value. Our analysis suggests that tweets about competitors with a negative valence are likely to have the highest permanent price impacts. Thus, by using permanent price impact as a metric to evaluate the longer-term impact of tweets, social media managers can design campaigns that have a sustainable impact on firm financial outcomes. The design recommendations from this study complement [41] work on social media content scheduling, as well as [64], which addresses real-time social media marketing and provides firms with information regarding which tweets to disseminate during the day for long-term effectiveness. We recognize that not all intraday tweets will, nor should they, have permanent impacts on firms' stock price. Some tweets are aimed at the creation of social media buzz, which is related to the temporary price impacts we examined in this study. Firms can indeed achieve social media buzz by tweeting, as our findings reveal that tweets, in aggregate, mostly generate temporary price impacts. We urge caution, however, because temporary price impacts are linked with larger transaction costs ([14]) and increases in firm cost of capital ([17]). This suggests that the benefits of designing tweets to generate buzz and incorporate information into stock price must be carefully managed. To support marketing managers in their intraday social media strategy design, Table 7 is designed as a set of insights based on our findings.
Graph
Table 7. Suggested Insights for Marketing Managers.
| | Permanent Price Impact | Temporary Price Impact | Research Findings | Recommendation | Expected Outcome | Interaction Effects | Permanent Price Impact | Temporary Price Impact | Research Findings |
|---|
| Valence | Positive valence | | | Positive and negative valence-only FGC contributes to more noise in a firm's stock price. | Add subject matter (e.g., competitor orientation such as "competition," "peer") | Increased permanent price impact | Positive valence and competitor orientation | | | Interaction of valence and subject matter increases/generates permanent price impact and amplifies temporary price impact. Permanent price impact is more pronounced than temporary price impact. The financial implication of these outcomes is a reduction in transaction and firm capital costs. |
| Negative valence | | | Add subject matter (e.g., consumer orientation such as "customer," "consumer," "buyer") | Increased permanent price impact | Negative valence and consumer orientation | | |
| Subject Matter | Consumer orientation | | | Subject matter-only FGC contributes to the noise component in a firm's stock price. | Add valence (e.g., positive valence such as "help," "solution," "best") | Increased permanent price impact | Positive valence and consumer orientation | | |
| Competitor orientation | | | Add valence (e.g., negative valence such as "attack," "stop," "threat") | Increased permanent price impact | Negative valence and competitor orientation | | |
11 = no price impact; = negative impact on stock price component; = positive impact on stock price component; = increased positive impact on stock price component.
We conclude by encouraging future research to address the limitations of our empirical study. One potential limitation of our analysis is its focus on firms in the IT sector. We recognize that these findings may not apply to other sectors. Future research could extend our analysis to other sectors to confirm whether similar price impacts hold. Second, the impact of tweets could depend on whether Twitter was the first source through which a firm released an important piece of news. For example, tweets could have been published in response to a competitor's tweet. In some cases, a firm's tweet could lead to a number of successive tweets, in which case the subsequent tweets might not be as impactful as the first. We do not discount the possibility that there could be some carryover or dampening effect in such situations. We note that, if this is the case, it would be highly unlikely for the magnitude of the effects we observe to occur, especially given the granular level of analysis that our market microstructure approach entails. Third, future work could explore high-frequency data generated by firms' use of FGC other than tweets, such as Facebook posts, where it has been reported firms post up to 80 times a day (Hutchinson 2018). It would be interesting to see whether the effect of FGC across social media platforms is consistent or if it varies. In addition to social media, it would be useful to examine firms' use of other online communication tools, such as webpages and blogging platforms. Researchers could also explore various types of FGC, including video content, as well as its characteristics, including emotions ([72]). As [36] show, there is an array of online marketing communication practices, and future research could therefore study the "echoverse" at a fine-grained level of analysis. Finally, we note that researchers can apply market microstructure to study user-generated content (UGC) in future research. We welcome future research that addresses the following questions: What is the real-time impact of UGC on firm value? What are the UGC attributes capable of generating permanent and temporary price impacts? Are these attributes the same as for FGC, or do they differ? Our research highlights the importance of interaction effects when examining the impact of FGC attributes on firm value; therefore, investigating the optimal mix of UGC attributes capable of generating temporary and permanent price impacts should be an interesting endeavor.
sj-pdf-1-jmx-10.1177_00222429211042848 - Supplemental material for Measuring the Real-Time Stock Market Impact of Firm-Generated Content
Supplemental material, sj-pdf-1-jmx-10.1177_00222429211042848 for Measuring the Real-Time Stock Market Impact of Firm-Generated Content by Ewelina Lacka, D. Eric Boyd, Gbenga Ibikunle and P.K. Kannan in Journal of Marketing
Footnotes 1 Hari Sridhar
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship and/or publication of this article.
4 P.K. Kannan https://orcid.org/0000-0003-0738-0766
5 The two examples in Figure 1 illustrate FGC attributes. First, the valence of FGC varies from positive (e.g., "we were so happy to be a part of it." [CA Technologies]) to negative (e.g., "have you lost trust in tech?" [CA Technologies]) throughout the day. Second, the subject matter of FGC also varies: Adobe's subject matter ranges from focusing on the consumer (e.g., "we want to know what inspires you") to focusing more on competitive positioning (e.g., "as of today, all fonts included with Creative Cloud can be used on iOS13.1.").
6 We excluded a total of 1,179 tweets,.77% of the total sample (see Web Appendix F), from the analysis because of excessive stock price return volatility on the days they occurred. This is a standard approach commonly employed in market microstructure literature. We also defined an exclusion criterion to exclude tweets that occurred within 60 seconds of each other. However, because none of the tweets in our data occurred within 60 seconds of each other, no tweet was excluded on the basis of the exclusion criteria.
7 There were three types of observations in our data set. The first were the buy/bid and sell/offer quotes (or orders), and the second were the transactions or trades, which featured directly in the state-space modeling and were generated as a result of the orders being executed. For the model, we only employed the prices of the 520,356,393 transactions in the data set. The third type of observations were cancellations issued to cancel previously submitted orders. All the observations were captured using timestamps to the nearest millisecond.
8 Although v (stock price in our model) was observable, its permanent and transitory components, which we aimed to characterize, were unobservable; that is, we could not acquire them as we would observable variables such as stock price or volume. Thus, we aimed to observe the evolution of that one variable—v—that we could observe and use this evolution within time intervals (i.e., one second and one minute) to estimate its components.
9 In addition to modeling the natural logarithm of price as an observable variable in the state-space representation, for robustness, we also employed percentage change in price and first difference of price. Our inferences were the same regardless of the approach we employed, and indeed, all the estimates obtained were qualitatively similar.
According to Merton's (1987) model, when investors hold underdiversified portfolios, idiosyncratic risk should be priced. in Equation 2 captures the effect of idiosyncratic risk as a function of information, and it is different from non-information-based (temporary) evolution in stock price captured by .
Some adjustments are required. When there are instances of missing (or irregularly spaced) observations in , the Kalman filter is unable to use the measurement equation (Equation 1); however, the transition equation (Equation 2) can be used because it depends on the previously estimated state ( depends on ). Indeed, Kalman filtering suggests that with missing observations in , the best estimation for is simply the evaluation of the transition equation. The estimated state-space model's source code in SAS is presented in [67].
The code we estimated is available via a public GitHub repository here: https://github.com/akataehonda/Twitter-Project.git.
Note that the percentage change is from a trade at t − 1 before the tweet-trade to 30 seconds after the tweet-trade. Varying this measurement for up to five trades at t − 5 before the tweet-trade did not significantly impact the estimates obtained, and neither did varying the time threshold to include 45- and 60-second percentage changes. Estimating the effects of tweet-trades within subminute to minute windows avoided the methodological issues associated with the occurrence of confounding events. Given the fine-grained level of measurement, it is highly unlikely that any other relevant event could have been driving the effects we captured.
To ensure that the results were not driven by unusual trading days, we excluded days for which stock return volatility was greater than one standard deviation of the average stock return volatility over the surrounding (−30, +30) trading days. We measured daily volatility as the standard deviation of intraday stock return.
References Amemiya Takeshi , Wu Roland Y.. (1972), " The Effect of Aggregation on Prediction in the Autoregressive Model ," Journal of the American Statistical Association , 67 (339), 628 – 32.
Amihud Yakov. (2002), " Illiquidity and Stock Returns: Cross-Section and Time-Series Effects ," Journal of Financial Markets , 5 (1), 31 – 56.
Atuahene-Gima Kwaku. (2005), " Resolving the Capability–Rigidity Paradox in New Product Innovation ," Journal of Marketing , 69 (4), 61 – 83.
Barclay Michael J. , Warner Jerold B.. (1993), " Stealth Trading and Volatility: Which Trades Move Prices? " Journal of Financial Economics , 34 (3), 281 – 305.
Berger Jonah , Humphreys Ashlee , Ludwig Stephan , Moe Wendy W. , Netzer Oded , Schweidel David A.. (2020), " Uniting the Tribes: Using Text for Marketing Insights ," Journal of Marketing , 84 (1), 1 – 25.
Bhagwat Yashoda , Warren Nooshin L. , Beck Joshua T. , Watson George F. IV. (2020), " Corporate Sociopolitical Activism and Firm Value ," Journal of Marketing , 84 (5), 1 – 21.
Blankespoor Elizabeth , Miller Gregory S. , White Hal D. , (2014), " The Role of Dissemination in Market Liquidity: Evidence from Firms' Use of Twitter ," The Accounting Review , 89 (1), 79 – 112.
Borah Abhishek , Banerjee Sourindra , Lin Yu-Ting , Jain Apurv , Eisingerich Andreas B.. (2020), " Improvised Marketing Interventions in Social Media ," Journal of Marketing , 84 (2), 69 – 91.
Boyd D. Eric , Kannan P.K. , (2018), " (When) Does Third-Party Recognition for Design Excellence Affect Financial Performance in Business-to-Business Markets? " Journal of Marketing , 82 (3), 108 – 23.
Breitung Jorg , Swanson Norman S.. (2002), " Temporal Aggregation and Spurious Instantaneous Causality in Multiple Time Series Models ," Journal of Time Series , 23 (6), 651 – 65.
Brogaard Jonathan , Hendershott Terrence , Riordan Ryan. (2014), " High-Frequency Trading and Price Discovery ," Review of Financial Studies , 27 (8) , 2267 – 306.
Budish Eric , Cramton Peter , Shim John. (2015), " The High-Frequency Trading Arms Race: Frequent Batch Auctions as a Market Design Response ," Quarterly Journal of Economics , 130 (4), 1547 – 621.
Chan Louis K.C. , Lakonishok Josef. (1993), " Institutional Trades and Intraday Stock Price Behavior ," Journal of Financial Economics , 33 (2), 173 – 99.
Chan Hing Kai , Wang Xiaojun , Lacka Ewelina , Zhang Min. (2016), " A Mixed-Method Approach to Extracting the Value of Social Media Data ," Production and Operations Management , 25 (3), 568 – 83.
Chordia Tarun , Roll Richard , Subrahmanyam Avanidhar. (2001), " Market Liquidity and Trading Activity ," Journal of Finance , 56 (2), 501 – 30.
Chordia Torun , Roll Richard , Subrahmanyam Avanidhar. (2008), " Liquidity and Market Efficiency ," Journal of Financial Economics , 87 (2), 249 – 68.
Colicev Anatoli , Kumar Ashish , O'Connor Peter. (2019), " Modeling the Relationship Between Firm and User Generated Content and the Stages of Marketing Funnel ," International Journal of Research in Marketing , 36 (1), 100 – 116.
Colicev Anatoli , Malshe Ashwin , Pauwels Koen , O'Connor Peter. (2018), " Improving Consumer Mindset Metrics and Shareholder Value Through Social Media: The Different Roles of Owned and Earned Media ," Journal of Marketing , 82 (1) , 37 – 56.
Diamond Douglas W. , Verrecchia Robert E.. (1991), " Disclosure, Liquidity, and the Cost of Capital ," Journal of Finance , 46 (4), 1325 – 59.
Du Rex Yuxing , Netzer Oded , Schweidel David A. , Mitra Debanjan. (2021), " Capturing Marketing Information to Fuel Growth ," Journal of Marketing , 85 (1), 163 – 83.
Durbin James , Koopman Siem Jan. (2012), Time Series Analysis by State Space Models. Oxford, UK : Oxford University Press.
Elliott W. Brooke , Grant Stephanie M. , Hodge Frank D.. (2018), " Negative News and Investor Trust: The Role of $Firm and #CEO Twitter Use ," Journal of Accounting Research , 56 (5), 1483 – 519.
Fama Eugene F. (1970), " Efficient Capital Markets: A Review of Theory and Empirical Work ," Journal of Finance , 25 (2), 383 – 417.
Frino Alex , Jarnecic Elvis , Lepone Andrew. (2007), " The Determinants of the Price Impact of Block Trades: Further Evidence ," Abacus , 43 (1), 94 – 106.
Gebhardt Gary F. , Carpenter Gregory S. , Sherry John F.. (2006), " Creating a Market Orientation: A Longitudinal, Multifirm, Grounded Analysis of Cultural Transformation ," Journal of Marketing , 70 (4) , 37 – 55.
Glosten Lawrence R. , Milgrom Paul R.. (1985), " Bid, Ask, and Transaction Prices in a Specialist Market with Heterogeneously Informed Traders ," Journal of Financial Economics , 14 (1), 71 – 100.
Gong Shiyang , Zhang Juanjuan , Zhao Ping , Jiang Xuping. (2017), " Tweeting as a Marketing Tool: A Field Experiment in the TV Industry ," Journal of Marketing Research , 54 (6), 833 – 50.
Gordon Brett R. , Jerat Kinshuk , Katona Zsolt , Narayanan Sridhar , Shin Juwoong , Wilbur Kenneth. (2021), " Inefficiencies in Digital Advertising Markets ," Journal of Marketing , 85 (1), 7 – 25.
Groß-Klußmann Alex , König Stephan , Ebner Markus. (2019), " Buzzwords Build Momentum: Global Financial Twitter Sentiment and the Aggregate Stock Market ," Expert Systems with Applications , 136 , 171 – 86.
Grossman Sanford J. , Miller Merton H.. (1988), " Liquidity and Market Structure ," Journal of Finance , 43 (3), 617 – 33.
Grossman Sanford J. , Stiglitz Joseph E.. (1980), " On the Impossibility of Informationally Efficient Markets ," American Economic Review , 70 (3), 393 – 408.
Hill, Joseph (2020), " How Tech Dominates the US Stock Market ," Hargreaves Lansdown (October 6), https://www.hl.co.uk/news/articles/how-tech-dominates-the-us-stock-market.
Hasbrouck Joel. (1991), " Measuring the Information Content of Stock Trades ," Journal of Finance , 46 (1), 179 – 207.
Hasbrouck Joel , Saar Gideon. (2013), " Low-Latency Trading ," Journal of Financial Markets , 16 (4), 646 – 79.
Hausman Jerry , Stock James H. , Yogo Motohiro. (2005), " Asymptotic Properties of the Hahn-Hausman Test for Weak-Instruments ," Economics Letters , 89 (3), 333 – 42.
Hedge Sjantaram P. , McDermott John B.. (2003), " The Liquidity Effects of Revisions to the S&P 500 Index: An Empirical Analysis ," Journal of Financial Markets , 6 (3), 413 – 59.
Hendershott Terrence , M. Jones Charles , Menkveld Albert J.. (2011), " Does Algorithmic Trading Improve Liquidity? " Journal of Finance , 66 (1), 1 – 33.
Hendershott Terrence , Menkveld Albert J.. (2014), " Price Pressures ," Journal of Financial Economics , 114 (3), 405 – 23.
Henry-Nickie, Makada, Kwadwo Frimpong, and Hao Sun (2019), " Trends in the Information Technology Sector ," Brookings (March 29), www.brookings.edu/research/trends-in-the-information-technology-sector/.
Hewett Kelly , Rand William , Rust Roland T. , van Heerde Harald J.. (2016), " Brand Buzz in the Echoverse ," Journal of Marketing , 80 (3) , 1 – 24.
Hoffman Donna L. , Fodor Marek. (2010), " Can You Measure the ROI of Your Social Media Marketing? " MIT Sloan Management Review , 52 (1), 41 –4 9.
Holthausen Robert W. , Leftwich Richard W. , Mayers David. (1990), " Large-Block Transactions, the Speed of Response, and Temporary and Permanent Stock-Price Effects ," Journal of Financial Economics , 26 (1), 71 – 95.
Homburg Christian , Theel Marcus , Hohenberg Sebastian. (2020), " Marketing Excellence: Nature, Measurement, and Investor Valuations ," Journal of Marketing , 84 (4), 1 – 22.
Hutchinson Andrew. (2018), " New Report Suggests Optimal Facebook Posting Frequency Following Algorithm Shift ," Social Media Today (August 21), https://www.socialmediatoday.com/news/new-report-suggests-optimal-facebook-posting-frequency-following-algorithm/530514/.
Hutto C.J. , Gilbert Eric. (2014), " VADER: A Parsimonious Rule-Based Model for Sentiment Analysis of Social Media Text ," paper presented at Eighth International Conference on Weblogs and Social Media (ICWSM-14) , Ann Arbor, MI (June 2–4).
Ibikunle Gbenga. (2015), " Opening and Closing Price Efficiency: Do Financial Markets Need the Call Auction? " Journal of International Financial Markets, Institutions, and Money , 34 , 208 – 27.
Kanuri Vamsi K. , Chen Yixing , Sridhar Shrihari. (2018), " Scheduling Content on Social Media: Theory, Evidence, and Application ," Journal of Marketing , 82 (6), 89 – 108.
Kirilenko Andrei , Kyle Albert S. , Samadi Mehradad , Tuzun Tugkan. (2017), " The Flash Crash: High-Frequency Trading in an Electronic Market ," Journal of Finance , 72 (3), 967 – 98.
Kirmani Amna , Rao Akshay R.. (2000), " No Pain, No Gain: A Critical Review of the Literature on Signaling Unobservable Product Quality ," Journal of Marketing , 64 (2) , 66 – 79.
Koopman Siem Jan , Durbin James. (2003), " Filtering and Smoothing of State Vector for Diffuse State-Space Models ," Journal of Time Series Analysis , 24 (1), 85 – 98.
Kumar Ashish , Bexawada Ram , Rishika Rishika , Janakiraman Ramakumar , Kannan P.K.. (2016), " From Social to Sale: The Effects of Firm-Generated Content in Social Media on Consumer Behavior ," Journal of Marketing , 80 (1) , 7 – 25.
Kumar V. , Jones Eli , Venkatesan Rajkumar , Leone Robert P.. (2011), " Is Market Orientation a Source of Sustainable Competitive Advantage or Simply the Cost of Competing? " Journal of Marketing , 75 (1), 16 – 30.
Kyle Albert S.. (1985), " Continuous Auctions and Insider Trading ," Econometrica , 53 (6), 1315 – 35.
Lam Son K. , Kraus Florian , Ahearne Michael. (2010), " The Diffusion of Market Orientation Throughout the Organization: A Social Learning Theory Perspective ," Journal of Marketing , 74 (5), 61 – 79.
Lamberton Cait , Stephen Andrew T.. (2016), " A Thematic Exploration of Digital, Social Media, and Mobile Marketing: Research Evolution From 2000 to 2015 and an Agenda for Future Inquiry ," Journal of Marketing , 80 (6) , 146 – 72.
Lambrecht Anja , Tucker Catherine , Wiertz Caroline. (2018), " Advertising to Early Trend Propagators: Evidence From Twitter ," Marketing Science , 37 (2), 177 – 99.
Li Ting , van Dalen Jan , van Rees Pieter Jan. (2018), " More Than Just Noise? Examining the Information Content of Stock Microblogs on Financial Markets ," Journal of Information Technology , 33 (1), 50 – 69.
Madhavan Ananth , Richardson Matthew , Roomans Mark. (1997), " Why Do Security Prices Change? A Transaction-Level Analysis of NYSE Stocks ," Review of Financial Studies , 10 (4), 1035 – 64.
Magill Paul , Moorman Christine , Avdiushko Nikita. (2019), " 8 Ways Marketers Can Show Their Work's Financial Results ," Harvard Business Review (July 31), https://hbr.org/2019/07/8-ways-marketers-can-show-their-works-financial-results.
Marcellino Marcellino. (1999), " Some Consequences of Temporal Aggregation in Empirical Analysis ," Journal of Business and Economic Statistics , 17 (1), 129 – 36.
Marketing Science Institute. (2018), " 2018–2020 Research Priorities: Marketers' Strategic Imperatives ," (May 13), https://www.msi.org/articles/marketers-top-challenges-2018-2020-research-priorities/
Marketing Science Institute (2020), "Research Priorities 2020 –2022," (accessed October 26, 2021), https://www.msi.org/wp-content/uploads/2020/06/MSI%5fRP20-22.pdf.
Meire Matthijs , Hewett Kelly , Ballings Michel , Kumar V. , van den Poel Dirk. (2019), " The Role of Marketer-Generated Content in Customer Engagement Marketing ," Journal of Marketing , 83 (6), 21 – 42.
Menkveld Albert J. (2013), " High Frequency Trading and the New Market Makers ," Journal of Financial Markets , 16 , 712 – 41.
Menkveld Albert J. , Koopman Siem Jan , Lucas Andre. (2007), " Modeling Around-the-Clock Price Discovery for Cross-Listed Stocks Using State Space Methods ," Journal of Business and Economic Statistics , 25 (2), 213 – 25.
Merton Robert C.. (1987), " A Simple Model of Capital Market Equilibrium with Incomplete Information ," Journal of Finance , 42 (3), 483 – 510.
Moorman Christine , Kirby Lauren. (2019), " How Marketers Can Overcome Short-Termism ," Harvard Business Review (November 21), https://hbr.org/2019/11/how-marketers-can-overcome-short-termism.
Moorman Christine , van Heerde Harald J. , Page Moreau C. , Palmatier Robert W.. (2019), " Challenging the Boundaries of Marketing ," Journal of Marketing , 83 (5), 1 – 4.
O'Hara Maureen. (2003), " Presidential Address: Liquidity and Price Discovery ," Journal of Finance , 58 (4), 1335 – 54.
Pavia-Miralles Jose Manuel. (2010), " A Survey of Methods to Interpolate, Distribute, and Extrapolate Time Series ," Journal of Service Science and Management , 3 (4) , 449 – 63.
Qian Hang. (2013), " Vector Autoregression with Mixed Frequency Data ," working paper, Munich Personal RePEc Archive.
Rao Ram C. (1986), " Estimating Continuous Time Advertising-Sales Models ," Marketing Science , 5 (2), 125 – 42.
Rust Roland T. , Rand William , Huang Ming-Hui , Stephen Andrew T. , Brook Gillian , Chabuk Timur. (2021), " Real-Time Reputation Tracking Using Social Media ," Journal of Marketing , 85 (4), 21 – 43.
Rzayev Khaladdin , Ibikunle Gbenga. (2019), " A State-Space Modeling of the Information Content of Trading Volume ," Journal of Financial Markets , 46 (100507), 1 – 19.
Saboo Alok R. , Grewal Rajdeep. (2013), " Stock Market Reactions to Customer and Competitor Orientations: The Case of Initial Public Offerings ," Marketing Science , 32 (1), 70 – 88.
Selukar Rajesh. (2011), " State Space Modeling Using SAS ," Journal of Statistical Software , 41 (12), 1 – 13
Silvestrini Andrea , Veredas David. (2008), " Temporal Aggregation of Univariate and Multivariate Time Series Models: A Survey ," Journal of Economic Surveys , 22 (3), 458 – 97.
Smith, Kit (2020), " 60 Incredible and Interesting Twitter Stats and Statistics ," Brandwatch (January 2), www.brandwatch.com/blog/twitter-stats-and-statistics/.
Statista (2020), " Social Media Spend ," (accessed October 25, 2020), www.statista.com/statistics/736971/social-media-ad-spend-usa.
Stock James H. , Wright Jonathan H. , Yogo Motohiro. (2002), " A Survey of Weak Instruments and Weak Identification in Generalized Method of Moments ," Journal of Business and Economic Statistics , 20 (4), 518 – 29.
Sul Hong Kee , Dennis Alan R. , Yuan Lingyao. (2017), " Trading on Twitter: Using Social Media Sentiment to Predict Stock Returns ," Decision Sciences , 48 (3), 454–88.
Tellis Gerard J. , MacInnis Deborah J. , Tirunillai Seshadri , Zhang Yanwei. (2019), " What Drives Virality (Sharing) of Online Digital Content? The Critical Role of Information, Emotion, and Brand Prominence ," Journal of Marketing , 83 (4), 1 – 20.
Tellis Gerard J. , Wernerfelt Birger. (1987), " Competitive Price and Quality Under Asymmetric Information ," Marketing Science , 6 (3) , 240 – 53.
Tiao George C. , Wei William S.. (1976), " Effect of Temporal Aggregation on the Dynamic Relationship of Two Time Series Variables ," Biometrika , 63 (3), 513 – 23.
Tirunillai Seshadri , Tellis Gerard J.. (2012), " Does Chatter Really Matter? Dynamics of User-Generated Content and Stock Performance ," Marketing Science , 31 (2), 198 – 215.
Van Heerde Harald , Gijsbrechts Els , Pauwels Koen. (2015), " Fanning the Flames? How Media Coverage of a Price war Affects Retailers, Consumers, and Investors ," Journal of Marketing Research , 52 (5) , 674 – 93.
Vermeer Susan A.M. , Araujo Theo , Bernritter Stefan F. , van Noort Guda. (2019), " Seeing the Wood for the Trees: How Machine Learning Can Help Firms in Identifying Relevant Electronic Word-of-Mouth in Social Media ," International Journal of Research in Marketing , 36 (3), 492 – 508.
Voss Glenn B. , Voss Zannie G.. (2000), " Strategic Orientation and Firm Performance in an Artistic Environment ," Journal of Marketing , 64 (1), 67 – 83.
Wedel Michel , Kannan P.K.. (2016), " Marketing Analytics for Data-Rich Environments ," Journal of Marketing , 80 (6) , 97 – 121.
Whitler Kimberly A. , Krause Ryan , Lehmann Donald R.. (2018), " When and How Board Members with Marketing Experience Facilitate Firm Growth ," Journal of Marketing , 82 (5), 86 – 105.
~~~~~~~~
By Ewelina Lacka; D. Eric Boyd; Gbenga Ibikunle and P.K. Kannan
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 92- Mickey D's Has More Street Cred Than McDonald's: Consumer Brand Nickname Use Signals Information Authenticity. By: Zhang, Zhe; Patrick, Vanessa M. Journal of Marketing. Sep2021, Vol. 85 Issue 5, p58-73. 16p. 1 Diagram, 1 Chart, 2 Graphs. DOI: 10.1177/0022242921996277.
- Database:
- Business Source Complete
Mickey D's Has More Street Cred Than McDonald's: Consumer Brand Nickname Use Signals Information Authenticity
Consumers often observe how other consumers interact with brands to inform their own brand judgments. This research demonstrates that brand relationship quality–indicating cues, such as brand nicknames (e.g., "Mickey D's" for McDonald's, "Wally World" for Walmart), enhance perceived information authenticity in online communication. An analysis of historical Twitter data followed by six experiments (using both real and fictitious brands across different online platforms [e.g., online reviews, social media posts]) show that brand nickname use in user-generated content signals a writer's relationship quality with the target brand from the reader's perspective, which the authors term "inferred brand attachment." The authors demonstrate that inferred brand attachment boosts perceived information authenticity and leads to positive downstream consequences, such as purchase willingness and information sharing. The authors also find that this effect is attenuated when brand nicknames are used in firm-generated content. How consumers' relationships with brands are portrayed and perceived in a social context (e.g., via brand nickname use) serves as a novel context to examine user-generated content and provides valuable managerial insight regarding how to leverage consumers' brand attachment cues in brand strategy and online information management.
Keywords: brand nickname; information authenticity; inferred brand attachment; user-generated content
Sifting "real" information from that which is "fake" is a challenging task for consumers in today's digital landscape. The large amount of fraudulent information related to brands and products, whether fake reviews or copycat websites, increases consumer information search costs, violates consumer privacy, and enhances the likelihood that consumers may be misled to make less optimal choices ([20]; [36]; [39]). To help consumers identify fake information, researchers have identified possible quantitative factors, including the frequency with which first-person pronouns, emotional words, and conjunctions are used ([ 3]; [ 9]; [44]), and have suggested protocols for consumers to follow when navigating the digital world. Despite this, consumers' success rate for detecting fictitious online information remains low, at around 49%–52%—not much better than guessing by chance ([33]).
Consumers' general inability to accurately identify fake online content has led researchers to ask another fundamental question: What factors influence consumers' perception and judgment of authentic versus fake brand-related information in the digital world? In other words, how do consumers sift the grain from the chaff when seeking brand-related information online? One answer to that question lies in understanding what it means to be a socially aware human being and how one might transfer knowledge and experiences from the offline world to inform judgments in online contexts. Indeed, a growing body of research now identifies some social and psychological factors that may affect an individual's judgment of fake information. For example, [29] show that perceived social presence reduces people's likelihood of fact-checking statements in social settings.
In the current work, we build on this stream of research to demonstrate that consumers rely on interpersonal communication norms in the social world to evaluate brand-related information they encounter in online communication. Specifically, we show that consumers pick up on a popular relationship quality–indicating cue—brand nicknames—to evaluate the authenticity of online information. Brand nicknames are the "street names" or monikers that serve as the informal substitutes for brands' trademarked formal names, such as "Mickey D's" for McDonald's, "Bdubs" for Buffalo Wild Wings, and "Timmie's" for Tim Hortons. Considering that prior research shows consumers typically use brand nicknames in the marketplace in a positive manner ([55]), the current work focuses on the use of common brand nicknames that do not have negative connotations (the general discussion looks at possible future research directions to examine negative nicknames) and their contexts, such as online recommendations and positive word of mouth (WOM).
Drawing on the theory of cross-domain knowledge transfer ([26]), we theorize that consumers transfer their social knowledge of offline personal nickname use to the realm of brand-related online communication to infer brand relationship quality based on whether one uses a brand nickname in user-generated content (UGC). We show that when a writer (message sender) uses a nickname to refer to a brand, the reader (message receiver) is likely to infer that the writer has a genuine and close relationship with the brand. We conceptualize the reader's inference of the writer's relationship quality with the brand as inferred brand attachment (IBA) and show that IBA enhances the perceived information authenticity and leads to downstream consequences such as the reader's increased purchase intent. Furthermore, drawing on consumer persuasion knowledge theory, we show that when brand nicknames are used in firm-generated content (FGC) as an attempt to persuade, they may no longer be viewed as a relationship signal but rather a promotion tactic, thereby attenuating the effect. Figure 1 presents the full conceptual framework.
Graph: Figure 1. Conceptual framework.
By examining a popular yet understudied marketing phenomenon—namely, brand nickname use—this research highlights the importance of brand nicknames as a communication signal among consumers in the digital world. As such, we make two theoretical contributions. First, we extend the brand attachment literature into the interpersonal consumer context. While prior research treats brand attachment as the consequence of a private and binary relationship between a consumer and the brand ([ 2]), the current research demonstrates a novel role of brand attachment in peer-to-peer social interactions (i.e., between consumers online). Specifically, this research captures the social nature of the consumer–brand–consumer interaction by introducing the notion of IBA: the means by which third-party consumers infer the quality of the relationship between a consumer and a brand in a social context. Second, we demonstrate the process by which consumers rely on IBA to discern information authenticity in a social context, thereby highlighting the value of understanding IBA in consumers' social interactions in today's connected marketplace.
From a managerial perspective, we underscore the importance of consumer lingo, such as brand nicknames, in effective marketing communication. Prior research has mainly examined the use of language variation on an individual consumer's personal relationship with the brand (e.g., [49]), overlooking the influence of consumers' linguistic choice on other consumers' brand-related judgments in a social context. We build on [55] to show when and how brand nickname use in online communication can serve as a means by which brand-related information can be communicated authentically and credibly. Findings from the current research offer practical managerial insights regarding how consumer lingo should be strategically used and communicated in the digital era.
The remainder of the article is organized as follows. We first briefly introduce the phenomenon of consumer brand nickname use. We then theorize how nicknames in UGC in online communication may result in heightened IBA and explain why IBA facilitates consumers' perception of information authenticity. We present a historical Twitter data analysis with three real-world brands, together with a series of six studies to test our hypotheses. We conclude with a discussion of our findings' theoretical contributions and emphasize the managerial implications of brand nickname use in the social environment for marketers.
Many brands are known and referred to by their popular nicknames. Well-known examples include "Big Blue" for IBM, "Wally World" for Walmart, "Chevy" for Chevrolet, and "Tarjay" for Target (for more examples, see [55]). The Cambridge dictionary defines "nickname" as "an informal name for someone or sometimes something, used esp. to show affection, and often based on the person's name or a characteristic of the person." In the context of branding, brand nicknames are defined as "the informal and descriptive names that serve as a substitute for a brand's trademarked formal name" ([55]). While prior studies show that brand nickname use forges consumers' attachment to the target brand, little research to our knowledge has addressed the role of brand nickname use in a social context.
The online communication context is fertile ground to study brand nickname use. The open, social, and somewhat informal nature of social media and digital communication channels lends a novel context to investigate how informal brand elements such as brand nicknames can be used and perceived. A pilot study (N = 123, %female = 37%, Mage = 33.0 years; see Web Appendix 1a for details) revealed that brand nicknames are used frequently in digital channels. When asked to indicate "How often do you see brand nicknames in online communication, such as online reviews and social media posts, and such? (never, rarely, sometimes, frequently, all the time)," 79% of the participants indicated that they encounter brand nickname use at least "sometimes." Perhaps most interestingly, and most relevant to the current research, participants inferred that when a brand was referred to by its nickname in an online post, the poster was thought to have a closer and stronger relationship with the brand (Mnickname = 5.28, Mformal name = 3.99, t(121) = 5.67, p <.001, d = −1.02; 1 = "weak/distant relationship," and 7 = "strong/close relationship"), and the content was believed to be more authentic (Mnickname = 5.26, Mformal name = 4.74, t(121) = 2.61, p =.01, d = −.47; 1 = "fake/lying," and 7 = "authentic/telling the truth"). In the section that follows, we elaborate on why and how different name references (i.e., formal name vs. nickname) can affect perceived relationship quality and information authenticity in a social context.
Names are powerful ways to establish social connections and to indicate relationship qualities. [19] argues that names and naming are an essential "part of the fabric of daily life" and serve as important social markers that display the nature of certain relationships and shape social perceptions of those relationships. In terms of nicknames, studies have documented their use in a variety of contexts, including the workplace ([21]), online interactions ([ 1]; [ 5]), sports ([18]; [45]), the automotive industry ([50]), and education ([47]). Names, particularly nicknames, can help people understand the nature of a given relationship in a social context.
The literature on personal idioms reveals that the use of nicknames in interpersonal communication can reflect the intimate nature of a relationship ([ 6]; [45]). In the context of social relationships, [ 4] suggests that certain words and phrases provide psychological seclusion and convey relationship exclusivity. Human nicknames carry unique meanings and serve as important shorthand for affection that can cue close relational associations ([11]; [12]). In interpersonal relationships involving romantic partners ([12]), friends ([ 6]), family members ([35]), and even celebrities with their fans ([48]), nicknames serve as a linguistic cue to suggest relationship closeness and intimacy. As a product of social interactions, human nicknames have been documented as relationship "tie signs" that make "evident the (close) nature of the relationship to others" ([ 6], p. 310).
What about brand nickname use in the marketplace? Drawing on an understanding of analogy-based knowledge transfer, we theorize that consumers transfer the social signaling value of human nickname use to brand nickname use in consumer–brand interactions within their online communication. Specifically, a message sender's choice of brand reference (using a nickname versus a formal name) can serve as an indicator for other consumers to infer the sender's relationship quality with the brand. Analogical learning refers to the process by which consumers transfer their existing knowledge from a familiar domain (the base) to a novel domain (the target) ([24]; [26]; [41]). During this process, consumers are likely to categorize the novel domain into a similar and familiar domain and use their existing knowledge from the familiar domain to make inferences and judgments about the targets in the novel domain ([41]).
With respect to the interpretation of consumers' brand nickname use, we argue that the message receiver (i.e., the reader) may apply their knowledge of human nickname use to the realm of brand nickname use, which helps the receiver infer the message sender's (i.e., the writer's) relationship with the brand. [ 8], p. 3) suggest that language use reflects and signals information about the message sender, which can provide insight into the sender's "relationship with other attitude objects," such as brands. The sender's linguistic choice of brand reference, therefore, will simultaneously impact the message receiver's attitude and perceptions of the sender's relationship with the brand. Human nicknames usually indicate interpersonal relationship closeness; therefore, we expect that the receiver (the reader) is more likely to infer that the writer (the sender) has a close and genuine relationship with the brand when a brand nickname is used. We refer to this inference as IBA.
Conceptualized as such, IBA builds on and extends the concept of consumers' brand attachment from a dyadic (brand–consumer) perspective to a triadic (brand–reviewing consumer–observing consumer) perspective, which illustrates how an individual consumer's brand relationship is displayed, communicated, and perceived in a social context. [46], p. 2) define brand attachment as "the strength of the bond connecting the brand with the self." Accordingly, we define IBA as one consumer's perception of the strength of the bond connecting another consumer with the target brand. This conceptualization of IBA is grounded in prior literature that suggests that self–brand connection and brand prominence are the two fundamental components of consumers' brand attachment ([31]; [46]). As such, IBA captures the inferred brand connection and prominence to indicate an individual's (e.g., the writer's) overall relationship quality with a brand from a third party's (e.g., the reader's) perspective. Consider an example in which a consumer might personally feel a weak attachment or indifference to Walmart, but that same consumer is able to infer from how a friend speaks about Walmart that the friend has a strong attachment to Walmart (IBA).
In the context of online communication, we expect that the writer's use of brand nicknames (vs. formal names) may result in enhanced IBA. [46], p. 2) argue that a consumer's relationship bond with a brand is "inherently emotional." As brand nicknames are linguistic cues that signal affection and intimacy, readers are more likely to infer the writer has a stronger emotional connection with the target brand when a nickname is used. In addition, the use of a nickname may also reflect one's cognitive closeness with a brand. When referring to a brand, its formal name and nickname can be used interchangeably, as the two are logically equivalent. However, the fact that the writer chooses to use the brand nickname instead of its formal name suggests the salience of a more casual and intimate relationship between the writer and the brand. Taken together, the use of a brand nickname indicates a closer and stronger relationship that is salient in the writer's mind and thus leads to an enhanced IBA from the reader's perspective. Consistent with prior research ([46]), we also note that IBA (an indicator of brand relationship quality) is conceptually different from inferred brand knowledge (an indicator of the possession of brand-related information). We discuss this conceptual difference subsequently and empirically disentangle these two constructs in Study 4b. We thus hypothesize the following:
- H1: Brand nickname use in online communication results in enhanced IBA in comparison to brand formal name use.
The concept of authenticity generally captures the "dimensions of truth or verification" ([42], p. 9) and "encapsulates what is genuine, real, and/or true" ([10], p. 839). While the broad sense of authenticity pertains to the assessment of truthfulness, being truthful alone is usually "insufficient to capture the complex and varied way in which the concept (of authenticity) is often put to work" ([43], p. 610). Therefore, [42] suggests that the verification of authenticity can be examined through three fundamental lenses: historical authenticity, categorical authenticity, and values authenticity. Historical authenticity usually applies to the origins of works of art or historical artifacts, which involve "the evaluation of an object's unique spatiotemporal history" ([43], p. 612). An example is an authentic Picasso painting. Categorical authenticity involves the observer's verification of whether an entity meets the expectation of the category or type that is claimed. This type of authenticity pertains "mostly to objects or physical entities" ([43], p. 613), such as authentic Chinese food. Lastly, values authenticity, the authenticity that is most relevant to the current work, is based on authenticity stemming from acting in accordance with one's true beliefs and values ([16]) and from being "genuinely committed" to the task or object at hand ([42], p. 10).
With respect to information authenticity for UGC, we propose that it comes from the assessment that the information genuinely reflects the writer's real experiences and thoughts about the brand (values authenticity), and thus is deemed truthful. Given that a major source of fake information online is from people who are incentivized by a company to "promote" the brand ([38]; [52]), we surmise that the values authenticity lens is critical to the evaluation of UGC because it indicates whether the object (e.g., an online review) is created based on the agent's (e.g., the review writer's) true beliefs. Therefore, for UGC, the information is less likely to be perceived as fabricated or fake when the reader believes the information is in accordance with the writer's own experiences, opinions, and thoughts about the brand. The information thus comes across as truthful and authentic.
Building on the preceding arguments, we propose that a heightened IBA increases the perceived authenticity of the brand-related information. [46] argue that as consumers develop a stronger bond with the brand, they generate "a sense of oneness with brand" ([46], p. 2), and those brand-related thoughts and memories are more accessible in consumers' minds. Applying this to the context of brand nickname use, heightened IBA can serve as an indicator of a real relationship between the writer and the brand, which reflects the writer's true thoughts and opinions about the brand. In other words, we posit that brand nickname use signals the writer's actual interactions and experiences with the brand over time such that the information is retrieved in a casual and instinctive manner. As a result, a heightened IBA suggests that the writer can provide factual information based on his or her spontaneous thoughts, making the content appear authentic. Formally, we hypothesize the following:
- H2: A higher IBA results in enhanced perceptions of information authenticity.
Furthermore, when brand-related information is perceived to be more authentic, it also increases the perceived utility of that information. Prior research suggests that the increased diagnosticity of information can increase consumers' confidence in making a decision ([28]). As such, we expect that nickname use will result in important downstream consequences, such as willingness to purchase, enhanced information helpfulness ([40]), or information sharing ([53]). We expect the following:
- H3: Brand nickname use (vs. formal name use) in online communication leads to heightened IBA and enhanced perceived information authenticity, which results in positive downstream consequences.
We use an analysis of a historical Twitter data set followed by six experiments to test our hypotheses using six real-world brands and two fictitious brands with their corresponding nicknames (see Table 1). We situate these studies in different digital platforms (e.g., online reviews, Twitter, Instagram) to examine our hypothesized nickname effect and showcase the distinct downstream consequences relevant to each context (e.g., information sharing, review helpfulness, willingness to purchase). The analysis of three brands using the Twitter data set (Study 1) provides support for the focal effect—that tweets using brand nickname hashtags (vs. formal name hashtags) lead to more likes and retweets. Studies 2a and 2b replicate the main effect with another real brand (McDonald's) and provide evidence that brand nickname use (Mickey D's) enhances perceived information authenticity. Study 3 replicates the main findings and demonstrates that it is the concept of a nickname, rather than the specific word used for the nickname, that serves as the driving force for this effect. This finding helps rule out alternative explanations such as phonetic differences. To confirm that IBA is the underlying mechanism, we use a real brand (Bloomingdale's) and a fictitious brand in Studies 4a and 4b, respectively. We show that using a brand's nickname leads the reader to infer that the writer has a stronger attachment to the target brand than when a brand's formal name is used (Study 4a). This enhanced IBA further boosts the reader's perception of information authenticity and leads to downstream consequences (Study 4b). Study 5 explores an important managerially relevant boundary condition regarding whether companies should include their popular nicknames in FGC (vs. UGC): an experiment using the Walmart brand suggests that the nickname effect diminishes when companies use the nickname in FGC because of the consumer-based nature embedded in brand nicknames.
Graph
Table 1. Overview of Studies.
| Study | Brand and Nickname | Platform | Study Design and Main Findings |
|---|
| Study 1 | Chevrolet (Chevy), Buffalo Wild Wings (Bdubs), New England Patriots (Pats) | Twitter | Historical Twitter data analysis (n = 10,703 tweets) Tweets with brand nickname hashtags are liked more (p <.001) and retweeted more (p <.001)
|
| Study 2a | McDonald's (Mickey D's) | online review | 2 (nickname vs. formal name) between-subjects (n = 209 MTurkers) The online review using the brand nickname is less likely to be reported as fake (H2: p =.001)
|
| Study 2b | McDonald's (Mickey D's) | Twitter | 2 (nickname vs. formal name) between-subjects (n = 201 MTurkers) Nickname use increases perceived information authenticity (H2: p =.004)
|
| Study 3 | fictitious brand | online review | 3 (nickname, formal name, control) between-subjects (n = 300 MTurkers) A "nickname" but not any word implies relationship association (H2: p =.01)
|
| Study 4a | Bloomingdale's (Bloomies) | Instagram | 2 (nickname vs. formal name) between subjects (n = 287 MTurkers) Nickname use enhances IBA (H1: p <.001)
|
| Study 4b | fictitious brand | online review | 2 (nickname vs. formal name) between subjects (n = 215 MTurkers) IBA mediates the relationship between nickname use and information authenticity (H1: p =.003; H2: p <.001)
|
| Study 5 | Walmart (Wally World) | Instagram | 2 (nickname vs. formal name) × 2 (UGC vs. FGC) between subjects (n = 320 MTurkers) The nickname effect is attenuated for FGC (H4: interaction for information authenticity: p =.001; interaction for downstream behavior: p =.034)
|
| Supplemental Study 1 (Web Appendix 5) | Houston (Htown) | discussion forum | 2 (nickname vs. formal name) between subjects (n = 734 students) Nickname use leads to enhanced IBA and results in positive downstream consequences (H1: p <.001)
|
| Supplemental Study 2 (Web Appendix 5) | fictitious brand | online review | 2 (nickname vs. formal name) between subjects (n = 251 MTurkers) Nickname use increases information authenticity and leads to positive downstream consequences (H2: p =.001)
|
Study 1 provides real-world evidence to support our main argument that brand nickname use facilitates perceived information authenticity and results in downstream consequences such as information sharing. To do so, we used a paid service from twitonomy.com to collate one month of historical Twitter data from June 11 to July 10, 2019. We chose three brands based on a pilot study (see Web Appendix 2) that showed that "Chevy" for Chevrolet, "Bdubs" for Buffalo Wild Wings, and "Pats" for New England Patriots are considered popular nicknames for each brand.
For each brand, we collected tweets that used either the brand formal name or the nickname as a hashtag in the post (e.g., tweets containing either #Chevrolet or #Chevy). We chose to collect tweets with brand name hashtags (rather than those containing the brand names but with no brand name hashtags) because hashtags serve as the "keywords" in tweets and are designed to highlight the topic and help users easily find relevant content they are interested in. If no brand name hashtag appeared in a post, it is more likely that the brand is not central to the content. For instance, Chevrolet is less likely to be used as a hashtag in a tweet like "There's a major traffic delay on Hwy 59 near the Baytown Chevrolet dealer. #traffic" than in "Just went to the Baytown Chevrolet near Hwy 59, awesome seasonal sale there! #Chevrolet," as the first tweet is mainly about the traffic but not the brand. In addition, hashtags have been argued to serve as a reflection of one's sentiment to the public ([13]), which is consistent with our theorizing of IBA. Therefore, we expect that hashtags are more suitable and focal to the phenomenon of interest.
The collection yielded 12,095 total brand-related tweets (7,163 tweets with formal name hashtags, 4,932 tweets with nickname hashtags). We further categorized the tweets into consumer posts and nonconsumer posts by examining the account names (we considered a Twitter account name containing the brand name a nonconsumer account; e.g., "Tom Gill Chevrolet"). Our focus in this study is on consumer posts, so we excluded nonconsumer accounts from the data analysis, although including these tweets does not change the significance of the results (for the analysis with the full data set, see Web Appendix 2). As a result, the data set of consumer posts we used for hypothesis testing contained 10,703 tweets (6,315 with formal name hashtags and 4,388 with nickname hashtags).
For each tweet, we collected the following measures: ( 1) the number of retweets (shares), ( 2) the number of likes, and ( 3) the number of account followers. We used the number of retweets and likes as the key dependent variables for our analysis, as these real-world behavioral measures are good indicators of the readers' perceived authenticity of the information. We used the number of account followers as the control variable because posts from accounts with more followers might be more likely to be shared or liked due to higher exposure.
We ran an analysis of covariance (ANCOVA) with the name condition (nickname vs. formal name) as the independent variable, the number of retweets (i.e., shares) as the dependent variable, and the number of followers per individual account as a covariate. This analysis revealed that tweets with brand nickname hashtags were retweeted (Mnickname = 1.24) significantly more than tweets with brand formal name hashtags (Mformal =.58, F( 1, 10,700) = 33.73, p <.001, η =.003). A similar ANCOVA analysis with the number of likes as the dependent variable showed that tweets with brand nickname hashtags also received more likes (Mnickname = 8.11) than those with formal name hashtags (Mformal = 2.72, F( 1, 10,700) = 47.65, p <.001, η =.004). We obtained similar results when we analyzed each brand separately (all ps <.01); see Web Appendix 2 for details.
Across three real-world brands, the historical Twitter data provides initial empirical evidence that when consumers use brand nicknames in social media communication, their posts are shared more and liked more—indicators of perceived information authenticity—than when they use the formal brand names. In the experimental studies that follow, we replicate this finding with both real and fictitious brands to provide evidence for the underlying mechanism, rule out alternative explanations, and identify a boundary condition for our effect.
We designed Study 2 to demonstrate that brand nickname use in online communication can enhance perceived information authenticity and lead to downstream behaviors. To show the robustness of the nickname effect, we replicate the finding on the key dependent variable of information authenticity by using both a behavioral measure (Study 2a) and a scale measure (Study 2b). Using a real-world brand (McDonald's) as the stimuli, we tested the proposed effect in two online sharing contexts: online reviews (Study 2a) and social media posts (Study 2b).
We designed Study 2a to capture consumers' behavioral responses to an online review. We based this design on the notion that brand nickname use can serve as a filter by which consumers sift real reviews from those that are fake, with the goal of capturing whether consumers are more likely to report a new product review as fake depending on whether it used the brand nickname or formal name. The practice of reporting problematic information (such as a possible fake review) is common, and many professional review platforms (e.g., Google Reviews, Tripadvisor.com, Amazon.com) offer this option to readers to help manage review quality.
Two hundred nine paid Amazon Mechanical Turk (MTurk) workers participated in this between-subjects study (%female = 52%, Mage = 38.2 years). Participants learned that McDonald's recently introduced a new "Mozzarella Chicken Sandwich" to its menu. To find out more about this new item, participants were directed to a review website and read an online review about the new sandwich posted by someone named Alex Smith. Alex recommended the sandwich and mentioned an ongoing promotion (buy one, get one free). Depending on the condition, participants read a review in which the brand was referred to by either its formal name (McDonald's) or its nickname (MickeyD's); see Appendix 3a for details of the stimuli.
We used participants' real clicking behavior to measure perceived information authenticity in this study. Participants were informed that the review website was public and anyone could post reviews. They were then cautioned that the website could contain fake reviews. We were interested in capturing perceived information authenticity, so we expected that participants who thought Alex's review was potentially fake would click on the "Report Review" button to report it to the website. In contrast, if participants believed that the review was authentic, they would not click the button.
A chi-square test showed that the writer's brand name choice (nickname vs. formal name) significantly influenced whether the review was reported as being fake (χ2( 1) = 10.73, p =.001, φ =.23). Specifically, 49.5% of participants in the formal name condition clicked on the button to report the review as fake, whereas only 27.5% of participants did so in the nickname condition. This result supports our theory that the use of the brand nickname suggests to a reader that the information provided is authentic. In Study 2b, we replicate this result by measuring perceived information authenticity.
Two hundred one paid MTurk workers participated in Study 2b's between-subjects experiment (%female = 51%, Mage = 38.9 years). Participants learned that when browsing Twitter, they came across a tweet from Alex Smith. Alex tweeted about the iced coffee from McDonald's, using either the brand formal name (McDonald's) or its nickname (Mickey D's), depending on the condition; see Appendix 3b for details of the stimuli.
We used a scale measure to capture perceived information authenticity in this study ("To what extent do you think...: "Alex's post is genuine and sincere," "Alex's post seems fake" [reverse coded], and "Alex's post is a paid advertisement" [reverse coded]; we later combined these items into a perceived information authenticity scale [α =.85][ 6]). Participants indicated how likely they would be to "take the recommendation from the post, and give the iced coffee a try," which serves as a measure of willingness to purchase. For all measures, 1 = "not at all," and 7 = "very much."
A t-test with perceived information authenticity as the dependent variable showed that participants in the nickname condition (MMickey D's = 3.81) perceived Alex's post to be significantly more authentic than did those in the formal name condition (MMcDonald's = 3.14, t(199) = −2.91, p =.004, d = −.41). Furthermore, a similar t-test showed that participants were more willing to get an iced coffee from McDonald's when Alex used the brand nickname Mickey D's in the post (MMickey D's = 3.79, MMcDonald's = 3.06, t(199) = −2.58, p =.011, d = −.36). Finally, as predicted, the result of a mediation analysis ([27], model 4: 5,000 bootstrapped samples; independent variable [IV] = name condition, mediator [M] = perceived information authenticity, dependent variable [DV] = willingness to purchase) showed a positive and significant indirect effect (ab =.4267, 95% CI [.1394,.7234]).
Studies 2a and 2b use a real brand and its nickname to replicate and extend the Twitter study findings (supplemental study 2 in Web Appendix 5 also serves as a replication using a fictitious brand). They provide support for our main hypothesis that how a brand is referred to (nickname vs. formal name) in online communication influences the reader's perception of information authenticity, captured via a behavioral measure (Study 2a) and a scale measure (Study 2b). These findings provide empirical support for H2 and H3. Taken together, the results suggest that nickname use can make online information appear more authentic and lead to positive downstream consequences.
Study 2, however, does not answer a key question associated with this effect: whether the observed effect is driven by phonetic differences between the formal name and the nickname (i.e., McDonald's and Mickey D's sound different). To demonstrate that the nickname effect stems from the relational associations it implies rather than phonetic differences, Study 3 uses a fictitious brand and includes an additional condition in which the nickname is used as the brand formal name. Because we theorize that it is the concept of the nickname and not the actual word used as the nickname that matters, we expect that the same name when used as a formal name would not serve as a relationship quality cue to influence information authenticity.
Three hundred paid MTurk workers participated in Study 3's between-subjects experiment (%female = 46%, Mage = 39.5 years). We randomly assigned participants to one of three experimental conditions (formal name, nickname, and nickname as formal name conditions; for ease of exposition, we refer to the last condition as the control condition). In the formal name and nickname conditions, participants learned that they were looking to buy a portable humidifier (Sunnwal Ultrasonic Portable Mist Air Humidifier from the brand Sunnwal). They then learned that Sunnwal has a popular nickname—"Sunny"—among consumers due to the product's bright yellow color. In the control condition, participants were told that the humidifier they were considering was the Sunny Ultrasonic Portable Mist Air Humidifier from the brand Sunny (i.e., "Sunny" as the formal brand name). Participants in the control condition also learned that the Sunny product was a bright yellow color. All the participants were shown a picture of the product so they could visualize the humidifier.
Participants were then directed to read a review written by a consumer named Alex from a review website. In all the conditions, Alex's review stated that the humidifier was easy to use and was an easy solution to relieve dryness. The only difference across the conditions was how the brand was referred to in the review ("Sunnwal" in the formal name condition, "Sunny" in the nickname condition, and "Sunny" in the control condition).
We measured perceived review authenticity (α =.90) using the same scale as in Study 2b. We assessed two important downstream consequences that are pertinent to the online review context: perceived review helpfulness ("Was the review helpful to you?") and WOM (Would you "recommend Alex's review to another friend who is also thinking about buying the mini humidifier?"). For all the measures, 1 = "not at all," and 7 = "very much."
A one-way ANOVA with perceived review authenticity as the dependent variable (F( 2, 297) = 4.65, p =.01, η =.03) showed that participants in the nickname condition (Mnickname = 4.89) perceived the review to be significantly more authentic than participants in the formal name condition (Mformal = 4.22, t(297) = −2.98, p =.003, d = −.42). These results replicated our findings from the previous studies. More importantly, participants in the nickname condition believed Alex's review to be more authentic than those in the control condition, although they viewed the same review with the same brand name "Sunny" (Mnickname = 4.89, Mcontrol = 4.42, t(297) = −2.07, p =.039, d = −.31). In contrast, the difference between the formal name and control conditions was not significant (p >.35, d = −.13). These results suggest that it is not the specific word "Sunny" that drives this effect; rather, it is because "Sunny" is used as a nickname.
A similar one-way ANOVA with review helpfulness as the dependent variable (F( 2, 297) = 4.12, p =.017, η =.027) revealed that participants in the nickname condition (Mnickname = 5.43) perceived the review to be significantly more helpful than participants in the formal name condition (Mformal = 4.81, t(297) = −2.86, p =.005, d = −.40) and marginally more helpful than participants in the control condition (Mcontrol = 5.07, t(297) = −1.68, p =.094, d = −.25). The difference between the formal name and control conditions was not significant (p >.23, d = −.16).
A similar one-way ANOVA with WOM as the dependent variable (F( 2, 297) = 4.37, p =.014, η =.029) revealed that participants in the nickname condition (Mnickname = 4.86) were more likely to pass along the information to another friend than participants in both the formal name condition (Mformal = 4.08, t(297) = −2.89, p =.004, d = −.42) and the control condition (Mcontrol = 4.31, t(297) = −2.02, p =.044, d = −.29). The difference between the formal name and control conditions was not significant (p >.37, d = −.12).
The results of a mediation analysis ([27], model 4: 5,000 bootstrapped samples; IV = name condition, M = perceived review authenticity, DV = review helpfulness) showed that the indirect effect was positive and significant (ab =.2203, 95% CI [.0689,.3831]). A similar mediation with DV = WOM also showed a positive and significant indirect effect (ab =.2499, 95% CI [.0826,.4260]).
The results of Study 3 demonstrate that the observed effect comes from the writer's brand nickname use, rather than the specific word used for the nickname. When the same word was used as the brand's formal name, the influence on perceived information authenticity and downstream consequences was diminished. In the two studies that follow, we aim to empirically illustrate the proposed mechanism underlying this effect. In Study 4a, we demonstrate that brand nickname use leads to a heightened IBA. In Study 4b, we test the complete model.
In Study 4a, we aim to demonstrate that the use of a brand nickname in online communication can lead readers to make inferences about the writer's relationship quality with the target brand (IBA). Using the real-world brand Bloomingdale's, we show that brand nickname use can heighten IBA and lead to perceived information authenticity.
Two hundred eighty-seven paid MTurk workers participated in this between-subjects study (%female = 43%, Mage = 38.2 years). We used a real-world brand, Bloomingdale's, as the stimulus. Participants were told that when browsing Instagram, they came across a post from Alex Smith. Alex posted something about shopping for shoes at Bloomingdale's department store, using either the brand formal name (Bloomingdale's) or its nickname (Bloomie's), depending on the condition; see Web Appendix 3d for details of the stimuli.
Participants reported their IBA using a scale adapted from [46] ("To what extent do you feel that this brand is part of Alex and who Alex is?" "To what extent do you feel that Alex is personally connected to this brand?" "To what extent do you think Alex's thoughts and feelings toward this brand are automatic, coming to his or her mind seemingly on their own?" and "To what extent do you think Alex's thoughts and feelings toward this brand come to him or her naturally and instantly?"; 1 = "not at all," and 7 = "very much"). We then combined all items to generate an IBA scale (α =.90). Participants reported the perceived authenticity of Alex's post using the information authenticity scale used in previous studies (α =.79).
A t-test with IBA as the DV revealed that participants in the nickname condition reported a significantly higher IBA (Mnickname = 5.22) than those in the formal name condition (Mformal = 4.49, t(285) = −4.59, p <.001, d = −.55). These results support H1. A similar t-test with perceived information authenticity as the DV showed that participants perceived Alex's post with the nickname (Mnickname = 4.13) to be marginally more authentic than the one with the formal name (Mformal = 3.80, t(285) = −1.87, p =.063, d = −.22).
Mediation analysis ([27], model 4: 5,000 bootstrapped samples; IV = name condition, M = IBA, DV = perceived information authenticity) results showed that the indirect effect was positive and significant (ab =.3461, 95% CI [.1807,.5463]). These results suggest that IBA could be the driving force for the effect of brand nickname use on information authenticity, in support of H2.
Study 4a shows that brand nickname use can positively influence readers' inference of the writer's relationship quality with the brand, captured as IBA in the current study. We further demonstrate that the positive influence of brand nickname use on information authenticity can be explained by IBA. In the following study, we extend these findings by testing the complete model, and we rule out inferred brand knowledge (via nickname use) as an alternative mechanism.
Given brand nicknames are relationship indicators, we expect that they do not systematically influence the reader's inferences of the writer's brand knowledge for two reasons. First, many popular brand nicknames are not technical terms; thus, the use of these nicknames does not require specific expertise or unique knowledge about the target brand. For instance, a chemist is likely to use "sodium bicarbonate" to refer to baking soda because of (and to signal) their knowledge and expertise in the specific area. However, given their "street" nature, many brand nicknames constitute common consumer language used in everyday life (e.g., "Chevy" for Chevrolet, "Mickey D's" for McDonald's) and thus are more likely to serve as a relationship signal than a knowledge indicator.
Second, brand knowledge alone seems insufficient for consumers to make inferences about information authenticity. [30] conceptualizes brand knowledge as the descriptive and evaluative brand-related information stored in a consumer's memory (e.g., brand awareness, general brand attitudes). However, these basic cognitive representations do not necessarily indicate a consumer's relationship quality with a brand, nor do they reflect whether the information is in accordance with ones' genuine and true thoughts about the brand. Therefore, we expect that IBA may serve as a more comprehensive explanation than inferred brand knowledge to explain the observed relationship between brand nickname use and perceived information authenticity.
We designed Study 4b with three objectives in mind. First, we aimed to test the full model with a focus on showing that IBA is the driving force for brand nickname use and resulting perceptions of information authenticity. Second, to demonstrate that nicknames signal one's brand relationship but not their brand knowledge, we measured inferred brand knowledge with a multi-item scale to empirically rule out this competing mechanism. Third, we extended our measure of downstream consequences by showing that increased information authenticity can also improve the acceptance and persuasiveness of the information (e.g., readers are more likely to take the writer's advice about the product recommendations).
Two hundred fifteen paid MTurk workers participated in this between-subjects study (%female = 47%, Mage = 33.3 years). The manipulation and procedure were similar to Study 3. Specifically, participants were told that they were looking to buy a new smart speaker and were considering the AcouTech Voice Activated Smart Speaker from the brand AcouTech. All participants viewed pictures of the AcouTech smart speakers and were told that due to the unique product design (ball-shaped) and advanced artificial intelligence technology (it is a smart device), the speaker is referred to by its popular nickname "Magic Ball" among consumers. Participants were then directed to a popular review website, where they read a product review written by a consumer named Alex. We randomly assigned participants to either the nickname or formal name condition. In both review conditions, Alex said that the speaker was easy to use, responded quickly to voice commands, and was versatile in performing a variety of tasks. Again, the only difference between the conditions was the formal name (AcouTech) and nickname (Magic Ball) Alex used to refer to the brand in the reviews.
We measured inferred brand attachment using the scale (α =.83) from Study 4a and perceived authenticity of the review using the information authenticity scale from previous studies (α =.90). We assessed three important factors pertinent in an online review context: perceived review helpfulness ("Was the review helpful to you?"), intent to take the advice of the review writer ("I would take Alex's advice about the product recommendations"), and WOM ("I would recommend Alex's review to another friend who is also thinking about buying the smart speaker").
We measured inferred brand knowledge using a multi-item scale that included the following: "To what extent do you think Alex is an expert of this brand?"; "Alex knows the brand well"; "Alex has more knowledge about this brand than other consumers"; and "Alex is very familiar with this brand." We later combined these items into an inferred brand knowledge scale (α =.92).
A t-test with IBA as the DV revealed that participants in the nickname condition reported a significantly higher IBA (Mnickname = 4.90) than those in the formal name condition (Mformal = 4.36, t(213) = −2.97, p =.003, d = −.40). A similar t-test with perceived review authenticity as the DV showed that participants perceived the review in the nickname condition (Mnickname = 4.31) to be more authentic than that in the formal name condition (Mformal = 3.18, t(213) = −4.90, p <.001, d = −.67). A t-test with perceived review helpfulness as the DV revealed that participants in the nickname condition (Mnickname = 4.64) indicated that the review was more helpful than participants in the formal name condition (Mformal = 4.13, t(213) = −2.02, p =.045, d = −.28). Results of a similar t-test showed that participants in the nickname condition were also more likely to take the advice from the review than those in the formal name condition (Mnickname = 4.02, Mformal = 3.35, t(213) = −2.67, p =.008, d = −.37). Furthermore, a similar t-test with WOM as the DV revealed that participants in the nickname condition were more likely to recommend the review to other friends than were those in the formal name condition (Mnickname = 4.20, Mformal = 3.78, t(213) = −1.96, p =.052, d = −.27). However, a t-test with inferred brand knowledge as the DV revealed no significant difference between the nickname and formal name conditions (Mnickname = 4.28, Mformal = 4.35, t(213) =.32, p >.70, d =.05). The differences for individual items in the scale were not significant (ps >.30). These results suggest that nickname use does not directly influence the reader's inference about the writer's brand knowledge. Figure 2 shows the results of a serial mediation analysis with IBA and inferred brand knowledge as parallel mediators ([27], model 80: 5,000 bootstrapped samples); they indicate that only IBA mediated the observed effects.
Graph: Figure 2. Serial mediation analysis for Study 4b.*** p <.01.Notes: Advice-taking: ab =.86, 95% CI with M1a [.0733,.4033], 95% CI with M1b [−.0284,.0286]. Helpfulness: ab =.83, 95% CI with M1a [.0691,.3976], 95% CI with M1b [−.0263,.0233]. WOM: ab =.47, 95% CI with M1a [.0349,.2298], 95% CI with M1b [−.0154,.0134].
Taken together, Studies 4a and 4b implicate IBA as the process mechanism underlying our hypothesized effect. A similar study reported in Web Appendix 5a (supplemental study 1) also shows that consumers infer greater brand attachment and are more willing to spread WOM of a recommended local store when the recommender uses the city's popular nickname (Htown) rather than its formal name (Houston). Importantly, these studies extend research on brand attachment to the realm of individual consumers' brand relationship in a social environment in which IBA serves as a communication signal for assessing information authenticity. In particular, IBA helps explain why and how brand nickname use can enhance information authenticity and lead to downstream consequences. Furthermore, Study 4b shows that while nickname use positively influences readers' assessment of information authenticity through IBA, it does not systematically change readers' inference about the writer's brand knowledge. This argument, supported by our data, is consistent with the definition of brand attachment [46] propose and provides additional empirical support to differentiate the construct of brand attachment from that of brand knowledge.
So far, we have shown that brand nickname use can positively influence readers' judgments of information authenticity in online communication. However, one important premise of this finding is that the brand nickname is used in UGC, such as in consumers' online reviews and social media posts. It is reasonable then to ask: Should companies employ this nickname strategy in FGC, such as their own social media posts or marketing campaigns ([34])? For example, if Target refers to itself as "Tarjay" in its own tweets or Instagram posts, would it influence the reader's perception of information authenticity in the same way as when a consumer uses the nickname? The results from a pilot survey (N = 241, %female = 37%, Mage = 36.9 years; see Web Appendix 1b) show that a majority of the participants believed that brand nicknames originate from consumers (81%), not companies (19%), and are used mostly by consumers (88%), not companies (12%). These findings imply that brand nicknames may not work for FGC because they are street names used by consumers and tend to come from a consumer source, perhaps lending to their street cred.
In line with persuasion knowledge theory, we expect that the nickname effect may be attenuated when brand nicknames are used (or even adopted) by companies to promote their own goods and services. Consumers' persuasion knowledge refers to consumers' beliefs and theories of marketers' motives, tactics, and persuasion attempts ([22]). [14] argue that persuasion knowledge helps consumers infer and explain marketers' motives and behaviors with respect to their intent to persuade consumers.
When used in UGC, a brand nickname serves as a cue of relationship quality: it signals one consumer's relationship quality with the brand to another consumer. In contrast, when a company itself uses the nickname, the relationship quality signal is lost, and brand nickname use becomes yet another promotion tactic the company uses to influence the consumer. For this reason, we expect that nickname use in FGC may activate consumers' persuasion knowledge and damage information authenticity. Specifically, when brand nicknames are used in FGC, consumers may interpret this as a marketing tactic in which firms utilize consumer lingo to achieve a marketing goal ([14]). We thus hypothesize the following:
- H4: The positive effect of brand nickname use on perceived information authenticity is attenuated when the information is FGC (vs. UGC).
Three hundred twenty paid MTurk workers participated in this 2 (information type: UGC vs. FGC) × 2 (name type: formal name vs. nickname) between-subjects design study (%female = 45%, Mage = 37.8 years). We used another real brand, Walmart, as the stimuli in this study. Participants were asked to imagine that they were planning a party and were looking for some ideas online. When browsing Instagram, they came across a post that recommended the cupcakes from Walmart. To manipulate information type, participants were told that the post is from either Walmart's Instagram account (FGC) or another consumer named Alex Smith (UGC). Participants then saw an Instagram post with the profile picture of either Walmart or Alex, depending on the condition. In addition, we manipulated name type by how the brand was referred to (Walmart vs. Wally World) in the post ([37]); see Web Appendix 3f for details of the stimuli.
We measured perceived information authenticity using the same authenticity scale from previous studies with minor changes to adapt to the study context ("To what extent do you think...:" "the post is genuine and sincere," "the post is an advertisement," and "the post seems fake"; α =.66). Participants indicated how likely it was that they would "take the recommendation from the post and give the cupcake a try," which served as the downstream consequence for willingness to purchase.
Perceived information authenticity: A two-way ANOVA revealed a significant interaction of information type and name type (F( 1, 316) = 10.77, p =.001, η =.033). The main effects of information type (p =.18, η =.006) and name type (p =.25, η =.004) were not significant. For UGC, nickname use significantly increased information authenticity (MUGC formal = 3.40, MUGC nickname = 4.07, F( 1, 316) = 9.87, p =.002, η =.030). However, for FGC, this difference disappeared, and brand nickname use did not increase perceived information authenticity (MFGC formal = 3.69, MFGC nickname = 3.37, F( 1, 316) = 2.27, p =.13, η =.007). In addition, nickname use in UGC significantly increased information authenticity compared with nickname use in FGC (MUGC nickname = 4.07, MFGC nickname = 3.37, F( 1, 316) = 10.73, p =.001, η =.033). However, we observed no significant difference between UGC and FGC for formal name use (MUGC formal = 3.40, MFGC formal = 3.69, F( 1, 316) = 1.89, p =.17, η =.006). Figure 3 presents the results graphically.
Graph: Figure 3. Results of Study 5 on perceived information authenticity.***p <.01.Notes: Error bars = ± 1 SE.
Willingness to purchase: A two-way ANOVA revealed a significant interaction of information type and name type (F( 1, 316) = 4.53, p =.034, η =.014). The main effects of information type (p =.71, η <.001) and name type (p =.71, η <.001) were not significant. For UGC, nickname use increased participants' willingness to purchase the product (MUGC formal = 3.33, MUGC nickname = 3.86, F( 1, 316) = 3.15, p =.077, η =.010). However, for FGC, the effect was diminished (MFGC formal = 3.71, MFGC nickname = 3.33, F( 1, 316) = 1.53, p =.22, η =.005). In addition, nickname use in UGC increased participants' willingness to purchase the product compared with nickname use in FGC (MUGC nickname = 3.86, MFGC nickname = 3.33, F( 1, 316) = 3.15, p =.077, η =.010). However, we observed no significant difference between UGC and FGC for formal name use (MUGC formal = 3.33, MFGC formal = 3.71, F( 1, 316) = 1.53, p =.22, η =.005). Figure 4 presents the results graphically.
Graph: Figure 4. Results of Study 5 on willingness to purchase.*p <.1Notes: Error bars = ± 1 SE.
A moderated mediation analysis ([27], model 7: 5,000 bootstrapped samples; IV = name type, M = perceived information authenticity, DV = willingness to purchase, W = information type) revealed a significant moderated mediation (index of moderated mediation =.6571, 95% CI [.2746, 1.0857]). The indirect effect of brand nickname use on willingness to purchase via perceived information authenticity was significant for UGC (95% CI [.1307,.7839]). However, the indirect effect was not significant for FGC (95% CI [−.4594,.0150]); see Web Appendix 4a for regression coefficients.
Study 5 explores a boundary condition regarding whether companies should adopt and include their nicknames in FGC. The results suggest that brand nicknames work for UGC but not for FGC, possibly due to the activation of consumers' persuasion knowledge in FGC. Therefore, while brand nicknames are popular with consumers, companies should be careful when appropriating consumer lingo, as it could be perceived as deliberate and less authentic.
A recent Washington Post investigation suggests that on some online commerce platforms such as Amazon.com, the number of fraudulent UGC-like consumer reviews exceeds the number of authentic ones for some popular product categories ([17]). As such, communicating brand-related information authentically and building continuous consumer trust is a crucial issue facing marketers today. By investigating brand nickname use in the marketplace, the current research introduces the new concept of IBA to the literature and demonstrates its influence on consumers' judgment of information authenticity. Specifically, we show that the different ways in which a brand is referred to by a message sender (e.g., the writer) can influence the receiver's (e.g., the reader's) perception of the sender's brand attachment, which in turn shapes the evaluation of information authenticity and results in effects on downstream consequences such as willingness to purchase, review helpfulness, and information sharing.
We present a study using historical Twitter data and a set of six experimental studies with both real and fictitious brands to support our theorizing. These studies provide converging evidence that, in online communication, readers infer the writer's brand relationship quality (captured as IBA from the reader's perspective; Studies 4a and 4b) according to how the brand is referred to. These inferences subsequently affect the readers' evaluation of online information authenticity and lead to downstream consequences (Studies 1–5). We demonstrate that the power of a brand nickname lies in the notion that it is an informal way to address a brand (Study 3) and that it is consumer based (Study 5). The studies also rule out the explanation of inferred brand knowledge (Study 4b) as a competing mechanism underlying this effect. Moreover, we report two supplemental studies in Web Appendix 5 that provide additional evidence for our hypothesized effects. Overall, the current research finds that brand nickname use serves as a means by which consumers infer the authenticity of online information and underscores the significance of recognizing consumers' brand relationship cues in marketing communication.
By treating brand attachment as a social signal and examining it from the message receiver's perspective, the current research offers two theoretical contributions to the branding and marketing communications literature. First, it places brand attachment in a social context so as to shed light on a novel function of brand attachment within consumers' interpersonal communication. We demonstrate that brand attachment cues can signal a consumer's relationship quality with the brand and further influence how message receivers perceive and process the sender's information. By investigating how an individual's brand attachment via language variation in the social environment may affect other consumers' judgment and perception, brand attachment is no longer examined as the consequence of the consumer–brand relationship; rather, it serves as an antecedent to important marketing consequences such as information authenticity. This novel perspective opens a new avenue and serves as the basis for future research in the study of brand attachment as a social signal.
Second, by switching the research focus from the consumer–brand relationship to peer-to-peer consumer interaction, the current research highlights the importance of brand attachment in successful marketing communication with respect to UGC. Results from this research suggest that IBA positively shapes the message receiver's perceived information authenticity. In addition, while previous research has typically focused on how consumers use possessions (e.g., products, brands) to show social status and identity, the current work suggests that relationship-indicating cues, such as brand nicknames, may have the same signaling effects and are thus used by consumers in their social interactions.
Beginning in the 1910s, Coca-Cola initiated a campaign with the theme "Coca Cola: Ask for it by its full name" and engaged in a 30-year marketing effort to dissuade consumers from using the nickname "Coke" ([51]). For Chevrolet, an internal memo showed that the company had a "swear jar" in the company hallway to "accept a quarter every time someone uses 'Chevy'" ([15]). A number of brands dissuade nickname use out of fear that these unofficial names might dilute the brand equity and confuse consumers. However, the current findings suggest that brand nicknames reflect genuine consumer language and resonate well in conversations between consumers. We show that brand nickname use can make the brand information appear more authentic and lead to desirable downstream consequences for brands. Therefore, brands should embrace their popular nicknames and be open about consumers using these nicknames. Furthermore, brands can be strategic and creative in how they rely on nicknames to communicate trustworthy brand information. For example, consumer reviews that use brand nicknames could be placed at the top of webpages and labeled as the "top reviews" so they are read first. Brands can also consider highlighting nicknames in other types of peer-to-peer interactions, like referral programs, to convey that the messaging is natural and authentic.
Findings from the current research suggest that brand nicknames seem more useful in UGC than FGC, highlighting the importance of limiting brand nickname use to consumers. While adopting or even trademarking a nickname may seem to be a convenient way for brands to engage consumers or rebrand, it could potentially discount the street cred of these monikers. As a case in point, McDonald's rebranded itself after its well-known nickname "Golden Arches" in China ([23]). In Argentina, the popular nickname "Pecsi" became the new name for Pepsi in order to match the pronunciation of Spanish consumers ([54]). On one hand, the company's adoption of a nickname may facilitate consumers' brand recognition and memory (as it naturally is a consumer language). At the same time, the "contamination" of firm elements into consumer lingo could damage the street cred of brand nicknames and result in the loss of their future marketing value. The reason brand nicknames resonate with consumers largely lies in its "street" nature. If a nickname is frequently used in advertisements or shown on the product package, it might not be viewed as consumer-based language anymore and may no longer be perceived as organic and authentic.
Finally, the current research suggests that informal brand-related language is popular in online communication among consumers. Therefore, brands should leverage the power of consumer-based language, such as brand nicknames, to maximize brand presence in the digital landscape. While the current research mainly examined nickname use in UGC, one could imagine that nicknames, as a casual way to refer to a brand, can be used by consumers in other types of online activities. For instance, instead of typing in a brand's formal name, consumers may input its nickname on search engines to look for brand related information. Therefore, marketers should keep brand nicknames in mind during their search engine optimization process to improve the quantity and quality of website traffic.
Brand nickname use in the marketplace is a broad and nuanced phenomenon. Unsurprisingly, the current research does not fully address every aspect of the phenomenon and, therefore, has some limitations. First, we explore the use of brand nicknames in the context of positive UGC. However, consumers may also include brand-related monikers—even neutral or positive ones—in negative situations such as product failure or brand betrayal. The use of a positive nickname in a negative context could come across as humorous or ironic and influence how the message is understood. Further, not all brand nicknames are positive. Some brand nicknames that convey negative consumer sentiment do exist in the marketplace; consider "Needless Markup" for Neiman Marcus and "Fix it again, Tony!" for Fiat. Future research could explore how negative nicknames are communicated between consumers to provide a more complete picture of brand nickname use in the marketplace.
Second, the current research deals with well-known and popular brand nicknames, and so for the nicknames we use, we find that brand knowledge does not adequately explain the observed effect. However, considering that brand nicknames are "street names," their widespread use in the marketplace relies on consumers sharing the cultural capital needed to pick up on the significance of the nickname as a relationship quality cue. Relatedly, some brand nicknames may serve as subtle signals of status and being the "in the know" (e.g., "Bolly Darling" for the champagne brand Bollinger, the "255" bag for the iconic Chanel handbag; [ 7]). We posit that brand nickname use for specialty goods, like luxury or pharmaceutical products, could serve as an indicator of specialized brand knowledge. Consider that a regular consumer may refer to the Chanel quilted handbag as a Chanel bag, while an expert luxury consumer may use its nickname—the "255" bag—to signal their special brand knowledge (they know the story behind the nickname: Coco Chanel launched the bag in February 1955, giving it a nickname that corresponds to the date). Similarly, nickname use for some drugs might only be decoded by those who share the same brand knowledge or experience, such as "Vitamin V" for Viagra or "Addys" for Adderall. Future research could examine the role of brand nicknames as a means to subtly convey status via signaling specialized brand knowledge.
The current research also raises some interesting questions that could be examined in future research. First, the current research mainly focuses on one important dimension of marketing communication: perceived information authenticity. However, brand attachment signals might also influence other dimensions of marketing communication, such as information accuracy and communication efficiency, which may take place in a variety of communication channels. For instance, when a salesperson uses a brand nickname in a sales pitch, consumers may perceive the salesperson as unprofessional and the information to be less credible. Future research, therefore, could explore how brand attachment cues can affect other aspects of consumers' information processing. Relatedly, in accordance with prior research ([46]), we conceptualized IBA in terms of its two components, self–brand connection and brand prominence. Future research could determine the weights of these two components in terms of how they shape and contribute to information interpretation.
Second, it would be worth investigating how to strategically use brand attachment cues without overdoing it. Brand nicknames are relatively subtle linguistic cues in the manner in which they signal brand attachment. It is possible that overtly signaling brand attachment could backfire. Future research could examine the notion that different levels of IBA might result in an inverted-U shape of information authenticity: too little IBA may suggest the information is fictitious, but too much IBA might backfire and lead to the perception of favoritism or bias.
Third, it would be worthwhile to explore potential moderators that could allow firms to benefit from using their nicknames in FGC. We briefly touch on brand nickname use in FGC (in the context of promoting and selling a product) and show that it may not be an optimal strategy. However, certain conditions could make nickname use desirable and beneficial in a firm's own brand messaging. For example, perhaps nickname use in FGC related to charity or corporate social responsibility (vs. merely trying to advertise a product) could enhance perceived brand warmth and lead to more positive consumer responses. Individual differences may also affect consumer response to brand nickname use in FGC. Future research could examine factors such as brand loyalty, consumer activist tendencies ([32]), or whether the FGC gains negative traction on social media inciting consumer groups ([25]). It could also be worthwhile for future research to investigate brand nickname use in other types of paid content; for instance, if used by (paid) micro-influencers, do brand nicknames still convey authenticity?
Closely related to the issue regarding firms' brand nickname use is the importance of understanding whether and how firms should promote nickname use among consumers. While we suggest that marketers should not discourage or restrict consumers from using brand nicknames, it is unclear whether it is wise to take the opposite approach by actively encouraging consumers to use brand nicknames. For example, a brand can create a "hashtag us by our nickname" campaign on Twitter. In this case, the brand nickname still appears in UGC eventually; however, because its use is initiated and prompted by the company, it is unclear whether the nickname would still lead to similar benefits as when it is used by consumers organically. On one hand, once companies get involved, it is possible that it dilutes the street cred of brand nickname use. On the other hand, perhaps a certain degree of encouragement from the firm is worthwhile as long as the action is not directly associated with external incentives (e.g., associating using the nickname hashtag with the chance of winning a prize offered by the company). Future research, therefore, could look into whether, how, and to what extent companies should be involved in consumers' brand nickname use process.
It is often said that "trust decreases transaction costs." In the current research, we find that when consumers refer to a brand by its nickname, it conveys a true relationship with the brand and increases how much other consumers find the information to be authentic. Given the widespread proliferation of fake information online, it might be useful for brands to rely on the insight revealed herein to establish measures and develop communication strategies that convey and capture true brand attachment, which, in turn, may serve as a means by which more authentic brand messages could be communicated in the current digital era.
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921996277 - Mickey D's Has More Street Cred Than McDonald's: Consumer Brand Nickname Use Signals Information Authenticity
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921996277 for Mickey D's Has More Street Cred Than McDonald's: Consumer Brand Nickname Use Signals Information Authenticity by Zhe Zhang and Vanessa M. Patrick in Journal of Marketing
Footnotes 1 This research comprises part of the first author's dissertation work under the supervision of the second author.
2 Jonah Berger
3 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
4 The author(s) received no financial support for the research, authorship, and/or publication of this article.
5 Online supplement: https://doi.org/10.1177/0022242921996277
6 The original authenticity scale contained an additional item: "Alex's post is done fairly." We removed this item at the request of the review team given its low conceptual mapping on the underlying construct as well as its lack of fit with the other scale items.
References Aggarwal Akshay. (2016), " Enhancing Social Media Experience by Usage of User-Defined Nicknames as Additional Identifiers for Online Interaction ," Journal of Computer Sciences and Applications , 4 (1), 1 – 8.
Albert Noel , Thomson Matthew. (2018), " A Synthesis of the Consumer–Brand Relationship Domain: Using Text Mining to Track Research Streams, Describe Their Emotional Associations, and Identify Future Research Priorities ," Journal of the Association for Consumer Research , 3 (2), 130 – 46.
Anderson Eric T. , Simester Duncan I.. (2014), " Reviews Without a Purchase: Low Ratings, Loyal Customers, and Deception ," Journal of Marketing Research , 51 (3), 249 – 69.
Baxter Leslie A.. (1987), " Symbols of Relationship Identity in Relationship Cultures ," Journal of Social and Personal Relationship , 4 (3), 261 – 80.
Bechar-Israeli Haya. (1995), " From <Bonehead> to <cLoNehEad>: Nicknames, Play, and Identity on Internet Relay Chat ," Journal of Computer-Mediated Communication , 1 (2), https://doi.org/10.1111/j.1083-6101.1995.tb00325.x
Bell Robert A. , Healey Jonathan G.. (1992), " Idiomatic Communication and Interpersonal Solidarity in Friends' Relational Cultures ," Human Communication Research , 18 (3), 307 – 35.
7 Berger Jonah , Ward Morgan. (2010), " Subtle Signals of Inconspicuous Consumption ," Journal of Consumer Research , 37 (4), 555 – 69.
8 Berger Jonah , Humphreys Ashlee , Ludwig Stephan , Moe Wendy W. , Netzer Oded , Schweidel David A.. (2019), " Uniting the Tribes: Using Text for Marketing Insight ," Journal of Marketing , 84 (1), 1 – 25.
9 Berzack Antony. (2011), "Language Use of Successful Liars," master of science thesis , Cornell University Graduate School, Cornell University.
Beverland Michael B. , Farrelly Francis J.. (2010), " The Quest for Authenticity in Consumer: Consumers' Purposive Choice of Authentic Cues to Shape Experienced Outcomes ," Journal of Consumer Research , 36 (5), 838 – 56.
Bolin Aaron. (2005), " The Effects of First Name Stereotypes on Ratings of Job Applicants: Is There a Difference between Bill and William? " American Journal of Psychological Research , 1 (1), 11 – 20.
Bruess Carol J. S. , Pearson Judy C.. (1993), " 'Sweet Pea' and 'Pussy Cat': An Examination of Idiom Use and Marital Satisfaction over the Life Cycle ," Journal of Social and Personal Relationships , 10 (4), 609 – 15.
Campbell Anita. (2018), " What Is a Hashtag? And What Do You Do with Hashtags? " Small Business Trends (December 24), https://smallbiztrends.com/2013/08/what-is-a-hashtag.html
Campbell Margaret C. , Kirmani Amna. (2000), " Consumers' Use of Persuasion Knowledge: The Effects of Accessibility and Cognitive Capacity on Perceptions of an Influence Agent ," Journal of Consumer Research , 27 (1), 69 – 83.
Chang Richard S.. (2010), " Saving Chevrolet Means Sending 'Chevy' to Dump ," The New York Times (June 9), https://www.nytimes.com/2010/06/10/automobiles/10chevy.html
Dutton Denis. (2003), " Authenticity in Art ," in The Oxford Handbook of Aesthetics , Levinson Jerrold , ed. New York : Oxford University Press , 258 – 74.
Dwoskin Elizabeth , Timberg Craig. (2018), " How Merchants Use Facebook to Flood Amazon with Fake Reviews ," The Washington Post (April 23), https://www.washingtonpost.com/business/economy/how-merchants-secretly-use-facebook-to-flood-amazon-with-fake-reviews/2018/04/23/5dad1e30-4392-11e8-8569-26fda6b404c7%5fstory.html
Dzikus Lars , Smith Allison B. , Evans Jonathan. (2017), " What It Means to Be a 'Lady': Defending the 'Lady Vols' Nickname and Logo ," Sociology of Sport Journal , 34 (1), 35 – 45.
Finch Janet. (2008), " Naming Names: Kinship, Individuality and Personal Names ," Sociology , 42 (4), 709 – 25.
Flanagin Andrew J. , Metzger Miriam J. , Pure Rebekah , Markov Alex. (2011), " User-Generated Ratings and the Evaluation of Credibility and Product Quality in Ecommerce Transactions ," in Proceedings of the 44th Hawaii International Conference on System Sciences. Washington, DC : IEEE Computer Society , 1 – 10.
Fortado Bruce. (1998), " Interpreting Nicknames: A Micropolitical Portal ," Journal of Management Studies , 35 (1), 13 – 34.
Friestad Marian , Wright Peter. (1994), " The Persuasion Knowledge Model: How People Cope with Persuasion Attempts ," Journal of Consumer Research , 21 (1), 1 – 31.
Fuhrmeister Chris. (2017), " McDonald's Officially Changes Its Name to 'Golden Arches' in China ," Eater (October 26), https://www.eater.com/2017/10/26/16552772/mcdonalds-china-name-change-golden-arches
Gentner Dedre. (1989), " The Mechanisms of Analogical Learning ," in Similarity and Analogical Reasoning , Vosniadou Stella , Ortony Andrew , eds. New York : Cambridge University Press , 199 – 241.
Gerbaudo Paolo. (2012), Tweets and the Streets: Social Media and Contemporary Activism. London : Pluto Press.
Gregan-Paxton Jennifer , John Deborah Roedder. (1997), " Consumer Learning by Analogy: A Model of Internal Knowledge Transfer ," Journal of Consumer Research , 24 (3), 266 – 84.
Hayes Andrew F.. (2017), Introduction to Mediation, Moderation, and Conditional Process Analysis. New York : The Guilford Press.
Hovland Carl I. , Janis Irving L. , Kelley Harold H.. (1953), Communication and Persuasion: Psychological Studies of Opinion Change. New Haven, CT : Yale University Press.
Jun Youjung , Meng Rachel , Johar Gita Venkataramani. (2017), " Perceived Social Presence Reduces Fact-Checking ," PNAS , 114 (23), 5976 – 81.
Keller Kevin Lane. (2003), " Brand synthesis: The Multidimensionality of Brand Knowledge ," Journal of Consumer Research , 29 (4), 595 – 600.
Khamitov Mansur , Wang Xin (Shane) , Thomson Matthew. (2019), " How Well Do Consumer–Brand Relationships Drive Customer Brand Loyalty? Generalizations from a Meta-Analysis of Brand Relationship Elasticities ," Journal of Consumer Research , 46 (3), 435 – 59.
Kozinets Robert V. , Handelman Jay M.. (2004), " Adversaries of Consumption: Consumer Movements, Activism, and Ideology ," Journal of Consumer Research , 31 (3), 691 – 704.
Kronrod Ann , Lee Jeffrey K. , Gordeliy Ivan. (2017), " Detecting Fictitious Consumer Reviews: A Theory-Driven Approach Combining Automated Text Analysis and Experimental Design ," Working Paper No. 17-124 , Marketing Science Institute.
Kumar Ashish , Bezawada Ram , Rishika Rishika , Janakiraman Ramkumar , Kannan P. K.. (2016), " From Social to Sale: The Effects of Firm-Generated Content in Social Media on Customer Behavior ," Journal of Marketing , 80 (1), 7 – 25.
Landau Elizabeth. (2015), " Why Do We Use Pet Names in Relationships? " Scientific American (February 12), https://blogs.scientificamerican.com/mind-guest-blog/why-do-we-use-pet-names-in-relationships/
Malbon Justin. (2013), " Taking Fake Online Consumer Reviews Seriously ," Journal of Consumer Policy , 36 (2), 139 – 57.
Mathews Zane. (2018), " Grand Junction's Favorite Nicknames for Walmart ," (accessed December 28, 2020), https://kool1079.com/grand-junctions-favorite-nicknames-for-walmart/
Miranda Leticia. (2019), " Some Amazon Sellers are Paying $10,000 a Month to Trick Their Way to the Top ," BuzzFeed News (April 24), https://www.buzzfeednews.com/article/leticiamiranda/amazon-marketplace-sellers-black-hat-scams-search-rankings
Miyazaki Anthony D. , Ana Fernandez. (2001), " Consumer Perceptions of Privacy and Security Risks for Online Shopping ," Journal of Consumer Affairs , 35 (1), 27 – 44.
Moore Sarah G.. (2015), " Attitude Predictability and Helpfulness in Online Reviews: The Role of Explained Actions and Reactions ," Journal of Consumer Research , 42 (1), 30 – 44.
Moreau C. Page , Markman Arthur B. , Lehmann Donald R.. (2001), " 'What Is It?' Categorization Flexibility and Consumers' Responses to Really New Products ," Journal of Consumer Research , 27 (4), 489 – 98.
Newman George E.. (2019), " The Psychology of Authenticity ," Review of General Psychology , 23 (1), 8 – 18.
Newman George E. , Smith Rosanna K.. (2016), " Kinds of Authenticity ," Philosophy Compass , 11 (10), 609 – 18.
Newman Matthew L. , Pennebaker James W. , Berry Diane S. , Richards Jane M.. (2003), " Lying Words: Predicting Deception from Linguistic Styles ," Personality and Social Psychology Bulletin , 29 (5), 665 – 75.
Nyambi Oliver. (2018), " Characterization of Unconventional Exor-Anthroponyms in Sport: The Spectacle and Aesthetics of Player Nicknames in Zimbabwean Soccer ," African Identities , 16 (3), 260 – 74.
Park C. Whan , MacInnis Deborah J. , Priester Joseph , Eisingerich Andreas B. , Iacobucci Dawn. (2010), " Brand Attachment and Brand Attitude Strength: Conceptual and Empirical Differentiation of Two Critical Brand Equity Drivers ," Journal of Marketing , 74 (6), 1 – 17.
Reyes Angela. (2013), " Corporations Are People: Emblematic Scales of Brand Personification Among Asian American Youth ," Language in Society , 42 (2), 163 – 85.
Roland Driadonna. (2016), " 'Hip Hop Honors' Sparks Debate: Is It the Beehive or the Beyhive? " Revolt (July 12), https://revolt.tv/stories/2016/07/12/hip-hop-honors-sparked-debate-beehive-beyhive-f341c32384
Sela Aner , Christian Wheeler S. , Sarial-Abi Gülen. (2012), " 'We' Are Not the Same as 'You and I': Causal Effects of Minor Language Variations on Consumers' Attitudes Toward Brands ," Journal of Consumer Research , 39 (3), 644 – 61.
Seppälä Janne. (2018), " Car Nicknames and Their Relation to the Brand ," in Language, Media and Economy in Virtual and Real Life: New Perspectives , Cotticelli Paola , Rizza Alfredo , eds. Newcastle upon Tyne, UK : Cambridge Scholars Publishing , 241.
Smith Shireen. (2010), " How Safe Is Coca Cola's Trade Mark Coke? " Azrights (June 28), https://azrights.com/media/news-and-media/blog/intellectual-property/2010/06/how-safe-is-coca-colas-trade-mark-coke/
Stevens Fakheera. (2018), " Report Finds Companies Paying for Positive Reviews ," Consumers' Research (October 30), http://consumersresearch.org/report-finds-companies-paying-for-positive-reviews/
Tellis Gerard J. , MacInnis Deborah J. , Tirunillai Seshadri , Zhang Yanwei. (2019), " What Drives Virality (Sharing) of Online Digital Content? The Critical Role of Information, Emotion, and Brand Prominence ," Journal of Marketing , 83 (4), 1 – 20.
Vescovi Valentina. (2009), " In Argentina, Pepsi Becomes 'Pecsi,' " AdAge (July 15), https://adage.com/article/global-news/pepsi-pecsi-argentina/137946
Zhang Zhe , Patrick Vanessa M.. (2018), " Call Me Rollie! The Role of Brand Nicknames in Shaping Consumer–Brand Relationship ," Journal of the Association for Consumer Research , 3 (2), 147 – 62.
~~~~~~~~
By Zhe Zhang and Vanessa M. Patrick
Reported by Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 93- Minimum Payments Alter Debt Repayment Strategies Across Multiple Cards. By: Hirshman, Samuel D.; Sussman, Abigail B. Journal of Marketing. Mar2022, Vol. 86 Issue 2, p48-65. 18p. 2 Black and White Photographs, 1 Diagram, 2 Charts, 3 Graphs. DOI: 10.1177/00222429211047237.
- Database:
- Business Source Complete
Minimum Payments Alter Debt Repayment Strategies Across Multiple Cards
U.S. households currently hold $770 billion in credit card debt, often managing repayments across multiple accounts. The authors investigate how minimum payment requirements (i.e., the requirement to allocate at least some money to each account with a balance) alter consumers' allocation strategies across multiple accounts. Across four experiments, they find that minimum payment requirements cause consumers to increase dispersion (i.e., spread their repayments more evenly) across accounts. The authors term this change in strategy "the dispersion effect of minimum payments" and provide evidence that it can be costly for consumers. They find that the effect is partially driven by the tendency for consumers to interpret minimum payment requirements as recommendations to pay more than the minimum amount. While the presence of the minimum payment requirement is unlikely to change, the authors propose that marketers and policy makers can influence the effects of minimum payments on dispersion by altering the way that information is displayed to consumers. Specifically, they investigate five distinct information displays and find that choice of display can either exaggerate or minimize dispersion and corresponding costs. They discuss implications for consumers, policy makers, and firms, with a particular focus on ways to improve consumer financial well-being.
Keywords: credit cards; choice architecture; debt; financial decision making; financial well-being
Imagine that you have accumulated debt across several different credit cards and are now allocating your monthly budget toward repaying this debt. As you look at your bills, each debt has a different total balance, interest rate, and minimum payment. How do you decide how much to allocate to each one? You might focus on balance amounts, for example, paying the smallest balance first to feel like you are making progress. Alternatively, you might focus on interest rates, paying the highest-interest card first to minimize total interest paid. Regardless of how you choose to prioritize your debts, you will likely consider the minimum payment amounts across all cards. You might plan to pay at least the minimum to avoid fees, interest rate increases, and credit score implications associated with failing to make the minimum payments. You might even think the presence of the minimum payments suggests you should pay more than the minimum.
U.S. households currently hold $770 billion in credit card debt ([13]). Rather than being consolidated into a single monthly payment on one card, the average American household with at least one credit card must manage decisions across an average of four accounts ([ 8]), and 91.7% of revolving balances are held by people with two or more cards ([16]). Relying solely on interest rates and minimum payment requirements can lead consumers to a strategy that would minimize total interest costs. However, it is not clear that consumers use the cost-minimizing strategy and may instead use other strategies such as heuristics based on balance amounts (e.g., [ 3]; [16]).
Minimum payment requirements are a central element of the credit card statement and, more broadly, of the financial system surrounding debt repayment. Specifically, credit card companies require borrowers to make a minimum payment each month. Missing these payments is used to classify accounts as in default and leads to late fees and credit score penalties (see [21]). Beyond avoiding penalties, there may be benefits to consumers from paying some amount every month, though paying exactly the minimum on a given credit card tends to reduce that credit card's debt very slowly ([32]).
We propose that minimum payments fundamentally alter allocation strategies across several credit card debt accounts. In field data, consumers' repayments are more dispersed (i.e., spread more evenly across accounts) than a strategy that minimizes interest costs or other known heuristic strategies would suggest ([16]; [23]; see also Web Appendix A, Study F). In the current research, we use controlled experiments to isolate the effect of minimum payments on dispersion. We find that the presence of minimum payments on all accounts increases dispersion relative to a control with no minimum present, which we term "the dispersion effect of minimum payments." This strategy change tends to redirect payments from the highest-interest cards to lower-interest cards, thus increasing overall interest costs. The dispersion effect occurs in addition to anchoring effects previously associated with minimum payments on single cards (e.g., [35]). We provide evidence that a perceived recommendation to pay more than the minimum drives the effect. Finally, we examine the role that alternative choice architectures (e.g., [33]) can play in altering dispersion and corresponding interest costs in the presence of minimum payments.
Policy makers have used a variety of behavioral science tools to influence consumers' repayments, particularly through regulations affecting the credit card statement. For example, the CARD Act of 2009 required credit card companies to provide additional information to consumers including the total interest costs associated with paying only the minimum payment, the time required to pay off the full credit card balance if paying only the minimum, and the repayment amount required to pay off the full credit card balance in three years. One goal of the CARD Act disclosures was to nudge consumers to make larger repayments, and these small changes to the information environment did affect repayment decisions, though to a somewhat limited extent ([ 1]; [ 2]).
The way credit card bills display information can impact consumers' understanding and repayments. [34] find that consumers have inaccurate perceptions of the growth of debts over time due to a misunderstanding of compound interest. The additional information mandated by the CARD Act corrects some but not all of consumers' misunderstanding of the time it takes to repay debt.
Focusing on the provision of the repayment amount required to pay off the full credit card balance in three years, [32] finds that the provision of that three-year cost information has the desired effect of increasing repayments above the minimum. However, the information about time required to pay off the full credit card balance if paying only the minimum has an unintended consequence of moving repayments below the three-year repayment amount.
The prevalence of online repayments has diminished the likelihood that consumers even see CARD Act disclosures ([ 8]). This context motivated [33] to investigate the way online payment modules with default options are constructed and the consequences of these displays for repayment decisions. They find that active-choice displays, which consist of distinct salient options to pay the minimum and the total debt amount, increase the likelihood that consumers will repay a single card in full.
Recent research has also examined how the presence of minimum payments on credit card statements influences the amount allocated toward debt repayment. To understand this question, researchers typically randomly assign participants to a single debt that either has a minimum payment or does not, and they ask participants how much they would allocate toward debt (e.g., [35]). Consumers faced with minimums in this context tend to repay less than those without them (e.g., [20]; [30]; [35]). Subsequent work has shown similar effects in the field, with consumers paying at or just above the minimum when the minimum required payments change (e.g., [24]).
One explanation for these results is that consumers anchor on the minimums and, as a result, pay just above the minimum ([30]; [35]). Probing this anchoring effect further, [20] propose that the salient numbers on credit card statements operate as a recommendation or appropriate payment amount, akin to a default amount serving as a recommendation in other contexts (e.g., [25]). In other words, consumers may take the required minimum values as the amount they should pay and adjust their repayments from that implied recommendation.
Each of these demonstrations examining the effect of credit card statements on payment decisions investigates the decision of how much money to repay toward a single card (e.g., [20]; [24]; [30]; [35]). If a person only has one outstanding debt account, then making lower payments toward this account will increase overall interest costs. However, when determining the most efficient allocations across multiple accounts, the strategy for minimizing interest costs changes. While it is possible that other fees can drive costs higher in specific situations, we focus on interest as a main driver of costs associated with credit card debt. Assuming a given amount of money is available to repay debt across cards, the strategy that leads to the lowest amount of accrued interest over time is to pay the minimum required amount on each card first and then to use the discretionary funds (i.e., repayments in excess of the minimum payments on all cards) to allocate any remaining money to the debt associated with the highest-interest-rate card. If, after doing so, funds remain, the consumer would repay the debt associated with the second-highest-interest-rate card, and so on. As a result, paying only the minimum on lower-interest-rate cards can be consistent with the cost-minimizing strategy. Studies using only one card cannot determine whether consumers are responding to minimum payments efficiently when considering repayment strategies across a consumers' portfolio of cards. A separate line of research examines the question of whether consumers use the optimal strategy across their portfolio of cards and, more generally, aims to better understand what strategy consumers do use. These articles tend not to examine the relevance of the minimum payment to repayment strategy decisions.
A variety of studies both in the lab (e.g., [ 3]; [ 5]; [ 6]) and in the field ([16]) have shown that consumers do not use the cost-minimizing strategy to make debt repayment decisions. Using field data from the United Kingdom, [16] argue that consumers are balance matching—that is, they behave as if they are paying in proportion to their account balances.[ 5] For example, a consumer who holds a $1,000 balance on one card and a $500 balance on another would allocate $200 to the first debt and $100 to the second if they had $300 available to repay. The authors argue that this heuristic is consistent with other matching heuristics shown in humans and animals (e.g., probability matching; [18]; [19]). Notably, this heuristic predicts that people should make their largest repayment to the debt with the largest balance.
In contrast, both academic researchers and financial advice gurus such as Dave Ramsey have documented positive motivational effects of paying off credit card accounts in full, typically by paying off the debt with the smallest balance. Under this strategy, described as "debt account aversion," consumers repay more money toward their debts with the smallest balances so they can close accounts and feel a sense of progress ([ 3]; [ 5]; [ 6]). For example, sorting multiple tasks from smallest to largest can reduce the time it takes to complete the tasks ([ 7]). Providing further evidence, consumers in a debt repayment program were more likely to remain in the program when they had a debt account closed by being fully repaid, even though the debt repayments were decided by the program ([15]). The motivational benefits of focusing on smaller debt balances can manifest, even in cases where consumers cannot pay debts off in full, as long as consumers make a repayment toward the card with the lowest balance ([23]) or complete a subgoal by repaying a purchase in full ([11]). As a result, there may be ways for consumers to improve their outcomes by deviating from the interest-cost-minimizing strategy.
Finally, drawing on literature from investment savings, consumers may rely on the 1/N heuristic as a way to simplify their decision-making process ([ 4]). Under this strategy, consumers would divide their budgets evenly across debt accounts. More recent work shows that consumers may allocate evenly only across considered options ([27]). Field evidence suggests that consumers may also use the 1/N heuristic when choosing how to allocate payments across credit cards ([16]).
Substantial attention has been devoted to documenting how minimum payments affect debt repayment decisions in single-card settings and, separately, to debt repayment decisions across multiple cards. However, an important gap remains in understanding whether and how minimum payments alter debt repayment strategies across multiple cards. In multicard settings, anchoring on the minimums could help consumers if it leads them to reduce their repayments to lower-interest debts while maintaining or increasing payments to higher-interest debts. Alternatively, it could hurt consumers if it leads them to reduce payments to higher-interest-rate cards. This could occur either by reducing overall allocations toward debt repayment or by causing consumers to shift funds from higher- to lower-interest debts. This article examines the effect of minimum payment requirements on repayment strategies when borrowers have multiple credit card debts. Specifically, we examine the relationship between minimum payments and dispersion (i.e., spreading payments across accounts) as well as consequences for interest costs.
Our focus on dispersion is motivated by patterns of repayments in field data. The cost-minimizing repayment strategy in any given payment cycle typically requires concentrating repayments on only one or a small number of consumers' highest-interest debts. By contrast, field data suggests that consumers' repayments are too dispersed relative to the cost-minimizing approach ([16]; [23]; see also Web Appendix A, Study F). Given the ubiquity of minimum payments in the context of credit card debt, each of the field investigations described previously examines cases where the minimum payment requirement is present, with no comparison to a no-minimum control. In the current research, we hypothesize that these minimum payment requirements may contribute to the observed dispersion in repayments.
- H1: Consumers make payments that are more dispersed across debts when repaying credit card accounts that all have (vs. do not have) minimum payment requirements.
More concentrated repayment strategies do not inherently lead consumers to pay lower interest costs, though the cost-minimizing strategy described previously is typically concentrated. Instead, terms of the specific accounts on which consumers choose to concentrate their repayments will affect the amount of interest paid. For example, if consumers' starting point was making the largest payments toward an account that happened to have the lowest interest (e.g., in the case that this was their smallest debt amount), then increasing dispersion would likely reduce interest costs overall by directing money toward higher-interest accounts. While consumers do not have a strong understanding of how interest compounds (e.g., [34]), there is evidence that consumers do attend to interest to some degree ([ 1]). This attention to interest rates suggests that consumers direct repayments toward higher-interest accounts. Some, but potentially insufficient, attention to interest rates is also consistent with results from a pilot study we conducted examining intended repayment strategies (see Web Appendix A, Study G). To the extent that consumers begin by allocating more funds toward the highest-interest-rate cards (vs. lower interest-rate cards), increasing dispersion (e.g., through the introduction of the minimum payment) should shift repayments away from the highest-interest-rate accounts toward those with lower interest rates. Consequently, we predict that minimum payments will increase consumers' total interest costs. Correspondingly,
- H2a: The presence of the minimum payment on all cards decreases repayments to the highest-interest-rate debt when compared with a no-minimum control.
- H2b: The presence of the minimum payment on all cards increases interest costs when compared with a no-minimum control.
While our hypotheses directly address situations with and without minimum payments, having minimum payments on only some accounts may lead to similar patterns.
Why might minimum payments lead to an increase in dispersion? In the absence of minimum payments, consumers would likely allocate available funds based on one of the previously mentioned heuristics. Whether focused on interest or amounts, these repayment strategies tend to be concentrated on a small number of accounts. To either minimize interest costs or maximize the motivation for paying off debt ([ 3]; [23]), it is helpful to repay certain accounts in full while repaying small or zero amounts to others. Consistent with this intuition, consumers describing their intended repayment strategies typically include plans to pay all or most of their available funds to a single debt (for a study examining lay beliefs about debt repayment strategies, see Web Appendix A, Study G).
Once minimum payments are introduced, however, we propose that consumers will perceive a recommendation to repay more than the minimum toward every account. While the primary focus of the literature on minimum payments has been on anchoring (e.g., [35]), more recent work has emphasized how a variety of numbers presented to customers on credit card statements can be perceived as recommended amounts (e.g., [20]; [26]; [33]). In multiple-card settings, consumers may act in accordance with a recommendation to repay more than the minimum on each account. As a result, accounts that would not have received an allocation at all without a required minimum payment may instead receive an allocation that is even greater than the minimum. This will lead allocations in the minimum-payment condition to be more dispersed. These repayments may appear similar in nature to a 1/N repayment strategy of splitting evenly but will not necessarily be divided into equal amounts. Further, we propose that the motivation to pay something to every card stems at least in part from a perceived recommendation. More formally,
- H3: A perceived recommendation to pay more than the minimum to every card mediates the relationship between the presence of the minimum requirement on all cards and payment dispersion.
We investigate these hypotheses across four experiments reported in the main text. Seven additional studies reported in the Web Appendix corroborate these key findings and provide evidence of robustness to variations in the experimental design. In Experiment 1, using a nationally representative sample and an incentive-compatible design, we find that participants' allocations in the minimum-payment condition are more dispersed than in the no-minimum-payment control (H1). We also find that the presence of minimum payments leads participants to repay less to the highest-interest-rate debt (H2a) and to incur larger interest costs (H2b). In Experiment 2, we allow consumers to choose their budgets for debt repayment and replicate these findings (H1, H2a, and H2b). Further, consistent with prior literature, we find that minimum payments decrease the budget allocated to debt repayment (e.g., [35]). In Experiment 3, we provide evidence that consumers take minimum payment requirements as recommendations to pay more than that amount toward each card. Further, these beliefs mediate the relationship between the minimum-payment condition and payment dispersion (H1, H2a, H2b, and H3). Finally, in Experiment 4, we aim to identify potential policy interventions by examining how five different information presentations modeled on real-world debt repayment environments influence the dispersion effect. We replicate findings of the dispersion effect of minimum payments (H1) and associated interest costs (H2a, H2b) using the same presentation as in Experiments 1–3. In addition, we find that a presentation modeled on the paper statements currently in use leads to worse outcomes for consumers relative to all other presentations. By contrast, a presentation modeled on newer online interfaces that use active choice alters perceived recommendations and reduces dispersion (H3). The exact materials and data for all studies are available on OSF at https://osf.io/xuf59/.
This article makes several contributions. First, we contribute to the literature on how consumers repay credit card debt (e.g., [39]; [ 5]; [16]) by examining the impact of minimum payments in situations where consumers have multiple credit cards. In addition, we provide a deeper understanding of the psychological mechanisms underlying heuristics previously identified using field data (e.g., [16]) by using lab experiments to examine the causal impact of minimum payments on dispersion. Second, we contribute to the literature on consumer inferences from choice architecture ([20]; [25]; [33]) by documenting dispersion and corresponding interest costs resulting from consumer inferences of recommendations from minimum payment requirements. Finally, we contribute to the literature on using choice architecture to improve consumers' financial outcomes ([28]; [38]) by identifying how different choice environments can exaggerate or minimize the excess dispersion and interest costs associated with minimum payment requirements. Consequently, our findings have practical implications for providing credit card companies or third-party financial management apps (e.g., Tally, Mint) additional strategies for helping consumers repay their debts.
When making debt repayment decisions, most consumers are choosing how to allocate payments across multiple cards. While it is difficult to find variation in the presence or absence of minimum payments in the field because they are almost universally required, examining repayments in the lab enables us both to assess the causal impacts of minimum payments and simultaneously to gain a better understanding of the underlying psychological drivers of payment strategies. In a nationally representative population, we use a debt management game to test participants' strategic responses to minimum payments.
Four hundred thirteen participants from a market research panel aggregator operated by CloudResearch completed the study. Of these, 50.7% were female with a median age of 48 years, a modal education level of a "Some college (no degree)," and median income of $30,000–$39,999. The sample was selected to be approximately nationally representative of U.S. adults on age, gender, and income. Seventy-nine percent reported having at least one credit card, with a median of two cards per participant. The sample was limited to U.S. participants.
Participants played a three-round debt management game modeled on a task from [ 3] but modified to mimic the average U.S. consumer's debt more closely. Participants were provided with information on six debt accounts including interest rates and amounts. We drew interest rates at random from a CFPB database of national credit card terms and drew debt amounts from a normal distribution designed to add up to $16,000, approximately the amount of credit card debt for the average indebted American household ([12]). In each round, participants received a budget of $3,000 dedicated to debt repayment, and they selected the amount they wanted to allocate to each debt. Their allocations in each round were forced to equal their budget (for participants' view of the task, see Figure 1).
Graph: Figure 1. Example participant screen in Experiment 1.
Participants were randomly assigned to either a minimum-payment condition or a no-minimum-payment control condition. All participants were instructed that their goal in the task was to reduce their debt as much as possible. After reading the instructions, all participants answered one comprehension check question about their goal in the task. In addition, participants in the minimum-payment condition answered a question about the size of the minimum payment fee. If a participant answered a comprehension check question incorrectly the first time, the question was shown again with the relevant section of the instructions.
All participants saw a table indicating the balance and interest rates for each of six credit card accounts. Participants in the minimum-payment condition saw an additional column in the table with the minimum payment amounts. The minimum payments were set at 2% of the total debt amount on all accounts, consistent with the typical range of 1%–4% in the field ([24]). Participants faced a $25 fee for each minimum that they failed to pay, similar to the initial "safe harbor" late fee set by the [ 8]. After each round, the task was updated to reflect the decisions participants had made in the previous round. Participants were told that they would be paid a bonus that ranged from $0–$1 based on their performance in the task.[ 6]
We excluded participants who failed to answer the comprehension check questions twice (N = 13) and those who allocated more than they owed to any debt in more than one round (N = 26) because these repeated errors suggest inattention to the task. These exclusions did not differ significantly by condition (no minimum: 10% vs. minimum: 9%; z = .33, p= .741). Analyses with all participants are included in the Web Appendix, and results do not substantially change (for robustness to exclusion criteria, see Web Appendix B, Table W2).
As a measure of dispersion, we build on the concentration metric described in [23], defining dispersion as one minus concentration:
Graph
where xikt is the allocation made by participant i to debt k in round t, is the mean allocation made by participant i in round t, and Ait is the number of accounts with nonzero balances for participant i in round t. Intuitively, the metric captures the ratio of the variance in repayments to the mean of the repayments, with a normalization to create a measure bounded between zero and one. This measure of dispersion has several advantages: it builds on information-theoretic concepts, is continuous over the range of zero to one, and allows for meaningful comparisons between participants with different numbers of accounts.
To account for mechanical differences between conditions on this measure that may bias it in favor of our hypothesis, we adjust the dispersion metric of participants in the minimum-payment condition who use the cost-minimizing strategy to match the dispersion of cost-minimizing repayments in the control. Specifically, participants in the minimum-payment condition who repay using the cost-minimizing strategy in round one are adjusted to have the same dispersion as participants in the control who use the cost-minimizing strategy. In rounds two and three, participants in the minimum-payment condition who repay using the cost-minimizing strategy in all three rounds are adjusted to have the same dispersion as control participants who use the cost-minimizing strategy in all three rounds (for additional details on this adjustment, see Web Appendix C). Because the measure is bounded between zero and one, we analyze dispersion using a fractional regression in a panel over the three rounds, controlling for the round of the task, with heteroskedasticity-robust standard errors clustered at the subject level ([ 9]).
Ideally, our measure of dispersion would produce similar results for people who try to implement similar repayment strategies, would be implemented the same way for all participants regardless of condition or strategy used, and would not require additional exclusions. However, many of these factors trade off against one another. The adjusted measure that we present in the main text has the benefit of producing similar results for similar repayment strategies across conditions. In addition, it allows us to interpret allocations of people who do not make the minimum payments, enabling us to include them in the analysis. However, it requires an ad hoc adjustment for those using the cost-minimizing strategy. As an alternative, the unadjusted dispersion measure can be applied uniformly, but it is biased in favor of H1 for some strategies, most importantly the cost-minimizing strategy. In contrast, a version of this metric that focuses on only discretionary repayments in the minimum-payment condition (i.e., by subtracting out the minimum payments in the minimum-payment condition prior to computing the dispersion metric) is substantially biased in the opposite direction for the cost-minimizing strategy as well as for any person repaying more than the minimums. In addition, when looking at only discretionary repayments, it is not clear how to use data from participants in the minimum-payment condition who fail to make the minimum payments, potentially inducing a sampling bias.
We present the adjusted dispersion metric throughout, and Web Appendix B, Table W4, reports the unadjusted metric and the discretionary metric excluding people who fail to make the minimum payments. Importantly, while each of these approaches has advantages and disadvantages, they produce qualitatively similar results. This converging evidence suggests that differences in dispersion as a function of the minimum payment requirement are robust to the specific operationalization of dispersion.
Connecting dispersion to interest costs, we focus on two measures. First, we examine the amount allocated to the highest-interest debt by condition using a linear regression. This analysis controls for the round of the task and uses heteroskedasticity-robust standard errors clustered at the subject level. Second, to capture the overall costs associated with the repayment decisions in each condition, we examine the natural log of interest paid per round by condition using a linear regression, controlling for the round of the task, with heteroskedasticity-robust standard errors clustered at the subject level. We use the natural log because this measure could be positively skewed depending on the strategies consumers use to repay debts, with many consumers clustered around low amounts but some paying large amounts of interest. We do not include the penalty fees in this measure as there are no penalty fees in the control condition. However, participants who fail to make minimum payments may face larger debts in subsequent rounds as a result of the fees.[ 7] In the following results, we report both the average amount in natural logs, which is the measure for our statistical comparison, as well as the dollar levels, in brackets, for interpretability.
Participants in the minimum-payment condition allocated their repayments in a more dispersed way than those in the control condition (Mcontrol = .73 vs. Mmin = .79; B = .31, 95% confidence interval [CI] = [.038,.57], t(373) = 2.24, p = .025), consistent with the hypothesized dispersion effect (see Figure 2). Turning to interest, we find that participants in the minimum-payment condition paid significantly less to the highest-interest-rate debt (Mcontrol = $1,085, Mmin = $777; B = −307.75, 95% CI = [−409.80, −205.70], t(373) = 5.91, p < .001). These repayment patterns translated to significantly higher interest paid per round in the minimum-payment condition (Mmin = 7.00 [$1,128]) than in the control condition (Mcontrol = 6.96 [$1,084]; comparison of natural log amounts: B = .044, 95% CI = [.024,.064], t(373) = 4.36, p < .001).
Graph: Figure 2. Main results from Experiment 1.
Experiment 1 demonstrates the effect of minimum payments on dispersion and interest across multiple cards. Minimum payments shifted participants' debt repayment strategies when allocating a fixed sum across multiple accounts, leading to higher dispersion. In addition, minimum payments decreased the money allocated to the highest-interest-rate debt and increased interest costs. These results are robust to accounting for financial literacy, number of credit cards, and other demographic controls (see Web Appendix B, Table W3).
In Experiment 1, participants were given a fixed budget to allocate across six accounts and were required to allocate their full budget to debt repayment in each round. Consequently, participants could not pay exactly the minimum requirements for all debts or otherwise vary the total amount allocated to debt repayment. Prior literature on debt repayment has identified consumers' tendency to pay at or just above the minimum payment in a single-card setting (e.g., [35]). Further, it has examined effects of paying off accounts in full on motivation to repay, altering decisions of how much money to allocate to debt repayment ([ 3]; [ 5]; [ 7]; [23]). The design of the current study allows us to build on this prior work by giving participants the opportunity to save some of their budget. This design has the additional benefit that it is more consistent with the decision people face in the real world, which requires determining both how much money to allocate to debt repayments and how to distribute this money across debt accounts.
Four hundred two participants completed the study on Amazon's Mechanical Turk (MTurk). Fifty-four percent of our participants were female, with a median age of 33 years, median income in the range of $50,000–$59,000, and modal education of a bachelor's degree. Eighty-four percent of our participants report having at least one credit card, with a median of two cards per participant.
Participants played a debt game similar to that from Experiment 1. However, in this version of the debt game, participants had the option to allocate money to a savings account if they desired. They were not given information about interest earned in this savings account. We instructed participants to allocate funds as they would in their everyday life. This instruction was included to allow us to gain a realistic picture of consumer behavior and avoid having participants treat this as an optimization problem, which could bias them against using the savings account. Participants were randomly assigned either to a minimum-payment condition or a no-minimum-payment control condition as in Experiment 1. In addition to the measures from the prior experiment, we also examined the amount and likelihood of savings across conditions, as well as how dispersion in prior rounds affected subsequent savings decisions. This analysis allowed for a better understanding of how the dispersion effect relates to literature on concentration and motivation ([23]).
Prior to examining the data, we excluded participants (N = 9) who allocated more than they owed to any debt in more than one round.[ 8] Consistent with results from Experiment 1, participants in the minimum-payment condition (Mmin = .72) allocated their chosen budgets to debt repayment in a significantly more dispersed fashion than the no-minimum-payment control (Mcontrol = .65; B = .33, 95% CI = [.094,.56], t(391) = 2.74, p = .006). The amount allocated toward the highest-interest debt was significantly lower in the minimum-payment condition than in the control condition (Mcontrol = $1,179, Mmin = $888; B = −291.20, 95%CI = [−403.20, −179.20], t(392) = 5.10, p < .001). These repayment patterns translated to significantly higher interest paid per round in the minimum-payment condition than in the control (Mcontrol = 6.99 [$1,119], Mmin = 7.06 [$1,206]; B = .076, 95% CI = [.042,.111], t(392) = 4.37, p < .001).
Prior work (e.g., [35]) suggests that people will allocate less money toward debt repayment when minimum payments are present. We replicate this finding in a multiple-card setting by examining the log of the amount allocated to debt in each round (i.e., the budget of $3,000 less any amount saved). We impute $0 allocated to debt to $1 because the log(0) is undefined, but this occurs in only three observations. Participants in the minimum-payment condition (Mmin = 7.79) allocated significantly less to debt than those in the no-minimum-payment control (Mcontrol = 7.91; B = −.12, 95% CI = [−.215, −.025], t(392) = 2.48, p = .014). Because their overall budget was fixed, this implies participants in the minimum-payment condition saved more. Participants in the minimum-payment condition were also significantly more likely to allocate any money to the savings account (Mcontrol = 47%, Mmin = 62%; B = .61, 95% CI = [.25,.97], t(392) = 3.31, p = .001). Correspondingly, we see increased dispersion in the minimum-payment condition when including the savings account in our dispersion metric (Mcontrol = .66, Mmin = .71; B = .23, 95% CI = [.02,.45], t(392) = 2.10, p = .036).
To better understand the relationship between amount allocated to debt repayment and dispersion, we regressed the log of the amount allocated to debt in round t on condition, dispersion in round t − 1, and log of the amount allocated to debt in round t − 1 with a round control. Unsurprisingly, prior-round allocation to debt predicted current-round allocation to debt (B = .751, 95% CI = [.543,.959], t(391) = 7.07, p < .001). In addition, prior-round dispersion was negatively correlated with future allocation to debt (B = −.07, 95% CI = [−.126, −.015], t(391) = 2.51, p = .013), suggesting that participants with less dispersed strategies were more focused on repaying debt. Controlling for these two effects, there was no effect of condition on allocation to debt (B = −.007, 95% CI = [−.026,.013], t(391) = .68, p = .495). This suggests that the differences in dispersion may contribute to the increased use of the savings option and the reduction in money allocated to debt in the minimum-payment condition. The positive relationship between dispersion and savings is consistent with the negative motivational effects of dispersed debt repayments. In other words, more dispersed prior-round repayments were associated with less focus on debt repayment in the subsequent round ([23]).
Prior literature on minimum payments in single-card settings finds that the presence of minimum payments reduces the amount repaid toward debt. We replicate that finding and also show that minimum payments lead consumers to allocate their remaining budget in a more dispersed way. Thus, minimum payments induce two separate costs across multiple accounts. First, they decrease the amount of money allocated to debt repayment. This pattern is consistent with an increase in "co-holding" in which consumers simultaneously hold money in low-interest bearing savings accounts and high-interest debt accounts ([17]; [36]). Second, they lead borrowers to spread their repayments more evenly across accounts, consistent with the proposed dispersion effect. Our findings further suggest that the more dispersed repayments may themselves be associated with reduced money allocated to debt repayment to the extent that they reduce motivation to repay debt.
In the current experiment, we aim to develop a better understanding of why the dispersion effect occurs. Specifically, we examine the inferences consumers draw from the minimum payments. Prior work suggests that minimum payments and other numbers included on debt statements can be perceived as a recommendation for how much to pay ([20]; [33]). Consequently, we test whether consumers in the minimum-payment condition perceive a recommendation to pay more than the minimum amount. Perceiving a recommendation to pay more than the minimum amount on every card could lead consumers to increase allocations from the minimum, corresponding to increased dispersion.
In addition, the current experiment includes two new elements to further explore the source and the robustness of the dispersion effect. To examine how much of the effect is driven by the mandatory nature of the required minimum payment, we add a suggested-minimum-payment condition. In this condition, participants are told a suggested amount for each card, which is the same 2% amount as the required minimum payment, but there is no penalty for failing to pay it. We hypothesize that this condition will be perceived more as a recommendation to pay exactly the suggested amount than the required-minimum condition.
In addition, we test the robustness of the dispersion effect to the distribution of debt amounts. Because debt amounts tend to be skewed ([23]), we want to examine how a more skewed distribution of debt amounts affects dispersion. We do this by drawing an additional set of debt amounts from a log-normal distribution with similar total debt to the prior studies, leading to a more skewed distribution of debt balances (see Figure 3). The study was preregistered at aspredicted.org.[ 9]
Graph: Figure 3. Debt values displayed in the high-skew condition.
Twelve hundred seven participants completed the study. Forty-six percent were female, with a median age of 37 years, a median income of $50,00–$59,999, and a modal education level of a bachelor's degree. Eighty-six percent of participants reported having at least one credit card, with a median of two credit cards per participant.
Participants completed a debt management game similar to that in Experiment 1. They were randomly assigned to a no-minimum control, a required-minimum-payment, or a suggested-minimum-payment condition. The no-minimum control and required-minimum-payment conditions were the same as in Experiment 1. Participants in the suggested-payment condition saw the same 2% minimum payment as those in the required-minimum condition, but these payments were not required. The instructions read, "On each page you will see a suggested minimum payment amount. There is no consequence for not making these payments, but they are suggested." In addition, participants were randomly assigned to see debt amounts drawn from a normal distribution (normal skew) as in previous studies or a log-normal distribution (high skew). Thus, the experiment had a 3 (minimum: control, required, suggested) × 2 (skew: normal vs. high) design.
After completing the debt management task, participants responded to questions about recommendations they perceived from the task. The main mediation question was "The instructions provided a recommendation to pay MORE THAN [X] to every card."[10] Participants responded on a seven-point scale from "strongly disagree" ( 1) to "strongly agree" ( 7). In the control condition, X was replaced with "$0"; in the suggested minimum condition, it was replaced with "THE SUGGESTED MINIMUM," and in the required minimum condition, it was replaced with "THE REQUIRED MINIMUM." We maintained capitalization in the survey to emphasize differences across questions. The experiment included two ancillary questions which are reported and discussed in the Web Appendix.
Note that our questions focused on recommendations inferred from the instructions, which contrasts somewhat from prior work on defaults as recommendations asking about inferences from policy makers' beliefs (e.g., [25]). Given widely known policy requiring minimum payments on credit cards in the United States, we directed the current question toward the debt game's instructions so that it would be easily interpretable in all conditions, even those (i.e., suggested minimum and control conditions) that deviate from known policy.
We excluded participants (N = 5) who failed to answer the comprehension check questions twice. We also excluded participants who allocated more than they owed to any debt in more than one round (N = 49).[11]
For our main analyses, in addition to the dummy variables for each round we included in prior studies, we also included a dummy variable for whether the debt amounts were drawn from a high-skew distribution. Examining the two different distributions of debt amounts with an interaction shows qualitatively similar results, with a somewhat larger dispersion effect of minimum payments in the high-skew condition. The Web Appendix contains additional analyses detailing the robustness of our results to different specifications.
Payments in the required-minimum condition were significantly more dispersed than those in the no-minimum control (Mcontrol = .58, Mmin = .66; B = .37, 95% CI = [.22,.51], t( 1,152) = 4.88, p < .001). The suggested minimum (Msuggested = .63) fell between the required-minimum and the control, with payments significantly more dispersed than in the control condition (B = .20, 95% CI = [.052,.34], t( 1,152) = −2.66, p = .008) and significantly less dispersed than in the required-minimum condition (B = −.17, 95% CI = [−.32, −.015], t( 1,152) = 2.16, p = .031).
Participants allocated less to the highest-interest-rate debt in both the required-minimum condition (Mcontrol = $1,239, Mmin = $1,029; B = −211.81, 95%CI = [−276.71, −146.92], t(1145) = 6.40, p < .001) and the suggested-minimum condition (Msuggested = $1,144; B = −96.33, 95% CI = [−170.11, −22.55], t( 1,152) = 2.56, p = .01) when compared with the control. Those in the required-minimum condition allocated significantly less to the highest-interest-rate debt than those in the suggested-minimum condition did (B = 115.48, 95% CI = [50.13, 180.83], t( 1,152) = 3.46, p < .001).
These allocations led participants in the required-minimum and suggested-minimum conditions to pay more interest than those in the control (Mcontrol = 7.20 [$1,424], Mmin = 7.22[$1,439], Msuggested = 7.22 [$1,449]; Bcontrol vs. min = .023, 95% CI = [.012,.033], t( 1,152) = 4.37, p < .001; Bcontrol vs. suggested = .024, 95% CI = [.014,.035], t( 1,152) = 4.60, p < .001). There was no significant difference between the minimum-payment and suggested-minimum conditions (B = .002, 95% CI = [−.009,.012], t( 1,152) = .355, p = .722), see Figure 4. Typically, lower dispersion and more focus on the highest-interest-rate debt correspond to lower interest paid. However, because our measure of interest is a composite of repayments on all accounts, this pattern may be driven by participants in the suggested-minimum condition making less efficient allocations to the non-highest-interest accounts.
Graph: Figure 4. Main results from Experiment 3.
To better understand the underlying process, we next examine the perceived recommendation to pay more than the minimum as a potential mediator of the differences in dispersion across conditions. As we expected, participants in the required-minimum condition report significantly higher agreement with the statement that there was a recommendation to pay more than the minimum when compared to the control participants (Mmin = 2.80, Mcontrol = 2.08; B = .72, 95% CI = [.47,.97], t( 1,152) = 5.68, p < .001), as do those in the suggested-minimum condition (Msuggested = 2.64; B = .56, 95% CI = [.32,.80], t( 1,152) = 4.56, p < .001).[12] However, there was no significant difference between the required and suggested minimums on this measure (B = -.16, 95% CI = [-.43,.10], t( 1,152) = 1.22, p = .221). Notably, average responses in each condition are below the scale midpoint, suggesting overall levels of disagreement. However, there is substantially more agreement, defined by responses at or above the scale midpoint, in the required-minimum condition (Mmin = 35% vs. Mcontrol = 22%; B = .64, 95% CI = [.32,.96], t( 1,152) = 3.96, p < .001) and the suggested-minimum condition (Msuggested = 32%; B = .50, 95% CI = [.18,.83], t( 1,152) = 3.07, p = .002) than in the control condition.
We use the Mediation package in R to conduct mediation analyses that separately compares each of the minimum-payment conditions with the no-minimum-payment control. The package allows us to cluster standard errors at the subject level using a quasi-Bayesian Monte Carlo simulation to estimate the variance ([39]).[13] We use 5,000 samples. Comparing the required-minimum and control conditions, the perceived recommendation to pay more than the minimum partially mediates the difference in dispersion (indirect effect = .0211, 95% CI = [.012,.032], p < .001; direct effect = .064, 95% CI = [.029,.097], p < .001). In a separate model, the perceived recommendation to pay more than the minimum fully mediates the difference between the control and suggested-minimum conditions (indirect effect = .0215, 95% CI = [.012,.033], p < .001; direct effect = .025, 95% CI = [−.008,.059], p = .135). These findings, illustrated in Figure 5, demonstrate that participants in the minimum-payment conditions were more likely to perceive a recommendation in the instructions to pay more than the minimum. Further, a substantial portion of the differences in dispersion between the no-minimum-payment control and the two minimum-payment conditions is explained by these differences in the perceived repayment recommendation.
Graph: Figure 5. Condition effects in mediation analysis in Experiment 3.
Endorsement of a perceived recommendation to pay more than the minimum accounts for some of the differences in dispersion we observe. Given the low average ratings on the perceived recommendation scale and the fact that the recommendation to pay more than the minimum does not significantly mediate the difference between the required and suggested minimums (indirect effect = .004, 95% CI = [−.002,.011], p = .219; direct effect = .036, 95% CI = [.003,.069], p = .036), this mechanism likely operates in concert with other factors. We=discuss potential additional factors in the "General Discussion" section.
Results from Experiment 3 suggest an important role for policy makers in designing decision environments. Specifically, while the role of minimum payments is commonly considered to be providing a base level of repayment necessary to avoid defaulting, their presence can be interpreted by consumers as a recommendation to pay more than the minimum. We find that the level of this perceived recommendation varies as a result of the details of the choice environment, for example, carrying greater influence when the minimum payment is required. In our next experiment, we examine how choice environments can exacerbate or reduce the consequences of minimum payments for consumers.
Experiments 1–3 documented a new cost to minimum payments—namely, that they lead participants to make more dispersed repayments and increase interest costs incurred. Experiment 3 further showed that the strength of the perceived recommendation to pay more than the minimum amount partially drives this effect. However, minimum payments serve important functions for the financial system, for example, by allowing credit card companies to classify accounts as in default and potentially reducing prices in the market overall. Consequently, in spite of the challenges that they create for consumers, it is unrealistic, from a policy perspective, to eliminate these requirements. The most efficient path for policy makers, firms, and consumer welfare advocates to improve consumer decision making or mitigate the costs of errors may not be through the elimination of the minimum payment requirement. Instead, they should reconsider the design of repayment interfaces. From an implementation perspective, third parties such as financial technology applications interested in enhancing consumer welfare may be most likely to produce a decision aid. Understanding of the psychology underlying debt repayment decisions, together with the role of information displays, can help inform these changes.
In the current experiment, we return to the setting from Experiment 1 but introduce four new conditions with required minimum payments that represent possible choice architectures policy makers, marketers, or app designers could use: a high-interest-salience condition, which sorts debts by their interest rates; an active-choice condition ([33]), which provides the minimum payment and full balance as two distinct salient options; a combination of the active-choice and salient-interest conditions; and a standard-paper-statement condition, which attempts to approximate the consumer experience of searching for relevant information in credit card debt statements. Each of these conditions provides insight into the way policy makers can choose to format decision environments to help—but also potentially hinder—consumers.
One thousand seven hundred ninety-one participants completed the main study on MTurk. Fifty-one percent of our participants were female, with a median age of 37 years, median income in the range of $60,000–$69,000, and modal education of a bachelor's degree. Eighty-eight percent of our participants reported having at least one credit card, with a median of two cards per participant.
Participants were randomly assigned to one of six conditions. The first was the no-minimum-payment control condition. The other five conditions had minimum payments. The second condition was the same as the standard minimum-payment condition from Experiment 1. We include this condition to examine the size of the baseline dispersion effect in this setting and to have a benchmark of comparison relative to the newly introduced designs. In addition, we introduce four new conditions that attempt to capture the way that different features of choice architecture can alter the impacts of minimum payments. The interest-salience condition aimed to increase the perceived importance of interest rates by ordering the debts by interest rates with the highest-interest-rate debt on top. We expected and found in a pretest that participants in this condition would be more likely to perceive a recommendation to repay according to interest rates (see Web Appendix A, Study H). If consumers underweight the importance of interest and thus allocate less to high-interest rate debts, this condition will decrease interest costs associated with minimum payments.
The active-choice condition was modeled on the current format for online repayments, which allows participants to easily select the minimum amount, select the full statement balance, or enter a different value (see Figure 6). Online repayment of credit card debts is growing, with about 10% of cardholders using online portals ([33]). However, the regulations and nudges that appear in paper statements are not visible to the consumer when making repayments in these interfaces. Prior work examining these interfaces suggests that they can highlight paying debt amounts in full. By nudging people to repay in full, we predict that consumers will focus their repayments on a smaller number of debts, reducing dispersion. To the extent that consumers focus on high-interest debts, this choice architecture may help consumers overcome the interest-cost consequences of minimum payments. We expected and found in a pretest that participants did perceive a recommendation to repay accounts in full (see Web Appendix A, Study H). Building on our findings from Experiment 3 (H2), we anticipated that shifting this perceived recommendation from a focus on the minimum payment to the full statement amount would minimize the dispersion effect. The active choice × salient interest condition combined active-choice repayment mode with the debt table sorted by interest rate.
Graph: Figure 6. Participant display for the active choice condition in Experiment 4.
Finally, we were interested in understanding how each of these highly designed settings that highlighted specific information to consumers compared with a standard paper statement. A key feature of the standard paper credit card statement is that it highlights debt amounts and not interest rates. While debt balances, minimum payments, and the minimum payment warning disclosure are shown on the front page of credit card statements by regulation, interest rates are frequently found on the last page of the statement. As a result, consumers need to exert effort to find the interest information. To approximate the need for information search, we created a table similar to the information table in Experiment 1 except that only amounts and minimum payments were featured on the initial screen. Participants could click a link to see additional information. The information acquired through the link included the interest rate along with the amount consumers would have to pay to pay off the debt in full in three periods (similar to the three-year number reported on credit card statements after the CARD Act of 2009), the previous balance, the last payment, and the amount of interest paid (for the participants' screens, see Figure 7). The goal of this condition was to increase the ecological validity of the task by adding an element of information search, though its impact on dispersion was not clear ex ante. We expected that by highlighting the amounts more than interest, participants would be more likely to perceive a recommendation to repay according to debt amounts and confirmed this intuition in our pretest (see Web Appendix A, Study H). This study was preregistered at aspredicted.org.[14]
Graph: Figure 7. Participant display for the standard statement condition in Experiment 4.
We excluded participants who failed our attention checks twice (N = 4) and participants who allocated more than their debt amount in more than one round (N = 48).[15] The effects of minimum payments varied across information presentation conditions (see Figure 7). We first examine the effects of the paper statement condition, because this maps most closely onto the most common information presentation for consumers in the field. The paper statement condition was the most costly to consumers. Though it did not substantially affect dispersion relative to the standard-minimum-payment condition (B = .053, 95% CI = [−.14,.24], t( 1,738) = .540, p = .589), it did significantly increase dispersion relative to the no-minimum-payment condition (B = .369, 95% CI = [.186,.552], t( 1,738) = 3.95, p <.001). Further, it reduced the amount allocated to the highest-interest-rate debt (minimum: B = −157.2, 95% CI = [−225.28, −89.13], t( 1,738) = 4.53, p < .001; control: B = −490.33, 95% CI = [−577.81, −402.85], t( 1,738) = 10.99, p < .001) and significantly increased the interest paid per round (minimum: B = .047, 95% CI = [.031,.064], t( 1,738) = 5.65, p < .001; control: B = .067, 95% CI = [.051,.084], t( 1,738) = 8.04, p < .001) when compared with both the standard-minimum-payment condition and the no-minimum-payment condition. Notably, these results could be driven by participants in this condition either failing to seek out additional information or using the information less effectively than participants in the standard-minimum condition.
We next turn to the condition that showed the most promise in reducing dispersion and interest costs: the active-choice condition. When compared with the standard-paper-statement condition, the active-choice condition significantly outperforms on all of our measures (dispersion: B = −.28, 95% CI = [−.47., −.10], t( 1,738) = 3.02, p = .002; highest-interest payments: B = 224.74, 95% CI = [157.16, 292.32], t( 1,738) = 6.52, p < .001; log interest paid: B = −.054, 95% CI = [−.071, −.037], t( 1,738) = 6.29, p < .001). Further, when compared with the standard-minimum condition, the active-choice condition significantly reduced dispersion (B = −.231, 95% CI = [−.41, −.055], t( 1,738) = 2.57, p = .01). It also significantly increased the amount allocated to the highest-interest-rate debt (B = 67.53, 95% CI = [3.28, 131.79], t( 1,738) = 2.06, p = .04) but only directionally reduced the amount of interest paid (B = −.007, 95% CI = [−.022,.008], t( 1,738) = .89, p = .373). By introducing an additional perceived recommendation, the recommendation to pay off debts in full, the active-choice condition partially moderated the effect of minimum payments on dispersion.[16] As Figure 8 shows, the other minimum-payment conditions fell somewhere between the active choice and the standard paper statement (for regressions, see Web Appendix B, Table W27).
Graph: Figure 8. Main results from Experiment 4.
The differences across conditions with minimum payment requirements suggests a role for policy makers and firms in designing consumers' choice environments and their implied recommendations. In particular, the current shrouding of interest rates in paper statements is especially costly for consumers because it leads them to repay less to high-interest debts. These results suggest that highlighting information on the core characteristics of debts, particularly interest rates, can help consumers repay in a less costly way, even in the presence of minimum payments.
In addition, the active-choice condition, by presenting multiple default options, can decrease dispersion and increase repayments to the highest-interest debts. This design also outperforms the standard-paper-statement condition on all measures examined. Choice architects can take advantage of this format to help consumers increase consideration of interest rates and possibly provide motivational benefits, as documented in Experiment 2 and in prior literature ([23]). At the same time, credit card companies can profit from higher-interest payments and therefore may have an incentive to maintain current user interfaces. Consequently, making repayment environments more user friendly may require government intervention or additional third parties like financial aggregation apps (e.g., Mint, NerdWallet, Tally).
In addition to the studies reported previously, we conducted five experiments to further explore the dispersion effect of minimum payments, using a variety of alternative designs.[17] We briefly summarize results here and provide complete details in the Web Appendix. The first additional experiment replicated our findings in an MTurk sample using a similar design to Experiment 1 (Web Appendix A, Study A). The second additional experiment aimed to enhance the ecological validity of the experimental design. Because credit card bills tend to arrive at different points throughout the month, we extend our findings to a version of our task in which participants make allocation decisions one at a time (Web Appendix A, Study B). We find consistent patterns using this design, allowing us to rule out the possibility that the effect is driven by participants responding to all debts in the same elicitation. This enables us to generalize the dispersion effect and more closely approximate credit card bill timing from the world.
The third additional experiment aimed to address the possibility that a difference in the mathematical sophistication required in the minimum-payment condition versus the control condition drives the dispersion effect. We examined a scenario where all the minimum payments are round numbers ($25). We found a marginally significant increase in dispersion and a significant decrease in the amount paid to high-interest debt, though the log of interest paid per round did not reach significance (Web Appendix A, Study C). These findings suggest that even when the minimum payments are round numbers, they still increase dispersion, though the effect may be somewhat weaker.
The fourth additional experiment introduced a condition where the minimum payment amount was filled in by default (Web Appendix A, Study D). Dispersion and interest costs in the default condition fell directionally between the minimum payment and control condition. Thus, the default payment interface appears to have similar outcomes for consumers as the active-choice interface. The default payment interface may operate in a similar way to the active-choice condition, shifting perceptions of recommended payment amounts. However, because participants can also ignore debts with the minimum payments already filled in, it may also reduce dispersion by reducing the number of accounts for which they have to make allocation decisions.
Finally, our hypotheses specifically ask about the presence of minimum requirements on all cards. However, there could be situations in which some cards have minimum requirements but others do not. In a one-round version of our task, we found that the dispersion effect of minimum payments persisted when there were minimum payments on some, but not all, accounts (Web Appendix A, Study E). This suggests that the minimum payment requirement can spill over to debts that do not have minimum payments.
This article presents new insights into the influence of minimum payments on consumer debt decisions. First, we document a dispersion effect of minimum payments that leads consumers to spread their discretionary allocations across more accounts. Second, we show that the dispersion effect tends to lead to larger interest costs. Further, we provide evidence that the perceived recommendation to pay more than the minimum partially underlies this effect. Together, our experimental results suggest that minimum payment requirements may contribute to the overdispersion of repayments observed in field data (e.g., [16]).
Our findings also document a novel path, beyond anchoring, through which minimum payments can harm consumers by increasing allocation dispersion. Consumers facing minimum payments fundamentally alter their strategic approach to repayments across accounts. Although minimum payments serve an important role in ensuring that debts are not neglected, they may negatively impact consumer financial well-being in part by suggesting a recommendation to pay above the minimum on each account. Importantly, we find that information and allocation environments can influence the impact of the minimum payment on dispersion and interest costs. Alternative information presentations can help consumers reduce interest costs in the presence of minimum payments but can also exacerbate the interest cost consequences for consumers. In fact, consumers in the condition modeled on current paper statements performed the worst, suggesting there are substantial opportunities to improve consumer outcomes through choice architecture.
We contribute to the literature on heuristics in debt repayment by studying the role of minimum payments in changing allocations across multiple cards. Previous research suggests that consumers allocate money to the accounts with the lowest amount of debt first ([ 3]; [ 5]; [ 6]). While some participants in our data show a strong focus on the accounts with the lowest amount of debt, most focus on their highest-interest-rate debts but allocate less to those debts in the minimum-payment condition.
Recent work has also provided evidence for a balance-matching heuristic, which implies that consumers will make their largest allocation to the account with the largest debt ([16]). The dispersion effect we document is closest in nature to that heuristic. However, because consumers are mostly focused on their highest-interest-rate debts, our results suggest that balance matching may not be an intentional decision strategy. Instead, it may be a consequence of the structure of debt repayment in the world. Specifically, minimum payments and the implied recommendation to pay more than the minimum may play a role in driving consumers to repay as if balance matching in the world, because they increase dispersion.
Finally, our participants do not appear to be using a pure 1/N heuristic with equal allocations across all accounts motivated by a desire to diversify ([ 4]). Instead, they seem to be focusing on salient interest costs and, when in the minimum-payment condition, moving repayments from this higher-interest debt to lower-interest ones. For example, in Experiment 1, participants allocate their largest repayment to the highest-interest-rate debt in 46% of rounds in the minimum-payment condition. By contrast, they allocate exactly evenly to all cards in only 12% of rounds. Thus, the effect of minimum payments appears to be a deliberate strategy shift that reflects their desire to act in accordance with the perceived recommendation from the minimum payment as opposed to a shift to 1/N heuristic use that relies on diversification in the absence of more informative cues.
Our studies have several limitations that suggest avenues for future work. First, our paradigm includes only three allocation periods. There may be an opportunity to examine the impact of learning and feedback on repayment decisions by extending the paradigm to more rounds. Second, while we have evidence for two distributions of debt amounts, there may be boundary conditions induced by particularly low or high debt amounts. This avenue may be of particular interest given that we find substantially more dispersion in our studies than has been documented in field data ([23]). It may be that the way consumers accrue debt and the dispersion of their debt accounts influences the dispersion of their repayments. There may be additional scope for examining what specific component of the minimum payment leads consumers to perceive a recommendation to repay more than the minimum. For example, different types of minimum payments—such as round numbers or flat, consistent amounts across cards—may alter perceptions of recommendations. We present data on suggested minimum payments as well as a flat minimum of $25 (see Web Appendix A, Study C) that lead to similar levels of dispersion as documented in the previous studies. Finally, it is not clear how moving from an environment with all minimum payments to one in which only some cards have minimums would affect consumers. While our hypotheses concern minimums on all versus no cards, data reported in the Web Appendix suggest that the presence of minimum payments on only a subset of accounts may similarly increase dispersion (Web Appendix A, Study E). Capturing these details will help future work explore the dynamics of debt repayment decisions more thoroughly, with direct implications for policy.
In addition, although we have identified perceived recommendations to repay more than the minimum as one mechanism underlying the dispersion effect, other mechanisms are likely to operate in concert. For example, some additional dispersion may come from using more approximate strategies, such as rounding, as a result of increased decision complexity ([16]). Another possibility is that minimum payments may alter the size of the consideration set, which could potentially lead to an increase in the use of 1/N type heuristics ([27]). Further, the default repayment experiment (Web Appendix A, Study D) provides suggestive evidence that increasing the number of active allocation decisions that a participant needs to make may also increase dispersion. Future research should examine these additional mechanisms in greater detail.
Although we have been examining effects on average across the population, consumers do not all use the same strategies to make these decisions. Understanding heterogeneity in repayment decisions and how choice architectures can nudge consumers to make better financial decisions is important, particularly in light of recent findings that low-financial-literacy consumers are the most likely to be affected by certain nudges ([28]).
Finally, there may be opportunities to bring the insights from the debt repayment heuristics literature to other similar allocation decisions, particularly investing. Our results suggest that highlighting important decision features—for example, fees in an investment context—may help consumers focus on those features. As a result, they may allocate their retirement dollars to higher-yield, lower-cost funds, which will leave them better off than using strategies such as the 1/N heuristic ([ 4]). In a charitable giving context, highlighting ratings such as Charity Navigator's may direct money toward higher-impact charities. In addition, if people simultaneously donate to a popular charity and other less well-known ones, adding a suggested donation to each charity may lead people to spread their donations more evenly.
This article has several important implications for practice. First, the results highlight the importance of considering what inferences consumers will draw from choice architecture. We show that minimum payment requirements lead consumers to infer a recommendation to pay more than the minimum, which causes increased dispersion and higher interest costs. When policy makers revise requirements for credit card statements, our results suggest that it is important to test the inferences consumers draw and how it changes their decision making. Doing so will allow policy makers to produce desired effects while heading off unanticipated consequences.
Second, while it is unlikely that the government would mandate the removal of minimum requirements, several decision aids could mitigate the costs associated with the dispersion effect. One possible choice architecture is adopting active-choice decision environments that highlight interest rates. Active-choice environments led to decreased dispersion in Experiment 4. Other work has shown consumers with an active-choice format are more likely to repay debts in full ([33]). This may increase the amount consumers dedicate to debt repayment in the future ([23]).
Relative to a statement modeled on current paper statements, our participants tended to focus more on higher-interest-rate debts when the interest information was displayed in a format that facilitated comparison (Experiment 4). Featuring interest rates on the front page of credit card statements may help consumers focus on minimizing interest costs instead of amount-based heuristics. Currently, interest rate information is not shown on the first page of credit card statements, unlike minimum payments and other debt amount information. Making interest more salient could offset some of the costs associated with increased dispersion.
Financial technology firms and account aggregators (e.g., Mint, NerdWallet), which consumers give account access to, may be in a good position to aid consumers by aggregating their credit card debt information, including interest rates. By focusing consumers on interest, our data suggest that such a decision aid could provide a buffering effect for consumers, reducing the consequences of dispersed allocations. A more extreme solution already being offered in the marketplace is automated credit debt management. For example, users of the app Tally pay a single sum to the app, which then allocates the lump sum to minimum payments and the consumer's highest-interest-rate debt ([37]). However, it is not clear that consumers are fully aware that they are paying excess interest, which may reduce demand for these products.
Importantly, our results suggest another channel by which the number of debt accounts could influence financial well-being. Recently, there has been substantial interest in defining and measuring financial well-being, using both subjective assessments and administrative data (e.g., [10]; [29]; Netemeyer et al. 2018). Our work suggests that, holding both total debt amount and number of accounts constant, the amount of dispersion may serve as a cue to differences in long-term interest costs that a borrower is likely to incur. In addition, reducing the total number of debt accounts may blunt the dispersion effect simply because there are fewer debts to disperse over. This could also lead to lower accumulated interest costs over time, even given an identical debt balance initially. As a result, the total number of active debt accounts and dispersion across these accounts may be useful indicators for measuring and intervening on financial well-being. Drawing firms' attention to these metrics may change the way they help consumers plan, budget, and understand their best options.
Finally, our work highlights the importance of considering the consumers' situation and inferences when designing policy. Most consumers have multiple debt accounts, so nudges targeted at increasing repayment on individual accounts may lead consumers to incur additional costs. In this instance, policy makers need to consider the impact of nudges on the full portfolio of debt accounts when they make policy changes. More broadly, applying marketing research to policy requires careful consideration of consumers' decision environment. We recommend that marketers make use of descriptive research not just to motivate their research questions but also in their empirical designs to maximize the likelihood of real-world impact.
sj-pdf-1-jmx-10.1177_00222429211047237 - Supplemental material for Minimum Payments Alter Debt Repayment Strategies Across Multiple Cards
Supplemental material, sj-pdf-1-jmx-10.1177_00222429211047237 for Minimum Payments Alter Debt Repayment Strategies Across Multiple Cards by Samuel D. Hirshman and Abigail B. Sussman in Journal of Marketing
Footnotes 1 Cait Lamberton
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work is supported by the True North Communications, Inc. Faculty Research Fund and the Beatrice Foods Co. Faculty Research Fund at the University of Chicago Booth School of Business.
4 Samuel D. Hirshman https://orcid.org/0000-0002-0330-9375 Abigail B. Sussman https://orcid.org/0000-0002-5333-558X
5 The argument that [16] make is not necessarily that people strategically intend to pay in this way but, rather, that their observed payments are best captured by the balance-matching model. This is akin to [14], who argues that the outcomes of people's decisions may appear rational even in cases when their decision-making process does not map onto the steps or calculations that a rational model would take.
6 The algorithm for determining the bonus was the ratio of the participants' amount of debt cleared minus the worst possible strategy to the interest cost-minimizing strategy minus the worst possible strategy. The worst possible debt clearance amount was constructed by using a strategy in which debts were paid in order of lowest to highest interest rate and minimum payments (when applicable) were not made. This was not described to the participants in detail.
7 An analysis limited to participants who made all minimum payments and, as a result, did not incur any fees yielded similar results (see Web Appendix B, Table W5).
8 This exclusion did not differ significantly by condition (minimum 2.4% vs. no minimum 2%, z = .28, p = .782)
9 See https://aspredicted.org/b9ap6.pdf. In the main text, we deviate from this analysis by adding a measure of the amount of money allocated to the highest-interest-rate debt and by reporting a simpler mediation analysis than was preregistered. The preregistered analysis is available in Web Appendix B, Tables W19 and W20, and yields similar conclusions.
As an example, in the required-minimum condition, this statement would read, "The instructions provided a recommendation to pay MORE THAN THE REQUIRED MINIMUM to every card."
These exclusions did not differ significantly by condition (see Web Appendix B, Table W13).
In Web Appendix B, Tables W19–W22, we present two alternative formulations for the mediation that incorporate the ancillary questions about perceived recommendations. The results are similar using each of these measures.
As suggested by [22], the coefficient for the indirect effects corresponds to the average causal mediation effect on the linear scale and the direct effect is also on the linear scale (see their Appendix F).
See https://aspredicted.org/me84a.pdf. We also included the amount of money allocated to the highest-interest-rate debt, which was not in our preregistration.
These exclusions did not differ significantly by condition (see Web Appendix B, Table W24).
Given the similarities between the active choice and the active choice × salient interest condition, it is worth noting that although the active choice × salient interest condition is directionally more dispersed, it is not significantly different from the active choice condition on any of our measures.
To summarize these results as well as those presented in the article itself, we include a table that features the key statistics for all of the studies (see Web Appendix B, Table W1).
References Agarwal Sumit , Chomsisengphet Souphala , Lim Cheryl. (2017), " What Shapes Consumer Choice and Financial Products? A Review ," Annual Review of Financial Economics , 9 (November) , 127 – 46.
Agarwal Sumit , Chomsisengphet Souphala , Mahoney Neale , Stroebel Johannes. (2015), " Regulating Consumer Financial Products: Evidence from Credit Cards ," Quarterly Journal of Economics , 130 (1), 111 – 64.
Amar Moty , Ariely Dan , Ayal Shahar , Cryder Cynthia E. , Rick Scott I.. (2011), " Winning the Battle but Losing the War: The Psychology of Debt Management ," Journal of Marketing Research , 48 (SPL), S38 – 50.
Benartzi Shlomo , Thaler Richard H.. (2001), " Naive Diversification Strategies in Defined Contribution Saving Plans ," American Economic Review , 91 (1), 79 – 98.
Besharat Ali , Carrillat François A. , Ladik Daniel M.. (2014), " When Motivation Is Against Debtors' Best Interest: The Illusion of Goal Progress in Credit Card Debt Repayment ," Journal of Public Policy & Marketing , 33 (2), 143 – 58.
Besharat Ali , Varki Sajeev , Craig Adam W.. (2015), " Keeping Consumers in the Red: Hedonic Debt Prioritization Within Multiple Debt Accounts ," Journal of Consumer Psychology , 25 (2), 311 – 6.
Brown Alexander L. , Lahey Joanna N.. (2015), " Small Victories: Creating Intrinsic Motivation in Task Completion and Debt Repayment ," Journal of Marketing Research , 52 (6), 768 – 83.
CFPB (2019), " The Consumer Credit Card Market ," (August 27), https://www.consumerfinance.gov/data-research/research-reports/the-consumer-credit-market-2019/.
Clark Michael. (2019), " Fractional Regression, " (August 19), https://m-clark.github.io/posts/2019-08-20-fractional-regression/
Comerton-Forde Carole , Ip Edwin , Ribar David C. , Ross James , Salamanca Nicolas , Tsiaplias Sam. (2018), " Using Survey and Banking Data to Measure Financial Wellbeing ," Commonwealth Bank of Australia and Melbourne Institute Financial Well-Being Scales Technical Report 1.
Donnelly Grant E. , Lamberton Cait , Bush Stephen , Chance Zoe , Norton Michael I.. (2020), " 'Repayment-by-Purchase' Helps Consumers to Reduce Credit Card Debt ," working paper.
El Issa Erin. (2017), " 2017 American Household Credit Card Debt Study ," NerdWallet (accessed October 21, 2021), https://www.nerdwallet.com/blog/credit-card-data/household-credit-card-debt-study-2017/.
Federal Reserve Bank of New York (2021), " Quarterly Report on Household Debt and Credit," (May).
Friedman Milton. (1953), Essays in Positive Economics. Chicago : University of Chicago Press.
Gal David , McShane Blakeley B.. (2012), " Can Small Victories Help Win the War? Evidence from Consumer Debt Management ," Journal of Marketing Research , 49 (4), 487 – 501.
Gathergood John , Mahoney Neale , Stewart Neil , Weber Jörg. (2019), " How Do Individuals Repay Their Debt? The Balance-Matching Heuristic ," American Economic Review , 109 (3), 844 – 75.
Gross David B. , Souleles Nicholas S.. (2002), " Do Liquidity Constraints and Interest Rates Matter for Consumer Behavior? Evidence from Credit Card Data." The Quarterly Journal of Economics , 117 (1), 149 – 185.
Herrnstein Richard J. (1970), " On the Law of Effect ," Journal of the Experimental Analysis of Behavior , 13 (2), 243 – 66.
Herrnstein Richard J. , Prelec Drazen. (1991), " Melioration: A Theory of Distributed Choice ," Journal of Economic Perspectives , 5 (3), 137 – 56.
Hershfield Hal E. , Roese Neal J. , (2015) " Dual Payoff Scenario Warnings on Credit Card Statements Elicit Suboptimal Payoff Decisions ," Journal of Consumer Psychology , 25 (1), 15 – 27.
Homonoff Tatiana , O'Brien Rourke , Sussman Abigail B.. (2021), " Does Knowing Your FICO Score Change Financial Behavior? Evidence from a Field Experiment with Student Loan Borrowers ," Review of Economics and Statistics , 103 (2), 236 – 50.
Imai Kosuke , Keele Luke , Tingley Dustin. (2010), " A General Approach to Causal Mediation Analysis ," Psychological Methods , 15 (4), 309 – 34.
Kettle Keri L. , Trudel Remi , Blanchard Simon J. , Häubl Gerald. (2016), " Repayment Concentration and Consumer Motivation to Get Out of Debt ," Journal of Consumer Research , 43 (3), 460 – 77.
Keys Benjamin J. , Wang Jialan. (2019), " Minimum Payments and Debt Paydown in Consumer Credit Cards ," Journal of Financial Economics , 131 (3), 528 – 48.
McKenzie Craig R.M. , Liersch Michael J. , Finkelstein Stacey R.. (2006), " Recommendations Implicit in Policy Defaults ," Psychological Science , 17 (5), 414 – 20.
Medina Paolina C. , Negrin Jose L.. (2021), " The Hidden Role of Contract Terms: The Case of Credit Card Minimum Payments in Mexico ," Management Science (published online August 5), https://doi.org/10.1287/mnsc.2021.4006.
Morrin Maureen , Inman J. Jeffrey , Broniarczyk Susan M. , Nenkov Gergana Y. , Reuter Jonathan. (2012), " Investing for Retirement: The Moderating Effect of Fund Assortment Size on the 1/N Heuristic ," Journal of Marketing Research , 49 (4), 537 – 50.
Mrkva Kellen , Posner Nathaniel A. , Reeck Crystal , Johnson Eric J.. (2021), " Do Nudges Reduce Disparities? Choice Architecture Compensates for Low Consumer Knowledge ," Journal of Marketing , 85 (4), 67 – 84.
Muggleton Naomi K. , Quispe-Torreblanca Edika G. , Leake David , Gathergood John , Stewart Neil. (2020), " Evidence from Mass-Transactional Data That Chaotic Spending Behaviour Precedes Consumer Financial Distress ," PsyArXiv (May 10), https://doi.org/10.31234/osf.io/qabgm.
Navarro-Martinez Daniel , Salisbury Linda Court , Lemon Katherine N. , Stewart Neil , Matthews William J. , Harris Adam J.L.. (2011), " Minimum Required Payment and Supplemental Information Disclosure Effects on Consumer Debt Repayment Decisions ," Journal of Marketing Research , 48 (Special Issue), S60 – 77.
Netemeyer Richard G. , Warmath Dee , Fernandes Daniel , Lynch John G. Jr.. (2018), " How Am I Doing? Perceived Financial Well-Being, Its Potential Antecedents, and Its Relation to Overall Well-Being ," Journal of Consumer Research , 45 (1), 68 – 89.
Salisbury Linda Court. (2014), " Minimum Payment Warnings and Information Disclosure Effects on Consumer Debt Repayment Decisions ," Journal of Public Policy & Marketing , 33 (1), 49 – 64.
Salisbury Linda Court , Zhao Min. (2020), " Active Choice Format and Minimum Payment Warnings in Credit Card Repayment Decisions ," Journal of Public Policy & Marketing , 39 (3), 284 – 304.
Soll Jack B. , Keeney Ralph L. , Larrick Richard P.. (2013), " Consumer Misunderstanding of Credit Card Use, Payments, and Debt: Causes and Solutions ," Journal of Public Policy & Marketing , 32 (1), 66 - 81.
Stewart Neil. (2009), " The Cost of Anchoring on Credit-Card Minimum Repayments ," Psychological Science , 20 (1), 39 – 41.
Sussman Abigail , O'Brien Rourke L.. (2016), " Knowing When to Spend: Unintended Financial Consequences of Earmarking to Encourage Savings ," Journal of Marketing Research , 53 (5), 790 – 803.
Tally (2020), " How Tally Works, " (accessed November 8, 2021), https://www.meettally.com/how-tally-works.
Thaler Richard H. , Sunstein Cass R.. (2009), Nudge: Improving Decisions About Health, Wealth, and Happiness. New York : Penguin.
Tingley Dustin , Yamamoto Teppei , Hirose Kentaro , Keele Luke , Imai Kosuke. (2014), " Mediation: R Package for Causal Mediation Analysis ," Journal of Statistical Software , 59 (5), 1 – 38.
~~~~~~~~
By Samuel D. Hirshman and Abigail B. Sussman
Reported by Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 94- Mobilizing the Temporary Organization: The Governance Roles of Selection and Pricing. By: Ghazimatin, Elham; Mooi, Erik A.; Heide, Jan B. Journal of Marketing. Jul2021, Vol. 85 Issue 4, p85-104. 20p. 1 Diagram, 3 Charts, 2 Graphs. DOI: 10.1177/0022242920982545.
- Database:
- Business Source Complete
Mobilizing the Temporary Organization: The Governance Roles of Selection and Pricing
Many marketing transactions between buyers and suppliers involve short-term collaborations or so-called temporary organizations. Such organizations have considerable value-creation potential but also face challenges, as evidenced by their mixed performance records. One particular challenge involves relationship governance, and in this respect, temporary organizations represent a conundrum: On the one hand, they pose significant governance problems due to the need to manage numerous independent specialists under time constraints. On the other hand, temporary organizations lack the inherent governance properties of other organizational forms such as permanent organizations. The authors conduct an empirical study of 429 business-to-business construction projects designed to answer two specific questions: First, how are particular selection and pricing strategies deployed in response to monitoring and coordination problems? Second, does the joint alignment between the two mechanisms and their respective attributes help mitigate cost overruns? The authors follow a formal hypothesis test with a series of in-depth interviews to explore and to gain insight into the validity of the key constructs, explanatory mechanisms, and outcomes. Managerially, the authors answer the long-standing question of how to mobilize a temporary organization. Theoretically, they develop an augmented "discriminating alignment" heuristic for relationship management involving multiple governance mechanisms and attributes.
Keywords: business-to-business marketing; discriminating alignment; governance; partner selection; pricing terms; projects; temporary organizations
It is commonly accepted that marketing performance, or value creation more broadly, depends on how the buyer–supplier interface is organized or governed. A large body of research shows how both formal organizational structures ([57]; [74]) and long-term relationships ([19]; [27]) play important governance roles.
Recent research shows, however, that much marketing activity takes place outside of formal organizations and long-term relationships through so-called temporary organizations ([ 4]; [37]), namely a "temporally bounded group of interdependent actors" ([ 9], p.1237) assembled for the purpose of performing a particular task ([31]). Temporary organizations have existed since antiquity, and they have been used for a variety of purposes, including the deployment of military campaigns such as those of Alexander the Great, the construction of iconic structures such as the Great Wall of China and the Sydney Opera House, the execution of the Manhattan Project and the Apollo Moon landing, and the development of well-known advertising campaigns and new products ([22]; [32]; [61]). Spending on temporary organizations is substantial—a McKinsey study ([25]) estimates a worldwide expenditure of $57 trillion from 2015 to 2030 on infrastructure projects alone.
Surprisingly, despite their versatility and prominence, temporary organizations have been subject to limited systematic theorizing. In particular, their specific governance practices are poorly understood. Researchers (e.g., [ 6]) have noted how past accounts of temporary organizations (e.g., [49]; [79]) are limited to stating their general impetus, such as knowledge sharing, cost containment, and flexibility, but they are silent on the specific mechanisms by which such outcomes come about. Consequently, the question originally raised by [30], p. 108) about "how to mobilize a temporary organization" remains largely unanswered.
Temporary organizations represent a governance conundrum. First, they must coordinate and monitor the efforts of multiple specialist suppliers without the benefit of the mechanisms that are available to other organizational forms, such as hierarchical structures, rules, and long-term compensation ([93]). In addition, they must manage the relevant relationships under time pressure ([62]). These challenges have led some researchers to describe temporary organizations as inherently ephemeral and unstable ([ 6]; [49]), using descriptors such as "one-night stands" ([68], p. 167). There is industry evidence to support this position: studies of major construction projects show that nine out of ten have significant cost overruns ([22]; [23]).
Other researchers, however, have challenged the instability thesis and suggested that temporary organizations may draw on unique sources of support, such as prior ties between their members ([20]). This line of work, however, has stopped short of specifying these mechanisms' specific effects and performance implications.
Our main contribution is to shed light on the governance conundrum of temporary organizations. In doing so, we build on recent work by [37], which examined such organizations through a governance lens. Our specific focus is on so-called "hybrid" temporary organizations that exist outside of the boundaries of a permanent firm (unlike "fully embedded" organizations) yet may exhibit prior ties among their members (unlike "stand-alone" organizations, which are assembled from scratch). We develop and empirically test a novel two-stage model of hybrid governance. The first stage, the deployment stage, considers ( 1) the monitoring and coordination problems that follow from a temporary organization's size (the number of involved suppliers or subcontractors) and ( 2) how these problems drive governance choices in terms of selection and pricing. These effects, in turn, are contingent on a hybrid's particular features, namely ( 1) the availability of prior ties among its members at different levels, including buyer, general contractor, and subcontractor, and ( 2) its particular time horizon or contracted duration.
The second stage in our framework, the performance stage, captures whether the mechanisms of selection and pricing, when jointly aligned with their respective attributes (called plural discriminating alignment), help minimize cost overruns. In general, our model captures the unique logic of temporary organizations, which involves solving multiple governance problems under time constraints. This logic has been succinctly expressed by [62], p. 437) as "the need to make things happen."
Beyond the general goal of developing and testing a novel model of temporary organizations, we make the following three contributions: First, our conceptualization advances the theoretical notions of plural governance and alignment. Since the publication of [ 7] seminal article, researchers (e.g., [12]; [39]) have embraced the idea that buyer–supplier interfaces comprise multiple governance mechanisms. Previous research, however, has focused on the observed relationships between the mechanisms themselves ([78]). We advance the plural-forms thesis by showing that performance is not necessarily a function of multiple governance mechanisms per se but of whether the different mechanisms are simultaneously aligned with their theoretically specified antecedents. Practically, as noted by one of our interviewees, this question addresses "how we can get things right." Theoretically, it adds precision to the plural forms thesis and to the alignment thesis in governance theory ([95]) more generally.
Second, we generate fine-grained insights into two fundamentally different approaches to governance: partner selection and incentives. These two mechanisms work in quite different ways, by either ( 1) identifying the "right" party in the first place or ( 2) inducing action through financial rewards ([76]; [91]). Our particular analyses enable us to undertake a precise comparison of the two approaches and to examine whether aligned selection has a more significant performance effect than aligned incentives. Because projects can fail spectacularly ([66]), taking a significant financial as well as human toll, understanding performance is a key concern.
Third, and building on the previous point, the two mechanisms that we study, selection and pricing, are managerially relevant and can be readily deployed within the time frame of a temporary organization. Their quick deployment has intrinsic benefits. [ 6] study of movie projects shows the challenge of achieving coordination under time constraints, as vividly illustrated by a comment from one of her subjects during a movie shoot: "We have two days to do it—we need to get things done right away." Despite their quick deployment, we show that these mechanisms have significant long-term performance implications in the form of ex post costs that manifest themselves after project completion.[ 5]
This article is organized as follows: The next section provides a general discussion of governance in a temporary organization context. This is followed by a presentation of our conceptual framework and research hypotheses. Next, we describe the research method used to test the hypotheses, including a unique data set of 429 construction projects spanning the time period from 2001 through 2015. We follow our formal hypothesis tests with a set of in-depth interviews with construction managers to gain insights into the validity of our hypothesized constructs, explanatory mechanisms, and outcomes. Collectively, the projects we study are worth $33 billion, yet they suffered from $4 billion in cost overruns. We conclude by discussing the study's contributions and limitations and create an agenda for future research on temporary organizations.
Temporary organizations raise complex questions that are not addressed by standard theory. In particular, they raise questions regarding the properties that its governance mechanisms must possess, compared with permanent firms and long-term relationships.
First, a temporary organization's discrete time horizon requires mechanisms that possess a deployment property, namely short-term marketing tools ([45]) that can ( 1) be rapidly implemented at the organization's initiation stage ([38]) and ( 2) take effect "swiftly" ([68], p. 175). The mechanisms of selection and pricing, which we focus on, possess such properties. Parenthetically, a deployment property contrasts sharply with other, more commonly studied, governance mechanisms that require time-consuming design efforts, such as permanent organizational structures ([93]) and relational norms ([63]).
Second, the emphasis on governance deployment has implications for mechanism content—namely, a need for mechanisms that are formal, rather than informal, in nature ([78]). For instance, pricing provisions in business-to-business settings are typically built into formal contracts ([71]) that can be crafted and brought to bear on an organization quickly. In contrast, informal governance mechanisms such as norms and expectations possess a distinct social component ([48]), and their availability requires prolonged interaction between parties.
The third property that we identify has to do with the level at which a given governance mechanism originates. Building on and extending the "discriminating alignment" principle of transaction cost economics ([95]), we posit that temporary organizations require mechanisms that are organization-specific in nature, in the sense that they match the focal organization's particular attributes. That said, prior research suggests that some temporary organizations, sometimes described as "hybrids" ([84]), can also draw on preexisting, more general properties that can be "activated" or brought to bear on the organization in question. For instance, [20], in his analysis of so-called "quasi-firms," shows how construction projects often exhibit recurring ties between its members. Similar patterns have been shown in movie production, where repeated interactions between actors and producers ([21]) serve governance purposes through the role structures and social bonds that emerge over time. Indeed, some researchers (e.g., [48]; [49]) have described temporary organizations as exemplars of social forms of governance. Despite the intuitive appeal of such arguments, important questions remain regarding ( 1) the specific form that prior ties can take and ( 2) the juxtaposition between prior ties and organization-specific governance mechanisms. We turn to these questions next.
Our main theoretical anchoring is twofold: First, we rely on the extant literature on temporary organizations (e.g., [37]; [46]) to identify relevant antecedents or attributes such as the focal organization's size ([43]; [53]). Second, we draw on the "new institutional economics" (e.g., [94]) literature to link the relevant attributes with governance mechanisms and performance outcomes.
Figure 1 presents our conceptual framework. Its key attribute or driver is organizational size, captured by the number of suppliers or subcontractors that are involved ([43]). For instance, Boston's "Big Dig"—a highway tunnel under the central part of Boston—involved more than 200 subcontractors ([42]). Similarly, developing the Boeing 737 required the involvement of hundreds of individual suppliers ([60]).
Graph: Figure 1. Conceptual framework.
Organizational size gives rise to different kinds of governance problems. Some of them have to do with controlling the actions of the individual suppliers or agents, as per established governance theory ([94]). For instance, the larger the temporary organization, the greater the difficulty of monitoring individual members' contributions. [77], p. 395) argues that as the number of involved partners increases, monitoring problems are exacerbated because of the cost of assigning accountability for individual actions ([ 2]).
Importantly, however, the governance challenges of temporary organizations go beyond managing individual actions per se; they also include ensuring consistency between the different parties' actions ([36]). Conceptually, the latter represents a particular form of governance problem, namely coordination ([88]). While researchers (e.g., [75]) have noted that coordination problems have received limited attention in the governance literature, they are ubiquitous in temporary organizations, in part because of interdependencies between the different agents' tasks ([65]). To use an obvious example, plumbing cannot begin until a building's foundation work has been completed.
As Figure 1 shows, the monitoring and coordination challenges that follow from organizational size require the deployment of governance mechanisms with particular properties. We draw on [37] framework to focus on the particular selection criteria and pricing provisions used. Both of these mechanisms possess the properties discussed previously, namely being readily deployable, formal, and organization-specific. With regard to selection, we distinguish between ( 1) ex ante assessments focused on supplier ability ([44]) and ( 2) evaluations that emphasize the price charged ([18]). With regard to the actual pricing provisions, we compare ( 1) variable pricing contracts, which reimburse the supplier for costs plus a margin, and ( 2) fixed pricing, which offers the supplier a prespecified price. These two formats possess different governance properties, in that the latter focuses strictly on the final output, while the former contains a built-in price adjustment mechanism ([ 3]; [13]).
The first part of our framework, which captures the governance deployment decision, specifies the relationship between organizational size and the two governance mechanisms. As Figure 1 shows, we also identify two different contingent influences, namely, ( 1) the temporary organization's time horizon, as indicated by the construction project's contracted duration, and ( 2) the presence of preexisting ties between the organization's members ([20]; [84]). The second part of the framework pertains to the performance implications of the governance choices made. Next, we consider each part of the framework in turn.
We argue that the main effect of organizational size, as measured by the number of suppliers involved, is to increase the likelihood of ability-based selection, because such a selection process proactively addresses monitoring and coordination problems. Consider first the problem of monitoring. As an example, the larger the number of individual suppliers involved in a building project, the greater the difficulty of monitoring their individual contributions ([59]; [69]). Extant governance research suggests that such problems can be managed through purposeful ex ante supplier selection on criteria such as skills, competencies, past performance, and reputation (e.g., [97]). In practice, such ability-based selection reduces the need for ongoing supplier handholding and follow-up. Essentially, having identified the "right" supplier in the first place economizes on ongoing monitoring efforts.[ 6]
Next, consider how organizational size gives rise to coordination problems. The larger the temporary organization, the greater the number of intraorganizational linkages that exist, and the greater the difficulty of coordinating the focal tasks ([80]). [54] discusses this issue in terms of the "O-ring property," where the organization's overall output requires that each input perform up to a certain level—if anything fails, the value of the project as a whole may be severely diminished. As an example, the size of the aforementioned "Big Dig" gave rise to extraordinary coordination problems between the different suppliers.
We draw on [36] to posit that the coordination needs that follow from organizational size can be addressed proactively through ability-based supplier selection. The more stringent the ex ante assessment of supplier ability, the greater the likelihood of identifying suppliers whose actions are unlikely to cause ongoing coordination problems. As noted by [18], the selection of such suppliers is an important way to curb potential supplier opportunism and facilitate coordination. Importantly, such information would not be revealed by a selection process that focused on the price charged.
As we have discussed, our baseline (main effect) expectation is that organizational size will increase the likelihood of an ability-based (as opposed to a price-based) supplier selection process. We posit, however, that the specific nature of this relationship is contingent on ( 1) a given organization's time horizon, as reflected in its contracted duration, and ( 2) the number of past collaborations between its key parties.
Consider first how contracted duration modifies the expected positive relationship between organizational size and ability-based selection. We expect the governance benefits of ability-based supplier selection to be the greatest for organizations that have a short contracted duration, where the parties have limited time to overcome size-related monitoring and coordination problems. In such situations, selection that emphasizes ability increases the likelihood that a given supplier will possess attributes that ensure frictionless interaction.
For longer-lasting collaborations, the ongoing interactions between the parties serve socialization purposes that ( 1) reduce the need for monitoring and ( 2) allow mutual learning that facilitates coordination. In turn, this reduces the need to explicitly select on ability. Instead, the relevant actions are induced through the organization's time horizon. Thus, although ability-based selection serves governance purposes for large temporary organizations in general, we expect the actual effect of size to be contingent on the contracted duration.
- H1: The positive effect of a temporary organization's size on ability-based selection is weakened as contractual duration increases.
Consider next the effect of an organization's past, as reflected in the number of prior collaborations between its key parties: the buyer, the general contractor, and the subcontractors. Many temporary organizations, due to their "one-off" nature, have idiosyncratic requirements and require that an organization be assembled from scratch. To the extent that the focal suppliers have not worked together before, they are, for all practical purposes, "interdependent strangers" ([68], p. 169). If so, the monitoring and coordination problems that follow from size will be significant; the parties will lack firsthand information about each other, and they cannot rely on established routines to manage their ongoing interactions. This, in turn, places a premium on ability-based selection to proactively mitigate the relevant governance problems.
Research suggests, however, that temporary organizations may involve repeated collaborations between its members (e.g., [21]). [20] specifically discusses the "persistent inter-firm relationships" ([43], p. 465) that sometimes exist in the construction industry. Importantly, however, prior ties may occur at different levels, namely between ( 1) the buyer and a general contractor, ( 2) the general contractor and a subcontractor, and ( 3) a buyer–general contractor–subcontractor. If prior collaborations have taken place between a fully matched set or "triad" of ties, it mitigates monitoring and coordination problems and reduces the need for costly and time-consuming ability-based selection. Essentially, firsthand observation and direct interaction on prior collaborations represent exogenous governance benefits ([35]; [77]), which reduce the need for organization-specific governance efforts in the form of purposeful selection. Previous research on teams suggests that prior collaborations create knowledge about "who is good at what," which helps promote partner-specific coordination routines ([52]; [84]). As one of our industry contacts stated, prior ties simply provide a "high level of certainty," as it "takes time to understand beliefs, how they [subcontractors] operate, how they like you to behave." Absent prior ties, however, the inherent monitoring and coordination challenges of large organizations will prevail, and ability-based selection will play a key role.
We stress that our arguments are based on a specific configuration of prior ties, namely a fully matched triad of parties. While [20] original thesis focused on the repetition of individual (dyadic) ties, we expect such a partial reconstruction ([16]) of a temporary organization to produce more limited governance benefits than a more comprehensive one. Stated differently, we expect that the specific form of prior ties matters. We propose the following hypothesis:
- H2: The positive effect of a temporary organization's size on ability-based selection is weakened as the number of prior collaborations increases.
Next, we introduce the main effect of organizational size on pricing and consider how size-related monitoring and coordination problems can be alleviated through the choice of pricing format. Similar to selection, pricing represents a "short-term marketing tool" ([45], p. 102) that can be readily deployed in a temporary organization context. However, as noted previously, there are important differences between the available pricing contracts, namely a fixed versus a variable format.
Consider first the pricing decision against the backdrop of an organization's monitoring problem. As we have discussed, for large organizations it is difficult to measure individual parties' contributions. We posit that variable pricing under such conditions provides inherent monitoring benefits because of the particular incentive structure it generates, which discourages shirking ([95]). Conceptually, variable pricing represents low-powered incentives because the supplier's actions are unrelated to their outcomes ([29]). In contrast, a fixed-pricing contract represents high-powered incentives because the supplier benefits directly and immediately from shirking. In essence, the incentive structure that is induced through variable pricing alleviates the need for ongoing monitoring in the first place ([77]).
Consider next pricing and the coordination problem. The greater the number of suppliers, the more interdependencies that must be managed, and the higher the likelihood that unilateral actions on the part of one supplier will impact others. [34] suggest that such coordination challenges can be managed through governance mechanisms that generate feedback and keep the parties informed of each other's actions on an ongoing basis. This can be achieved more readily through variable than fixed pricing, because a variable format by design involves regular checks on suppliers' costs and actions. [10], p.7) describe this in terms of "progress reports" that hold individual contractors accountable.
Importantly, variable pricing involves coordination benefits per se, but it also controls costs by preventing sizable and unexpected renegotiations ([ 3]; [13]). From a governance standpoint, variable pricing represents a built-in adjustment mechanism that provides flexibility when the organization's attributes (i.e., size) require it. As one of our industry contacts stated, "Variable pricing releases funds that are required to proceed with construction," which involves intensive and regular communication. In contrast, under fixed pricing, the focal point is the final output, and there is no built-in mechanism for managing the underlying process. Thus, all else being equal, we expect organizational size to promote the use of variable pricing.
As with selection, however, we expect the specific nature of this effect to be contingent on the organization's contracted duration. Conceptually, crafting a variable pricing contract means infusing a market relationship with hierarchical elements. In effect, variable pricing provisions generate communications that "will be taken as authoritative" ([86] p. 165), thereby facilitating coordination. At the same time, a variable pricing scheme is costly, time-consuming, and administratively burdensome, with regard to both its initial setup and ongoing communication requirements. Consequently, variable pricing is more easily justifiable when the contracted duration is long. Conversely, shorter duration makes it more difficult to justify hierarchical mechanisms, which may slow down communication and decision making. Thus, while organizational size motivates the use of variable pricing, as per the previous discussion, we expect this effect to be contingent on duration. We propose:
- H3: The positive effect of a temporary organization's size on variable pricing is strengthened as contractual duration increases.
Finally, consider the moderating effect of prior collaborations on pricing. As discussed in H2, given preexisting ties between all of the relevant parties (i.e., the buyer, the general contractor, and the subcontractors), certain monitoring and coordination benefits are supplied to the temporary organization exogenously. In turn, this reduces the need to specifically craft administratively burdensome monitoring and coordination mechanisms in large projects through variable pricing. Thus, we expect the general tendency of organizational size to promote variable pricing, as per our previous discussion, to be weaker the higher the number of prior collaborations between the relevant parties.
- H4: The positive effect of a temporary organization's size on variable pricing is weakened as the number of prior collaborations increases.
The second part of our conceptual framework describes a temporary organization's performance implications, as reflected in its cost overruns. Cost overruns represent a measure of supplier (non)performance, as expressed by the deviation between the project's actual and contracted cost.
Drawing on [ 7], p. 98) "plural forms" thesis about governance mechanisms as "building blocks," a nascent body of literature (e.g., [39]; [92]) shows how firms rely on combinations of governance mechanisms. While [ 7] original work failed to specify the exact nature of such combinations, subsequent research has operationalized the plural forms thesis by formally testing interactions between individual mechanisms ([11]; [78]). This body of research has significantly enhanced our understanding of plural governance, but extant studies have also limited their focus to the relationships among the mechanisms themselves, without accounting for the underlying attributes that motivate their deployment in the first place. This represents a limitation, especially when considered against the backdrop of transaction cost economics' principle of "discriminating alignment," which deals specifically with the relationships between governance mechanisms and their corresponding attributes.
Although prior studies (e.g., [27]) have examined the performance implications of discriminating alignment, they have limited their focus to single mechanism–attribute combinations. Extending the logic of discriminating alignment to the context of plural governance poses certain conceptual and (as we subsequently discuss) empirical challenges. Regarding the former, the specific challenges involve ( 1) articulating each mechanism's unique properties and ( 2) considering how a larger constellation of mechanisms and their respective attributes impacts performance.
With regard to selection, its main purpose is to identify suppliers who possess the right abilities or characteristics to complete a particular task. Although it seems intuitive that buyers should always select the most skilled and experienced suppliers (i.e., use ability-based selection), that may be neither necessary nor efficient for a relatively simple project. In fact, suppliers who are selected on the basis of such a process may be naturally inclined to conduct tasks or pursue project features that were not requested (or strictly required). In the project management literature, this phenomenon is referred to as "gold-plating" ([51]). In practice, it involves "overprovision" of quality ([83]), and it may be systematically induced by the particular supplier selection practices that were used in the first place. As such, we argue that only aligned selection, that is, the use of the particular selection strategy that is predicted by the focal organization's attributes, will enhance performance. Conversely, a failure to align will compromise performance.
In contrast to selection, the goal of pricing is to impact a supplier's motivation; the assumption being that a lack of motivation (or insufficient incentives) may compromise performance. As we have discussed, however, different types of pricing contracts (fixed vs. variable) impact motivation differently and must therefore be considered in the context of the focal organization's attributes.
While, in principle, transactional outcomes follow from the joint presence of motivation and ability ([67]), a full-fledged test of performance must account for the relevant drivers of each. In our framework, when the strategies used for selection and pricing are simultaneously aligned with their respective attributes, the necessary levels of supplier ability and motivation are both brought to bear on the temporary organization. This argument is logically consistent with both [ 7] "plural forms" thesis and [94] notion of discriminating alignment, but we expand on them by jointly accounting for multiple mechanisms and attributes.
In practice, because we study two governance mechanisms, we must account for 2 × 2 or four different scenarios. In principle, for a given temporary organization, all of the governance mechanisms can be properly aligned with their respective attributes (hereinafter called "joint alignment"), all can be misaligned (hereinafter called "joint misalignment"), or some combination of alignment and misalignment may occur. As a hypothetical example, for a large project with many suppliers, ability-based selection may be chosen (consistent with our predictions), and a fixed pricing format may be used (contrary to our predictions).[ 7] In general, when both selection and pricing are aligned with their respective attributes, the appropriate levels of supplier ability and motivation are both brought to bear on the focal project. Conceptually, the relevant "gaps" ([56]) have been filled, which should produce higher performance, as reflected in the smallest cost overruns. In summary,
- H5: Joint alignment of selection and pricing decreases ex post cost overruns.
H5 specifies how an appropriate joint deployment of selection strategy and pricing format will promote performance in the form of minimizing cost overruns. It does not, however, address the separate effects of selection and pricing on performance. For instance, does matched (or aligned) selection have a greater effect on performance than matched (or aligned) pricing? Stated differently, is there a greater payoff from having identified the "right" supplier than from administering the "right" incentive structure? From a resource allocation standpoint, this question has obvious practical implications: should a temporary organization prioritize selection over incentive design? In addition, this question is of considerable theoretical importance, since the two mechanisms represent fundamentally different approaches to relationship governance ([ 7]; [38]; [76]). Indeed, research suggests that selection and pricing reflect two different decision logics ([64]). As [41] note, selection efforts are inherently associated with a logic of appropriateness and the identification (or creation) of a "friend." Pricing, in contrast, is based on a "logic of consequences" and on inducing a "businessperson's" action through appropriately crafted incentives ([70]).
Consider the likely performance difference of aligned selection versus aligned incentives. Our specific expectation is that the beneficial effects that follow from ability (correct selection) will exceed those that follow from motivation (correct pricing). This is because finding the right party (and inducing a logic of appropriateness) represents a governance baseline or a performance prerequisite. In contrast, getting the incentives right (and inducing a logic of consequences) will contribute to performance, but its individual effect may be limited in the absence of the right partner abilities.
Beyond representing a performance baseline, we expect aligned selection to have a relatively greater performance effect because of its broader scope of influence. [41] note how selection, in addition to helping identify the "right" suppliers, may also serve a socialization purpose due to the interactions that take place during the process. As such, proper selection may impact supplier ability as well as motivation. In contrast, pricing is likely to have a narrower effect, involving only supplier motivation. In summary, with regard to the individual effects of the two mechanisms, we expect selection to have stronger governance properties than pricing. Drawing on the preceding arguments, we posit:
- H6: Alignment of selection (only) decreases ex post cost overruns more than alignment of pricing (only).
We test our research hypotheses using data on construction projects. These are complex from both a technical and a governance perspective, featuring a supplier (typically called a general contractor) who contracts tasks to other suppliers or subcontractors. Our data source is the Design-Build Institute of America (DBIA), whose stated purpose is the promotion of design-build project delivery in the United States. Given what the DBIA and its members deem the "critical ingredients" to delivering such projects, its members are asked to continuously submit structured information on completed projects—information that is then subject to verification. The DBIA database contains 429 completed projects spanning from 2001 through 2015 and represents contracted budgets ranging from $687,520 to $999,000,000 (https://www.dbia.org/resource-center/Pages/Project-Database.aspx, accessed June 2017).
Prior research points to the significant obstacles involved in obtaining data on construction projects. One particular constraint is that much of the relevant data are proprietary ([73]), which, as noted by [23], p. 73), tends to "keep [project] data from the hands of scholars." As a specific example, this obstacle required Flyvbjerg, Holm, and Buhl to spend four years collecting data on 258 transport infrastructure projects. Using the DBIA data enables us to conduct large-scale quantitative analyses using a rich variety of infrastructure, commercial, and industrial projects. We augment the rich DBIA data with data from the U.S. Census Bureau, the Construct Connect database, and metrics from Google. The unit of analysis is an individual construction project.
Table 1 summarizes the measures and data sources for all the study variables, and Table 2 presents the correlation matrix and descriptive statistics. We discuss each of the variables next.
Graph
Table 1. Variables and Data Sources.
| Variable | Conceptual Definition of Key Theoretical Variables | Measure | Data Source |
|---|
| Ex post cost overruns | Project (non)performance relative to the agreed-on project cost as per the formal contract. | A continuous variable indicating the percentage difference between the actual cost of the project and the contracted cost. | DBIA |
| Selection | A buyer's ex ante efforts to screen or verify a supplier prior to entering a relationship (Heide and John 1990). | A dichotomous variable indicating the main criteria on which the buyer selects the supplier(s) (0 = price-based selection, and 1 = ability-based selection). | DBIA |
| Pricing | The payment arrangement for the delivery of the project (Ghosh and John 2009; John 2008). | A dichotomous variable indicating the pricing terms guiding the project (0 = fixed pricing, and 1 = variable pricing). | DBIA |
| Project size | The size of the project organization. | Count of the number of suppliers included in the project. | DBIA |
| Contracted duration | The degree of time compression to which a project is subject. | The number of days between the start and end date of the project defined on the contract. | DBIA |
| Buyer–general contractor prior ties | | A count variable indicating the number of times a specific buyer–general contractor pair shares a prior tie. | DBIA |
| General contractor–subcontractor prior ties | | A count variable indicating the number of times a specific general contractor–subcontractor pair shares a prior tie. | DBIA |
| Buyer–general contractor–subcontractor prior ties | | A count variable indicating the number of times a specific triad of buyer–general contractor–subcontractor triad shares a prior tie. | DBIA |
| Project category | | A categorical variable classifying construction projects into commercial/institutional, industrial process facility, and civil infrastructure, with other projects being the base category. | DBIA |
| Environmental uncertainty | | The dollar value of U.S. construction spending regressed on the year. The resultant standard errors of the regression slope coefficient indicate environmental uncertainty. | U.S. Census Bureau |
| Number of state regulations | | A count variable indicating the number of rules and regulations that specify standards for constructed objects such as buildings and nonbuilding structures. | Construct Connect |
| Municipality | | A binary variable indicating whether a project is bought by municipality (= 1) or not. | DBIA |
| Nonprofit corporation | | A binary variable indicating whether a project is bought by a nonprofit corporation (= 1) or not. | DBIA |
| Developer | | A binary variable indicating whether a project is bought by a developer (= 1) or not. | DBIA |
| Buyer–project distance | | The distance (in miles) between the address of the buyer and the focal project. | Google |
| Government | | A binary variable indicating whether a buyer is government (= 1) or not. | DBIA |
| Defense | | A binary variable indicating whether a buyer is a defense organization (= 1) or not. | DBIA |
Graph
Table 2. Correlations and Descriptive Statistics.
| Variables | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 |
|---|
| 1. Ex post cost overrun | — | | | | | | | | | | | | | | | | |
| 2. Selection | −.074 | —a | | | | | | | | | | | | | | | |
| 3. Pricing | .066 | .275 | —a | | | | | | | | | | | | | | |
| 4. Project size | .029 | −.019 | .042 | — | | | | | | | | | | | | | |
| 5. Contracted duration | .445 | −.143 | −.118 | .101 | — | | | | | | | | | | | | |
| 6. Buyer–general contractor prior ties | −.035 | −.056 | −.039 | −.072 | .018 | — | | | | | | | | | | | |
| 7. General contractor–subcontractor prior ties | −.044 | −.095 | −.116 | .172 | .014 | .342 | — | | | | | | | | | | |
| 8. Buyer–general contractor–subcontractor prior ties | −.028 | −.087 | −.023 | −.000 | .049 | .611 | .408 | — | | | | | | | | | |
| 9. Project category | −.036 | −.034 | −.094 | −.067 | .156 | −.036 | −.106 | −.033 | —a | | | | | | | | |
| 10. Environmental uncertainty | −.002 | .014 | .099 | −.049 | −.190 | .165 | .050 | −.020 | −.111 | — | | | | | | | |
| 11. Number of state regulations | −.001 | −.011 | .129 | −.058 | −.103 | −.060 | −.133 | −.174 | −.001 | .147 | — | | | | | | |
| 12. Municipality | −.044 | .084 | .065 | −.020 | .057 | −.044 | −.096 | −.031 | .283 | .039 | .178 | —a | | | | | |
| 13. Nonprofit corporation | −.023 | .091 | .044 | .019 | −.017 | .038 | −.057 | .098 | −.102 | −.049 | −.024 | −.066 | —a | | | | |
| 14. Developer | −.002 | .026 | .108 | −.043 | −.035 | .023 | −.093 | .027 | −.058 | −.001 | −.011 | −.072 | −.036 | —a | | | |
| 15. Buyer–project distance | −.053 | −.086 | −.072 | −.016 | .083 | .049 | .058 | −.014 | .037 | −.073 | −.042 | −.062 | −.055 | −.000 | — | | |
| 16. Government | .041 | −.091 | −.137 | .134 | .134 | −.099 | −.095 | −.071 | .178 | −.057 | −.035 | −.237 | −.118 | −.129 | −.059 | —a | |
| 17. Defense | .061 | −.274 | −.216 | −.041 | .082 | .244 | .267 | .152 | −.104 | .019 | .061 | −.210 | −.104 | −.115 | .037 | −.376 | —a |
| M | 12.591 | .289 | .454 | 10.992 | 685.450 | .142 | 1.059 | .371 | 1.531 | 48,198.550 | 9.542 | .116 | .041 | .044 | 167.821 | .283 | .250 |
| SD | 74.092 | .454 | .499 | 10.873 | 389.747 | .404 | 2.950 | 1.633 | .941 | 2,407.653 | 3.020 | .321 | .199 | .205 | 796.167 | .451 | .433 |
| Minimum | −37.723 | .000 | .000 | 1.000 | 91.000 | .000 | .000 | .000 | .000 | 44,570.560 | 3.000 | .000 | .000 | .000 | .000 | .000 | .000 |
| Maximum | 1,046.207 | 1.000 | 1.000 | 142.000 | 4,856.000 | 3.000 | 24.000 | 13.000 | 3.000 | 51,865.810 | 15.000 | 1.000 | 1.000 | 1.000 | 7,039.000 | 1.000 | 1.000 |
- 30022242920982544 a The correlations in these columns are with binary or categorical variables and are therefore indicative only.
- 40022242920982540 Notes: n = 328. Correlations with an absolute value greater than.10 are significant at p <.05 (two-tailed).
"Ex post cost overruns" is defined as a project's level of (non)performance relative to the agreed-on project cost, as per the formal contract. It is measured in terms of the percentage deviation between the actual cost of the project and the contracted cost. Descriptive evidence suggests that 76.03% of projects experience cost deviations, of which overruns are the most common. The mean cost overrun is 21.79%, while the mean cost underrun is −4.91%.
"Selection" is defined as a buyer's ex ante efforts to screen or verify a supplier prior to entering a relationship ([40]). Our specific measure indicates whether buyers in their supplier selection emphasize ( 1) capabilities, skills, and past performance (ability-based selection) or ( 2) the price they offer on the project (price-based selection). Importantly, ability- and price-based selection are not mutually exclusive; the DBIA database reports on what criterion was the most important in a given situation. In our database, 71.09% of projects emphasize price-based selection, while 28.91% emphasize ability-based selection.
"Pricing" reflects the payment arrangement for the delivery of the project ([28]; [47]). Specifically, it captures whether the project relies on a fixed or a variable pricing contract. Fixed and variable pricing are grounded measures of contract design and incompleteness ([ 5]; [14]; [27]) and show whether the contract is subject to adjustments and renegotiations. Fixed pricing makes no allowance for adjusting the initial prices, while variable pricing allows for some mutually acceptable adjustment, either ex ante using adjustment formulas or ex post through negotiated adjustment ([27]). We captured fixed versus variable pricing using a single direct measure collected by the DBIA. Fixed prices were used for 54.55% of projects, while 45.45% of projects used variable pricing.
"Project size" captures the size of the project organization. The actual measure is a count of the number of suppliers involved in the project. The size of the project ranges from 1 to 142.
Our conceptual framework includes two moderators that describe different aspects of a project's time dimension. The first of these describes prior ties between the project's members ([20]). Specifically, we control for prior ties at three different levels: ( 1) "buyer–general contractor prior ties" is a count variable indicating the number of times a buyer has done business previously with the same general contractor, ( 2) "general contractor–subcontractor prior ties" is a count variable indicating the number of times a specific pair of general contractor–subcontractor share a prior tie, and ( 3) "buyer–general contractor–subcontractor prior ties", also as a count variable, indicates the number of times a specific combination of buyer–general contractor–subcontractors share a prior tie.
Second, we capture time through a project's "contracted duration." Conceptually, duration refers to a given project's degree of time compression. Our specific measure is the number of days between the project's contracted start and end dates. Shorter projects involve greater time compression, ceteris paribus, which has governance implications.
We include an extensive set of control variables. We account for the "project category" to absorb unobserved heterogeneity across different types of projects (including commercial/institutional, industrial process facility, civil infrastructure, and others). "Environmental uncertainty" describes the unpredictability of the project's environment. Following previous work in marketing (e.g., [81]), we use fluctuations of spending in the construction industry as a measure of "uncertainty" by ( 1) regressing total construction spending on the related years and ( 2) using the standard errors of the regression slope coefficient as a measure. The "number of regulations" refers to the number of building regulations in the project's state and may have different effects on cost overruns. For example, regulations may guide the conduct of the project and reduce cost overruns but may also complicate on-cost completion. In addition, we control for whether projects are purchased by "municipalities," which may favor local contractors. We also control for projects that are purchased by "nonprofit corporations" as these do not have shareholders and thus buyer value (and overruns) may not be their primary concern. We also control for whether the projects were purchased by "developers." A developer may not be a project's final owner but, rather, builds in anticipation of prospective buyers. This has potential implications for the governance mechanisms deployed and cost overruns incurred. "Buyer–project distance" is measured as the number of miles between the buyer and the project location ([ 8]). Using Google Maps, we extracted the geographical distance between a buyer and a project using the exact address of the buyers and the focal projects, as made available by the DBIA.
To test our hypotheses, our model specification must meet four different requirements. The first of these is the potential endogeneity of selection and pricing.[ 8] Second, our database consists of 429 projects across 291 buyers. Therefore, we need to account for clustering ([96]). Third, the drivers of the pricing, selection, and ex post cost equations are likely correlated due to unobserved factors. Fourth, selection and pricing are binary variables, and the ex post cost variable is a continuous metric.
We accommodate all of these requirements by employing [82] conditional (recursive) mixed-process (CMP) regression procedure, which uses a simulated maximum likelihood algorithm to estimate multiple equations simultaneously. To address concerns of endogeneity, we leverage CMP's ability to act as a control function estimator ([50]). This involves the use of instruments to explain selection and pricing, and the residuals of both equations are used subsequently as controls in the outcome equation. Our specification is consistent with prior use in governance research (e.g., [71]). The instruments used are government and defense projects. Per federal rules, such projects must feature fixed pricing and price-based selection, regardless of the project's attributes ([89]). This potentially makes them good instruments if they satisfy the relevance and orthogonality conditions. It is noteworthy that they are two distinct instruments, both theoretically and empirically (r = −.36). For instance, a government project such as the construction of a new library can attract a large set of potential bidders and is subject to federal (specifically, Federal Acquisition Rules, Chapter 1, Title 48), state, and potentially local rules. Defense projects, such as the construction of a new naval base, are also subject to federal rules, but a larger set of federal codes is involved (Code of Federal Regulations 10), and they may draw on fewer suppliers ([26]). We provide empirical evidence regarding the orthogonality and appropriateness of these instruments as part of our analyses.
Critical to our research is an empirical formulation of plural discriminating alignment. [94] classic alignment argument involves a single governance mechanism–attribute combination and suggests that appropriate matches have positive performance implications. Expanding on this argument, we argue that when firms deploy multiple governance mechanisms, each individual mechanism must be aligned with its particular attributes (i.e., the entire spectrum of variables included in the first stage [as per Equations 2 and 3]) for superior performance. When two governance mechanisms are jointly aligned with their hypothesized attributes, we refer to the resulting mechanisms–attributes constellation as plural discriminating alignment.
We previously identified the possibility of four different scenarios; one scenario in which both selection and pricing are aligned (predicted) by project attributes, a second in which both selection and pricing are misaligned, and two "off-diagonal" scenarios in which pricing or selection is misaligned but the counterpart is aligned. Thus, we construct a binary variable, joint alignment, for the empirical test. Following [58] and [72], we use a two-step procedure where, in the first step, we use a bivariate probit model to predict selection and pricing, using project attributes and instruments while mean-centering the key variables of interest and their lower-order terms for interpretation purposes, as follows:
P(pricingi=1 and selectionj=1)=Φ(β1Xi, β2Xi, ρ),1
where Φ is the cumulative bivariate standard normal distribution, β1 and β2 are the vector of estimated coefficients for project attributes, and ρ is the correlation coefficient between the residuals. Using the predictions from Equation 1, we construct a joint alignment variable that takes a value of 1 when the observed choices of selection and pricing are both identical to the predicted choices, and 0 otherwise. We observe 179 cases of joint alignment.
To test the hypotheses, we insert the joint alignment, aligned pricing, and aligned selection variables in the outcome equation of the CMP model. We use cluster-robust standard errors to allow for possible heteroskedasticity and clustering, which deals with potential correlations when the same buyer is observed more than once ([96]). Overall, the CMP model accounts for potential endogeneity of selection and pricing, as well as clustered observations, and simultaneously estimates pricing, selection, and ex post cost overruns (that are measured on different scales). We specify our model as follows:
Selectioni=β10+β11Project size+β12Contracted duration+β13Buyer-general contractor prior ties+β14General contractor-subcontractor prior ties+β15Buyer-general contractor-subcontractor prior ties+β16(Project size×Contracted duration)+β17(Project size×Buyer-general contractor-subcontractor prior ties)+β18,19,110Project category dummies+β111Environmental uncertainty+β112Number of state regulations+β113Municipality+β114Nonprofit corporation+β115Developer+β116Buyer-project distance+β117Government+β118Defense+∊1,2
Pricingi=β20+β21Project size+β22Contracted duration+β23Buyer-general contractor prior ties+β24General contractor-subcontractor prior ties+β25Buyer-general contractor-subcontractor prior ties+β26(Project size×Contracted duration)+β27(Project size×Buyer-general contractor-subcontractor prior ties)+β28,29,210Project category dummies+β211Environmental uncertainty+β212Number of state regulations+β213Municipality+β214Nonprofit corporation+β215Developer+β216Buyer-project distance+β217Government+β218Defense+∊2,3
Ex post cost overruni=β30+β31Joint alignment+β32Aligned pricing+β33Aligned selection+β34Project size+β35Contracted duration+β36Buyer-general contractor prior ties+β37General contractor-subcontractor prior ties+β38Buyer-general contractor-subcontractor prior ties+β39(Project size×Contracted duration)+β310(Project size×Buyer-general contractor-subcontractor prior ties)+β311-313Project category dummies+β314Environmental uncertainty+β315Number of state regulations+β316Municipality+β317Nonprofit corporation+β318Developer+β319Buyer-project distance+∊3.4
Table 3 contains the results of our CMP estimates. The results provide support for most of our key hypotheses. The significant Wald chi-square statistic of 198.28 (p <.01) demonstrates that the CMP model is significant. With regard to the deployment stage of our model, we find that the interaction between project size and contracted duration on selection is significant and negative (p <.05), consistent with H1. [ 1] recommendations, we conduct a simple slope analysis to gain a better understanding of the nature of this interaction. For this purpose, we tested the simple slopes at their minimum, low (−1 SD), mean, high (+1 SD), and maximum values of project contracted duration. Given the scaling of contracted duration, we thus cover the range and typical (minimum, −1 SD, mean, and +1 SD, maximum) values ([85]). Figure 2, Panel A, presents the moderating effect of contracted duration on the relationship between project size and selection. As the plot indicates, the effect is nonmonotonic over the range of project contracted duration. Specifically, the effect is positive at the minimum (p <.05) and −1 SD levels (p <.05), where time compression is the greatest, turning negative as time compression decreases or contracted duration is at its observed maximum (p <.05).
Graph
Table 3. CMP Regression Estimates.
| Variables | Selection | Pricing | Ex Post Cost Overruna (%) |
|---|
| β (Robust SE) | β (Robust SE) | β (Robust SE) |
|---|
| Joint alignment | | | (H5) −19.057 (9.648)** |
| Aligned pricing | | | 8.466 (7.558) |
| Aligned selection | | | −20.454 (7.669)*** |
| Project size | .014 (.008)** | .012 (.010) | −.884 (.565)* |
| Contracted duration | −.001 (.001)*** | −.000 (.000)* | .093 (.048)** |
| Prior Ties | | | |
| Buyer–general contractor prior ties | .520 (.292)** | −.145 (.398) | −19.754 (12.808)* |
| General contractor–subcontractor prior ties | −.043 (.086) | −.167 (.125)* | −.413 (4.061) |
| Buyer–general contractor–subcontractor prior ties | −.211 (.180) | .262 (.268) | −1.299 (6.150) |
| Interactions | | | |
| Project size × Contracted duration | (H1) −.000 (.000)** | (H3).000 (.000)** | .003 (.002)* |
| Project size × Buyer–general contractor–subcontractor prior ties | (H2) −.072 (.029)*** | (H4) −.011 (.051) | 2.979 (1.659)** |
| Control Variables and Instruments | | | |
| Project Category | | | |
| Commercial/institution building | .454 (.276)* | .005 (.322) | −4.2034 (10.489) |
| Industrial process facility | .458 (.293)* | .456 (.445) | 4.546 (13.934) |
| Civil infrastructure project | .617 (.301)** | −.167 (.384) | −20.859 (15.127)* |
| Environmental uncertainty | −.000 (.000)** | .000 (.000)* | .003 (.002)** |
| Number of state regulations | −.024 (.020) | .072 (.027)*** | .550 (1.203) |
| Municipality | .203 (.199) | −.282 (.289) | −16.855 (8.720)** |
| Nonprofit corporation | .285 (.270) | −.190 (.430) | −13.948 (13.211) |
| Developer | .086 (.234) | .266 (.404) | −7.153 (12.335) |
| Buyer–project distance | −.020 (.034)*** | .002 (.035) | −2.595 (1.858)* |
| Government | −.290 (.137)** | −.765(.222)*** | |
| Defense | −.604 (.244)*** | −1.066 (.270)*** | |
| Intercept | 1.505 (1.284) | −2.400 (1.649)* | −114.894 (91.4789)* |
- 50022242920982540 *p <.10.
- 60022242920982540 **p <.05.
- 70022242920982540 ***p <.01.
- 80022242920982540 a This variable is scaled as a percentage.
- 90022242920982540 Notes: Number of observations = 336. One-tailed significance if hypothesized. Robust standard errors are reported in parentheses.
Graph: Figure 2. Illustrations of the contingency effects of project size.Notes: One-tailed tests of significance.
H2 is also supported, as indicated by the negative and significant (p <.01) coefficient. We again rely on [ 1] approach. As shown in Figure 2, Panel B, we find that when buyer–general contractor–subcontractor prior ties are absent (p <.01) and at −1 SD (p <.01), the effect is positive. From +1 SD (p <.05) to the highest (p <.01) observed values of buyer–general contractor–subcontractor prior ties, this effect becomes negative and stronger.
Turning to H3, as we hypothesized we obtain a positive and significant (p <.05) interaction between project size and contracted duration on pricing. To understand if this effect is nonmonotonic, we again rely on [ 1]. Figure 2, Panel C, shows a negative and significant coefficient at the lowest (p <.1) values where time compression is strongest, in support of H3. From +1 SD (p <.01) to the highest (p <.01) values, this effect becomes positive.
As Table 3 shows, we do not find support for H4 (p >.1), which involved the expectation that the positive relationship between project size on variable pricing is weakened as the number of prior collaborations among all the involved parties increases.
To understand the managerial impact of the three significant results, we calculated the marginal effects when the contracted duration increases from one year to five years. These results show that the probability of choosing ability-based selection decreases by 6.39% (p <.1), while the probability of choosing variable pricing increases by 8.89% (p <.05). Similarly, we calculated the marginal effect for prior ties when these increase from zero (no prior ties) to five. The calculated impact shows that the probability of choosing ability-based selection decreases by 12.87% (p <.01).
Turning to our performance stage hypotheses, H5 is supported (β = −19.057; p <.05), which shows that the joint alignment of selection and pricing significantly decreases ex post cost overruns. We also find support for H6 (β = −2.686; p <.01), which shows that aligned selection (alone) decreases ex post cost overruns more than aligned pricing (alone).
We probe all scenarios that are raised by alignment of selection and pricing in more detail by considering ( 1) joint alignment, ( 2) aligned pricing and misaligned selection, ( 3) misaligned pricing and aligned selection, and ( 4) joint misalignment. Specifically, we calculated the mean predicted cost overrun for each scenario and carried out pairwise comparisons using Tukey's testing procedure. The results, as illustrated in Figure 3, indicate that the mean cost overrun of projects characterized by joint misalignment is 39.55%, which is significantly higher than the mean cost overrun of all other scenarios. Interestingly, no differences appear between the two scenarios that involve alignment on one mechanism (selection or pricing) and misalignment on the other (selection or pricing) (p >.10). We return to this finding in the "Discussion" section.
Graph: Figure 3. The costs of misalignment.Notes: This matrix is symmetric, and the lower triangle is a mirror of the upper triangle.
In addition, we find the correlation coefficient between the residuals of selection and cost overrun equations to be significant (p <.01), suggesting the appropriateness of applying an endogeneity-correcting procedure. However, the correlation coefficient of the residuals between the selection and pricing equations as well as the correlation coefficient associated with pricing and cost overrun equations are insignificant (p >.1).
Both government and defense projects significantly relate to selection and pricing, which suggests instrument relevance. To understand instrument relevance further, we consider instrument strength using the approach of [15]. For nonlinear models, such as those used to estimate selection and pricing, traditional indicators of instrument strength such as those developed by [87] are not applicable. Danaher et al. suggest comparing nonlinear models through the use of Akaike information criterion (AIC) and Bayesian information criterion (BIC) fit indices, where a "marked" increase in fit indicates good instrument strength. Heeding this approach, we estimated a model explaining pricing with and without these two instruments. Subsequently, we calculated the AIC and BIC fit statistics for both models and assessed whether fit had increased markedly (i.e., the ▵AIC and ▵BIC > 10). This was the case for both the selection and pricing estimates, which suggests that they are strong instruments. To assess the orthogonality of the instruments, we also estimated a separate model where we added government and defense projects to Equation 4. In this alternative model, the coefficients of government and defense are insignificant, suggesting that the instruments are orthogonal and thus satisfy an important criterion.
Hypothetically, a simple interaction between pricing and selection may explain ex post costs. We considered this possibility by adding the multiplicative interaction of selection and pricing to Equation 4. The parameter estimate was not significant (p >.1), suggesting that simple interactions cannot explain the more complex and nuanced effects our plural discriminating alignment thesis suggests. We return to the question of plural discriminating alignment in the "Discussion" section.
To complement our formal hypothesis tests, we conducted three in-depth interviews with senior managers in the construction industry. We selected these managers using an industry contact to ensure ( 1) manager experience, ( 2) willingness to engage, and ( 3) variation across project types (i.e., residential, government, large infrastructure, and commercial construction). As summarized in the Web Appendix, the three had significant experience with managing construction projects generally, as well as with the governance of subcontractor relationships. These managers' seniority allowed them to consider our questions against the backdrop of a wide range of construction projects, and to comment on the key aspects of our model. The interviews started with a brief introduction that outlined general aims, followed by a semistructured interview that took about 50 minutes and which focused on the four constructs of selection, pricing, prior ties, and cost overruns. Notes were taken, and from these we extracted salient aspects and representative quotes. We present our findings from these interviews in the Web Appendix and summarize the key insights next.
First, with regard to selection, the managers expressed views that support our contention that ability-based selection reduces subsequent coordination problems. Moreover, we learned that the benefits of ex ante assessments of ability go beyond pure screening; the selection process itself, including the interviews with the focal suppliers, serves a relationship-building (socialization) purpose. One interviewee specifically described selection in terms of "this is where the relationship resides." Thus, ability-based selection, while focusing on a set of tangible (ability-based) partner attributes, also establishes and promotes partner motivation.
Second, with regard to pricing, our interview data clearly show the managers' beliefs that variable pricing involves more ongoing communication than fixed pricing. One interviewer specifically noted the "back and forth" involved in this type of pricing contract.
Third, with regard to the role of prior ties, we learned that previous interactions help reduce ambiguity and, as noted by one manager, ensure that "parties are of the same mind." This, in turn, has governance implications for a new project, because prior ties reduce the need for mechanisms that "protect one's position."
Finally, the interviews suggested that the impact of cost overruns is both significant and multifaceted. Specifically, cost overruns have a significant impact on a contractor's reputation and ability to secure future work. Cost overruns also raise the possibility of subsequent conflict and costly litigation. Cost overruns may also, as a future safeguard, cause rigid and formalized relationships.
Marketing decisions and their outcomes depend crucially on the organizational context within which the focal decisions are made ([27]; [57]; [74]). Stated differently, organizational form matters. To date, however, marketing's focus has been limited to two forms: permanent organizations and long-term relationships. With some exceptions ([37]), very little attention has been given to a third, quite common form: temporary organizations. Recent reviews (e.g., [61]) note how fields such as strategy, operations, management, and engineering have started to build a literature on temporary organizations. To date, however, marketing has contributed little to our understanding of this important organizational form.
The importance of temporary organizations stems from their ability to ( 1) solicit inputs from a team of specialist suppliers, ( 2) deploy these inputs within a particular time frame, and ultimately ( 3) deliver outputs that would be unattainable for a single firm. In many respects, temporary organizations are uniquely suited to deliver customer value. At the same time, they pose significant challenges, as evidenced by their mixed performance record. For instance, Facebook and Apple's MobileMe initiative is widely considered to be a significant failure, despite the very significant investments made by both firms ([24]).
Our particular focus was on hybrid temporary organizations. Theoretically, such organizations raise interesting questions because of their particular constellation of governance problems and solutions. On the one hand, hybrids do not possess the extended time horizon of fully embedded temporary organizations that exist within a permanent firm and therefore must deploy governance mechanisms subject to time compression. However, unlike stand-alone temporary organizations that are assembled from scratch, hybrids may benefit from prior ties among its members.
We aimed to capture the unique features of hybrids through a two-stage conceptual framework that comprised ( 1) the governance deployment decision and ( 2) the resulting performance implications. An empirical test of the framework in the context of construction projects provided good support for our hypotheses. In the following subsection, we highlight our key empirical findings and their managerial implications. Next, we discuss theoretical implications and limitations and suggest future research topics.
Our primary goal with this study was to gain insight into temporary organizations' governance practices. This question was posed originally by [30] and framed in terms of how to "mobilize" a temporary organization. Since that time, however, this question has gone largely unanswered. Some researchers have asserted, largely without formal evidence, that temporary organizations are inherently unstable. Our findings cast doubt on this assertion. We show that temporary organizations have particular governance mechanisms at their disposal, and these mechanisms have significant and predictable effects on performance.
Importantly, however, given a temporary organization's unique time dimension, its governance mechanisms must possess particular properties. We focused on two broad categories of mechanisms: First, drawing on the "new institutional economics" literature, we considered organization-specific mechanisms that match a given organization's attributes ([95]). These mechanisms, however, must be capable of being deployed and of taking effect quickly. Our specific focus was on supplier selection strategy and pricing provisions. Our second category consisted of exogenous or preexisting mechanisms that could be activated and brought to bear on a new organization. Our specific focus was on the role of prior ties between the temporary organization's members.
Our conceptual framework was based in part on the juxtaposition between the two categories of mechanisms, including the possibility that the deployment of organization-specific mechanisms is contingent on the presence of prior ties. Our empirical findings provided good support for our framework, and in doing so refuted the idea that temporary organizations lack structure. They do show, however, that temporary organization governance is more complex than frequently assumed, and that it involves complex interactions between mechanisms at different levels.
Consider the governance mechanisms that we studied in more detail. As we have discussed, prior ties serve useful governance purposes due to the partner knowledge and social fabric that is retained from prior collaborations. This makes prior ties valuable, consistent with [33] "embeddedness" thesis. We show, however, that the specific form of prior ties matters crucially. Specifically, our results suggest that only a fully matched triad of existing ties—rather than a dyadic tie between an individual buyer and general contractor—provides discernible benefits. This finding also highlights the importance of broadening the focus beyond individual dyads ([55]; [90]).
Beyond shedding light on individual governance mechanisms and their properties, an important conclusion from this study is the importance of jointly accounting for combinations of mechanisms and their theoretically specific attributes—what we referred to as "plural discriminating alignment." Specifically, we show that selection and pricing, when jointly matched with their theoretically specified attributes, impact cost overruns. A simple, model-free analysis of our data, involving an analysis of variance test for fixed/variable pricing and price/ability-based selection on cost overruns, suggests no clear effects (F( 4, 360) =.91, p >.10). Yet our more advanced analyses, built on an operationalization of plural alignment, revealed a nuanced pattern of performance implications, where cost overruns followed from complex constellations of governance mechanisms and organizational attributes. From a theoretical standpoint, this both advances transaction cost economics' general discriminating alignment ([94]) argument and takes the "plural forms" thesis ([ 7]; [12]) to its logical conclusion.
From a managerial standpoint, the governance mechanisms that we studied (selection and pricing) represent decision variables that firms can readily deploy. However, their deployment is not straightforward. Our results show that controlling cost overruns is not simply a function of deploying multiple governance mechanisms per se. Rather, performance requires that all the governance mechanisms be aligned with their corresponding attributes. We consider the specific implications of selection and pricing in turn.
First, with regard to pricing, previous research in marketing has established the importance of pricing as a decision variable. However, the focus of past research has been on how pricing impacts supplier profit, not necessarily on specific aspects of buyer value such as cost overruns. Further, we show that it is not necessarily the pricing level that matters in creating buyer value, but the specific pricing format used. Interestingly, this suggests that the same price level can actually produce entirely different buyer outcomes. From the perspective of a supplier, managing overruns through appropriate choices on pricing format is equally important, because cost overruns impact supplier reputation and thus the likelihood of attracting future business.
The marketing implications of selection are significant. To buyers, managing cost overruns does not follow from selecting suppliers on ability or price per se, but by appropriately aligning selection with key organizational attributes. For instance, we show that less stringent selection, involving price only, can actually be beneficial, but only when matched to ( 1) particular attributes and ( 2) other governance mechanisms. This adds nuance to prior work that has recommended strict selection as a strategy ([17], [18]). This finding helps explain why firms may choose less stringent selection for reasons other than economizing on search costs ([71]).
In general, our current research helps expand the toolkit available to business-to-business marketers. From a practical standpoint, selection and pricing have attractive decision-making properties, because they can be deployed under time compression. At the same time, firms' choices with regard to selection and pricing have long-term effects. In our study, performance outcomes were effectuated months and even years after the initial governance deployment. As such, these governance choices are both durable and important. Conversely, we found that making the (theoretically) wrong governance choices given the prevailing attributes resulted in greater cost overruns. Importantly, the implications of overruns go beyond narrow financial metrics. Our interviews pointed to potential reputational damage, litigation, and future (over)reliance on rigid and formalized relationship features.
To generate fine-grained insights into the managerial importance of selection and pricing, we considered four specific performance scenarios: ( 1) joint alignment, ( 2) misaligned selection and aligned pricing, ( 3) aligned selection and misaligned pricing, and ( 4) joint misalignment. Although we established that scenario 1 is preferable from the perspective of minimizing cost overruns, our current research also demonstrates which mechanism (selection or pricing) matters more, through a "what-if" analysis involving counterfactual computations. These shed light on the performance implications of two scenarios 2 and 3 by comparing a chosen alternative with the outcomes obtained had a different selection or pricing choice been made.
As an observation of scenario 2, the construction of a power transmission line was governed by capability-based selection and variable pricing. In this project, selection was misaligned with our model-predicted choice while pricing was aligned. Per our model, had the project instead been governed by aligned selection and aligned pricing, the cost overrun would have been 37.57% lower. Conversely, in the renovation project of San Francisco International Airport (Terminal 2), as a manifestation of scenario 3, the observed choice of selection was aligned with our model-predicted choice, while the observed choice of pricing was misaligned. Again, we calculated the counterfactual cost overrun, which showed that if this project had used aligned selection and aligned pricing, the cost overrun would have been 14.93% lower.
These two examples show the explanatory power of selection and pricing as individual governance mechanisms, as reflected in the substantial reductions in cost overruns (37.57% and 14.93%, respectively) that potentially were available. Further, they show that selection (only) decreases ex post cost overruns more than alignment of pricing (only). Because the benefits from getting selection "right" outweigh those that result from getting pricing "right," it suggests that selection should be a higher strategic priority for a firm.
We close with a discussion of some research limitations and possible extensions. First, certain exogenous governance benefits may be afforded by national culture. Specifically, cultural variables may facilitate both coordination and monitoring, but our U.S.-centric database prevented us from examining this. Relatedly, [ 6] show how a temporary organization may benefit from industry roles that facilitate member coordination. General industry codes and professional certifications may serve similar purposes. It is noteworthy, however, that some researchers (e.g., [32]) have questioned such mechanisms' properties. While roles and codes may be capable of solving governance problems that involve coordination, they may possess constraints relative to more fundamental problems of cooperation ([36]).
Second, unanswered questions pertain to the effect of prior ties. While such ties may be beneficial, existing ties may come under strain, depending on the nature of an organization's task. If a temporary organization's task involves a meaningful degree of repetition relative to an earlier one ([16]), existing ties may indeed continue to serve governance purposes. In fact, our interviews revealed that prior ties help parties "read contracts correctly." However, radically new tasks may require a recalibration of past rules, and could, potentially, even be reflected in the selection deployed. If so, existing ties are associated with boundary conditions that diminish their value, and a new temporary organization may need to fall back on organization-specific (and short-term) mechanisms such as selection and pricing.
Third, the degree to which selection induces specific roles and behaviors among project members is difficult to discern using secondary data. Although we attempted to ascertain this through interviews, an experiment might generate more nuanced insights.
Fourth, future research could go beyond our current focus on cost overruns to examine additional performance outcomes—for instance, in the form of time delays, innovation, and financial returns. Similarly, future studies of temporary organization would also benefit from comparing the choice of governance and the resultant performance outcomes of stand-alone or fully embedded temporary organizations against hybrid ones.
Fifth, the focus of this study was on the ex ante resolution of ex post costs. However, such costs could be managed formally, for example through court ordering, or informally through renegotiation or private enforcement. Future work could usefully focus on such ex post resolution mechanisms and show the conditions under which they are effective.
Finally, given our anchoring in marketing, our main focus was on the buyer. However, projects feature multiple parties, including first- and second-tier suppliers. The impact of project governance on these particular parties is likely a fruitful avenue for further research.
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242920982545 - Mobilizing the Temporary Organization: The Governance Roles of Selection and Pricing
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242920982545 for Mobilizing the Temporary Organization: The Governance Roles of Selection and Pricing by Elham Ghazimatin, Erik A. Mooi and Jan B. Heide in Journal of Marketing
Footnotes 1 Mark Houston
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Online supplement https://doi.org/10.1177/0022242920982545
5 The average project completion time is two years. Thus, the implications of the different ex ante governance choices are quite durable.
6 As our industry interviews suggest, buyers' selection processes, due to the interactions that take place, also help establish and induce supplier motivation. We return to this question subsequently.
7 For reasons of collinearity, the performance outcomes of all four cells cannot be tested simultaneously.
8 In theory, size could be endogenous. However, it is not entirely clear how this would impact our dependent variable of ex post cost overruns, because size may be associated with competing predictions. For instance, having a small number of suppliers may reduce cost overruns by virtue of more limited communication needs. At the same time, a small organization may be more prone to overruns than a large organization given its limited slack resources. The implication of these competing expectations is that a buyer's strategy is not a given a priori, and the potential impact of endogeneity is unclear. Relatedly, because the choice of the number of suppliers is contingent on factors such as potential incompatibilities with other subcontractors, local availability of subcontractors, and various task features, any buyer wishing to "deploy" the number of suppliers would need a formidable degree of foresight—foresight we believe is lacking in practice.
References Aiken Leona S., West Stephen G. (1991), Multiple Regression: Testing and Interpreting Interactions. Thousand Oaks, CA: SAGE Publications.
Alchian Armen A., Demsetz Harold. (1972), "Production, Information Costs, and Economic Organization," American Economic Review, 62 (5), 777–95.
Bajari Patrick, Tadelis Steven. (2001), "Incentives Versus Transaction Costs: A Theory of Procurement Contracts," RAND Journal of Economics, 32 (3), 387–407.
Bakker Rene M., DeFillippi Robert J., Schwab Andreas, Sydow Jörg. (2016), "Temporary Organizing: Promises, Processes, Problems," Organization Studies, 37 (12), 1703–19.
Banerjee Abhijit V., Duflo Esther. (2000), "Reputation Effects and the Limits of Contracting: A Study of the Indian Software Industry," Quarterly Journal of Economics, 115 (3), 989–1017.
Bechky Beth A. (2006), "Gaffers, Gofers, and Grips: Role-Based Coordination in Temporary Organizations," Organization Science, 17 (1), 3–21.
Bradach Jeffrey L., Eccles Robert G. (1989), "Price, Authority, and Trust: From Ideal Types to Plural Forms," Annual Review of Sociology, 15 (1), 97–118.
Brickley James A., Dark Frederick H. (1987), "The Choice of Organizational Form: The Case of Franchising," Journal of Financial Economics, 18 (2), 401–20.
9 Burke Catriona M., Morley Michael J. (2016), "On Temporary Organizations: A Review, Synthesis and Research Agenda," Human Relations, 69 (6), 1235–58.
Calvo Eduard, Cui Ruomeng, Serpa Juan Camilo. (2019), "Oversight and Efficiency in Public Projects: A Regression Discontinuity Analysis," Management Science, 65 (12), 5651–75.
Cannon Joseph P., Achrol Ravi S., Gundlach Gregory T. (2000), "Contracts, Norms, and Plural Form Governance," Journal of the Academy of Marketing Science, 28 (2), 180–94.
Cao Zhi, Lumineau Fabrice. (2015), "Revisiting the Interplay Between Contractual and Relational Governance: A Qualitative and Meta-Analytic Investigation," Journal of Operations Management, 33/34, 15–42.
Corts Kenneth S., Singh Jasjit. (2004), "The Effect of Repeated Interaction on Contract Choice: Evidence from Offshore Drilling," The Journal of Law, Economics, and Organization, 20 (1), 230–60.
Crocker Keith J., Reynolds Kenneth J. (1993), "The Efficiency of Incomplete Contracts: An Empirical Analysis of Air Force Engine Procurement," RAND Journal of Economics, 24 (1), 126–46.
Danaher Peter J., Smith Michael S., Ranasinghe Kulan, Danaher Tracey S. (2015) "Where, When, and How Long: Factors that Influence the Redemption of Mobile Phone Coupons," Journal of Marketing Research, 52 (5), 710–25.
DeFillippi Robert, Sydow Jorg. (2016), "Project Networks: Governance Choices and Paradoxical Tensions," Project Management Journal, 47 (5), 6–17.
Dekker Henri C. (2004), "Control of Inter-Organizational Relationships: Evidence on Appropriation Concerns and Coordination Requirements," Accounting, Organizations and Society, 29, 27–49.
Dekker Henri C. (2008), "Partner Selection and Governance Design in Interfirm Relationships," Accounting, Organizations and Society, 33 (7/8), 915–41.
Dwyer Robert F., Schurr Paul H., Oh Sejo. (1987), "Developing Buyer-Seller Relationships," Journal of Marketing, 51 (2), 11–27.
Eccles Robert G. (1981), "The Quasifirm in the Construction Industry," Journal of Economic Behavior & Organization, 2, 453–75.
Faulkner Robert R., Anderson Andy B. (1987), "Short-Term Projects and Emergent Careers: Evidence from Hollywood," American Journal of Sociology, 92 (4), 879–909.
Flyvbjerg Bent. (2014), "What You Should Know About Megaprojects and Why: An Overview," Project Management Journal, 45 (2), 6–19.
Flyvbjerg Bent, Holm Mette K. Skamris, Buhl Søren L. (2003), "How Common and How Large Are Cost Overruns in Transport Infrastructure Projects?" Transport Reviews, 23 (1), 71–88.
Fortt Jon. (2011), "Why Apple's iCloud Announcement Matters," CNBC (June 6), https://www.cnbc.com/id/43294274].
Garemo Nicklas, Hjerpe Martin, Mischke Jan. (2015), "The Infrastructure Conundrum: Improving Productivity," Rethinking Infrastructure: Voices from the Global Infrastructure Initiative, 2, 60–67.
Garner Birtice A. (2009), "Alternative to Low Bid Selection in Air Force Reserve Military Construction: Approach to Best Value Procurement," dissertation, Georgia Institute of Technology, https://smartech.gatech.edu/handle/1853/29744.
Ghosh Mrinal, John George. (2005), "Strategic Fit in Industrial Alliances: An Empirical Test of Governance Value Analysis," Journal of Marketing Research, 42 (3), 346–57.
Ghosh Mrinal, John George. (2009), "When Should Original Equipment Manufacturers Use Branded Component Contracts with Suppliers?" Journal of Marketing Research, 46 (5), 597–611.
Gilliland David, Kim Stephen. (2014), "When Do Incentives Work in Channels of Distribution?" Journal of the Academy of Marketing Science, 42 (4), 361–79.
Goodman Lawrence Peter, Goodman Richard Alan. (1972), "Theater as a Temporary System," California Management Review, 15 (2), 103–08.
Goodman Richard A., Goodman Lawrence P. (1976), "Some Management Issues in Temporary Systems: A Study of Professional Development and Manpower—The Theater Case," Administrative Science Quarterly, 21 (3), 494–501.
Grabher Gernot. (2002), "The Project Ecology of Advertising: Tasks, Talents and Teams," Regional Studies, 36 (3), 245–62.
Granovetter Mark. (1985), "Economic Action and Social Structure: The Problem of Embeddedness," American Journal of Sociology, 91 (3), 481–510.
Gulati Ranjay, Lawrence Paul R., Puranam Phanish. (2005), "Adaptation in Vertical Relationships: Beyond Incentive Conflict," Strategic Management Journal, 26 (5), 415–40.
Gulati Ranjay, Singh Harbir. (1998), "The Architecture of Cooperation: Managing Coordination Costs and Appropriation Concerns in Strategic Alliances," Administrative Science Quarterly, 43 (4), 781–814.
Gulati Ranjay, Wohlgezogen Franz, Zhelyazkov Pavel. (2012), "The Two Facets of Collaboration: Cooperation and Coordination in Strategic Alliances," Academy of Management Annals, 6 (1), 531–83.
Hadida Allègre L., Heide Jan B., Bell Simon J. (2019), "The Temporary Marketing Organization," Journal of Marketing, 83 (2), 1–18.
Heide Jan B. (1994), "Interorganizational Governance in Marketing Channels," Journal of Marketing, 58 (1), 71–85.
Heide Jan B. (2003), "Plural Governance in Industrial Purchasing," Journal of Marketing, 67 (4), 18–29.
Heide Jan B., John George. (1990), "Alliances in Industrial Purchasing: The Determinants of Joint Action in Buyer-Supplier Relationships," Journal of Marketing Research, 27 (1), 24–36.
Heide Jan B., Wathne Kenneth H. (2006), "Friends, Businesspeople, and Relationship Roles: A Conceptual Framework and a Research Agenda," Journal of Marketing, 70 (3), 90–103.
Hollmer Mark. (2002), "Big Dig Is Big on Safety," Insurance Times, 21 (19), 1–3.
Holloway Samuel S., Parmigiani Anne. (2016), "Friends and Profits Don't Mix: The Performance Implications of Repeated Partnerships," Academy of Management Journal, 59 (2), 460–78.
Ireland R. Duane, Hitt Michael A., Vaidyanath Deepa. (2002), "Alliance Management as a Source of Competitive Advantage," Journal of Management, 28 (3), 413–46.
Jackson Barbara B. (1985), Winning and Keeping Industrial Customers. Lexington, MA: Lexington Books.
Jacobsson Mattias, Burström Thommie, Wilson Timothy L. (2013), "The Role of Transition in Temporary Organizations: Linking the Temporary to the Permanent," International Journal of Managing Projects in Business, 6 (3), 576–86.
John George. (2008), "Designing Price Contracts for Procurement and Marketing of Industrial Equipment," in Review of Marketing Research, Malhotra Naresh K., ed. Vol. 4. Bingley, UK: Emerald Group Publishing, Ltd.
Jones Candace, Hesterly William S., Borgatti Stephen P. (1997), "A General Theory of Network Governance: Exchange Conditions and Social Mechanisms," Academy of Management Review, 22, 911–45.
Kanter Rosabeth Moss. (1995), World Class: Thriving Locally in the Global Economy. New York: Simon & Schuster.
Kashyap Vishal, Antia Kersi D., Frazier Gary L. (2012), "Contracts, Extracontractual Incentives, and Ex Post Behavior in Franchise Channel Relationships," Journal of Marketing Research, 49 (2), 260–76.
Kautz Karlheinz. (2009), "The Impact of Pricing and Opportunistic Behavior on Information Systems Development," Journal of Information technology Theory and Application, 10 (3), 24–41.
Kellogg Ryan. (2011), "Learning by Drilling: Interfirm Learning and Relationship Persistence in the Texas Oilpatch," Quarterly Journal of Economics, 126 (4), 1961–2004.
Knoben Joris, Gössling Tobias. (2009), "Proximity in Temporary Organizations," in Temporary Organizations: Prevalence, Logic and Effectiveness, Kenis Patrick, Janowicz-Panjaitan Martyna, Cambré Bart, eds. Cheltenham, UK: Elgar, 155–70.
Kremer Michael. (1993), "The O-Ring Theory of Economic Development," Quarterly Journal of Economics, 108 (3), 551–75.
Kumar Alok, Heide Jan B., Wathne Kenneth H. (2011), "Performance Implications of Mismatched Governance Regimes Across External and Internal Relationships," Journal of Marketing, 75 (2), 1–17.
Lazzarini Sergio G., Miller Gary J., Zenger Todd R. (2004), "Order with Some Law: Complementarity Versus Substitution of Formal and Informal Arrangements," Journal of Law, Economics, and Organization, 20 (2), 261–98.
Lee Ju-Yeon, Kozlenkova Irina, Palmatier Robert. (2015), "Structural Marketing: Using Organizational Structure to Achieve Marketing Objectives," Journal of the Academy of Marketing Science, 43 (1), 73–99.
Leiblein Michael J., Reuer Jeffrey J., Dalsace Frédéric. (2002), "Do Make or Buy Decisions Matter? The Influence of Organizational Governance on Technological Performance," Strategic Management Journal, 23 (9), 817–33.
Li Dan, Eden Lorraine, Hitt Michael A., Ireland R. Duane, Garrett Robert P. (2012), "Governance in Multilateral R&D Alliances," Organization Science, 23 (4), 1191–1210.
Linn Allison. (2010), "Hundreds of Suppliers, One Boeing 737 Airplane," NBC (April 28), http://www.nbcnews.com/id/36507420/ns/business-us%5fbusiness/t/hundreds-suppliers-one-boeing-airplane/#.XisECUZKick.
Lundin Rolf A., Arvidsson Niklas, Brady Tim, Ekstedt Eskil, Midler Christophe, Sydow Jörg. (2015), Managing and Working in Project Society: Institutional Challenges of Temporary Organizations. Cambridge, UK: Cambridge University Press.
Lundin Rolf A., Söderholm Anders. (1995), "A Theory of the Temporary Organization," Scandinavian Journal of Management, 11 (4), 437–55.
Macneil Ian R. (1980), The New Social Contract: An Inquiry into Modern Contractual Relations. New Haven, CT: Yale University Press.
March James G. (1981), "Decisions in Organizations and Theories of Choice," in Perspectives on Organizational Design and Behavior, Van de Ven Andrew, Joyce William, eds. New York: John Wiley & Sons.
Mathieu John, Maynard M. Travis, Rapp Tammy, Gilson Lucy. (2008), "Team Effectiveness 1997-2007: A Review of Recent Advancements and a Glimpse into the Future," Journal of Management, 34 (3), 410–76.
Matta Nadim E., Askenas Ronald N. (2003), "Why Good Projects Fail Anyway,"Harvard Business Review, 81 (9), 109–14.
Merton Robert K. (1957), Social Theory and Social Structure. Glencoe, IL: The Free Press.
Meyerson Debra, Weick Karl E., Kramer Roderick M. (1996), "Swift Trust and Temporary Groups," in Trust in Organizations: Frontiers of Theory and Research, Kramer Roderick M., Tyler Thomas R., eds. Thousand Oaks, CA: SAGE Publications, 166–95.
Mishra Anant, Das Sidhartha R., Murray James J. (2016), "Risk, Process Maturity, and Project Performance: An Empirical Analysis of US Federal Government Technology Projects," Production and Operations Management, 25 (2), 210–32.
Montgomery James D. (1998), "Toward a Role-Theoretic Conception of Embeddedness," American Journal of Sociology, 104 (1), 92–125.
Mooi Erik A., Ghosh Mrinal. (2010), "Contract Specificity and Its Performance Implications," Journal of Marketing, 74 (2), 105–20.
Mooi Erik A., Gilliland David I. (2013), "How Contracts and Enforcement Explain Transaction Outcomes," International Journal of Research in Marketing, 30 (4), 395–405.
Mooi Erik A., Sarstedt Marko, Mooi-Reçi Irma. (2017), Market Research: The Process, Data, and Methods Using Stata. Singapore: Springer.
Moorman Christine, Day George S. (2016), "Organizing for Marketing Excellence," Journal of Marketing, 80 (6), 6–35.
Oliveira Nuno, Lumineau Fabrice. (2017), "How Coordination Trajectories Influence the Performance of Interorganizational Project Networks," Organization Science, 28 (6), 1029–60.
Ouchi William G. (1980), "Markets, Bureaucracies, and Clans," Administrative Science Quarterly, 25 (1), 129–41.
Oxley Joanne E. (1997), "Appropriability Hazards and Governance in Strategic Alliances: A Transaction Cost Approach," Journal of Law, Economics, and Organization, 13 (2), 387–409.
Poppo Laura, Zenger Todd. (2002), "Do Formal Contracts and Relational Governance Function as Substitutes or Complements?" Strategic Management Journal, 23 (8), 707–25.
Powell Walter W. (1990), "Neither Market Nor Hierarchy: Network Forms of Organization," Research in Organizational Behavior, 12, 295–336.
Puranam Phanish, Raveendran Marlo. (2013), "Interdependence and Organization Design," in Handbook of Economic Organization: Integrating Economic and Organization Theory, Grandori Anna, ed. Cheltenham, UK: Elgar, 193–209.
Raassens Néomie, Wuyts Stefan, Geyskens Inge. (2012), "The Market Valuation of Outsourcing New Product Development," Journal of Marketing Research, 49 (5), 682–95.
Roodman David. (2011), "Fitting Fully Observed Recursive Mixed-Process Models with CMP," Stata Journal, 11 (2), 159–206.
Rubin Paul H. (1990), Managing Business Transactions: Controlling the Cost of Coordinating, Communicating, and Decision Making. New York: The Free Press.
Schwab Andreas, Miner Anne S. (2008), "Learning in Hybrid-Project Systems: The Effects of Project Performance on Repeated Collaboration," Academy of Management Journal, 51 (6), 1117–49.
Spiller Stephen A., Fitzsimons Gavan J., Lynch John G.Jr, McClelland Gary H. (2013), "Spotlights, Floodlights, and the Magic Number Zero: Simple Effects Tests in Moderated Regression," Journal of Marketing Research, 50 (2), 277–88.
Stinchcombe Arthur L. (1985), "Contracts as Hierarchical Documents," in Organization Theory and Project Management, Stinchcombe Arthur L., Heimer Carol A., eds. Oslo: Norwegian University Press, 121–71.
Stock James H., Yogo Motohiro. (2002). "Testing for Weak Instruments in Linear IV Regression," National Bureau of Economic Research (accessed January 31, 2020), http://www.nber.org/papers/t0284.
Thompson James D. (1967), Organizations in Action: Social Science Bases of Administrative Theory. New York: McGraw-Hill.
Wang Chong, San Miguel Joseph G. (2013), "Are Cost-Plus Defense Contracts (Justifiably) Out of Favor?" Journal of Governmental & Nonprofit Accounting, 2 (1), 1–15.
Wathne Kenneth H., Heide Jan B. (2004), "Relationship Governance in a Supply Chain Network," Journal of Marketing, 68 (1), 73–89.
Wathne Kenneth H., Heide Jan B., Mooi Erik A., Kumar Alok. (2018), "Relationship Governance Dynamics: The Roles of Partner Selection Efforts and Mutual Investments," Journal of Marketing Research, 55 (5), 704–21.
Weber Libby, Mayer Kyle J., Macher Jeffrey T. (2011), "An Analysis of Extendibility and Early Termination Provisions: The Importance of Framing Duration Safeguards," Academy of Management Journal, 54 (1), 182–202.
Williamson Oliver E. (1975), Markets and Hierarchies: Analysis and Antitrust Implications. New York: The Free Press.
Williamson Oliver E. (1985), The Economic Institutions of Capitalism: Firms, Markets, Relational Contracting. New York: The Free Press.
Williamson Oliver E. (1991), "Comparative Economic Organization: The Analysis of Discrete Structural Alternatives," Administrative Science Quarterly, 36 (2), 269–96.
Wooldridge Jeffery M. (2003), Introductory Econometrics: A Modern Approach, 2nd ed. Mason, OH: Thomson South-Western.
Wuyts Stefan, Geyskens Inge. (2005), "The Formation of Buyer–Supplier Relationships: Detailed Contract Drafting and Close Partner Selection," Journal of Marketing, 69 (4), 103–17.
~~~~~~~~
By Elham Ghazimatin; Erik A. Mooi and Jan B. Heide
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 95- Navigating the Last Mile: The Demand Effects of Click-and-Collect Order Fulfillment. By: Gielens, Katrijn; Gijsbrechts, Els; Geyskens, Inge. Journal of Marketing. Jul2021, Vol. 85 Issue 4, p158-178. 21p. 8 Charts, 3 Graphs. DOI: 10.1177/0022242920960430.
- Database:
- Business Source Complete
Navigating the Last Mile: The Demand Effects of Click-and-Collect Order Fulfillment
Many retailers are rushing into the click-and-collect (C&C) format, where shoppers place orders online and pick up the goods themselves later. The authors study the demand implications of C&C and postulate how different ways of organizing this format—each with its own convenience features—appeal to households with different shopper characteristics. Using two data sets, each covering the introduction of two C&C fulfillment types by a major grocery retailer in a large number of local markets, the authors compare the impact of in-store fulfillment (pickup at existing stores), near-store fulfillment (pickup at outlets adjoining stores), and stand-alone fulfillment (pickup at free-standing locations). The authors find that the shift in online consumer spending significantly differs between the three order fulfillment types, as does the impact on total spending. No one order fulfillment type systematically dominates; the effects depend heavily on shopper characteristics. The study's results provide guidance on which C&C fulfillment type(s) to operate under what conditions and caution retailers not to take the easy in-store route routinely.
Keywords: brick and mortar; click and collect; distribution channels; grocery shopping; e-commerce; online; order fulfillment; retailing
Although online retail sales boomed over the past decade, grocery retail, representing almost 40% of global retail business, did not follow suit. In the United States, online grocery sales are growing, but still, only a modest 5.9% of grocery spending in 2019 occurred online. In Europe, online made up 7.7% of grocery sales in the United Kingdom, followed by France at 7.4%, with many European countries not exceeding 1% (http://retailinsight.ascentialedge.com).
Probably the biggest obstacle to the success of online grocery shopping is "the last mile" problem, as consumers want fast delivery of their groceries yet are reluctant to pay delivery fees ([22]). In response, retailers are launching a new format that can mark the tipping point for online grocery ([39]): "click and collect," also known as "buy online, pick up in-store," "drive-thru," or "drive."[ 6] Click-and-collect (C&C) shoppers place orders online and pick up the goods themselves later. Unlike home delivery, C&C often does not come with an extra fee, while prices online are the same as in-store.
Business analysts believe that C&C may become the road to online success in the grocery industry ([34]). In the United States, grocery retailers are racing to build C&Cs, with frequent reports of supermarkets across the country venturing into the format. Walmart set out to extend U.S. grocery pickup services to 3,100 stores by the end of 2019, while also Kroger and Albertsons are quickly growing their pickup points ([35]). The speedy expansion of the C&C format is, however, not necessarily indicative of its viability. Indeed, at the end of the day, the critical question is whether consumers will respond positively ([30]). This consumer response may well depend on the way order collection is organized ([35]). Realizing this, especially in the pioneering (European) countries, retailers have experimented with three different C&C types—from pickup at existing stores (the "in-store" type) to drive-thru outlets adjoining stores (the "near-store" type, also known as curbside pickup) to free-standing locations with dedicated warehouses (the "stand-alone" type).
A rigorous analysis of consumer response to these alternative fulfillment types is currently lacking, and we intend to fill this gap. Specifically, we address the following questions: ( 1) How does C&C order fulfillment affect shoppers' online spending and total spending at a retailer? ( 2) How do different types of C&C in the form of in-store, near-store, and stand-alone C&C affect these outcomes? and ( 3) What is the impact of these different C&C types for different types of shoppers? Using two unique data sets, each covering a major grocery retailer's rollout of two of the three C&C order fulfillment types in a large number of local markets, we find that the demand implications sharply differ between the three types.
Our substantive contribution is threefold. First, despite the almost daily business reports about the potential importance of the C&C format, it has been glossed over in academic research. The single exception is [40], who home in on the informational function of a "buy online, pick up in-store" format of a consumer durables retailer and find that it mostly serves to inform consumers about item availability before visiting the store. We complement [37] by focusing on the fulfillment function of the C&C format and analyzing how different ways of organizing C&C fulfillment may affect online consumer demand. Although order fulfillment has been recognized in academic research as a critical channel function ([10]) and in the business press as a crucial differentiator that ought to be prioritized ([49]), it has rarely been studied.
Second, we lay out the different "channel benefits" ([25]) or "service outputs" ([56]) of (alternative types of) C&C. Building on the shopping convenience typology of [54], we argue that the three order fulfillment types differ in terms of the access, collection, and adjustment convenience they offer to shoppers. We then show how the C&C types are valued differently by shoppers, depending on their need or desire for these convenience features.
Third, we recognize that the impact of C&C is not limited to online and study how the C&C types influence shoppers' total (i.e., online plus brick-and-mortar) spending at the retailer, and how this extrapolates to retailer revenues in different local markets. As such, we provide a more comprehensive picture of how retailers' C&C operations affect their overall performance.
Managerially, while "e-commerce is transforming business...for supermarket companies," even the bigger players (e.g., Kroger in the United States) "struggle with the online grocery upheaval" ([43]). Our findings may provide grocery retailers with insights on how to avoid mistakes when jumpstarting the introduction of C&C. As retailers are racing to build C&Cs, they are mostly opting for fulfillment within existing stores for the sake of quick, low-cost rollout. Indeed, because in-store C&C can rely on existing infrastructure and processes, it is the easiest to implement ([20]). However, the pursuit of speed without knowing which type is best in terms of demand may lead to the demise of the format. Besides, while most retailers tend to opt for one type of C&C across all markets, we find that a one-size-fits-all approach is not advisable. Instead, the impact depends on shoppers' needs for fulfillment convenience and the retailer's focal performance metric. Our study may help retailers in finding the right mix.
Retailers launch C&C as an additional channel, next to their existing brick-and-mortar and home-delivery channels, in an attempt to increase online sales by tapping into consumers' ever-increasing convenience needs ([15]). We first compare the convenience benefits of the C&C format to a retailer's prevailing brick-and-mortar and home-delivery channels. Next, we introduce the alternative C&C types—in-store, near-store, and stand-alone C&C—and their distinctive convenience features. We then argue how shoppers value the C&C types differently depending on their need for these convenience features, and how this affects their online spending at the retailer. We conclude by briefly reflecting on the impact of C&C on alternative performance indicators.
[54] offer a framework that distinguishes different features of shopping convenience: access convenience (i.e., the ease of reaching a retailer), transaction convenience (i.e., the ease of effecting and amending transactions at a retailer), search convenience (i.e., the ease of identifying products at a retailer), and possession convenience (i.e., the ease of obtaining desired products from a retailer). In a C&C setting, where shoppers place orders online and pick up the goods later, access convenience pertains to the time to, at, and from a pickup location, while transaction convenience comprises two distinct aspects: the physical effort to collect the order at the pickup point (which we refer to as collection convenience) and the ease with which shoppers can adjust their online orders, by adding items ("top-up shopping"), or returning or replacing them, upon pickup (denoted by adjustment convenience). Search convenience captures the ease of identifying and selecting groceries online. In a C&C context, possession convenience is mainly subsumed within search convenience, because shoppers are informed about a product being in or out of stock at the time they place their order.
Compared with brick-and-mortar shopping, C&C may enhance consumers' search and collection convenience by allowing them to select and order the desired items online and by having the order physically prepared for them by store personnel. Moreover, to the extent that C&C can take effect through dedicated lanes in easy-to-reach locations, it may come with higher access convenience than brick-and-mortar shopping.
Compared with the home-delivery channel, C&C may offer more access-convenience benefits. While home delivery provides consumers with the advantage of not having to leave their houses, they do have to wait at home for their order to be delivered. In contrast, C&C may allow consumers to reduce shopping time without having to wait. Besides, to the extent that C&C allows for top-up shopping and item replacement, it may offer more adjustment convenience than home delivery.
By better catering to consumers' convenience needs, C&C is expected to increase shoppers' online spending at the retailer. First, previous brick-and-mortar shoppers (at the focal retailer's or competing retailers' brick-and-mortar channels) may shift their purchases online ([ 7]). Second, shoppers at competing retailers' home-delivery channels may switch to the more convenient retailer's C&C channel. Third, C&C might even increase shoppers' online spending at the retailer through its effect on their total grocery spending, because of the time savings (which may lead to extra trips; [15]) and the reduced shopping effort (which may enhance spending per trip; [55]). As [15] indicates, travel time is a fixed cost, and consumers will only travel to a store when their basket size is large enough to amortize this cost. Thus, as the time cost drops, households will shop more often. Moreover, by shifting the task of picking and packing the order to the store personnel, C&C reduces a shopper's variable shopping cost, which is known to enhance spending per trip ([11]). These changes in shopping patterns may create a "primary-demand effect": they may increase shoppers' home inventories and, as a result, their consumption ([ 2]; [ 3]; [21]; [61]) and total grocery spending ([42]).
Click and collect can be organized in three major ways. In an in-store C&C, store employees pick and pack ordered items and make the order available at the traditional store location. Shoppers are required to go into the store for pickup rather than wait in the car. In a near-store C&C, the fulfillment center is colocated at a retail store. Shoppers pull up to a drive-thru area with a roof overhead, where they find a touch screen kiosk to alert store staffers. The groceries are then loaded up by retail employees, without the shoppers ever having to leave their car. A stand-alone C&C consists of a warehouse and collection point located wholly separate from the grocer's retail store, but at a convenient site (e.g., a commuting route, a parking lot, or an office block). Professional pickers assemble orders from a dedicated fulfillment center or black store. Shoppers drive up and have the order delivered to their car.
Table 1 summarizes how these C&C types rate on the convenience features. Search and possession convenience pertain to the informational function of C&C and are the same across in-store, near-store, and stand-alone C&C types. In contrast, access, collection, and adjustment convenience originate from the fulfillment function of C&C, and differ across the three C&C types. Consequently, we propose that consumer response to the alternative C&C types rests on a trade-off between these three fulfillment-convenience features.
Graph
Table 1. How C&C Delivers Convenience.
| Convenience Feature | Shopping Elements | Type of C&C Format |
|---|
| In-Store | Near-Store | Stand-Alone |
|---|
| Search convenience: Shopper ease of identifying and selecting groceries online | Search and order online | Yes | Yes | Yes |
| Access convenience: Shopper ease in traveling to, at, and from a pickup location | No need to wait at home | Yes | Yes | Yes |
| Dedicated pickup lane | No | Yes | Yes |
| Easy-to-reach location | No | No | Yes |
| Collection convenience: Shopper ease of collecting the order at the pickup location | Store staff pack order | Yes | Yes | Yes |
| Store staff load car | No | Yes | Yes |
| Adjustment convenience: Shopper ease in adding or changing items to online order upon pickup | Ability to add, return, or replace items | Yes | Yesa | No |
50022242920960430 a The possibility to add, return, or replace items is present but requires more effort from the shopper than in-store C&C.
In terms of the differences between the C&C types, stand-alone C&C excels in terms of access convenience, followed by near-store C&C, with in-store C&C being the inferior option. Stand-alone C&Cs typically open in easy-to-reach locations that are part of consumers' daily journeys, with no congestion from consumers driving up to the regular brick-and-mortar store. Near-store C&Cs are not located in select, easy-to-access locations but still avoid congestion through dedicated pickup lanes. Stand-alone and near-store C&Cs also outperform in-store C&C in terms of collection convenience by eliminating the physical burden of handling shopping baskets. In addition, consumers are not limited to what they can physically carry when using stand-alone or near-store C&C, as their orders are lifted into their cars. Finally, near-store and in-store C&C offer more adjustment convenience than stand-alone C&C, by allowing shoppers to return or replace unsatisfactory products on the spot and to top up on their purchases at the retailer's on-site brick-and-mortar store. Especially with in-store C&C, top-up shopping is feasible with little extra effort.
No C&C type universally excels on all three fulfillment-convenience features. As argued in classic channel texts ([25], p. 36; see also [56]), "the very basis of all channel strategies lies in an understanding of the benefits that end-users desire in how they want to buy." Thus, which of the C&C types yields the highest consumer utility—and therefore demand—depends not just on the convenience features they offer (the supply side) but also on how a specific shopper, with certain shopper characteristics, values each bundle of convenience features (the demand side). Next, we argue how the relative demand effects of the three C&C types vary with shopper characteristics that reflect their fulfillment-convenience needs. Table 2 summarizes our expectations.
Graph
Table 2. Framework Predictions.
| Shopper Characteristic | Impact on Convenience | Click and Collect Type's Rating on Convenience Features | Expectations: Effect on Online Spending Lift |
|---|
| In-Store | Near-Store | Stand-Alone |
|---|
| Rural shoppersWeekend shoppers | Access convenience is more important | Low | Medium | High | For (1) rural shoppers and (2) weekend shoppers, stand-alone shopping increases online spending at the retailer more than near-store shopping, and near-store shopping increases online spending at the retailer more than in-store shopping. |
| Basket sizeBulkiness | Collection convenience is more important | Low | High | High | For shoppers who buy (1) more items and (2) more bulky items per shopping trip, stand-alone and near-store shopping increase online spending at the retailer more than In-Store shopping. |
| Household sizePerishabilityImpulse nature | Adjustment convenience is more important | High | Medium | Low | For shoppers with (1) larger families, (2) larger basket shares of perishables, and (3) larger basket shares of impulse goods, in-store shopping increases online spending at the retailer more than near-store shopping, and near-store shopping increases online spending at the retailer more than stand-alone shopping. |
Consumers that live in more rural markets and that shop mostly on weekends are more likely to value access convenience. As such, online spending at the retailer may increase if these consumers shift from brick-and-mortar shopping (at the focal or competing retailers) and/or home-delivery shopping (at competing retailers) to the more access-convenient stand-alone shopping at the focal retailer. Rural shoppers may switch as they typically have to travel a longer distance to shop at brick-and-mortar stores and may, therefore, save on driving or waiting time. The same holds for consumers who have higher opportunity costs of time, typically those who shop on weekends and face more crowded stores and congested parking lots ([57]). In addition, the convenient access of stand-alone C&Cs (on commuting routes, near office blocks, near highway exits) may lead them to incur extra trips ([15]), as they can stop for pickup on their way home from work with hardly any time loss. These extra trips may spur primary demand, by keeping up continuous and plentiful supply in consumers' home pantry and, as a result, expanding their consumption ([ 2]; [ 3]). Thus, we expect that for rural and weekend shoppers, stand-alone shopping lifts online spending at the retailer more than near-store and, especially, in-store shopping.
Large-basket shoppers and shoppers who buy more bulky items are more likely to value collection convenience. As such, they may switch from brick-and-mortar shopping (at the focal or competing retailers) to near-store and stand-alone shopping (at the focal retailer) to reduce the physical onus of shopping. Moreover, the lower physical effort that comes with near-store and stand-alone shopping may stimulate these shoppers to increase their order sizes by purchasing higher quantities ([ 2]), which has been shown to increase consumption ([ 9]; [21]; [61]), especially in the case of large package sizes ([60]). Thus, we expect that for shoppers with larger and bulkier baskets, online spending at the retailer is higher with C&C types that excel in facilitating transactions (i.e., with stand-alone and near-store compared with in-store C&C).
Larger families and shoppers with more perishable and impulse items in their baskets are more likely to value adjustment convenience. As such, they are more likely to switch from home delivery (at competing retailers) to the focal retailer's near-store and especially in-store C&C.[ 7] Ceteris paribus (which includes keeping basket size constant), larger households have greater variety needs to satisfy the heterogeneous tastes of multiple household members ([ 2]). As they face more complex shopping tasks, they are also more likely to forget items. Near-store and especially in-store C&C allow them to top up in case of forgotten family needs. In a similar vein, consumers who buy more perishable goods—whose expiration dates and freshness can vary a great deal ([24])—are more likely to shift their purchases to the C&C types that offer adjustment convenience (i.e., in-store and near-store C&C). Moreover, in-store's and near-store's option to easily replace unsatisfactory items on the spot decreases the risk to include products of varying quality in their online orders, which may increase their spending per trip. In addition, shoppers who tend to spend larger shares of their baskets on impulse goods may be more attracted to in-store than near-store and stand-alone C&C because of the ability to fulfill their impulse needs when collecting their orders. In all, we expect that larger families and shoppers with more perishable and impulse items in their baskets spend more online at the focal retailer with in-store than with near-store and, especially, stand-alone C&C.
Studying the effect of the C&C types on consumers' online spending helps shed light on whether C&C may be the route to online success. However, it provides an incomplete account of the implications for the retailer. On the one hand, C&C shopping may lower (i.e., cannibalize) consumers' brick-and-mortar spending ([ 7]) and thus total spending at the retailer. Near-store and in-store C&Cs in particular may constitute a mere shift away from the brick-and-mortar store, because the C&C pickup location is the same as the retailer's brick-and-mortar store. At the same time, stand-alone shopping may lead to more systematic/planned buying and less impulse buying. This may not just shift consumer purchases away from the retailer's brick-and mortar-stores to the C&C format but also reduce consumers' order sizes, thereby attenuating shoppers' total spending lift.
On the other hand, C&C shopping might increase consumer spending at the retailer's brick-and-mortar stores and thus enhance shoppers' total spending lift. The adjustment convenience offered by in-store and, to a lesser extent, near-store C&C may create positive spillovers. Once on site to pick up their online orders, in-store and near-store shoppers can engage in top-up shopping at the retailer's same-site brick-and-mortar store ([ 6]). Moreover, in-store C&C enables shoppers to impulse-buy in response to in-store offers, rather than only use the store as a destination for collection. Extant research has shown that when consumers make more planned purchases (which they do through online ordering), they may balance that type of self-control with other, more indulgent purchases during the same trip ([42]).
Which of these forces prevails is an empirical question that may again be contingent on shopper characteristics. As a corollary, the C&C type that performs best in boosting consumers' online spending is not necessarily the one that performs best in terms of increasing consumers' total spending, nor the retailer's market revenues. We therefore also investigate the impact of the C&C types on C&C shoppers' total (i.e., online plus brick-and-mortar) spending at the retailer and how this extrapolates to retailer revenues in different local markets.
In France, Europe's largest grocery market and the birth nation of C&C, retailers are extensively experimenting with alternative C&C types ([33]). Ideally, we would study the differences between in-store, near-store, and stand-alone C&C within a retailer, as this allows us to control for price- and assortment-positioning differences between retailers and concentrate squarely on the differences between the three C&C types.[ 8] However, no retailer operates all three types. Still, two leading retailers each use two types: Intermarché uses in-store and near-store C&C, while Leclerc uses both near-store and stand-alone C&C. To ensure that we do not confound interretailer with C&C-type differences, we analyze the data by retailer.
The two retailers studied are quite similar. Leclerc is the second-largest French retailer, accounting for 14.4% of the grocery market in 2017, operating a network of 1,779 stores (mainly hypermarkets), and generating revenues of approximately €44 billion. Intermarché is the third-largest French retailer, representing 12.8% of the market with 2,806 stores (mostly supermarkets) and revenues of approximately €33 billion (www.planetretail.net). Leclerc, the number-one C&C player in France, opened its first pickup point in 2007 and rolled out the format over the next few years. Intermarché, the number-three C&C player in France, opened its first C&C in 2008. The price positioning of both retailers is below the market average ([ 1]), with Leclerc perceived as somewhat lower in price than Intermarché ([29]). Both retailers maintain a uniform pricing policy: prices in the C&C format and in the brick-and-mortar stores in a given market are identical and, thus, do not affect consumers' channel choice (for a similar argument, see [22]]). Neither retailer adds a service charge.
We use scanner panel data from Kantar France. The data cover 91 four-week periods between 2008 and 2014. We complement these data with information from the marketing research companies Experian and A3Distrib to identify the opening date of each C&C location, its type (in-store, near-store, or stand-alone C&C), and the local market in which it operates.[ 9] We focus on markets where the focal retailer operates only one of the three C&C types (which represent over 95% of all markets), and where the first opening by that retailer occurred during our observation window. To allow for a one-year period to assess postadoption effects, we only consider local markets where C&Cs opened before January 2014, resulting in 101 near-store and 82 stand-alone markets for Leclerc and 66 in-store and 212 near-store markets for area Intermarché. Next, we retain households living in these markets that are consistently active in the panel throughout our observation window, which leads to 3,674 and 3,555 households in Leclerc's near-store and stand-alone markets and 2,224 and 7,667 households in Intermarché's in-store and near-store markets. Of the Leclerc households, 10.6% and 11.8% started shopping at near-store and stand-alone C&Cs, respectively, whereas 3.6% and 3.8% of the Intermarché households started shopping at in-store and near-store C&Cs.[10]
The dependent variable is online spending (OnlineSpend) at the retailer. It includes all household purchases ordered on the retailer's website within a four-week period.
The dummy variable C&C shopper is one from the time a household adopts C&C shopping and zero otherwise. Households that were merely attracted once (e.g., by trial promotions) but then backed out are not marked as C&C shoppers (for a similar practice, see, e.g., [45]] and [58]]). The dummy variable Type indicates which C&C type a retailer operates in a local market; it equals 0 for near-store C&C, and 1 for the other type operated by the retailer (i.e., stand-alone C&C for Leclerc, in-store C&C for Intermarché).
Rural shopper (Rural) is measured by a dummy variable that indicates whether a household lives in a rural market (1 = rural market, 0 = other), while weekend shopper (Weekend) is operationalized as the fraction of a household's shopping trips that are made during weekends. Basket size (Basket) is measured as a household's average number of items purchased per shopping trip. Bulkiness (Bulky) is the average number of bulky items in a household's shopping basket. Household size (Hhsize) captures the number of members in the household.[11] Finally, perishability (Perish) and impulse nature (Impulse) are measured as the fraction of a household's shopping basket that is spent on perishables and impulse-natured goods. All moderator variables are measured in the year before the C&C opened and calculated across all retailers. Table 3 provides a detailed description of the operationalizations.
Graph
Table 3. Measurement.
| Variable | Data Source | Operationalization |
|---|
| Dependent Variablesa |
| OnlineSpend h, l, c, r, t | Kantar France | Online spending (including home delivery and C&C) by household h, living in local market l with C&C type c, at retailer r and time t (in euros). |
| OfflineSpend h, l, c, r, t | Kantar France | Brick-and-mortar spending by household h, living in local market l with C&C type c, at retailer r and time t (in euros). |
| TotalSpend h, l, c, r, t | Kantar France | Total (i.e., online plus brick-and mortar) spending by household h, living in local market l with C&C type c, at retailer r and time t (in euros). |
| Focal C&C Variables |
| CCShopper h, l, r, t | Kantar France; Experian | Step dummy variable that equals 1 from the time t household h starts shopping at a C&C operated by retailer r in local market l. We only consider shoppers with at least one repeat purchase; 0 otherwise. |
| Type l, c, r | Kantar France; Experian | Dummy variable that equals 1 if retailer r operates C&C type c (c = in-store or stand-alone; near-store = reference case) in local market l, and 0 otherwise. |
| Type_Open l, c, r, t | Kantar France; Experian | Step dummy variable that equals 1 if C&C type c (c = in-store or stand-alone; near-store = reference case) is already opened by retailer r in local market l at time t, and 0 otherwise. |
| Moderator Variables: Shopper Characteristicsb, c |
| Ruralh,0 | Kantar France | Dummy variable that equals 1 if household h lives in a rural market, and 0 otherwise. Data provider Experian classifies local markets in nine types, reflecting different levels of urbanization. We code the two types "Rural attractive" and "Rural in decline" as "rural markets." |
| Weekend h,0 | Kantar France | Fraction of household h's shopping trips executed during weekends (at any retailer). |
| Basket h,0c | Kantar France | Average number of items purchased per shopping trip by household h (at any retailer). |
| Bulkyh,0c | Kantar France | Average number of bulky items in household h's shopping basket (at any retailer). |
| Hhsizeh,0 | Kantar France | Number of members in household h. In Equation 1, we partial out the effect of basket size by regressing household size on basket size and using the residual as the regressor. |
| Perishh,0d | Kantar France | Fraction of household h's shopping basket (at any retailer) spent on perishable items. |
| Impulseh,0d | Kantar France | Fraction of household h's shopping basket (at any retailer) spent on impulse-natured items. |
| Control Variables | | |
| Own C&Cs l, r, t | Experian | Number of C&Cs operated by retailer r in local market l at time t. |
| Competing Stand-Alone C&Cs l, r, t | Experian | Number of stand-alone C&Cs operated by retailer r's competitors in local market l at time t. |
| Competing Near-Store C&Cs l, r, t | Experian | Number of near-store C&Cs operated by retailer r's competitors in local market l at time t. |
| Competing In-Store C&Cs l, r, t | Experian | Number of in-store C&Cs operated by retailer r's competitors in local market l at time t. |
| Own super/hypermarkets l, r, t | Experian | Number of super- and hypermarkets operated by retailer r in local market l at time t. |
| Competing super/hypermarkets l, r, t | Experian | Number of super- and hypermarkets operated by retailer r's competitors in local market l at time t. |
| Independent local retailers l, te | INSEE France | Number of independent local retailers (e.g., mom- and- pop stores, butchers, bakeries) operating in local market l at time t. |
| Distance to nearest brick-and-mortar store l, r | Experian; Retailer sites | Distance in km from retailer r's C&C in local market l to the nearest brick-and-mortar store (only included in the spending models for the stand-alone C&C type). |
| Age h,0 | Kantar France | Age of main shopper in household h in the year before the C&C opens in its local market. |
| Trend l, c, r, t | Experian | A linear trend from the time t retailer r's C&C type c opens in local market l. |
| Instrumental Variables | | |
| Cumulative adoption own C&Cs l, r, t | Kantar France | Number of adopters of C&Cs opened by retailer r in local markets that share borders with local market l, at time t. |
| Cumulative adoption competing C&C l, r, t | Kantar France | Number of adopters of C&Cs opened by retailer r's competitors in local markets that share borders with local market l, at time t. |
| Num_hh l, t | Experian | Number (in thousands) of households living in local market l at time t. |
| Availability of construction sites l, te | INSEE France | Number of construction sites in local market l at time t. |
| Price of construction sites l, te | INSEE France | Average price of construction sites per square meter in local market l at time t (in thousand euros). |
| Price of real estate l, te | INSEE France | Average price of real estate per square meter in local market l at time t (in thousand euros). |
| Availability of labor l, te | INSEE France | Number (in thousands) of job seekers in local market l at time t. |
| Labor costs l, te | INSEE France | Average labor costs in local market l in the year prior to the opening of the C&C (in thousand euros). |
- 40022242920960430 Notes: C&C = click and collect.
- 50022242920960430 aWe calculate the dependent variables by four-week period t.
- 60022242920960430 b We measure the moderator variables in the year before the C&C opens in the household's local market.
- 70022242920960430 cWe calculate the averages across all four-week periods in the year before the C&C opening.
- 80022242920960430 d Perishability and impulse nature are based on an unpublished 2018 Mturk survey conducted by Gielens and van Lin.
- 90022242920960430 e Since these data were not available at the local-market but at the zip-code level, we matched the local markets to the corresponding zip codes.
We model the extent to which C&C shopping changes a household's online spending at the retailer. For each retailer, we estimate the online spending model across the different C&C types operated by that retailer, using interaction terms to evaluate the differences between the C&C types while accounting for potential sources of endogeneity.
To estimate the online-spending model, we consider all households in local markets where the retailer opened a C&C within our observation window with at least one year of data available before and after the date of the opening. As such, we can track for each of the households (those who eventually do and do not adopt C&C shopping) their online spending at the focal retailer at least one year before the opening of the C&C until at least one year after the household adopted that C&C (or, in case of no C&C shopping, the end of the data window).
We model the impact of C&C shopping on household online spending at the retailer as
OnlineSpendh,l,c,r,t=α0,r+α1,rCCShopperh,l,r,t+α2,rTypel,c,r* CCShopperh,l,r,t+CCShopperh,l,r,t×(β1,rRuralh,0+β2,rWeekendh,0+β3,rBasketh,0+β4,rBulkyh,0+β5,rHhsizeh,0+β6,rPerishh,0+β7,rImpulseh,0)+CCShopperh,l,r,t*Typel,c,r×(γ1,rRuralh,0+γ2,rWeekendh,0+γ3,rBasketh,0+γ4,rBulkyh,0+γ5,rHhsizeh,0+γ6,rPerishh,0+γ7,rImpulseh,0)+controls+β1,r'Type_Openl,c,r,t+β2,r'CF_Openl,c,r,t+β3,r'CF_CCShopperh,l,r,t+μh+νt+ϵh,l,c,r,t.1
where h captures the household, l the local market, c the C&C type, r the retailer, and t the time period (month). reflects household h's online spending at time t at retailer r in local market l where the retailer opened C&C type c. is a dummy variable indicating which C&C type c retailer r operated in local market l; it equals 0 for the Near-Store type, and 1 for the other type operated by retailer r (i.e., in-store C&C for Intermarché; stand-alone C&C for Leclerc). Thus, the near-store type is the reference C&C type for both retailers.
is a step dummy variable indicating whether household h started shopping at (any type of) C&C from time t onward. Because the online-spending effect of C&C shopping may depend on the C&C type, we include an interaction term: . Because all continuous variables are mean-centered, the parameter associated with captures consumers' average change in online spending at the retailer after they started near-store shopping. The interaction coefficient indicates how this spending impact differs from the retailer's other C&C type.
The interactions between , on the one hand, and , on the other hand, reflect how the shopper characteristics moderate the effects of near-store C&C shopping on online spending. The three-way interactions with capture how these moderating effects differ for the other C&C type operated by the retailer.
Household online spending at a retailer may depend on the availability of competing retailers' C&Cs as well as on the presence of brick-and-mortar stores in the local market. To account for these influences, we control for the number of the focal retailer's and competing retailers' C&Cs, their super- and hypermarkets, and independent local retailers at time t in market l (for details, see Table 3). For Stand-Alone C&Cs, we control for the distance to the retailer's nearest brick-and-mortar store (which is 0 for in-store and near-store). We add a linear trend from the time the C&C pickup point opens in the local market. All continuous variables (except the trend) are mean-centered. is a step dummy variable that equals 0 before the C&C opened in the local market, and 1 afterward. We use household fixed effects ( ) to control for unobserved (time-invariant) household differences and time fixed effects ( ) to control for unobserved (household-invariant) temporal confounds ([53]).[12]
The retailer's decision to open a particular C&C in a specific market may be driven by variables that also influence households' online spending at the retailer and, if unaccounted for, may bias our estimates. Insofar as these variables are observable, we can add them as controls to Equation 1. However, some of the variables that simultaneously drive the retailer's decision to open a certain C&C type and the consumer's spending decision may be unobservable. To deal with this issue, we use a control-function approach ([63]). We first estimate, per retailer, a multinomial probit selection model across all periods and local markets, including those markets without a C&C. The three choice alternatives are whether the retailer opens one of the two C&C types or none at all, and the explanatory variables are the exogenous variables in Equation 1 plus several instruments. As instruments, we use the number of households living in the local market, and the setup and operational costs of opening a C&C in the market. The setup costs are captured by the availability of construction sites and the average price of both construction sites and real estate in the local market. The costs of running a C&C are reflected in the availability of labor in the local market and labor costs. All six variables are time-varying and meet the criteria for instrument variables (IVs; [62], p. 112): we expect them to influence the retailer's decision to open a specific C&C type in a local market but to be uncorrelated with households' online spending at the retailer. The coefficients of the IVs are statistically significant. Moreover, the likelihood-ratio tests for the restricted (without instruments) versus full model (with instruments) is highly significant for both retailers (Intermarché: = 365.7, p <.01; Leclerc: = 3,247.2, p <.01), which further confirms the incremental explanatory power of the instruments combined. Sargan tests for overidentifying restrictions confirm the validity of our IVs (ps >.10). We add the resulting generalized residuals as an independent variable to Equation 1 ([31]; [63]). Web Appendix W1 provides the auxiliary regressions and their estimation results.
Unobserved characteristics that drive a household's C&C shopping decision may also affect its spending. In other words, the adoption variable in Equation 1 may also be endogenous. We therefore estimate a probit model across all households in these markets and all periods in which the C&C type is available, with as the dependent variable and the same exogenous variables as in Equation 1 plus two instruments that influence the likelihood of C&C shopping but do not directly affect the household's amount spent at the retailer. The instruments are the number of C&C shoppers of both ( 1) the focal retailer and ( 2) competing retailers in neighboring markets of market l (i.e., markets that share boundaries with market l) at time t. The social contagion and local-neighborhood effects literature (e.g., [23]; [46]) has shown that imitation among consumers—including emulation in trial behavior for an internet retailer ([12])—is more likely when consumers are geographically proximate. As such, a household is more prone to engage in C&C shopping at a retailer when many other households already engage in C&C shopping in neighboring markets. However, as households' grocery outlays are idiosyncratic ([14]), the number of C&C shoppers in neighboring markets is not likely to affect a household's spending at the retailer. Again, the IVs are significant individually and combined as shown by the coefficient p-values (see Web Appendix W2) and the likelihood-ratio tests (Intermarché: = 1,868.1, p <.01; Leclerc: = 3,566.8, p <.01), which further confirms their strength. Sargan tests for overidentifying restrictions confirm the validity of our instruments (ps >.10). Having estimated these auxiliary models, we add the associated generalized residuals ( ) to Equation 1 ([31]; [63]). Web Appendix W2 provides details.[13]
Table 4 provides descriptives for both Intermarché and Leclerc markets. None of the moderator or control variables differ systematically between markets with different C&C types. Figure 1 portrays the average C&C shopper's ( 1) online, ( 2) brick-and-mortar, and ( 3) total spending at Intermarché (Panel A) and Leclerc (Panel B) in the year before and after they started C&C shopping. Regardless of the retailer and C&C type involved, C&C shopping leads to a clear upswing in online spending and a weaker increase in total spending at the retailer. Brick-and-mortar spending at the retailer remains rather stable. These plots should, however, be interpreted with caution: they do not control for factors that concurrently change with the retailer's C&C opening or the households' decision to start C&C shopping, nor for possible endogeneity in these decisions. Moreover, they cannot reveal differences in spending effects between shoppers with different convenience needs. Our model accounts for these issues.
Graph
Table 4. Descriptives.
| Intermarché | Leclerc |
|---|
| In-Store | Near-Store | Stand-Alone | Near-Store |
|---|
| Mean | SD | Mean | SD | Mean | SD | Mean | SD |
|---|
| Moderator Variablesa | | | | | | | | |
| Access convenience | | | | | | | | |
| Rural (%) | .11 | | .11 | | .06 | | .12 | |
| Weekend shopping (% of trips) | .24 | .16 | .24 | .16 | .24 | .16 | .23 | .23 |
| Collection convenience | | | | | | | | |
| Basket size (# of items) | 15.13 | 7.93 | 15.13 | 7.93 | 15.47 | 8.56 | 15.28 | 15.28 |
| Bulky items in basket (#) | 2.56 | 1.49 | 2.56 | 1.49 | 2.60 | 1.54 | 2.60 | 2.60 |
| Adjustment convenience | | | | | | | | |
| Household size (#) | 2.40 | 1.13 | 2.40 | 1.13 | 2.47 | 1.15 | 2.46 | 2.46 |
| Perishables in basket (%) | .76 | .06 | .76 | .06 | .76 | .06 | .76 | .76 |
| Impulse items in basket (%) | .66 | .08 | .66 | .08 | .68 | .07 | .67 | .67 |
| Competition-Related Variablesb | |
| Own C&Cs (#) | .93 | 2.00 | .54 | 1.25 | .99 | 1.55 | .79 | 1.20 |
| Competing stand-alone C&Cs (#) | .04 | .19 | .08 | .27 | .03 | .16 | .02 | .14 |
| Competing near-store C&Cs (#) | .05 | .22 | .13 | .34 | .10 | .30 | .11 | .31 |
| Competing in-store C&Cs (#) | .01 | .12 | .03 | .17 | .04 | .20 | .08 | .27 |
| Own super- and hypermarkets (#) | 1.86 | 1.73 | 1.64 | 1.46 | 2.44 | 1.83 | 3.07 | 2.73 |
| Competing super- and hypermarkets (#) | 14.00 | 14.94 | 19.16 | 15.65 | 9.94 | 10.43 | 12.76 | 13.43 |
| Independent local retailers (#) | 49.61 | 53.88 | 63.39 | 57.10 | 50.49 | 74.65 | 46.53 | 48.24 |
| Focal Variables | |
| C&C shopperc (%) | 3.60 | 3.80 | 10.60 | 11.80 |
| Online spending before adoption (€/month) | 101.55 | 113.87 | 105.48 | 85.59 |
| Online spending after adoption (€/month) | 110.71 | 151.44 | 127.55 | 115.12 |
- 100022242920960430 Notes: C&C = click and collect.
- 100022242920960430 aMeasured the year before C&C opening.
- 100022242920960430 bMeasured over the entire time window. cPenetration of C&C in year after opening.
Graph: Figure 1. Average four-weekly online, brick-and-mortar, and total spending at the retailer by click and collect shoppers before and after adoption.Notes: B&M = brick and mortar.
Table 5 shows the effects of C&C shopping on online retailer spending. The coefficient of the "CC shopper" variable captures the impact of near-store shopping (the reference case in both models) on online spending at the retailer, for the average household in our sample. This effect is positive and significant for both retailers (Intermarché: = 36.04, p <.01; Leclerc: = 42.73, p <.01). Thus, near-store shopping increases online spending at the retailer. The interaction effect between "CC shopper" and "Alternative C&C type" reflects the difference in the online-spending effect between the reference C&C type and the alternative C&C type. This effect is positive for in-store C&C at Intermarché ( = 11.23, p <.01), indicating that in-store shopping leads to a higher lift in online spending at the retailer than near-store shopping. In contrast, the interaction effect is not significant for stand-alone C&C at Leclerc ( = −.35, p >.10), implying that stand-alone and near-store shopping have the same online spending lift. Interestingly, as to the control variables, we find small, positive spillover effects for both the number of in-store and stand-alone C&Cs of competitors. When the number of C&Cs in a local market increases, consumers spend more at the focal retailer's C&C. This may point to an "awareness effect": as more C&C pickup points open in the vicinity, consumers become more aware of the format and are more prone to use it.
Graph
Table 5. The Effect of C&C Shopping on Consumer Spending at the Retailer.
| Online Spending | Brick-and-Mortar Spending | Total Spending |
|---|
| Dependent Variable | Intermarché | Leclerc | Intermarché | Leclerc | Intermarché | Leclerc |
|---|
| C&C shopper (α1) | 36.04*** | 42.73*** | 39.46*** | 2.64*** | 75.50*** | 45.37*** |
| Access convenience | | | | | | |
| × Rural (β1) | −27.49*** | −1.11** | −8.28*** | 15.34*** | −35.77*** | 14.22*** |
| × Weekend shopping (β2) | 49.13*** | −38.23*** | −31.08*** | 26.80*** | 18.04*** | −11.43*** |
| Collection convenience | | | | | | |
| × Basket size (β3) | 2.62*** | 1.76*** | 1.07*** | −1.51*** | 3.69*** | .25* |
| × Bulkiness (β4) | .39** | 4.28*** | −13.16*** | −7.09*** | −12.77*** | −2.81** |
| Adjustment convenience | | | | | | |
| × Household size (β5) | 7.28*** | 4.62*** | 10.11*** | −3.33*** | 17.39*** | 1.30* |
| × Perishability (β6) | 72.66*** | −2.89 | 118.29*** | −8.27 | 190.96*** | −11.16 |
| × Impulse nature (β7) | −93.83*** | −77.90*** | −167.80*** | 76.11*** | −261.63*** | −1.78 |
| C&C shopper× Alternative C&C type (α2) | 11.23*** | −.35 | −39.08*** | −1.96* | −27.85*** | −2.31** |
| Access convenience | | | | | | |
| × Rural living (γ1) | −.96 | 19.62*** | 66.08*** | −8.25** | 65.11*** | 11.38*** |
| × Weekend shopping (γ2) | −83.75*** | 5.31** | 67.88*** | 1.87 | −15.87*** | 7.18 |
| Collection convenience | | | | | | |
| × Basket size (γ3) | −.42*** | .64*** | −6.04*** | −.08 | −6.46*** | .56*** |
| × Bulkiness (γ4) | 1.04*** | 2.84*** | .22 | 9.12*** | 1.25*** | 11.95*** |
| Adjustment convenience | | | | | | |
| × Household size (γ5) | 18.66*** | −1.00*** | −30.22*** | 3.83*** | −11.56*** | 2.83*** |
| × Perishability (γ6) | 109.49*** | −60.34*** | −286.71*** | 67.99*** | −177.22*** | 7.65 |
| × Impulse nature type (γ7) | 306.12*** | 138.85*** | −306.15*** | −128.59*** | −.03 | 10.27 |
| Control Variables | | | | | | |
| Own C&Cs | −.01 | .21*** | .27*** | −.47*** | .28*** | .26* |
| Competing stand-alone C&Cs | −.00 | .46** | 1.90*** | −1.57** | 1.90*** | −1.11 |
| Competing near-store C&Cs | −.03 | −.15 | 3.89*** | −2.31*** | 3.87*** | −2.46*** |
| Competing in-store C&Cs | .28*** | .09 | 1.72*** | 4.17*** | 2.00*** | 4.26*** |
| Trend | −.00*** | −.02*** | −.19*** | −.33*** | −.19*** | −.35*** |
| C&C opening | .17*** | .53*** | .71*** | −.36 | .88*** | .17 |
| CF C&C opening | .26*** | .12 | 2.56*** | −11.27*** | 2.82*** | −11.15*** |
| CF C&C shopper | −6.55*** | 12.31*** | −74.27*** | −26.23*** | −80.82*** | −13.92** |
- 140022242920960430 *p <.10.
- 150022242920960420 **p <.05.
- 160022242920960420 ***p <.01.
- 170022242920960420 Notes: Two-sided tests of significance. CF = control-function regressor. All models include household and time fixed effects. The reference C&C type is near-store. The "alternative C&C type" is in-store for Intermarché, and stand-alone for Leclerc.
Rural shoppers show a below-average spending lift after they start shopping at near-store C&Cs (Intermarché: = −27.49, p <.01; Leclerc: = −1.11, p <.05). As we expected, for these rural shoppers, stand-alone shopping increases online spending more than near-store shopping ( = 19.62, p <.01); however, in-store shopping does not entail lower spending lifts than near-store shopping ( = −.96, p >.10). Weekend shopping does not have a unison impact on the online spending lift of near-store shopping (Intermarché: = 49.13, p <.01; Leclerc: = −38.23, p <.01). Still, its influence on the relative appeal of different C&C types is as expected: for weekend shoppers, in-store shopping leads to lower online spending lifts ( = −83.75, p <.01) and stand-alone shopping to higher online spending lifts ( = 5.31, p <.05), than near-store shopping.
The online spending lift from near-store shopping is higher for shoppers with larger baskets (Intermarché: = 2.62, p <.01; Leclerc: = 1.76, p <.01) and those buying more bulky items (Intermarché: =.39, p <.05; Leclerc: = 4.28, p <.01) than for the average shopper. Larger basket shoppers spend less online with in-store than near-store C&C ( = −.42, p <.01), confirming our expectations. However, households with more bulky items in their basket spend more online when in-store than near-store shopping (Intermarché: = 1.04, p <.01), even if the latter is relatively more effortful. Maybe these households strategically split their baskets and order their bulky items—typically more routinely needed stock-up products—online, to have more time to seek out the remaining products upon order pickup in-store. With stand-alone shopping, large-basket and bulky-item shoppers show larger online spending lifts than with near-store shopping ( =.64, p <.01; = 2.84, p <.01), although both C&C types rate equally high on collection convenience. Possibly, the very presence of stand-alone C&Cs at new, high-drive-through locations increases familiarity with the retailer name (a "billboard" effect; see [ 7]) and generates extra appeal. In addition, because these shoppers mostly engage in single-purpose trips, they may prefer the bare-bones stand-alone location over the more commercial environment of near-store C&Cs, which they may consider a nuisance rather than an asset ([54]).
Near-store shopping leads to a larger increase in online spending for larger households (Intermarché: = 7.28, p <.01; Leclerc: = 4.62, p <.01), keeping basket size constant. As we have postulated, these households spend even more online when in-store than near-store shopping ( = 18.66, p <.01) but less when stand-alone shopping ( = −1.00, p <.01). Near-store shopping also leads to a higher online spending lift among shoppers with a larger share of perishables (Intermarché: = 72.66, p <.01, Leclerc: = −2.89, p >.10). In line with expectations, this lift is higher for in-store than near-store shopping ( = 109.49, p <.01) but lower for stand-alone shopping ( = −60.34, p <.01). Finally, near-store shoppers with a larger share of impulse goods spend less online (Intermarché: = −93.83, p <.01; Leclerc: = −77.90, p <.01). As we anticipated, these households spend more when in-store (vs. near-store) shopping ( = 306.12, p <.01) but also, surprisingly, when stand-alone (vs. near-store) shopping ( = 138.85, p <.01). In hindsight, some impulse shoppers may value the shift to stand-alone shopping as a means to control their unplanned spending.
We performed multiple robustness checks. First, as an alternative way to correct for market selection, we reestimated the market selection model on pooled data across the two retailers using four choice options (no C&C opening or an opening of one of the three C&C types). We then used the generalized residuals from this analysis as control-function regressors in Equation 1. Second, to alleviate the concern that serial correlation in the unobserved household components would overstate the significance of our effects, we ran a simple before–after spending model to measure the impact of C&C shopping on online spending. Following [13] and similar to [ 4], we removed the time dimension by collapsing the time-series data for each household into a pre- and posttreatment period. Third, we used a difference-in-difference approach with propensity score matching ([52]) to compare the change in online spending of households who adopt C&C shopping with that of control households (i.e., nonadopters with a similar adoption propensity). Fourth, instead of assessing the impact of a household's C&C adoption, we reran our models with the local opening of a C&C pickup point as the treatment variable. Finally, consumers' online spending may well evolve over time as they gain more experience, or as the execution of the format improves. Adding an extra interaction between the adoption variable and time-since-adoption points to a small negative effect. Thus, after the initial upswing, consumers slightly shift some of their purchases back to the store.
Web Appendix W3 provides details on the various robustness checks. With one exception, results remained substantively the same, thereby underscoring the robustness of our findings. Only for bulkiness did the interaction with in-store C&C shopping become insignificant (instead of positive and significant) in a number of instances, indicating that this effect (which was unexpected to begin with) should be interpreted with caution.
To glean more insight into the economic significance of the findings, we use the estimates from Table 5 to calculate the effect sizes. Column 1 in Table 6, Panel A, reports the average change in online spending after households start C&C shopping. On average, households spend about €40 extra online per month at the retailer. While the spending lift is significant for each C&C type, it is highest for in-store C&C. At first sight, this seems to validate the (many) retailers that are investing in in-store C&C. However, closer inspection reveals a more nuanced picture.
Graph
Table 6. Online and Total Spending Implications for Different Click and Collect Types and Convenience Needs.
| A: Online Spending | B: Total Spending |
|---|
| Average Spending Lift | Difference in Spending Lift Between Shoppers with High Versus Low Needs | Average Spending Lift | Difference in Spending Lift Between Shoppers with High Versus Low Needs |
|---|
| Access- Convenience Needs | Collection- Convenience Needs | Adjustment- Convenience Needs | Access- Convenience Needs | Collection- Convenience Needs | Adjustment- Convenience Needs |
|---|
| 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 |
|---|
| Intermarché |
| In-store (IS) | +44.58 | −41.79 | +48.92 | +136.49 | +50.43 | +30.18 | −95.56 | −31.72 |
| Near-store (NS) | +33.44 | −8.57 | +53.55 | +12.47 | +72.12 | −28.81 | +28.44 | +25.22 |
| Sign. ▵ IS − NS | ✔ | ✔ | ✔ | ✔ | ✔ | ✔ | ✔ | ✔ |
| Leclerc |
| Stand-alone (SA) | +43.98 | +5.83 | +72.59 | +12.07 | +45.26 | +23.96 | +48.17 | +12.2 |
| NS | +42.63 | −15.84 | +49.96 | −2.68 | +46.59 | +9.82 | −4.96 | +1.54 |
| Sign. ▵ SA − NS | ✔ | ✔ | ✔ | ✔ | n.s. | ✔ | ✔ | ✔ |
180022242920960420 Notes: Figures in regular font are significant at p <.05, figures in italics are not. The significance of the difference between C&C types is indicated in a separate row, where ✔ = the difference is significant at p <.05; n.s. = not significant. In Panel A, Column 1 reports the online spending lift for a shopper with average convenience needs. For example, for the average shopper, near-store (in-store) shopping increases online spending at Intermarché by €33.44 (€44.58) per month. The difference in online spending lift between these two C&C types is significant, as indicated by the ✔ in the row below. Columns 2–4 report the effect sizes of the convenience features (i.e., the difference in online spending lift between shoppers with high vs. low needs) and whether those effect sizes differ significantly between C&C types. For instance, for near-store C&C at Intermarché, +€53.55 is obtained as the difference between ( 1) the online spending lift for shoppers with high collection-convenience needs (€63.17 in Figure 2) and ( 2) the corresponding change for shoppers with low collection-convenience needs (€9.62 in Figure 2). This difference reflects how the online spending implications of near-store shopping depend on shoppers' need for collection convenience.
Graph: Figure 2. Impact of shoppers' convenience needs on online spending at the retailer.Notes: The figure reads as follows. For low collection convenience, the near-store bar at Intermarché equals €9.62. This means that households with low collection-convenience needs spend €9.62 per month more online at Intermarché after they start near-store shopping. For high collection convenience, the near-store bar at Intermarché equals €63.17. Thus, households with high collection-convenience needs spend €63.17 per month more online at Intermarché after they start near-store shopping. The difference between these two bars (€63.17 − €9.62 = €53.55) reflects how the online spending implications of near-store shopping depend on shoppers' need for collection convenience, and is reported in Table 6.
The spotlight analysis in Figure 2 displays the changes in online spending following C&C shopping, by C&C type, for shoppers with low versus high needs for each convenience feature.[14] Columns 2 to 4 in Table 6, Panel A, report the corresponding effect sizes of the convenience features (i.e., the difference in online spending lift between shoppers with high vs. low needs) and whether those effect sizes differ between C&C types. Shoppers with a high access-convenience need (weekend shoppers, shoppers in rural areas) spend less online than shoppers low on that need when near-store shopping (Intermarché: €18.81 vs. €27.38, a difference of −€8.57; Leclerc: €33.47 vs. €49.31 = −€15.84) and especially in-store shopping (−€41.79), but more so when stand-alone shopping (+€5.83). This difference between C&C types stems mostly from rural shoppers. Shoppers with higher collection-convenience needs (those with large and bulky baskets) always spend more online, regardless of the C&C type, but significantly more so when near-store than in-store shopping (Intermarché: near-store = +€53.55, in-store = +€48.92) and when stand-alone than near-store shopping (Leclerc: stand-alone = +€72.59, near-store = +€49.96). While we find that the online spending lift across C&C types is mostly driven by basket size, bulkiness drives the extra lift for stand-alone C&C. Shoppers who score higher on adjustment-convenience needs (large households shopping more perishables and impulse goods) spend significantly more online when in-store shopping (+€136.49), while the difference between low- and high-adjustment-need shoppers is less pronounced for stand-alone and near-store C&Cs. The complexity of household needs plays a role here: larger households in particular spend more online when in-store shopping, which enables them to purchase forgotten items upon pickup. Basket composition, too, matters: shoppers with more perishable and impulse items in their baskets spend much more online when in-store shopping and less so when near-store or stand-alone shopping.
In all, we find that C&C shopping boosts online spending at the retailer. On average, in-store shopping produces the highest online spending lift, especially among shoppers with high adjustment-convenience needs. For shoppers with high access- or collection-convenience needs, stand-alone shopping yields the highest online-spending gain.
Click-and-collect shopping enhances households' online spending at the retailer. However, if C&C shopping cannibalizes brick-and-mortar spending, the retailer's gain may be negligible. To explore this, we reestimate Equation 1 with shoppers' brick-and-mortar and total spending at the retailer as the dependent variables. We summarize the estimation results in Table 5. Table 6, Panel B, and Figure 3 document the ensuing changes in total spending.
Graph: Figure 3. Impact of shoppers' convenience needs on total spending at the retailer.Notes: The figure reads as follows. For low collection convenience, the near-store bar at Intermarché equals €60.74. This means that households with low collection-convenience needs spend €60.74 per month more in total at Intermarché after they start near-store shopping. For high collection convenience, the near-store bar at Intermarché equals €89.18. Thus, households with high collection-convenience needs spend €89.18 per month more in total at Intermarché after they start near-store shopping. The difference between these two bars (€89.18 − €60.74 = €28.44) reflects how the total spending implications of near-store shopping depend on shoppers' need for collection convenience, and is reported in Table 6.
For households with average convenience needs, C&C shopping enhances total spending at the retailer. For stand-alone and in-store shopping, the online and total spending lift do not significantly differ, while for near-store shopping, the increase in total spending significantly exceeds the online lift (Intermarché: €72.12 compared with €33.44, p <.01; Leclerc: €46.59 compared with €42.63, p <.05,). Thus, cannibalization does not appear to be an issue, and positive spillovers from C&C to brick and mortar can occur.
However, this average picture conceals substantial shopper heterogeneity. The impact of shoppers' convenience needs on their total spending at the retailer (Table 6, Panel A, columns 6–8) is very different from that on their online spending (Table 6, Panel B, columns 2–4). In a similar vein, Figure 3 shows that across shoppers with different convenience needs, the pattern of changes in total spending does not mimic that of changes in online spending displayed in Figure 2. For households that value access convenience, the total spending lift (in Figure 3) always exceeds the online lift (in Figure 2), suggesting positive spillovers to the retailer's brick-and-mortar stores. The time savings of C&C shopping may attract access-oriented shoppers to the retailer and, having become familiar with the retailer or being on-site for C&C order collection, they may use some of the saved time to seek out additional items at the retailer's brick-and-mortar stores. A higher collection-convenience need typically leads to lower total than online spending lifts, indicating that part of the household's online spending lift is at the expense of brick-and-mortar spending. Such cannibalization appears especially prevalent with in-store C&C. For this C&C type, the online spending lift (€71.63, see Figure 2) is entirely dissipated by the fact that shoppers with high collection-convenience needs spend less at the retailer's physical outlets (leading to a total spending lift of −€1.56; see Figure 3). Likewise, for households with high adjustment-convenience needs, the total spending lift from in-store shopping (€40.50, see Figure 3) is far lower than the online spending lift (€111.66, see Figure 2), suggesting strong cannibalization. While adjustment-oriented shoppers tend to be impulse-driven, the ability of controlling their impulse tendencies online results in a relatively larger brick-and-mortar loss.
In summary, C&C shopping not only affects a household's online but also affects its brick-and-mortar spending at the retailer in ways that critically depend on the C&C type and that household's convenience needs. Thus, which C&C type to operate in a local market should be guided by the total (online plus brick-and-mortar) spending implications and the local shopper profile.
To get a feel for these local-market implications, Table 7 extrapolates the four-week household-level total spending effects to yearly market-level retailer revenues for each retailer and C&C type, and for different shopper profiles. Market-level revenues are obtained by multiplying the C&C shoppers' total spending with the observed penetration rates for the considered C&C type and retailer. These figures are extrapolated to the local-market level by multiplying with the number of households in an average-sized market. The table shows that when shoppers in a local market have average convenience needs, stand-alone C&C is the highest-performing option: it yields slightly higher market revenues than near-store C&C (+€143,000), which, in turn, strongly outperforms in-store C&C (+€328,000). Interestingly, whereas in-store C&C outperforms the other two types in terms of online spending, it leads to the smallest market-revenue lift for the retailer.
Graph
Table 7. Retailer Market-Revenue Implications for Different Shopper Profiles.
| A: In-Store Versus Near-Store Click and Collect (Intermarché) |
|---|
| Shopper Profile | | Average Convenience Needs | High Access- Convenience Needsa | High Collection- Convenience Needsa | High Adjustment- Convenience Needsa |
|---|
| In-store (IS) | Market penetration | .036 | .036 | .036 | .036 |
| ▵ Total spending per C&C shopper (€/month) | 50.428*** | 77.455*** | −1.557 | 40.495*** |
| ▵ Revenue at local-market level (millions €/year) | .644*** | .989*** | −.020 | .517*** |
| Near-store (NS) | Market penetration | .038 | .038 | .038 | .038 |
| ▵ Total spending per C&C shopper (€/month) | 72.121*** | 43.504*** | 89.183*** | 91.774*** |
| ▵ Revenue at local-market level (millions €/year) | .972*** | .586*** | 1.202*** | 1.236*** |
| ▵ (IS − NS) | ▵ Total spending per C&C shopper (€/month) | −21.693 | 33.950 | −90.741 | −51.279 |
| | ▵ Revenue at local-market level (mill €/year) | −.328*** | .402*** | −1.221*** | −.720*** |
| B: Stand-Alone Versus Near-Store Click and Collect (Leclerc) |
| Shopper Profile | | AverageConvenienceNeeds | HighAccess-ConvenienceNeedsa | HighCollection-ConvenienceNeedsa | HighAdjustment-ConvenienceNeedsa |
| Stand-alone (SA) | Market penetration | .118 | .118 | .118 | .118 |
| ▵ Total spending per C&C shopper (€/month) | 45.264*** | 67.749*** | 70.245*** | 51.772*** |
| ▵ Revenue at local-market level (mill €/year) | 1.894*** | 2.834*** | 2.939*** | 2.166*** |
| NS | Market penetration | .106 | .106 | .106 | .106 |
| ▵ Total spending per C&C shopper (€/month) | 46.593*** | 57.152*** | 44.226*** | 47.582*** |
| ▵ Revenue at local-market level (mill €/year) | 1.751*** | 2.148*** | 1.662*** | 1.788*** |
| ▵ (SA − NS) | ▵ Total spending per C&C shopper (€/month) | −1.328 | 10.597*** | 26.019*** | 4.190*** |
| ▵ Revenue at local-market level (mill €/year) | .143 | .686*** | 1.277*** | .378*** |
- 190022242920960420 *p <.10.
- 200022242920960420 **p <.05.
- 210022242920960420 ***p <.01.
- 220022242920960420 a And average needs for the other two convenience features.
- 230022242920960420 Notes: Two-sided tests of significance. Revenue expressed in millions of euros per year for an average-sized local market.
Stand-alone C&C yields especially higher market revenues than near-store when the shoppers in a local market strongly value collection convenience (a €1,277,000 revenue advantage) or access convenience (a €686,000 revenue advantage). In turn, near-store C&C outperforms in-store C&C for all shopper profiles, except those oriented toward access convenience. When, on average, the shoppers in a local market highly value adjustment convenience, the market-revenue advantage of near-store over in-store C&C amounts to €720,000. For such a shopper profile, in-store C&C generates the largest boost in online spending, but mostly at the expense of the retailer's brick-and-mortar stores. Thus, when it comes to the retailer's market revenue, local markets with this shopper profile are better served with near-store and especially stand-alone C&C.
With order fulfillment as a major impediment for growth in online grocery, C&C has been advanced as the new fulfillment mantra. Accordingly, many retailers have rushed (or are rushing) into this format to secure their piece of the online grocery pie. However, C&C remains a relatively young phenomenon, and its demand-side effects are not well documented. Moreover, because "the last mile is fast becoming the ultimate battleground" ([51]), retailers are looking for advice on whether and how to implement these programs. Drawing on two data sets, each covering the introduction of different C&C types by a major grocery retailer, we tracked the spending implications of three C&C types. Our study centered on the following questions: ( 1) How does C&C order fulfillment affect shoppers' online spending and total spending at a retailer? ( 2) How do different types of C&C in the form of in-store, near-store, and stand-alone C&C affect these outcomes? and ( 3) What is the impact of these different C&C types for different types of households?
As to the online spending impact of the C&C format, we find that C&C can be an effective means to boost online spending at the retailer. Thus, C&C may indeed be the long-awaited road to online success for grocery retailers ([26]), overcoming the last-mile problems associated with home delivery. As to the total spending impact of the C&C format, consultants have long thought that "the increase in...click-and-collect will come primarily at the expense of brick-and-mortar sales" ([50]; see also, e.g., [ 6]; [38]). In contrast, we find cannibalization to be minimal. Overall, this bodes well for C&C. By blending the convenience features of home delivery and brick-and-mortar channels, C&C enhances households' total spending at the retailer. The C&C format thus constitutes a valuable addition to a retailer's channel mix.
When it comes to fulfillment, not all C&Cs are alike. While we find that the C&C format has a significant impact on consumer spending at the retailer, the effects are strongly shaped by ( 1) the C&C type and ( 2) shopper characteristics. In-store shopping typically produces the highest online spending lift. As expected, this holds especially among shoppers with high adjustment-convenience needs (i.e., larger households that buy more perishables and impulse items). Interestingly, this lift also stems from an increase in shopping frequency, as these shoppers complement their C&C order collection with store purchases. Stand-alone shopping yields the highest online spending gain among shoppers with high collection-convenience needs (i.e., those buying bulky and large baskets), and especially access convenience (i.e., rural and weekend shoppers). This C&C type does, indeed, come with lower shopping effort than in-store C&C and allows for more time-efficient order pickup than both in-store and near-store order fulfillment.
Looking beyond the online-spending effects, we find that in-store shopping has the most pronounced impact on the retailer's brick-and-mortar sales. However, depending on shoppers' convenience needs, the brick-and-mortar effect can go both ways—either attenuating or enhancing the total spending lift. For shoppers with high access-convenience needs, in-store C&C generates positive spillovers for the retailer's brick-and-mortar stores. Especially for rural shoppers, in-store C&C shopping goes along with higher spending in the retailer's brick-and-mortar stores and, thus, an increase in total spending. In contrast, for shoppers with high collection-convenience needs, in-store shopping strongly cannibalizes brick-and-mortar sales. When using this C&C type, large-basket shoppers for bulky items seem to merely shift their purchases from brick-and-mortar to online, such that total spending at the retailer hardly goes up. Also for shoppers with high adjustment-convenience needs, the total spending effects of in-store shopping are not promising. Especially for large households buying more impulse items, in-store C&C shopping goes along with a dramatic reduction in brick-and-mortar spending. Thus, the C&C type that yields the biggest boost in online spending at the retailer (i.e., in-store C&C) is not necessarily the best-performing option for the retailer overall.
These insights can be useful for retailers setting up or expanding their C&C operations. Table 8 lays out which C&C type performs best depending on the shoppers' characteristics and the retailer's key performance metric (i.e., shoppers' online spending, brick-and-mortar spending, or total spending). Because C&C types differ in terms of setup and order-preparation costs, a possible concern is that even if stand-alone (or near-store) C&Cs yield higher market revenues, these may be offset by higher costs. To shed some light on this issue, we perform back-of-the-envelope calculations to trade off each C&C type's market revenues (calculated using our estimates) against their costs (taken from industry reports; [18]; [19]; [36], [37]). We then compare the profitability of the C&C types in the short run and the long run (i.e., after the set-up investments have been depreciated) and add these insights to Table 8. Web Appendix W5 shows the underlying profitability calculations.
Graph
Table 8. Best Performing C&C Types.
| Shopper Profile | Online Spending at the Retailer | B&M Spending at the Retailer | Total Spending at the Retailer | Retailer Short-Run Profita | Retailer Long-Run Profita |
|---|
| Average convenience needs | In-store | Near-store or stand-alone | Near-store or stand-alone | Near-store | Stand-alone |
| High access-convenience needs | Stand-alone | In-store | In-store or stand-alone | In-store or stand-alone | In-store or stand-alone |
| High collection-convenience needs | In-store or stand-alone | Stand-alone | Stand-alone | Stand-alone | Stand-alone |
| High adjustment-convenience needs | In-Store or stand-alone | Near-store or stand-alone | Stand-alone | Near-store | Stand-alone |
240022242920960420 a Web Appendix W5 shows the profit calculations.
Overall, while in-store C&Cs excel at increasing (especially adjustment-convenience-oriented) shoppers' online spending at the retailer, they lead to lower total spending. Stand-alone C&Cs yield the highest total-spending increase for households with high collection- or adjustment-convenience needs. As to profit, stand-alone C&C appears to be the more profitable option. Many retailers have jumped the bandwagon by quickly setting up the in-store type. Our results caution managers not to take this easy route routinely.
In laying out our expectations, we argued that C&C shopping may increase primary demand through larger baskets and/or extra shopping trips. To deepen our insights, we reestimated Equation 1 with households' ( 1) total grocery spending (across all retailers), ( 2) total number of trips (across all retailers), and ( 3) spending per trip as the dependent variables (for the estimates, see Web Appendices W4-A, W4-B, and W4-C). For both retailers, we find that, on average, C&C shopping positively affects households' total grocery spending, and this manifests in an increase in both total trip frequency and spending per trip. These findings contradict the common fear that online shopping may be detrimental for primary demand because of lower impulse and unplanned purchases. Instead, our additional probe suggest that the time convenience leads households to shop more often while the reduction in shopping effort drives up order sizes, thus increasing consumption and spending ([ 2]; [21]; [42]; [61]). Thus, C&C shopping can also increase the size of the grocery pie.
Our insights may be particularly relevant in the light of the COVID-19 pandemic, which has accelerated the shift to online that was already taking place. Fulfillment options with limited contact, such as C&C, have become increasingly popular as shoppers feel wary of walking store aisles. According to the Adobe Digital Economy Index, "buy online, pick up in store" orders in the United States surged in 2020 by 62% between February 24 and March 21, compared with the same period a year earlier (https://www.adobe.com/experience-cloud/digital-insights/digital-economy-index.html). With the adoption rate of C&C spiking, many retailers have accelerated the rollout of C&C pickup points or have added C&C to their channel mix. Because the pandemic created urgent demand for C&C, retailers predominantly turned to fulfillment options that they could quickly launch or expand (i.e., in-store C&C and curbside, near-store options). With the sharp increase in demand, online orders fulfilled with store inventory increasingly risk out-of-stocks on store shelves ([32]). Thus, as retailers try to permanently convert the increased C&C demand, they should carefully (re)consider which C&C type performs best depending on their performance outcome of interest and their local markets' shopper characteristics.
Extant research has focused on internet channels with home delivery (e.g., [17]; [22]; [24]), with several studies investigating the effects of online channel usage on consumer shopping behavior (e.g., [16]; [41]). While home delivery does not require shoppers to travel to a store and queue up, it also has disadvantages: consumers have to commit to delivery times, and returning products can be challenging. The C&C format does not suffer from these drawbacks. We contribute to the online-channel literature by laying out how different C&C fulfillment options rate differently on different convenience features and how consumers value these features. In so doing, we expand the shopping convenience typology of [54] by distinguishing two aspects of transaction convenience that are particularly relevant in a C&C setting: the physical effort to collect the order at the pickup point (collection convenience) and the ease with which shoppers can adjust their online orders, by adding, returning, or replacing items upon pickup (adjustment convenience). In doing so, we recognize that efficient order fulfillment has become an ever more important shopper convenience need that is no longer obscured in the supply chain but has moved to the forefront to capture, delight, and retain shoppers in an omnichannel retail world.
Our findings are consistent with classic channel texts that advise no single channel is best ([25]; [56]). Rather, consumers differ in how they want to buy. We contribute by adding that it is better to excel in one convenience feature (and score low on others) than to rate average on all features. Indeed, near-store C&C seldom comes out as the best option. Thus, while no single C&C option is best, the "middle" option is definitely worst—conforming to Porter's (1980) "stuck-in-the-middle" principle.
Furthermore, prior research has studied whether online and offline channels are complementary or synergetic (e.g., [ 7]; [28]; [48]; [59]). In their synthesis of the literature, [47] conclude that substantial channel differentiation may alleviate cannibalization occurring between similar channels (e.g., mobile and online channels). However, such differentiation also leads consumers to perceive lower consistency in the retailer's channel mix and a less seamless experience during multichannel shopping, which may reduce retailer performance. Thus, [47] conclude that finding the right balance between channel cannibalization and synergy is a major unresolved issue. We demonstrate that whether the online and the offline channels are cannibalistic or synergetic depends on matching C&C order fulfillment types with shoppers' convenience needs. By tailoring the C&C type to the local-market shopper profile (e.g., by not setting up in-store C&Cs in markets where consumers have, on average, high collection- or adjustment-convenience needs), retailers can avoid cannibalization while maintaining a consistent channel offer and substantially enhance their market revenues and profit.
Our study is subject to several limitations, some of which open up new questions that warrant further research. First, our model includes household fixed effects, which ensures that cross-sectional household-specific components are removed. It is still possible, though, that there is some serial correlation left in the errors within each household, which may overstate the significance of the effects. Still, the simple before–after model in which the time dimension is aggregated out reassures us that our effect identification is not biased by (higher-order) temporal error correlation within households ([13]).
Second, we analyzed the impact of three C&C order fulfillment types in two data sets, each involving one retailer. In doing so, we join empirical work in multichannel research that analyzes a single retailer (e.g., [22]; [37]). Our focus was on three C&C types with distinct fulfillment modalities. Future studies, across more retailers, could formally investigate to what extent retailer characteristics, such as their price and quality positioning and their assortment breadth and depth (factors that were constant in our comparisons), affect demand for the C&C types.
Third, our focus was on comparing alternative C&C types. Given that the option to order online with home delivery was scarcely used by the panelists in our data set in the country and period considered, we did not consider that option in our analysis. Future studies could compare the C&C types with a broader set of options to buy not purely inside the store, including home delivery. In addition, the uptick in retailers experimenting with various autonomous delivery options to crack the last-mile problem due to COVID-19, including self-driving robots and drones ([27]), provides numerous research opportunities.
Fourth, based on the convenience features of the different C&C types, we conjectured which shoppers would be most inclined to value these features and empirically assessed their behavioral response without intermediate "process" measures. Future studies could directly verify which consumer segments attach more importance to access, collection, and adjustment convenience and how this relates to their preference for alternative C&C types.
Finally, our empirical results reflect the current market situation. The size of the spending lift may well evolve as more retailers roll out their C&C operations, and new types of C&C order fulfillment gain way. We leave the analysis of these developments, including the study of competitive reactions, as a fruitful topic for future research.
Supplemental Material, Web_Appendices_--JM-18-0623-R5--_Navigating_the_Last_Mile_in_Grocery_Shopping_PDF - Navigating the Last Mile: The Demand Effects of Click-and-Collect Order Fulfillment
Supplemental Material, Web_Appendices_--JM-18-0623-R5--_Navigating_the_Last_Mile_in_Grocery_Shopping_PDF for Navigating the Last Mile: The Demand Effects of Click-and-Collect Order Fulfillment by Katrijn Gielens, Els Gijsbrechts and Inge Geyskens in Journal of Marketing
Footnotes 1 Kusum Ailawadi
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Inge Geyskens https://orcid.org/0000-0002-7657-863X
5 Online supplement: https://doi.org/10.1177/0022242920960430
6 Click and collect includes pure online players, but most of the operations are run by brick-and-mortar retailers. The latter are also the focus of this study.
7 Although their incentive to switch to C&C shopping is less pronounced, households valuing adjustment convenience may view near-store and in-store C&C as an opportunity to shift to online while preserving the flexibility of top-up shopping. Thus, they may shift their shopping from brick and mortar to C&C. For shoppers with a high need for adjustment convenience, however, we do not expect a primary-demand lift. If anything, the shift to C&C may reduce their total grocery spending because of the reduced opportunities for top-up and impulse buying.
8 We control for within-retailer changes in price and assortment over time through retailer-specific time fixed effects.
9 According to Experian, a local market represents the area within which daily life is organized. These local markets do not correspond to zip code areas but represent neighborhoods in which a certain geographic, social, cultural, and economic coherence exists and in which consumer needs are relatively homogenous.
The different usage rates of the C&Cs for Leclerc and Intermarché are an artifact of the different timing of the retailers' rollout. Leclerc's C&Cs opened earlier in our observation window than Intermarché's.
Because larger households typically have larger basket sizes, we orthogonalize these two variables by regressing household size on basket size and using the residuals (instead of household size per se) in Equation 1. For a similar "partialing-out" procedure, see [8] and [44].
We use the absorption technique to estimate Equation 1 with household fixed-effects, which implies that we do not estimate separate constants for each household. Instead, we control for household differences through within-household centering of the variables, with an appropriate standard error adjustment for the degrees of freedom lost in estimating N individual means ([5], p. 224).
Adding the two control-function regressors to the main model assumes that the residuals of the retailer-decision and the household-adoption models are independent ([31]; [62]). The correlation between these two residuals is indeed very small (−.06 for Leclerc,.12 for Intermarché).
We set the group of shopper characteristics associated with a specific convenience feature (Table 2) to values capturing low versus high needs for that feature, while all other shopper characteristics are held constant at their average values. For continuous variables, we set the "low" and "high" values to the 10th and 90th percentiles of the variables' distributions across panelists. For dummy variables, we set "low" to 0 and "high" to 1. For instance, for access convenience, a low-need shopper corresponds to a shopper who lives in a nonrural area (0 for the variable "rural") and who shops infrequently on weekends (10th percentile for the variable "weekend shopping"). All combinations of variables at the selected 10% and 90% values are within the range of our data.
References A3Distrib (2012), "Drive Insights," https://www.olivierdauvers.fr/2012/07/05/la-v2-de-drive-insights-est-disponible/.
Ailawadi Kusum L., Ma Yu, Grewal Dhruv. (2018), "The Club Store Effect: Impact of Shopping in Warehouse Club Stores on Consumers' Packaged Food Purchase,"Journal of Marketing Research, 55 (2), 193–207.
Ailawadi Kusum L., Neslin Scott A. (1998), "The Effect of Promotion on Consumption: Buying More and Using It Faster," Journal of Marketing Research, 35 (3), 390–98.
Ailawadi Kusum L., Zhang Jie, Krishna Aradhna, Kruger Michael W. (2010), "When Wal-Mart Enters: How Incumbent Retailers React and How This Affects Their Sales Outcomes," Journal of Marketing Research, 47 (4), 577–93.
Angrist Joshu D., Pischke Jörn-Steffen. (2009), Mostly Harmless Econometrics. An Empiricist's Companion. Princeton, NJ: Princeton University Press.
Ankeny Jason. (2017), "Are Retailers Eating Themselves Alive? 8 Experts Chew on the Question," Retail Dive. (March 16), https://www.retaildive.com/news/are-retailers-eating-themselves-alive-8-experts-chew-on-the-question/438174/.
Avery Jill, Steenburgh Thomas J., Deighton John, Caravella Mary. (2012), "Adding Bricks to Clicks: Predicting the Patterns of Cross-Channel Elasticities over Time," Journal of Marketing, 76 (3), 96–111.
Batra Rajeev, Ramaswamy Venkatram, Alden Dana L., Steenkamp Jan-Benedict E.M. (2000), "Effects of Brand Local and Nonlocal Origin on Consumer Attitudes in Developing Countries," Journal of Consumer Psychology, 9 (2), 83–95.
Bell David R., Chiang Jeongwen, Padmanabhan V. (1999), "The Decomposition of Promotional Response: An Empirical Generalization," Marketing Science, 18 (4), 504–26.
Bell David R., Gallino Santiago, Moreno Antonio. (2018), "Offline Showrooms in Omni-Channel Retail: Demand and Operational Benefits," Management Science, 64 (4) 1639–51.
Bell David R., Ho Teck-Hua, Tang Christopher S. (1998), "Determining Where to Shop: Fixed and Variable Costs of Shopping," Journal of Marketing Research, 35 (3), 352–69.
Bell David R., Song Sangyoung. (2007), "Neighborhood Effects and Trial on the Internet: Evidence from Online Grocery Retailing," Quantitative Marketing and Economics, 5 (4), 361–400.
Bertrand Marianne, Duflo Esther, Mullainathan Sendhil. (2004), "How Much Should We Trust Differences-in-Differences Estimates?" Quarterly Journal of Economics, 119 (1), 249–75.
Briesch Richard A., Chintagunta Pradeep K., Fox Edward J. (2009), "How Does Assortment Affect Grocery Store Choice?" Journal of Marketing Research, 46 (2), 176–89.
Bronnenberg Bart. (2018), "Retailing and Consumer Demand for Convenience," in Handbook of Research on Retailing, Gielens Katrijn, Gijsbrechts Els, eds. Cheltenham, UK:Edward Elgar Publishing, 17–43.
Campbell Dennis, Frei Frances. (2009), "Cost Structure, Customer Profitability, and Retention Implications of Self-Service Distribution Channels: Evidence from Customer Behavior in an Online Banking Channel," Management Science, 56 (1), 4–24.
Campo Katia, Breugelmans Els. (2015), "Buying Groceries in Brick and Click Stores: Category Allocation Decisions and the Moderating Effect of Online Buying Experience," Journal of Interactive Marketing, 31 (3), 63–78.
Capital (2018), "Le 'Drive', un Cadeau Empoisonné Pour La Grande Distribution," (July 27), https://www.capital.fr/entreprises-marches/le-drive-un-cadeau-empoisonne-pour-la-grande-distribution-814745.
Caussil Jean-Noël. (2013), "Le Drive: Un Modèle en Questions," LSA (January 17), https://www.lsa-conso.fr/le-drive-un-modele-en-questions,137405.
Chaity Nadia. (2015), "Exclusive Report: Click & Collect Drive Through Model and Its Application to the US Market," http://www.terrapinn.com/template/Live/documents/6783/13724?#sthash.VcZNPqt5.dpbs.
Chandon Pierre, Wansink Brian. (2002), "When Are Stockpiled Products Consumed Faster? A Convenience-Salience Framework of Post-Purchase Consumption Incidence and Quantity," Journal of Marketing Research, 39 (8), 321–35.
Chintagunta Pradeep K., Chu Junhong, Cebollada Javier. (2012), "Quantifying Transaction Costs in Online/Offline Grocery Channel Choice," Marketing Science, 31 (1), 96–114.
Choi Jeonghye, Hui Sam K., Bell David R. (2010), "Spatiotemporal Analysis of Imitation Behavior across New Buyers at an Online Grocery Retailer," Journal of Marketing Research, 47 (1), 75–89.
Chu Junhong, Arce-Urriza Marta, Cebollada-Calvo José-Javier, Chintagunta Pradeep K. (2010), "An Empirical Analysis of Shopping Behavior Across Online and Offline Channels for Grocery Products: The Moderating Effects of Household and Product Characteristics," Journal of Interactive Marketing, 24 (4), 251–68.
Coughlan Anne T., Jap Sandy D. (2016), A Field Guide to Channel Strategy: Building Routes to Market. CreateSpace Independent Publishing Platform.
Craft Kathy. (2019), "Click-and-Collect Is King, and Grocery Stores Must Shift Their Back-of-House Operations," Grocery Dive (October 28), https://www.grocerydive.com/news/click-and-collect-is-king-andgrocery-stores-must-shift-their-back-of-hous/565911.
Dekimpe Marnik G., Geyskens Inge, Gielens Katrijn. (2020), "Using Technology to Bring Online Convenience to Offline Shopping," Marketing Letters, 31, 25–29.
Deleersnyder Barbara, Geyskens Inge, Gielens Katrijn, Dekimpe Marnik G. (2002), "How Cannibalistic Is the Internet Channel? A Study of the Newspaper Industry in the United Kingdom and the Netherlands," International Journal of Research in Marketing, 19 (4), 337–48.
Delvallée Julie. (2016), "Image-prix des Enseignes Alimentaires Passée au Peigne Fin," LSA(May 26), https://www.lsa-conso.fr/l-image-prix-des-enseignes-alimentaires-passee-au-peigne-fin,239145.
Donato-Weinstein Nathan. (2015), "Exclusive: Amazon Planning Drive-Up Grocery Stores with the First Likely Coming to Sunnyvale—Sources," Silicon Valley Business Journal(July 23), http://www.bizjournals.com/sanjose/news/2015/07/23/exclusive-amazon-planning-drive-up-grocery-stores.html.
Ebbes Peter, Papies Dominik, van Heerde Harald J. (2016), "Dealing with Endogeneity: A Nontechnical Guide for Marketing Researchers," in Handbook of Market Research, Homburg C., Klarmann M., Vomberg A., eds. Cham, Switzerland: Springer International Publishing, 1–37.
Edge by Ascential (2020), "Supply Chain & Fulfillment. COVID-19 Update," (June 12), https://retailinsight.ascentialedge.com/research?searchTerm=Supply%20Chain%20%26%20Fulfillment.%20COVID-19%20Update&contentType=.
Edge Retail Insight (2018), "France Omnichannel," Go-to-Market Report, (November 23), https://retailinsight.ascentialedge.com/research/Market%20Report/iiavjrmqs9ez8elkx6a6kf/France%20Omnichannel%20Go-To-Market%20Report,%20November%202018.
Edge Retail Insight (2019a), "The Rise and Evolution of Click & Collect," (April 12),https://retailinsight.ascentialedge.com/research/Analyst%20Update/1g2qce1mj3b2pcknzqe0kq/The%20rise%20and%20evolution%20of%20click%20&%20collect.
Edge Retail Insight (2019b), "The Rise of Online Grocery and What CPGs can Learn from Drive," (December 12),https://retailinsight.ascentialedge.com/research/Analyst%20Update/q70yburoa4i1ykxpxgtsab/The%20rise%20of%20online%20grocery%20and%20what%20CPGs%20can%20learn%20from%20drives.
Editions Dauvers (2013a), "Drive et Rentabilité," [Available athttps://www.olivierdauvers.fr/wp-content/uploads/2014/07/Dossier-Grande-Conso-Drive-Renta.pdf].
Editions Dauvers (2013b), "Drive: Les 7 Questions que l'On me Pose le Plus Souvent," (June)[Available athttps://www.olivierdauvers.fr/wp-content/uploads/2013/06/Dossier-Grande-Conso-Drive1.pdf].
Espiner Tom. (2015), "Is Click-and-Collect 'Cannibalising' Retailers?" BBC News (July 3), https://www.bbc.com/news/business-33371265.
Galante Nicolò, López Enrique García, Munby Sarah. (2013), "The Future of Online Grocery in Europe," McKinsey. (March 1), https://www.mckinsey.com/industries/retail/our-insights/the-future-of-online-grocery-in-europe.
Gallino Santiago, Moreno Antonio. (2014), "Integration of Online and Offline Channels in Retail: The Impact of Sharing Reliable Inventory Availability Information," Management Science, 60 (6), 1434–51.
Gensler Sonja, Leeflang Peter, Skiera Bernd. (2012), "Impact of Online Channel Use on Customer Revenues and Costs to Serve: Considering Product Portfolios and Self-Selection," International Journal of Research in Marketing, 29 (2), 192–201.
Gijsbrechts Els, Campo Katia, Vroegrijk Mark. (2018), "Save or (Over-)Spend? The Impact of Hard-Discounter Patronage on Consumer Grocery Spending," International Journal of Research in Marketing, 35 (2), 270–88.
Haddon Heather. (2019), "Online Grocery Upheaval Hits America's Biggest Chain," The Wall Street Journal (April 22), https://www.wsj.com/articles/americas-biggest-supermarket-company-struggles-with-online-grocery-upheaval-11555877123.
Keller Kristopher O., Geyskens Inge, Dekimpe Marnik G. (2020), "Opening the Umbrella: The Effects of Rebranding Multiple Category-Specific Private-Label Brands to One Umbrella Brand," Journal of Marketing Research, 57 (4), 677–94.
Lambrecht Anja, Seim Katja, Tucker Catherine. (2011), "Stuck in the Adoption Funnel: The Effect of Interruptions in the Adoption Process on Usage," Marketing Science, 30 (2), 355–67.
Lee Jae Young, Bell David R. (2013), "Neighborhood Social Capital and Social Learning for Experience Attributes of Products," Marketing Science, 32 (6), 960–76.
Liu Huan, Lobschat Lara, Verhoef Peter. (2018), "Multichannel Retailing: A Review and Research Agenda," Foundations and Trends in Marketing, 12 (1), 1–79.
Pauwels Koen, Neslin Scott A. (2015), "Building with Bricks and Mortar: The Revenue Impact of Opening Physical Stores in a Multichannel Environment," Journal of Retailing, 91 (2), 182–97.
PlanetRetail (2015), "Fulfillment of the Future," (December 15).
Porter Michael E. (1980), Competitive Strategy: Techniques for Analyzing Industries and Competitors. New York: The Free Press,
Post & Parcel (2016), "UK Online Delivery and Click-and-Collect to Double by 2025, Claims New Report," (March 14), https://postandparcel.info/71804/news/uk-online-delivery-and-click-and-collect-to-double-by-2025-claims-new-report.
Rosenbaum Paul R., Rubin Donald B. (1983), "The Central Role of the Propensity Score in Observational Studies for Causal Effects," Biometrika, 70 (1), 41–55.
Rossi Peter E. (2014), "Even the Rich Can Make Themselves Poor: A Critical Examination of IV Methods in Marketing Applications," Marketing Science, 33 (5), 655–72.
Seiders Kathleen, Berry Leonard, Gresham Larry G. (2000), "Attention, Retailers! How Convenient Is Your Convenience Strategy?" MIT Sloan Management Review, 41 (Spring), 79–89.
SKUlocal (2018), "How Online Grocery Has Impacted Basket Size and Frequency Among Shoppers," (November 7), http://www.skulocal.com/insights/how-online-grocery-has-impacted-basket-size-and-frequency-among-shoppers.
Stern Louis W., El-Ansary Adel I. (1988), Marketing Channels, Englewood Cliffs, NJ: Prentice Hall.
Stratton Leslie S. (2012), "The Role of Preferences and Opportunity Costs in Determining the Time Allocated to Housework," American Economic Review, 102 (3), 606–11.
Van Nierop Erjen, Leeflang Peter S.H., Teerling Marije L., Huizingh Eelko K.R.E. (2011), "The Impact of the Introduction and Use of an Informational Website on Offline Customer Buying Behavior," International Journal of Research in Marketing, 28 (2), 155–65.
Wang Kitty, Goldfarb Avi. (2017), "Can Offline Stores Drive Online Sales?" Journal of Marketing Research, 54 (5), 706–19.
Wansink Brian. (1996), "Can Package Size Accelerate Usage Volume?" Journal of Marketing, 60 (3), 1–14.
Wertenbroch Klaus. (1998), "Consumption Self-Control by Rationing Purchase Quantities of Virtue and Vice," Marketing Science, 17 (4), 317–37.
Wooldridge Jeffrey M. (2010), Econometric Analysis of Cross Section and Panel Data, Cambridge, MA: MIT Press.
Wooldridge Jeffrey M. (2015), "Control Function Methods in Applied Econometrics," Journal of Human Resources, 50 (2), 420–45.
~~~~~~~~
By Katrijn Gielens; Els Gijsbrechts and Inge Geyskens
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 96- Onboarding Salespeople: Socialization Approaches. By: Wiseman, Phillip; Ahearne, Michael; Hall, Zachary; Tirunillai, Seshadri. Journal of Marketing. Mar2022, p1. DOI: 10.1177/00222429221076437.
Ahead of Print- Database:
- Business Source Complete
Record: 97- Order Matters: Rating Service Professionals First Reduces Tipping Amount. By: Chen, Jinjie; Xu, Alison Jing; Rodas, Maria A.; Liu, Xuefeng. Journal of Marketing. Jun2022, p1. DOI: 10.1177/00222429221098698.
Ahead of Print- Database:
- Business Source Complete
Record: 98- Penny for Your Preferences: Leveraging Self-Expression to Encourage Small Prosocial Gifts. By: Rifkin, Jacqueline R.; Du, Katherine M.; Berger, Jonah. Journal of Marketing. May2021, Vol. 85 Issue 3, p204-219. 16p. 1 Black and White Photograph, 1 Chart, 2 Graphs. DOI: 10.1177/0022242920928064.
- Database:
- Business Source Complete
Penny for Your Preferences: Leveraging Self-Expression to Encourage Small Prosocial Gifts
Prior approaches that leverage identity to motivate prosocial behavior are often limited to the set of people who already strongly identify with an organization (e.g., prior donors) or by the costs and challenges associated with developing stronger organization-linked identities among a broader audience (e.g., encouraging more people to care). In contrast, this research demonstrates that small prosocial gifts, such as tips or small donations, can be encouraged by framing the act of giving as an opportunity to express identity-relevant preferences—even if such preferences are not explicitly related to prosociality or the organization in need. Rather than simply asking people to give, the "dueling preferences" approach investigated in this research frames the act of giving as a choice between two options (e.g., cats vs. dogs, chocolate vs. vanilla ice cream). Dueling preferences increases prosocial giving by providing potential givers with a greater opportunity for self-expression—an intrinsically desirable opportunity. Seven experiments conducted in the laboratory, online, and in the field support this theorized process while casting doubt on relevant alternatives. This research contributes to work on self-expression and identity and sheds light on how organizations can encourage prosocial behavior.
Keywords: donation; identity; prosocial; self-expression; tipping
Prosocial behavior, defined as "actions intended to benefit one or more people other than oneself" ([ 4], p. 463), is critical for society to function. Prosocial organizations "aid populations and people who would otherwise be overlooked, and they fill the gaps where public programs cannot provide sufficient support" ([11]). After Hurricanes Katrina and Rita, for instance, Catholic Charities agencies heavily relied on volunteering and prosocial gifts to help victims ([16]); similarly, service workers depend on tips to make minimum wage and stave off poverty ([28]).
But while it is of practical, managerial, and societal interest to increase prosocial giving, motivating prosocial behavior is often difficult. Younger consumers are tipping less than prior generations ([63]), for example, and most charities report that they do not have enough funds to meet need ([47]). In this research, we examine two prosocial behaviors, tipping and donating, which are acts of giving intended to help others. We focus on small gifts in particular (e.g., a few dollars or less), which, according to recent research (e.g., [55]) and popular press outlets ([67]; [72]), are common and important in both tipping and charitable giving contexts.
One marketing solution to increase prosocial giving is identity-based appeals (e.g., [ 1]; [18]; [21]; [35]; [70]; [71]). In particular, organizations can increase prosocial giving by appealing to people who strongly identify with them (e.g., donated to them previously; [36]) or by growing the set of people who view an organization as linked with their own identity ([ 2]; [48]).
These approaches, however, have challenges. Appealing to people who already identify with an organization can promote giving, but this strategy is limited to the group of people who already care. Encouraging more people to strongly identify with an organization can lead more people to care, but shifting identities can be difficult and costly (e.g., [61]).
To address some of these challenges, we examine a novel approach, occasionally observed in the field, that we call "dueling preferences." Rather than simply asking people to tip, some cafés have started to frame tipping as a choice between two options (e.g., summer vs. winter; for more examples, see https://tinyurl.com/exampleduelingpreferences). Similarly, the American Society for the Prevention of Cruelty to Animals recently asked people to donate money by expressing their preference for cats or dogs ("Vote for your Paw-sident").
While this approach is intriguing, little is known about its effectiveness and underlying mechanism. Does dueling preferences actually bolster giving compared with traditional approaches? If so, what psychological mechanism might drive this effect? And are all dueling pairs equally effective, or are some more effective than others?
Seven experiments conducted in the laboratory, online, and in the field examine whether dueling preferences increases small prosocial gifts, and if so, why. The results demonstrate that dueling preferences increases the rate of giving and the amount given versus standard giving appeals. Furthermore, we explore how self-expression drives these effects: Dueling preferences frames the act of giving as a choice between two options, which can offer givers a heightened opportunity to say something about who they are (i.e., self-express). This additional, intrinsically desirable opportunity to express oneself, in turn, increases prosocial giving. As a result, duels are more effective than standard appeals when they are seen as providing greater opportunities for self-expression and presented to people who value self-expression to a greater extent.
This work makes three key contributions. First, while there is continual interest in strategies for motivating prosocial giving, prior solutions often suffer from issues of limited scope and difficulty of implementation. To our knowledge, we are the first to systematically examine the effectiveness and underlying mechanism of dueling preferences, an identity-based solution that is relatively flexible and easy to employ. In doing so, we provide insights into how to most effectively implement this strategy.
Second, we add to the academic literature on how identity motivates behavior. While prior work has shown that evoking a specific identity (e.g., helper) can motivate identity-congruent behavior (e.g., helping; [14]; [49]), we illustrate that this identity–behavior congruence relationship can be decoupled. Our work demonstrates that prosocial behavior can even be motivated by identities that are not traditionally associated with prosociality (e.g., cat person).
Third, we shed light on the behavioral consequences of self-expressive choice. Prior work has found that people make qualitatively different selections when choices are viewed as more (vs. less) self-expressive (e.g., selecting more variety [[37]], choosing brand-name vs. generic products [[38]]). We further demonstrate that self-expressiveness can influence whether people make a choice at all: enhanced self-expressiveness, elicited by dueling preferences, increases people's likelihood of tipping or donating.
Identity is a key driver of consumer behavior ([41]; [43]; [44]; [50]; [51]; [62]). The identity-based motivation model suggests that people are motivated to behave congruently with their identities ([49]; [60]). Someone who identifies as an athlete, for instance, will be motivated to act consistently with that identity by going to the gym or by purchasing new running shoes.
Some identities are chronically active, while others can become salient in a given situation ([50]). The aforementioned athlete, for example, may also be a mother. If her mother identity is situationally cued (e.g., by spending time with her child), she may be more motivated to behave congruently with that identity (e.g., by purchasing something for her child) relative to her athlete identity. Organizations often leverage this principle. By creating an association between their product (e.g., Jif peanut butter) and a specific identity (e.g., "choosy moms"), for example, marketers can increase the likelihood that consumers with that identity will purchase their product ([19]; [22], [23]; [51]).
Consistent with the role of identity in driving congruent behavior, research has examined how evoking prosocial identities can elicit prosocial behavior ([ 1]; [18]; [21]; [26]; [35]; [71]). Some work advocates for appealing to potential donors who already strongly identify with an organization. Reminding people that they previously donated to an organization, for example, enhances the likelihood that they will donate to that organization again ([36]). Similarly, employees who strongly identify with a company are more likely to volunteer time and effort to their company ([48]). In both cases, relying on existing strong identities evokes subsequent prosocial behavior and support.
Unfortunately, however, this approach is limited to people who already strongly identify with an organization. Reminding blood donors that they donated previously may encourage them to donate again, for instance, but by focusing solely on existing donors, this approach is limited in scope. Similarly, employees who already identify with their company may volunteer at company events, but that number of potential volunteers is finite.
Another identity-based approach aims to bolster identification among people who may only weakly identify with an organization. College students may not initially feel connected to their university, for example, which might prevent desired donation rates. Universities can bolster identification, though, through school-themed events such as homecoming ([ 2]), thereby increasing alumni donations. Similarly, companies can implement employee team-building and appreciation programs to strengthen employees' identification with their company and, in turn, promote prosocial behavior towards the company (e.g., volunteering time; [ 2]; [48]). While this approach can broaden the number of potential givers, it has its own challenges. For instance, efforts to strengthen identification among potential givers can be financially burdensome and take years to develop ([61]).
To address these issues, some work has tried to develop identity-based appeals that are easier to implement than prior approaches. Compared with a verb-based appeal (e.g., "please help"), for example, simply invoking the equivalent prosocial identity (e.g., "please be a 'helper'") encourages prosocial behavior ([14]; [15]). However, this approach, like all of the aforementioned approaches, still relies on the principle of identity–behavior congruence—prosocial identities motivate prosocial behavior. Might there be an even broader identity-based approach?
Leveraging an approach that we call "dueling preferences," we suggest that identity–behavior congruence can be decoupled, such that any valued identity (e.g., cat person) can motivate giving—even if that identity is unrelated to prosociality. In other words, rather than being restricted to the set of people that already hold a strong organization-linked identity or exerting resources to grow that set, we suggest that, by employing the dueling preferences approach, a broader set of valued identities can motivate prosocial behavior. Instead of simply asking people to give, the dueling preferences approach frames the act of prosocial giving as an opportunity to choose between multiple—almost always two—options.
We propose that, compared with a standard giving appeal, dueling preferences can boost small prosocial gifts by increasing the perceived opportunity to self-express valued identities or preferences. Tipping or donating in and of itself can be an opportunity to express oneself. People like to view themselves as prosocial (e.g., [ 7]), and giving to a charity or tipping a barista provides the opportunity to express that. Thus, even a standard prosocial giving appeal provides some opportunity for self-expression.
By framing a tip or donation as a means to express a preference between multiple options, we suggest that dueling preferences can heighten the perceived opportunity for self-expression even further. Beyond simply showing that one is generous or considerate via donating or tipping, dueling preferences can transform the act of giving into an opportunity to express something additional about oneself—for instance, whether one prefers dogs or cats, or vanilla or chocolate ice cream.
Choice, by its very nature, is self-expressive ([12]), as it involves stating a preference for one option over another. [20] explore vice–virtue choices, for example, and note that "[the choice of] virtue and vice become more meaningful in the presence of each other as available opportunities" (p. 24). In other words, the self-expressiveness of choosing an action (e.g., jogging) is greater when it is chosen in the presence of a competing option (e.g., watching TV). Applied to our context, compared with the question "Do you like dogs?," we suggest that the answer to an either/or question such as "Which do you like more, dogs or cats?" should be perceived as more self-expressive. Put simply, stating that you like dogs should feel like it says more about you when in the presence of another available and plausible option, like cats. Consequently, the either/or format of dueling preferences should amplify the perceived opportunity for self-expression.
We suggest that this greater opportunity for self-expression, in turn, can increase prosocial giving. On average, people enjoy ([31]) and have an intrinsic desire to express ([65]) who they are, and they engage in specific behaviors to communicate identity-relevant information to themselves and/or others. These behaviors include choosing, wearing, and talking about products and experiences that reflect their identities ([ 5]; [ 8], [ 9]; [25]; [43]; [59]; [68], [69]).
Because people generally find self-expression intrinsically motivating (and, thus, fundamentally desirable), they should be willing to pay to engage in it. Neuroimaging work underscores that people are wired to crave self-expression opportunities and that such opportunities are processed like rewards, much like opportunities for food or money ([65]). Therefore, just as people pay more for any other desirable product attribute (e.g., a brand they like, a product in their favorite color), they should be willing to pay more for an opportunity to express something valuable about themselves. We propose that this intrinsic value of self-expression underlies the effect of dueling preferences.
Consequently, with the dueling preferences approach, we extend the identity-behavior congruence principle to show that identities can motivate seemingly unrelated behaviors, due to the intrinsically motivating pull of self-expression. Specifically, we predict that
- H1: Compared with a standard appeal, the dueling preferences appoach increases prosocial giving.
- H2: This effect (H1) occurs because, relative to a standard appeal, dueling preferences can provide greater opportunity for self-expression.
Our theoretical perspective suggests when and for whom dueling preferences should be most effective. Duels can enhance the opportunity for self-expression, and therefore increase giving, by virtue of involving choice. Critically, though, this effect should only occur if the dueling options are seen as relatively self-expressive by potential givers. The extent to which dueling options are perceived as self-expressive may generalize across individuals broadly (e.g., the choice of cats vs. dogs is likely seen as more self-expressive than A vs. B) or vary from person to person (e.g., some individuals find animals more self-expressive than others). The extent to which dueling options are perceived as self-expressive should have a direct impact on how effective the appeal will be.
Second, our perspective suggests that dueling preferences should be more effective among people who chronically value the act of self-expression to a greater extent or in situations that heighten one's need for self-expression. In other words, while duels often provide a greater opportunity for self-expression, that opportunity should be more desirable, and thus more likely to boost giving, for someone who strongly values self-expression (either chronically or for situational reasons). For someone who does not value self-expression, dueling preferences should be less likely to increase giving.
Finally, we are not suggesting that duels are the only way to provide givers with an opportunity for self-expression. Indeed, an opportunity for self-expression can be provided without the presence of choice. For example, one could express that they love dogs in the absence of cats as a competing option, and, accordingly, an appeal leveraging this "nonchoice" self-expression option should increase giving versus a standard appeal, according to our theory. We directly test such an expressive, nonchoice appeal in Experiment 4. Rather, we are suggesting that, by virtue of involving choice, the dueling preferences appeal can provide people with an even greater perceived opportunity to express something about who they are and, thus, can be especially effective at increasing giving.
Seven experiments test our theorizing. Experiment 1 demonstrates that dueling preferences increases tipping (vs. a standard appeal) in a café, and Experiment 2 demonstrates that dueling preferences increases incentive-compatible donations in a more controlled setting.
The next five experiments examine the underlying process. Experiment 3 measures the perceived self-expressiveness of the giving opportunity and tests the mechanism through mediation. Experiment 4 provides further evidence for our framework by testing the underlying role of self-expressiveness and how the dueling preferences format uniquely amplifies this mechanism. Experiments 5, 6a, and 6b leverage both mediation and moderation to further provide evidence for our theory, testing the role of individual differences in a duel domain's self-expressiveness (Experiment 5), as well as the chronic (Experiment 6a) and situational (Experiment 6b) need to self-express.
The experiments also test several alternative mechanisms, including that the duel increases the feeling of choice ([13]; [34]; [52]), is novel ([32]), has a game-like feeling, and features competition ([10]; [33]; [66]). Specifically, we cast doubt on these alternatives by measurement (Experiment 3) and in situations where many of these alternatives are held constant (Experiments 4–6b).
Finally, across experiments, we employed best practices regarding data cleaning. Following suggestions in [45], we excluded observations containing ( 1) IP addresses that appeared two or more times (i.e., duplicate or multiple entries), ( 2) memory check failures, and ( 3) outliers based on experiment timing.[ 6] Samples reported in each experiment are after data cleaning occurred, and original samples sizes prior to data cleaning are reported in Web Appendix A. Unless otherwise specified in Web Appendix A, results are substantively unchanged without employing these exclusions.
Experiment 1 tests whether dueling preferences boosts prosocial giving in the field. While paying for their beverages, café customers were shown either a standard giving appeal (i.e., tip jar) or a preference duel. We predicted (H1) that dueling preferences would encourage more people to tip and increase tip revenue overall.
We conducted a two-cell (standard appeal vs. dueling preferences) between-subjects field experiment during a single business day (7:00 a.m.–4:00 p.m.) at a locally owned café in Durham, North Carolina. Tipping opportunities were placed at their typical location by the cash register. Employees of the café were blind to our hypotheses.
The only difference between conditions was whether patrons encountered a tip jar (standard appeal condition; Web Appendix B) or a cats versus dogs duel (dueling preferences condition; Figure 1). Conditions were alternated every 45 minutes during "peak" times, as reported by the owner before the start of the experiment, and every hour during "off-peak" times (Web Appendix C). This resulted in ten time periods (five standard appeal periods, five duel periods), and each appeal type appeared for 4.5 hours in total.
Graph: Figure 1. Field Experiment 1 dueling preferences condition.
We inconspicuously observed customers and recorded whether they tipped. Because the café's point-of-sale system did not allow people to tip via credit card, we focused on cash-paying customers (N = 44).[ 7] Logistical constraints prevented us from recording individual tip amounts, but we were able to measure the total amount tipped during each time period.
As we predicted (H1), compared with the standard appeal, dueling preferences led more customers to tip (Pduel = 77.3% vs. Pstandard = 40.9%; χ2( 1, N = 44) = 6.02, p =.014). Furthermore, the duel more than doubled the amount of money raised overall during the business day (Totalduel = $18.61 vs. Totalstandard = $7.87; Web Appendix C).
Experiment 1 provides preliminary support for our theorizing in the field. Relative to a standard appeal, dueling preferences led more consumers to tip and more than doubled the amount of money tipped overall (H1). This experiment also provides initial evidence that identity–behavior congruence can be decoupled, given that the desired behavior (tipping a barista) was motivated by leveraging the expression of something conceptually unrelated (preference for dogs vs. cats).
Experiment 2 uses a more controlled setting and an alternative duel to investigate whether dueling preferences can increase another form of prosocial giving: donations. Participants were given an incentive-compatible donation opportunity. We predicted (H1) that, compared with a standard appeal, a duel would result in greater rates of donation and in greater amounts, thereby raising more money for charity.
One hundred seventy-eight Amazon Mechanical Turk (MTurk) workers (63.5% male; Mage = 33.5 years) were randomly assigned to a two-cell (standard appeal vs. dueling preferences) between-subjects design. In addition to their $.50 payment for completing the survey, all participants were given a $.10 bonus to use for a potential donation.
First, we manipulated appeal type. All participants were given an opportunity to donate to the American Red Cross. After reading introductory information about the charity, participants were shown images depicting their donation opportunity. In the standard appeal condition, the image displayed a jar labeled "donations" featuring a Red Cross logo and the caption, "make your donation!" In the dueling preferences condition, the image displayed two jars, one labeled "chocolate ice cream" and the other labeled "vanilla ice cream," with the caption "vote with your donation!" and the Red Cross logo beside it (Web Appendix B).
Participants then completed the dependent variable, donation behavior. Participants selected how much of their $.10 bonus they wanted to donate (ranging from $.00 to $.10 in $.01 increments). We examined both whether people donated and how much they chose to donate. Because donation amount was heavily skewed toward zero (Kolmogorov–Smirnov test of a single distribution indicating nonnormality: D(178) =.33, p <.001), we performed a nonparametric test (Mann–Whitney test of equality of two distributions) on donation amount.
As we predicted (H1), compared with the standard appeal, the duel led more participants to donate (Pduel = 54.0% vs. Pstandard = 37.4%; χ2( 1, N = 178) = 4.98, p =.026). The duel also directionally increased how much people donated (Mduel = $.037 vs. Mstandard = $.030; U = 3,462.50, Z = −1.59, p =.111, d =.17). While this difference was not statistically significant, the total amount donated in the duel condition was 28% greater than the total amount donated in the standard appeal condition (Totalduel = $3.83 vs. Totalstandard = $2.99).
Experiment 2 demonstrates that dueling preferences can also increase donations. Compared with a standard appeal, dueling preferences encouraged more people to donate (H1). Furthermore, dueling preferences raised more money for the American Red Cross, illustrating the value of this approach for nonprofit organizations. The experiment also provides further evidence that this approach is able to decouple identity-behavior congruence: The desired behavior (giving to the Red Cross) was motivated by leveraging the expression of something conceptually unrelated (ice cream preference).
Experiment 3 tests whether self-expression underlies the positive effect of dueling preferences on prosocial giving (H2) as well as several potential alternative explanations. Rather than providing a heightened opportunity for self-expression, one could argue that duels boost prosocial giving because they are more novel ([32]). Alternatively, competition can increase motivation ([10]; [33]; [66]), so perhaps duels increase giving because they stimulate competition or feel like a game. Similarly, embedding a choice between multiple options into the appeal could enhance a feeling of choice, which could also bolster giving. Indeed, providing choice to consumers has been shown to encourage action by increasing their feelings of agency and control ([52]). In this study, we measured all of these possibilities and tested whether they can explain the results on real donation behavior.
We also used a novel context to test generalizability. One could wonder whether the dueling preferences effect is somehow restricted to displaying two jars. To rule out this possibility, this experiment removed any mention or images of jars.
Sixty-three MTurk workers (68.3% male; Mage = 34.8 years) were randomly assigned to a two-cell (standard appeal vs. dueling preferences) between-subjects design. As in Experiment 2, in addition to their $.50 payment for completing the survey, all participants were given a $.10 bonus to use for a potential donation.
First, participants received information about Afterschool Alliance, a real charity that promotes access to affordable, quality afterschool programs. They were told that they had a donation opportunity based on the charity's recent online fundraising material for their "Play Outside" campaign.
Then, we manipulated appeal type. In the standard appeal condition, the materials said, "Donate TODAY to the Play Outside Campaign!" In the dueling preferences condition, the material said, "Which makes you happier: SUMMER or WINTER? Tell us how you feel with your donations to the Play Outside Campaign!" (Web appendix B). In both conditions, the material also featured the Afterschool Alliance logo and a tagline, "Encouraging kids to play outside safely all year round."
Second, participants completed the dependent variable, real donation behavior. As in Experiment 2, participants selected how much of their $.10 bonus they wanted to donate (ranging from $.00 to $.10 in $.01 increments). We examined both whether and how much people donated. Donation amount was heavily skewed toward zero (Kolmogorov–Smirnov test of a single distribution indicating nonnormality: D(63) =.29, p <.001), so we used a nonparametric test (Mann–Whitney test of equality of two distributions).
Third, we collected the proposed mediator. Participants reported how expressive the donation opportunity felt with two items (r =.82; averaged to create a self-expressiveness index): ( 1) "Compared to other donation situations I usually see, this situation gave me MORE of an opportunity to express something about who I am" (1 = "strongly disagree," and 7 = "strongly agree") and ( 2) "Compared to other donation situations you usually see, how much did this donation situation give you an opportunity to say something about who you are?" (1 = "this situation was LESS of an opportunity to say something about who I am," and 7 = "this situation was MORE of an opportunity to say something about who I am").
Finally, we measured potential alternative explanations in a linguistically similar format as the proposed mediator (i.e., "Compared to other donation situations I usually see..."). Participants reported the degree to which the appeal gave them more of an opportunity to participate in a competition, felt more novel, gave them more of an opportunity to make a choice, and felt more like a game (1 = "strongly disagree," and 7 = "strongly agree").
Finally, we collected a memory check that assessed whether participants recalled seeing a prompt to express their preference ("Did the donation opportunity you viewed mention anything about sharing your preference between two options?"; 1 = yes, 0 = no). Those in the standard appeal condition (who, in reality, were not shown a prompt to self-express) who responded "yes," and those in the duel condition (who, in reality, were shown a prompt to self-express) who responded "no," were coded as failing the memory check.
As we predicted (H1), compared with the standard appeal, the duel increased willingness to donate (Pduel = 69.0% vs. Pstandard = 38.2%; χ2( 1, N = 63) = 5.93, p =.015). The duel also increased how much people donated (Mduel = $.055 vs. Mstandard = $.028; U = 332.50, Z = −2.38, p =.017, d =.63) and, overall, raised 68% more money for Afterschool Alliance (Totalduel = $1.60 vs. Totalstandard = $.95).
Consistent with our theorizing, compared with the standard appeal, the duel was perceived to be a greater opportunity for self-expression (Mduel = 5.71 vs. Mstandard = 4.15; t(61) = 2.74, p =.008, d =.69). Furthermore, as we predicted (H2), a mediation analysis ([30]; Model 4 with 5,000 bootstraps) demonstrated that the effects on donation incidence (ab =.45, 95% confidence interval [CI] = [.10, 1.11]) and amount (ab =.68, 95% CI = [.21, 1.44]) were driven by perceived self-expressiveness.
Finally, we examined alternative explanations. Compared with the standard appeal, the duel was viewed as more novel (Mduel = 6.03 vs. Mstandard = 3.79; t(61) = 3.69, p <.001, d =.93), like a game (Mduel = 4.97 vs. Mstandard = 3.32; t(61) = 2.55, p =.013, d =.64), and competitive (Mduel = 4.59 vs. Mstandard = 2.56; t(61) = 3.35, p =.001, d =.84), and it marginally increased the feeling of choice (Mduel = 5.90 vs. Mstandard = 4.76; t(61) = 1.67, p =.099, d =.43). Importantly, although some alternatives mediated the observed effects when entered individually (i.e., without self-expression in the model), virtually none of them consistently drove the observed effects when entered in parallel with self-expression. Moreover, self-expression remained significant in these parallel mediation models (for detailed results, see Web Appendix D). Thus, while dueling preferences differs from standard appeals on several dimensions, the difference in perceived self-expressiveness appears to be the most important and predictive driver of the effects on prosocial giving.
Experiment 3 provides further evidence that dueling preferences can increase real prosocial giving behavior (H1) while providing support for the proposed underlying mechanism (H2). Dueling preferences increased real donations because it provided greater perceived opportunity for self-expression.
Additional analyses also cast doubt on a range of alternative explanations including competition, novelty, feelings of choice, or a game-like feeling. Furthermore, two larger conceptual replicates, including a version that randomized the presentation order of all potential mediators (i.e., the proposed self-expression mechanism and all potential alternatives), found similar results (Web Appendix E). Together, these studies provide convergent support for our proposed self-expression mechanism.
While Experiment 3 provides evidence for the role of self-expression and begins to cast doubt on several alternatives, one could still wonder if some other feature of duels, rather than perceived self-expressiveness, is driving these effects. To test this possibility, Experiment 4 adds a duel condition that provides less of an opportunity for self-expression—the choice between the letters A versus B. If our theorizing about the role of self-expressiveness is correct, the effect of dueling preferences on giving should be stronger (weaker) when a duel involves options that are seen as more (less) self-expressive.
As discussed previously, one might also wonder if choice is the only way to provide givers with an opportunity for self-expression. According to our theory, while duels are not the only way to provide an opportunity to self-express—one could simply express what one likes without the presence of a comparison option—they are particularly effective at doing so by virtue of their either/or format. To test this argument, Experiment 4 also adds a condition that provides a nonchoice opportunity for self-expression. We predicted that, due to the intrinsic appeal of self-expression, an expressive nonchoice appeal should increase giving relative to a standard appeal. Critically, however, we further predicted that a preference duel with expressive options should increase giving the most. In other words, we are expecting that the expressive preference duel should uniquely amplify perceived self-expressiveness and, in turn, prosocial giving. We also held constant the actual preference being expressed across the expressive duel and expressive nonchoice appeals in order to tease out the specific proposed role of choice in amplifying perceived self-expressiveness.
Four hundred sixty-three participants on Prolific Academic (42.8% male; Mage = 34.9 years) were randomly assigned to one of four appeal types (standard appeal, less expressive duel, expressive nonchoice, expressive duel). First, participants in the expressive nonchoice appeal condition indicated whether they preferred the mountains or the beach. We included this question so that we could later provide an expressive nonchoice appeal that matched their preference.
Second, all participants completed a nine-question filler task involving counting syllables and judging the brightness of colors (Web Appendix B). We included this filler task to create distance between the preference expression (in the expressive nonchoice condition) and the main study.
Third, all participants were given information about a real environmental conservation charity, the Environmental Investigation Agency (EIA), and were told that the charity was testing online fundraising materials. Fourth, we manipulated appeal type. All appeals (Web Appendix B) featured the EIA logo. In the standard appeal condition, the appeal said, "Donate TODAY!" and featured a button that said, "Click this button to donate." In the expressive duel condition, the appeal said, "Which do you love more: The MOUNTAINS or the BEACH?" and featured two "donation buttons" allowing donors to click the button that expresses their preference.
In the less expressive duel condition, the appeal involved a choice designed to be relatively low in self-expression—the preference for letters. It said, "Which do you love more: A or B?" and featured two donation buttons similar to those in the expressive duel condition. Both the expressive and less expressive duels offer the opportunity to make a choice between two options, but the expressive duel condition should provide a greater opportunity for self-expression. We later validate this assumption in our measure of the mediator.
The expressive nonchoice condition provided the opportunity for self-expression without choice. We took the information participants provided previously (i.e., whether they preferred the mountains or the beach) and piped that preference into the text of the appeal, so that the appeal read "Do you love the [beach/mountains]?" and featured a button stating, "I LOVE THE [BEACH/MOUNTAINS]." By controlling for the actual preference being expressed across the expressive duel and expressive nonchoice conditions (i.e., the fact that the person likes the mountains or beach), we isolate the specific proposed role of choice in amplifying perceived self-expressiveness.
Fifth, we collected the dependent measures. We measured donation likelihood with two items (r =.81): ( 1) "Based on the material that you viewed above, how likely would you be to donate any amount of money to Environmental Investigation Agency?" (1 = "not at all," and 7 = "very") and ( 2) "Rate your agreement with the following statement: This fundraising material makes me more likely to donate to the Environmental Investigation Agency" (1 = "strongly disagree," and 7 = "strongly agree"). In addition, we asked participants how much they would donate to the EIA (up to $10).
Sixth, we collected the mediator using the two items from Experiment 3 on nine-point scales (r =.83). Finally, we collected a memory check. All participants responded to the question "Which of the donation opportunities below did you see for the Environmental Investigation Agency earlier in this study?" and were shown photos of each condition's appeal. Those who recalled seeing a different appeal than the one they actually saw were coded as failing the memory check.
Omnibus tests revealed significant differences between conditions on both donation likelihood (F( 3, 459) = 10.87, p <.001) and amount (F( 3, 459) = 7.80, p <.001; Table 1). As in the prior experiments, compared with the standard appeal, the expressive duel increased donation likelihood (Mexp. duel = 3.01 vs. Mstandard = 2.22; t(459) = 4.26, p <.001, d =.56) and amount (Mexp. duel = $3.12 vs. Mstandard = $1.55; t(459) = 4.02, p <.001, d =.52), again confirming H1.
Graph
Table 1. Expressive Duels Provide Greater Opportunity for Self-Expression (vs. Other Appeals), Which Drives Giving (Experiment 4).
| Appeal Type | Donation Likelihood | Donation Amount ($) | Self-Expressiveness |
|---|
| Standard | 2.22 (1.13) | 1.55 (2.43) | 3.18 (1.77) |
| Less expressive duel | 2.06 (1.23) | 1.50 (2.51) | 3.67 (2.08) |
| Expressive nonchoice | 2.62 (1.39) | 2.32 (2.98) | 4.42 (1.96) |
| Expressive duel | 3.01 (1.66) | 3.12 (3.48) | 5.36 (2.21) |
10022242920928064 Notes: Statistics reported as M (SD).
Importantly, and as we expected, compared with the expressive duel, the less expressive duel significantly reduced giving (likelihood: t(459) =5.20, p <.001, d =.65; amount: t(459) = 4.20, p <.001, d =.53). In addition, the less expressive duel did not increase giving relative to the standard appeal (likelihood: Mless exp. duel = 2.06 vs. Mstandard = 2.22; t(459) =.89, p =.38, d =.14; amount: Mless exp. duel = $1.50 vs. Mstandard = $1.55; t(459) =.14, p =.89, d =.02), suggesting that choice alone, without self-expression, cannot boost prosocial giving.
Finally, we examined the expressive nonchoice appeal. Consistent with the hypothesized role of self-expression, the expressive nonchoice appeal increased donation likelihood relative to both the standard appeal (likelihood: Mexp. NC = 2.62; t(459) = 2.24, p =.025, d =.32; amount: Mexp. NC = $2.31; t(459) = 2.04, p =.042, d =.28) and the less expressive duel (likelihood: t(459) = 3.19, p =.001, d =.39; amount: t(459) = 2.21, p =.028, d =.49). However, as we predicted, the expressive nonchoice appeal was not as effective as the expressive duel (likelihood: t(459) = 2.19, p =.029, d =.25; amount: t(459) = 2.15, p =.032, d =.25). In other words, although a nonchoice appeal that provides the opportunity for self-expression increases giving somewhat, an expressive duel increases giving even more.
Again, omnibus tests revealed significant differences between conditions on self-expressiveness (F( 3, 459) = 24.87, p <.001; Table 1). Consistent with Experiment 3, compared with the standard appeal, the expressive duel was seen as an enhanced opportunity for self-expression (Mexp. duel = 5.36 vs. Mstandard = 3.18; t(459) = 8.06, p <.001, d = 1.09), and mediation analysis ([30]; Model 4 with 5,000 bootstraps) revealed that this drove the effects on donation likelihood (ab =.77, 95% CI = [.55, 1.02]).
The expressive duel also was seen as providing greater opportunity for self-expression than the less expressive duel (Mless exp. duel = 3.67; t(459) = 6.30, p <.001, d =.79), which also drove the effect on donation likelihood (ab =.60, 95% CI = [.38,.85]). This result underscores the role of self-expression, and not the presence of choice alone, in explaining the effectiveness of the dueling preferences approach.
Also, as expected, the expressive nonchoice appeal was seen as more self-expressive than the standard appeal (Mexp. NC = 4.42; t(459) = 4.72, p <.001, d =.66), and this drove its positive effect on donation likelihood (ab =.44, 95% CI = [.26,.63]). Thus, even without the presence of dueling options, providing people with a chance to self-express can boost giving. The expressive nonchoice appeal also boosted self-expressiveness (t(459) = 2.86, p =.004, d =.37) and, in turn, giving (marginally; ab =.26, 90% CI = [.08,.45]), relative to the less expressive duel.
Most importantly, and critical to our theory, the expressive duel was perceived as the most self-expressive (vs. expressive nonchoice appeal: t(459) = 3.66, p <.001, d =.45), and this difference drove the effect on donation likelihood (ab =.33, 95% CI = [.14,.54]). In other words, holding the act of expression constant, the dueling format amplifies self-expressiveness and, in turn, giving. Thus, the expressive duel is best able to boost giving versus all other appeals.
Mediation results on donation amount were identical and will thus not be discussed for the sake of brevity.
Experiment 4 further demonstrates self-expression as the mechanism underlying this effect and underscores the value of the dueling preferences format in amplifying this mechanism. As with prior studies, the expressive duel increased donation likelihood relative to the standard appeal. Moreover, consistent with the notion that this effect is driven by self-expression, the expressive duel also outperformed a duel containing less expressive options. This result further underscores that the presence of dueling options (i.e., choice) alone cannot fully explain the pattern of results.
We also found that an expressive nonchoice appeal, which allowed givers to express a liking for the beach or mountains, increased giving compared with the standard appeal—again supporting the powerful and motivating role of self-expression. Importantly though, and key to the present research, the dueling format generated the greatest perceived self-expressiveness and, thus, the greatest levels of prosocial giving. Notably, across the expressive nonchoice appeal and the expressive duel, participants were engaging in the exact same act of expression (i.e., stating that they like the beach or mountains, depending on their preference), and we varied only whether this act was situated in the form of a duel (i.e., choice). These results demonstrate that although duels are not the only way to provide self-expression and enhance giving, they are a privileged type of appeal that, through their structure, can uniquely boost expression and, thus, prosocial giving.
Thus far, we have shown that duels can be effective because they provide a heightened opportunity for self-expression relative to standard appeals. While Experiment 4 varied duel content (i.e., contrasting duels that were more vs. less self-expressive), in Experiment 5 we hold the duel constant and leverage individual differences to test our theory.
For any given duel, individuals should differ in the extent to which they perceive the duel's domain as self-expressive. While many people find pet preferences (e.g., whether they prefer cats or dogs) self-expressive, for instance, others may find them less so. In this study, we test the proposed mechanism by measuring individual differences in perceived self-expressiveness of the duel's domain and predict that it will moderate the effects on prosocial giving. Specifically, we predicted that the effectiveness of dueling preferences would be amplified among those who perceive the duel's domain to be more self-expressive and attenuated among those who perceive the duel's domain to be less self-expressive.
One hundred twenty-one laboratory participants at the University of Pennsylvania (36.4% male; Mage = 24.6 years) were randomly assigned to one of two appeal types (standard appeal vs. duel) in a 2 (appeal type) × continuous (duel domain self-expressiveness) between-subjects design.
First, all participants imagined ordering a beverage at a café and approaching the cash register to pay. Second, we manipulated the appeal type using a standard appeal and a duel (cats vs. dogs), depending on condition, similar to those presented to participants in Experiment 1 (Web Appendix B).
Third, we collected the dependent measures. We asked participants if they would tip (1 = yes, 0 = no), and if so, how much (open ended). One participant indicated that they would tip $150 (+5 SD from mean) and was thus removed from all analyses. We also asked participants how likely they would be to tip (1 = "not at all," and 9 = "extremely"). The likelihood and amount measures showed identical results to the incidence measure and thus are not discussed further, for the sake of brevity.
Fourth, using three items (α =.77; averaged to create a domain self-expressiveness index) we measured the expressiveness of the duel domain as an individual difference: ( 1) "How much do you care about the choice of cats versus dogs?"; ( 2) "How strong is your preference for one over the other?"; and ( 3) "How much does your choice of pets say something about who you are?" (1 = "not at all," and 7 = "very much"). This index did not vary by condition (t(119) = 1.50, p =.14).
Finally, we collected a memory check. All participants responded to the question "How many tip jars did you see in the tipping situation?" (0, 1, 2). Those in the standard appeal condition who responded "0" or "2" and those in the duel conditions who responded "0" or "1" were coded as failing the memory check.[ 8]
First, a binary logistic regression regressing tip incidence on appeal type (−1 = standard appeal, 1 = duel), domain self-expressiveness (mean-centered), and their interaction revealed the predicted effect of appeal type (b =.51, Wald χ2( 1, N = 121) = 6.40, p =.011). Consistent with the other experiments, compared with a standard appeal, dueling preferences increased willingness to engage in prosocial giving (Pduel = 52.4% vs. Pstandard = 31.0%), further confirming H1.
Second, and more importantly, this effect was qualified by the predicted (marginal) interaction (b =.27, Wald χ2( 1, N = 121) = 3.42, p =.065, Johnson–Neyman point at 4.06; [64]). Spotlight analysis provides more insight into the pattern of effects (Figure 2). As we predicted, among people who found the domain highly self-expressive (+1 SD), dueling preferences increased tipping (Pduel = 75.2% vs. Pstandard = 32.8%; b =.94, Wald χ2( 1, N = 121) = 9.90, p =.002). Consistent with the underlying role of self-expressiveness, however, dueling preferences no longer garnered greater tipping likelihood among people who found the duel's domain less self-expressive (−1 SD: Pduel = 33.1% vs. Pstandard = 28.8%; b =.13, Wald χ2( 1, N = 121) =.18, p =.67).
Graph: Figure 2. Duel domain expressiveness moderates the effect (Experiment 5).
Experiment 5 uses process by moderation to further underscore how dueling preferences shapes prosocial giving. Consistent with the hypothesized role of self-expression, the effect of the duel was moderated by individual differences in perceived expressiveness of the duel's domain. Among people for whom the duel domain felt more self-expressive, the duel increased willingness to tip. Among people for whom the duel felt less self-expressive, however, the duel did not increase willingness to tip.
This moderation also casts further doubt on several alternative explanations. While one might argue that the beneficial effects of dueling preferences are driven by characteristics inherent to most duels (e.g., they are novel, they feature competition and choice), such alternatives cannot easily explain why domain self-expressiveness would moderate the effect.
Our final two experiments further test the hypothesized underlying process by moderating the second link in our proposed causal chain. We have suggested that duels can enhance prosocial giving by providing an additional opportunity for self-expression (H2). Experiments 4 and 5 tested the "a-link" of this theoretical process, or whether duels are more effective when they are seen as providing more of an opportunity for expression. Experiments 6a and 6b test moderation on the "b-link," or the extent to which enhanced self-expressiveness motivates greater levels of giving.
Although we have argued and research demonstrates that people generally have an intrinsic desire to express themselves ([31]; [65]), there is some variation in this desire or need. People can vary, either chronically, as an individual difference, or situationally, in the degree to which they value ([39]; [40]) or need self-expression ([17]; [29]). This variation should impact people's responses to a self-expression opportunity. If our theorizing about the underlying role of self-expressiveness is correct, dueling preferences should have a stronger (weaker) effect on giving among people who value self-expression to a greater (lesser) extent, both as an individual difference and situationally. In other words, both individual and situational variation in value for self-expression should affect how intrinsically desirable the self-expression opportunity is, thus influencing how much people give to an appeal that incorporates such an opportunity. In Experiment 6a, we measured how much people chronically value self-expression and test whether it moderates our effect. In Experiment 6b, we manipulated need for self-expression.
Five hundred thirty-four MTurk workers (52.1% male; Mage = 36.9 years) were randomly assigned to one of two appeal types (standard appeal vs. duel) in a 2 (appeal type) × continuous (value of self-expression) between-subjects design. First, we measured how much each participant values self-expression using five items (α =.90; averaged to create a value of self-expression [VSE] index) adapted from the Value of Expressiveness Questionnaire ([40]; Web Appendix B). Second, to create distance between the individual difference measure and the main experiment, all participants completed a brief filler task similar to the one used in Experiment 4.
Third, we manipulated the appeal type. Participants imagined that they were browsing social media and came upon a local animal shelter's recent post "regarding their yearly fundraising campaign." All participants viewed a fictional post that contained the shelter's name ("Smile Animal Shelter"), two images of cats and dogs, and the focal caption, "Donate today by clicking here!" In the standard appeal condition, this was the entirety of the post. In the duel condition, two additions were made: Above the cats and dogs images, text was added that said, "Which do you like more: Cats or Dogs?" and the focal caption read, "Share your preference and donate today by clicking here!" (Web Appendix B).
Fourth, we collected the dependent measure, donation likelihood (1 = "not at all," and 9 = "extremely"). Fifth, we collected the mediator using the items from Experiments 3 and 4 (r =.79).[ 9] Finally, we collected a memory check that assessed whether participants recalled seeing a prompt to express their preference between cats and dogs, similar to the one used in Experiment 3.
First, regressing donation likelihood on appeal type (−1 = standard appeal, 1 = duel), VSE (mean-centered), and their interaction revealed the predicted effect of appeal type (b =.28, t(530) = 3.46, p =.001). Consistent with the first five experiments, compared with a standard appeal, dueling preferences increased the likelihood of prosocial giving (Mduel = 3.52 vs. Mstandard = 2.98), again confirming H1.
Second, and more importantly, this effect was qualified by the predicted (marginal) interaction (b =.11, t(530) = 1.80, p =.072, Johnson–Neyman point at 3.53; Figure 3). As we predicted, among people who highly value self-expression (+1 SD), dueling preferences increased willingness to donate (Mduel = 4.16 vs. Mstandard = 3.31; b =.43, t(530) = 3.73, p <.001). Among people who value self-expression to a lesser extent (−1 SD), however, the effect was attenuated (Mduel = 2.89 vs. Mstandard = 2.63; b =.13, t(530) = 1.15, p =.251). Examining the simple slopes provides further insight into this interaction: The relationship between valuing self-expression and donation likelihood was greater in the dueling preferences condition (b =.46, t(530) = 6.05, p <.001) than in the standard appeal condition (b =.25, t(530) = 2.72, p =.007).
Graph: Figure 3. Value of self-expression moderates the effect (Experiment 6a).
We conducted a similar regression on self-expressiveness. As we predicted, there was an effect of appeal type (b =.50, t(530) = 5.86, p <.001). As in prior experiments, compared with the standard appeal, dueling preferences increased the perceived self-expressiveness of the giving opportunity (Mduel = 5.55 vs. Mstandard = 4.56).[10]
Finally, a moderated mediation analysis ([30]; Model 14 with 5,000 bootstraps) using VSE as the "b-link" moderator supports the hypothesized underlying process. Results demonstrated that the effect of dueling preferences on donation likelihood was driven by self-expressiveness (H2), and that this effect was stronger (weaker) for individuals who place greater (lesser) value on self-expression (index of moderated mediation =.04, 95% CI = [.02,.07]). Specifically, while the self-expressiveness of the donation opportunity mediated the effects at all levels of VSE (at −1 SD: ab =.16, 95% CI = [.10,.24]; at +1 SD: ab =.28, 95% CI = [.18,.38]), the size of the indirect effect is amplified among people who value self-expression to a greater extent. In other words, because they provide a greater opportunity for self-expression than standard appeals, duels exert a stronger (weaker) positive influence on prosocial giving among individuals who strongly (weakly) value self-expression.
Experiment 6a further underscores the underlying role of self-expressiveness. As we predicted (H2), the effect on prosocial giving was both mediated by the self-expressiveness of the donation opportunity and moderated by the importance an individual chronically places on self-expression. While the duel featured in this experiment was generally perceived to be a greater opportunity for self-expression (relative to the standard appeal), people's willingness to give money for that opportunity depended on how strongly they chronically valued self-expression. Among people who strongly value self-expression, dueling preferences' effects were amplified, and among people who weakly value it, the effects were attenuated.
Experiment 6b experimentally manipulates the need for self-expression. We leveraged a satiation manipulation ([17]) in which participants were given the opportunity to express themselves prior to a donation scenario, thereby temporarily reducing their need for self-expression. We predicted that the effect of dueling preferences would be attenuated when the need for self-expression was reduced.
Three hundred twenty-four participants on Prolific Academic (49.1% male; Mage = 33.7 years) were randomly assigned to condition in a 2 (appeal: standard vs. duel) × 2 (need for expression: baseline vs. satiated) between-subjects design.
First, participants in the satiated condition completed a self-expression task adapted from [17] shown to satiate consumers' need to self-express. Specifically, participants were asked to report several their favorite things, including their favorite sports teams, musical artist, color, hobby, subject in school, television show, book, and movie (Web Appendix B). Participants in the baseline condition did not complete this task, meaning their baseline need to self-express, which we have leveraged in prior studies, was not satiated. A separate test (N = 141 on Prolific Academic) using the manipulation check items from [17]; Experiment 5) confirmed that the satiation condition does indeed decrease participants' need to express (Mbaseline = 4.84 vs. Msatiated = 4.36; t(139) = 2.46, p =.016, d =.42).
Second, we manipulated the appeal type. To do so, we used the same hypothetical donation scenario for the EIA as in Experiment 4. Participants were shown a standard appeal or a mountains versus beach duel, depending on condition.
Third, we collected our dependent variables. Participants were asked to report their donation likelihood using the two measures from Experiment 4 (r =.78) as well as whether they would donate (0 = no, 1 = yes). The incidence measure showed substantively similar results to the likelihood measure and thus will not be discussed further.
Finally, we also collected a memory check using the same measure from Experiment 4. Those who recalled seeing a different appeal than the one they actually saw were coded as failing the memory check.
A two-way ANOVA including the need for self-expression and appeal type on donation likelihood revealed the predicted (marginal) interaction (F( 1, 320) = 2.95, p =.087). First, consistent with the prior experiments, in the baseline condition, participants reported a greater likelihood to donate to the duel versus the standard appeal (Mduel = 3.10 vs. Mstandard = 2.17; F( 1, 320) = 18.21, p <.001, d =.64). Second, and consistent with the proposed underlying role of self-expression, when the need for expression was previously satiated, this effect was attenuated (Mduel = 2.51 vs. Mstandard = 2.11; F( 1, 320) = 3.40, p =.07, d =.30).
Examining the simple effects within appeal condition yields further insight. Participants' likelihood to give in the standard appeal condition was equal across baseline and satiation conditions (F( 1, 320) =.08, p =.78, d =.05). In the duel condition, however, previously satiating the desire for self-expression reduced participants' reported likelihood to engage in prosocial giving (F( 1, 320) = 7.44, p =.007, d =.40). In other words, attenuating the need to self-express reduced participants' response to the duel appeal only.
Using a process-by-moderation design, Experiment 6b provides convergent support for the underlying role of self-expressiveness. Satiating the need for expression moderated the effect of dueling preferences. As with prior studies, dueling preferences enhanced prosocial giving in the baseline condition (i.e., when we leveraged people's natural desire to express themselves). When participants had the chance to express themselves prior to the giving opportunity, however, this effect was attenuated.
These results also provide a final argument against several of the alternative explanations discussed in this work. While our satiation manipulation reduced participants' need for self-expression (thereby attenuating dueling preferences' effects), the manipulation seems less likely to have influenced participants' responses to aforementioned possibilities such as choice, gamification, or novelty.
Marketing academics and practitioners have long been interested in motivating prosocial giving—developing better marketing tactics for a better world. While several identity-based techniques exist, many suffer from either constraints of scope or difficulty of implementation. This paper explores whether a novel technique, dueling preferences, can alleviate some of these challenges by using choice to leverage the motivating power of self-expression.
Seven experiments illustrate that dueling preferences increases prosocial giving. Across real (Experiments 1–3) and imagined (Experiments 4–6b) donations (Experiments 2–4, 6a–b) and tips (Experiments 1 and 5), dueling preferences increased the rate of giving and the total amount raised for the entity in need. Importantly, this approach does not require a public display to increase giving: Not only do we show robust results in both in-person (i.e., relatively public) and online (i.e., relatively private) settings across studies, but a supplemental study (Web Appendix G) also provides initial evidence to suggest that the effect of duels is not moderated by whether the appeal was encountered in private (with no observers) or public (with observers). Together, these results demonstrate the intrinsic appeal of dueling preferences.
The experiments also illustrate the process underlying this effect. Dueling preferences encourages giving by increasing the perceived opportunity to self-express valued identities and preferences (Experiments 3–6b). As such, the strategy is more effective when duels are seen as more self-expressive (Experiments 4–5) and among those who strongly value self-expression chronically or situationally (Experiments 6a–b).
In addition, the findings cast doubt on a range of alternative explanations (i.e., feeling of choice, novelty, competition, and gamification) as the crucial drivers of this effect. These alternatives do not mediate the effect to the extent that self-expression does (Experiment 3) and have trouble explaining why the self-expressiveness of the duel (Experiments 4–5) and the value of/need for self-expression (Experiments 6a–6b) moderate the effects.
Finally, demonstrating these effects across multiple duels (cats vs. dogs, vanilla vs. chocolate ice cream, summer vs. winter, mountains vs. beach), operationalizations (side-by-side jars, online appeals, and captioned social media posts), samples (café-goers, students, and online panels), and giving behaviors (cash tips and small online donations) underscores the generalizability of the effect.
This work makes several theoretical contributions. First, we contribute to the research on self-expression. While prior work has shown that evoking specific identities (e.g., helper) can motivate congruent behavior (e.g., helping; [14]; [49]), we demonstrate that an evoked identity need not be congruent with a desired behavior to have motivating power. Across our experiments, we consistently show that prosocial behavior can be motivated by identities not traditionally associated with prosociality (e.g., cat person, chocolate ice cream lover). This provides new insight into the appeal of self-expression, demonstrating an increased scope of identities that may motivate prosocial behavior.
Second, we shed additional light on the behavioral consequences of self-expressive choice. Prior work has shown that more (vs. less) self-expressive choices lead people to make different types of selections (e.g., more variety; [37]). Building on this work, we demonstrate that enhanced self-expressiveness, elicited especially effectively through preference duels, also influences whether people choose (e.g., to donate or tip) in the first place.
Third, we contribute to the ongoing discussion about the benefits of providing choice. For instance, while consumers generally like choice, it can become overwhelming (e.g., [34]) or make people selfish ([53]; [54]). Adding to this discussion, we show that specific types of choice—choices that provide an opportunity for self-expression—can promote prosocial behaviors. We also add to this literature by demonstrating dueling preferences' effects in situations where a giver's "choice" has no actual influence on the outcome of the prosocial gift (cf. choice of charity; [52]) and by directly imbuing self-expressive choice into the giving appeal, rather than making self-expression salient in advance (cf. priming comparative mindsets; [73]).
This work also has practical implications for managers interested in increasing prosocial giving. While some cafés and organizations have started using this approach, its actual effectiveness and underlying mechanism remains undetermined. We demonstrate that dueling preferences can boost prosocial giving while alleviating some of the practical challenges with prior approaches. Rather than expending resources to build a connection between givers and the organization, firms can leverage what potential givers already care about by creating an appropriate preference duel. By employing a tailored preference duel for only a single day, for instance, our field experiment more than doubled a local café's tips (Experiment 1). Similarly, rather than limiting communication to people who have organization-linked identities, the dueling preferences approach leverages a variety of existing identities, which need not be associated with prosociality, to be effective.
Essential to its managerial relevance, our research not only reveals the effectiveness of dueling preferences, but it also illustrates several considerations that can make this strategy more or less effective. As documented across experiments, dueling preferences has a more positive impact on behavior when it provides people the opportunity to express themselves in important domains, among individuals that value self-expression, and in situations where the need to self-express is relatively high. Perhaps most importantly, our results suggest that opportunities to self-express will be most effective at eliciting giving when situated within an either/or choice. Consequently, managers interested in using dueling preferences should consider what choice options their targets see as expressive, the extent to which their retail setting sparks a need for self-expression, and the extent to which their target audience values self-expression. Cafés in college towns, for instance, could leverage important rivalries or sports events to craft identity-relevant duels that will increase tipping.
Finally, we note that dueling preferences could be understood by managers as an implementation of "task unification" ([27]). Task unification occurs when an existing product feature or component is designed to accomplish a second feature or task, making a single feature do "double duty" (p. 6). By layering the opportunity for self-expression onto a giving appeal, dueling preferences can be thought of as one instance of task unification—unifying the action of giving with the action of self-expression. Importantly, though, this novel instance of task unification differs from existing examples in two key ways: first, it does not economize on production or materials costs; second, the tasks being unified are not product features, they are consumer actions (giving and expression).
Several directions deserve future study. A first direction to explore is when and why dueling preferences can be ineffective or even backfire (i.e., elicit less giving than a standard appeal). The results of Experiments 4 and 5 suggest that less expressive duels fail to increase giving versus a standard appeal; however, future work might explore whether and when duels reduce giving. We speculate that this may occur when duels are seen as invasive, such as if they ask potential givers to express private, embarrassing, or taboo aspects of themselves, or if they are seen as gimmicky attempts at persuasion or overly trivial in light of the cause.
The number of dueling options could also lead to backfiring, thus warranting future research. We focused on two options, as that is what is most commonly seen in the field, but future work might explore the effects of providing additional dueling options. While more dueling options might enhance the perceived opportunity for self-expression, it might also induce a feeling of choice overload (e.g., [34]; [58]; [57]), thereby undermining dueling preferences' benefits.
A second area to explore in the future is the over-time and downstream consequences of the dueling preferences effect. Two questions seem particularly relevant: First, how well does this strategy work if employed repeatedly over time? Managers may be concerned that dueling preferences would cease to be effective if employed repeatedly. We asked this question to a barista at a café in Milwaukee, Wisconsin, that uses dueling preferences daily to elicit tips and gleaned one major insight: He believes that dueling preferences can remain effective over time if the content periodically changes. In fact, the barista we interviewed changes the content of the duel every morning when he opens the café. To contrast his approach, he described a different local business that also uses dueling preference to elicit tips but, unlike him, infrequently changes their duel content. He said that, upon seeing a commonly repeated duel involving the local sports teams at this other business, he thinks to himself, "I already told you that I like the [Milwaukee] Brewers." His reaction to this repeated duel suggests that potential givers may feel that their desire to express is satiated when they repeatedly encounter the same content. In support of this possibility, we find that when self-expression is previously satiated, dueling preferences is less effective at increasing giving (Experiment 6b). To avoid this issue of satiation, we suggest that firms and employees mimic the practice of the barista who we interviewed, regularly rotating the content of their duels. The question of how often the content needs to be rotated to maintain an optimal effect, though, remains open for future research.
Second, how might giving to a preference duel at one point in time shape future support for the organization? On the one hand, some work suggests that self-oriented benefits can "taint" altruistic acts ([46]; [56]) and discourage future prosocial behavior ([42]), suggesting that giving to a preference duel may fail to promote future support for an organization. On the other hand, it is also possible that a single act of giving might create a foot-in-the-door effect. Given that people are motivated to act consistently with prior actions ([ 6]; [24]), an initial act of self-expression (i.e., giving to a preference duel) may ultimately engender ongoing support for the organization.
A third area for future research to explore is whether these effects extend to other outcomes. Rather than soliciting small gifts, as we have done in this work, would duels succeed at garnering larger monetary gifts? It is unclear whether consumers would pay much more than a nominal amount to self-express something relatively trivial, such as a preference for chocolate or vanilla ice cream. Certain larger sums of money may serve as a boundary condition to this effect, only eliciting donations from those few who perceive the duel as extremely self-expressive. For instance, if a fundraising director were trying to garner $100 donations, they may need to employ dueling options that are extremely significant to the target population.
Similarly, might duels encourage giving not only money but also time or other prosocial behaviors? Organizations may be able to encourage volunteerism by framing sign-ups as a preference duel between identity-relevant options. Similarly, some people might be more likely to recycle if they are given the opportunity to drop their items in either a Red Sox or Yankees bin. Indeed, BallotBin, a U.K.-based company, designs custom preference duels for the disposal of cigarette butts (https://ballotbin.co.uk/). For instance, one recent "ballot bin" in London asks smokers whether flying or invisibility is the better superpower, allowing them to express their preference by depositing their cigarette butts in one of two labeled compartments.
Dueling preferences might motivate behavior even outside of prosocial domains. If Nikon wants to engage consumers on social media, for example, they could ask, "Will you use your Nikon on a beach vacation or a mountain vacation?" By turning consumers' responses into an opportunity for self-expression, such messaging might strengthen engagement.
Finally, future work might test duels against other kinds of giving appeals, such as an "urgent plea" for donations or a statement that donations will help an organization make progress to reach some target goal. This future direction would allow researchers to directly test the strength of self-expression as a motivating factor relative to other known motivating factors.
In conclusion, this research explores a novel technique to motivate prosocial giving. By framing the act of giving as a choice between two identity-relevant options and leveraging the inherent appeal of self-expression, dueling preferences can encourage greater rates of giving. By understanding what consumers value, organizations may be able to earn a penny for their preferences and, ultimately, get one step closer to building a better world.
Supplemental Material, PennyPref_JM_Final_Web_Appendix - Penny for Your Preferences: Leveraging Self-Expression to Encourage Small Prosocial Gifts
Supplemental Material, PennyPref_JM_Final_Web_Appendix for Penny for Your Preferences: Leveraging Self-Expression to Encourage Small Prosocial Gifts by Jacqueline R. Rifkin, Katherine M. Du and Jonah Berger in Journal of Marketing
Footnotes 1 Correspondence should be addressed to the first two authors, who contributed equally and whose order was determined by a coin flip.
2 Andrea Morales
3 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
4 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was funded by the Fuqua School of Business at Duke University, the Lubar School of Business at the University of Wisconsin–Milwaukee, and The Wharton School at the University of Pennsylvania.
5 Online supplement: https://doi.org/10.1177/0022242920928064
6 To determine outliers based on experiment timing, we used the iterative procedure described in [45], which included removing those who took greater than ±2.5 SD in terms of experiment timing in a first wave, followed by recalculating the timing distribution and removing a second wave of greater than ±2.5 SD in terms of timing.
7 The proportion of cash- (vs. credit-) paying customers did not differ across appeal type conditions (χ2(1, N = 94) =.34, p =.561).
8 We also collected a measure of the proposed mediator and used this to test for moderated mediation, although this was not the central focus of this study. Results were consistent with our theory. For measures and moderated mediation results, see Web Appendix F.
9 Tests of discriminant validity demonstrate that the mediator was distinct from the moderator. [3], p. 436) suggest that "discriminant validity among traits is achieved when the trait correlation differs significantly from 1.0." Significant difference from 1.0 is assessable by running a correlation between the supposed constructs and collecting a 95% confidence interval, and discriminant validity between constructs is suggested when the confidence interval does not include 1.0. This method suggests that VSE (the moderator) and self-expressiveness (the mediator) are distinct constructs (i.e., feature discriminant validity; r =.18, 95% CI = [.09,.28]). Principal axis factor analysis with direct oblimin rotation corroborates that the mediator and moderator items load onto two distinct factors (Eigenvalue for factor 2 = 1.68).
There was also an interaction (b =.15, t(530) = 2.33, p =.020) of appeal type and VSE on self-expressiveness, suggesting that the duel was viewed as more self-expressive than the standard appeal only among those who valued self-expression at least a small amount (Johnson–Neyman point at 2.76; 88% of participants were at and above this point). Because only a small percentage of our sample (12%) had low-enough VSE to undermine the perceived self-expressiveness of the duel, we do not focus on a-link moderation in our discussion of the results. Of potential interest, however, we do find that this interaction indeed drives the effect of dueling preferences ([30]; Model 7 with 5,000 bootstraps index of moderated mediation =.07, 95% CI = [.01,.14]). This result further supports our theory that seeing the duel as self-expressive is crucial in driving the effect of dueling preferences.
References Aaker Jennifer L., Akutsu Satoshi. (2009), "Why Do People Give? The Role of Identity in Giving," Journal of Consumer Psychology, 19 (3), 267–70.
Arnett Dennis B., German Steve D., Hunt Shelby D. (2003), "The Identity Salience Model of Relationship Marketing Success: The Case of Nonprofit Marketing," Journal of Marketing, 67 (2), 89–105.
Bagozzi Richard P., Yi Youjae, Phillips Lynn W. (1991), "Assessing Construct Validity in Organizational Research," Administrative Science Quarterly, 36 (3), 421–58.
Batson C. Daniel, Powell Adam A. (2003), "Altruism and Prosocial Behavior," in Handbook of Psychology, Vol. 5, Millan Theodore, Lerner Melvin J., eds. Hoboken, NJ: John Wiley & Sons, 463–84.
Belk Russell W. (1988), "Possessions and the Extended Self," Journal of Consumer Research, 15 (2), 139–68.
Bem Daryl J. (1972), "Self-Perception Theory," in Advances in Experimental Social Psychology, Vol. 6, Berkowitz Leonard, ed. New York: Academic Press, 1–62.
Bénabou Roland, Tirole Jean. (2011), "Identity, Morals, and Taboos: Beliefs as Assets," Quarterly Journal of Economics, 126 (2), 805–55.
Berger Jonah, Heath Chip. (2007), "Where Consumers Diverge from Others: Identity Signaling and Product Domains," Journal of Consumer Research, 34 (2), 121–34.
Berger Jonah, Heath Chip. (2008), "Who Drives Divergence? Identity Signaling, Outgroup Dissimilarity, and the Abandonment of Cultural Tastes," Journal of Personality and Social Psychology, 95 (3), 593–607.
Berger Jonah, Pope Devin. (2011), "Can Losing Lead to Winning?" Management Science, 57 (5), 817–27.
Biberdorf Jeremy. (2017), "Why Charity Is More Important than Ever," International Policy Digest (July 11), https://intpolicydigest.org/2017/07/11/charity-important-ever/.
Bodner Ronit, Prelec Drazen. (2003), "Self-Signaling and Diagnostic Utility in Everyday Decision Making," in The Psychology of Economic Decisions, Vol. 1, Brocas Isabelle, Carrillo Juan D., eds. Oxford, UK: Oxford University Press, 105–23.
Botti Simona, Iyengar Sheena S. (2004), "The Psychological Pleasure and Pain of Choosing: When People Prefer Choosing at the Cost of Subsequent Outcome Satisfaction," Journal of Personality and Social Psychology, 87 (3), 312–26.
Bryan Christopher J., Master Allison, Walton Gregory M. (2014), "'Helping' Versus 'Being a Helper': Invoking the Self to Increase Helping in Young Children," Child Development, 85 (5), 1836–42.
Bryan Christopher J., Walton Gregory M., Rogers Todd, Dweck Carol S. (2011), "Motivating Voter Turnout by Invoking the Self," Proceedings of the National Academy of Sciences, 108 (31), 12653–56.
Catholic Charities USA (2006), "Catholic Charities Network Helps 300,000 Victims of Hurricanes Katrina and Rita," ReliefWeb (February 22), https://reliefweb.int/report/united-states-america/catholic-charities-network-helps-300000-victims-hurricanes-katrina-and.
Chernev Alexander, Hamilton Ryan, Gal David. (2011), "Competing for Consumer Identity: Limits to Self-Expression and the Perils of Lifestyle Branding," Journal of Marketing, 75 (3), 66–82.
Darley John M., Batson Daniel C. (1973), "'From Jerusalem to Jericho': A Study of Situational and Dispositional Variables in Helping Behavior," Journal of Personality and Social Psychology, 27 (1), 100–108.
Deshpandé Rohit, Stayman Douglas M. (1994), "A Tale of Two Cities: Distinctiveness Theory and Advertising Effectiveness," Journal of Marketing Research, 31 (1), 57–64.
Dhar Ravi, Wertenboch Klaus. (2012), "Self-Signaling and the Costs and Benefits of Temptation in Consumer Choice," Journal of Marketing Research, 49 (1), 15–25.
Duclos Rod, Barasch Alixandra. (2014), "Prosocial Behavior in Intergroup Relations: How Donor Self-Construal and Recipient Group Membership Shape Generosity," Journal of Consumer Research, 41 (1), 93–108.
Escalas Jennifer E., Bettman James R. (2003), "You Are What They Eat: The Influence of Reference Groups on Consumers' Connection to Brands," Journal of Consumer Psychology, 13 (3), 339–48.
Escalas Jennifer E., Bettman James R. (2005), "Self-Construal, Reference Groups, and Brand Meaning," Journal of Consumer Research, 32 (3), 378–89.
Festinger Leon. (1957), A Theory of Cognitive Dissonance. Stanford, CA: Stanford University Press.
Gal David. (2015), "Identity-Signaling Behavior," in The Cambridge Handbook of Consumer Psychology, Norton Michael I., Rucker Derek D., Lamberton Cait, eds. Cambridge, UK: Cambridge University Press, 257–81.
Gergen Kenneth J., Gergen Mary M., Meter Kenneth. (1972), "Individual Orientations to Prosocial Behavior," Journal of Social Issues, 28 (3), 105–30.
Goldenberg Jacob, Horowitz Roni, Levav Amnon, Mazursky David. (2003), "Finding Your Innovation Sweet Spot," Harvard Business Review, 81 (3), 120–29.
Gould Elise, Cooper David. (2018), "Seven Facts About Tipped Workers and the Tipped Minimum Wage," Economic Policy Institute (May 31), https://www.epi.org/blog/seven-facts-about-tipped-workers-and-the-tipped-minimum-wage/.
Grewal Lauren, Stephen Andrew T., Coleman Nicole Verocchi. (2019), "When Posting About Products on Social Media Backfires: The Negative Effects of Consumer Identity Signaling on Product Interest," Journal of Marketing Research, 56 (2), 197–201.
Hayes Andrew F. (2013), Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach. New York: Guilford Press.
He Daniel, Melumad Shiri, Pham Michel Tuan. (2019), "The Pleasure of Assessing and Expressing Our Likes and Dislikes," Journal of Consumer Research, 46 (3), 545–63.
Hirschman Elizabeth C. (1980), "Innovativeness, Novelty Seeking, and Consumer Creativity," Journal of Consumer Research, 7 (3), 283–95.
Huang Szu-chi, Etkin Jordan, Jin Liyin. (2017), "How Winning Changes Motivation in Multiphase Competitions," Journal of Personality and Social Psychology, 112 (6), 813–37.
Iyengar Sheena S., Lepper Mark R. (2000), "When Choice Is Demotivating: Can One Desire Too Much of a Good Thing?" Journal of Personality and Social Psychology, 79 (6), 995–1006.
Kaikati Andrew M., Torelli Carlos J., Winterich Karen Page, Rodas Maria A. (2017), "Conforming Conservatives: How Salient Social Identities Can Increase Donations," Journal of Consumer Psychology, 27 (4), 422–34.
Kessler Judd B., Milkman Katherine L. (2016), "Identity in Charitable Giving," Management Science, 64 (2), 1–16.
Kim Heejung S., Drolet Aimee. (2003), "Choice and Self-Expression: A Cultural Analysis of Variety-Seeking," Journal of Personality and Social Psychology, 85 (2), 373–82.
Kim Heejung S., Drolet Aimee. (2009), "Express Your Social Self: Cultural Differences in Choice of Brand-Name Versus Generic Products," Personality and Social Psychology Bulletin, 35 (12), 1555–66.
Kim Heejung S., Markus Hazel Rose. (2002), "Freedom of Speech and Freedom of Silence: An Analysis of Talking as a Cultural Practice," in Engaging Cultural Differences: The Multicultural Challenge in Liberal Democracies, Shweder Richard A., Minow Martha, Markus Hazel Rose, eds. New York: Russell Sage Foundation, 432–52.
Kim Heejung S., Sherman David K. (2007), "'Express Yourself': Culture and the Effect of Self-Expression on Choice," Journal of Personality and Social Psychology, 92 (1), 1–11.
Kleine Rober E., Kleine Susan Schultz, Kernan Jerome B. (1993), "Mundane Consumption and the Self: A Social-Identity Perspective," Journal of Consumer Psychology, 2 (3), 209–35.
Kristofferson Kirk, White Katherine, Peloza John. (2013), "The Nature of Slacktivism: How The Social Observability of an Initial Act of Token Support Affects Subsequent Prosocial Action," Journal of Consumer Research, 40 (6), 1149–66.
Levy Sidney J. (1959), "Symbols for Sale," Harvard Business Review, 37 (4), 117–24.
Malhotra Naresh K. (1988), "Self-Concept and Product Choice: An Integrated Review," Journal of Economic Psychology, 9 (1), 1–28.
Meyvis Tom, Osselaer Stijn M.J. Van. (2018), "Increasing the Power of Your Study by Increasing the Effect Size," Journal of Consumer Research, 44 (5), 1157–73.
Newman George E., Cain Daylian M. (2014), "Tainted Altruism: When Doing Some Good Is Evaluated as Worse Than Doing No Good At All," Psychological Science, 25 (3), 648–55.
Nonprofit Finance Fund (2018), "State of the Nonprofit Sector Survey 2018," https://nff.org/surveydata.
O'Reilly Charles A., Chatman Jennifer. (1986), "Organizational Commitment and Psychological Attachment: The Effects of Compliance, Identification, and Internalization on Prosocial Behavior," Journal of Applied Psychology, 71 (3), 492–99.
Oyserman Daphna. (2009), "Identity-Based Motivation: Implications for Action-Readiness, Procedural-Readiness, and Consumer Behavior," Journal of Consumer Psychology, 19 (3), 250–60.
Reed Americus, II. (2004), "Activating the Self-Importance of Consumer Selves: Exploring Identity Salience Effects on Judgments," Journal of Consumer Research, 31 (2), 286–95.
Reed Americus, II, Forehand Mark R., Puntoni Stefano, Warlop Luk. (2012), "Identity-Based Consumer Behavior," International Journal of Research in Marketing, 29 (4), 310–21.
Robinson Stefanie Rosen, Irmak Caglar, Jayachandran Satish. (2012), "Choice of Cause in Cause-Related Marketing," Journal of Marketing, 76 (4), 126–39.
Savani Krishna, Rattan Aneeta. (2012), "A Choice Mind-Set Increases the Acceptance and Maintenance of Wealth Inequality," Psychological Science, 23 (7), 796–804.
Savani Krishna, Stephens Nicole M., Markus Hazel Rose. (2011), "The Unanticipated Interpersonal and Societal Consequences of Choice: Victim Blaming and Reduced Support for the Public Good," Psychological Science, 22 (6), 795–802.
Savary Jennifer, Goldsmith Kelly. (2020), "Unobserved Altruism: How Self-Signaling Motivations and Social Benefits Shape Willingness to Donate," Journal of Experimental Psychology: Applied(published online January 9), DOI:10.1037/xap0000261
Savary Jennifer, Li Charis X., Newman George E. (2020), "Exalted Purchases or Tainted Donations? Self-Signaling and the Evaluation of Charitable Incentives," Journal of Consumer Psychology, (published online January 18), DOI:10.1002/jcpy.1157.
Scheibehenne Benjamin, Greifeneder Rainer, Todd Peter M. (2010), "Can There Ever Be Too Many Options? A Meta-Analytic Review of Choice Overload," Journal of Consumer Research, 37 (3), 409–25.
Schwartz Barry. (2004), The Paradox of Choice: Why More Is Less. New York: Ecco.
Shavitt Sharon. (1990), "The Role of Attitude Objects in Attitude Functions," Journal of Experimental Social Psychology, 26 (2), 124–48.
Shavitt Sharon, Torelli Carlos J., Wong Jimmy. (2009), "Identity-Based Motivation: Constraints and Opportunities in Consumer Research," Journal of Consumer Psychology, 19 (3), 261–66.
Shefska Zach. (2016), "Fundraising Nightmare: The Cost of Donor Acquisition," Fundraising Report Card (August 26), https://fundraisingreportcard.com/donor-acquisition-cost/.
Sirgy M. Joseph. (1982), "Self-Concept in Consumer Behavior: A Critical Review," Journal of Consumer Research, 9 (3), 287–300.
Spector Nicole. (2018), "Dollars and Sense: Why Are Millennials Tipping Less Than Older Generations?" NBC News (June 27), https://www.nbcnews.com/better/business/dollars-sense-why-are-millennials-tipping-less-older-generations-ncna886966.
Spiller Stephen A., Fitzsimons Gavan J., Lynch John G.Jr, McClelland Gary H. (2013), "Spotlights, Floodlights, and The Magic Number Zero: Simple Effects Tests in Moderated Regression," Journal of Marketing Research, 50 (2), 277–88.
Tamir Diana I., Mitchell Jason P. (2012), "Disclosing Information About the Self Is Intrinsically Rewarding," Proceedings of the National Academy of Sciences, 109 (21), 8038–43.
Triplett Norman. (1898), "The Dynamogenic Factors in Pacemaking and Competition," American Journal of Psychology, 9 (4), 507–33.
United States Public Interest Research Group (2019), "Small Donors Driving 2020 Presidential Race," United States Public Interest Research Group (November 7), https://uspirg.org/reports/usp/small-donors-driving-2020-presidential-race.
White Katherine, Dahl Darren W. (2006), "To Be or Not to Be? The Influence of Dissociative Reference Groups on Consumer Preferences," Journal of Consumer Research, 16 (4), 404–14.
White Katherine, Dahl Darren W. (2007), "Are All Out-Groups Created Equal? Consumer Identity and Dissociative Influence," Journal of Consumer Research, 34 (4), 525–36.
White Katherine, Habib Rishad, Dahl Darren W. (2020), "A Review and Framework for Thinking About the Drivers of Prosocial Consumer Behavior," Journal of the Association for Consumer Research, 5 (1), 2–18.
Winterich Karen Page, Mittal Vikas, Aquino Karl. (2013), "When Does Recognition Increase Charitable Behavior? Toward a Moral Identity-Based Model," Journal of Marketing, 77 (3), 121–34.
Wohlfeill Gary. (2018), "Why Micro-Donations Are the Future of Fundraising," Philanthropy Journal (July 2), https://pj.news.chass.ncsu.edu/2018/07/02/why-micro-donations-are-the-future-of-fundraising/.
Xu Alison Jing, Wyer Robert S.Jr. (2008), "The Comparative Mind-Set: From Animal Comparisons to Increased Purchase Intentions," Psychological Science, 19 (9), 859–64.
~~~~~~~~
By Jacqueline R. Rifkin; Katherine M. Du and Jonah Berger
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 99- Platform Exploitation: When Service Agents Defect with Customers from Online Service Platforms. By: Zhou, Qiang; Allen, B.J.; Gretz, Richard T.; Houston, Mark B. Journal of Marketing. Mar2022, Vol. 86 Issue 2, p105-125. 21p. 8 Charts, 1 Graph. DOI: 10.1177/00222429211001311.
- Database:
- Business Source Complete
Platform Exploitation: When Service Agents Defect with Customers from Online Service Platforms
Online, pure-labor service platforms (e.g., Zeel, Amazon Home Services, Freelancer.com) represent a multibillion-dollar market. An increasing managerial concern in such markets is the opportunistic behavior of service agents who defect with customers off platform for future transactions. Using multiple methods across studies, the authors explain this platform exploitation phenomenon. In Study 1, they utilize a theories-in-use approach to clarify why and when platform exploitation occurs and derive some hypotheses. Study 2 empirically tests these hypotheses using data from a health care platform that connects nurses and patients. The results indicate that high-quality, long-tenured service agents may enhance platform usage, but customers also are more likely to defect with such agents. Platform exploitation also increases with greater customer–agent interaction frequency (i.e., building stronger relationships). This phenomenon decreases agents' platform usage due to capacity constraints caused by serving more customers off platform. These effects are stronger as service price increases (because higher prices equate to more fee savings), as service repetitiveness increases, and as the agent's on-platform customer pool comprises more repeat and more proximal customers. Finally, the authors use two scenario-based experiments to establish some managerial strategies to combat platform exploitation.
Keywords: customer defection; disintermediation; hazard model; opportunism; service platforms; sharing economy; theories in use
Online service platforms help customers and service agents connect and thereby facilitate transactions of physical assets (e.g., Airbnb), combinations of assets and labor (e.g., Uber), or pure labor (e.g., TaskRabbit) ([ 4]). Pure-labor platforms are growing in markets for home services (e.g., Handy), medical services (e.g., Heal), and specialized outsourcing (e.g., Freelancer.com) that grant customers access to "qualified" (e.g., background-checked) service agents and provide service agents with access to jobs. For both parties, platforms reduce search costs and transaction uncertainty and provide scheduling and payment services ([ 4]), usually in return for a commission on each transaction.
But a growing challenge for these service platforms, especially pure-labor platforms, is that a matched agent and customer may knowingly break platform rules and engage in subsequent transactions off the platform to avoid platform fees, a phenomenon we call "platform exploitation." Such exploitation reduces vital platform revenue and can even threaten the platform's survival ([23]; [46]). Anecdotal evidence of it abounds. [34] quotes an agent who admits, "I quit [using the platform] because … once people found me, they would just call me direct." The chief executive officer of Openbay has lamented, "A car owner who finds a great mechanic through Openbay may just call the mechanic directly the next time her car needs a repair, rather than book … through [us]," and Rover's chief executive officer similarly noted, "A dog owner who books a sitter through Rover.com may just call that sitter directly the next time" ([28]). In response to these behaviors, many platforms prohibit agents from transacting with customers off platform, suspend the accounts of agents found doing so, and explicitly require that the payment be made via the platform (see Table 1). Platforms often inform customers of the risks of transacting off platform and ask them to report agents who ask for direct payment. To incentivize customers to stay, platforms attempt to provide more value, with assurances (e.g., Amazon's Happiness Guarantee) or enhanced functionality (e.g., Rover's Dog Walking Map).
Graph
Table 1. Platform Rules to Prevent Exploitive Disintermediation.
| Platform Rule | Rule Summary from Terms and Conditions | Example Platforms |
|---|
| Exclusive subsequent bookings | All subsequent appointments between agents and customers must be made through the platform. | Zeel, Soothe |
| Exclusive communication channel | Agents must only communicate with users via the platform. | Freelancer |
| Prohibition of transacting outside the platform | Nurses are prohibited from transacting with platform patients in private. Penalties of breaching include a warning if suspected and permanent account suspension if confirmed. | Our nursing platform |
| Prohibition of exchanging contact information | Users are not allowed to share identifiable personal information with each other. | TaskRabbit, Tutor.com, Glamsquad |
| "Referral fee" charge | A very high "referral fee" will be charged for transactions outside the platform (e.g., $1,000 per instance). | Wag |
| Withholding payment | Payment to agents will be held if any agent activity threatens the platform. | Rover |
1 Notes: These rules reflect various platforms' terms and conditions. The consequences for violating them include suspended accounts, ceased cooperation, and withheld payments.
Despite the prevalence and importance of platform exploitation, as well as existing calls for research ([13]), the phenomenon has prompted few scholarly examinations ([21]; [22]; [46]). No research explicates its exploitative nature or tests strategies for combating it. We examine platform exploitation to shed light on why and when platform exploitation is likely to occur, clarify the relational and transaction dynamics surrounding it, and propose managerial interventions. We address platform exploitation in general but conduct empirical tests with pure-labor platforms, which are especially vulnerable because the services they involve typically require agents and customers to interact personally (e.g., in-home care), involve commission-based fees, allow relationships to form through communication and coordination, and rely on an agent's individual skill.
We use multiple methods to achieve these insights. In Study 1, we identify drivers of platform exploitation and uncover the theories-in-use held by platform participants on an in-home health care (patient–nurse) platform ([43]). Findings reveal that platform exploitation exists and is pervasive; they also indicate factors that lead customers and agents to defect. We combine these insights with existing theory to derive hypotheses tested in Study 2.
Study 2 uses the same setting to analyze 17,636 platform transactions among 12,523 unique patients (customers) and 2,009 nurses (agents). As customer–agent interaction frequency increases, platform exploitation increases. For platforms, high-quality or long-tenured agents who prompt platform usage can be a double-edged sword; they may also entice customers off the platform. These effects are exacerbated at higher prices and with more repetitive services. Among agents, interaction frequency, agent quality, and tenure also reduce agent return rates due to capacity constraints. That is, as agents take more customers off platform, they have less time for on-platform orders. These effects are magnified for highly priced services. The negative effects also are exacerbated by variables that reflect exploitation opportunities in the agent's pool of recent on-platform customers, such as when more of them are repeat customers or geographically close. In a robustness check, these core findings hold when we test them as predictors of the likelihood that a specific customer–agent dyad will return to the platform.
Finally, we use the insights from Studies 1 and 2 to inform scenario-based experiments in a different pure-labor context (dog walking platform) and test the efficacy of two interventions: a sliding-scale fee (financial mechanism) and service–agent community building (nonfinancial mechanism). Both interventions are effective in combating platform exploitation.
We extend prior literature by introducing platform exploitation, its driving factors, and a theoretical lens to understand it. These insights can generalize to service platforms that require close contact, communication, and skilled agents. We provide managerial implications related to the conditions in which platform exploitation is most likely and strategies for counteracting it.
Studies that examine situations in which buyers and sellers avoid intermediaries and transact directly tend to take a neutral stance toward such disintermediation, focusing more on overall market welfare ([ 5]; [31]). For example, [39] consider how bypassing traditional publishers in the book industry affects product variety and quality. Few studies take the intermediary's perspective.
Platform exploitation involves disintermediation, but it is unique due to its opportunistic nature. In using the term "exploitation," we explicitly aim to connote opportunistic misuses of a platform. Traditional disintermediation involves legal choices by customers, producers, or channel members to bypass an intermediary, without violating existing agreements, but with platform exploitation, users knowingly break their agreement with the platform, consistent with Wathne and Heide's (2000) idea of active opportunism (see common platform rules in Table 1). For example, the user agreements for Zeel and Freelancer specify that all customer–agent communications and future transactions must occur on the platform. The agreements also spell out consequences for violating these rules, including account suspension and withheld payments. Despite agreeing to these rules, agents exploit the platform to match with good customers and then take those customers off platform. Acquisition costs mean that platforms must rely on repeat customers and agents for profitability. Platform exploitation thus keeps current and future transaction-based revenue from the platform, hurts customer retention, and reduces agents' availability for on-platform customers, threatening the platform's profits and its very viability.
Although managerial thought pieces acknowledge that disintermediation may threaten the survival of service-based platforms ([22]; [46]), they do not investigate the antecedents of disintermediation, discuss the problems it causes, or examine the practice in a platform context. Notably, [21] examine users of a freelance project platform, who begin a discrete project and then complete that project either on or off the platform. Using a randomized control trial study, with treatment and control conditions that are not fully unique, they suggest a role for trust. When a freelancer has higher customer satisfaction scores, the likelihood that customers and freelancers finish the transaction off platform increases. Although these scholars do not simultaneously test all the statistical interactions that are key to their conclusions, their findings align with and help support our view.
Overall, the issues surrounding platform exploitation remain poorly understood. First, no study explicates the opportunistic nature of the behavior, the contextual conditions that alter its likelihood, or its long-term consequences for firms. Second, platforms differ widely in their design and service type, the presence and style of reputation systems, and the information shared prior to a transaction. For example, Amazon Home Services does not offer reputation scores or allow for customer–agent communication prior to transactions. Thus, we consider which types of platforms are more vulnerable to exploitation and how it manifests in practice. Third, little is known about interventions platform managers might take. In their conclusion, [21] propose a few strategies but do not test them empirically, a critical gap that we address.
Although service platforms differ in the resources they monetize, most research focuses on asset-based platforms that involve transactions of underutilized physical assets (e.g., [38]). However, pure-labor platforms have unique aspects ([ 4]) that make them susceptible to exploitation. The quality of the customer experience relies solely on the agent, because there is no physical asset quality (e.g., comfortable, clean car) to compensate for concerns about the agent (e.g., erratic Uber driver) ([13]). In turn, customers likely perceive the agent (not the platform) as the source of value and may develop interpersonal trust with agents ([21]). Services also require substantial communication and coordination between parties to clarify work specifications, which enhances the social aspects of these transactions. Agents and customers experience personal vulnerability that requires them to exhibit trust, especially as pure-labor services often occur in private settings (e.g., in-home). Thus, these aspects of pure-labor services can promote interpersonal loyalty.
Such enhanced trust and loyalty decrease the need for the platform. Initially, the platform makes matches and reduces search costs and uncertainty (e.g., by providing guarantees), serving as a formal institution that establishes rules to overcome transaction obstacles, as predicted by transaction cost economics ([41]). However, as trust between parties develops, the trust itself can serve as an informal institution ([16]) that effectively overcomes transaction hurdles ([21]). With trust, the need for the formal protections of the platform declines. For the agent, once matched with a desirable customer, the value of the platform for further transactions with that customer diminishes greatly. Payments and scheduling are easy to handle in other ways. Furthermore, agency theory predicts that if the platform fee (∼15%–30% of price) exceeds the value of added benefits ([37]), agents are likely to act opportunistically at the expense of the platform's interests. Yet the agent is the face of the platform to customers, so agents have both motivation and opportunity to pursue their own self-interest. The platform risks becoming a mere customer prospecting tool.
Figure 1 summarizes these incentives for agents and customers to defect from the platform and links those motivations to our research questions. Even if we can argue theoretically why platform exploitation is likely, we know little about how and when it manifests in practice. Therefore, with Study 1, we directly solicit insights from practitioners regarding ( 1) the degree to which platform exploitation occurs, ( 2) how and when customer–agent relationships form, and ( 3) which customer and agent characteristics increase the likelihood of platform exploitation.
Graph: Figure 1. Theoretical underpinnings of platform exploitation and research questions.
The context of Study 1 is a platform in China that connects patients and nurses for in-home health care services (e.g., injections, infusions). In China, medical services are confined to a few big hospitals in cities, increasing patients' transaction costs (e.g., long waits). Launched in 2015, the platform is one of the largest platforms for in-home nursing services in China, according to the Baidu Mobile Assistant app market. Patients can register for free; the platform reviews patients' medical documents to verify an order and then makes it available to nurses on the platform. A nurse who accepts an order contacts the patient and performs the service in the patient's home. The platform sets the price for each service (e.g., dressing change = ¥139, or ∼$21) and receives a 30% commission on all orders. It has explicit policies that nurses who perform off-platform transactions will be excluded from future platform use. Nurses usually work full-time at medical institutions and use the platform to earn extra money. To register on the platform, nurses must be certified, have three years of experience, and submit to a background check.
According to the summary of business models and designs of various pure-labor platforms in Table 2, the focal platform is not unique in its rules and operations. For example, many platforms similarly set prices, take percentage-based commissions, and rely on platform-matched agents to serve customers. We note some variations in whether agent information is revealed to customers before a transaction, but generally, platforms that set pricing tend not to reveal agent information before scheduling the transaction (e.g., Wag, Zeel).
Graph
Table 2. Business Models of Pure-Labor Service Platforms.
| Service Type | Platform Company | Revenue Stream (% of Price Retained by Platform) | Who Sets Service Price? | Who Matches Agent When Fulfilling Order? | Agent Info. Visible to Customer Before Order? | Payment by Platform Required? | Valuation (in Millions of U.S. Dollars) |
|---|
| Home services | Amazon Home Services | Transaction-based revenue (15%–20%) | Platform | Platform | No | Yes | N.A. |
| Handy | Transaction-based revenue (around 20%) | Platform | Platform | No | Yes | 500 |
| TaskRabbit | Transaction-based revenue (15%) | Agent | Customer | Yes | Yes | 125 |
| Thumbtack | Quote fee from agent | Agent | Customer | Yes | No | 1,700 |
| Work outsourcing | Freelancer.com | Transaction-based revenue (13%–20%) | Agent | Customer | Yes | Yes | 290 |
| Fiverr | Transaction-based revenue (around 20%) | Agent | Customer | Yes | Yes | 525 |
| Dog care | Rover | Transaction-based revenue (15%–25%) | Agent | Customer | Yes | Yes | 970 |
| Wag | Transaction-based revenue (around 40%) | Platform | Platform | No | Yes | 650 |
| Tutoring | Wyzant | Transaction-based revenue (20%–25%) | Platform/agent | Platform/customer | Yes | Yes | 85 |
| Tutor.com | Transaction-based revenue (N.A.) | Platform | Platform/customer | Yes | Yes | 66 |
| Massage | Soothe | Transaction-based revenue (around 30%) | Platform | Platform | No | Yes | 272 |
| Zeel | Transaction-based revenue (N.A.) | Platform | Platform | No | Yes | 129 |
| Medical care | Doctor on Demand | Transaction-based revenue (around 25%) | Platform | Platform/customer | Yes | Yes | N.A. |
| Heal | Transaction-based revenue (N.A.) | Platform | Platform | No | Yes | 228 |
| Our nursing platform | Transaction-based revenue (30%) | Platform | Platform | No | Yes | N.A. |
| Beauty care | Glamsquad | Transaction-based revenue (around 40%) | Platform | Platform | No | Yes | 123 |
| Stylebee | Transaction-based revenue (20%–33%) | Platform | Platform | No | Yes | N.A. |
| Psychotherapy | Talkspace | Subscription by customer | Platform | Platform/customer | Yes | Yes | 210 |
| Babysitting | Urbansitter | Subscription by agent and customer | Agent | Customer | Yes | No | 110 |
| Caregiving | Care.com | Subscription by agent and customer | Agent | Customer | Yes | No | 867 |
2 Notes: N.A. = not applicable. Most information is based on the platforms' "Terms & Conditions"; valuation data are from the PrivCo database (in millions of dollars).
Given this nascent area, we begin with a theories-in-use approach ([43]). We interviewed nurses, patients, and platform managers from cities served by the platform. Because our goal was discovery, the sample included participants with varying demographics and experiences with the platform. Following McCracken's (1988) guidance to continue interviews until no further significant insights arise, we interviewed 35 participants: 15 nurses, 15 patients, and 5 platform managers (each was paid $15). The sample details are in Web Appendix Tables W1, W2, and W3. Interviews were conducted in Chinese and ranged from 10 to 30 minutes. As recommended by [43], we began by asking generally how the participant feels about using the platform, probing for pros and cons. After building rapport, we asked nurses (patients) whether they had ever worked with a patient (nurse) initially through the platform and then moved future transactions off platform or if they had heard of others doing so. If they answered affirmatively, we followed up to uncover reasons why they (or others) took the transactions private and when patient–nurse relationships develop. We used a similar protocol for managers but also asked about how pervasive off-platform transactions were, whether they affect the platform, and the actions the platform takes in response. Given the topic's sensitive nature, we assured participants of confidentiality. Interviews were recorded and transcribed, and we used the transcripts to search for emerging themes. The results inform our hypothesis development. Table W4 in the Web Appendix contains summary data.
Platform exploitation is pervasive: 12 of 15 nurses (80%) and 10 of 15 patients (67%) had personally transacted off platform with partners they met on the platform (and nearly all had heard of others doing so). This pervasiveness exists despite explicit policies, clearly communicated to nurses and patients, that prohibit off-platform transactions. The platform managers noted that nurses would be "permanently blocked from taking orders on our platform" and that customers are told "do not make a private deal with nurses." Patients are aware, acknowledging that "[they] don't want us to transact directly." As a nurse explained,
Of course, the platform doesn't want these cases [transactions in private] to happen. They are trying to prevent it. But this is unavoidable. The nurses are not stupid. They know if they build relationships with the customers, they control the resources.
Managers view platform exploitation as a major problem that reduces platform revenue and profit, such that "every year we are missing three to four times the profit we are making right now" because the platform constantly recruits new nurses and patients to replace those who leave. Recruiting and on-boarding (e.g., background checks) is expensive. For example, the platform's primary recruiting method for new nurses is to pay a bonus for referrals from current nurses. Managers estimate the combined cost of acquiring a new customer and a new nurse at ¥110–¥150. For a typical order—say, a ¥139 injection—the platform earns 30%, or ¥42. Using injections as an example, for the platform to break even on its acquisition costs, a new nurse and patient must complete three to four injections on the platform. Thus, retention and repeat ordering are critical to the platform's profitability.
Platform exploitation also creates consequences that are not themselves direct violations of platform policies but are potentially harmful to the platform. First, some potential customers never try the platform. Nurses noted that when they have a good relationship with a patient, the patient often introduces relatives and friends to them directly, bypassing the platform (corroborated by patient interviews). Second, high-quality nurses may use the platform less. These nurses reduce their platform usage when they have "a stable [off-platform] customer base," because they have "no time to take more orders [on the platform]." This trend affects the quality of the platform's portfolio of nurses and increases recruiting costs to maintain enough high-quality nurses. Third, platform exploitation might alter the quality of the platform's patients. Nurses have incentives to take "good-quality customers" off platform, skimming the best patients and leaving riskier ones ([ 1]). As explained by one nurse,
As long as the nurse starts taking orders in private, she'll keep going, not [using] the platform. Unless she feels the risk is particularly high, such as the patient or his family members are particularly mean and the patient's condition is serious … once she has assessed the patient and found that the family is okay and the patient is okay, she will always consider accepting orders in private.
Various relational dynamics and motivations underlie defections from the platform. First, nurses and patients frame affiliations with one another as relational (e.g., nurse: "It's not like a relationship of serving and being served. It's more like friends"; patient: "We are like friends"). Service platforms may promote such relationship building because the person-to-person nature of service provision requires extensive communication and coordination. A patient described the process as "making friends at the beginning." Relationships grow through multiple interactions across service visits between the same nurse and patient.
Second, once built, relationship trust reduces information asymmetry and uncertainties in future transactions (e.g., "If we trust each other … it's better to contact directly. I feel secure if we know each other"). Using the same partner increases communication efficiency, saving time:
When I do infusion[s] … I need to know the specific condition of the patients before I go. But the platform will not offer me very detailed information. If I [work with] patients directly, it's much more convenient since [I] know what they need … specifically.
It is "inconvenien[t] for patients to change nurses every time" because they have to repeatedly share information. Through private dealings, patients have on-demand access to service, easily reached by "the phone at [their] disposal." By "contact[ing] this nurse directly, I can just tell them what I need, there's no need to use a third party."
Operational and participant characteristics within a transaction also exert effects. First, platform exploitation is heavily driven by the potential for monetary savings. Taking orders privately enables "patients [to] pay less, and nurses [to] earn more." As detailed by one nurse, a particular service "was 130 yuan per time. The patient saves 38 yuan, I earn about 20 to 30 yuan more. It was a long-term and multitime [procedure]." Higher-priced services are more likely to be conducted off platform, because the savings earned from avoiding the fixed-commission fee are higher. Nurses similarly prefer to take higher-value patients off platform, to avoid competing for them, noting that "it's not the case you can grab [a customer's] order on the platform every time."
Second, in terms of agent characteristics, patients seek to defect with high-quality nurses, defined by their "good operation skill" and "good serving attitude," which are important determinants of trust. Nurses who have used the platform longer also appear more likely to defect, because they "know more about the processes" and how the platform truly works.
Third, opportunities for platform exploitation depend on the characteristics of the customers the nurse currently serves. For example, agents who work with many customers who are proximal have a customer pool ripe for exploitation; as a patient noted, nurses "can decide whether they want to work with me based on whether their location is close to my place."
Although platform exploitation is pervasive, nurses and patients also highlight some benefits of remaining on the platform. The platform offers a large pool of nurses, increasing patients' constant access to service. It can help resolve "customer disputes" and "offer protection for both patients and nurses" in every transaction. Thus, not everyone seeks to leave. Platform exploitation is a strategic decision, based on the situation and any patient–nurse relationship.
We integrate these interview findings with existing theory to derive a set of testable hypotheses about when platform exploitation is most likely. Figure 2 illustrates our conceptual model. Table W5 in the Web Appendix summarizes the representative quotes and theory for each hypothesis.
Graph: Figure 2. Conceptual framework: drivers of platform exploitation.
The interview data suggest that platform exploitation manifests differently for each party. Customers' primary motivation is to find a good agent with whom to defect, and thereafter, they are unlikely to return to the platform. Thus, we predict factors that might decrease customer retention on the platform. However, the agents engaged in platform exploitation are unlikely to leave completely because the platform provides ongoing opportunities for customer prospecting. Still, agents do not have unlimited service capacity ([ 9]); when an agent takes more customers private (i.e., builds a base of off-platform customers), an increasing amount of that agent's capacity is used, so the agent likely is slower to return to the platform. Thus, we predict factors that decrease agents' rate of returning to the platform.
We begin with three main-effect variables, prominent in the interviews: number of prior dyadic transactions, agent quality, and agent tenure.
Interviewees noted that platform exploitation becomes more likely as trust within a customer–agent dyad builds over repeat interactions (e.g., nurse: "It is hard to establish a good relationship during the first meeting"; patient: "It's like making friends … after [a few] times"). Service transactions are prone to relationship building because they involve high levels of interpersonal communication ([29]), and repeated transactions provide evidence that reinforces trust ([24]). Greater customer–agent trust also reduces uncertainties, so the platform becomes less necessary for future transactions. Thus, repeated, dyadic transactions between an agent and customer should increase platform exploitation, in the form of reduced customer retention and agent return rates.
- H1: As the number of prior platform dyadic transactions increases (number of times an agent transacts with the same customer on the platform), the (a) probability of customer retention and (b) agent return rate to the platform decrease.
The interviewees noted that patients are most likely to defect with high-quality nurses, stressing qualities such as "good operation skill" as keys to growing trust, so that patients "want to contact you directly for services." This finding is congruent with Gu and Zhu's (2020) arguments, but the trust we investigate arises more from social interaction and direct customer experiences of quality, not an agent's public rating. High-quality agents engender trust, reducing transaction uncertainty and making platform exploitation more likely.
Thus, platforms face a paradox. Agent quality drives platform growth ([11]), and platforms can evoke more customer responses by featuring agents in advertisements ([12]). Yet high agent quality may decrease customer retention. The paradox implies a nonlinear, inverted U-shaped relationship between agent quality and retention: bad experiences with low-quality agents drive customers away, and initial increases in quality should increase retention, but after a certain level of agent quality, customers grow more likely to defect with these superior agents. We anticipate similar effects on the agent return rate, such that high-quality agents are more desirable to customers and thus more capable of growing a private customer base, reducing their return rate to the platform.
- H2: Inverted U-shaped relationships exist for agent quality relative to both (a) the probability of customer retention and (b) agent return rate, such that greater agent quality increases these outcomes up to a certain point but then reduces them.
The length of time an agent has served on the platform (i.e., platform tenure) should increase the likelihood of platform exploitation because experienced agents, as stated by a nurse interviewed for Study 1, "know more about the processes, and what services they can offer to the customers and what the customer can get." Their insights into the platform's processes and institutional knowledge of how to use the system enable them to attain personal gains, as well as maximize value for customers. When agents first join the platform, they rely on it to acquire customers, and they may be reluctant to take customers private. Over time, as they meet more customers, identify desirable ones, and learn that the likelihood of exploitation activity being discovered is low, they feel less dependent on the platform and more confident taking customers private. This logic also applies to the agent return rate. As longer-tenured agents gain more off-platform customers, more of their capacity gets taken up, and they return to the platform more slowly. Formally:
- H3: As agent platform tenure increases, the (a) probability of customer retention and (b) agent return rate decrease.
We next examine several moderators to these main effects that follow from our interviews. We consider four moderators: one relevant to both parties (i.e., transaction price), one for customers (i.e., service repetitiveness), and two for agents (i.e., on-platform customers that are repeat and proximal).
Transaction price enhances platform exploitation motivations for both parties. Patients and nurses both identified avoiding fees as a major benefit of skirting the platform to work with a trusted partner. When fees are a fixed percentage of price, potential savings increase with price. As [23] notes for the cleaning service platform Homejoy, "The higher the commission, the higher the incentive for cleaners and customers to strike a better deal directly." However, even though price is a strong motivating factor, we do not think it is sufficient to drive platform exploitation; customers and agents leave the platform only after finding a good match. Thus, higher prices increase the financial incentives and motivations of both parties to move the relationship off platform given a trusting relationship. Thus, we propose price as a moderator of all three main-effect relationships involving customer retention and agent return rates.
- H4: The effect of prior dyadic transactions on (a) customer retention and (b) agent return rate intensifies as price increases.
- H5: The effect of agent quality on (a) customer retention and (b) agent return rate intensifies as price increases.[ 6]
- H6: The effect of agent tenure on (a) customer retention and (b) agent return rate intensifies as price increases.
We define service repetitiveness as the frequency with which customers require a particular service. The Study 1 interviews indicate that moving future transactions with a given agent off platform reduces the hassle of repeating the order process, offers greater service delivery flexibility, and reduces coordination effort. Thus, we expect that greater service repetitiveness increases customer motivation to defect with a good agent and thus will strengthen the negative main effects on customer retention.
- H7: The effects of (a) prior dyadic transactions, (b) agent quality, and (c) agent tenure on customer retention intensify with greater service repetitiveness.
The last two moderators provide insights into when agents might move their platform customers off platform. As agents acquire a larger on-platform pool of customers, they become less dependent on the platform for customer acquisition, and this asymmetric dependence makes opportunistic behavior more likely ([32]). Certain characteristics of the agent's on-platform pool in turn might indicate increased exposure to desirable customers and thus increase exploitation opportunities.
The interviews suggest that nurses prefer to work with patients who are geographically close, for efficiency reasons: "Nurses want to have a stable and regular customer base. If we can meet patients in our area through the app, then we [know] each other." Patients recognize this desire for geographic proximity, noting that nurses "can decide whether they want to work with me based on whether their location is close to my place." The percentage of on-platform customers who live in close geographic proximity to a particular agent represents the potential pool of desirable customers for this agent to take off platform. However, customers might be willing to leave with these motivated agents only if they represent a good match, so we propose it as a moderator.
- H8: The effects of (a) prior dyadic transactions, (b) agent quality, and (c) agent tenure on agent return rate intensify as the percentage of an agent's current pool of proximal on-platform customers grows.
Finally, assuming good customer–agent matches have been established on the platform, the percentage of repeat on-platform customers within the agent's portfolio represents targets who are particularly susceptible to be taken off platform. Agents who build relationships with repeat customers become less dependent on the platform and more able to act opportunistically, so that "only after I've been on this platform for a while, I started to offer services to the patients directly," whereas "right at the beginning … you don't have your customer base … not many people know you." Agents with a higher percentage of repeat customers have a pool of good prospects for off-platform transactions because their prior interactions have helped establish trust ([29]). Trusted agents have more motivation and opportunity to move customers off platform.
- H9: The effects of (a) prior dyadic transactions, (b) agent quality, and (c) agent tenure on agent return rate intensify as the percentage of an agent's current pool of repeat on-platform customers grows.
We analyze transaction data from the platform from Study 1 to test our hypotheses. We observe all transactions on the platform between July 1, 2017 and July 4, 2018, which encompasses 17,636 transactions by 2,009 unique nurses and 12,523 unique patients. In this set, 22.3% of patients place multiple orders (41.9% of observations). There are 15,002 unique patient–nurse dyads, and 1 in every 3.43 transactions involves dyads observed multiple times.
We jointly model customer retention and agent return rates, which provides stronger evidence of platform exploitation while also enabling us to address the specific hypotheses for each. If a customer defects after working with a desirable agent but the agent's platform usage remains the same, then platform exploitation likely is not driving defection (e.g., patients might recover faster due to high-quality care). However, if both parties decrease platform usage, the pattern indicates platform exploitation. As a robustness check, we also estimate the probability that each dyad returns, which reflects the joint decision by a customer and an agent to continue their relationship through the platform.
There are four unique strengths of our data set. First, customers are financially motivated, because they pay for the service themselves (i.e., no insurance payments). Second, in-home service is an attractive option for customers who are unable to travel and who find the service quality offered by large hospitals in China insufficient ([15]). Third, no significant new competition emerged or public policy changed ([45]) during the study period. Fourth, the rate at which a customer requires service is dictated by the type of service they need, not the agent. The platform mostly deals in routine services needed at regular intervals.
For each customer, we observe the number of days until the next order by the same customer, . Variables are right-censored because we do not observe additional orders after the sample ends. For right-censored observations, equals days from the last order to the end of the sample period. An indicator variable marks censored observations, such that = 1 if the customer places another order, and 0 otherwise (censored).
For each agent, we observe the number of days between orders they filled, . We let = 1 if we observe the agent fill another order, and 0 otherwise. Accordingly, is the days from the agent's last order to the end of the sample time frame when = 0.
We observe the number of days between each dyad's appearance on the platform, . We let = 1 if we observe the dyad again, and 0 otherwise, and here, is the days from the last order to the end of the sample when = 0.
We operationalize prior transactions as the number of times, prior to the current order, that the customer received service from the agent who fills a current order ( ).
The platform provides a continuous measure of quality ( ) for each nurse. The measure is proprietary to the platform and not revealed to customers, which lessens endogeneity concerns because customers cannot rely on this quality metric when making decisions about future orders. According to the platform managers, factors that contribute to include the nurse's job title, customer ratings (not visible to customers or agents), years of experience, and number of platform orders filled. This measure captures quality differences because it considers both experience independent of the platform and performance on the platform. It also is managerially relevant and used to evaluate agents. We normalize from 0 (worst) to 100 (best).
We use the number of days since the agent joined the platform to operationalize agent tenure ( ).
We obtain each order's total price ( ). Prices for each service do not change over the observed time frame, which eliminates endogeneity issues. However, some customer-specific variation in service prices might occur, due to surcharges imposed if agents must travel more than 5 km or provide medical supplies (e.g., bandages). These actual costs do not enable the agent to make extra money. We separately control for medical supply surcharges ( ), which might signal particularly complex or risky services. By controlling for distance and equipment cost, thus captures the unchanging platform charge and associated fees that could be saved by circumventing the platform.
We measure the degree to which a given service is required repeatedly by customers. A service is "new" for customers the first time they receive it and "repeat" all other times. We operationalize service repetitiveness ( ) as the percentage of times a service is repeated on other customers by other agents in a given month. We calculate the measure after excluding data from the customer and agent of the focal observation so that it is not influenced by their characteristics. The monthly time frame addresses seasonality in service provision (e.g., infusions are more common during flu season).
The current pool of proximal on-platform customers ( ) equals the fraction of customers served by the agent in the last 30 days who are within 5 km of the agent's location. Five kilometers is what the platform considers distant when imposing a travel surcharge; 30 days captures recent opportunities.
Similarly, the agent's current pool of repeat on-platform customers ( ) is the fraction of repeat customers (i.e., repeat divided by total customers) served by the agent in the past 30 days.
Table 3 lists the summary statistics (correlations are in Web Appendix Table W6) for our controls. We include several variables managers use when computing agent quality to address potential nonquality confounds. For example, an agent's years of nursing experience likely relates to their salary at their full-time job, which then may correlate with their desire to use the platform to earn extra money. Controlling for these observables ensures that reflects the inputs to platform's quality measure that more strongly relate to overall excellence of the nurse (e.g., customer ratings and job titles). We also include service dummies (24 services), city dummies (188 cities), and time dummies by month.[ 7] The time dummies capture platform-wide shocks in a period.
Graph
Table 3. Variables Names, Definitions, and Descriptive Statistics.
| Variable Name | Definition | M | SD | Min | Max |
|---|
| Dependent Variables | | | | |
| Number of days between orders for the customera | 131.884 | 120.603 | .001 | 368.701 |
| Dummy = 1 if the customer places another order after current order | .290 | .454 | .000 | 1.000 |
| Number of days between orders filled by the same agenta | 29.728 | 71.416 | .001 | 368.560 |
| Dummy = 1 if the agent fills an additional order after current order | .886 | .318 | .000 | 1.000 |
| Number of days between times the same dyad is on the platforma | 154.609 | 118.919 | .001 | 368.701 |
| Dummy = 1 if the dyad returns to the platform after current order | .149 | .356 | .000 | 1.000 |
| Key Independent Variables and Moderators | | | | |
| AgentQuality | Continuous agent quality measure given by the platform | 11.730 | 16.705 | .000 | 100.000 |
| AgentTenure | Number of days since the agent was first observed on the platform | 406.043 | 196.269 | .000 | 966.000 |
| PriorDyadVisits | Number of times the customer is observed receiving service from the same agent that fills the current order prior to the current order | .460 | 2.079 | .000 | 43.000 |
| Price | Total order price (¥) including distance and medical supply surcharges | 308.753 | 381.764 | 70.000 | 4434.000 |
| %ServiceRepeat | Percentage of time the service is performed as repeat on other customers by other agents in a given month | .431 | .204 | .000 | 1.000 |
| %CloseCustomers | Fraction of customers served by the agent in the last 30 days within 5 km | .519 | .340 | .000 | 1.000 |
| %RepeatCustomers | Fraction of repeat customers served by the agent in the last 30 days | .052 | .134 | .000 | 1.000 |
| Additional Control Variables | | | | |
| Customer characteristics | | | | | |
| CustomerSameService | Number of times prior to the order the customer orders the same service | 1.097 | 5.129 | .000 | 102.000 |
| CustomerPriorOrders | Number of orders by the customer prior to the order | 1.591 | 5.788 | .000 | 102.000 |
| TimeOnApp | Number of days since customer was first observed on the platform | 75.332 | 147.578 | .000 | 889.000 |
| Agent characteristics | | | | | |
| AgentSameService | Number of times prior to the order the agent performs the same service | 8.556 | 16.367 | .000 | 121.000 |
| AgentOrderNum | Number of times the agent fills an order, including current order | 37.700 | 56.068 | 1.000 | 332.000 |
| AgentHomeVisits | Total number of times agent has performed a home visit on the platform | 324.585 | 465.472 | .000 | 2624.000 |
| AgentProExp | Years of nursing experience for the agent | 3.218 | 1.793 | 3.000 | 28.000 |
| PotentiallyStolen Customers | Number of customers served by the agent prior to the order that have not been observed returning to the platform | 23.115 | 35.205 | .000 | 238.000 |
| Order characteristics | | | | | |
| EquipmentCosts | Medical equipment and distance surcharges for the order | 15.078 | 33.963 | .000 | 750.000 |
| MultiVisit | Dummy = 1 if the order included multiple in-home visits | .167 | .373 | .000 | 1.000 |
| Distance | The distance in meters between the customer and agent in the order | 7,056.917 | 9,248.474 | .000 | 98,218.00 |
| How order was filled | | | | | |
| Request | Dummy = 1 if the patient requests a nurse and the nurse accepts | .185 | .389 | .000 | 1.000 |
| Scramble | Dummy = 1 if the order is available as first-come, first-serve | .762 | .426 | .000 | 1.000 |
| RequestScramble | Dummy = 1 if the patient requests a nurse who is unavailable after which the order is made available as first-come, first-serve | .052 | .223 | .000 | 1.000 |
- 3 aNumber of days from the last observation for the customer, agent, or dyad to the end of the sample period for censored observations.
- 4 Notes: N = 17,636.
We also use dummies to control for how customers and agents connect on the platform. The platform's process for connecting parties mitigates endogeneity concerns. Neither party can select partners on the basis of unobservables that might also affect customer retention or the agent return rate. Agents fill orders on a first-come, first-served basis (76.22%), after reviewing the service type, travel distance, and date/time requested. No other information is available until an agent accepts. It also is difficult and costly to cancel orders. Selection issues thus are addressed by observable factors. Consumers also can request an agent, who accepts (18.54%) or not (5.23%); following a rejection, the process reverts to first-come, first-served.[ 8] Only customers can make requests; agents are unable to pick customers. The dummy that indicates an agent accepts a request captures variance due to customer preference for an agent.
At the customer level, we model the probability that a customer returns after the most recent order, regardless of whether they use the same agent. Customer return to the platform thus represents the "event" in hazard-model terminology ([35]). We have multispell data, such that some customers experience the event many times ([ 3]).[ 9] At the agent level, we model how quickly agents return for more orders, regardless of whether they match with different customers. We use a continuous-time hazard approach for both customers and agents because we observe the exact order time, and there is no set order schedule ([14]).
Hazard models assume that all individuals experience the event, so we follow [30] and adapt a multispell hazard model with a cure fraction (i.e., percentage of individuals who will not return; [ 3]). We use the cure fraction to model the probability that a customer returns to the platform.[10] Specifically, for customer i, we observe each order j on the platform; is the time between j and j + 1. For the cure fraction, we define the latent variable = 1 if customer i returns to the platform after order j. The probability that = 1 is , where is a customer-specific effect, and is a vector of covariates that contain characteristics of customer i's jth order.[11]
The survival function, or the probability that a customer will not order again at least until time , is used to derive the hazard, or the likelihood that a customer orders at time , given that this customer has not done so yet. Given customer i's repeat order on the platform, this survival function is , where is the customer effect and are covariates. The survival function that includes the cure is
Graph
( 1)
The first two terms in Equation 1 capture the probability that customer i defects; the last term is the probability that customer i returns to the platform but has not placed another order by time . The hazard then is ([ 3]), or
Graph
( 2)
where is the probability density function associated with , which is the hazard conditional on customer i returning to the platform.
According to Equations 1 and 2, the log-likelihood for the N customers on the customer side of the problem with right-censoring ([25]) is
Graph
( 3)
There is not a cure with agents, in that platform exploitation means they slow their return rather than abandon the platform. For agent k, we observe each order m placed on the platform; the time between m and m + 1 is . The survival function and hazard for agent k returning to the platform is and , where is an agent-specific effect, and is a vector of covariates.[12] The log-likelihood function for all K agents is
Graph
( 4)
An Expo-power distribution for allows for a flexible hazard.[13] The probability density function and survival functions, given customer i's return, are ([33])
Graph
( 5)
and
Graph
( 6)
where > 0, > 0, and are the scale, shape, and location parameters. We use a proportional hazard specification in which ([33]) and is a vector of coefficients. We also assume the agent-side hazard follows an Expo-power distribution with associated parameters , , and , where is the coefficient vector.
We use logit to model the cure fraction for customers ([14]). It is akin to a panel logit, in that we include a customer-specific effect to account for unobserved heterogeneity that affects customer retention. Formally, with as the coefficient vector,
Graph
( 7)
Finally, we model customer-specific effects, and , and the agent-specific effect, , as random effects, where , and . Random effects address unobserved heterogeneity at the customer and agent levels in their respective equations; failure to do so in multispell hazard models can lead to biased estimates ([ 3]).[14] We allow customer and agent random effects to correlate with each other through and . A positive indicates that unobserved agent characteristics associated with longer (shorter) times between orders also are associated with a higher (lower) probability the customer defects.
For the cure fraction, we specify
Graph
( 8)
where consists of the dummy variables discussed previously and the additional controls in Table 3.[15] However, we do not include in the customer equation, because it is an agent side–specific control, as we discuss subsequently.
We include the same variables in the customer-side hazard, , as the customer cure fraction. It provides a stronger test of our hypotheses because the variables are not constrained to affecting only return probability; otherwise, results could be biased. For example, agent quality's true effect may be to decrease the hazard. If we failed to allow for this effect, it might instead manifest as a negative impact on retention and provide undue evidence of platform exploitation.
For the hazard on the agent side, we specify
Graph
( 9)
where includes variables in , , and its square. Here, is the number of customers served by the agent prior to the order who have yet to return to the platform. We include this variable on the agent side because these customers are potentially in the agent's off-platform customer pool. Linear and quadratic terms capture an agent's initial prospecting opportunity and capacity constraint.
We estimate the model by summing Equations 3 and 4, using the functional forms in Equations 5–9, and maximizing with respect to all coefficients, including random effect variances and correlations, using STATA's maximum likelihood estimation command.[16] Variables are centered so main effects are interpreted for the other variables at their means.
Table 4 contains information on orders, customers, and agents by month. Orders fluctuate monthly, with some growth toward the end; the monthly percentage change averages 2.6%. Yet each month, 923.1 new customers on average join the platform. Many new customers, coupled with tenuous growth in the number of orders, suggests that relatively few new customers place multiple orders. In addition, note that, on average, 58.42 new agents join each month. Together, new agents and low growth in orders implies that existing agents take fewer orders. While the percentage of existing agents who return to the platform averages 27.9% per month, it declines initially from a high of 56.9%, stabilizes, and then declines again toward the end to finish at 23.5%. Patterns for both customers and agents are consistent with platform exploitation. The platform-level dynamics also hurt profitability, in that acquisition costs are increasing over time, while existing customers and agents are using the platform less.
Graph
Table 4. Orders, Customers, and Agents by Month.
| Month/Year | 7/17 | 8/17 | 9/17 | 10/17 | 11/17 | 12/17 | 1/18 | 2/18 | 3/18 | 4/18 | 5/18 | 6/18a | Avg. |
|---|
| Orders | 1,689 | 1,343 | 1,300 | 1,314 | 1,402 | 1,420 | 1,535 | 952 | 1,476 | 1,535 | 1,981 | 1,689 | 1,470 |
| % Δ in Orders from Prior Month | | −20.5% | −3.2 | 1.1 | 6.7 | 1.3 | 8.1 | −38.0 | 55.0 | 4.0 | 29.1 | −14.7 | 2.6% |
| New Customers | 853 | 835 | 859 | 846 | 862 | 906 | 952 | 564 | 951 | 959 | 1,392 | 1,098 | 923.1 |
| % Δ in New Cust. from Prior Month | | −2.1% | 2.9 | −1.5 | 1.9 | 5.1 | 5.1 | −40.8 | 68.6 | .8 | 45.2 | −21.1 | 5.8% |
| % Orders by Existing Cust. | 42.5 | 26.7 | 23.2 | 25.2 | 24.5 | 25.6 | 25.3 | 29.2 | 24.6 | 26.6 | 21.6 | 24.2 | 26.6 |
| New Agents | 79 | 74 | 60 | 42 | 50 | 46 | 38 | 32 | 49 | 58 | 75 | 98 | 58.42 |
| % Existing Agents Using Platform | 56.9 | 32.6 | 26.4 | 27.0 | 25.1 | 25.2 | 25.5 | 18.3 | 24.9 | 25.1 | 23.9 | 23.5 | 27.9 |
- 5 aIncludes the first four days of 7/18.
- 6 Notes: New Agents = agents who fill orders for the first time in the month; Existing Agents = agents who filled orders before the month; New Customers = customers ordering for the first time in the month; Existing Customers = customers who placed orders before the month.
Table 5 contains results. We focus on Model 2a, which includes both main and moderator effects. The main effects of prior dyad visits ( = −.0867, p <.05) and agent tenure ( = −.0006, p <.01) provide strong support for H1a and H3a. Customers are likely to defect when they repeatedly receive service from the same agent or if they interact with longer-tenured agents.
Graph
Table 5. Probability of Customer Return and Hazard Estimation Results.
| Probability of Customer Return | Customer Return Hazard |
|---|
| Base 1a | Moderators 2a | H (sign) | Support | Base 1b | Moderators 2b |
|---|
| βs | SE | βs | SE | δs | SE | δs | SE |
|---|
| PriorDyadVisits | −.1227*** | .0335 | −.0867** | .0377 | H1a (−) | Yes | −.0073 | .0076 | −.0127 | .0102 |
| AgentQuality | −.0070* | .0040 | .0081 | .0058 | H2a (+) | | −.0021 | .0034 | .0017 | .0050 |
| AgentQuality2 | | | −.0003*** | .0001 | H2a (−) | Yes | | | −.0001 | .0001 |
| AgentTenure | −.0007*** | .0002 | −.0006*** | .0002 | H3a (−) | Yes | .0001 | .0001 | .0001 | .0002 |
| PriorDyadVisits × ln(Price) | | | −.0956** | .0426 | H4a (−) | Yes | | | −.0266 | .0205 |
| AgentQuality × ln(Price) | | | .0239*** | .0054 | H5a (+) | | | | −.0072 | .0044 |
| AgentQuality2 × ln(Price) | | | −.0004*** | .0001 | H5a (−) | Yes | | | .0002** | .0001 |
| AgentTenure × ln(Price) | | | −.0007** | .0003 | H6a (−) | Yes | | | .0001 | .0002 |
| PriorDyadVisits × %ServiceRepeat | | | −.5294*** | .1543 | H7a (−) | Yes | | | −.0493 | .0441 |
| AgentQuality × %ServiceRepeat | | | .0333* | .0193 | H7b (+) | | | | .0236 | .0155 |
| AgentQuality2 × %ServiceRepeat | | | −.0006* | .0004 | H7b (−) | Yes | | | −.0006** | .0003 |
| AgentTenure × %ServiceRepeat | | | −.0014* | .0008 | H7c (−) | Yes | | | −.0002 | .0007 |
| ln(Price) | −.5815*** | .0854 | −.4361*** | .0933 | | | −.4835*** | .0982 | −.4761*** | .0960 |
| %ServiceRepeat | −.1392 | .4612 | .0101 | .4762 | | | .7949** | .3607 | .7768** | .3708 |
| %CloseCustomers | −.1533 | .0943 | −.1643* | .0944 | | | −.0706 | .0779 | −.0808 | .0780 |
| %RepeatCustomers | .3351 | .2516 | .3119 | .2486 | | | −.1442 | .1466 | −.1485 | .1435 |
| CustomerSameService | −.1870*** | .0449 | −.1819*** | .0450 | | | .0048 | .0089 | .0061 | .0088 |
| CustomerPriorOrders | .3041*** | .0351 | .2992*** | .0336 | | | −.0042 | .0084 | −.0046 | .0082 |
| TimeOnApp | .0007*** | .0002 | .0007*** | .0002 | | | −.0007*** | .0003 | −.0007*** | .0002 |
| AgentSameService | −.0018 | .0031 | −.0045 | .0032 | | | .0002 | .0025 | .0003 | .0026 |
| AgentOrderNum | .0005 | .0011 | −.0002 | .0012 | | | .0015* | .0009 | .0010 | .0009 |
| AgentHomeVisits | .0002 | .0002 | .0002 | .0002 | | | −.0001 | .0001 | −.0001 | .0001 |
| AgentProExp | −.0059 | .0142 | −.0069 | .0143 | | | .0092 | .0103 | .0091 | .0102 |
| ln(1 + EquipmentCosts) | .1560*** | .0213 | .1544*** | .0214 | | | .1020*** | .0199 | .0956*** | .0196 |
| MultiVisit | .2345** | .1187 | .1177 | .1212 | | | .0343 | .1265 | .0264 | .1208 |
| ln(1+Distance) | −.0096 | .0150 | −.0165 | .0151 | | | −.0222* | .0115 | −.0232** | .0115 |
| Scramblea | −.2691*** | .0881 | −.2884*** | .0886 | | | −.0215 | .0583 | −.0447 | .0596 |
| RequestScramblea | −.1309 | .1302 | −.1355 | .1308 | | | −.0390 | .0937 | −.0605 | .0939 |
| Constant | –1.4412*** | .5062 | –2.0310*** | .4748 | | | –2.0310*** | .4748 | –1.9295*** | .4757 |
| Service, City, and Time Fixed Effects | ✓ | | ✓ | | | | ✓ | | ✓ | |
| | | | | | | | .7272*** | .0136 | .7268*** | .0137 |
| | | | | | | | −.0028 | .0046 | −.0002 | .0046 |
| Random Effects Parameters | | | | | | | | | | |
| | 1.6065*** | .1990 | 1.6875*** | .2013 | | | | | | |
| | | | | | | | .7682*** | .0839 | .7891*** | .0816 |
| | .1731* | .0956 | .1031 | .0802 | | | .1731* | .0956 | .1031 | .0802 |
| | .7064*** | .0499 | .4552*** | .0462 | | | | | | |
| | | | | | | | .1689** | .0750 | .0469 | .0619 |
| Log-likelihood | –74,564.9 | –74,373.8 | | | –74,564.9 | –74,373.8 |
| Wald (p-value) | 621.68 (.000) | 710.12 (.000) | | | 621.68 (.000) | 710.12 (.000) |
| N | 17,636 | 17,636 | | | 17,636 | 17,636 |
- 7 *Significant at the 10% level.
- 8 **Significant at the 5% level.
- 9 ***Significant at the 1% level.
- 10 aRequest is the base category.
We find evidence of an inverted U-shaped effect of agent quality on retention, which supports H2a. High agent quality reduces retention because customers are more likely to go off platform with superior agents. However, low agent quality is likely associated with poor service, which also reduces retention. The coefficient on is negative ( = −.0003, p <.01), and the linear term is insignificant, so the effect of agent quality is close to a maximum at variable averages: an average-quality agent is best for retention for average-priced services with average repetitiveness. Deviations from average quality reduce retention. In summary, the main effects of the key independent variables are consistent with platform exploitation and support their hypotheses.
Turning to moderators, we expect the impact of prior dyad visits, agent quality, and agent tenure to be more pronounced at higher prices. After finding a good match, both parties have a stronger motivation to move the relationship off platform when the financial incentive is larger. Note that all interactions with are significant. The impact of prior dyad visits ( = −.0956, p <.05), the inverted U-shaped effect of agent quality ( =.0239, p <.01; = −.0004, p <.01), and the effect of agent tenure ( = −.0007, p <.05) all are amplified at higher prices, in support of H4a, H5a, and H6a. We illustrate the interaction of at different values of in Figure 3, revealing the strongest inverted U-shaped effect of agent quality on retention at high prices, whereas it does not hold at the lowest price. The results for low agent quality imply that customer dissatisfaction from working with an inferior agent is amplified at higher prices; for high agent quality, the financial incentive of higher prices increases the likelihood of going off platform. Figures W1A and W1B in the Web Appendix depict the predicted probability of customer return by and at different values of .
Graph: Figure 3. Probability of customer return to platform by agent quality for different order prices.
We hypothesize that once a customer finds a good agent, they should be more inclined to go off platform when they require repeat services ( ). The results show that as services become more repetitive, the negative impact of prior dyad visits increases ( = −.5294, p <.01; H7a); the inverted U-shaped effect of agent quality is marginally more pronounced ( =.0333, p <.10; = −.0006, p <.10; H7b); and the effect of agent tenure is marginally greater ( = −.0014, p <.10; H7c). See Figures W2A–W2C in the Web Appendix.
In summary, results for the probability of customer return provide strong evidence of platform exploitation. As additional evidence, we consider the customer return hazard. Platform exploitation suggests that our key variables of interest should have limited influence on the time between orders for customers who come back to the platform. With Moderators 2b, we examine the cumulative probability of customer return to the platform by day, following their most recent order (conditional on retention). It represents the (conditional) failure rate in hazard terminology. Agent quality, agent tenure, prior dyad visits, and most of the interactions with price and service repetitiveness have little influence. Rather, the time between orders depends mostly on service characteristics: restricting service dummy effects jointly to 0 in the hazard yields = 210.86, with p <.01. Platform exploitation affects retention rather than time between orders for patients.
Table 6 presents the agent-side results. Platform exploitation suggests that the main variables which decrease customer retention will also decrease agent return rate. For the main-effect hypotheses, the impact of prior dyad visits, agent quality, and agent tenure on agent return rate should mirror customer side results. We do not find support for H1b, as the main effect of prior dyadic transactions is insignificant ( =.0046, p >.10). However, the significant effects of agent quality ( =.0380, p <.01) and its square ( = −.0012, p <.01) support H2b; the negative and significant impact of agent tenure ( = −.0010, p <.01) supports H3b. Customers are more likely to defect with high-quality or long-tenured agents; these agents are slower to return to the platform as their private customer base grows.
Graph
Table 6. Hazard Estimation Results for Agent's Return to the Platform.
| Nurse Return Hazard |
|---|
| Base 1c | Moderators 2c | H (sign) | Support |
|---|
| θs | SE | θs | SE |
|---|
| PriorDyadVisits | −.0094** | .0046 | .0046 | .0068 | H1b (−) | No |
| AgentQuality | −.0060*** | .0016 | .0380*** | .0027 | H2b (+) | |
| AgentQuality2 | | | −.0012*** | .00003 | H2b (−) | Yes |
| AgentTenure | −.0010*** | .0001 | −.0010*** | .0001 | H3b (−) | Yes |
| PriorDyadVisits × ln(Price) | | | −.0014 | .0100 | H4b (−) | No |
| AgentQuality × ln(Price) | | | .0051*** | .0016 | H5b (+) | |
| AgentQuality2 × ln(Price) | | | −.0001* | .00003 | H5b (−) | Yes |
| AgentTenure × ln(Price) | | | −.00014* | .00008 | H6b (−) | Yes |
| PriorDyadVisits × %CloseCustomers | | | −.0342** | .0155 | H8a (−) | Yes |
| AgentQuality × %CloseCustomers | | | .0163*** | .0043 | H8b (+) | |
| AgentQuality2 × %CloseCustomers | | | −.0002*** | .0001 | H8b (−) | Yes |
| AgentTenure × %CloseCustomers | | | −.0002* | .0001 | H8c (−) | Yes |
| PriorDyadVisits × %RepeatCustomers | | | −.0513*** | .0189 | H9a (−) | Yes |
| AgentQuality × %RepeatCustomers | | | .0016 | .0097 | H9b (+) | |
| AgentQuality2 × %RepeatCustomers | | | −.0001 | .0002 | H9b (−) | No |
| AgentTenure × %RepeatCustomers | | | −.0012*** | .0004 | H9c (−) | Yes |
| PotentiallyStolenCustomers | .0094*** | .0013 | .0133*** | .0018 | | |
| PotentiallyStolenCustomers2 | | | −.00002*** | .00001 | | |
| ln(Price) | −.2688*** | .0279 | −.2062*** | .0298 | | |
| %ServiceRepeat | −.6499*** | .1264 | −.5928*** | .1258 | | |
| %CloseCustomers | −.0249 | .0321 | .0752* | .0438 | | |
| %RepeatCustomers | −.4363*** | .0733 | −.3369*** | .0963 | | |
| CustomerSameService | .0003 | .0049 | .0018 | .0050 | | |
| CustomerPriorOrders | .0025 | .0045 | .0009 | .0046 | | |
| TimeOnApp | −.0001** | .0001 | −.0001** | .0001 | | |
| AgentSameService | −.0001 | .0008 | −.0012 | .0008 | | |
| AgentOrderNum | −.0052*** | .0009 | −.0056*** | .0010 | | |
| AgentHomeVisits | .0030*** | .0001 | .0026*** | .0001 | | |
| AgentProExp | −.0308*** | .0051 | −.0151** | .0074 | | |
| ln(1 + EquipmentCosts) | .0370*** | .0063 | .0321*** | .0064 | | |
| MultiVisit | .2156*** | .0385 | .1483*** | .0395 | | |
| ln(1+Distance) | .0355*** | .0050 | .0338*** | .0050 | | |
| Scramblea | −.1259*** | .0266 | −.1198*** | .0271 | | |
| RequestScramblea | −.1530*** | .0432 | −.1515*** | .0434 | | |
| Constant | –2.7685*** | .2244 | –2.1736*** | .2168 | | |
| Service, City, and Time Fixed Effects | ✓ | | ✓ | | | |
| | .7556*** | .0053 | .7573*** | .0053 | | |
| | −.0309*** | .0012 | −.0311*** | .0012 | | |
| Random Effects Parameters | | | | | | |
| | .5864*** | .0259 | .4438*** | .0262 | | |
| | .7064*** | .0499 | .4552*** | .0462 | | |
| | .1689** | .0750 | .0469 | .0619 | | |
| Log-likelihood | –74,564.9 | –74,373.8 | | |
| Wald (p-value) | 621.68 (.000) | 710.12 (.000) | | |
| N | 17,636 | 17,636 | | |
- 11 *Significant at the 10% level.
- 12 **Significant at the 5% level.
- 13 ***Significant at the 1% level.
- 14 aRequest is the base category.
Similar to the customer side, we hypothesize that the impact of the main variables should be magnified at higher prices. The motivation for both parties to move off platform with a good match is stronger when the financial incentive is larger. We do not find support for H4b, as × is insignificant ( = −.0014, p >.10). However, price interactions with agent quality ( =.0051, p <.01; = −.0001, p <.10) and agent tenure ( = −.00014, p <.10) support H5b and offer marginal support for H6b. At higher prices, high-quality or long-tenured agents are in a better position to move customers off platform. To gauge effect sizes, Figures W3A and W3B in the Web Appendix depict the return rate in the seven days since the last on-platform order by and at different prices. Although the moderating effect of price on agent quality is statistically significant, it is relatively small. The inverted U-shaped effect is relatively stable, though it shifts down as price increases. In this example, the negative main effect of price on return rate dominates the moderation effect. Results are similar for agent tenure (i.e., significant but relatively small impact).
In contrast, characteristics of the agent's on-platform customer pool have greater moderating effects. Agents have more exploitation opportunities when they interact with a relatively larger pool of desirable customers on the platform. These characteristics likely strengthen the impact of the main variables because customers only leave the platform with agents that are a good match.
For example, agents prefer working with customers who are geographically close. There are more opportunities to take customers off platform when an agent serves a greater percentage of geographically close customers in their on-platform customer pool. The interactions of with prior dyad visits ( = −.0342, p <.05), agent quality ( =.0163, p <.01; = −.0002, p <.01), and agent tenure ( = −.0002, p <.10) are significant (marginal for agent tenure) with expected signs, in support of H8a–H8c. Figures W4A–W4C in the Web Appendix show that the moderating effects are relatively large. As increases, the impact of goes from positive to negative, the inverted U-shaped effect of agent quality is more distinct, and the effect of agent tenure is greater. Agents with a larger percentage of close customers in their on-platform customer pool are better able to leverage exploitation opportunities from more prior dyad visits, higher agent quality, and longer agent tenure.
Agents also prefer working with repeat customers. An agent with a larger percentage of repeat customers in their on-platform customer pool has more opportunities to take customers off platform. We find that negatively moderates the impact of ( = −.0513, p <.01) and ( = −.0012, p <.01), in support of H9a and H9c. However, we do not find support for H9b, as the interactions with the agent quality variables are insignificant. Figures W5A and W5B in the Web Appendix show that the significant moderation effects of are also relatively large. Agents with a larger percentage of repeat on-platform customers are in a better position to capitalize on prior dyad visit or agent tenure related exploitation opportunities.
Lastly, customer- and agent-specific random effects are significant in Tables 5 and 6. We find a positive correlation between agent and customer effects in the cure fraction ( =.4552, p <.01), which suggests that customers are less likely to return when they interact with agents who come back to the platform less frequently, which is consistent with platform exploitation.
Our theorizing generates distinct hypotheses for agents and customers, with their different motivations, in support for our approach of modeling them separately. However, the main-effect hypotheses and price interactions are congruent for both sides. Thus, we model returns at the customer–agent dyad level as a robustness check. Because each dyad contains a specific customer and a specific agent, it is not observed again if the customer defects, so customer-side moderators could influence the dyads. In contrast, agents try to build portfolios of offline customers and are unlikely to leave the platform completely, so we do not rely on the dyadic analysis to provide insights into how quickly agents return to serve additional customers.
The dyad model and results appear in Web Appendix, Table W7; they offer broad support for the main-effect hypotheses, price interactions, and most customer-side interactions. The results for the agent-side interactions are mixed, which is not surprising, as we indicated that these moderators are unlikely to influence both parties. In summary, the results are largely robust for dyads.
After documenting the existence and nature of platform exploitation, we nextinvestigate ways to potentially reduce it. We conduct two scenario-based experiments using a fictional platform that connects agents with dog-owning customers (DogGo; similar to Rover or Wag). We test a financial intervention that alters the incentive structure and a social intervention that aims to build agent–platform community. We screened respondents from Amazon Mechanical Turk to ensure that offering dog walking services for pay is something they would consider (so that the respondents' mindsets are similar to those held by real platform users). If not, they were eliminated from the study.
The nurses in Study 1 indicated that reduced fees would decrease their motivation to defect with customers and lead them to pursue more orders on the platform. Thus, we test a sliding-scale fee, such that the platform's commission decreases with more on-platform services.
For the independent variables, we manipulated customer–agent relationship quality (RQ) and the platform's fee structure. We assigned 324 participants randomly to a 2 (RQ: high vs. low) × 2 (fee structure: fixed percentage vs. sliding scale) factorial design. The scenario explained the platform rules and asked the participants to imagine being an agent on the platform (Web Appendix, Figure W6A). In the high-RQ condition, agents read that they had seen the customer several times and were familiar with their expectations. In the low-RQ condition, they instead read that they had not seen the customer before and were unsure of their expectations. Next, in the fixed-fee condition, agents learned that the platform commission is 30% of the transaction price, whereas the sliding-scale condition explained that when agents complete more orders per week, the commission rate per transaction decreases. We measured off-platform transaction intentions by asking respondents how likely they were to suggest future transactions off platform with the customer (seven-point scale; 1 = "extremely unlikely," and 7 = "extremely likely").[17] We concluded with manipulation checks.
For our manipulations, respondents in the high-RQ (vs. low-RQ) condition rated trust with customers higher (Mhigh = 5.7, Mlow = 4.0; t(320) = 10.77, p <.01). The percentage of participants who correctly identified the fee policy was high in both the sliding-scale (P = 94.0%, ( 2) = 250.90, p <.01) and fixed-fee (P = 80.9%, ( 2) = 177.02, p <.01) conditions.
An analysis of variance for off-platform transaction intentions reveals main effects of both RQ (F( 1, 320) = 16.27, p <.01) and fee structure (F( 1, 320) = 6.59, p <.05), as well as an interaction effect (F( 1, 320) = 4.04, p <.05) (Web Appendix, Figure W7A). Planned contrasts show that in the fixed-fee model, the findings match those from our main studies, in that participants indicate greater intentions to take customers off platform in the high-RQ (vs. low-RQ) condition (Mhigh = 4.95, Mlow = 3.66; t(320) = 4.43, p <.01). However, the sliding-scale fee mitigates this RQ effect, such that high-RQ no longer leads to higher off-platform intentions (Mhigh = 3.97, Mlow = 3.54; t(320) = 1.39, p =.17). Overall, the results suggest that a sliding-scale fee is a promising intervention, effective for counterbalancing the effect of high relationship quality.
Financial interventions can be costly, so we also test whether platforms might leverage social mechanisms to decrease platform exploitation. Because organizational commitment can be enhanced through employee community building and socialization ([ 6]), we test the effectiveness of a program that enables platform service agents to build stronger connections with other platform-affiliated agents and with the platform itself.
In this experiment, we manipulated customer–agent relationship quality and the presence of a community program. The 331 participants were randomly assigned to a 2 (RQ: high vs. low) × 2 (community program: yes vs. no) factorial design. The context and procedures are similar to those in the previous experiment, except that we manipulated the presence of a community program, such that participants read about the benefits (community status and offline social events) that could be earned with a certain number of platform orders in the program condition (Web Appendix, Figure W6B) but received no such information in the no-program condition.
The manipulations were successful; participants in the high-RQ (vs. low-RQ) condition rated their trust with customers higher (Mhigh = 5.4, Mlow = 3.7, t(327) = 10.96, p <.01). The percentage of subjects who correctly identified the presence of a community program was higher in both the community (P = 82.1%, ( 2) = 156.01, p <.01) and no-community (P = 78.5%, ( 2) = 173.32, p <.01) program conditions.
The analysis of variance for off-platform transaction intentions revealed a main effect of RQ (F( 1, 327) = 22.90, p <.01), replicating the results from the first experiment (Web Appendix, Figure W7B). The main effect of the community program significantly reduces off-platform intentions (F( 1, 327) = 5.77, p <.05). The lack of interaction effect (F( 1, 327) =.33, p =.56) indicates that the presence of the community program does not specifically mitigate the effects of high RQ, but that the intervention is effective regardless of the level of RQ (high RQ: Mno community = 4.60, Mcommunity = 3.96; low RQ: Mno community = 3.45, Mcommunity = 3.05).
The findings from these two experiments corroborate our theorizing. In a trusting relationship, agents are more likely to move future transactions with that customer off platform, but in low-quality relationships, the parties prefer the relative safety of the platform. We demonstrate this effect using a distinct service setting (dog walking), which suggests the generalizability of our findings from Studies 1 and 2.
The experiments also provide insights for how managers might reduce platform exploitation. First, a sliding-scale fee increases the likelihood that agents keep their future transactions on the platform. Such a policy appears most effective for high-quality relationships, for which the other benefits of using the platform are lower. Without an offsetting financial incentive, the agent is motivated to move off platform. Second, a program that builds a sense of community also can reduce the likelihood of platform exploitation, affecting both high- and low-RQ conditions. Regardless of trust levels, community building can enhance agents' loyalty to the platform.
Growing the base of users is critical for platform businesses ([11]), but defection by agents and customers remains an ongoing problem. Little scholarly research has addressed this issue. With this article, we establish the existence and nature of platform exploitation, isolate key drivers, and provide potential remedies.
We highlight four implications for managers of online service platforms. First, high-quality agents and those with longer platform tenure are more likely to leave. Ironically, platforms need high supply-side quality to keep customers ([13]), but doing so increases the risk of exploitation. Managers of platforms on which high-quality service prompts interpersonal trust must recognize that high-quality agents are double-edged swords in business models based on a fixed percentage-of-price fee. In addition to a sliding-scale fee, managers might segment agents on the basis of quality and tenure and design different incentive strategies to encourage on-platform transactions. Furthermore, subscriptions or flat-fee models might be designed to reduce customer incentives to move off platform, even if an agent seeks to do so.
Second, managers should regularly emphasize the benefits of staying on-platform for customers and seek to enhance those benefits; repeat customers are essential for platform profitability. But it would be a mistake to assume that customers appreciate all platform benefits. Existing customers, particularly those who consume high-priced or repetitive services, may need as much attention as new customers. Thus, managers could segment customers according to their consumption patterns and apply different messaging and incentive strategies to these segments (e.g., premium benefits such as guarantees, insurance, special platform features; [28]).
Third, high-quality, long-tenured agents who defect and take customers with them are less available for on-platform services. The situation worsens when they interact with susceptible (e.g., repeat, proximal) customer pools. Managers thus might need to reimagine the agent–platform relationship. Initially, they are codependent: the platform relies on agents to fill orders, and agents rely on the platform to access customers. However, agents become less dependent as they meet more customers. To strengthen the relationships, platforms might offer usage incentives or even consider directly employing agents, though our interviews and anecdotal evidence (e.g., Amazon Home Services; [18]) suggest mixed results of direct employment experiments.
Fourth, human needs for recognition and social belonging can be leveraged; in the program we tested, offering special status markers and opportunities to socialize with other members of the platform community reduced the likelihood that agents tried to take customers off platform. Perhaps other agent socialization approaches could be effective too, such as encouraging mentor–mentee relationships.
Very little research has discussed platform exploitation, a gap with important implications for building theory about relationships between platforms and their ecosystem members (e.g., service agents) ([13]). Previous platform and two-sided market research take a perspective that largely emphasizes the symbiotic relationship of the two sides—for example, how ecosystem members and platforms help each other ([20]; [26]). However, our research introduces a perspective that stresses the friction between the two. Through theories-in-use and secondary data analysis, we show how platform-dependent service agents can knowingly break platform rules and retain customers for themselves. Our results suggest that platforms experience financial harm from exploitation. Further, while previous research emphasizes how the two sides of a platform market are reliant on each other, our results suggest that platform exploitation may decrease service agents' dependence on the platform.
Our research also contributes to platform literature by examining pure-labor platforms and adding nuances to findings in extant research. Most of the platform research focuses on physical products, product sharing (e.g., Airbnb), or product-based service (e.g., Uber) platforms. For example, studies of product transactions indicate that high supply-side quality enhances platform growth ([20]). We clarify that in a service setting, quality helps ensure early attraction and retention but also increases the likelihood of defection. Scholars might test other findings generated in product contexts to see if new insights emerge in service contexts.
Finally, in studies of the negative consequences of close customer–agent relationships ([ 8]; [17]) and employee turnover ([ 7]), the focus is often employees of traditional firms, rather than agents with substantial options to behave opportunistically. This distinction is important. Online service markets are two-sided platforms, so managers need to build a customer base by increasing supply-side quality, but when they do so, they also risk increasing platform exploitation. We reveal the dark side of customer relationships in a context where these effects might not be readily apparent.
Further research might address limitations of our study. The study context is a typical service platform (see Table 2), but managers should take care in extrapolating findings across settings. Our results likely generalize to platforms that offer services rendered in close proximity, involve commission-based pricing, rely on the agent's skill, and involve customer–agent coordination. It would be helpful to test the robustness of our findings in other contexts, such as home services (e.g., TaskRabbit) or beauty care (e.g., Stylebee). Furthermore, we consider conditions that make platform exploitation more severe and use two experiments to test mechanisms to combat it. Managers also might leverage other moderators, which further research can identify and test, perhaps in field experiments that analyze whether directly employing agents or altering fee structures reduce defection in reality. Finally, even as pure-labor platforms grow more common, we lack detailed insights into the challenges of managing the dual agent–platform relationship (cooperative and competitive) and the long-term implications of platform exploitation. Thus, service platform research offers many promising avenues for scholars.
sj-pdf-1-jmx-10.1177_00222429211001311 - Supplemental material for Platform Exploitation: When Service Agents Defect with Customers from Online Service Platforms
Supplemental material, sj-pdf-1-jmx-10.1177_00222429211001311 for Platform Exploitation: When Service Agents Defect with Customers from Online Service Platforms by Qiang (Kris) Zhou, B.J. Allen, Richard T. Gretz and Mark B. Houston in Journal of Marketing
Footnotes 1 Stefan Wuyts
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Qiang (Kris) Zhou https://orcid.org/0000-0003-2230-514X
5 Online supplement: https://doi.org/10.1177/00222429211001311
6 Because we predict an inverted U-shaped main effect of agent quality, its moderation by price (and subsequent moderators), strengthening this effect, should intensify the inverted U-shaped effect. This results in a greater positive effect of the linear term and a greater negative effect of the quadratic term.
7 Each service and city dummy includes at least 50 observations to avoid bias issues due to a small n per fixed effect ([19]). We pool observations for services for which n < 50. Smaller cities with few observations also are merged with geographically close cities. These efforts involve few observations, and the results are robust to eliminating these observations.
8 These percentages include new platform customers; 41.4% of customers request an agent after their first visit.
9 Hazard models have a long history in marketing. [10] provide an early survey, and Kumar, Leszkiewicz, and Herbst (2018) offer a recent application to customer retention. These models deal with right-censoring; multispell hazard models address repeated events by the same individual ([3]).
Standard survival models (e.g., hazard) measure the time until an event happens (i.e., repeat customer on the platform). The cure fraction is a survival analysis that accounts for individuals who do not experience the event (i.e., do not repeat); in our case, this is the portion of patients who do not place additional orders on the platform.
This vector includes the characteristics of the agent who fills order j. For example, is the quality of the agent who fills customer i's jth order.
Here, includes characteristics of the customer that agent k serves in the mth order. For example, is the number of orders by the customer prior to being served in agent k's mth order.
The Expo-power distribution allows for a "monotonically increasing, monotonically decreasing, U-shaped, or inverted U-shaped" ([36], p. 372) baseline hazard.
Random effects are the standard way to deal with unobserved heterogeneity in repeated-event hazard models ([44]). Fixed effects can introduce severe biases, especially in a context like ours, with relatively few observations per panel and a censored last observation in each panel ([2]). However, because there is a fixed-effects alternative to random-effects logit ([19]), we can incorporate fixed effects in the cure fraction without introducing bias. We use the test suggested by [42] and reject the hypothesis that fixed effects are preferred over random effects in the cure ( = 37.90, p =.6516).
We log-transform , , and , because we compare disparate sets of services and distances with different values for customers and agents, which also address right-skewness. We add 1 to and to ensure that these transformations are defined when they equal 0.
We numerically integrate to incorporate the random effects using Gaussian–Hermite quadrature with 12 basis points for each random effect. It results in 1,728 = 12 × 12 × 12 total basis points. We provide the analytical gradient and hessian to speed up the optimization. The maximum likelihood estimation command is available from the authors on request.
We informed the participants that off-platform transactions sometimes occur. To check for robustness, in another, unreported study with the same design, we (1) did not inform subjects about the phenomenon and (2) asked them how likely they were to agree to an off-platform transaction if requested by a customer. The results did not change.
References Akerlof George A. (1970), " The Market for 'Lemons': Quality Uncertainty and the Market Mechanism ," Quarterly Journal of Economics , 84 (3), 488 – 500.
Allison Paul D. (2009), Fixed Effects Regression Models, Quantitative Applications in the Social Sciences. Thousand Oaks, CA : SAGE Publications.
Allison Paul D. (2014), Event History and Survival Analysis, Quantitative Applications in the Social Sciences. Thousand Oaks, CA : SAGE Publications.
Apte Uday M. , Davis Mark M.. (2019), " Sharing Economy Services: Business Model Generation ," California Management Review , 61 (2), 104 – 31.
Arya Anil , Mittendorf Brian , Sappington David E.M.. (2007), " The Bright Side of Supplier Encroachment ," Marketing Science , 26 (5), 651 – 59.
Bauer Talya N. , Bodner Todd , Erdogan Berrin , Truxillo Donald M. , Tucker Jennifer S.. (2007), " Newcomer Adjustment During Organizational Socialization: A Meta-Analytic Review of Antecedents, Outcomes, and Methods ," Journal of Applied Psychology , 92 (3), 707 – 21.
Bendapudi Neeli , Leone Robert P.. (2002), " Managing Business-to-Business Customer Relationships Following Key Contact Employee Turnover in a Vendor Firm ," Journal of Marketing , 66 (2), 83 – 101.
Brady Michael K. , Voorhees Clay M. , Brusco Michael J.. (2012), " Service Sweethearting: Its Antecedents and Customer Consequences ," Journal of Marketing , 76 (2), 81 – 98.
Cachon Gérard P. , Daniels Kaitlin M. , Lobel Ruben. (2017), " The Role of Surge Pricing on a Service Platform with Self-Scheduling Capacity ," Manufacturing & Service Operations Management , 19 (3), 368 – 84.
Chintagunta Pradeep K. , Dong Xiaojing. (2006), " Hazard/Survival Models in Marketing ," in The Handbook of Marketing Research , Vriens Marco , Grover Rajiv , eds. Thousand Oaks, CA : SAGE Publications , 441 – 54.
Chu Junhong , Manchanda Puneet. (2016), " Quantifying Cross and Direct Network Effects in Online Consumer-to-Consumer Platforms ," Marketing Science , 35 (6), 870 – 93.
Costello John P. , Reczek Rebecca Walker. (2020), " Providers Versus Platforms: Marketing Communications in the Sharing Economy ," Journal of Marketing , 84 (6), 22 – 38.
Eckhardt Giana M. , Houston Mark B. , Jiang Baojun , Lamberton Cait , Rindfleisch Aric , Zervas Georgios. (2019), " Marketing in the Sharing Economy ," Journal of Marketing , 83 (5), 5 – 27.
Eilert Meike , Jayachandran Satish , Kalaignanam Kartik , Swartz Tracey A.. (2017), " Does It Pay to Recall Your Product Early? An Empirical Investigation in the Automobile Industry ," Journal of Marketing , 81 (3), 111 – 29.
Fan Li-hua , Gao Lei , Liu Xin , Zhao Shi-hong , Mu Hui-tong , Li Zhe , et al. (2017), " Patients' Perceptions of Service Quality in China: An Investigation Using the SERVQUAL Model ," PLoS ONE , 12 (12).
Granovetter Mark. (1985), " Economic Action and Social Structure: The Problem of Embeddedness ," American Journal of Sociology , 91 (3), 481 – 510.
Grayson Kent. (2007), " Friendship Versus Business in Marketing Relationships ," Journal of Marketing , 71 (4), 121 – 39.
Green Dennis. (2018), " Amazon is Quietly Revamping One of Its Most Under-the-Radar Services—And It Should Terrify Yelp and Handy ," Business Insider (March 30), https://www.businessinsider.com/amazon-to-start-hiring-cleaners-2018-3.
Greene William H. (2018), Econometric Analysis , 8th ed. New York: Pearson.
Gretz Richard T. , Malshe Ashwin , Bauer Carlos , Basuroy Suman. (2019), " The Impact of Superstar and Non-Superstar Software on Hardware Sales: The Moderating Role of Hardware Lifecycle ," Journal of the Academy of Marketing Science , 47 (3), 394 – 416.
Gu Grace , Zhu Feng. (2020), " Trust and Disintermediation: Evidence from an Online Freelance Marketplace ," Management Science 67 (2), 794 – 807.
Hagiu Andrei , Rothman Simon. (2016), " Network Effects Aren't Enough ," Harvard Business Review , 94 (4), 64 – 71.
Huet Ellen. (2015), " What Really Killed Homejoy? It Couldn't Hold on to Its Customers ," Forbes (July 23), https://www.forbes.com/sites/ellenhuet/2015/07/23/what-really-killed-homejoy-it-couldnt-hold-onto-its-customers/#56d8dcbb1874.
Kanagaretnam Kiridaran , Mestelman Stuart , Khalid Nainar S.M. , Shehata Mohamed. (2010), " Trust and Reciprocity with Transparency and Repeated Interactions ," Journal of Business Research , 63 (3), 241 – 47.
Kumar V. , Leszkiewicz Agata , Herbst Angeliki. (2018), " Are You Back for Good or Still Shopping Around? Investigating Customers' Repeat Churn Behavior ," Journal of Marketing Research , 55 (2), 208 – 25.
Luo Xueming , Tong Siliang , Lin Zhijie , Zhang Cheng. (2021), " The Impact of Platform Protection Insurance on Buyers and Sellers in the Sharing Economy: A Natural Experiment ," Journal of Marketing , 85 (2), 50 – 69.
McCracken Grant. (1988), The Long Interview. New York : SAGE Publications.
Nobel Carmen. (2017), " The Most Pressing Issues for Platform Providers in The Sharing Economy ," Forbes (April 24), https://www.forbes.com/sites/hbsworkingknowledge/2017/04/24/the-most-pressing-issues-for-platform-providers-in-the-sharing-economy/#4a1a39c13cfa.
Price Linda L. , Arnould Eric J.. (1999), " Commercial Friendships: Service Provider–Client Relationships in Context ," Journal of Marketing , 63 (4), 38 – 56.
Prins Remco , Verhoef Peter C.. (2007), " Marketing Communication Drivers of Adoption Timing of a New E-Service Among Existing Customers ," Journal of Marketing , 71 (2), 169 – 83.
Rider Christopher I. , Samila Sampsa. (2018), "Envisioning Value: Certification, Matchmaking, and Returns to Brokerage," working paper. Georgetown University.
Rokkan Aksel I. , Heide Jan B. , Wathne Kenneth H.. (2003), " Specific Investments in Marketing Relationships: Expropriation and Bonding Effects ," Journal of Marketing Research , 40 (2), 210 – 24.
Saha Atanu , Hilton Lynette. (1997), " Expo-Power: A Flexible Hazard Function for Duration Data Models ," Economics Letters , 54 (3), 227 – 33.
Said Carolyn. (2015), " Could Client Poaching Undercut On-Demand Companies? " San Francisco Chronicle (April 24), https://www.sfchronicle.com/business/article/Could-client-poaching-undercut-on-demand-6222919.php.
Seetharaman P.B. (2004), " The Additive Risk Model for Purchase Timing ," Marketing Science , 23 (2), 234 – 42.
Seetharaman P.B. , Chintagunta Pradeep K.. (2003), " The Proportional Hazard Model for Purchase Timing: A Comparison of Alternative Specifications ," Journal of Business & Economic Statistics , 21 (3), 368 – 82.
Shapiro Susan P. (2005), " Agency Theory ," Annual Review of Sociology , 31 (1), 263 – 84.
Tian Lin , Jiang Baojun. (2018), " Effects of Consumer-to-Consumer Product Sharing on Distribution Channel ," Production and Operations Management , 27 (2), 350 – 67.
Waldfogel Joel , Reimers Imke. (2015), " Storming the Gatekeepers: Digital Disintermediation in the Market for Books ," Information Economics and Policy , 31 , 47 – 58.
Wathne Kenneth H. , Heide Jan B.. (2000), " Opportunism in Interfirm Relationships: Forms, Outcomes, and Solutions ," Journal of Marketing , 64 (4), 36 – 51.
Williamson Oliver E. (1985), The Economic Institutions of Capitalism. New York : Simon and Schuster.
Wooldridge Jeffrey M. (2010), Econometric Analysis of Cross Section and Panel Data. Cambridge, MA : MIT press.
Zeithaml Valarie A. , Jaworski Bernard J. , Kohli Ajay K. , Tuli Kapil R. , Ulaga Wolfgang , Zaltman Gerald. (2020), " A Theories-in-Use Approach to Building Marketing Theory ," Journal of Marketing , 84 (1), 32 – 51.
Zhang Yuchi , Moe Wendy W. , Schweidel David A.. (2017), " Modeling the Role of Message Content and Influencers in Social Media Rebroadcasting ," International Journal of Research in Marketing , 34 (1), 100 – 119.
Zhao Ruixue. (2018), " In-Home Nursing Rises as Latest Medical Care Trend ," China Daily (July 10), https://www.chinadaily.com.cn/a/201807/10/WS5b4405afa3103349141e1c16.html.
Zhu Feng , Iansiti Marco. (2019), " Why Some Platforms Thrive and Others Don't ," Harvard Business Review , 97 (1), 118 – 25.
~~~~~~~~
By Qiang Zhou; B.J. Allen; Richard T. Gretz and Mark B. Houston
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 100- Popping the Positive Illusion of Financial Responsibility Can Increase Personal Savings: Applications in Emerging and Western Markets. By: Garbinsky, Emily N.; Mead, Nicole L.; Gregg, Daniel. Journal of Marketing. May2021, Vol. 85 Issue 3, p97-112. 16p. 1 Chart, 2 Graphs. DOI: 10.1177/0022242920979647.
- Database:
- Business Source Complete
Popping the Positive Illusion of Financial Responsibility Can Increase Personal Savings: Applications in Emerging and Western Markets
People around the world are not saving enough money. The authors propose that one reason people undersave is because they hold the positive illusion of being financially responsible. If this conjecture is correct, then deflating this inflated self-view may increase saving, as people should become motivated to restore perceptions of financial responsibility. After establishing that people do hold the illusion of financial responsibility, the authors developed an intervention that combats this self-enhancing bias by triggering people to recognize their frequent engagement in superfluous spending. This superfluous-spender intervention increased saving by enhancing people's motivation to restore their diminished perceptions of financial responsibility. Consistent with theorizing, the intervention increased saving only when superfluous spending was under one's control and among those who were motivated to perceive themselves as financially responsible. In addition to increasing saving in Western countries, the superfluous-spender intervention increased saving of earned income and a financial windfall over time among chronically poor coffee growers in rural Uganda. Collectively, this work shows that people view their financial responsibility through rose-colored glasses, which can undermine their financial well-being. It also endows stakeholders with a simple, practical, and inexpensive intervention that offsets this bias to increase personal savings.
Keywords: behavioral intervention; emerging markets; financial well-being; positive illusions; saving; self-regulation
People around the world are not saving enough money. In North America, Europe, and Japan, personal savings rates have fallen ([25]; [62]). People in developing countries face the saving challenges of meager income and lack of secure savings facilities. Nevertheless, they too spend a portion of their scant income on nonnecessities, money that could otherwise be saved ([ 3]; [32]). For many households, a lack of savings means that they may not have enough money to cover an unexpected expense, which can result in a downward spiral of costly debt as they try to cover necessary spending such as housing, car, and child-related expenses ([31]).
A lack of savings can undermine personal and societal welfare, so many researchers have considered why people do not save enough. For example, research has shown that people's willingness to save money is influenced by relatively stable factors such as their financial status ([ 9]), childhood economic environment ([27]), level of self-control ([63]), or even the language they speak ([13]). Relatively less well studied is how practitioners can leverage psychological processes that govern saving to create implementable and scalable interventions to help people save more. Given that practitioners from a wide range of organizations across the world (e.g., fintech companies, commercial banks, nongovernmental organizations, governments) want to help people save more ([ 8]), the question of how to facilitate saving in a simple yet substantive way is important but relatively understudied.
In the current article, we aim to shed light on both of these issues. First, we leverage classic work on positive illusions to theorize that people hold the positive illusion of being financially responsible. In other words, even when they spend money superfluously, many people believe that they manage and save their money in a responsible fashion because this enables them to feel good about themselves. Second, drawing from classic work on self-regulation, we hypothesize and show that offsetting people's inflated perceptions of financial responsibility motivates them to restore those tempered self-views through increased saving. More specifically, we develop an intervention that increases people's motivation to save by drawing attention to the fact that they regularly engage in superfluous and thus irresponsible spending. Consequently, in addition to illuminating an antecedent to saving (i.e., the illusion of financial responsibility), we illustrate what financial stakeholders can do to combat this self-enhancing bias in an effort to encourage consumers to (accurately) reflect on their financial situation, realize their (many) surpassed savings opportunities, and commit to saving more.
People have a need to view themselves positively ([ 5]; [56]). Previous research has demonstrated the pervasiveness of these positive illusions, revealing that most people hold positively distorted views of themselves, overly optimistic expectations about the future, and an exaggerated sense of personal control ([59]). Most relevant to the current investigation is the finding that people are especially likely to develop distorted positive self-views when they are faced with stressful events ([59]). For this reason, the cultivation of positive self-views may be particularly evident in the financial domain, as financial matters are a common source of stress in daily life ([41]).
This raises the question of how these positive self-views manifest in the financial domain. Here, we suggest that people hold the positive illusion that they are financially responsible, which we define as the unrealistically positive self-perception that one is good at managing one's money, saving one's money, and using one's money in a sensible, nonsuperfluous manner. To illustrate, we anticipate that most people view themselves as more financially responsible than their average peer, which is statistically impossible (i.e., the better-than-average effect; [ 1]). Our hypothesis that people hold the positive illusion of financial responsibility is rooted in several lines of work.
First, people tend to hold positive illusions of themselves for attributes that are valued by others and society ([61]). A core aspect of financial responsibility, saving money, is a desirable attribute that people look for in a partner ([39]). Second, people are particularly likely to hold inflated self-views for attributes that are vague and abstract because the fuzziness of those constructs enables people to construe the attribute in a self-serving way ([15]; [30]). This tendency thus leads people to reach overly flattering conclusions about themselves that fail to align with objective evidence ([15]). Although financial actions are concrete, how those actions map onto more abstract judgments about financial responsibility is far from straightforward. In this way, people who want to perceive themselves as financially responsible can easily construe their financial behaviors in a flattering, self-serving way.
Third, prior work has shown that people try to preserve their perceptions of financial responsibility ([55]). More specifically, [55] found that consumers preferred to borrow from a high-interest credit rate option rather than dip into their savings when those savings were earmarked for a goal that they considered to be responsible (e.g., children's education). Although the purpose of our research is to examine how perceptions of financial responsibility affect the decision to save in the first place rather than people's willingness to preserve savings, this previous research clearly supports the notion that consumers perceive saving as financially responsible. Taken together, these lines of reasoning lead us to our first hypothesis:
- H1: People hold the positive illusion of financial responsibility (i.e., overly positive perceptions of how financially responsible they are).
At first glance, one might expect that the positive illusion of financial responsibility would encourage consistent behavior and thus facilitate saving. However, drawing from classic work on positive illusions and self-regulation, we conjecture that inflated perceptions of financial responsibility may contribute to undersaving.
People who hold positive illusions about themselves believe they are more likely than others to experience positive events and see themselves as less vulnerable to threatening events ([59]; [60]). Because positive illusions cultivate the belief that bad things happen to others rather than to oneself, individuals who hold the illusion of financial responsibility may feel less of a need to save money for unexpected (emergency) expenses. Similarly, people's rose-tinted glasses engender the belief that future conditions will be favorable and thus they can act in an ideal manner at a later point in time (e.g., saving money; [57]). In this way, people who hold the positive illusion of financial responsibility may not feel the need to save in the present because they think they will save in the future. Finally, for those who need to save the most, taking a cold, hard look at their finances is likely a sobering and unpleasant process. The desire to feel good about oneself and avoid discomfort—basic components of the positive illusion ([60])—may lead people to postpone or even avoid taking such steps.
Research on self-regulation tells a similar story. People monitor their standing on important dimensions (e.g., how they are doing as a parent, teacher, or researcher; [12]). When people do not perceive a discrepancy between their behavior and their standard, they do not change their behavior; when they do perceive a discrepancy, they try to bring their behavior in line with their desired standard (e.g., they devote more time and energy to their children, class, or research projects). Because positive illusions cause people to perceive themselves in a flattering, self-serving way ([59]; [58]; [61]), they may not perceive that they are falling short of their desired standard—in the current context, that they are financially responsible. In this way, people will not perceive a need to change their behavior and thus will not take actions toward increasing their savings.
If the positive illusion of financial responsibility undermines people's inclination to save, then offsetting the positive illusion of financial responsibility—namely, by drawing attention to their routine superfluous spending—should increase saving. As alluded to in the previous paragraph, when people perceive that their behavior has fallen short of their desired standard, they become motivated to enact behaviors that will close the gap ([11]; [20]; [29]; [34]). Because self-regulation processes persist until one's standing on the threatened dimension has been achieved ([11]; [24]; [37]), offsetting people's perceptions of financial responsibility should motivate them to restore those diminished perceptions through increased saving. Note that because there is no ultimate attainment of identity goals ([23]), people who are striving to restore perceptions of financial responsibility may continue to save beyond the initial threat. We thus hypothesize the following:
- H2: Diminishing the positive illusion of financial responsibility (by highlighting that one frequently engages in superfluous spending) increases savings intentions and behavior.
- H3: The desire to restore diminished perceptions of financial responsibility mediates the effect of the superfluous-spender intervention on savings.
In addition to mediation, we also aimed to test our proposed process by examining two boundary conditions that follow directly from our theorizing (i.e., process through moderation; [52]). We selected these boundary conditions because they provide theoretical value and illustrate managerial relevance.
Situational circumstances alter whether people perceive a discrepancy between their enacted behavior and their desired self-views ([14]). Most relevant to the current work, when people perceive that their behavior is not reflective of themselves, they do not feel that they have fallen short of their desired standards and thus they do not perceive the need to engage in restorative actions ([64]).
Spending is sometimes freely chosen, but other times it is outside of people's control. Previous research suggests that people do not use their past behavior as a clue to their inner states when they believe that their behavior is under external control ([ 2]; [ 7]; [36]). We thus hypothesize that when people can attribute their past superfluous spending to external forces, their perceptions of financial responsibility should remain intact, and they do not need to bolster those perceptions through increased saving. More formally,
- H4: Highlighting frequent engagement in superfluous spending increases saving when superfluous spending is under one's control, but not when superfluous spending is under external control.
This proposed moderator not only has theoretical value but also is practically relevant in the domain of financial decision making, as people's financial decisions may be determined by external factors.
In addition to identifying a situational variable, we wanted to identify an individual difference variable that predicts variation in the desire and importance placed in viewing oneself as financially responsible. When a specific self-view is important, people try to restore it when it has been cast into doubt ([24]; [51]; [54]). By contrast, when the threatened self-view is not important, people can engage in general self-worth restoration. In this way, identifying a relevant individual difference variable can shed light on our proposed underlying process.
For some people, happiness and satisfaction come from spending money on material goods instead of from saving (i.e., materialistic happiness, a subfactor of materialism; [43]; [44]). For these individuals, we anticipated (and confirmed) that they are relatively less motivated to view themselves as financially responsible. If diminishing perceptions of financial responsibility increases saving because people are then motivated to restore those self-perceptions, then increased saving should be observed among those who are motivated to perceive themselves as financially responsible—namely, those low (but not high) on materialistic happiness. More formally,
- H5: Highlighting frequent engagement in superfluous spending increases saving, but only among those who are motivated to perceive themselves as financially responsible (i.e., those scoring low on materialistic happiness).
In addition to its theoretical value, this moderator has managerial value. Generally speaking, knowing which consumers will respond positively to the intervention is helpful for using limited resources effectively. More specifically, this segmentation basis may be useful to marketing managers, as policy makers often include materialism-like constructs in national surveys and polls to interpret social trends and concerns ([43]).
Our proposed conceptual model suggests that diminishing inflated perceptions of financial responsibility will increase saving specifically. However, the self-affirmation literature suggests that, in response to self-threats, people can restore general feelings of self-worth ([50]; [53]). The prediction from that literature, then, is that people can address threats to their perceptions of financial responsibility by engaging in any action that bolsters self-worth, such as donating to charity or buying something nice for someone special ([16], [17]; [17]). Although some people may choose to bolster feelings of self-worth generally rather than perceptions of financial responsibility specifically, our conceptual model proposes that most people will choose the latter route when their perceptions of financial responsibility have been cast into doubt. This expectation is grounded in classic work on self-regulation, and we test the specificity of our model in our empirical investigation.
Although bolstering unrelated aspects of the self can improve general feelings of self-worth, it does not close the gap between one's behavior (in this case, superfluous spending) and desired self-view (in this case, financial responsibility). In this way, self-regulation processes that are geared toward achieving the important standard of financial responsibility should direct people toward restoring perceptions of financial responsibility until their standing on that dimension has been achieved ([11]; [24]; [37]). Furthermore, according to the principle of multifinality ([33]; [49]), people will choose whichever means serves the most goals. When financial responsibility is considered a positive attribute, the act of saving can restore both perceptions of financial responsibility and general self-worth, and thus it should be preferred over means that fulfill general self-worth but not financial responsibility. Indeed, when given the option to address a threat specifically or bolster feelings more generally, people chose to address the threat specifically ([54]).
Across eight studies, we tested the hypothesis that offsetting the positive illusion of financial responsibility with the superfluous-spender intervention increases people's motivation to restore their tempered perceptions of financial responsibility through increased saving. To begin, we conducted two pilot studies to test and confirm that ( 1) people hold the positive illusion of financial responsibility and ( 2) our superfluous-spender intervention diminishes inflated perceptions of financial responsibility. Then, we turned to our primary hypothesis testing.
In Study 1, we tested the hypothesis that the superfluous-spender intervention increases intentions to save relative to a control and a financially responsible condition. In Study 2, we examined the effect of the superfluous-spender intervention on saving of earned income in a three-week diary study with chronically poor coffee growers in rural Uganda. We replicated the effectiveness of the intervention in a follow-up study (also conducted in rural Uganda) examining saving of a monetary windfall, as windfalls represent a prime opportunity to set aside money for unexpected emergency expenses.
Studies 3–5 tested our proposed process through mediation and moderation. More specifically, Study 3 tested the hypothesis that the superfluous-spender intervention (vs. control procedure) increases saving through the desire to restore diminished perceptions of financial responsibility. Studies 4 and 5 tested theoretically and practically relevant boundary conditions for our basic effect. We expected the superfluous-spender intervention to increase saving intentions, but only when the superfluous spending behaviors were perceived to be under one's control and thus attributable to the self (Study 4), and only among those who were relatively motivated to view themselves as financially responsible (i.e., those scoring low in materialism; Study 5).
In all studies, data analysis occurred after we reached our desired sample size. For our sample size determination in each study, see Web Appendix A. Consistent with previous research, we used attention check questions when relevant and excluded those who did not pass the check ([40]); details on the attention check and all exclusions applied are provided in each study. In all studies with participant exclusions, chi-square analyses indicated that exclusion did not differ by experimental condition.
In this study, we tested the hypothesis that people hold the positive illusion of being financially responsible (H1). One of the most common ways of assessing people's inclination to perceive themselves in an overly positive fashion is with the better-than-average paradigm. In the direct version, respondents evaluate themselves as better (or worse) than their average peer on a number of characteristics, attributes, skills, or traits. Although only half of the population can be above average on a given trait, the motivation to view oneself positively causes the majority of people to report themselves as above average ([ 1]; [47]). Because people's desire to feel good about themselves drives these higher-than-average ratings, the better-than-average effect emerges only for characteristics that are positive and valued by society.
We posit that most people view financial responsibility as a positive trait and thus are motivated to claim it for themselves. If this hypothesis is correct, reducing the attractiveness of financial responsibility should decrease people's proclivity to rate themselves as more financially responsible than others. We therefore manipulated whether saving, a core part of financial responsibility ([55]), was valued by society. If participants are motivated to view themselves as financially responsible, then self-evaluations of financial responsibility should be lower among those who learn that saving money has fallen out of favor (saving-is-less-desirable condition) compared with those who learn that saving money is valued by society (saving-is-good condition). We included a control condition to confirm the direction of our effects. If participants view financial responsibility as a positive trait, then self-evaluations of financial responsibility should be similar across the control and saving-is-good conditions.
We conducted this study within a larger research session. We made our study available to all participants who took part in the study session. Four hundred North American university students (Mage = 19 years; 56% female) completed the study in exchange for partial course credit.
To minimize suspicion and experimental demand, we framed the study as a test of consumer memory. Participants started by reading one of three mock newspaper articles (Web Appendix B) for which their memory would ostensibly be tested later in the study. Participants randomly assigned to the saving-is-good condition read an article that depicted saving as positive and likable, whereas those in the saving-is-less-desirable condition read an article suggesting that saving is not viewed as positively as it used to be. Participants in the control condition read an article about jellyfish. A pretest (Web Appendix B) with 87 participants drawn from a similar population confirmed that the manipulation had its intended effect.
After reading one of the three articles, we assessed self-evaluations of financial responsibility. Specifically, respondents indicated how they compared with the average university student on three financial responsibility items: "How good are you at saving your money?," "How financially responsible are you?," and "How much are you a saver?" Participants rated themselves on each attribute using a nine-point scale (0 = "much less than the average student," 4 = "about the same as the average student," and 8 = "much more than the average student"). These three items were related, so we combined them into an index of financial responsibility (α =.86).
To reduce experimental demand, we mixed the three target (financial) traits with six filler (nonfinancial) traits. Three of the filler traits were positive (polite, considerate, and respectful) and three were negative (unpleasant, dishonest, and disrespectful). We randomized the order of the nine traits. As a further precaution against demand, we framed the self-evaluation questions as unrelated to the savings article: participants were told that the attribute measures were designed to clear short-term memory before the memory test. These steps were effective for reducing suspicion.
Before analyzing the data, we excluded participants who met the exclusion criteria: 14 (3.5%) failed the attention check, 37 (9.3%) failed the article comprehension check (i.e., they did not know the takeaway message of the article), 13 (3.3%) believed that the article was fake, and 3 (.1%) guessed that we were trying to change their self-perceptions through the article. Three-hundred thirty-five responses remained.
As an initial test of the hypothesis that people hold the positive illusion of financial responsibility, we compared the average self-evaluation of financial responsibility with the midpoint of the scale ( 4) for the entire sample. In support of H1, the average respondent in this sample considered themselves to be more financially responsible than the average university student (M = 5.04, SD = 1.58; t(334) = 11.975, p <.001). Thus, this result supports the hypothesis that people hold the positive illusion of financial responsibility.
Second, we tested the assumption that people are motivated to view themselves as financially responsible because this enables them to feel good about themselves. If so, self-evaluations of financial responsibility should be reduced when saving is framed as relatively less socially desirable. Predicting self-evaluations of financial responsibility, an omnibus analysis of variance (ANOVA) revealed a main effect of experimental condition (F( 2, 332) = 3.326, p =.037; η2 =.020). In support of our theorizing, participants who read the saving-is-less-desirable article reported lower self-evaluations of financial responsibility (M = 4.72, SD = 1.70) compared with participants who read the saving-is-good article (M = 5.22, SD = 1.45; t(332) = 2.373, p =.018, d =.30) and participants who read the control (jellyfish) article (M = 5.16, SD = 1.56; t(332) = 2.078, p =.038, d =.27). Consistent with the notion that people view financial responsibility as desirable, participants in the control and saving-is-good conditions rated themselves similarly (t(332) =.278, p =.781).
Notably, each condition's mean was higher than the midpoint (ps <.001). In the saving-is-less-desirable condition, this could mean that the motivation to perceive oneself as financially responsible is very strong and robust, that our manipulation was not very powerful, or most likely a combination of the two. We present analyses for the filler traits, for which we did not have a priori predictions, in Web Appendix B.
The results of Pilot Study 1 provide experimental support for the basic notion that people hold the positive illusion of financial responsibility. With this result in hand, we next offset this illusion by developing an intervention.
In response to the need to develop simple, effective, and nonphysical interventions that tackle psychological barriers to saving ([32]), we aimed to develop an intervention that both offsets the positive illusion of financial responsibility and can be implemented in practice. To do so, we aimed to highlight frequent superfluous spending, which we define as consumers' perceptions that they are unnecessarily and commonly spending more than saving (e.g., going out to eat instead of cooking at home). To instill these perceptions, we drew on previous research suggesting that, when answering surveys, respondents extract information about themselves from their placement on rating scales ([45]), as they assume that the middle of a rating scale refers to the "average" or "usual" frequency of a behavior. Thus, those who respond toward the extreme end of the scale (i.e., far right or far left) infer that they are more or less extreme on that particular dimension ([38]; [45]; [46]).
The current intervention required participants to indicate how often they engage in five spending behaviors that are common but superfluous. We selected the response-scale anchors to ensure that the majority of participants would endorse the upper ends of the response scale and thus consider themselves to be engaging in excessive superfluous spending. We present details of how we selected the items and anchors for the intervention as well as the percentage of participants falling at the upper ends of the scale in each study (Web Appendix C). Table 1 shows the intervention for each population sampled in this research (North American adults, Ugandans, and North American students). The aim of the present pilot studies was to check the assumption that this superfluous-spender intervention does in fact reduce inflated perceptions of financial responsibility in the populations studied in this research.
Graph
Table 1. Superfluous-Spender Intervention.
| Studies 1 and 4: North American Adults | Study 2: Ugandan Coffee Growers | Study 3: North American Students | Study 5: North American Adults |
|---|
| Choose to eat at a more expensive restaurant instead of cheaper one | Buy a soda or soft drink instead of drinking water | Buy something you want but do not need (e.g., clothes, accessories for dorm, etc.) instead of forgo the purchase | Go out to eat instead of cook at home |
| Buy something at full price instead of waiting for it to go on sale | Borrow money from friends/family instead of waiting and saving | Buy something at full price instead of wait for it to go on sale | Purchase a branded product instead of a generic one |
| Buy something you want instead of forgoing the purchase | Visit a local restaurant for lunch/dinner instead of cooking at home | Buy new products instead of buy used ones | Buy something to drink instead of use a water fountain |
| Purchase a more expensive brand instead of a cheaper one (e.g., store brand) | Give discretionary money to partner/children instead of using for savings | Purchase a more expensive brand instead of a cheaper one (e.g., store brand) | Buy something at full price instead of wait for it to go on sale |
| Go out to eat instead of cook at home | Catch a bodaa instead of walking to travel locally | Order takeout from a restaurant (that does not accept flex points) instead of use your meal plan | Buy a lunch instead of pack one |
| Response Scale |
| 1 = Once every 18 months (or less) | 1 = Once a year or less | 1 = Never | 1 = Never |
| 2 = Once every 15 months | 2 = 2–3 times a year | 2 = Once a month | 2 = Once a year |
| 3 = Once every 12 months | 3 = 4–5 times a year | 3 = 2 times a month | 3 = 2 times a year |
| 4 = Once every 9 months | 4 = 6–7 times a year | 4 = 3 times a month | 4 = 3 times a year |
| 5 = Once every 6 months | 5 = 8–9 times a year | 5 = 4 times a month | 5 = 4 times a year |
| 6 = Once every 3 months | 6 = 10–11 times a year | | 6 = 5 times a year |
| 7 = Once a month (or more) | 7 = 12+ times a year | | 7 = 6+ times a year |
- 50022242920979650 a Ugandan motorbike.
- 60022242920979650 Notes: Text in the table is exactly as it appeared to participants.
In one test, 100 students from a North American university (Mage = 20 years; 62% female) were randomly assigned to either the superfluous-spender condition or the baseline condition (used in Studies 1 and 3). In the baseline condition, participants did not answer any questions about their financial behaviors before completing the financial responsibility items.
In another test, 59 coffee growers from Uganda were randomly assigned to either the superfluous-spender condition or the control condition (used in Studies 2 and 5). In the control condition, participants reported how frequently they engaged in the same five superfluous spending behaviors. However, instead of using the response scale to indicate their answer, control participants reported their engagement in an open-ended format. In this way, it is less likely that they made inferences about themselves from their (superfluous) spending.
After the manipulation, all participants indicated their agreement with the following three statements: "I feel good about how I manage my money," "I manage my money responsibly," and "I am a financially responsible person" (1 = "not at all," and 7 = "very much so"). These items were related (North America: α =.88; Uganda: α =.82), so we combined them into an index of financial responsibility.
First, consistent with Pilot Study 1, the majority of North American and Ugandan participants held unrealistically positive perceptions of their financial responsibility, as evidenced by their scores falling higher than the midpoint in each sample (North America: M = 4.97, SD = 1.18; t(99) = 8.18, p <.001; Uganda: M = 6.42, SD =.68; t(58) = 27.33, p <.001). Second, the superfluous-spender intervention diminished inflated perceptions of financial responsibility. North American participants in the superfluous-spender condition reported lower perceptions of financial responsibility (M = 4.73, SD = 1.23) compared with their counterparts in the baseline condition (M = 5.21, SD = 1.09; t(98) = 2.07, p =.042, d =.41). Similarly, Ugandans reported lower self-perceptions of financial responsibility in the superfluous-spender condition (M = 6.18; SD =.55) than the control condition (M = 6.67; SD =.72; t(57) = 2.932, p =.005, d =.76). Thus, relative to baseline or a procedure that controlled for reminders of superfluous spending, the superfluous-spender intervention tempered the positive illusion of financial responsibility.
Study 1 tested the hypothesis that diminishing inflated perceptions of financial responsibility with the superfluous-spender intervention increases saving (H2). We randomly assigned participants to one of three conditions (superfluous spender, responsible spender, or baseline) and observed the effect of the manipulation on their savings intentions. The dependent variable was the percentage of monthly income that participants were willing to put into their savings account ([21]).
We predicted that those in the superfluous-spender condition would intend to save a greater percentage of their monthly income than those in the responsible-spender and baseline conditions. Drawing on our assumption that most consumers perceive themselves to be financially responsible and thus undersave, we also hypothesized that those in the baseline condition would intend to save a similar amount to those in the responsible-spender condition.
Amazon Mechanical Turk workers from the United States (n = 296; Mage = 35 years; 46% female) completed this study for $.50. We randomly assigned all participants to one of three conditions (superfluous spender vs. responsible spender vs. baseline). To create the responsible-spender condition, we reversed the superfluous-spender items (e.g., "Cook at home instead of going out to eat"). Participants in the baseline condition did not answer any questions before completing the dependent variable.
To assess savings intent, participants indicated their monthly income, and of this amount, how much they would be willing to put in their savings account at that moment. Consistent with prior work ([21]), savings intent was the amount of money participants indicated they would save, divided by their monthly income (M = 26.0%, SD = 21.8%).[ 6] Finally, participants completed an attention check measure ("Please tick 'Disagree' to show that you are paying attention") and demographic questions.
Before conducting the analyses, we excluded 12 participants (4% of the sample) because they ( 1) failed the attention check, ( 2) indicated they would save more money than they make, or ( 3) indicated an unrealistic monthly income (e.g., $2). Two hundred eighty-four participants remained in the final sample.
An ANOVA revealed the predicted main effect of the experimental manipulation on saving (F( 2,281) = 3.88, p =.022, η2 =.027). Planned contrasts showed that participants in the superfluous-spender condition saved a significantly greater percentage of their monthly income (M = 30.6%, SD = 26.5%) than did their counterparts in the responsible-spender condition (M = 21.9%, SD = 16.4%; t(281) = 2.74, p =.007, d =.39). Furthermore, participants in the superfluous-spender condition saved a marginally greater percentage of their monthly income (M = 30.6%, SD = 26.5%) than those in the baseline condition (M = 24.8%, SD = 20.9%; t(281) = 1.83, p =.068, d =.24). Consistent with the notion that people are inclined to see themselves as financially responsible and thus undersave, those in the baseline condition saved a similar amount to those in the responsible-spender condition (t(281) =.93, p =.35).
The results of Study 1 support the idea that inflated perceptions of financial responsibility contribute to undersaving. Offsetting the positive illusion of financial responsibility with the superfluous-spender intervention increased people's inclination to save money. In addition, baseline participants saved a similar amount to those induced to feel financially responsible. Participants in both the baseline and financially responsible conditions saved less than the superfluous-spender condition, casting doubt on the possibility that perceptions of financial responsibility generally encourage consistent behavior and thus saving. In Study 2, we tested whether our intervention could sufficiently shift actual savings behavior over time.
Chronically poor people who live in rural areas face many barriers to saving. Nevertheless, they do want to increase their savings ([ 3]; [32]). We tested the effectiveness of the superfluous-spender intervention among coffee growers in the Mount Elgon region of eastern Uganda. The income of respondents in this sample mainly comes from agriculture, so their income and expenditures fluctuate throughout the year, making it particularly important for them to save (rather than spend) surplus money.
After recording their daily savings and income in a financial diary for one week, participants completed either the superfluous-spender intervention or the control procedure. Participants completed the financial diary for another two weeks after the intervention, allowing us to test whether the superfluous-spender intervention (vs. the control procedure) causes people to save more of their household income than they would otherwise.
We recruited 250 coffee growers from the Mount Elgon region of eastern Uganda. In exchange for their participation, they received a small monetary payment at the end of the study. Four participants could not be located at the start of the study, which meant that 246 participants undertook the study (57 women). The median weekly household income was approximately 56,000 Ugandan shillings (approximately US$15).
This study was a 2 (experimental condition: superfluous-spender vs. control) × 2 (time of measurement: pre- vs. postintervention) mixed-measures design. Before the enumerators began to visit households (i.e., before the study began), we assigned households an identification number and randomly assigned each one to either the superfluous-spender or the control condition. We aimed to measure preintervention savings for one week and postintervention savings for two weeks. For logistical reasons, enumerators could not reach each household on the same day. The most common length of the preintervention period was 8 days; the most common length of the postintervention period was 13 days.
The enumerator introduced the study to participants as a money management study; we did not tell participants that we were interested in savings specifically. As an additional precaution against experimental demand, we embedded the target savings measure within a larger financial diary.
To measure savings, all participants recorded their daily savings and income in a monetary diary both before and after the intervention. Consistent with Study 1, we measured savings as the percentage of income set aside for saving ([21]). Because we recorded daily data in this study, we analyzed savings at the daily level (i.e., daily savings divided by daily income).
In both conditions, the enumerator read aloud each of the five superfluous spending behaviors and asked respondents to take a moment to think of how many times they engaged in that behavior every year (see Table 1). In the superfluous-spender condition, the enumerator held up a response scale (Web Appendix E); respondents indicated their answer by pointing to their response on the scale. In the control condition, participants verbally reported how many times they engaged in the behavior. In both conditions, the enumerator recorded participants' responses, thereby holding constant social desirability concerns. Although the enumerators were not blind to participants' experimental condition, they were blind to the hypotheses of this study.
Fifteen participants did not complete the diary for various reasons (i.e., dropped out of the study for a variety of reasons such as moving away for work), leaving a sample of 231 respondents. Sixteen percent attrition is a relatively low rate of attrition for a mobile population with limited means of communication such as this one.
Initial inspection of the data revealed a severe outlier for both income and savings (i.e., 40+ standard deviations from the means). We excluded this household from all analyses, leaving 230 respondents. In total, we collected 4,640 unique daily savings observations and 4,640 unique daily income observations.
Given the nested nature of the data (i.e., daily savings across multiple days for each household), we used multilevel modeling to test our hypotheses. Before computing our measure of savings, we checked to ensure that household income did not differ by experimental condition, time of measurement, or the interaction between the two. It did not, thus confirming the effectiveness of random assignment (ps >.248). Therefore, we created our dependent measure of daily percentage saved by dividing daily amount saved by daily amount earned (i.e., percentage saved; [21]). We report effect sizes using standardized coefficients, which were obtained by standardizing the dependent measure ([19]).
To test the hypothesis that the superfluous-spender intervention increases daily saving (H2), we entered experimental condition (superfluous spender vs. control), time of measurement (preintervention vs. postintervention), and the interaction between the two as fixed predictors in a multilevel model predicting daily percentage saved. This model revealed the anticipated interaction between experimental condition and time of measurement (β =.148, B =.065, SE =.025, p =.010; see Figure 1). In the main model, there was a main effect of time of measurement (β =.155, B =.068, SE =.017, p <.001); experimental condition was not a significant predictor of daily percentage saved (β = −.019, B = −.008, SE =.024, p =.723).
Graph: Figure 1. Pre- and posttreatment savings in Study 2.*p =.013.***p <.001.Notes: This figure depicts the interaction between experimental condition and time of savings measurement when predicting daily percentage saved (daily amount saved divided by daily amount earned) in a multilevel model. Error bars represent ±1 SEs.
Conceptually replicating Study 1 with actual savings and a different population, participants who received the superfluous-spender intervention (vs. control procedure) saved more of their daily income in days following the intervention (β =.123, B =.057, SE =.022, p =.013). The opposite pattern was apparent prior to the intervention, although only descriptively so (β = −.167, B = −.074, SE =.043, p =.089).
Looked at a different way, participants in the superfluous-spender condition saved more of their daily income after the intervention than before the intervention (β =.303, B =.134, SE =.018, p <.001). In this way, the intervention caused coffee growers to save more of their income than they otherwise would. In contrast, and consistent with our expectations, participants in the control condition saved similar amounts of their daily income in the pre- and postintervention time periods (β =.007, B =.003, SE =.040, p =.935).
As a conceptual check, we ensured that our intervention was most effective among those who endorsed the upper ends of the superfluous-spender scale. This was the case, suggesting that the effect of the intervention is due to inferences made from scoring high on the scale specifically rather than reminders of superfluous spending more generally. For full details, see Web Appendix E.
Conceptually replicating and extending Study 1, Study 2 found that diminishing the positive illusion of financial responsibility with the superfluous-spender intervention (vs. control procedure) increased saving of earned income among chronically poor coffee growers in Uganda. In this field study, treatment participants showed a continued increase in savings of their earned income after the intervention, up to 13 days' time. A follow-up field study extended this basic result to savings of a monetary windfall among a different group of Ugandan coffee growers (for complete details, see Web Appendix F). Given the simplicity of this intervention, especially compared with far more complex ones, these studies suggest that the superfluous-spender intervention has a potentially meaningful effect on people's actual saving.
It is worth noting that over the course of the two-week period after the intervention, participants in Study 2 had other means to bolster their positive self-views. Nevertheless, participants in the superfluous-spender condition saved more than those in the control condition, ostensibly because saving enabled them to feel financially responsible. This result suggests that the superfluous-spender intervention leads people to restore threatened self-views of financial responsibility specifically (i.e., through saving). We test this aspect of our model more directly in Study 3.
We designed Study 3 to achieve two goals. The first was to test our proposed psychological process. More specifically, we hypothesized that the superfluous-spender intervention (vs. baseline) motivates people to restore their diminished perceptions of financial responsibility, which in turn causes them to save more money (H3). We therefore assessed this motivation and evaluated its viability as a statistical mediator.
The second was to evaluate an alternative model that suggests that, in response to threats, people can restore general feelings of self-worth ([50]; [53]). To evaluate our proposed model against the alternative model, we gave participants the opportunity to allocate a $200 lottery between saving and spending opportunities that would enable them to restore general feelings of self-worth—namely, money on something nice for someone special or donating money to charity. This lottery was executed after the completion of the study, enabling us to assess a consequential savings decision. We hypothesized that participants in the superfluous-spender condition (vs. baseline) would save more money to restore diminished self-perceptions of financial responsibility. This study presents a conservative test of our hypothesis given that people tend to treat windfalls as disposable income ([10]).
Two hundred five North American undergraduate students (Mage = 19 years; 57% female) completed this paper-and-pencil study for the chance to win $200. We randomly assigned participants to the superfluous-spender condition or the baseline condition. Those in the superfluous-spender condition answered five questions before completing the lottery allocation task (see Table 1), whereas those in the baseline condition did not complete any questions prior to this task. No participants were excluded from analyses.
To make this study incentive compatible, we told all participants that if they won the lottery, they could allocate their winnings between the following four options: ( 1) an Amazon gift card for themselves, ( 2) money for their savings account, ( 3) an Amazon gift card for someone special, or ( 4) a donation to the charity of their choice. Pretest results from a separate set of participants (presented in Web Appendix G) indicated that putting money into a savings account was considered the most financially responsible option, whereas spending money on others or donating money to charity were considered the best ways to bolster general feelings of self-worth. Participants could allocate the entire $200 to a single option or they could split the $200 however they wished among the four options, but we stressed that their decision was final and unchangeable if they won. We executed the lottery one week later.
After the lottery allocation, we measured the putative mediator, the desire to restore perceptions of financial responsibility. All participants were directed to focus specifically on the amount of money they indicated they would save when answering three questions: "To what extent was your savings decision based on..." ( 1) "your desire to feel more financially responsible?," ( 2) "your desire to take control of your personal finances?," and ( 3) "the feeling that you have been financially irresponsible?" (1 = "not at all," and 7 = "very much"). These three items were related, so we averaged them to create an index of desire to restore perceptions of financial responsibility (α =.78).
We first tested the hypothesis that the intervention will affect how participants decided to allocate their winnings among the four categories. A multivariate ANOVA confirmed that it did. Consistent with theorizing and the results of Studies 1 and 2, participants in the superfluous-spender condition committed to saving more money (M = $110.25, SD = $81.29) than did participants in the baseline condition (M = $82.61, SD = $73.12; F( 1, 203) = 6.56, p =.011, η2 =.031). Conversely, those in the baseline condition committed to spending more money both on themselves (M = $42.71, SD = $61.05) and on someone special (M = $21.06, SD = $37.39) compared with those in the superfluous-spender condition (Mspend_self = $24.25, SD = $56.91; Mspend_other = $7.85, SD = $21.43; Fspend_self( 1, 203) = 5.01, p =.026, η2 =.024; Fspend_other( 1, 203) = 9.50, p =.002, η2 =.045). There was no significant difference in the amount of money participants wanted to donate to charity based on condition (F( 1, 203) =.21, p =.65).
Next, we tested our prediction that the desire to restore perceptions of financial responsibility is statistically responsible for the effect of the superfluous-spender intervention on increased saving. Consistent with our predictions, those in the superfluous-spender condition indicated a greater desire to restore their perceptions of financial responsibility (M = 3.95, SD = 1.72) than those in the baseline condition (M = 3.44, SD = 1.49; t(203) = 2.53, p =.025, d =.32). Perhaps more importantly, participants' desire to restore their perceptions of financial responsibility mediated the effect of the superfluous-spender intervention (vs. baseline) on saving, as the 95% confidence interval with 10,000 bootstrap resamples excluded zero (1.99, 29.79).
Two pieces of evidence provide clear support for our theory that the superfluous-spender intervention increase saving because it motivates people to restore diminished perceptions of financial responsibility. First, the effect of the superfluous-spender intervention on increased savings was statistically explained by self-reported motivation to restore perceptions of financial responsibility. Second, the intervention caused people to engage in actions that would restore self-perceptions of financial responsibility specifically (i.e., saving money), as opposed to restoring self-worth more generally (i.e., spending on others or donating money to a charity). In Studies 4 and 5, we aim to provide further evidence for our proposed process by following a moderation-of-process approach ([52]), which is a particularly effective tool for examining psychological processes to establish causal chains.
In Study 4, we sought further support for our proposed process by altering the diagnostic value of the superfluous-spender intervention. More specifically, our findings rest on the notion that when participants observe themselves responding at the upper ends of the scale, they infer that they are less financially responsible than they ought to be and are thus not living up to their desired standards. However, people only make inferences about themselves when they observe behaviors that are perceived to be freely chosen ([ 7]). In this way, endorsing instances of superfluous spending that were outside one's control (e.g., "About how often do you have to go out to eat, instead of cook at home") should not offset the positive illusion of financial responsibility and hence should not facilitate saving (H4).
Six hundred two Mechanical Turk workers from the United States (Mage = 36 years; 45% female) completed the main study for $.30. We randomly assigned participants to one of three conditions (superfluous-spender vs. responsible-spender vs. forced-superfluous-spender).
To create the forced-superfluous-spender condition, we modified the phrasing of the superfluous-spender items chosen for this sample by emphasizing instances in which consumers had to spend money. A pretest confirmed that participants in the forced-superfluous-spender condition reported less perceived control over these financial behaviors compared with those in the superfluous-spender condition and responsible-spender condition (for full details, see Web Appendix H). In all three conditions, participants indicated how frequently they engaged in each behavior using the same seven-point scale (see Table 1).
To assess savings intent, participants imagined that they received $100; of this amount, they indicated how much they would put in their savings account at that moment ([21]). The main dependent variable was the amount of money participants indicated they would save (M = $51.03, SD = $33.42). Finally, participants completed an attention check ("Please tick 'Disagree' to show that you are paying attention") and demographic questions.
Before conducting the analyses, we excluded 23 participants (3.8% of the sample) because they failed the attention check. This left us with 579 participants in the final sample.
An omnibus ANOVA revealed the anticipated main effect of experimental condition on saving intentions (F( 2,576) = 3.97, p =.019, η2 =.014). Conceptually replicating the finding from Study 1, participants in the superfluous-spender condition intended to save significantly more of their $100 windfall (M = $57.06, SD = $31.58) than did their counterparts in the responsible-spender condition (M = $49.20, SD = $34.71; t(576) = 2.33, p =.020, d =.24). Next, in support of our hypothesis that the superfluous-spender intervention increases saving intentions only when the spending behaviors are under one's control and thus diagnostic of the self (H4), participants in the superfluous-spender condition reported higher savings intentions than participants in the forced-superfluous-spender condition (M = $48.61, SD = $32.88; t(576) = 2.52, p =.012, d =.26).
The results of Study 4 conceptually replicated and extended the results of the previous studies. When the superfluous-spending behaviors were perceived to be under one's control and thus were diagnostic of one's irresponsible spending, the superfluous-spender intervention increased savings intentions as compared with a control condition (in this case, a responsible-spending comparison condition). In contrast, and in support of our theorizing, when the superfluous spending behaviors were perceived to be relatively less under one's control, and therefore were relatively less diagnostic of one's irresponsible spending, the superfluous-spender intervention did not increase savings intentions, arguably because perceptions of financial responsibility remained intact.
Study 5 examined the moderating role of an individual difference variable that predicts variation in the desire to perceive oneself as financially responsible. We hypothesized that individuals who believe that happiness comes from acquiring material goods (i.e., materialistic happiness, a subfactor of materialism) will be less inclined to hold the positive illusion of financial responsibility. A pretest confirmed that inverse relationship: the more individuals derive happiness from material goods, the less they hold the positive illusion of financial responsibility (r(55) = −.286, p =.033; for complete details, see Web Appendix I).
When people's self-views are threatened, they try to restore those threatened self-views specifically, but only when the self-views are important ([24]). If the superfluous-spender intervention increases saving because it spurs people to restore self-views of financial responsibility, then this effect should only be apparent among those who are motivated to view themselves that way (i.e., those who score low on materialistic happiness).
Six hundred fifty Prolific Academic workers from the United States (57% female; Mage = 34 years) completed this study in exchange for $.70. We randomly assigned participants to the superfluous-spender condition (see Table 1) or the control condition. Those in the control condition reported their frequency of engaging in the same five superfluous spending behaviors in an open-ended format (by typing their answer into a textbox; see Pilot Study 2).
To assess savings intentions, we relied on the same measure used in Study 1—the amount of money participants indicated they would save divided by their monthly income (M = 23.75%, SD = 20.82%). Participants then completed an attention check measure (i.e., they were instructed to type "none" in the question box "What is today's date?").
After the attention check, participants completed the 15-item materialism scale ([42]). Even though we were only interested in the happiness-from-materialism subscale, we administered the whole scale given that the happiness subscale was validated within the larger scale. The items for the focal subscale showed acceptable reliability (happiness α =.83).
Before conducting the analyses, we excluded the following participants using the same criteria applied in Study 1 (which used the same dependent variable): 87 participants (13% of the sample) for failing the attention check, 1 participant (.2% of the sample) for saving more money than they earn, and 46 participants (7% of the sample) for reporting either no monthly income or an unrealistic monthly income. The final sample included 529 participants.
The manipulation did not affect scores on the happiness from materialism scale (p =.882). Therefore, we proceeded with testing our hypothesis that the superfluous-spender intervention (vs. control) would increase saving intentions—but only among those scoring low on the happiness-from-materialism scale.
We regressed percentage saved on experimental condition, happiness scores, and the interaction between the two using Hayes's PROCESS macro ([28]). There was a main effect of experimental condition (β =.382, t(525) = 2.371, p =.018); consistent with Studies 1–4, participants in the superfluous-spender (vs. control) condition reported higher savings intentions. As we expected, this main effect was qualified by the predicted interaction between materialism happiness scores and experimental condition (β = −.358, t(525) = 2.146, p =.032; Figure 2). Materialism happiness scores were not significantly associated with savings intentions in this model (β =.101, t(526) = 1.630, p =.104).
Graph: Figure 2. Materialism moderator (Study 5).Notes: This figure depicts the interaction between experimental condition and individual differences in materialistic happiness when predicting percentage of monthly income intended to save (amount saved divided by monthly income) in a regression model. The vertical line depicts the Johnson–Neyman point.
Simple effects analyses yielded results that were supportive of our conceptual model. Consistent with H5, the superfluous-spender intervention (vs. control) increased savings intentions among those scoring low on materialistic happiness (−1 SD from the mean; β =.142, t(525) = 2.315, p =.021). By contrast, and consistent with our expectations, the intervention had no effect among those who scored relatively high (+1 SD from the mean) on materialistic happiness (p =.471). Johnson–Neyman analyses told a similar story. The experimental manipulation increased saving intentions among those scoring at or below 2.694 on the materialism scale (−.44 SD; 31% of the sample). The results are descriptively unchanged when we control for the other facets of materialism (for full results, see Web Appendix I).
Looked at a different way, in the control condition, materialistic-happiness scores were positively related to saving intentions, (β =.287, t(525) = 2.082, p =.038). In other words, those who were most inclined to hold the positive illusion of financial responsibility (those scoring low on materialistic happiness) reported the lowest levels of saving intentions. This relationship was then offset by the superfluous-spender intervention (β = −.085, t(525) = 1.403, p =.161).
The results of Study 5 provide two pieces of supportive evidence for our conceptual model. First, in the control condition, materialistic happiness scores were positively associated with saving intentions. Ironically, the participants who were most motivated to perceive themselves as financially responsible intended to save the least. Second, the tendency to undersave among participants who scored low on materialistic happiness was offset by instilling in them perceptions of superfluous spending. In other words, the superfluous-spender intervention (vs. control procedure) increased savings intentions among those who were most motivated to perceive themselves as financially responsible (i.e., those scoring low on materialistic happiness).
This research began with the conjecture that the tendency to view one's financial responsibility through rose-colored glasses may contribute to undersaving. If that conjecture is correct, then tempering this positive illusion of financial responsibility may paradoxically help people save more money because it should motivate them to restore this important but diminished self-view through increased saving. From poor coffee growers in Uganda to students and adults in North America, highlighting people's frequent engagement in superfluous spending (vs. various controls) increased both saving intentions and behavior (Studies 1–5). In support of our conceptual model, the superfluous-spender intervention increased saving because it motivated people to restore diminished perceptions of financial responsibility (Study 3). Furthermore, the superfluous-spender intervention increased saving intentions only when spending behaviors were perceived to be under one's control (Study 4) and among people who were motivated to perceive themselves as financially responsible (Study 5). Notably, effect sizes for the effect of the intervention on savings behavior were larger than the effect sizes for savings intentions, attesting to the fact that in real life, this intervention can make a positive difference for people's saving.
In addition to contributing a novel answer to the question of why consumers fail to save, our article also provides a complementary perspective. Although researchers have identified important reasons for undersaving, such as one's financial status ([ 9]), childhood economic environment ([27]), and cultural influences ([13]), the majority of these reasons represent relatively stable individual differences. In the current research, we identify a self-enhancing bias—the positive illusion of financial responsibility—which undermines saving. As demonstrated in this work, this bias is malleable and can be offset with a simple intervention to increase saving.
Increasing consumer saving is difficult in part because consumers' desire to buy products (especially in the heat of the moment) is often the automatic response that needs to be counteracted by the fallible psychological mechanism of self-control ([63]). Indeed, for many people, intentionally saving money is difficult because it is a cold, calculated, and controlled process, which requires one to forgo pleasures in the moment to reap rewards in the future. The superfluous-spender intervention in the current work may be effective in part because it renders saving as the automatic, gut response by eliciting the motivation to restore positive financial self-views through saving.
Western society tends to encourage and even indulge positive self-views. Thus, some may find our recommendation to temper inflated perceptions of financial responsibility to be a surprising one. While it may be tempting to think that the savings silver bullet is to make consumers feel better about themselves in the financial domain, the empirical data—from this research and others—suggest otherwise ([18]). For example, an exhaustive review of interventions that were designed to boost positive self-views to increase desirable outcomes such as academic success and healthy behaviors found no clear benefits or improved outcomes from those interventions ([ 6]). In some cases, the evidence led to the opposite conclusion, with interventions leading to undesirable consequences through the fostering of narcissism.
In line with these previous findings, the aim of this work was to develop an intervention that would encourage a relatively more accurate and realistic view of one's spending behavior. To be clear, we did not set out to make people feel bad about themselves, and the intervention did not do so. Indeed, the results of Pilot Study 2 indicate that participants who received the superfluous intervention still reported perceptions of financial responsibility that were above the midpoint of the scale. In this way, the intervention simply helped people realize that they have not been acting as financially responsibly as they could be, which in turn led them to change their behavior accordingly.
Although we did not deliver the intervention at a large scale, we do believe this could be done because of its brevity and because technology enables practitioners to deliver questions to consumers for little cost. (We advocate that policy makers obtain consent before proceeding with the intervention.) For example, commercial banks or pension funds could survey a small subset of their clients to gain insight into common and frequent superfluous spending behaviors that are under their clients' control. The results of our pretests for each population we tested (see Web Appendix C) demonstrate that this task is relatively simple and straightforward. Once the stakeholder selects these five behaviors and corresponding scale anchors, customers can be asked if they want to opt in to receiving prompts that encourage them to reflect on their past spending before making decisions about their saving. More specifically, banks could deliver the intervention when consumers are opening a savings account, setting a savings goal, or planning for retirement.
In developing countries, the intervention could be delivered through cell phones for subscribers to mobile banking initiatives or through community-based savings and loans organizations. Both savings tools are increasingly prevalent across urban and rural areas and have the potential to be delivered at a large scale. Administering this intervention in developing countries may especially be important, as an emerging policy option is to "shock" struggling households with the provision of large monetary transfers which may have conditions attached ([32]). While the intentions behind these practices are good, in reality they have less than ideal success ([ 4]). Given that the superfluous-spender intervention increased saving of a monetary windfall, the current work provides a novel monetary tool that policy makers could use to increase saving of monetary assistance.
The intervention may have unintended negative consequences if people do not have the opportunity to save money after receiving the intervention (e.g., because they do not have monetary resources at that time, because they are facing external constraints beyond their control). People who experience uncontrollable situations subsequently give up ([22]). Thus, if people feel the need to save after receiving the intervention but lack a clear means for doing so, it could potentially demotivate them from saving altogether. For this reason, we encourage those implementing the intervention to consider the timing and community external constraints (e.g., whether it is time to harvest or pay for children's education). To maximize chances of success and minimize potential downsides, the intervention should be implemented at a time when contextual or household factors make it feasible to save.
Finally, we should note that while we observed a positive increase in saving as a function of our intervention, we have not tested the effects of the intervention on people's well-being. However, we surmise that, for many, the effects of the intervention will be positive for well-being. A study on saving and well-being in 38 countries with 50,000 respondents found that as societal poverty increased, well-being decreased ([35]). Perhaps more importantly, that same study found saving to greatly improve well-being in high-poverty countries.
We believe that this research provides several opportunities for future work. First, future research could examine the processes through which the illusion of financial responsibility reduces saving. Better understanding those processes would provide fruitful insight for generating targeted interventions that can help diminish the bias and increase saving. Developing alternative routes to shifting perceptions of financial responsibility would complement and build on the effectiveness of the superfluous-spender intervention over time.
Second, our intervention is based on the assumption that saving is considered a socially desirable attribute. While we did test that assumption in two very different contexts (wealthy North America and Sub-Saharan Africa), some cultures may not view saving as a positive attribute, and other cultures may discourage self-enhancement. In those cultures, then, the current intervention is unlikely to be effective. Indeed, in an additional study we conducted, we found that the superfluous-spender intervention did not increase savings when participants were told that saving money was not socially desirable. In this way, the effectiveness of the intervention depends on the broader social norms, given that people are motivated to engage in behaviors that are socially desirable and thus reflect well on the self. Moving forward, we encourage researchers to continue to investigate the link between positive illusions and savings in other cultures. For example, it would be interesting to examine whether, at the national level, there is a relationship between self-enhancement and personal-savings rates.
Third, future research could investigate dependent variables above and beyond saving. Our conceptual model predicts that threatening perceptions of financial responsibility causes consumers to restore their threatened (financial) self-view. Although saving money is a primary action associated with financial responsibility (see pretest in Web Appendix G), other financial behaviors (e.g., getting an extra job, selling off unwanted goods) could also enable consumers to restore this particular self-view. In this way, our superfluous-spender intervention could be applied not only to saving, but to any financial behavior that enables one to restore perceptions of being financially responsible.
Finally, examining how our intervention compares with or complements other interventions aimed to increase saving could be a worthwhile investigation. Prior research has demonstrated the role of choice architecture in influencing savings decisions. More specifically, it has been shown that decreasing the number of funds offered in a 401(k) makes people more likely to participate ([48]). Similarly, presenting choices that highlight options for putting money into savings causes people to save a greater portion of their tax refund ([26]). Future research could investigate whether the administration of the superfluous-spender intervention prior to such choice curations has the potential to increase savings beyond these existing interventions.
People have a fundamental need to view themselves positively. This bias leads them to pay attention to, encode, and selectively remember information that supports, rather than disconfirms, those positive self-views. Most people's financial reality is not particularly positive or pleasant, making finance a prime target for positive illusions. Gently correcting people's overly positive illusions of financial responsibility, however, can increase both willingness to save and actual saving over time. In this way, practitioners have the power to improve consumers' long-term well-being by highlighting past instances in which they engaged in regular superfluous spending.
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242920979647 - Popping the Positive Illusion of Financial Responsibility Can Increase Personal Savings: Applications in Emerging and Western Markets
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242920979647 for Popping the Positive Illusion of Financial Responsibility Can Increase Personal Savings: Applications in Emerging and Western Markets by Emily N. Garbinsky, Nicole L. Mead and Daniel Gregg in Journal of Marketing
Footnotes 1 Amitava Chattopadhyay
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The Uganda studies in this article were undertaken as part of the Value Chains for Innovation Platforms project funded by the Australian Centre for International Agricultural Research (project ID #FST/2014/093). This research was also funded by a Social Sciences and Humanities Research Council of Canada (SSHRC) grant awarded to NL Mead (Grant #430-2020-00829).
4 Emily N. Garbinsky https://orcid.org/0000-0002-9781-0063
5 Online supplement: https://doi.org/10.1177/0886109920979647
6 Because this was the first study we conducted in July 2017, we also assessed two potential alternative mechanisms and one possible moderator, none of which were statistically significant. For a complete list of measures, please refer to Web Appendix D.
References Alicke Mark D., Govorun Olesya. (2005), "The Better-Than-Average Effect," in The Self in Social Judgment, Alicke Mark D., Dunning David A., Krueger Justin I., eds. New York: Psychology Press, 85–106.
Aronson Elliot, Carlsmith J. Merrill. (1963), "Effect of the Severity of Threat on the Devaluation of Forbidden Behavior," Journal of Abnormal and Social Psychology, 66 (6), 584–88.
Banerjee Abhijit V., Duflo Esther. (2007), "The Economic Lives of the Poor," Journal of Economic Perspectives, 21 (1), 141–68.
Banerjee Abhijit, Duflo Esther, Glennerster Rachel, Kinnan Cynthia. (2015), "The Miracle of Microfinance? Evidence from a Randomized Evaluation," American Economic Journal: Applied Economics, 7 (1), 22–53.
Baumeister Roy F. (1998), "The Self," in The Handbook of Social Psychology, Vol. 1, Fiske Susan T., Gilbert Daniel T., Lindzey Gardner, eds. Hoboken, NJ: John Wiley & Sons, 680–740.
Baumeister Roy F., Campbell Jennifer D., Krueger Joachim I., Vohs Kathleen D. (2003), "Does High Self-Esteem Cause Better Performance, Interpersonal Success, Happiness, or Healthier Lifestyles?" Psychological Science in the Public Interest, 4 (1), 1–44.
7 Bem Daryl J. (1972), "Self-Perception Theory," Advances in Experimental Social Psychology, 6, 1–62.
8 Bernartzi Shlomo. (2020), "People Don't Save Enough for Emergencies, but There Are Ways to Fix That," The Wall Street Journal(February 17), https://www.wsj.com/articles/people-dont-save-enough-for-emergencies-but-there-are-ways-to-fix-that-11581951601.
9 Bertrand Marianne, Mullainathan Sendhil, Shafir Eldar. (2006), "Behavioral Economics and Marketing in Aid of Decision Making Among the Poor," Journal of Public Policy & Marketing, 25 (1), 8–23.
Carlsson Fredrik, He Haoran, Martinsson Peter. (2013), "Easy Come, Easy Go," Experimental Economics, 16 (2), 190–207.
Carver Charles S. (2001), On the Self-Regulation of Behavior. Cambridge, UK: Cambridge University Press.
Carver Charles S., Scheier Michael F. (1998), On the Self-Regulation of Behavior. Cambridge, UK: Cambridge University Press.
Chen M. Keith. (2013), "The Effect of Language on Economic Behavior: Evidence from Savings Rates, Health Behaviors, and Retirement Assets," American Economic Review, 103 (2), 690–731.
Cooper Joel, Fazio Russell H. (1984), "A New Look at Dissonance," Advances in Experimental Social Psychology, 17, 229–68.
Critcher Clayton R., Helzer Erik G., Dunning David. (2011), "Self-Enhancement Via Redefinition: Defining Social Concepts to Ensure Positive Views of the Self," in Handbook of Self-Enhancement and Self-Protection, Alicke M.D., Sedikides C., eds. New York: Guilford Press, 69–91.
Dunn Elizabeth W., Aknin Lara B., Norton Michael I. (2008), "Spending Money on Others Promotes Happiness," Science, 319 (5870), 1687–88.
Dunn Elizabeth W., Norton Michael I. (2014), Happy Money: The Science of Happier Spending. New York: Simon and Schuster.
Fernandes Daniel, John G. LynchJr, Netemeyer Richard G. (2014), "Financial Literacy, Financial Education, and Downstream Financial Behaviors," Management Science, 60 (8), 1861–83.
Ferron John M., Hogarty Kristin Y., Dedrick Robert F., Hess Melinda R., Niles Jonhn D., Kromrey Jeffrey D. (2008), "Reporting Results from Multilevel Analyses," in Multilevel Modeling of Educational Data, O'Connell Ann A., Betsy McCouch D., eds. Charlotte, NC: Information Age Publishing Inc., 391–426.
Festinger Leon. (1957), A Theory of Cognitive Dissonance, Vol. 2. Stanford, CA: Stanford University Press.
Garbinsky Emily N., Klesse Anne Kathrin, Aaker Jennifer. (2014), "Money in the Bank: Feeling Powerful Increases Saving," Journal of Consumer Research, 41 (3), 610–23.
Glass David C., Singer Jerome E., Friedman Lucy N. (1969), "Psychic Cost of Adaptation to an Environmental Stressor," Journal of Personality and Social Psychology, 12 (3), 200–210.
Gollwitzer Peter M., Kirchhof Oliver. (1998), The Willful Pursuit of Identity. New York: Cambridge University Press.
Gollwitzer Peter M., Marquardt Michael K., Scherer Michaela, Fujita Kentaro. (2013), "Identity-Goal Threats: Engaging in Distinct Compensatory Efforts," Social Psychological and Personality Science, 4 (5), 555–62.
Graham Luke. (2017), "Europeans Aren't Saving Enough, Warns ING Report," CNBC(January 24), https://www.cnbc.com/2017/01/24/europeans-arent-saving-enough-warns-ing-report.html.
Grinstein-Weiss Michal, Cryder Cynthia, Despard Mathieu R., Perantie Dana C., Oliphant Jane E., Ariely Dan. (2017), "The Role of Choice Architecture in Promoting Saving at Tax Time: Evidence From a Large-Scale Field Experiment," Behavioral Science & Policy, 3 (2), 20–38.
Griskevicius Vladas, Ackerman Joshua M., Cantú Stephanie M., Delton Andrew W., Robertson Theresa E., Simpson Jeffry A., et al. (2013), "When the Economy Falters, Do People Spend or Save? Responses to Resource Scarcity Depend on Childhood Environments," Psychological Science, 24 (2), 197–205.
Hayes Andrew F. (2013), Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach. New York: Guilford Press.
Higgins E. Tory. (1997), "Beyond Pleasure and Pain," American Psychologist, 52 (12), 1280–300.
Horowitz Leonard M., Turan Bulent. (2008), "Prototypes and Personal Templates: Collective Wisdom and Individual Differences," Psychological Review, 115 (4), 1054–68.
Johnson Angela. (2013), "76% of Americans Are Living Paycheck-to-Paycheck," CNN(June 24), http://money.cnn.com/2013/06/24/pf/emergency-savings/index.html.
Karlan Dean, Ratan Aishwarya Lakshmi, Zinman Jonathan. (2014), "Savings by and for the Poor: A Research Review and Agenda," Review of Income and Wealth, 60 (1), 36–78.
Kruglanski Arie W. (1996), "Goals as Knowledge Structures," in Linking Cognition and Motivation to Behavior: The Psychology of Action, Gollwitzer Peter M., Bargh John A., eds. New York: Guilford Press, 599–618.
Locke Edwin A., Latham Gary P. (2002), "Building a Practically Useful Theory of Goal Setting and Task Motivation: A 35-Year Odyssey," American Psychologist, 57 (9), 705–17.
Martin Kelly D., Hill Ronald Paul. (2015), "Saving and Well-Being at the Base of the Pyramid: Implications for Transformative Financial Services Delivery," Journal of Service Research, 18 (3), 405–21.
Mead Nicole L., Patrick Vanessa M. (2016), "The Taming of Desire: Unspecific Postponement Reduces Desire For and Consumption of Postponed Temptations," Journal of Personality and Social Psychology, 110 (1), 20–35.
Moskowitz Gordon B., Li Peizhong, Ignarri Courtney, Stone Jeff. (2011), "Compensatory Cognition Associated with Egalitarian Goals," Journal of Experimental Social Psychology, 47 (2), 365–70.
Nelson Leif D., Morrison Evan L. (2005), "The Symptoms of Resource Scarcity: Judgments of Food and Finances Influence Preferences for Potential Partners," Psychological Science, 16 (2), 167–73.
Olson Jenny, Rick Scott. (2017), "A Penny Saved Is a Partner Earned: The Romantic Appeal of Savers," working paper, SSRN, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2281344.
Oppenheimer Daniel M., Meyvis Tom, Davidenko Nicolas. (2009), "Instructional Manipulation Checks: Detecting Satisficing to Increase Statistical Power," Journal of Experimental Social Psychology, 45 (4), 867–72.
Papp Lauren M., Cummings E. Mark, Goeke-Morey Marcie C. (2009), "For Richer, for Poorer: Money as a Topic of Marital Conflict in the Home," Family Relations, 58 (1), 91–103.
Richins Marsha L. (1994), "Special Possessions and the Expression of Material Values," Journal of Consumer Research, 21 (3), 522–33.
Richins Marsha L. (2004), "The Material Values Scale: Measurement Properties and Development of a Short Form," Journal of Consumer Research, 31 (1), 209–19.
Richins Marsha L., Dawson Scott. (1992), "A Consumer Values Orientation for Materialism and Its Measurement: Scale Development and Validation," Journal of Consumer Research, 19 (3), 303–16.
Schwarz Norbert. (1999), "Self-Reports: How the Questions Shape the Answers," American Psychologist, 54 (2), 93–105.
Schwarz Norbert, Hippler Hans J., Deutsch Brigitte, Strack Fritz. (1985), "Response Scales: Effects of Category Range on Reported Behavior and Comparative Judgments," Public Opinion Quarterly, 49 (3), 388–95.
Sedikides Constantine, Alicke Mark D. (2012), "Self-Enhancement and Self-Protection Motives," in The Oxford Handbook of Human Motivation, Ryan R.M., ed. New York: Oxford University Press.
Sethi-Iyengar Sheena, Huberman Gur, Jiang Wei. (2004), "How Much Choice Is Too Much? Contributions to 401 (k) Retirement Plans," Pension Design and Structure: New Lessons From Behavioral Finance, 83, 84–87.
Shah James Y., Kruglanski Arie W. (2000), "Aspects of Goal Networks: Implications for Self-Regulation," in Handbook of Self-Regulation, Boekaerts M., Pintrich P.R., Zeidner M., eds. New York: Elsevier, 85–110.
Sherman David K., Cohen Geoffrey L. (2006), "The Psychology of Self-Defense: Self-Affirmation Theory," Advances in Experimental Social Psychology, 38, 183–242.
Sherman Steven J., Gorkin Larry. (1980), "Attitude Bolstering When Behavior Is Inconsistent with Central Attitudes," Journal of Experimental Social Psychology, 16 (4), 388–403.
Spencer Steven J., Zanna Mark P., Fong Geoffrey T. (2005), "Establishing a Causal Chain: Why Experiments Are Often More Effective Than Mediational Analyses in Examining Psychological Processes," Journal of Personality and Social Psychology, 89 (6), 845–51.
Steele Claude M. (1988), "The Psychology of Self-Affirmation: Sustaining the Integrity of the Self," Advances in Experimental Social Psychology, 21 (2), 261–302.
Stone Jeff, Wiegand Andrew W., Cooper Joel, Aronson Elliot. (1997), "When Exemplification Fails: Hypocrisy and the Motive for Self-Integrity," Journal of Personality and Social Psychology, 72 (1), 54–65.
Sussman Abigail B., O'Brien Rourke L. (2016), "Knowing When to Spend: Unintended Financial Consequences of Earmarking to Encourage Savings," Journal of Marketing Research, 53 (5), 790–803.
Tajfel Henri, Turner John C. (1979), "An Integrative Theory of Intergroup Conflict," in Organizational Identity: A Reader, Hatch Mary Jo, Schultz Majken, eds. Oxford, UK: Oxford University Press, 56–65.
Tanner Robin J., Carlson Kurt A. (2009), "Unrealistically Optimistic Consumers: A Selective Hypothesis Testing Account for Optimism in Predictions of Future Behavior," Journal of Consumer Research, 35 (5), 810–22.
Taylor Shelley E. (1994), "Positive Illusions and Well-Being Revisited: Separating Fact from Fiction," Psychological Bulletin, 116 (1), 21–27.
Taylor Shelley E., Armor David A. (1996), "Positive Illusions and Coping with Adversity," Journal of Personality, 64 (4), 873–98.
Taylor Shelley E., Brown Jonathon D. (1988), "Illusion and Well-Being: A Social Psychological Perspective on Mental Health," Psychological Bulletin, 103 (2), 193–210.
Taylor Shelley E., Collins Rebecca L., Skokan Laurie A., Aspinwall Lisa G. (1989), "Maintaining Positive Illusions in the Face of Negative Information: Getting the Facts Without Letting Them Get to You," Journal of Social and Clinical Psychology, 8 (2), 114–29.
Thompson Derek. (2016), "Why Don't Americans Save More Money?" The Atlantic(April 19), https://www.theatlantic.com/business/archive/2016/04/why-dont-americans-save-money/478929/.
Vohs Kathleen D., Faber Ronald J. (2007), "Spent Resources: Self-Regulatory Resource Availability Affects Impulse Buying," Journal of Consumer Research, 33 (4), 537–47.
Zanna Mark P., Cooper Joel. (1974), "Dissonance and the Pill: An Attribution Approach to Studying the Arousal Properties of Dissonance," Journal of Personality and Social Psychology, 29 (5), 703–09.
~~~~~~~~
By Emily N. Garbinsky; Nicole L. Mead and Daniel Gregg
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 101- Portraying Humans as Machines to Promote Health: Unintended Risks, Mechanisms, and Solutions. By: Weihrauch, Andrea; Huang, Szu-Chi. Journal of Marketing. May2021, Vol. 85 Issue 3, p184-203. 20p. 1 Diagram, 1 Chart, 3 Graphs. DOI: 10.1177/0022242920974986.
- Database:
- Business Source Complete
Portraying Humans as Machines to Promote Health: Unintended Risks, Mechanisms, and Solutions
To fight obesity and educate consumers on how the human body functions, health education and marketing materials often highlight the importance of adopting a cognitive approach to food. One strategy employed to promote this approach is to portray humans as machines. Five studies (and three replication and follow-up studies) using different human-as-machine stimuli (internal body composition, face, appearance, and physical movement) revealed divergent effects of human-as-machine representations. While these stimuli promoted healthier choices among consumers who were high in eating self-efficacy, they backfired among consumers who were low in eating self-efficacy (measured in Studies 1 and 3–5; manipulated in Study 2). This reversal happened because portraying humans as machines activated consumers' expectation of adopting a cognitive, machine-like approach to food (Studies 3 and 4)—an expectation that was too difficult to meet for those with low (vs. high) eating self-efficacy. We tested a solution to accompany human-as-machine stimuli in the field (Study 5): we externally enhanced how easy and doable it was for consumers low in eating self-efficacy to adopt a cognitive approach to food, which effectively attenuated the backfire effect on their lunch choices at a cafeteria.
Keywords: artificial intelligence; dehumanization; eating self-efficacy; health; human-as-machine representations; machine; (performance) expectation
More than two-thirds of adults and one-third of preschoolers in the United States are overweight or obese ([20]); similar rates exist in many other countries worldwide ([105]). To combat obesity, governments, marketers, and consumer welfare organizations invest a substantial amount of resources to encourage consumers to make food choices in a cognitive manner and to use their head instead of their heart (e.g., "Eat to fuel your body, not to feed your emotions"). These cognitive, head-based approaches to food such as reading nutrition labels and computing calories are believed to be optimal health strategies ([31]; [105]). Accordingly, major health interventions and programs have invested a lot of resources into promoting cognitive approaches that are analytical, rule-focused, and free of emotions ([33]; [61]; [81]; [85]).
One popular strategy employed to promote a cognitive approach to food is to portray humans as machines and to depict human body parts using mechanistic components. A wide variety of examples can be found in the recent campaigns by the American Heart Association, the Centers for Disease Control and Prevention, Men's Health Week, and GBCHealth (for a list of recent health campaigns using human-as-machine stimuli, see Web Appendix 1). These materials are aimed to leverage people's existing associations about machines—that machines make decisions using their head (cognition) and not their heart (emotion)—to help consumers approach food in a more cognitive, machine-like manner, with the goal of encouraging healthier choices. National Geographic's series "The Incredible Human Machine" even describes unhealthy behaviors as (human) "errors" in the maintenance of our bodily machine.
Similarly, companies and marketers have started using human-as-machine representations. For instance, Centrum asks consumers to "power the human machine" with healthy food supplements. Nestlé encourages indulgence as the "human" (vs. a machine-like) thing to do; their tagline "Working like a machine? Have a Kit Kat" motivates consumers to be more like humans and have a chocolate bar—a choice that rational machines would not make (for additional examples by Snickers, Red Bull, Anheuser-Busch, and others, see Web Appendix 1).
Furthermore, consumers experience human-as-machine representations not only in targeted advertisements but also in everyday life. With rapid improvements in technology, virtual telepresence systems show people as human faces with mechanistic bodies, human enhancement technologies (e.g., augmented reality goggles, transcranial simulation headbands) represent humans as more machine-like, and artificial intelligence software further blurs the line between humans and machines ([19]; [68]; [70]). These technological advances are entering the retail and restaurant sectors ([78]) and are driving consumption decisions.
Despite the existence of human-as-machine representations in public policy, education, food marketing, and consumers' daily lives, research has yet to systematically examine how consumers react to such representations. The previous examples suggest a possible lay belief among practitioners that by making humans look more like machines, people would choose food in a cognitive manner and thus make healthier choices. How accurate is this lay belief? We aim to answer three questions in this research: ( 1) Does representing humans as machines indeed encourage healthier choices? ( 2) Might there be heterogeneity in how consumers respond to these stimuli? ( 3) What psychological processes drive these effects?
To achieve this aim, we spotlight an important individual difference variable: consumers' eating self-efficacy (i.e., confidence in one's own ability to choose healthy food; also referred to as healthy eating efficacy, healthy diet efficacy, and dieting self-efficacy; [ 4]; [95]). We theorize that, contrary to practitioners' lay beliefs, human-as-machine representations could create divergent effects on consumers' food choices, depending on a person's chronic level of eating self-efficacy.
This hypothesized divergent effect is driven by the following process: ( 1) being exposed to human-as-machine stimuli brings to mind the expectation that one should behave more machine-like (i.e., adopting a cognitive, head-based approach to food); ( 2) importantly, this expectation can be motivating (i.e., leading to healthier choices) only if consumers believe that they can meet it. While consumers with high levels of eating self-efficacy believe in their abilities to choose food in a cognitive, machine-like manner (and thus would be motivated to fulfill this expectation), consumers with low levels of eating self-efficacy tend to struggle with a cognitive approach to food. As a result, this latter and more vulnerable consumer segment would anticipate failure in fulfilling this expectation, leading them to contradict it and choose unhealthier options ([14]; [15]; [17]; [86]). Because low levels of eating self-efficacy have been linked to overweight and obesity ([32]), the very segment that the human-as-machine marketing communication aims to educate is the one that does not benefit from this approach, revealing a critical dark side of these representations on consumers' well-being.
Obesity has been considered one of the most critical global crises in the twenty-first century, with detrimental health consequences to individuals as well as serious economic costs collectively ([105]). As a result of this, governments, policy makers, nongovernmental organizations, and marketers have developed a variety of materials and programs to encourage consumers to make healthier food choices. One key trend in these materials and programs is to push toward a more cognitive approach to food ([33]; [61]; [81]; [85]); for instance, to encourage consumers to choose food with their head and not their heart (e.g., "H.A.L.T. before eating," in which H.A.L.T. stands for hungry, angry, lonely, and tired) and to highlight the importance of nutrition labels and calorie tracking ([105]). The rise of artificial intelligence–based technologies and devices in the health industry ([83]) further promotes the notion that health-related decisions (such as food choices) should be based on analytics and that considering humans' unique characteristics and emotions might hinder optimal decision making ([68]).
In line with this general push toward a cognitive approach to food, one popular strategy is to portray humans as machines and depict human body parts using mechanistic components (see Web Appendix 1). This process of altering humans' physical dimensions to make humans look more like machines is conceptualized as mechanistic dehumanization ([42]; [44]) and can be treated as a reverse process to anthropomorphism (i.e., making objects/machines look more like humans; [ 1]; [29]). While anthropomorphism has received considerable attention in the marketing literature ([ 1]; [64]), dehumanization is mostly studied in psychology with a focus on intergroup relations and threat as a top-down, motivated bias that affects how in-group members may compare out-group members to objects/machines ([40]; [44]; [65]; [102]).
More recent work has begun to acknowledge that perceptions of machine-likeness in humans can also be driven by a bottom-up process, such as through an exposure to a visual cue (without specific intergroup conflicts or biases). For instance, facial configurations (e.g., width-to-height ratio [[27]]; see also [52]; [69]) or movement speed ([48]; [90]) can influence how machine-like a human is perceived. Our work builds on these recent findings by ( 1) exploring other dimensions that can shift how humans are perceived (i.e., changing the body composition and appearance); ( 2) homing in on the impact of representing humans as machines on food choices that consumers make daily, beyond the traditional context of intergroup relations and threat; and ( 3) further theorizing the driving role of consumers' idiosyncratic differences in eating self-efficacy. Importantly, we argue that being exposed to human-as-machine representations, with more machine-like ( 1) internal body composition, ( 2) face, ( 3) appearance, and ( 4) physical movement, can affect consumers' choices because it changes their expectation of how they should behave when it comes to food.
We posit that being exposed to human-as-machine representations changes not only consumers' perceptions but also their expectations of how they should behave. This is because alterations of physical features elicit schemas (either of humans or machines) and prompt individuals to apply normative behavioral expectations accordingly ([ 1], [ 2]; [55]; [57]). For instance, machines that look more human-like (e.g., anthropomorphized computers) are expected to interact like humans, such as by making small talk ([18]), and anthropomorphized automobiles are trusted more ([103]). In contrast, when humans are portrayed as machines, this brings to mind the expectation that one should behave like a machine. If a runner is portrayed as a machine, one expects them to be a strong entity without "human weakness" ([36]; [49]). Patients who are perceived as machines are expected to experience less "human" pain, which would allow doctors to maintain their professional distance and objectivity ([41]; [63]).
While humans can surely hold a variety of schemas and expectations about machines, one of the most prominent schemas, we conjecture, is that machines rely solely on their "head" (cognition), as they lack a human heart (emotion); in contrast, emotion and cognition are both fundamental elements of humans' decision making ([22]). These associations are formed from early childhood and are continuously reinforced through common language usage and mass media. For instance, in The Wizard of Oz, all that the Tin Man wants is a human heart. Data, Star Trek's android character, wants to let go of rationality to experience human emotions. Likewise, when a human possesses machine-like features, such as Iron Man (Tony Stark), he struggles with the effects of becoming too rational and losing human emotionality.
To empirically verify consumers' existing schema that machines rely on their head (cognition) and not their heart (emotion), we conducted a pilot study and asked 305 U.S.-based adults and students (46.6% female; Mage = 36.08 years), on three seven-point scales, to indicate the extent to which they consider machines' decisions and humans' decisions to be based on emotion (1 = "emotional, nonanalytical, warm") compared with cognition (7 = "unemotional, analytical, cold"; [42]; [43]; Cronbach's alpha =.92). Results verified that people believed that machines' decisions were more cognitive and head-based (M = 6.35, SD =.84) than humans' decisions (M = 3.34, SD = 1.26; t(303) = 24.87, p <.001, d = 2.82). We also included the classic Heart Versus Mind Scale ([91]; Cronbach's alpha =.94) and found that these two sets of scales were highly correlated (r(303) =.85, p <.001) and provided consistent results: machines' decisions were perceived as being based more on thoughts, cognition, and the head (M = 4.57, SD = 1.20) than humans' decisions (M = 2.00, SD = 1.35; t(303) = 17.20, p <.001, d = 1.97).
We posit that this popular association that machines rely on their head (cognition) can activate an expectation for one's own behaviors because the human-as-machine stimuli either explicitly or implicitly establish a connection between humans and machines. By visually transforming humans' body composition, appearance, and movement characteristics into machines, the human-as-machine stimuli bring to mind schemas about machines (e.g., a cognition-driven decision approach) and activate an expectation that these schemas should apply to humans, much the way anthropomorphism—by portraying objects as humans—activates an expectation that human schemas should apply to the focal objects ([ 1], [ 2]; [55]; [57]).
In summary, we posit that when humans are portrayed as machines in health or food marketing, it activates an expectation that one should behave like a machine, relying on one's head (cognitive) instead of the heart (emotion) when choosing food. Importantly, we argue that this expectation can lead to more complicated consequences than originally anticipated: the effect depends on consumers' chronic level of eating self-efficacy.
Having an expectation of making cognitive, machine-like food choices can motivate healthier choices only if consumers believe that the expected behavior is doable ([ 5]; [66]; [79]). Specifically, when facing an expectation, consumers go through an evaluation process, in which they assess their abilities to successfully meet the expectation (e.g., using their past behaviors as a proxy; [ 6]). This evaluation process thus involves predicting future outcomes to determine one's choices and behaviors. If consumers believe that they can meet the expectation, they then anticipate success in fulfilling it ([ 6]; [ 8]), which operates as a positive motivator, facilitating the engagement of behaviors that will help meet the expectation ([ 7]; [10]; [66]).
In contrast, if consumers believe that they cannot meet the activated expectation (e.g., because their past performances were unsuccessful), they instead anticipate failure in fulfilling it ([ 6]; [ 8]). The anticipation of failure, critically, serves as a negative motivator ([ 6]; [ 8]), leading to disengagement ([51]; [67]) and often opposite behaviors. Two lines of research suggest that a backfire effect—going against the activated expectation to choose unhealthier food—would likely occur in this case. First, anticipating failure can trigger aggression toward the self and reactance against the activated standard or expectation ([14]; [15]). Because an impossible standard/expectation induces feelings of impairment regarding one's abilities, people would opt to reestablish their freedom by behaving "in the way they want" (and not in the way they are expected to; [86]). In the context of health, this would result in a backfire or boomerang effect that goes against the communicated message ([17]; [86]). Second, and more specific to the food domain, knowing that one will fall short of an internal or external expectation leads to an unflattering and aversive evaluation of the self, which is often accompanied by negative emotions and emotional distress ([12]; [47]). Dietary disinhibition and overeating can then occur as a way to escape from these unpleasant states ([73]; [89]; [96]). Feeling unable to meet the body-shape expectations activated by a super-thin magazine model, for instance, led women to unhealthy overeating to make themselves feel better ([59]).
Many traits can affect how consumers respond to the expectation of making cognitive, machine-like food choices. We propose that consumers' chronic level of eating self-efficacy constitutes one critical trait. Self-efficacy is broadly defined as belief in one's ability to achieve a particular outcome or goal ([ 7]). Eating self-efficacy, accordingly, refers to a consumer's belief in his or her specific ability to choose healthy food ([ 4]; [95]).
More importantly, eating self-efficacy is linked to several eating habits essential to a cognitive, machine-like approach to food. Consumers high (vs. low) in eating self-efficacy are less likely to succumb to emotional eating ([24]; [97]) or to use food to respond to negative emotional events (e.g., an argument with family; [94]) and anxiety ([23]; [37]). While consumers low in eating self-efficacy use food to deal with boredom ([37]), those high in eating self-efficacy have less difficulty staying focused on the functional (cognitive) aspect of food. Similarly, consumers high (vs. low) in eating self-efficacy do better with analytics-based consumption, such as estimating portion size ([60]), evaluating caloric needs ([95]), and calculating nutritional values ([104]).
Because existing habits and past behaviors are the basis for assessing one's ability to meet an expectation ([ 6]), it is likely that consumers chronically high in eating self-efficacy would consider a cognitive, machine-like approach to food an easy expectation to meet, whereas the same standard would seem extremely difficult or impossible for consumers low in eating self-efficacy. We empirically verified this in another pilot study: Consumers high (vs. low) in eating self-efficacy indeed felt more (vs. less) able to make food decisions in a cognitive, machine-like manner (for method and results of the pilot study, see Web Appendix 2).
In summary, we propose the following three hypotheses:
- H1: Exposure to human-as-machine (vs. human or control) representations leads to healthier (vs. unhealthier) food choices for consumer high (vs. low) in eating self-efficacy.
- H2: Exposure to human-as-machine (vs. human or control) representations activates an expectation that one should adopt a cognitive, machine-like approach to food.
- H3: Consumers high (vs. low) in eating self-efficacy feel able (vs. unable) to meet the activated expectation of adopting a cognitive, machine-like approach to food, resulting in healthier (vs. unhealthier) food choices.
Figure 1 illustrates our conceptual model.
Graph: Figure 1. Overview of the model and hypotheses.
We conducted five studies (and three replication and follow-up studies) with a variety of incentive-aligned food choices and multiple human-as-machine stimuli. Study 1 (and two replications) and Study 2 tested our key hypothesis: human-as-machine representations led to healthy or unhealthy food choices depending on consumers' level of eating self-efficacy (H1). We measured consumers' chronic level of eating self-efficacy in Study 1 and directly manipulated eating self-efficacy in Study 2.
Studies 3 and 4 (and a follow-up study) tested the proposed mechanisms through moderated mediation analyses: exposure to human-as-machine stimuli activated an expectation to approach food in a cognitive, machine-like manner in all consumers (H2). Activating this expectation led to divergent effects: whereas consumers high in eating self-efficacy made healthier food choices, consumers low in eating self-efficacy went against the expectation and made unhealthier choices (H3). Studies 3 and 4 also ruled out alternative accounts such as perception of food (as a source of pleasure or energy), hunger, people's beliefs about what they could digest, emotionality, and perception of humans' competence.
Finally, Study 5 explored a theory-based solution in the field by accompanying human-as-machine stimulus with a message that made consumers feel that they could meet the expectation to make food choices in a cognitive, head-based manner. The intervention message successfully attenuated the backfire effect on lunch choices at a cafeteria and enabled an effective use of human-as-machine stimuli to facilitate healthier choices for all.
Study 1 tested our key proposition, that human-as-machine representations would facilitate healthier choices among consumers high in eating self-efficacy but would backfire and result in unhealthier choices among consumers low in eating self-efficacy (H1). To set a baseline of what people choose without exposure to any stimulus related to humans or machines, we also included a control condition in which participants viewed a neutral visual.
Three hundred U.K.-based adults (64.0% female; Mage = 36.70 years) recruited from Prolific Academic participated in this study. The study used a 3 (stimulus: human as machine vs. human vs. control) × 1 (eating self-efficacy [measured as a continuous variable]) between-subjects design. For this and all following studies, target sample sizes were determined in advance of data collection on the basis of participant availability, study design, and collection method ([92]). Herein, we report all data exclusions, manipulations, and measures; all stimuli can be found in Web Appendix 3, and all data sets are available upon request.
Inspired by health marketing stimuli used in the real world (see Web Appendix 1) and following procedures from anthropomorphism research ([72]), we created human-as-machine stimuli by altering an image of the human digestive system (i.e., the internal body composition) in this study. In the human-as-machine condition, the digestive system was illustrated as a machine; in the human condition, the digestive system was illustrated as human organs. For the stimulus pretest, we also included a third human condition, a human upper body with no organs showing, to ensure that showing human organs in the human condition did not make the image seem less human (for all stimuli, see Web Appendix 3).
In the stimulus pretest, we measured human versus machine perception using scales from the anthropomorphism literature ([ 1]; [56]; [87]). Participants rated one of the images on three seven-point Likert scales ("The human [body]..." 1 = "looks like a machine," and 7 = "looks like a human"; 1 = "does not look alive at all," and 7 = "looks very alive"; 1 = "contains mainly machine-like features," and 7 = "contains mainly human-like features"). For comprehensiveness, we also included classic measures of dehumanization ("The human [body] is represented as unemotional, cold, rigid, fungible (lacking individuality), superficial, passive, inert (lifeless)"; 1 = "strongly disagree," and 7 = "strongly agree"; [42]). The results verified that the digestive system presented as a machine was indeed perceived as more machine-like on the human–machine continuum (M = 3.54, SD = 1.81) than the digestive system presented as human organs (M = 5.18, SD = 1.24; t(64) = 5.51, p <.001, d = 1.36) and the human upper-body condition (M = 5.08, SD = 1.41; t(64) = 4.82, p <.001, d = 1.19); the latter two groups did not differ (t(64) =.31, p =.759, d =.08). Results were similar for the reverse-coded dehumanization scale (for results and scale correlations, see Web Appendix 3). Drawing on the pretest results, we used the images of the two digestive systems (without the upper body) in the main study to test the hypothesized effect.
In the main study, participants were told that they would view different visuals and representations of the human body and that they would share their honest thoughts and opinions about them. After completing a bot check, they saw one of the two images from the pretest (digestive system presented as a machine vs. as human organs). Following the procedures in prior literature ([34]; [93]), we had participants describe the digestive system in 100 words on basis of the image they saw to reinforce the manipulation and ensure attention to the stimulus. We also included a pure (no human and no machine) control condition, in which participants saw a map and were asked to describe the directions from home to their workplace in 100 words. The control condition ensured a similar amount of writing effort with no specific relation to either machine or human, allowing us to isolate the direction of changes in participants' food choices. Participants responded to two filler questions to reduce demand effects (for the variety of filler questions used in this and the following surveys, see Web Appendix 4).
To capture food choices, all participants were told at the end of this survey that, in addition to their regular compensation, they would be entered into a lottery for $9 worth of food coupons. They were asked to choose three snack items (each in a $3 portion size) out of a selection of ten and were promised the coupons for the three items they chose (incentive-compatible). For each snack item, participants read information on ingredients and caloric content per package. The calorie content of these ten items ranged from 30 calories (mini peeled carrots) to 250 calories (Snickers bar; for snack choices used, see Web Appendix 5). Participants selected three items and received a confirmation that they were now entered into the lottery.
Participants then proceeded to another set of filler questions before responding to the four-item eating self-efficacy scale by [ 4]; e.g., "I believe I have the ability to eat a low-fat diet in the next month," 1 = "definitely do not," and 7 = "definitely do"; "If it were entirely up to me, I am confident that I would be able to eat a healthy diet in the next month," 1 = "strongly disagree," and 7 = "strongly agree"; Cronbach's alpha =.91; for the full scale, see the Appendix). Before exiting the study, participants entered demographic information and reported any suspicions or questions they had. All participants were debriefed and entered a lottery to receive $9 additional payment (the monetary value of the coupons).
None of the participants raised any suspicions or questions. We summed the calorie content of the three snack items the participants chose as a proxy for how healthy their food choices were, as prior literature has shown that consumers use calorie information to assess the healthiness of food items ([21]). To ensure that this was indeed the case, we also conducted a posttest on the health perception of these ten snacks (for posttest results of the snacks, see Web Appendix 5). Replacing the sum of calories with the sum of health scores from the posttest as the dependent measure revealed consistent results.
We conducted a regression analysis with stimulus (human as machine vs. human vs. control), eating self-efficacy (continuous measure), and their interaction as predictors, with age and gender serving as control variables (Model 1, [46]). In this and all following studies, we included age and gender as covariates because both have been shown to affect how people feel about machines ([11]; [77]) as well as how they make food choices ([ 3]). Analyses without these variables revealed consistent patterns in all studies. We report results without age and gender in Web Appendix 6 for comprehensiveness.
The model revealed a main effect of eating self-efficacy (β = −45.22, SE = 8.86; t = −5.11, p <.001); people with higher eating self-efficacy chose lower-calorie snacks. The model also revealed two main effects of stimulus (human as machine vs. control: β = 270.48, SE = 71.50; t = 3.78, p <.001; human as machine vs. human: β = 205.09, SE = 81.32; t = 2.52, p <.001); the human-as-machine stimulus led participants to choose higher-calorie snacks compared with the other two conditions. With regard to control variables, results revealed that female participants chose lower-calorie snacks than male participants (β = −32.37, SE = 14.19; t = −2.28, p =.023). Age did not have an effect. More importantly, we found two significant stimulus × eating self-efficacy interactions, one between the human-as-machine condition and the human condition (β = −36.84, SE = 14.19; t = −2.66, p =.008) and another between the human-as-machine condition and the control condition (β = −50.32, SE = 12.56; t = −4.01, p <.001).
Further spotlight analyses on eating self-efficacy (M = 5.62, SD = 1.27) illustrated that among those with high eating self-efficacy (1 SD above the mean [+1 SD] = 6.89), the effect of the human-as-machine stimulus was facilitative such that participants chose lower-calorie snacks in the human-as-machine condition (M = 475.40) than in the control condition (M = 551.41; β = −76.01, SE = 23.89; t = −3.18, p =.002) or the human condition (M = 523.92; β = −48.52, SE = 23.03; t = −2.11, p =.036); the human and control condition did not differ (β = −27.48, SE = 24.56; t = −1.12, p =.264). In contrast, the human-as-machine stimulus backfired among participants with low eating self-efficacy (1 SD below the mean [−1 SD] = 4.35), such that they chose higher-calorie snacks after viewing the human-as-machine stimulus (M = 590.28) than after viewing either the control stimulus (M = 538.44; β = 51.84, SE = 22.46; t = 2.31, p =.022) or the human stimulus (M = 545.22; β = 45.05, SE = 25.54; t = 1.76, p =.079); the human and control conditions did not differ (β = 6.78, SE = 24.34; t =.28, p =.781; see Figure 2).
Graph: Figure 2. The effect of stimulus on snack choices for consumers low versus high in eating self-efficacy: measured (Study 1).
We further replicated these results in two follow-up studies with a different eating self-efficacy scale to increase generalizability. Self-efficacy and behavioral control are conceptually similar and often used interchangeably ([16]). Accordingly, we adopted a measure from the behavioral control literature and used five items of [74] scale that directly assessed eating self-efficacy; sample items included "It's easy for me to reduce my sodium intake" and "It's easy to eat fresh fruits and vegetables regularly" (1 = "strongly disagree," and 7 = "strongly agree"; Cronbach's alpha =.74). In the first follow-up study, mirroring Study 1, we captured participants' chronic level of eating self-efficacy at the very end of the survey session so that its measurement would not contaminate participants' interpretation of the stimuli or their food choices. In the second follow-up study, we measured participants' chronic level of eating self-efficacy first, then added filler items, and then exposed participants to the human-as-machine stimuli to account for any demand effects. For the method and results of these studies, see Web Appendix 7.
The results of Study 1 and the two replications provide support for the divergent effects of portraying the human body as a machine (H1), revealing a critical dark side of such representation. While consumers with a high level of eating self-efficacy reacted positively to this stimulus and made healthier choices, consumers with a low level of eating self-efficacy made worse food choices upon exposure to human-as-machine representations.
Study 2 served two objectives. First, we tested a different human-as-machine stimulus to enhance the generalizability of the support for H1: the face, which is often used in anthropomorphism research ([57]; [64]) and has been a keen focus of previous dehumanization research (e.g., [27]). We added machine-like features to a human face and tested the effect of this stimulus on food choices. This also ensured that the observed divergent effects would occur without a visual of the digestion system. Second, we directly manipulated individuals' perceived level of eating self-efficacy to rule out any other dispositional differences between these two types of consumers as alternative accounts.
Two hundred three undergraduate students (43.8% female; Mage = 20.32 years) came into the lab of a large Dutch university to participate in this study in exchange for study credits. The study used a 2 (stimulus: human as machine vs. human) × 2 (eating self-efficacy: high vs. low) between-subjects design.
Following the procedures in Study 1, participants in the pretest were randomly assigned to one of two conditions. In the human-as-machine condition, participants saw a human face with machine-like features. In the human condition, participants viewed the same human face without machine-like features. In both conditions, participants saw either a male or female face. The pretest, which used the same two machine-likeness and dehumanization scales as in Study 1's pretest ([ 1]; [42]; [56]; [87]), was successful. Those who saw the human-as-machine face evaluated the face as more machine-like than those who saw the human face (for stimuli pretest details and results, see Web Appendix 3).
In the main study, participants were told that there were multiple different surveys in the study and that they would complete all of them in order. They first went through a general survey about themselves, which incorporated an eating self-efficacy manipulation that we developed based on work by [ 9] and [13]. Participants answered a set of questions regarding their current eating habits (e.g., "How many of your meals in an average week include red meat," "How many of your weekly meals are likely high in sodium [because they are canned, packaged, or take-out options]"). They were then informed that a score was calculated based on their answer to these questions, reflecting how capable they were of eating healthily; participants were randomly assigned to see that they were classified as "very capable" (high eating self-efficacy) or "having difficulties" (low eating self-efficacy). For the manipulation, see Web Appendix 8.
After completing the eating self-efficacy manipulation, participants entered the second study, in which we randomly exposed them to either the human-as-machine face or the human face. Participants were not asked to write 100 words about the stimuli in this study, to further ensure that the observed effects could occur without mandatory reflection.
Finally, participants were asked to choose three snacks (each in a $3 portion size) out of a selection of ten, as in Study 1. We further included both eating self-efficacy scales as manipulation checks: the scale used in Study 1 ([ 4]; Cronbach's alpha =.92) and the scale used in the two replications ([74]; Cronbach's alpha =.64). The manipulation was successful (for results, see Web Appendix 8).
The session ended with demographic information and a probe for suspicion and questions. All participants entered a lottery for $9 additional payment. Because we informed some participants that they were not eating healthily, we included an extensive debrief and ensured that all participants read and understood that the score was arbitrary and unrelated to their actual behavior. We also allowed them to withdraw their data from analysis if desired (out of 203 undergraduate students, 7 opted to not be included, leaving a final sample of 196).
We conducted an analysis of covariance with stimulus (human as machine vs. human), eating self-efficacy (high vs. low), and their interaction as predictors, and age and gender as covariates. We again found a main effect of eating self-efficacy: participants who believed they were low in eating self-efficacy chose higher-calorie snacks (M = 475.88, SD = 139.96) than those who believed that they had high eating self-efficacy (M = 420.00, SD = 134.61; F( 1, 190) = 7.63, p =.006, η2 =.039). There was no main effect of stimulus, age, or gender in this study. Consistent with Study 1, we again observed the hypothesized stimulus × eating self-efficacy interaction (F( 1, 190) = 5.15, p =.024, η2 =.026).
Further contrast analysis revealed that among the participants who were led to perceive high eating self-efficacy, the caloric content of the snacks chosen was similar between the human-as-machine face condition (M = 404.25, SD = 132.74) and the human face condition (M = 434.23, SD = 135.98; t(97) = 1.11, p =.270, d =.22). In contrast, those who were led to perceive low eating self-efficacy chose snacks significantly higher in calories when they saw the human-as-machine face (M = 505.77, SD = 139.04) than when they saw the human face (M = 441.33, SD = 134.38; t(95) = 2.32, p =.023, d =.47) (see Figure 3).
Graph: Figure 3. The effect of stimulus on snack choices for consumers low versus high in eating self-efficacy: manipulated (Study 2).*p <.05.**p <.01.Notes: Error bars are ±1 SE.
Unlike Study 1 and the two replications, those who were manipulated to have a high level of eating self-efficacy did not make healthier choices upon exposure to the human-as-machine stimuli. Because we did not observe this pattern in any of our other studies, we discuss this discrepancy in the "General Discussion" section. Overall, Study 2 used another type of human-as-machine stimulus—a machine-like human face—and directly manipulated people's perceived level of eating self-efficacy; we found that while the results differed among those who were led to perceive high eating self-efficacy, the backfire effect was replicated for those who were led to perceive themselves as bad at eating healthily.
We hypothesized that viewing the human-as-machine stimulus leads to divergent effects depending on consumers' levels of eating self-efficacy because ( 1) the stimulus brings to mind an expectation to choose food in a cognitive, machine-like manner, and ( 2) this expectation motivates consumers with high eating self-efficacy (who feel capable of meeting the expectation) to make healthier choices but conversely leads consumers with low eating self-efficacy (who feel incapable of meeting the expectation) to act against it, resulting in unhealthier choices. In Study 3, we used the same human-as-machine stimulus as in Study 1 to capture the activated expectation; in Study 4, we used another type of human-as-machine stimulus to triangulate the proposed role of expectation. Both studies further rule out multiple alternative accounts, including the perception of food as a source of pleasure or energy, hunger, people's beliefs about what they could digest, emotionality, and perception of humans' competence.
Study 3 served multiple purposes. First, we used a moderated mediation approach to provide support for the role of expectation—namely, that exposure to a human-as-machine stimulus creates an expectation that one should adopt a cognitive, machine-like approach to food in all participants (H2), and that this expectation leads to divergent food choices on the basis of participants' chronic levels of eating self-efficacy (H3).
Second, we wanted to rule out several food-related alternative accounts, such as the human-as-machine representations changing how hungry participants felt and what they believed their body could digest. We also wanted to ensure that our stimuli did not affect how the participants thought about food (as a source of pleasure or energy). Therefore, we measured these alternative accounts for moderated mediation analyses and used a different set of food choices that varied in health perceptions but not in calorie content to further underscore that it was indeed "unhealthy" food choices, rather than high-energy food choices (which often correlate with high calories), that led to the observed divergent effects.
Third, we aimed to underscore the importance of seeing a human-as-machine representation, and not just a general prime of machine, to activate the expectation for how humans should behave when choosing food. Thus, we added a machine-only condition without relating it to humans to explore this possibility.
Two hundred ninety-five undergraduate students (47.8% female; Mage = 20.46 years) participated in this lab study for study credits at a large Dutch university. The study used a 3 (stimulus: human as machine vs. machine only vs. human) × 1 (eating self-efficacy [measured as a continuous variable]) between-subjects design.
Following the procedures in previous studies, participants in the main study first viewed one of the stimuli of Study 1 (the digestive system: human as machine vs. human), or a machine-only stimulus (with no visual reference to the human body; see Web Appendix 3). We again had participants describe the digestive system in 100 words based on the image they saw, to reinforce the manipulation and ensure attention to the stimulus ([34]; [93]). In addition to ensuring attention, this approach also enabled us to register the amount of time spent on writing ([54]), to assess whether any condition evoked greater effort than others (which could lead to unhealthier choices because of perceived reward entitlement; [84]). Participants answered a filler question to further minimize the possibility of a demand effect.
Afterward, participants were told that for their participation, they could choose one snack to bring home and viewed four snack options available for that day's session: an energy bar, a yogurt, a chocolate bar, and a bag of chips (for snack choices, see Web Appendix 5). Based on our posttest (n = 107; 67.3% female; Mage = 27.05 years), the first two snacks were perceived as similarly healthy, whereas the latter two were similarly unhealthy; calorie content was exactly the same across these snacks (for the posttest, see Web Appendix 5). Using a different set of snack options further enhanced the generalizability of our findings.
To cleanly capture the role of expectation, participants then went through another filler task and continued to the next part of the study. We were particularly interested in assessing whether ( 1) exposure to human-as-machine stimulus that did not specifically mention food or eating would activate an expectation about how food choices should be made, and ( 2) participants applied the activated expectation to themselves and not just to humans in general. Thus, we asked the participants to report their perceived expectation of adopting a cognitive, machine-like approach to food on three seven-point Likert scales ([42]; [43]; 1 = "strongly disagree," and 7 = "strongly agree"; Cronbach's alpha =.71): "I feel that I am expected to make my food choices..." "unemotional," "analytical," and "cold." Participants were also asked to judge the function of food (pleasure or energy), their body's ability to digest a variety of food (Cronbach's alpha =.85), and their current hunger level (for all scales, see the Appendix). All scales of potential process variables were presented in random order.
The session ended with demographic information and the eating self-efficacy scale used in Studies 1 and 2 ([ 4]; Cronbach's alpha =.95) and a probe for suspicion and questions. All participants received a snack when exiting the lab.
None of the participants raised any suspicion or questions about the study. Because all snack options contained the same amount of calories, we coded the choice of healthy snack as 0 and unhealthy snack as 1 to be consistent with previous studies (i.e., higher values represented unhealthier choices) and submitted this binary dependent variable to an analysis with stimulus (human as machine vs. machine only vs. human), eating self-efficacy, and their interactions as predictors and age and gender as control variables (Model 1, [46]). Similar to previous studies, results revealed a main effect of eating self-efficacy: Those high in eating self-efficacy were more likely to choose a healthy snack (β = −1.48, SE =.31; Z = −4.74, p <.001). We observed two main effects of stimulus (human as machine vs. machine only: β = 8.67, SE = 2.00; Z = 4.34, p <.001; human as machine vs. human: β = 9.40, SE = 1.90, Z = 4.97, p <.001). Gender had a main effect (as in Study 1, female participants chose healthier snacks; β = −.69, SE =.28; Z = −2.46, p =.014), but age did not have an effect. More importantly, the model revealed two hypothesized stimulus × eating self-efficacy interactions on snack choice, one between the human-as-machine and the machine-only conditions (β = −1.64, SE =.36; Z = −4.53, p <.001) and one between the human-as-machine and human conditions (β = −1.76, SE =.35; Z = −5.07, p <.001).
Further spotlight analyses on eating self-efficacy (M = 5.35, SD = 1.45) illustrated that among those with high eating self-efficacy (+1 SD = 6.80), the effect of the human-as-machine stimulus was facilitative such that participants chose healthier snacks in the human-as-machine condition than in the machine-only condition (β = −2.46, SE =.60; Z = −4.10, p <.001) or in the human condition (β = −2.54, SE =.61; Z = −4.17, p <.001).
In contrast, the human-as-machine stimulus again backfired among participants with low levels of eating self-efficacy (−1 SD = 3.90), such that they chose unhealthier snacks after viewing the human-as-machine stimulus than in the machine-only condition (β = 2.27, SE =.66; Z = 3.46, p =.001) or the human condition (β = 2.53, SE =.62; Z = 4.13, p <.001). The machine-only condition did not differ from the human condition for either group of consumers, indicating that mere exposure to a machine (without any visual reference to humans) did not affect food choices.
We conducted the same analyses on expectation. The model revealed only a main effect of stimulus, such that all participants in the human-as-machine condition experienced a higher expectation to choose food in a cognitive, machine-like manner, compared with the machine-only condition (β = 1.92, SE =.74; t = 2.61, p =.009) and the human condition (β = 3.26, SE =.61; t = 5.30, p <.001). The latter two conditions did not differ, suggesting that mere exposure to a machine (without any visual reference to humans) did not affect participants' expectation to adopt a machine-like approach to food. There was no effect of eating self-efficacy or interaction with stimuli.
We performed the same analyses on the alternative accounts (function of food, digestion capability, and hunger). These analyses revealed no differences between the three stimuli, eating self-efficacy, and no interaction effects (see full results in Web Appendix 6).
We proceeded to conduct a bias-corrected moderated mediation analysis (Model 15, [46]): the stimulus predicted the perceived expectation of adopting a machine-like approach to food, and individuals' eating self-efficacy moderated the effect of this expectation on food choice, with age and gender serving as control variables. Results without control variables again revealed consistent effects and are reported in Web Appendix 6 for completeness.
The results supported our predictions. The first part of the model showed that viewing the human-as-machine stimulus heightened the expectation to adopt a cognitive, machine-like approach to food compared with the machine-only condition (β = 1.36, SE =.17; t = 7.90, p <.001) and the human condition (β = 2.27, SE =.17; t = 13.21, p <.001).
The second part of the model showed that for food choices, there were two direct effects of stimulus (human as machine vs. machine only: β = 6.37, SE = 2.17; Z = 2.93, p =.003; human as machine vs. human: β = 4.59, SE = 2.34; Z = 1.96, p =.050), and two interactions with eating self-efficacy, respectively (β = −1.20, SE =.39; Z = −3.07, p =.002; β = −.86, SE =.43; Z = −2.00, p =.046). Expectation also significantly affected food choices (β = 1.98, SE =.58; Z = 3.39, p =.001).
Importantly, whether expectation led to healthier or unhealthier choices depended on individuals' level of eating self-efficacy, as captured by a significant expectation × eating self-efficacy interaction in the full model (β = −.38, SE =.11; Z = −3.54, p <.001). The conditional indirect effects for eating self-efficacy (M = 5.35, SD = 1.45) between the human-as-machine versus machine-only conditions showed that a heightened expectation of adopting a machine-like approach to food led to healthier choices for those high in eating self-efficacy (+1 SD = 6.80; β = −.78, 95% confidence interval [CI] = [−1.50, −.29]) but led to unhealthier choices for those low in eating self-efficacy (−1 SD = 3.90; β =.70, SE =.36; 95% CI = [.13, 1.55]; index of moderated mediation: β = −.51, SE =.19; 95% CI = [−.98, −.29]). The same applied for the human-as-machine versus human conditions: a heightened expectation led to healthier choices for those high in eating self-efficacy (+1 SD = 6.80; β = −1.29, 95% CI = [−2.49, −.47]) but unhealthier choices for those low in eating self-efficacy (−1 SD = 3.90; β = 1.17, SE =.36; 95% CI = [.20, 2.59]; index of moderated mediation: β = −.85, SE =.19; 95% CI = [−1.63, −.38]).
We again conducted the same moderated mediation analyses with the alternative account variables (function of food, digestion capability, and hunger) as the mediator. There were no effects of stimulus on either of these variables, nor were there any significant (moderated) mediation effects (we report the results in Web Appendix 6).
In addition to these analyses, we also compared how long participants spent writing about the stimulus they saw in each condition. A regression analysis with time spent on writing as the outcome variable, stimulus (human as machine vs. machine only vs. human), eating self-efficacy, and their interaction as predictors and age and gender as control variables (Model 1, [46]) revealed that there was no effect of stimulus, eating self-efficacy, or their interactions.
Employing a moderated mediation approach, we demonstrated that exposure to a human-as-machine stimulus led to a heightened expectation to adopt a cognitive, machine-like approach for all individuals, irrespective of their level of eating self-efficacy. The effect of expectation on food choice, however, was moderated by eating self-efficacy—it motivated those high in eating self-efficacy to make healthier choices but backfired among those low in eating self-efficacy. Priming machine alone did not lead to these effects, suggesting that consumers apply the expectation of making food choices in a cognitive, machine-like way to themselves only if the visual represented a human as a machine. Seeing a machine-only visual did not trigger an expectation for how humans should behave, just as seeing a human-only visual did not trigger expectations for how humans may need to behave differently. The observed effects also cannot be explained by altered food perceptions, hunger, or digestive capability.
Study 4 provided additional evidence on the proposed role of expectation (H2) and the divergent consequences it produces on food choices (H3) with yet another human-as-machine stimulus—altering human appearance and physical movement. Specifically, we used a virtual telepresence machine, which is gaining popularity in consumers' daily lives and in business interactions, to design our stimulus (see Web Appendix 3). As mentioned previously, these types of technological advances will soon be used in the retail sector and in restaurants ([78]), where food choices are often made. The chosen stimulus therefore has high relevance for practice and further expands the scope of our examination beyond the body's internal composition and face.
Furthermore, we focused on food-related alternative accounts in Study 3 but acknowledge that exposure to human-as-machine stimuli could alter one's level of emotionality or the perception of how competent humans in general are (vs. machines). We therefore wanted to measure and rule out these possibilities. Finally, to further enhance generalizability, we used another food choice in this study: yogurts that varied in calories, sugar, and fat.
Three hundred three U.K.-based adults (67.0% female; Mage = 38.26 years) participated in the study through Prolific Academic. This study constituted a 2 (stimulus: human as machine vs. human) × 1 (eating self-efficacy [measured as a continuous variable]) between-subjects design.
In this study, we created a different human-as-machine stimulus by altering appearance and physical movement ([ 2]; [39]; [75]). In the human-as-machine condition, the appearance was illustrated as a robotic skeleton and a human face, just as seen in virtual telepresence machines; in the human condition, the appearance was illustrated in a regular human form (see Web Appendix 3; we included different genders to enhance generalizability). To incorporate the dimension of physical movement, we then showed participants a video clip of this person (in either a human-as-machine form or a human form) moving through an apartment for 45 seconds. In the human-as-machine condition, the movement was choppy/mechanistic; in the human condition, the movement was smooth/fluent (adopted from [98]]).
The pretest, using the same two scales as in previous studies' pretests, was successful. The human-as-machine stimulus was perceived as more machine-like than the human stimulus (for stimuli pretest details and results, see Web Appendix 3).
Following the procedures in previous studies, participants first viewed one of the stimuli (human as machine or human, randomly assigned to a female or male version of the stimulus irrespective of their own gender) and watched the 45-second clip of this person moving through an apartment. Similar to the procedures in Study 2, participants were not asked to write 100 words about the stimuli, to further ensure that the observed effects could occur without mandatory reflection. Participants answered a filler question and then proceeded to enter their food choices.
Participants read a short introduction about a new yogurt company. They were told that the researchers had agreed to conduct a market study for this company to assess students' preferences for yogurts. They were then asked to choose one out of nine yogurts that they would like to receive and try. Yogurts differed in their level of healthiness, indicated by caloric, sugar, and fat content, ranging from 80 calories to 256 calories, with an increase of 22 calories between each choice and the next-higher-calorie choice (for the yogurt choices, see Web Appendix 5). To ensure that higher-calorie yogurts were indeed perceived as less healthy, we again conducted a posttest on the health perceptions of these yogurt options. As in prior studies, replacing calorie count with the health score from the posttest as the dependent measure revealed consistent results (for the posttest, see Web Appendix 5).
After selecting their choice of yogurt, participants were asked to report their perceived expectation of adopting a cognitive, machine-like approach to food as in Study 3 ([42]; [43]; Cronbach's alpha =.71). Although the stimuli in Study 3 did not specifically mention food or eating, they utilized digestive system visuals, which could activate thoughts related to food. In this study, the human-as-machine stimulus was not related to the digestive system, food, or eating, further underscoring that even a stimulus that was unrelated to digestion/food could activate an expectation about how food choices should be made. We also asked participants to respond to statements about their emotionality (Cronbach's alpha =.77) and perception of human competence (Cronbach's alpha =.59) to rule out these alternative accounts; see Appendix for full scales. All scales were presented in random order. The survey ended with demographic information, the eating self-efficacy scale ([ 4]; Cronbach's alpha =.93), and a suspicion probe.
None of the participants raised any suspicion or questions about the study. We submitted yogurt choice (1 = "healthiest option," and 9 = "unhealthiest option") as the dependent variable to an analysis with stimulus (human as machine vs. human), eating self-efficacy (continuous measure), and their interaction as predictors and age and gender as covariates (Model 1, [46]). Similar to prior studies, results revealed a main effect of eating self-efficacy: those high in eating self-efficacy chose lower-calorie yogurts (β = −.42, SE =.10; t = −4.27, p <.001). We again observed a main effect of stimulus (human as machine vs. human: β = 2.68, SE =.55; t = 4.86, p <.001). We also found a main effect of age, with older participants choosing healthier yogurts (β = −.03, t = −2.48, p =.014), and no gender effect. More important, the model again revealed the hypothesized stimulus × eating self-efficacy interaction on yogurt choice (β = −.49, SE =.10; t = −5.00, p <.001).
Further spotlight analyses on eating self-efficacy (M = 5.39, SD = 1.42) illustrated that the effect of the human-as-machine stimulus was again facilitative among those with high eating self-efficacy (+1 SD = 6.81): they chose healthier yogurts (M = 3.30) in the human-as-machine condition than in the human condition (M = 4.63; β = −.67, SE =.20; t = −3.38, p =.001). In contrast, the human-as-machine stimulus again backfired among participants with low levels of eating self-efficacy (−1 SD = 3.97): they chose less healthy yogurts (M = 5.88) in the human-as-machine condition than in the human condition (M = 4.44; β =.72, SE =.14; t = 3.65, p <.001).
We conducted the same analyses on expectation as in Study 3. As we hypothesized, we observed only a main effect of stimulus: participants in the human-as-machine condition experienced a higher expectation to choose food in a machine-like manner, compared with the human condition (β = 1.40, SE =.26; t = 5.32, p <.001). There was no effect of eating self-efficacy or interaction. We again conducted the same analyses on the alternative accounts (emotionality, perceived human competence), which revealed no differences between the two stimuli, eating self-efficacy, or any interaction (we report the results in Web Appendix 6).
We proceeded to conduct a bias-corrected moderated mediation analysis (Model 15; [46]) as in Study 3. Results replicated the findings in Study 3: the first part of the model showed that viewing the human-as-machine stimulus heightened the expectation to adopt a cognitive, machine-like approach to food (β = 1.28, SE =.67; t = 19.17, p <.001), irrespective of age and gender. The second part of the model showed that for food choices, there was an effect of both eating self-efficacy (β = −.76, SE =.40; t = −1.93, p =.055) and expectation (β = 1.66, SE =.54; t = 3.09, p =.002).
Most importantly, whether expectation led to healthier or unhealthier choices again depended on eating self-efficacy, as captured by a significant Expectation × Eating self-efficacy interaction in the full model (β = −.30, SE =.10; t = −3.10, p =.002). The conditional indirect effects for eating self-efficacy (M = 5.39, SD = 1.42) again showed that a heightened expectation of adopting a machine-like approach to food led to significantly healthier choices for those high in eating self-efficacy (+1 SD = 6.81; β = −.50, SE =.21; 95% CI = [−.91, −.06]) but conversely led to unhealthier choices for those low in eating self-efficacy (−1 SD = 3.97; β =.59, SE =.24; 95% CI = [.01, 1.09]; index of moderated mediation: β = −.38, SE =.13; 95% CI = [−.64, −.14]). Conducting the same moderated mediation analyses with the alternative account variables (emotionality and perception of human competence) as the mediator revealed no effects of stimulus or any (moderated) mediation effects.
So far, we have documented across three types of stimuli (internal body composition, face, and appearance and movement), three types of food choices, and a diverse group of participants from different countries that human-as-machine representations led to healthier food choices for consumers high in eating self-efficacy but backfired for consumers low in eating self-efficacy. We also provided evidence that these divergent effects were driven by an activated expectation to choose food in a cognitive, machine-like manner, which resulted in divergent food choices. In a follow-up study (Web Appendix 9), we replicated the moderated mediation results in this study and further measured whether participants anticipated success or failure in meeting the activated expectation. The results verified that whereas participants high in eating self-efficacy anticipated success in meeting the expectation and thus chose healthier options, those low in eating self-efficacy anticipated failure in meeting the expectation, which led to the backfire effect.
The final study tested a theory-driven solution: if the backfire effect occurred because consumers low in eating self-efficacy found the expectation of adopting a cognitive, machine-like approach to food too difficult to meet, then by making consumers feel that they can meet this expectation, the backfire effect should be attenuated. Testing this possibility provides not only additional support for the role of expectation but also a viable solution for marketers, educators, and policy makers; instead of withdrawing human-as-machine stimuli altogether or excluding specific consumer segments from these communications, interested parties can accompany a human-as-machine stimulus with an intervention message that makes everyone feel that they can meet the activated expectation.
In Study 5, we distributed flyers showing a human-as-machine representation (the digestive system stimulus used in Studies 1 and 3) to customers at a university-based cafeteria before they purchased lunch. Half of the flyers were accompanied by a message that aimed to make the activated expectation more doable, and half were not. This study further enhanced the external validity of our findings and its generalizability (from snack choices to lunch entrée choices), while testing a mechanism-driven solution.
We designed the intervention with the goal of making consumers who are low in eating self-efficacy believe that they can meet the expectation of adopting a cognitive, machine-like approach to food, without harming those high in eating self-efficacy. Specifically, in the intervention condition, an additional message stating "You CAN choose your food today with your head (not your heart)" was printed right under the human-as-machine visual (for the flyer, see Web Appendix 3). We informed participants that a head-based approach to food is easy and doable (instead of blatantly stating that a "cognitive" or "machine-like" approach is easy and doable) as this message is short, simple to process, and applicable for practical use. To ensure that adding this message indeed made the expectation activated by the human-as-machine stimulus seem more doable, we conducted a posttest. The posttest verified that when viewing the human-as-machine stimulus with the intervention message, participants indeed perceived it less difficult and more doable to meet the expectation of adopting a head-based, cognitive approach to food; see Web Appendix 3.
In the field study, which took place from January 13 to February 7, 2020, at a university-based cafeteria, research assistants approached customers before they entered the cafeteria and inquired about their interest in participating in a study in exchange for $7.00 (for pictures of the study's setup, see Web Appendix 3). Three hundred thirty-three customers (67.6% female; Mage = 41.07 years) participated. All customers were exposed to the human-as-machine stimulus (as the goal was to test the effectiveness of the intervention message); the study employed a 2 (intervention: yes vs. no) × 1 (eating self-efficacy [measured as a continuous variable]) between-subjects design.
Customers who were willing to participate received a survey about a flyer. Under the cover story that the school was testing different flyers for effectiveness and wanted to ensure that the flyers were relevant to the customers and had good printing quality, all participating customers were asked to review a flyer with a human-as-machine visual printed at the center (the digestive system stimulus used in Studies 1 and 3). The flyer either had an additional intervention message "You CAN choose your food today with your head (not your heart)" printed under the human-as-machine visual or did not have this message. The survey about the flyer included a few design-related questions (e.g., on color and clarity of the flyer), as well as questions on mood, hunger level, age, gender, and occupation/field of study. All customers listed the last three digits of their phone number and their initials (which served to link the surveys). Customers received $2 for this survey and proceeded to buy their lunch at the cafeteria. This cafeteria offered a wide variety of entrée choices, including a salad and soup bar, international bowls, pizza, burgers, sandwiches, and a grill station.
Right after customers purchased lunch and paid, they were invited to participate in the second part of this study to receive another $5, totaling $7. All customers who took the first survey participated in the second part. Research assistants took a picture of the lunch that the customers had just purchased while the customers completed the second survey. The second survey included a few questions about the lunch purchased, the overall impression of the cafeteria, the two eating self-efficacy scales ([ 4]; [74]), and phone number digits and initials to match their responses.
We asked two research assistants (blind to the hypotheses) to assess the healthiness of the lunch choices (1 = "very healthy," and 5 = "not at all healthy") based on the pictures. We averaged their scores (intercoder reliability was high; r =.72, p <.001) and then conducted a regression analysis with stimulus (human as machine without intervention message vs. with intervention message), eating self-efficacy ([ 4]), and their interaction as predictors and age and gender as control variables (Model 1, [46]). The model revealed a main effect of stimulus (β = −2.05, SE =.39; t = −5.25, p <.001). There was no direct effect of eating self-efficacy, age, or gender. More importantly, we found a significant stimulus × eating self-efficacy interaction (β = .31, SE =.07; t = 4.66, p <.001).
Further spotlight analyses (M = 5.78, SD =.94) showed that the intervention helped consumers low in eating self-efficacy (−1 SD = 4.84); they made healthier lunch choices when exposed to the human-as-machine stimulus with the intervention (M = 2.24) than without the intervention (M = 3.33; β = −.55, SE =.09, t = −6.12, p <.001). As we expected, there was no effect of the message (Mno int. = 2.73 vs. Mint. = 2.81) among those high in eating self-efficacy (+1 SD = 6.72; β = .04, SE =.09; t =.44, p =.657); they already felt that meeting the activated expectation was easy (see Figure 4).
Graph: Figure 4. The effect of stimulus and intervention message on lunch choice (Study 5).*p <.05.**p <.01.Notes: Error bars are ±1 SE.
We repeated these analyses with the alternative eating self-efficacy scale by [74], as tested in the two replications of Study 1 and in Study 2. The two eating self-efficacy scales were again correlated (r(333) =.40, p <.001), and results were consistent in both direction and significance (see Web Appendix 6).
Study 5 provided additional evidence for the proposed mechanism—that the divergent effects occurred because the consumers low (vs. high) in eating self-efficacy felt that it was difficult to meet the expectation of adopting a cognitive, machine-like approach to food. Most importantly, it also offers an effective solution for policy makers, educators, and marketers: by adding a message that makes a cognitive approach to food easier and more doable, the human-as-machine stimulus can lead to healthier choices for all consumers.
In an effort to fight obesity and educate consumers on how the human body functions, health marketing and education materials frequently portray humans as machines and encourage consumers to act more "machine-like." They use slogans like "Fuel your body, not your emotions," or visuals that literally present humans as machines (see Web Appendix 1).
In this work, we put this belief to a test and used a variety of human-as-machine representations inspired by anthropomorphism research, health education, marketing practice, and recent technological advancements. We uncovered critical divergent effects of exposure to human-as-machine representations—it was facilitative for consumers high in eating self-efficacy but backfired among consumers low in eating self-efficacy (Studies 1–5). We further showed that this divergent effect happened because exposure to human-as-machine stimuli activated the expectation that one should adopt a cognitive, machine-like approach to food (Studies 3 and 4), which would be difficult to meet for consumers low in eating self-efficacy. Importantly, this backfire effect was alleviated when human-as-machine stimuli were accompanied with an intervention message that made consumers feel that they could meet the expectation of adopting a cognitive, head-based approach to food (Study 5).
Our work echoes the growing interest in studying the push for a cognitive approach to food in consumer behavior research ([61]; [62]). We documented how using human-as-machine stimuli to promote this approach can create divergent effects on food choices, depending on consumers' chronic level of eating self-efficacy. Importantly, we further captured the underlying mechanisms accounting for these divergent responses: exposure to human-as-machine stimuli activates an expectation to adopt a cognitive, machine-like approach to food. Whereas this expectation is motivating to consumers high in eating self-efficacy, it backfires among those low in eating self-efficacy. Results of Studies 3–5 thus underscore the importance of this trait as the antecedent for how consumers would respond to an expectation about food consumption, resulting in expectation-aligned behaviors ([ 8]; [80]).
Our work thus provides important insights and inspires future research regarding the rich psychologies of consumers of different levels of eating self-efficacy. While consumers high in eating self-efficacy already make healthier food choices than those low in eating self-efficacy (i.e., a significant main effect in Studies 1–4, directional in Study 5), consumers high in eating self-efficacy could still benefit from human-as-machine stimuli and make healthier choices (in all studies except for Study 2, in which eating self-efficacy was manipulated).
One possibility for the inconsistent results could be related to our eating self-efficacy manipulation. While the specific treatment used to manipulate eating self-efficacy in Study 2—social comparison—can be powerful and pervasive ([100]), the feeling that one is currently ahead of others can conversely license one to indulge ([50]). If this occurs, it may cancel the originally positive effect of human-as-machine stimuli among these consumers. We encourage future research to explore how balancing/ licensing may interact with eating self-efficacy perceptions to affect food choices.
Another possibility could be that the manipulation of high eating self-efficacy did not induce sufficiently high self-perception on eating self-efficacy. In the original scale development ([ 4]), the sample mean of eating self-efficacy was 4.53 (SD = 1.45); more recent research using this measure ([76]) found a sample mean of 5.22 (SD =.89). A close examination of the means in all our studies using this scale (from 5.35 to 5.78) revealed an aggregate mean of 5.25 (SD = 1.35), which was consistent with prior literature. However, the manipulation check of the high-eating-self-efficacy condition in Study 2 only produced a mean of 5.01 (see Table 1). We further conducted a meta-analysis aggregating the eating self-efficacy scores across all studies that used this efficacy scale and had a continuous food-choice dependent variable (i.e., Studies 1, 2, 4, and follow-up); the threshold analysis of this aggregate data set revealed that the human-as-machine (vs. human) stimuli backfired for consumers with eating self-efficacy scores between 1.00 and 5.37 and were facilitative for consumers with eating self-efficacy scores between 6.07 to 7.00. Thus, the high-eating-self-efficacy condition in Study 2 may not be sufficiently high to produce a significant positive effect. We encourage future research to explore other ways to shift people's perception of eating self-efficacy.
Graph
Table 1. Regions of Significance for the Effect of Stimulus on Food Choice for Different Levels of Eating Self-Efficacy.
| Study | Sample | Eating Self-Efficacy Scale | Dependent Variable | Eating SE(–1 SD) | Eating SE(Mean) | Eating SE(+1 SD) | Johnson–Neyman Procedure Regions of Significance | Comments |
|---|
| Study 1 | Prolific Academic (United Kingdom) | Armitage and Connor (1999) | Calories snacks | 4.35 | 5.62 | 6.89 | 1.00 to 4.11 and 6.74 to 7.00 | |
| Study 1, Replication 1 | Crowdflower (United States) | Moorman and Matulich (1993) | Calories snacks | 2.85 | 4.05 | 5.25 | 1.00 to 3.10 and 5.20 to 7.00 | |
| Study 1, Replication 2 | Amazon's Mechanical Turk (United States) | Moorman and Matulich (1993) | Calories snacks | 2.82 | 4.12 | 5.42 | 1.00 to 3.24 and 5.10 to 7.00 | |
| Study 2 | Undergraduate students (Netherlands) | Armitage and Connor (1999) | Calories snacks | 4.71 | | 5.01 | N.A. | Eating self-efficacy was manipulated, scores reflect the manipulation check |
| Study 3 | Undergraduate students (Netherlands) | Armitage and Connor (1999) | Snack choice | 3.90 | 5.35 | 6.80 | 1.00 to 4.92 and 5.76 to 7.00 | Snack choices were binary (healthy/unhealthy) |
| Study 4 | Prolific Academic (United Kingdom) | Armitage and Connor (1999) | Yogurt choice | 3.30 | 5.39 | 6.81 | 1.00 to 4.84 and 6.06 to 7.00 | |
| Study 4, follow-up | Prolific Academic (United Kingdom) | Armitage and Connor (1999) | Yogurt choice | 4.05 | 5.45 | 6.85 | 1.00 to 5.35 and 6.23 to 7.00 | |
| Study 5 | Customer's university cafeteria (United States) | Armitage and Connor (1999), Moorman and Matulich (1993) | Lunch choice | 4.84 | 5.78 | 6.72 | N.A. | Study did not include a human condition |
70022242920974984 Notes: N.A. = not applicable.
For consumers low in eating self-efficacy, prior research has shown that these consumers have difficulties with eating rationally, unemotionally, and analytically ([23]; [24]; [37]; [60]; [95]; [97]; [104]). Our work suggests that these past experiences could lead consumers low in eating self-efficacy to act against human-as-machine stimuli and counter the expectation of adopting a cognitive, machine-like approach to food. Importantly, by adding an intervention message that made the expectation seem easier to meet (Study 5), we were able to attenuate the previously observed backfire effect. This field study not only provides a relevant solution for practitioners but also complements work on the importance of setting achievable expectations in inducing health-related behavioral change ([ 6]; [59]).
We chose to focus on eating self-efficacy because it is one of the most frequently used constructs in health behavior theories ([35]). Still, future research should explore the robustness of these effects using other related constructs, such as health behavioral control ([16]), eating self-control ([28]; [45]), emotional eating ([99]), and overall self-regulation ([101]). Finally, we note that consumers' past and current fitness levels, health conditions, and whether they are on a diet affect how they perceive their eating self-efficacy. Although we did not measure these habits and physical conditions in our studies, we encourage future research to take these variables into consideration when studying eating self-efficacy and healthy eating.
This research introduces the concept of mechanistic dehumanization—visually representing humans as machines—to the consumer behavior literature as a reverse process of anthropomorphism ([ 1]; [29]). Our findings echo those in anthropomorphism research that demonstrate that changes along the human–machine continuum prompt specific behavioral expectations ([ 2]; [55]; [57]). We found that when humans are portrayed as machines, it activates an expectation that one should behave in a machine-like way.
For dehumanization research, our work expands prior studies on dehumanization to underscore its relevance for consumer behavior research in three ways. First, while prior work in dehumanization has focused primarily on how changes in facial features and movements influence how humans are perceived on the human–machine continuum ([26]; [27]; [52]; [69]), our work tested other dimensions such as altering internal body composition and appearance. Our findings offer a rich set of stimuli for future work on dehumanization and marketing while bringing dehumanization literature closer to consumers' everyday lives. Second, we explored an important downstream consequence that is highly relevant for consumers and marketers and underscored how human-as-machine stimuli could activate unique expectations in the context of food, leading to both positive and negative effects on consumers' real-world choices. Third, we shed light on the importance of idiosyncratic differences. While previous research promotes the idea that feeling like a human is desirable and valuable for all individuals ([38]; [43]), we found that dehumanization stimuli can generate divergent effects.
Importantly, the direction of changes along the human–machine continuum warrants further investigation. When encountering a stimulus, individuals first make a binary choice to classify a stimulus as either a "human" or "nonhuman" ([71]). Drawing on this first-level assessment, they generate expectations (e.g., dehumanized humans should be more rational, anthropomorphized machines more emotional). In all our studies, we informed participants that they were evaluating a human (body, face, physical movements). As a result, our stimuli depict humans portrayed as machines. However, the line between humans and machines becomes blurrier, and many physical features convey conflicting signals ([30]; [40]). Future research should investigate the boundary at which a human or a machine is categorized as such and explore other dehumanization types. Examples include marketing messages with mechanical voices and artificial intelligence software that blurs the line between humans and machines ([70]; [83]).
Furthermore, researchers should examine the impact of human-as-machine representations on other types of food decisions as well as decisions in other domains. We focused on food choices (snack choices in Studies 1–4 and lunch purchases in Study 5) because of the relevance of human-as-machine representations in this context and the importance of uncovering unintended risks in this domain, but we believe that the documented effects and mechanisms could occur in other domains (e.g., financial, medical, and social decisions).
Finally, demographic and cultural differences should be further considered. Age and gender affect how people feel about machines ([11]; [77]) and how they make food choices ([ 3]). In our studies, we did not find consistent effects of these variables. While this could result from the natural variance in our samples (i.e., students vs. Prolific Academic population), we believe that future research is warranted. The same speculation applies to different cultures, which vary in the expectations they hold about machines (e.g., Asian vs. Western cultures; [53]; [58]). Culture also affects specific food-related expectations. While we showed that exposure to human-as-machine representations did not change whether food was construed as a source of energy or pleasure among participants from Western culture (Study 3), it is possible that the observed effects would differ in cultures that associate food with pleasure ([88]) or in contexts in which a cognitive, machine-like approach to food is not expected (e.g., buying a gift for someone, bringing food/snacks to a social gathering).
Important nonacademic stakeholder groups will find value in this research. Many stakeholders encourage consumers to make food choices in a cognitive (and less emotional) manner to battle the rise of obesity. We used stimuli available in the real world (digestive system illustrations used in health marketing, face morphing available in mobile apps, and teleconferencing agents used in business meetings and retail) and showed that while consumers indeed felt expected to adopt a more cognitive, machine-like approach to food, this expectation can backfire. Our results thus ring a cautionary bell for nonprofit organizations, policy makers, educators, and for-profit health marketers: a strategy used with good intentions of educating consumers and improving their health can have an unintended dark side that hurts a vulnerable segment of consumers. Our work thus echoes the insights from prior research, such that ( 1) confronting consumers with expectations on how they should behave can be risky if it is not aligned with their abilities, and ( 2) influencers should carefully tailor their content for target audiences (e.g., [82]).
There is hope, though, because the backfire effect documented in this research can be attenuated by altering the perception of one's relative level of eating self-efficacy (Study 2) and by reassuring consumers that meeting the expectation to make cognitive, head-based food choices is doable (Study 5). Our research thus provides practical solutions to help circumvent the backfire effect for various stakeholders who plan to use human-as-machine stimuli to encourage healthy eating. Finally, understanding the potential processes that cause indulgent food choices is also crucial for consumers, especially as human-as-machine stimuli become more prevalent in the lives of consumers around the world.
The following statements are related to your lifestyle and your behavior concerning your health. Please state to what extent you agree with the following statements.
- I believe I have the ability to eat a healthy diet in the next month (1 = "definitely do not," and 7 = "definitely do").
- To what extent do you see yourself as being capable of eating a healthy diet in the next month? (1 = "very unlikely," and 7 = "very likely").
- How confident are you that you will be able to eat a healthy diet in the next month? (1 = "very unsure," and 7 = "very sure").
- If it were entirely up to me, I am confident that I would be able to eat a healthy diet in the next month. (1 = "strongly disagree," and 7 = "strongly agree").
For the purpose of this research, we replaced [ 4] wording of "low fat diet" with "healthy diet."
Items rated on 1 = "strongly disagree," and 7 = "strongly agree."
- It's easy to cut back on snacks and treats.
- It's easy to eat fresh fruits and vegetables regularly.
- I find it hard to moderate my red meat consumption. (reverse-coded)
- It's easy to minimize the additives I consume.
- It's easy for me to reduce my sodium intake.
These five items from the original scale assessed participants' eating self-efficacy (other items pertained to general health behaviors and thus were not included to create the composite measure).
The main function of food is [to]...(1 = "provide pleasure/fun," and 7 = "satisfy hunger").
It is important that food...(1 = strongly agree, and 7 = "strongly disagree")
- Has a good taste.
- Has a pleasant appearance.
- Provides energy.
- Improves one's performance.
- I feel that my body can easily digest the food I consume.
- I feel that my body is prepared to digest a variety of food items easily.
- I feel that my body has no problem digesting what I choose to eat.
How hungry do you feel at the moment? (1 = "not at all," and 7 = "very much")
- How emotional did you feel when you looked at this image?
- How emotional did you feel when you made your food choice?
- How much was your food choice based on emotions/feelings?
- How competent are humans in general?
- How competent are humans in making good food choices?
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242920974986 - Portraying Humans as Machines to Promote Health: Unintended Risks, Mechanisms, and Solutions
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242920974986 for Portraying Humans as Machines to Promote Health: Unintended Risks, Mechanisms, and Solutions by Andrea Weihrauch and Szu-Chi Huang in Journal of Marketing
Footnotes 1 Kelly Haws
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Online supplement https://doi.org/10.1177/0022242920974986
References Aggarwal Pankaj, McGill Ann L. (2007), "Is That Car Smiling at Me? Schema Congruity as a Basis for Evaluating Anthropomorphized Products," Journal of Consumer Research, 34 (4), 468–79.
Aggarwal Pankaj, McGill Ann L. (2012), "When Brands Seem Human, Do Humans Act Like Brands? Automatic Behavioral Priming Effects of Brand Anthropomorphism," Journal of Consumer Research, 39 (3), 307–23.
Ares Gastón, Gámbaro Adriana. (2007), "Influence of Gender, Age and Motives Underlying Food Choice on Perceived Healthiness and Willingness to Try Functional Foods," Appetite, 49 (1), 148–58.
Armitage Christopher J., Conner Mark. (1999), "Distinguishing Perceptions of Control from Self-Efficacy: Predicting Consumption of a Low-Fat Diet Using the Theory of Planned Behavior," Journal of Applied Social Psychology, 29 (1), 72–90.
5 Atkinson John W. (1957), "Motivational Determinants of Risk-Taking Behavior," Psychological Review, 64 (6 Pt. 1), 359–72.
6 Bandura Albert. (1991), "Social Cognitive Theory of Self-Regulation," Organizational Behavior and Human Decision Processes, 50 (2), 248–87.
7 Bandura Albert. (1997), Self-Efficacy: The Exercise of Control. New York: Freeman.
8 Bandura Albert, Cervone Daniel. (1983), "Self-Evaluative and Self-Efficacy Mechanisms Governing the Motivational Effects of Goal Systems," Journal of Personality and Social Psychology, 45 (5), 1017–28.
9 Bandura Albert, Jourden F.J. (1991), "Self-Regulatory Mechanisms Governing the Impact of Social Comparison on Complex Decision Making," Journal of Personality and Social Psychology, 60 (6), 941–51.
Bandura Albert, Schunk Dale H. (1981), "Cultivating Competence, Self-Efficacy, and Intrinsic Interest Through Proximal Self-Motivation," Journal of Personality and Social Psychology, 41 (3), 586–98.
Bartneck Christoph, Suzuki Tomohiro, Kanda Takayuki, Nomura Tatsuya. (2007), "The Influence of People's Culture and Prior Experiences with Aibo on Their Attitude Towards Robots," AI and Society, 21 (1/2), 217–30.
Baumeister Roy F. (1997), "Esteem Threat, Self-Regulatory Breakdown, and Emotional Distress as Factors in Self-Defeating Behavior," Review of General Psychology, 1 (2), 145–74.
Ben-Ami Michal, Hornik Jacob, Eden Dov, Kaplan Oren. (2014), "Boosting Consumers' Self-Efficacy by Repositioning the Self," European Journal of Marketing, 48 (11/12), 1914–38.
Brehm Jack W. (1966), A Theory of Psychological Reactance. New York: Academic Press.
Brehm Sharon S., Brehm Jack W. (1981), Psychological Reactance: A Theory of Freedom and Control. New York: Academic Press.
Bui My, Droms Courtney M., Craciun Georgiana. (2014), "The Impact of Attitudinal Ambivalence on Weight Loss Decisions: Consequences and Mitigating Factors," Journal of Consumer Behaviour, 13 (4), 303–15.
Byrne Sahara, Hart Philip Solomon. (2009), "The Boomerang Effect: A Synthesis of Findings and a Preliminary Theoretical Framework," Annals of the International Communication Association, 33 (1), 3–37.
Cassell Justine, Bickmore Timothy. (2000), "External Manifestations of Trustworthiness in the Interface," Communications of the ACM, 43 (12), 50–56.
Castelo Noah, Schmitt Bernd, Sarvary Miklos. (2019), "Human or Robot? Consumer Responses to Radical Cognitive Enhancement Products," Journal of the Association for Consumer Research, 4 (3), 217–30.
Centers for Disease Control and Prevention (2016), "Health, United States, 2015: With Special Feature on Racial and Ethnic Health Disparities,"research report (May), https://www.cdc.gov/nchs/data/hus/hus15.pdf.
Chernev Alexander, Chandon Pierre. (2010), "Calorie Estimation Biases in Consumer Choice," in Leveraging Consumer Psychology for Effective Health Communications: The Obesity Challenge, Batra Rajeev, Keller Punam Anand, Strecher Victor J., eds. Armonk, NY: M.E. Sharpe, 104–21.
Cian Luca, Krishna Aradhna, Schwarz Norbert. (2015), "Positioning Rationality and Emotion: Rationality Is Up and Emotion Is Down," Journal of Consumer Research, 42 (4), 632–51.
Clark Matthew M., Abrams David B., Niaura Raymond S., Eaton Cheryl A., Rossi Joseph S. (1991), "Self-Efficacy in Weight Management," Journal of Consulting and Clinical Psychology, 59 (5), 739–44.
Costanzo Philip R., Reichmann Simona K., Friedman Kelly E., Musante Gerard J. (2001), "The Mediating Effect of Eating Self-Efficacy on the Relationship Between Emotional Arousal and Overeating in the Treatment-Seeking Obese," Eating Behaviors, 2 (4), 363–68.
Cramer Leonie, Antonides Gerrit. (2011), "Endowment Effects for Hedonic and Utilitarian Food Products," Food Quality and Preference, 22 (1), 3–10.
Deska Jason C., Almaraz Steven M., Hugenberg Kurt. (2017), "Of Mannequins and Men: Ascriptions of Mind in Faces Are Bounded by Perceptual and Processing Similarities to Human Faces," Social Psychological and Personality Science, 8 (2), 183–90.
Deska Jason C., Lloyd E. Paige, Hugenberg Kurt. (2018), "The Face of Fear and Anger: Facial Width-to-Height Ratio Biases Recognition of Angry and Fearful Expressions," Emotion, 18 (3), 453–64.
Dzhogleva Hristina, Lamberton Cait Poynor. (2014), "Should Birds of a Feather Flock Together? Understanding Self-Control Decisions in Dyads," Journal of Consumer Research, 41 (2), 361–80.
Epley Nicholas, Waytz Adam, Cacioppo John T. (2007), "On Seeing Human: A Three-Factor Theory of Anthropomorphism," Psychological Review, 114 (4), 864–86.
Ferrey Anne E., Burleigh Tyler J., Fenske Mark J. (2015), "Stimulus-Category Competition, Inhibition, and Affective Devaluation: A Novel Account of the Uncanny Valley," Frontiers in Psychology, 6, 249.
Food and Agriculture Organization, World Health Organization, and United Nations University (2004), "Human Energy Requirements: Report of a Joint FAO/WHO/UNU Expert Consultation,"research report, http://www.fao.org/3/a-y5686e.pdf.
Friedman Michael A., Brownell Kelly D. (1995), "Psychological Correlates of Obesity: Moving to the Next Research Generation," Psychological Bulletin, 117 (1), 3–20.
Gerrior Shirley, Juan Wenyen, Basiotis Peter. (2006), "An Easy Approach to Calculating Estimated Energy Requirements," Preventing Chronic Disease, 3 (4), A129.
Gino Francesca, Kouchaki Maryam, Galinsky Adam D. (2015), "The Moral Virtue of Authenticity: How Inauthenticity Produces Feelings of Immorality and Impurity," Psychological Science, 26 (7), 983–96.
Glanz Karen, Bishop Donald B. (2010), "The Role of Behavioral Science Theory in Development and Implementation of Public Health Interventions," Annual Review of Public Health, 31, 399–418.
Gleyse Jacques. (2013), "The Machine Body Metaphor: From Science and Technology to Physical Education and Sport, in France (1825–1935)," Scandinavian Journal of Medicine and Science in Sports, 23 (6), 758–65.
Glynn Shirley M., Ruderman Audrey J. (1986), "The Development and Validation of an Eating Self-Efficacy Scale," Cognitive Therapy and Research, 10 (4), 403–20.
Goldenberg Jamie L., Pyszczynski Tom, Greenberg Jeff, Solomon Sheldon, Kluck Benjamin, Cornwell Robin. (2001), "I Am Not an Animal: Mortality Salience, Disgust, and the Denial of Human Creatureliness," Journal of Experimental Psychology: General, 130 (3), 427–35.
Graham Susan A., Poulin-Dubois Diane. (1999), "Infants' Reliance on Shape to Generalize Novel Labels to Animate and Inanimate Objects," Journal of Child Language, 26 (2), 295–320.
Gray Heather M., Gray Kurt, Wegner Daniel M. (2007), "Dimensions of Mind Perception," Science, 315 (5812), 619.
Haque Omar Sultan, Waytz Adam. (2012), "Dehumanization in Medicine: Causes, Solutions, and Functions," Perspectives on Psychological Science, 7 (2), 176–86.
Haslam Nick. (2006), "Dehumanization: An Integrative Review," Personality and Social Psychology Review, 10 (3), 252–64.
Haslam Nick, Bain Paul, Douge Lauren, Lee Max, Bastian Brock. (2005), "More Human Than You: Attributing Humanness to Self and Others," Journal of Personality and Social Psychology, 89 (6), 937–50.
Haslam Nick, Loughnan Steve. (2014), "Dehumanization and Infrahumanization," Annual Review of Psychology, 65 (June), 399–423.
Haws Kelly L., Davis Scott W., Dholakia Utpal M. (2016), "Control over What? Individual Differences in General Versus Eating and Spending Self-Control," Journal of Public Policy & Marketing, 35 (1), 37–57.
Hayes Andrew F. (2013), Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach. New York: Guilford Press.
Heatherton Todd F., Herman C. Peter, Polivy Janet. (1991), "Effects of Physical Threat and Ego Threat on Eating Behavior," Journal of Personality and Social Psychology, 60 (1), 138–43.
Heptulla Chatterjee Sheba, Freyd Jennifer J., Shiffrar Maggie. (1996), "Configural Processing in the Perception of Apparent Biological Motion," Journal of Experimental Psychology: Human Perception and Performance, 22 (4), 916–29.
Hoberman John M. (2001), Mortal Engines: The Science of Performance and the Dehumanization of Sport. Caldwell, NY: Blackburn Press.
Huang Szu-chi, Lin Stephanie C., Zhang Ying. (2019), "When Individual Goal Pursuit Turns Competitive: How We Sabotage and Coast," Journal of Personality and Social Psychology, 117 (3), 605–20.
Huang Szu-chi, Zhang Ying. (2011), "Motivational Consequences of Perceived Velocity in Consumer Goal Pursuit," Journal of Marketing Research, 48 (6), 1045–56.
Hugenberg Kurt, Young Steven, Rydell Robert J., Almaraz Steven, Stanko Kathleen A., See Pirita E., et al. (2016), "The Face of Humanity: Configural Face Processing Influences Ascriptions of Humanness," Social Psychological and Personality Science, 7 (2), 167–75.
Kaplan Frederic. (2004), "Who Is Afraid of the Humanoid? Investigating Cultural Differences in the Acceptance of Robots," International Journal of Humanoid Robotics, 1 (3), 465–80.
Kellogg Ronald T. (1987), "Effects of Topic Knowledge on the Allocation of Processing Time and Cognitive Effort to Writing Processes," Memory and Cognition, 15 (3), 256–66.
Kim Hyeongmin Christian, Kramer Thomas. (2015), "Do Materialists Prefer the 'Brand-as-Servant'? The Interactive Effect of Anthropomorphized Brand Roles and Materialism on Consumer Responses," Journal of Consumer Research, 42 (2), 284–99.
Kim Hye-young, McGill Ann L. (2017), "The Effect of Financial Status on Consumer-Perceived Anthropomorphism and Evaluation of Products with Marketer-Intended Anthropomorphic Features," working paper, Wharton School, University of Pennsylvania (accessed August 2018), https://marketing.wharton.upenn.edu/wp-content/uploads/2016/12/McGill-Ann-PAPER-MKTG-Camp.pdf.
Kim Sara, McGill Ann L. (2011), "Gaming with Mr. Slot or Gaming the Slot Machine? Power, Anthropomorphism, and Risk Perception," Journal of Consumer Research, 38 (1), 94–107.
Kitano Naho. (2006), "A Comparative Analysis: Social Acceptance of Robots Between the West and Japan," EURON Atelier on Roboethics, http://www.roboethics.org/atelier2006/docs/Kitano%20west%20japan.pdf.
Klesse Anne K., Goukens Caroline, Geyskens Kelly, de Ruyter Ko. (2012), "Repeated Exposure to the Thin Ideal and Implications for the Self: Two Weight Loss Program Studies," International Journal of Research in Marketing, 29 (4), 355–62.
Knäuper Bärbel. (2013), "The Diet Self-Efficacy Scale (DIET-SE): Measurement Instrument Database for the Social Sciences," (accessed December 8, 2020), https://www.midss.org/content/diet-self-efficacy-scale-diet-se.
Kozup John C., Creyer Elizabeth H., Burton Scot. (2003), "Making Healthful Food Choices: The Influence of Health Claims and Nutrition Information on Consumers' Evaluations of Packaged Food Products and Restaurant Menu Items," Journal of Marketing, 67 (2), 19–34.
Krishnamurthy Parthasarathy, Prokopec Sonja. (2010), "Resisting That Triple-Chocolate Cake: Mental Budgets and Self-Control," Journal of Consumer Research, 37 (1), 68–79.
Kumar Vinay, Abbas Abul K., Fausto Nelson, Aster Jon C. (2014), Robbins and Cotran Pathologic Basis of Disease. Amsterdam: Elsevier Health Sciences.
Landwehr Jan R., McGill Ann L., Herrmann Andreas. (2011), "It's Got the Look: The Effect of Friendly and Aggressive 'Facial' Expressions on Product Liking and Sales," Journal of Marketing, 75 (3), 132–46.
Leyens Jacques-Philippe, Paladino Paola M., Rodriguez-Torres Ramon, Vaes Jeroen, Demoulin Stéphanie, Rodriguez-Perez Armando. (2000), "The Emotional Side of Prejudice: The Attribution of Secondary Emotions to Ingroups and Outgroups," Personality and Social Psychology Review, 4 (2), 186–97.
Liberman Nira, Förster Jens. (2008), "Expectancy, Value, and Psychological Distance: A New Look at Goal Gradients," Social Cognition, 26 (5), 515–33.
Locke Edwin A., Latham Gary P. (2002), "Building a Practically Useful Theory of Goal Setting and Task Motivation: A 35-Year Odyssey," American Psychologist, 57 (9), 705–17.
Longoni Chiara, Bonezzi Andrea, Morewedge Carey K. (2019), "Resistance to Medical Artificial Intelligence," Journal of Consumer Research, 46 (4), 629–50.
Looser Christine E., Wheatley Thalia. (2010), "The Tipping Point of Animacy: How, When, and Where We Perceive Life in a Face," Psychological Science, 21 (12), 1854–62.
Luo Xueming, Qin Marco Shaojun, Fang Zheng, Qu Zhe. (2020), "Artificial Intelligence Coaches for Sales Agents: Caveats and Solutions," Journal of Marketing, published online October 14, https://doi.org/10.1177/0022242920956676.
Mathur Maya B., Reichling David B. (2016), "Navigating a Social World with Robot Partners: A Quantitative Cartography of the Uncanny Valley," Cognition, 146, 22–32.
McGill Ann L. (1998), "Relative Use of Necessity and Sufficiency Information in Causal Judgments About Natural Categories," Journal of Personality and Social Psychology, 75 (1), 70–81.
Mills Jennifer S., Polivy Janet, Peter Herman C., Tiggemann Marika. (2002), "Effects of Exposure to Thin Media Images: Evidence of Self-Enhancement Among Restrained Eaters," Personality and Social Psychology Bulletin, 28 (12), 1687–99.
Moorman Christine, Matulich Erika. (1993), "A Model of Consumers' Preventive Health Behaviors: The Role of Health Motivation and Health Ability," Journal of Consumer Research, 20 (2), 208–28.
Morewedge Carey K., Preston Jesse, Wegner Daniel M. (2007), "Timescale Bias in the Attribution of Mind," Journal of Personality and Social Psychology, 93 (1), 1–11.
Naughton Paul, McCarthy Mary, McCarthy Sinéad. (2015), "Acting to Self-Regulate Unhealthy Eating Habits. An Investigation into the Effects of Habit, Hedonic Hunger and Self-Regulation on Sugar Consumption From Confectionery Foods," Food Quality and Preference, 46, 173–83.
Nomura Tatsuya, Kanda Takayuki, Suzuki Tomohiro. (2006), "Experimental Investigation into Influence of Negative Attitudes Toward Robots on Human–Robot Interaction," AI and Society, 20 (2), 138–50.
O'Reilly Tim. (2017), "What Will Our Lives Be Like as Cyborgs? A Case for Embracing the 'Augmentation' of Human Minds and Bodies," The Atlantic (October 27), https://www.theatlantic.com/technology/archive/2017/10/cyborg-future-artificial-intelligence/543882/.
Oettingen Gabriele, Bulgarella Caterina, Henderson Marlone, Gollwitzer Peter M. (2004), "The Self-Regulation of Goal Pursuit," in Motivational Analyses of Social Behavior: Building on Jack Brehm's Contributions to Psychology, Wright Rex A., Greenberg Jeff, Brehm Sharon S., eds. Mahwah, NJ: Lawrence Erlbaum Associates, 225–44.
Ozer Elizabeth M., Bandura Albert. (1990), "Mechanisms Governing Empowerment Effects: A Self-Efficacy Analysis," Journal of Personality and Social Psychology, 58 (3), 472.
Parker Jeffrey R., Lehmann Donald R. (2014), "How and When Grouping Low-Calorie Options Reduces the Benefits of Providing Dish-Specific Calorie Information," Journal of Consumer Research, 41 (1), 213–35.
Pechmann Cornelia, Catlin Jesse R. (2016), "The Effects of Advertising and Other Marketing Communications on Health-Related Consumer Behaviors," Current Opinion in Psychology, 10, 44–49.
Puntoni Stefano, Reczek Rebecca Walker, Giesler Markus, Botti Simona. (2020), "Consumers and Artificial Intelligence: An Experiential Perspective," Journal of Marketing, (published online October 16), https://doi.org/10.1177/0022242920953847.
Racine Sarah E., Horvath Sarah A., Brassard Sarah L., Benning Stephen D. (2019), "Effort Expenditure for Rewards Task Modified for Food: A Novel Behavioral Measure of Willingness to Work for Food," International Journal of Eating Disorders, 52 (1), 71–78.
Reyna Valerie F., Nelson Wendy L., Han Paul K., Dieckmann Nathan F. (2009), "How Numeracy Influences Risk Comprehension and Medical Decision Making," Psychological Bulletin, 135 (6), 943–73.
Reynolds-Tylus Tobias. (2019), "Psychological Reactance and Persuasive Health Communication: A Review of the Literature," Frontiers in Communication, 4, 56.
Romero Marisabel, Craig Adam W. (2017), "Costly Curves: How Human-Like Shapes Can Increase Spending," Journal of Consumer Research, 44 (1), 80–98.
Rozin P., Fischler Claude, Imada Sumio, Sarubin A., Wrzesniewski Amy. (1999), "Attitudes to Food and the Role of Food in Life in the USA, Japan, Flemish Belgium, and France: Possible Implications for the Diet–Health Debate," Appetite, 33 (2), 163–80.
Seddon Lesley, Berry Neil. (1996), "Media-Induced Disinhibition of Dietary Restraint," British Journal of Health Psychology, 1 (1), 27–33.
Shiffrar Maggie, Freyd Jennifer J. (1990), "Apparent Motion of the Human Body," Psychological Science, 1 (4), 257–64.
Shiv Baba, Fedorikhin Alexander. (1999), "Heart and Mind in Conflict: The Interplay of Affect and Cognition in Consumer Decision Making," Journal of Consumer Research, 26 (3), 278–92.
Simmons Joseph P., Nelson Leif D., Simonsohn Uri. (2011), "False-Positive Psychology: Undisclosed Flexibility in Data Collection and Analysis Allows Presenting Anything as Significant," Psychological Science, 22 (11), 1359–66.
Smith Pamela K., Jostmann Nils B., Galinsky Adam D., van Dijk Wilco W. (2008), "Lacking Power Impairs Executive Functions," Psychological Science, 19 (5), 441–47.
Stich Christine, Knäuper Bärbel, Tint Ami. (2009), "A Scenario-Based Dieting Self-Efficacy Scale: The DIET-SE," Assessment, 16 (1), 16–30.
Stotland Stephen, Zuroff David C., Roy Marguerite. (1991), "Situational Dieting Self-Efficacy and Short-Term Regulation of Eating," Appetite, 17 (2), 81–90.
Strauss Jaine, Doyle Alysa E., Kreipe Richard E. (1994), "The Paradoxical Effect of Diet Commercials on Reinhibition of Dietary Restraint," Journal of Abnormal Psychology, 103 (3), 441.
Toray Tamina, Cooley Eric. (1997), "Weight Fluctuation, Bulimic Symptoms, and Self-Efficacy for Control of Eating," Journal of Psychology, 131 (4), 383–92.
Tremoulet Patrice D., Feldman Jacob. (2000), "Perception of Animacy from the Motion of a Single Object," Perception, 29 (8), 943–51.
Van Strien Tatjana, Frijters Jan E., Bergers Gerard P., Defares Peter B. (1986), "The Dutch Eating Behavior Questionnaire (DEBQ) for Assessment of Restrained, Emotional, and External Eating Behavior," International Journal of Eating Disorders, 5 (2), 295–315.
Vartanian Lenny R., Spanos Samantha, Herman C. Peter, Polivy Janet. (2015), "Modeling of Food Intake: A Meta-Analytic Review," Social Influence, 10 (3), 119–36.
Vohs Kathleen D., Heatherton Todd F. (2000), "Self-Regulatory Failure: A Resource-Depletion Approach," Psychological Science, 11 (3), 249–54.
Waytz Adam, Gray Kurt, Epley Nicholas, Wegner Daniel M. (2010), "Causes and Consequences of Mind Perception," Trends in Cognitive Sciences, 14 (8), 383–88.
Waytz Adam, Heafner Joy, Epley Nicholas. (2014), "The Mind in the Machine: Anthropomorphism Increases Trust in an Autonomous Vehicle," Journal of Experimental Social Psychology, 52, 113–17.
Wilson-Barlow Lindsay, Hollins Tishanna R., Clopton James R. (2014), "Construction and Validation of the Healthy Eating and Weight Self-Efficacy (HEWSE) Scale," Eating Behaviors, 15 (3), 490–92.
World Health Organization (2016), "World Health Statistics 2016: Monitoring Health for the SDGs," research report, https://www.who.int/gho/publications/world%5fhealth%5fstatistics/2016/EN%5fWHS2016%5fTOC.pdf.
~~~~~~~~
By Andrea Weihrauch and Szu-Chi Huang
Reported by Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 102- Pretty Healthy Food: How and When Aesthetics Enhance Perceived Healthiness. By: Hagen, Linda. Journal of Marketing. Mar2021, Vol. 85 Issue 2, p129-145. 17p. 2 Color Photographs, 1 Chart. DOI: 10.1177/0022242920944384.
- Database:
- Business Source Complete
Record: 103- R2M Index 1.0: Assessing the Practical Relevance of Academic Marketing Articles. By: Jedidi, Kamel; Schmitt, Bernd H.; Ben Sliman, Malek; Li, Yanyan. Journal of Marketing. Sep2021, Vol. 85 Issue 5, p22-41. 20p. 2 Charts, 4 Graphs, 1 Map. DOI: 10.1177/00222429211028145.
- Database:
- Business Source Complete
R2M Index 1.0: Assessing the Practical Relevance of Academic Marketing Articles
Using text-mining, the authors develop version 1.0 of the Relevance to Marketing (R2M) Index, a dynamic index that measures the topical and timely relevance of academic marketing articles to marketing practice. The index assesses topical relevance drawing on a dictionary of marketing terms derived from 50,000 marketing articles published in practitioner outlets from 1982 to 2019. Timely relevance is based on the prevalence of academic marketing topics in practitioner publications at a given time. The authors classify topics into four quadrants based on their low/high popularity in academia and practice —"Desert," "Academic Island," "Executive Fields," and "Highlands"—and score academic articles and journals: Journal of Marketing has the highest R2M score, followed by Marketing Science, Journal of Marketing Research, and Journal of Consumer Research. The index correlates with practitioner judgments of practical relevance and other relevance measures. Because the index is a work in progress, the authors discuss how to overcome current limitations and suggest correlating the index with citation counts, altmetrics, and readability measures. Marketing practitioners, authors, and journal editors can use the index to assess article relevance, and academic administrators can use it for promotion and tenure decisions (see www.R2Mindex.com). The R2M Index is thus not only a measurement instrument but also a tool for change.
Keywords: information retrieval; marketing; marketing theory; marketing practice; relevance; topic modeling
Given that marketing is an applied discipline, articles published in academic journals should fulfill marketing practitioners' informational needs and be relevant to marketing practice. However, many articles do not seem to present relevant and important insights and findings that impact business practice ([17]; [21]; [23]; [42]; [44]). Prominent marketing scholars, including the founders of the annual Theory and Practice in Marketing (TPM) conference, have noted that "many observers complain that academia is far removed from addressing substantive problems of industry. This perception threatens to make our field irrelevant and undermine the legitimacy of research at business schools" ([11], p. 1).
For years, academic researchers have proposed changing the status quo. For example, they have proposed providing incentives and rewards for scholars to engage in relevant research, focusing less on technical sophistication and more on substantive issues, and improving communications and interactions between academics and practitioners ([21]; [24]; [25]; [32]; [39]; [50]). The editors of the Journal of Marketing (JM) recently highlighted specific actions to help infuse real-world perspectives into academic research ([44]). Scholars have also discussed how business schools can improve the practical importance of faculty research ([42]). A similar discussion has taken place among business school faculty in other fields (e.g., management, information systems, accounting), leading to analogous conclusions and proposals ([ 3]; [10]; [19]; [40]; [45]). However, an objective and easy-to-use measure of the practical relevance of articles has been missing from this debate. We believe that a simple and automated measure of relevance is likely the most effective way to change the status quo. In this article, we present version 1.0 of the Relevance to Marketing (R2M) Index, which measures the relevance of an academic article to marketing practice.
We define marketing relevance as the degree of the topical relation between the topics contained in an academic article and topics of marketing practice at a given time. Following this definition, an academic article is relevant if it is topically related to marketing practice and is timely (i.e., the information in the article satisfies marketing practitioners' current informational needs). The R2M Index is a dynamic index based on text-mining methodology. It uses a carefully constructed dictionary of more than 1,000 marketing terms derived from approximately 50,000 marketing articles published in important practitioner outlets such as Bloomberg Business Week, Financial Times, Forbes, Fortune, Harvard Business Review, McKinsey Quarterly, Marketing News, and The Wall Street Journal from 1982 to 2019, supplemented by other sources such as a Google search and Kotler and Keller's (2011) Marketing Management textbook. We validate the dictionary and the index with more than 350 executives enrolled in an Executive Master of Business Administration (EMBA) program as well as other marketing practitioners. The index allows us to measure which academic papers and topics are most relevant to marketing practice, whether academic marketing has become more or less relevant over time, and which academic marketing journals are most relevant. The index can help marketing practitioners (managers, consultants, and researchers in agencies) as well as the media and social influencers quickly identify whether an academic article is practically relevant to their own context and thus speaks to their informational needs. The index is also useful for academic scholars, journal editors, and administrators reviewing the relevance of academic publications. Finally, it contributes to the broader debate about the relevance of academic marketing. For R2M implementation, we developed a web application (www.R2Mindex.com) for scoring articles and searching for practically relevant research.
We first conceptualize relevance and show how text mining can be used to measure relevance to marketing. We then present the empirical study, including the construction of the marketing dictionary and the R2M Index. We show which topics are most popular over time in academia and marketing practice and use quadrant analysis to classify the topics into four quadrants based on their (low/high) popularity in academia and practice —"Desert," "Academic Island," "Executive Fields," and "Highlands." We also assess the overall relevance of marketing journals. We find that JM has the highest R2M score, followed by Marketing Science (MSC), Journal of Marketing Research (JMR), and Journal of Consumer Research (JCR). Next, we present validity and robustness checks. Finally, we discuss how limitations of this first version of the index can be addressed and how various stakeholders (e.g., managers, consultants, marketing researchers, authors, editors, administrators) can use the index, based on focus groups that we conducted with former organizers of the annual TPM conference.
Merriam-Webster's Dictionary defines "relevant" as ( 1) "a relation to the matter at hand," and ( 2) as "practical and especially social applicability" (https://www.merriam-webster.com/dictionary/relevance). Similarly, Lexico characterizes relevance as "the quality or state of being closely connected or appropriate" (https://www.lexico.com/en/definition/relevance), and Wikipedia states that the term refers to "the concept of one topic being connected to another topic in a way that makes it useful to consider the second topic when considering the first" (https://en.wikipedia.org/wiki/Relevance). The idea of a topical connection (or relation), which may lead to practical applicability, is central to the relevance concept in information science, and specifically information retrieval (IR) ([13]; [36]). Merriam-Webster also provides a third definition of relevance as it relates to IR, "the ability (as of an information retrieval system) to retrieve material that satisfies the needs of the user."
These definitions guided the development of the R2M Index, which is based specifically on the conceptualization of relevance in IR. In several comprehensive reviews of the relevance literature, [35], [36], [37]), a leading scholar in IR, conceptualizes relevance as "a property along which parts are related and may also be considered as a measure of the strength of the related connection" ([36], p. 10) and stresses that relevance is information that meets a target's needs. Other IR scholars have also emphasized the notion of relations and the importance of connecting content with users' needs ([ 5]; [29]; [38]). This view is consistent with a definition by the American Marketing Association's (AMA) Board of Directors, which states that marketing research "links the consumer, customer, and public to the marketer through information" (https://www.ama.org/the-definition-of-marketing-what-is-marketing/). Finally, in addition to topical relevance that serves informational needs, relevance is also conceptualized as being time and context dependent. [29] includes time as well as task (or context) as key dimensions in the interaction of the user and the IR system.
Some IR scholars and researchers in other fields have defined the concept of relevance more broadly in terms of impact and change. For example, [12], p. 603) states that "a phenomenon that is relevant changes the matter in some way; it adds information, or decreases information, offers a new perspective, or causes other kinds of cognitive change." In discourse theory, relevance refers to a linguistic discourse that produces change in knowledge and assumptions ([49]). The impact (or change) view of relevance is also central to [17], p. 212) definition of managerial relevance as "the degree to which a specific manager in an organization perceives academic knowledge to aid his or her job-related thoughts or actions in the pursuit of organizational goals." Relatedly, some authors distinguish relevance from importance ([21]; [41]). They suggest operationalizing importance by the number and status of stakeholders who are likely to change their behavior as well as the magnitude of the expected change ([21]).
The topic modeling approach we use in this article does not allow us to assess the broader meaning of relevance in terms of practical impact and change such as aiding or supporting decisions and actions related to organizational goals. However, topical relevance that provides timely information for marketing practitioners seems to be a necessary condition for practical impact and change. The index is also based on information that is important to practitioners. Outlets such as Financial Times, Harvard Business Review, McKinsey Quarterly, and The Wall Street Journal are read by numerous stakeholders, including high-status senior managers, to gain insight that they could use as part of their jobs. We therefore posit that marketers' need for topical and timely information can be captured by sourcing text (i.e., marketing terms from practitioner outlets such as practitioner-oriented journals, newspapers, and trade magazines) because the content in these outlets is written and published specifically for practitioners and reflects current topics of relevance to marketing practitioners.
Some scholars have measured the relevance of specific methodologies such as scanner data and conjoint analysis ([ 7]; [51]). [33] have used a measurement approach to address practical relevance more generally by asking practitioners (marketing managers and intermediaries) to rate academic articles. While the practitioners could identify practically relevant articles, the process of asking them to rate each article seemed tedious and time intensive. Only 20% of the managers and 37% of the intermediaries (e.g., consultants, agency researchers) in their study provided usable data. Because this process cannot be automated, constant practitioner input would be required to assess the relevance of new articles. In addition, the small sample of practitioners that might be recruited is unlikely to be representative and may be biased.
We build on these prior measurement approaches by pursuing an alternative, more efficient, and more contemporary approach based on text-mining and topic modeling ([ 4]), which allows for a continuous measurement of relevance. While there have been several prior text-mining analyses of article keywords or abstracts in marketing journals including MSC ([28]), JMR ([14]), and JCR ([47]), none of these analyses has focused on relevance. In this article, we use text mining to analyze full articles in MSC, JMR, JCR, and JM to assess the relevance of academic articles to marketing practice.
We measure practical relevance on the basis of the degree to which an academic article features relevant and current content to marketing practice. This measurement uses two components to capture relevance. The first, topical relevance, measures the degree to which concepts and ideas featured in the article are employed by marketing practitioners as part of their activities. The second, timely relevance, measures the degree to which the topics addressed in an article are current (or timely) to marketing practitioners within a given time period.
To assess topical relevance, the R2M Index uses a dictionary of more than 1,000 marketing terms derived from business and academic sources and validated by practitioners. The dictionary includes a wide variety of terms, ranging from words related to the "Three Cs" (company, competition, and customers) and strategy (segmentation, targeting, and positioning) to the "Four Ps" (product, pricing, place, and promotion) and other pertinent marketing concepts related to value, brand, innovation, choice, goals, culture, and measurement. To measure timely relevance, the index tracks the popularity of marketing topics in publications in practitioner outlets, assuming that these publications reflect topics of interest to practitioners within a given time period. From this input, the R2M Index is constructed to assess how relevant an academic article is to current marketing practice. This is done in two steps. First, we use topic modeling to identify topics in the marketing literature and assess topical relevance based on the dictionary of marketing terms. Second, the index assesses the articles on timely relevance based on the content published in practitioner outlets each year since the publication of the article. Thus, for each year, the R2M Index measures the degree to which an academic article includes marketing terms related to topics that are relevant and timely.
In summary, the R2M Index is an easy-to-use instrument to measure the practical relevance of an academic article. The index can be used to perform comparative analyses on the relevance of topics and entire journals and to conduct analyses over time. Next, we provide an overview of the empirical study, followed by the methodology, results, and validation and robustness checks.
As described in our conceptualization, we consider topical relevance (whether the content of an academic article covers topics that are associated with marketing practice) and timely relevance (whether the content relates to marketing practitioners' current interests). Regarding topical relevance, we constructed a dictionary of marketing words based on keywords extracted from marketing articles in practitioner outlets such as Bloomberg Business Week, Financial Times, Forbes, Fortune, Harvard Business Review, McKinsey Quarterly, Marketing News, and The Wall Street Journal. The database consisted of more than 50,000 marketing articles published in 15 practitioner outlets from 1982 to 2019. The final dictionary also included words from other sources (a Google search, an industry marketing dictionary, a textbook, and keywords from academic articles). The dictionary was validated with practitioners. Regarding timely relevance, we tracked the popularity of marketing topics published in practitioner outlets over time.
We text-mined the entire text of over 4,000 academic articles randomly sampled from four journals —JM, MSC, JMR, and JCR. We then scored these articles on their relevance to marketing practice. JM, MSC, and JMR are certainly marketing journals (the word "marketing" is in their titles); JCR focuses on consumer research, which is considered a subdiscipline of marketing ([26]). Hereinafter, we refer to the journals collectively as "marketing journals."
Using topic modeling, we derive a total of 40 marketing topics, which we capitalize in this article to identify them clearly as topics. Articles related to these topics vary in terms of their relevance to marketing (as measured by the R2M Index) and in topical and timely relevance (the components of the index). There are topics for which practitioner outlets are ahead of academic journals (e.g., Online Marketing) and others for which we observe the opposite effect (e.g., Conjoint Analysis). There are also topics with similar dynamics for practitioner outlets and academic journals (e.g., Market Entry, Branding). Across academic journals, we find that JM has the highest R2M score, followed by MSC, JMR, and JCR, in this order. Except for JCR, the marketing journals have progressed toward publishing more relevant topics.
Regarding validation, the R2M Index correlates well with observable measures of relevance, such as practice prize awards, [33]) list of 100 impactful papers, TPM conference submissions, and Marketing Science Institute (MSI) working papers and reports. Using a holdout set of articles from American Psychologist, American Economic Review, Psychological Review, and Quarterly Journal of Economics, we find, as expected, that these psychology and economics journals score lower than marketing journals. We also used executives to validate both the dictionary and the index. We surveyed executives in an EMBA program to confirm the practical relevance to marketing practice for each of the terms in our dictionary. In addition, we asked MSI corporate members and another group of EMBA individuals to judge the relevance of articles and observed a significant relationship between practitioners' judgments and the R2M Index.
The measurement methodology included four steps. First, we developed a dictionary of marketing terms based on 51,646 marketing articles, published in 15 practitioner outlets from 1982 to 2019, as well as other practice-related and academic sources, which were validated with marketing practitioners. Second, we employed text-mining techniques to extract key noun phrases ("words") from published articles and used latent Dirichlet allocation (LDA; [ 6]) to identify marketing topics. Each topic is defined by the set of words characterizing it, and each article has a probability of addressing each topic. Third, we scored the topics on their topical relevance by assessing the prevalence of practical marketing words in the topic. Fourth, we used the LDA estimates to predict the topic probabilities of each of the 51,646 practitioner articles. The average topic probability of all practitioner articles published in a particular year corresponds to its timely relevance score. The R2M Index for an article in a particular year is the product of the probability of the article to be associated with a topic and the topical and timely relevance scores of the topic, summed over all topics.
The data used to calibrate the LDA analysis included 4,229 articles, randomly sampled from a total of 5,495 (i.e., 77%), published in JM, MSC, JMR, and JCR over 34 years (from 1982 to 2015), employing JSTOR (dfr.jstor.org). This digital library provided the following information for each article: title, abstract, author(s), volume, issue, publication date, and full text. Our sample of articles was balanced over the years and journals with approximately 1,060 articles per journal and 125 articles per year. The cross-section and time-series nature of our data enabled us to compare journals and examine the evolution of the R2M Index over time.
We included the full text of each article for our textual analysis; the title, abstract, author(s), and references were removed. We preprocessed the PDF text of each paper by removing stop words, PDF markers, punctuation, plurals, author names, and references. We also fixed errors from converting the PDF files to text (e.g., the letter "h" in PDFs sometimes converts to "b" in text form). Following common practice in topic modeling, we also removed uncommon words that appeared in fewer than 20 (out of 4,229) papers or had a raw frequency across all papers (i.e., by counting duplicates) lower than 40. Note that if the infrequent word was related to a marketing concept, we looked for a marketing synonym and combined the two terms into one term without dropping the word. For example, "buzz marketing" was combined with "word of mouth marketing." The resulting tokenized text of each article was a bag of W = 16,080 key terms/words that occur at different frequencies across papers. This information was compiled in a spreadsheet of word counts with D = 4,229 rows (articles/documents) and W = 16,080 columns (words, key terms) where each element ndw represents the number of times word w (w = 1, ..., W) appears in document d (d = 1, ..., D). We utilized this spreadsheet, which we denote by X = ((ndw)), as input to the LDA analysis that we perform to identify the topics in the marketing field.
We employed a systematic process to construct and validate the marketing dictionary. Specifically, the dictionary was constructed based on term selections from practitioner-oriented articles and other business and academic sources that we describe next.
We used university library databases (e.g., ProQuest) to search for marketing articles published for practitioners. These archives contain subject indexing, keywords from articles, abstracts, and full texts of all published articles for each of the outlets. We searched for articles published in newspapers, magazines, and trade publications including Ad Age, Bloomberg Business Week, California Management Review, Entrepreneur, Fast Company, Financial Times, Forbes, Fortune, Harvard Business Review, Harvard Business School Publishing (for marketing cases), Inc., McKinsey Quarterly, Marketing News, MIT Sloan Management Review, and The Wall Street Journal. Our search retrieved 51,646 marketing-related articles that were published by these 15 outlets from January 1982 to April 2019. Using the subject terms and the keywords listed in the articles, we compiled a list of 18,200 terms.
We performed a Google search for definitions of marketing and selected the first ten documents that appeared. We used natural language processing to extract the most common words in these documents based on the frequency of occurrence (after removing stop words such as "and" and "the"), retaining the top 80 marketing words. These core marketing terms are shown as a word cloud in Figure A1 in the Web Appendix.
We included the 500 terms of the Common Language Marketing Dictionary (https://marketing-dictionary.org/), created as part of a partnership of the AMA, Marketing Accountability Standards Board, MSI, and the Association of National Advertisers. We also included 1,200 index terms from [22]Marketing Management, a standard textbook in marketing education, and 2,900 keywords from the articles in our corpus.
The combined list from all sources contained 22,880 (= 18,200 + 80 + 500 + 1,200 + 2,900) marketing terms, with a large degree of overlap. We used an elaborate process to reduce the number of terms by removing obvious nonmarketing words (e.g., "Congress," "women poets," "European Union") and words appearing fewer than 20 times in our corpus of 4,229 articles. We also checked whether words were properly used as marketing terms. For example, the word "distribution" connotes not only channel of distribution but also statistical distribution, and the word "chain" connotes not only retail chain but also Markov chain. We do not consider the statistical meanings as marketing terms. To resolve ambiguous instances, we created bigrams and trigrams to qualify the marketing use of the word. For example, we replaced "distribution" with "channel of distribution" whenever it was used in a channel context. Similarly, we dropped the word "relationship" because it is often used in a statistical or a psychological sense and replaced it with terms such as "customer relationship" and "firm relationship," depending on the context. Finally, synonymous terms were combined into one term (e.g., "ad," "advertisement," and "commercial"; "brand equity" and "equity of the brand").
Each of the authors evaluated the resulting list of marketing terms to ensure that it contained only relevant marketing words. Disagreements were resolved in a group setting using the Delphi approach.[ 5] The final dictionary contains 1,154 nonoverlapping marketing terms. Figure A2 in the Web Appendix shows the top 30 unigrams (e.g., "brand") and the top 30 bigrams and trigrams (e.g., "brand equity," "customer relationship management") in our marketing dictionary, ordered by how many times they appeared in our corpus.
To assess the evolution of the marketing terms in the dictionary, we tracked their word frequency in practitioner publications over time. Figure A3 in the Web Appendix displays this evolution from 1982 to 2019 in percentage terms. The dictionary of marketing terms stabilized between 2000 and 2010 (arguably due to the maturity of the field), with very few new marketing terms emerging after that.
We further validated our dictionary by surveying 247 executives enrolled in four sections of an EMBA course in a U.S. business school in the summer and fall semesters of 2019. This survey had two purposes: ( 1) to measure the extent to which each of the terms in our dictionary are related to marketing practice as perceived by business practitioners and ( 2) to use these measures to weigh the dictionary terms differently when we construct the R2M Index. The survey was administered in class using Qualtrics. We received 12,350 (= 247 × 50) observations, with each term in our marketing dictionary being evaluated by about 11 respondents. We did not offer compensation but randomly selected two students using a lottery to have a free dinner with the course professor, who is not an author of this article.
In the survey, we presented respondents with 50 marketing terms randomly drawn from our dictionary of 1,154 words and asked them to indicate whether the term (and, importantly, "the idea behind it") is relevant for the work of a marketing practitioner. Order of presentation of the terms was randomized for each respondent. The respondents had, on average, more than 9.59 years of business experience (2.17 years in a marketing-related job). They were asked how much their current job related to marketing; their average response was 3.94 on a 7-point scale (1 = "not all," and 7 = "very much"). The respondents had taken an average of 2.59 marketing courses in the past.
Each of the 1,154 marketing terms was judged to be relevant to marketing practice, on average, by 78% of the practitioners. For example, the term "brand equity" was judged to be relevant by 100% of the respondents who evaluated this term, "firm valuation" by 40%, and "accrual" by 0% (and thus was deemed irrelevant). Let be the marketing-term relevance score of word w (e.g., for "firm valuation"). Then we can use this information to weigh dictionary terms differently when we construct our R2M measure.
We use LDA to uncover the latent topic structure of publications in marketing journals. In LDA, each article can be viewed as a mixture of T latent topics. A topic is characterized by a set of words that is associated with it. A word or a term can be a single word (unigram) or a phrase (ngram). We use "word" and "term" interchangeably. For example, words most associated with the topic Customer Satisfaction/Customer Relationship Management (CRM) include "customer," "satisfaction," and "loyalty." An article can be associated with more than one topic (e.g., an article on channel coordination could be associated with Channel Management and with Analytical Models).
LDA uses the word count matrix X = ((ndw)), where ndw is the frequency of word w in article d, as input to generate two output matrices of probabilities. The first is a word-by-topic matrix PW = ((pwt)) where each element pwt ( ) indicates the probability that word w (w = 1, ..., W) is associated with topic t (t = 1, ..., T). Similar to factor loadings, this matrix characterizes the set of words associated with each topic and is generally used to interpret the derived topics. The second is a document-by-topic matrix QD = ((qdt)), where each element qdt ( indicates the probability that document d is associated with topic t. The QD matrix indicates the likely topic(s) to which an article can be assigned. It is akin to a factor score matrix.
LDA estimates QD = ((qdt)) and PW = ((pwt)) by assuming that the data are generated from a Dirichlet process. Each document has a probability qdt to be associated with topic t. The vector of topic probabilities for document d, Qd = (qd1, qd2, ..., qdT) is assumed to be distributed Dirichlet (α1, α2, ..., αT), where the αs are hyperparameters. For topic t, each word has a multinomial probability pwt to be associated with the topic. That is, the set of W words follows a multinomial distribution with parameters Pt = (p1t, p2t, ..., pWt) conditioned on topic t. We use the Gensim Python package (radimrehurek.com/gensim/) to estimate PW and QD for varying values of T. We pick the proper number of topics T* using a mix of criteria: minimum perplexity ([46]), variance of Pt across topics, and topic interpretability. Perplexity measures the degree of "uncertainty" an LDA model has in predicting a holdout text. Next, we discuss how to use the LDA estimates to construct the R2M Index.
Our bag of W = 16,080 words consists of marketing and nonmarketing terms. A topic is a distribution over a set of words that is differentiated from others. We measure topical relevance (Mt) by the preponderance of the marketing terms in topic t weighted by their marketing-term relevance score (rw) obtained from business practitioners. Let M denote the subset of marketing terms in our dictionary (M W). Then the topical relevance of topic t (t = 1, ..., T) is defined as
Graph
Thus, a topic that is associated with a larger set of practical marketing terms from our dictionary would have a higher topical relevance to marketing. (Note that
Timely relevance reflects how current a particular topic is for marketing practitioners in a given year. Articles that cover timely topics are judged to be more relevant to marketing practice than articles that treat nontimely topics. For example, the topic of Online Marketing is of more interest these days to marketing practitioners than the topic of Sales Promotions. In this regard, timely relevance captures which topics matter to practice in a given period and rewards academic articles at the forefront of these topics. Thus, we assess the timely relevance of a topic in a given year by assessing its prevalence in practitioner-oriented publications that year. We use the LDA estimates to predict the topic probabilities of each of the 51,646 marketing articles published in 15 business outlets from 1982 to 2019 based on the article's title, abstract and keywords. Let indicate the predicted probability that practitioner article d published in year y covers topic t. Let denote the number of practitioner articles published in year y. Then, we measure the timely relevance of a particular topic in a given year by its average topic probability across all practitioner articles published that year. That is,
Graph
Thus, topics that are more popular in a given year are considered to have higher timely relevance in that year than less popular topics. (Note that and
Whereas topical relevance ensures that an article covers marketing topics, timely relevance ensures that the content of the article is current. The two measures are thus conceptually different. Indeed, the sample correlation between the average timely relevance, and Mt is only.16 (p >.32), suggesting discriminant validity between the two measures. To illustrate, in our study Analytical Models scores relatively high on topical relevance (Mt =.31) because it employs marketing terms that are highly relevant to practitioners (e.g., profit, channel, sales, price, margin, competition). However, it scores relatively low on timely relevance ( suggesting low interest from practitioners in this topic.
An article is a probability mixture over the set of T topics. Each topic t is associated with two measures of relevance: topical relevance ( and timely relevance in year y ( ). Thus, to be practically relevant, an article needs to cover timely topics that are associated with marketing practice. Therefore, our R2M Index for article d in year y is given by
Graph
where qdt is the probability that article d is associated with topic t. We multiply by 100 because the elements in the sum are products of three probabilities, resulting into low numbers. Thus, articles that are timely and are associated with more substantive marketing topics are expected to have higher R2M scores.
The R2M measure of an article is not static but, rather, evolves over time from the year when the article is published to the present. For example, an article on Multiattribute Models/Conjoint would score low on R2M if published in the early 1980s but higher in the late 1990s and afterward, when this topic became popular among marketing practitioners. Conversely, an article on Marketing Theory and Policy published in early 1980s would score high on R2M given the buzz about marketing as a discipline at that time, but relatively lower in mid-1990s and afterwards as new marketing topics of interest to practitioners emerged. Although trending lower over time, the popularity of Marketing Theory and Policy is still relatively high in 2015 compared with other marketing topics. Thus, for our R2M measurement, ideally one should report the complete evolution of the R2M score of an article from its inception to the present as well as related summary statistics (i.e., minimum, mean, maximum, and standard deviation). However, for a point estimate, we use the mean R2M score throughout to assess the relevance of an article to practice.
Does our R2M measurement disadvantage leading-edge research relative to older, more seasoned research? This is unlikely. First, our dictionary is cumulative; it includes all the marketing terms until 2019, and the dictionary stabilized between 2000 and 2010 (see Figure A3 in the Web Appendix). Second, the timely relevance component of the R2M Index should benefit leading-edge papers especially if they address emerging marketing topics that are of current interest to practitioners. Third, as an empirical illustration, we compared the mean R2M scores for the top 100 papers in Online Marketing (currently a leading-edge topic) with those from Sales Promotions (an older topic). We find that the Online Marketing papers have a significantly higher mean R2M score than the Sales Promotion articles (.86 vs..75; p <.001).
Another question concerns the use of an indirect LDA approach rather than a straightforward measure based on the frequency of marketing words in a published article. A measure at the topic (vs. article) level is likely to create a more robust index because authors cannot easily inflate the relevance of an article by arbitrarily adding marketing terms. In other words, if the marketing jargon used in an article is not coherent with the topic, the work is less likely to be rewarded for it. Conversely, because a topic embodies a set of articles using similar language, the marketing terms used in the topic are more varied and are overall more exhaustive than those used in a single article. Thus, an article associated with the topic is less likely to be penalized if it misses some of the marketing jargon used in the topic because words that are synonymous are likely to appear in the same topic. In addition, a measure at the topic level (vs. article level) provides diagnostic information for why the R2M score for a journal (or article) is low or high, or why it is increasing or declining over time for a journal. Importantly, using LDA enables us to quantify the timeliness of the topics addressed in an article. Without LDA, it would be difficult to measure practitioners' interest in a topic in a particular time period.
We implemented LDA on our data to determine the topics that best characterize the articles published in marketing journals. Because the number of topics is unknown a priori, we performed the LDA analysis in two stages. First, we conducted a series of fivefold cross-validations to determine the perplexity per word for 5, 10, 15, 20, 25, ..., and 75, 80 topics. The perplexity plot in Figure A4 in the Web Appendix shows a U-shaped pattern with a plateau between 15 and 40 topics. Next, we examined the downward pattern of the variance of the posterior word-topic probabilities ( as we varied the number of Topics T from 1 to 100. Such variance moves closer to 0 after 40 topics (not shown). As such, using our judgment (interpretability of topics), the variance declining pattern, and the minimum perplexity criterion, we decided to retain T* = 40 topics.
In naming the 40 topics, we relied on ( 1) the top 30 words associated with the topic; ( 2) the top 30 papers that have the highest probability of loading on the topic; and ( 3) a comparative analysis with the topics generated in previous text analyses for MSC ([28]), JMR ([14]), and JCR ([47]). The 40 topics are meaningful and relatively easy to interpret. Importantly, these topics are consistent with the taxonomy by [ 9] using articles published in JMR from 2013 to 2019 ([ 9], Table 1, p. 987). Note that they found only 21 topics partly because they grouped consumer research topics under Consumer Psychology and empirical analysis topics under Research Methods. In addition, [33], pp. 128–29) listed 12 key marketing decision areas in firm management. All of these areas are included in our list of topics, and 11 of them rank among our 20 most relevant topics.
Graph
Table 1. Topics Relevance Overall and Over Decades.
| Overall Rank | Topic | | R2M Componentsa | Topic Rank by Decade |
|---|
| 100MtC¯t | Mt | C¯t | 1980s | 1990s | 2000s | 2010s |
|---|
| 1 | Advertising | 2.74 | .46 | .06 | 3 | 1 | 1 | 3 |
| 2 | Marketing Strategy | 2.56 | .52 | .05 | 2 | 2 | 2 | 1 |
| 3 | Market Orientation | 2.31 | .31 | .07 | 1 | 3 | 3 | 2 |
| 4 | Market Segmentation | 1.31 | .40 | .03 | 5 | 4 | 6 | 5 |
| 5 | Branding | 1.19 | .51 | .02 | 11 | 5 | 5 | 4 |
| 6 | Marketing Theory/Policy | 1.13 | .13 | .09 | 4 | 8 | 8 | 11 |
| 7 | Online Marketing | 1.09 | .37 | .03 | 17 | 7 | 4 | 6 |
| 8 | Sales Promotions | 1.03 | .41 | .03 | 7 | 6 | 7 | 9 |
| 9 | Market Entry | .88 | .38 | .02 | 8 | 9 | 10 | 17 |
| 10 | Product Management | .88 | .46 | .02 | 6 | 11 | 13 | 13 |
| 11 | Household Expenditure | .87 | .26 | .03 | 10 | 10 | 11 | 12 |
| 12 | New Products | .82 | .26 | .03 | 9 | 12 | 16 | 8 |
| 13 | Financial Impact | .81 | .27 | .03 | 13 | 13 | 9 | 10 |
| 14 | Pricing | .74 | .44 | .02 | 12 | 15 | 14 | 15 |
| 15 | Entertainment Marketing | .71 | .22 | .03 | 18 | 14 | 12 | 19 |
| 16 | Innovation | .68 | .37 | .02 | 14 | 18 | 17 | 14 |
| 17 | Customer Satisfaction/CRM | .68 | .36 | .02 | 15 | 16 | 18 | 16 |
| 18 | Consumer Culture | .65 | .12 | .05 | 19 | 17 | 15 | 18 |
| 19 | Sales Force Motivation | .56 | .24 | .02 | 16 | 19 | 20 | 21 |
| 20 | WOM and Social Media | .55 | .27 | .02 | 26 | 23 | 21 | 7 |
| 21 | Multiattribute Models/Conjoint | .51 | .26 | .02 | 23 | 20 | 19 | 20 |
| 22 | Household Purchase Behavior | .48 | .33 | .01 | 21 | 21 | 22 | 23 |
| 23 | Channel Management | .47 | .29 | .02 | 20 | 22 | 23 | 22 |
| 24 | Influence and Persuasion | .41 | .28 | .01 | 25 | 24 | 25 | 24 |
| 25 | Analytical Models | .41 | .31 | .01 | 22 | 26 | 26 | 26 |
| 26 | Bargaining and Negotiation | .40 | .24 | .02 | 27 | 25 | 24 | 25 |
| 27 | Sales Force Management | .36 | .25 | .01 | 24 | 27 | 27 | 27 |
| 28 | Consumer Choice | .31 | .25 | .01 | 28 | 28 | 28 | 28 |
| 29 | Behavioral Decision Theory | .29 | .21 | .01 | 31 | 29 | 29 | 30 |
| 30 | Affect and Emotions | .29 | .22 | .01 | 30 | 30 | 30 | 29 |
| 31 | Consumer Judgment | .28 | .20 | .01 | 29 | 31 | 31 | 31 |
| 32 | Family and Socialization | .24 | .12 | .02 | 34 | 32 | 32 | 33 |
| 33 | Information Processing | .24 | .19 | .01 | 32 | 33 | 33 | 32 |
| 34 | Measurement Scales | .21 | .13 | .02 | 33 | 36 | 36 | 37 |
| 35 | Cue Perception | .21 | .12 | .02 | 35 | 34 | 35 | 34 |
| 36 | Self | .21 | .15 | .01 | 36 | 35 | 34 | 35 |
| 37 | Dynamic Models | .18 | .11 | .02 | 37 | 37 | 38 | 38 |
| 38 | Consumer Goals and Motives | .18 | .13 | .01 | 38 | 38 | 37 | 36 |
| 39 | Empirical Estimation | .07 | .06 | .01 | 39 | 39 | 39 | 39 |
| 40 | Construct Measurement | .06 | .05 | .01 | 40 | 40 | 40 | 40 |
1 a These are the topical and average timely relevance scores for topic t. Their product (× 100) in column 3 gives the overall relevance of the topic to marketing practice.
Table A1 in the Web Appendix shows the labels for the 40 topic and reports selected frequent words associated with each topic. For example, the top words associated with Marketing Strategy are "firm," "competition," "strategy," "resource," "industry," "market," "target," and "business." Words associated with Branding are "branding," "category," "brand name," "private label," "extension," and "brand equity." The top words associated with Construct Measurement include "variable," "measure," "testing," "data," "factor," "correlation," "model," and "analysis."
Table 1 lists the 40 topics in descending order on the basis of their degree of relevance to practical marketing (100Mt ). For example, Advertising, Marketing Strategy, and Market Orientation have the highest relevance-to-marketing scores (100Mt = 2.74, 2.56, and 2.31, respectively). Construct Measurement has the lowest score (100Mt ), likely because it involves methodological issues (e.g., structural equation modeling may be of little interest to practitioners). As a closer examination of academic research revealed, this does not mean that practitioners do not care about having methodological tools. For example, Bagozzi and Yi's (1991) seminal article proposing how to use structural equation modeling to test multitrait-multimethod matrices to assess convergent and discriminant validities had an R2M score of.08, whereas a similar paper by [34], p. 7) addressing "practicing marketing researchers" had a score of.25.
Table 1 also reports the rank order of relevance to marketing topics per decade. Over the last decade, Marketing Strategy has become the most important topic; other topics, for example, Marketing Theory and Policy, have declined over time. Table 1 also reports the topical and timely components separately. Over the last four decades, Marketing Theory and Policy ( =.09) had the highest mean timely relevance score. Interestingly, Analytical Models also has relatively high topical marketing relevance but low timely relevance. Conversely, Consumer Culture is a topic of timely interest to practitioners, arguably because of its high specificity in exploring contemporary phenomena, but scores low on topical relevance.
We used multiple discriminant analysis to determine the topics that best differentiate the journals. The grouping variable is the journal where the article is published (JM, MSC, JCR, or JMR) and the independent variables are the probabilities of each article to be associated with each of the 40 topics, Qd- = (qd1, qd2, ..., qdT). The results indicate that all three discriminant functions are significant ( , p <.0001). For simplicity, we only retain the first two dimensions, which capture 93.7% of the variation in the data.
Figure 1 displays the journal centroids on the two dimensions as points and the topics as vectors. The vector coordinates are the "factor loadings" of the topics on each of the discriminant functions (i.e., structure matrix). Thus, topics with longer vectors better differentiate the journals than those with shorter ones. The orthogonal projection of a journal on a topic vector indicates the degree to which the journal is associated with the topic. For example, JM is most associated with Marketing Strategy, followed by MSC, JMR, and JCR. The figure highlights the topics with which each journal is mostly associated using different colors. Thus, almost all the topics in the lower-right quadrant of the figure are associated with JCR. Most of the topics in the lower-left quadrant are associated with MSC, and most of the topics in the upper half of the figure are associated with JM. Topics close to the center do not discriminate among topics. JMR seems to be central, with very few topics that clearly distinguish it.
MAP: Figure 1. Multiple discriminant analysis map of marketing journals.
The horizontal dimension, which explains 58.9% of the variability, contrasts consumer behavior (e.g., Information Processing, Consumer Goals and Motives) and analytical marketing topics (e.g., Analytical Models, Empirical Estimation). The vertical dimension, which explains 35% of the variability, contrasts managerial (e.g., Market Orientation) and nonmanagerial (e.g., Information Processing) topics. The marketing journals are well differentiated on the map. The triangular shape of the journal locations (with JMR at the center) is consistent with the standard classification of marketing scholarship into behavioral, quantitative, and managerial categories. JCR is mostly associated with behavioral topics, MSC with quantitative topics, and JM with managerial topics.
We next examine the popularity of each topic over the measurement period 1982–2015. Consider all the articles published in a given year, then the popularity of a topic in that year is given by the average probability ( ) of each of these articles to be associated with this topic. Figure 2 traces the popularity of the 40 topics over time in academic and practitioner publications.
Graph: Figure 2. Popularity of topics in academic and practitioner publications over time.
In the academic literature, research topics such as Behavioral Decision Theory, Innovations, Self, Online Marketing, Customer Satisfaction/CRM, and Word of Mouth (WOM) and Social Media have gained importance over time, whereas Advertising, Marketing Theory and Policy, Construct Measurement, and Information Processing have declined in popularity. Other topics, such as Market Orientation, Market Entry, and Channel Management, peaked around the early 2000s and then declined. By 2015, there was also a greater variety of topics, the most prominent ones being Self, WOM and Social Media, Online Marketing, Innovations, and Affect.
Topics that have been of relatively low interest to practitioners include Dynamic Models, Information Processing, Construct Measurement, and Analytical Models. Topics of relatively higher interest to practitioners include Household Expenditure, Marketing Theory and Policy, Advertising, Entertainment Marketing, Marketing Strategy, and Bargaining and Negotiations. Finally, there are similar evolution patterns in academic and practitioner publications for topics such as Market Entry, Salesforce Motivation, and Branding. In addition, there are topics where practitioners were ahead of the curve (e.g., Financial Impact, Innovations, Entertainment Marketing, Online Marketing, Pricing, WOM and Social Media). Conversely, academics were ahead of the curve with their articles related to topics such as Multiattribute Models/Conjoint and Sales Promotions.
To examine the topical relation between academic and practice topics, we suggest performing a quadrant analysis to create a map that displays topic popularity in academia and marketing practice. The four quadrants may be defined based on median splits along the two axes of topic popularity in academic journals and practitioner outlets. The quadrants may be labeled as follows: "Desert" (topics with low popularity in both academia and practice), "Academic Island" (topics with high popularity in academia but low popularity in practice), "Executive Fields" (topics with high popularity in practice but low popularity in academia), and "Highlands" (topics with high popularity in both academia and practice). From a theory-to-practice perspective, Highlands is the most desirable quadrant because the topical interests of scholars meet those of practitioners. To illustrate this approach and also depict the evolution of the topical relation over time, Figure 3 shows such a quadrant analysis for the recent time periods of 2000–2009 versus 2010–2015. The vectors in the figure indicate changes from the first to the second time period, and the colors of the vectors indicate the nature of the change.
Graph: Figure 3. Quadrant analysis of topic popularities.
We choose two topics for illustration. The first topic, Information Processing, was still popular in the early 2000s in academia but moved toward the Desert quadrant by becoming less popular in academia and barely changing its below-average popularity in practice during the second time period. The second topic, Online Marketing, moved from Executive Fields to the Highlands quadrant by becoming more popular in academia while maintaining its popularity in practice.
We also calculated the aggregate popularity of each quadrant (i.e., sum of the popularity scores of the topics in each quadrant for the whole period 2000–2015). As Figure 3 shows, the most popular quadrant for topics published in all academic journals is the Academic Island quadrant (45%), whereas 24% are in Highlands, 19% are in Executive Fields, and 13% are in the Desert quadrant. A closer analysis of topics published in each journal revealed that relatively few of the topics published in JM are in the Academic Island quadrant (28%) compared with MSC, JMR, and JCR (50%, 50%, and 51%, respectively). Conversely, 39% of the topics in JM are in Highlands, which is more than double the percentage of topics in Highlands in any other journal: 19%, 19%, and 18% for MSC, JMR, and JCR, respectively.
In this section, we illustrate what information can be obtained by scoring an academic marketing article on the R2M Index. We then examine the R2M distribution across marketing journals and over time. Finally, we provide validity and robustness checks for the R2M Index.
As Table 2 shows, for each article one can calculate its association with the topics (shown in the table for the top five topics), the mean R2M score, and the R2M evolution over time (here from 1982 to 2019). For illustration, we show two articles each from our four marketing journals that are associated with topics in different degrees, have different R2M scores, and, importantly, have opposite patterns over time. For example, consider the two selected JM articles. Hunt's (1983) article is primarily associated with Marketing Theory and Policy (probability =.59), has a mean R2M score of 1.03, and has a declining R2M score over time. In contrast, Keller's (1993) paper is associated with branding (.35), has a mean score of 1.0, but displays an increasing R2M score.
Graph
Table 2. Illustrative R2M Measurement for Academic Articles in the Marketing Journals.
| Article (Authors, Title, and Journal) | Top Five Topicsa | R2M Score over Time (1982–2019)b |
|---|
| Thomas (1982): Correlates of Interpersonal Purchase Influence in Organizations. JCR. | Market Orientation (.22)Sales Force Management (.18)Marketing Ethics (.15)Consumer Judgment (.11)Measurement Scales (.07) | |
| Cayla and Eckhardt (2008): Asian Brands and the Shaping of a Transnational Imagined Community. JCR. | Consumer Culture (.52)Household Expenditure (.17)Branding (.09)New Products (.06)Market Orientation (.06) | |
| Hunt (1983): General Theories and the Fundamental Explananda of Marketing. JM. | Marketing Theory/Policy (.59)Market Orientation (.07)Consumer Culture (.06)Channel Management(.05)Empirical Estimation (.04) | |
| Keller (1993): Conceptualizing, Measuring, and Managing Customer-Based Brand Equity. JM. | Branding (.35)Information Processing (.13)Product Management (.08)Marketing Theory/Policy (.07)Consumer Judgment (.05) | |
| Park and Hahn (1991): Pulsing in a Discrete Model of Advertising Competition. JMR. | Advertising (.32)Analytical Models (.18)Marketing Strategy (.13)Empirical Estimation (.10)Market Entry (.06) | |
| Maltz and Kohli (1996): Market Intelligence Dissemination Across Functional Boundaries. JMR. | Market Orientation (.51)WOM/Social Media (.14)Sales Force Management (.09)Dynamic Models (.05)Product Management (.04) | |
| Kanetkar, Weinberg, and Weiss (1992): Price Sensitivity and Television Advertising Exposures: Some Empirical Findings. MSC. | Sales Promotion (.18) Advertising (.18)H. Purchase Behavior (.16)Construct Measurement (.13)Branding (.09) | |
| Natter et al. (2008): Practice Prize Report—Planning New Tariffs at tele.ring: The Application and Impact of an Integrated Segmentation, Targeting, and Positioning Tool. MSC. | Behavioral Segmentation (.22)Market Orientation (.20)New Products (.11)Marketing Theory/Policy (.09)Analytical Models (.07) | |
- 2 a The number in parentheses is the probability that the article is associated with the topic.
- 3 b The graph depicts the evolution of the R2M score of an article since its publication.
The mean R2M score of all articles across journals and years is.63 (SD =.29, min =.08, and max = 1.91). Mean R2M scores vary significantly across journals (F = 432.33, p <.001). Articles in JM have the highest mean R2M score (.85), followed by articles in MSC (.63), JMR (.56), and JCR (.48) (see Figure 4, Panel A); all pairwise journal comparisons are significant (all Bonferroni ps <.001). Table A2 in the Web Appendix shows the ten most relevant articles, nine of which have been published in JM.
Graph: Figure 4. R2M findings and validity checks.
Has the relevance of academic marketing articles deteriorated or improved over time? A regression analysis that controls for journal effects indicates a modest annual increase of R2M scores (β =.001, p <.001). By journal, there is a positive trend toward more relevance for MSC and JMR (MSC: β =.006, p <.001; JMR: β =.004, p <.001), a relatively stable, barely significant trend for JM (β = −.002, p =.056), and a slightly significant declining trend for JCR (β = −.003, p <.001) (see Figure A5 in the Web Appendix for the R2M trend by journal and overall). As Figure A5 shows, while JCR was comparable in relevance to MSC and JMR in the early 1980s, it has widened the gap since 2000 and now is less relevant than MSC and JMR.
To examine the drivers behind the evolution of marketing relevance across journals over time, we generated positioning maps using correspondence analysis of the averaged topic probabilities for each of the four journals and each year. For each journal and year, we computed the average topic probability of all the articles published during that year. Given the large number of topics, we focused the analysis on the topics that are most associated with the journal (these topics capture about 80% of the topic probabilities for each journal). For readability, Figure 5 portrays only the ten topics that are most associated with each of the four journals over the years. In this two-dimensional map, years are depicted at the centroid of the topics they covered, and topics are depicted at the centroid of the years in which they were published. The size of the circles highlights the prevalence of the topic in the journal in which it is published. As in a heat map, the degree of relevance to marketing (computed as the product of and Mt in Table 1) for a topic is reflected by the shade darkness of the color within the circle.
Graph: Figure 5. Correspondence analysis showing evolution of topics over time across journals.
Across journals, Figure 5 shows that the horizontal dimension (Dimension 1) captures most of the variability of the input data. This dimension aligns well with the temporal sequence of the topics. For JM, for example, there is a U-shaped ("horseshoe") pattern starting in the early and mid-1980s in the top left corner of the figure, moving toward the 1990s and early 2000s in the bottom half of the figure, and toward the end moving back to more recent periods in the top right corner. We see similar horseshoe patterns for JCR, JMR, and MSC. As indicated by [14], p. 88), the reason for the "wraparound" patterns in the figure is that the topics in the center of the horseshoe (e.g., Market Orientation and Marketing Strategy for JM) can be thought of as forming a gravitational field that captures the topics regularly published by the journal, and the topics at the periphery reflect specialized topics in the years near them (e.g., Financial Impact for 2013–2015 for JM).
Judging by the size of the topic circles, the three most prevalent topics published in JCR over the last 34 years are Construct Measurement and Measurement Scales (in the 1980s), Information Processing (in the 1990s), Consumer Culture and Consumer Judgment (in the 2000s), and Consumer Goals and Motives (in the 2010s). Judging by the shade darkness of the circles, most of the topics published in JCR have relatively low relevance to marketing. This explains why the average marketing relevance for JCR has slightly declined from 1982 to 2015.
For JM, the most studied topics are Marketing Theory and Policy (in the 1980s); Market Orientation and Marketing Strategy (in the 1990s and early 2000s); and, more recently, Customer Satisfaction/CRM and Financial Impact (metrics). Judging by the shade darkness of the circles, most of the topics published in JM have relatively high relevance to marketing throughout the period of our study. This explains why the average marketing relevance for JM has remained high throughout the period. For JMR, the most studied topics are Construct Measurement (in the 1980s), Empirical Estimation (in the late 1990s), Sales Promotions (in the 2000s), and Consumer Goals and Motives (in the 2010s). Judging by the darkness of the topic circles, the Construct Measurement and Empirical Estimation topics explain why JMR had a low R2M score early on. However, most of the topics published later have relatively higher relevance to marketing scores, explaining the improved R2M score over time. A similar pattern for MSC topics has emerged: Construct Measurement (in the 1980s); Sales Promotions and Empirical Estimation (in the 1990s and early 2000s); and, more recently, Analytical Models. The low relevance of Construct Measurement and Empirical Estimation compared with recent topics explains why MSC (like JMR) had a low R2M score early on but a higher relevance score over time.
To check the validity of the index, we tested whether articles with practical impact (based on various measures) would have higher R2M scores, whether articles in nonmarketing journals would have lower scores, and whether the index correlates with marketing practitioners' judgments. For robustness checks, we tested whether the index could be easily "gamed."
To examine the extent to which the R2M Index is predictive of practically relevant articles, we compared the mean R2M scores for practice-award papers, Roberts, Kayande, and Stremersch's (2014) list of 100 impactful papers, TPM conference submissions, and MSI working papers and reports with the R2M score of all other marketing articles in our corpus. We expect that the four sets of practically relevant articles (hereinafter, the validation papers) to have higher R2M scores than the others.
We identified 250 practice award-winning papers from 1982 to 2015 using multiple sources, including the MSI Long Term Impact Award, ISMS-MSI Practice Prize, John D.C. Little Award, Frank M. Bass Dissertation Paper Award, AMA Paul E. Green Award, Harold H. Maynard Award, Annual William F. O'Dell Award, and Sheth Foundation/Journal of Marketing Award. While some of these awards are specifically focused on relevance to practice, others use multiple criteria including practical relevance.
[33], p. 131) selected 100 "marketing science" papers published in JM, JMR, MSC, Management Science, and International Journal of Research in Marketing from 1982 to 2003 based on their high academic (citations) and practice impact (as judged by practitioners).
We analyzed the submissions (175 in total) to the 2019 TPM conference hosted by Columbia Business School in collaboration with JM. Each submission includes about ten PowerPoint slides in which authors describe their research problem, method, findings, and how the research links theory and practice. In the call for conference submissions, TPM stresses the importance of addressing substantive business problems with broad relevance and sound methodology.
We used a set of 218 MSI working papers and reports published from 1997 to 2017. These papers are screened by MSI for their practical relevance and are targeted to practitioners.
An analysis of variance with publication year as covariate shows that there is a significant difference in the R2M scores of the five sets of papers (the validation papers and other marketing papers in our corpus) (F = 57.23, p <.001). We control for year of publication because Roberts, Kayande, and Stremersch's (2014) list of 100 impactful papers was published much earlier (1982–2003) than the MSI working papers and reports (1997–2017) and TPM conference submissions (2019). Figure 4, Panel B, depicts the mean R2M scores (adjusted for time). The mean R2M scores of the validation papers are significantly higher than the mean R2M score of all other marketing articles (all Bonferroni ps <.001). Among the validation papers, the only significant difference is between the mean R2M score of award-winning papers and that of MSI working papers and reports.
To assess the discriminant validity of the R2M Index, we calculated R2M scores on a holdout set of 700 articles randomly drawn from American Psychologist (132 articles), American Economic Review (131 articles), Psychological Review (99 articles), and Quarterly Journal of Economics (120 articles). These basic discipline journals were selected as relevant because marketing often draws from psychology and economics. If the R2M Index is valid, then marketing journals should have higher scores than nonmarketing journals. A one-way analysis of variance with year of publication as covariate shows that this is the case (F = 126.3, p <.0001; all pairwise Bonferroni ps <.01). Figure 4, Panel C, reports the time-adjusted R2M mean scores: marketing has a significantly higher R2M score than both economics and psychology (p <.001), and psychology has a significantly lower R2M score than economics (p <.001).
To examine the extent to which our R2M score correlates with practitioners' judgment, we sought MSI's help to survey trustees during the March 2018 Trustees Meeting as well as executives enrolled in an EMBA program.
We presented practitioners with five pairs of article abstracts (including the article title) and asked them to indicate the article in each pair they thought was more relevant to the practice of marketing. One abstract in each pair was drawn randomly from the set of articles ranked in the top third by our R2M Index while the other was drawn from the bottom third. The order of presentation of the pair of articles was randomized across the five choice tasks and across practitioners. We also asked them to indicate their occupation and marketing expertise.
The survey was sent to 70 MSI corporate members; only 14 respondents completed it (a 20% response rate). The practitioners were senior marketing managers (e.g., Marketing Director, Chief Marketing Officer, Vice President of Marketing) with expertise in various positions (e.g., brand management, new product development, marketing strategy) and working in different industries (e.g., pharmaceuticals, online media, consumer package goods, telecommunications, financial services, technology). On average, they had 18.5 years of marketing experience. Three of them did not disclose their background information, and one respondent indicated being a marketing professor and was excluded from the sample. In total, our statistical analysis was based on a sample of 65 observations (13 practitioners, each performing five choice tasks).
We conducted a logistic regression where the dependent variable indicates which article in a pair the practitioner judged to be more relevant, and the independent variable is the difference between the R2M scores of the two articles in the pair. The regression coefficient was positive and significant (β = 2.1, p <.001). In addition, the likelihood ratio test shows that the model fit difference between an intercept-only model and the R2M model is significant (χ2 = 82.3, p <.001). The hit rate is 71%, which is higher than the 50% chance criterion. In summary, while the sample is small, the statistical results provide evidence for the validity of the R2M measurement.
We also validated the index with 101 EMBA executives in a course in spring 2020. On average, the respondents had more than 9 years of business experience (3.5 years in a marketing-related job). We asked how much their job related to marketing; the mean was 3.4 on a 7-point scale. They had taken 2.3 marketing courses, on average. Because MSI trustees had reported that it was cumbersome for them to read five pairs of abstracts, we used a different procedure with the EMBA participants. We presented them with 20 triplets of article titles (no abstracts) and asked them to identify the title of the article in each triplet that they deemed most relevant and the one that was the least relevant to the practice of marketing. The first title in each triplet was drawn randomly from the set of articles ranked in the top third by our R2M Index; the second title was drawn from middle third; and the third title was drawn from the bottom third. The order of presentation of each triplet of articles was randomized across the 20 choice tasks and across EMBA participants.
To assess the validity of the R2M Index, we estimated an ordinal logit model where article rank in the triplet is the dependent variable and the R2M score of the article is the independent variable. In a choice set, there are 3! = (3 × 2) possible ways to order the three articles. Thus, there is a 16.67% (= 1/6) chance to randomly predict the correct ordering. We obtain a hit rate of 40.2% and a positive and significant R2M coefficient (β = 1.15, p <.001). This hit rate is significantly higher than chance. The likelihood ratio test shows that the model fit difference between an intercept-only model and the R2M model is significant (χ2 = 153.4, p <.001). Though modest, the results provide further evidence for the validity of the R2M Index.
Can the R2M Index be "gamed"? That is, similar to companies trying to affect Google PageRank outcomes, can authors increase the relevance of a research paper by simply adding words from our dictionary? To examine this possibility, we randomly sampled 200 journal articles from our corpus of 4,229 articles (about 5% of the articles). To "game the index," we first augmented the text of each of the 200 articles by adding 10% of the terms in our marketing dictionary. For each article, these additional marketing terms were randomly drawn from the 512 most relevant terms in our dictionary (i.e., those with marketing-term relevance weight rw = 1). We then used our LDA estimates to predict the topic probabilities for each article and calculate the new R2M score. If the index can be "gamed," we would expect the R2M scores for these 200 articles to be higher with "gaming" than without. Without "gaming" the mean R2M score is.64 and with "gaming" the mean R2M score across these 200 articles is.67. This score is only slightly higher and not significant (t = 1.35, p >.18). The robustness of the index stems from measuring the relevance to marketing at the topic rather than the article level because it is much more difficult to "game" the relevance to marketing of a collection of articles (which are subsumed under a topic) than that of a single article.
Using a text-mining methodology, we developed the R2M 1.0 Index to measure the relevance of academic marketing articles to marketing practice. This dynamic measure fulfills the desiderata for an ideal measure compiled by [ 1]. The index can be summarized by a single number: the mean R2M score of an article. The index and its key components (topical and timely relevance) are grounded in theory, in particular, in the field of information science. The index is also diagnostic and predictive and captures the potential of an article such as the likelihood of winning practice-related awards and being seen as relevant by practitioners. In addition, it is an easy-to-use, objective measure rather than a subjective judgment, and is based on readily available data (i.e., publicly available articles and information). Finally, the index is reliable and robust against "gaming" and has been validated against several other measures of relevance.
We found that the relevance of academic articles has slightly increased over time. We also found that articles published in JM have higher R2M scores than MSC and JMR, with JCR being the least relevant. In addition, of all journals JM has the highest percentage of topics covered in its articles in the Highlands quadrant, and the lowest percentage of topics in the Academic Island. The leadership role of JM in terms of practical relevance seems to be due to JM's long-standing commitment to serve not only academics but also practitioners. To continue this legacy, the current editor in chief has decided to focus on publishing marketing research that has important implications for firms, policy makers, and other societal stakeholders (see https://www.ama.org/editorial-guidelines-journal-of-marketing). In addition, the scores revealed that MSC and JMR have become more relevant over time, and JCR has become slightly less relevant. The positive trend for MSC and JMR is consistent with research indicating that quantitative marketing has responded well to the emergence of new industries and the availability of new data by introducing new relevant topics ([14]; [28]). The lower practical relevance of JCR and its slight decline in relevance over time may indicate that, during our time period, the journal deliberately chose to position itself as a purely academic publication. Our analysis also indicated that publications in JCR heavily focused on less relevant topics such as Information Processing, Consumer Judgment, and Consumer Goals and Motives and did not include more relevant marketing-related topics such as Branding, Online Marketing, New Products, and Innovation.
This first version of the R2M Index has limitations that call for future refinement. First, we only analyzed articles in four marketing journals from 1982 to 2015. Future research should include a broader set of marketing journals and longer periods. Second, because it was our goal to cover all marketing content, our dictionary of relevant terms is rather exhaustive but also quite long. Future research should create a shorter dictionary and customize the index to the needs of particular journals and audiences (e.g., focusing on technology relevance, marketing strategy relevance, public policy relevance, or consumer relevance). Third, we empirically observed that the dictionary entries became quite stable between 2000 and 2010; however, new terms will emerge, and the dictionary needs to be updated. We recommend frequent updating of publications in practitioners' outlets and a periodic major overhaul of the dictionary. As new academic articles are published, they should also be included and scored. Fourth, further efforts should be expanded to validate the index with more managers and increase the robustness against "gaming." Fifth, in the long term, the index could be made more sophisticated using machine learning and artificial intelligence by using an algorithm to decipher the meaning of marketing terms by learning interrelations among terms and topics. In summary, the R2M Index version 1.0 is a starting point, but also a work in progress that requires further refinement.
In addition, relevance is only one indicator of the merit of an academic marketing article. Other important indicators may include how often an article is cited, whether it is widely shared (e.g., because it provides a critical theoretical, substantive, or methodological contribution), and whether it is intellectually stimulating and inspiring to read. We therefore propose that scholars assess the relationship of the R2M Index to other article measures such as citation counts, altmetrics, and writing-clarity measures ([48]). Most importantly, topical and timely relevance are distinct from business impact. While relevance is a necessary condition for impact in business, impact is a broader concept ([17]; [39]). A measure of impact may require the creation of a system that includes marketing practitioners as judges, employs measurement scales centered on behavioral and organizational change, and objectively assesses impact as part of a theory-to-practice chain. We urge researchers to develop such an "impact-on-marketing" measurement system to assess which academic ideas and knowledge have significantly changed marketing practice.
We view the R2M Index as a useful and relevant tool for multiple stakeholders. We conducted two focus groups with senior marketing scholars, asking them to brainstorm about how the index could be used by various stakeholders.[ 6]
Important stakeholder groups include marketing managers in for-profit firms as well as marketing decision makers in nonprofit firms and public policy organizations. These groups can use the index to identify articles that satisfy their informational needs. In addition, market intermediaries such as market research firms and consultants can use the measure to identify relevant articles and compile curated relevant readings to satisfy their clients' needs for relevant information. Similarly, the media and social influencers who play an important role in bringing relevant academic work to the attention of decision makers can employ the index and the topics derived from it to screen academic articles and identify new trends in business and academic literature. Rating and accreditation agencies may also find the index useful for ranking business school research in marketing. For example, regulators in the United Kingdom, Australia, and the Netherlands have introduced systems that require universities to demonstrate relevance and impact ([16]).
Inside academia, authors who are concerned about the practical relevance of their research could use the R2M index to assess the topical and timely relevance of their research as they market their research among the scholarly community and pitch it to business and media outlets. Editors of academic journals could use the index as a tracking device to assess the status quo of their journals and to position or reposition them based on the results. Editors could also employ the index to evaluate papers and make suggestions during the review process to identify topics of relevance, and as a screening device for awards. Finally, department heads and business school deans could utilize the index for fundraising and as part of promotion and tenure decisions.
In summary, the index is valuable for a wide range of marketing stakeholders. Following the dictum ascribed to Peter Drucker, "What gets measured, gets done," we view the index not only as a measurement and assessment instrument but also as a tool for change. In this vein, we hope that the index will be broadly embraced so marketing can fulfill its mission as an applied discipline and publish articles in academic journals that serve marketing practitioners' needs. We encourage scholars in other fields of business that have examined and debated the relevance of their own academic research (such as management, information systems, and accounting) to develop similar indices to show which scholarly work in their field is relevant to business and has the potential to change business practice. Toward this endeavor, we created a web application (www.R2Mindex.com) where users can score articles on R2M and search for practically relevant research.
The reader may wonder how the present article scores on the R2M Index. The answer is:.89, which is slightly above average in relevance to marketing and high for a measurement article. (And we promise we have not "gamed" the score.)
Footnotes 1 Stefan Stremersch
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Malek Ben Sliman https://orcid.org/0000-0002-0294-4828
5 As part of this process, ten senior marketing executives from various industries also reviewed a preliminary version of our dictionary and crossed out any terms that they did not consider to be marketing terms. As a result, we eliminated 32 terms (e.g., "EBA," "valence," "volitional," "choice heuristic," "referent power") from the dictionary. In the survey, we also asked practitioners to suggest other terms they thought were missing from the dictionary. No additional terms were proposed.
6 The second author moderated the focus groups; the first author was present for technical clarification about the index. We thank Professors Randolph Bucklin, Sunil Gupta, V. Kumar, Don Lehmann, David Reibstein, Venkatesh Shankar, Nader Tavassoli, Rajkumar Venkatesan, Kim White, and Raghu Iyengar for their participation. All these scholars were organizers of prior TPM conferences.
References Ailawadi Kusum L. , Lehmann Donald R. , Neslin Scott A.. (2003), " Revenue Premium as an Outcome Measure of Brand Equity ," Journal of Marketing , 67 (4), 1 – 17.
Bagozzi Richard P. , Yi Youjae. (1991), " Multitrait-Multimethod Matrices in Consumer Research ," Journal of Consumer Research , 17 (4), 426 – 39.
Benbasat Izak , Zmud Robert W.. (1999), " Empirical Research in Information Systems: The Practice of Relevance ," MIS Quarterly , 23 (1), 3 – 16.
Berger Jonah , Humphreys Ashlee , Ludwig Stephan , Moe Wendy W. , Netzer Oded , Schweidel David A.. (2020), " Uniting the Tribes: Using Text for Marketing Insight ," Journal of Marketing , 84 (1), 1 – 25.
Berlund Pia. (2003), " The Concept of Relevance in IR ," Journal of the American Society for Information Science and Technology , 54 (10), 913 – 25.
Blei David M. , Ng Andrew Y. , Jordan Michael I.. (2003), " Latent Dirichlet Allocation ," Journal of Machine Learning Research , 3 (January), 993 – 1022.
7 Bucklin Randolph E. , Gupta Sunil. (1999), " Commercial Use of UPC Scanner Data: Industry and Academic Perspectives ," Marketing Science , 18 (3), 247 – 73.
8 Cayla Julien , Eckhardt Giana M.. (2008), " Asian Brands and the Shaping of a Transnational Imagined Community ," Journal of Consumer Research , 35 (2), 216 – 230.
9 Grewal Rajdeep , Gupta Sachin , Hamilton Rebecca. (2019), " The Journal of Marketing Research Today: Spanning the Domains of Marketing Scholarship ," Journal of Marketing Research , 57 (6), 985 – 98.
Gulati Ranjay. (2007), " Tent Poles, Tribalism, and Boundary Spanning: The Rigor-Relevance Debate in Management Research ," Academy of Management Journal , 50 (4), 775 – 82.
Gupta Sunil , Hanssens Dominique , Hauser John R. , Lehmann Donald , Schmitt Bernd. (2014), " Introduction to Theory and Practice in Marketing Conference Special Section of Marketing Science ," Marketing Science , 33 (1), 1 – 5.
Harter Stephen P. (1992), " Psychological Relevance and Information Science ," Journal of the American Society for Information Science , 43 (9), 602 – 15.
Huang Xiaoli , Soergel Dagobert. (2013), " Relevance: An Improved Framework for Explicating the Notion ," Journal of the American Society for Information Science and Technology , 64 (1), 18 – 35.
Huber Joel , Kamakura Wagner , Mela Carl F.. (2014), " A Topical History of JMR ," Journal of Marketing Research , 51 (1), 84 – 91.
Hunt Shelby D. (1983), " General Theories and the Fundamental Explananda of Marketing ," Journal of Marketing , 47 (4), 9 – 17.
Jack Andrew. (2020), " Academic Focus Limits Business Schools' Contribution to Society ," Financial Times (February 23), https://www.ft.com/content/5953739c-3b94-11ea-b84f-a62c46f39bc2.
Jaworski Bernard J. (2011), " On Managerial Relevance ," Journal of Marketing , 75 (4), 211 – 24.
Kanetkar Vinay , Weinberg Charles B. , Weiss Doyle L.. (1992), " Price Sensitivity and Television Advertising Exposures: Some Empirical Findings ," Marketing Science , 11 (4), 359 – 71.
Kaplan Robert S. (2011), " Accounting Scholarship that Advances Professional Knowledge and Practice ," The Accounting Review , 86 (2), 367 – 83.
Keller Kevin L. (1993), " Conceptualizing, Measuring, and Managing Customer-Based Brand Equity ," Journal of Marketing , 57 (1), 1 – 22.
Kohli Ajay , Haenlein Michael. (2021), " Factors Affecting the Study of Important Marketing Issues: Implications and Recommendations ," International Journal of Research in Marketing , 38 (1), 1 – 11.
Kotler Philip T. , Keller Kevin L.. (2011), Marketing Management , 14th ed. Upper Saddle River, NJ : Pearson.
Kumar V. (2017), " Integrating Theory and Practice in Marketing ," Journal of Marketing , 81 (2), 1 – 7.
Lehmann Donald R. , McAlister Leigh , Staelin Richard. (2011), " Sophistication in Research in Marketing ," Journal of Marketing , 75 (4), 155 – 65.
Lilien Gary L. (2011), " Bridging the Academic–Practitioner Divide in Marketing Decision Models ," Journal of Marketing , 75 (4), 196 – 210.
MacInnis Deborah J. , Folkes Valerie S.. (2010), " The Disciplinary Status of Consumer Behavior: A Sociology of Science Perspective on Key Controversies ," Journal of Consumer Research , 36 (6), 899 – 914.
Maltz Elliot , Kohli Ajay K.. (1996), " Market Intelligence Dissemination Across Functional Boundaries ," Journal of Marketing Research , 33 (1), 47 – 61.
Mela Carl F. , Roos Jason , Deng Yiting. (2013), " A Keyword History of Marketing Science ," Marketing Science , 32 (1), 8 – 18.
Mizzaro S. (1998), " How Many Relevances in Information Retrieval? " Interacting with Computers , 10 (3), 303 – 20.
Natter Martin , Mild Andreas , Wagner Udo , Taudes Alfred. (2008), " Practice Prize Report—Planning New Tariffs at tele.ring: The Application and Impact of an Integrated Segmentation, Targeting, and Positioning Tool ," Marketing Science , 27 (4), 600 – 609.
Park Sehoon , Hahn Minhi. (1991), " Pulsing in a Discrete Model of Advertising Competition ," Journal of Marketing Research , 28 (4), 397 – 405.
Reibstein David J. , Day George , Wind Jerry. (2009), " Guest Editorial: Is Marketing Academia Losing Its Way? " Journal of Marketing , 73 (4), 1 – 3.
Roberts John H. , Kayande Ujwal , Stremersch Stefan. (2014), " From Academic Research to Marketing Practice: Exploring the Marketing Science Value Chain ," International Journal of Research in Marketing , 31 (2), 127 – 40.
Rust Roland T. , Cooil Bruce. (1994), " Reliability Measures for Qualitative Data: Theory and Implications ," Journal of Marketing Research , 31 (1), 1 – 14.
Saracevic Tefko. (1975), " Relevance: A Review of and a Framework for the Thinking on the Notion of Information Science ," Journal of American Society for Information Science , 26 (6), 321 – 43.
Saracevic Tefko (2007a), " Relevance: A Review of the Literature and a Framework for Thinking on the Notion in Information Science. Part II: Nature and Manifestations of Relevance ," Journal of the American Society for Information Science and Technology , 58 (13), 1915 – 33.
Saracevic Tefko (2007b), " Relevance: A Review of the Literature and a Framework for Thinking on the Notion in Information Science. Part III: Behavior and Effects of Relevance ," Journal of the American Society for Information Science and Technology , 58 (13), 2126 – 44.
Schamber Linda , Eisenberg Michael B. , Nilan Michael S.. (1990), " A Re-Examination Relevance: Toward a Dynamic Situational Definition ," Information Processing and Management , 26 (6), 755 – 76.
Schmitt Bernd H. (2012), " Bridging Theory and Practice: A Conceptual Model of Relevant Research ," in Cracking the Code: Leveraging Consumer Psychology to Drive Profitability , Posavac Steve , ed. Armonk, NY : M.E. Sharpe.
Shapiro Debra L. , Kirkman Bradley L. , Courtney Hugh G.. (2007), " Perceived Causes and Solutions of the Translation Problem in Management Research ," Academy of Management Journal , 50 (2), 249 – 66.
Stremersch Stefan. (2021), " The Study of Important Marketing Issues: Reflections ," International Journal of Research in Marketing , 38 (1), 12 – 17.
Stremersch Stefan , Winer Russell S. , Camacho Nuno. (2021), " Faculty Research Incentives and Business School Health: A New Perspective from and for Marketing ," Journal of Marketing (published online April 28), https://doi.org/10.1177/00222429211001050.
Thomas Robert J. (1982), " Correlates of Interpersonal Purchase Influence in Organizations ," Journal of Consumer Research , 9 (2), 171 – 82.
Van Heerde Harald J. , Moorman Christine , Page Moreau C. , Palmatier Robert W.. (2021), " Reality Check: Infusing Ecological Value into Academic Marketing Research ," Journal of Marketing , 85 (2), 1 – 13.
Vermeulen Freed. (2005), " On Rigor and Relevance: Fostering Dialectic Progress in Management Research ," Academy of Management Journal , 48 (6), 978 – 82.
Wallach Hanna , Murray Iain , Salakhutdinov Ruslan , Mimno David. (2009), " Evaluation Methods for Topic Models ," in Proceedings of the 26 International Conference on Machine Learning. New York : Association for Computing Machinery , 1105 – 12.
Wang Xin (Shane) , Bendle Neil T. , Mai Feng , Cotte June. (2015), " The Journal of Consumer Research at 40: A Historical Analysis ," Journal of Consumer Research , 42 (1), 5 – 18.
Warren Nooshin L. , Farmer Matthew , Gu Tianyu , Warren Caleb. (2021), " Marketing Ideas: How to Write Research Articles that Readers Understand and Cite ," Journal of Marketing (published online May 7), https://doi.org/10.1177/00222429211003560.
Wilson Deirdre , Sperber Dan. (2004), " Relevance Theory ," in Handbook of Pragmatics , Ward Lawrence , Horn Gregory , eds. Oxford, UK : Blackwell , 607 – 32.
Winer Russell S. (1999), " Experimentation in the 21st Century: The Importance of External Validity ," Journal of the Academy of Marketing Science , 27 (3), 349 – 58.
Wittink Dick R. , Cattin Philippe. (1989), " Commercial Use of Conjoint Analysis: An Update ," Journal of Marketing , 53 (3), 91 – 6.
~~~~~~~~
By Kamel Jedidi; Bernd H. Schmitt; Malek Ben Sliman and Yanyan Li
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 104- Reality Check: Infusing Ecological Value into Academic Marketing Research. By: van Heerde, Harald J.; Moorman, Christine; Moreau, C. Page; Palmatier, Robert W. Journal of Marketing. Mar2021, Vol. 85 Issue 2, p1-13. 13p. 1 Diagram. DOI: 10.1177/0022242921992383.
- Database:
- Business Source Complete
Record: 105- Real-Time Brand Reputation Tracking Using Social Media. By: Rust, Roland T.; Rand, William; Huang, Ming-Hui; Stephen, Andrew T.; Brooks, Gillian; Chabuk, Timur. Journal of Marketing. Jul2021, Vol. 85 Issue 4, p21-43. 23p. 4 Diagrams, 6 Charts, 3 Graphs. DOI: 10.1177/0022242921995173.
- Database:
- Business Source Complete
Real-Time Brand Reputation Tracking Using Social Media
How can we know what stakeholders think and feel about brands in real time and over time? Most brand reputation measures are at the aggregate level (e.g., the Interbrand "Best Global Brands" list) or rely on customer brand perception surveys on a periodical basis (e.g., the Y&R Brand Asset Valuator). To answer this question, brand reputation measures must capture the voice of the stakeholders (not just ratings on brand attributes), reflect important brand events in real time, and connect to a brand's financial value to the firm. This article develops a new social media–based brand reputation tracker by mining Twitter comments for the world's top 100 brands using Rust–Zeithaml–Lemon's value–brand–relationship framework, on a weekly, monthly, and quarterly basis. The article demonstrates that brand reputation can be monitored in real time and longitudinally, managed by leveraging the reciprocal and virtuous relationships between the drivers, and connected to firm financial performance. The resulting measures are housed in an online longitudinal database and may be accessed by brand reputation researchers.
Keywords: brand driver; brand reputation tracker; corporate reputation; customer equity; relationship driver; social media mining; Twitter; value driver
Social media is playing an increasingly important role in enabling conversations about brands ([15]; [21]; [25]). In the social media age, listening online to how brands are talked about is critical for brand management because the comments are from people who care about brands, and they are real-time and dynamic.
Many brand measures currently exist, but they are mostly at the aggregate level, are survey-based, and are typically only available annually. Examples of these are the Interbrand "Best Global Brands" list, the Forbes "The World's Most Valuable Brands" list, the Kantar Millward Brown "BrandZ Top 100 Global Brands" list, among many others. These brand measures rank leading companies in terms of their overall brand value.
Other brand measures (e.g., the Y&R Brand Asset Valuator) measure multiple dimensions of customer brand perceptions using surveys, but the actionability of the dimensions is limited, given that the measured dimensions (differentiation, relevance, esteem, and knowledge) do not map in a straightforward way to strategic business decisions. Nevertheless, researchers who have attempted to research brand perceptions over time have resorted to such measures (and the Brand Asset Valuator in particular) to conduct their research (e.g., [23]; [33]; [35]; [51]; [52]).
To provide a brand measure that exploits this new social environment, we develop a real-time longitudinal brand tracker using social media. The tracker provides a new window into what stakeholders think and feel about brands, in their own voice. Such data provide timely, actionable information about how firms should manage brands. The tracker facilitates both longitudinal analysis and exploration of brands in a real-time and granular way.
This new metric is compiled by mining Twitter data for 100 leading global brands for comments on specific drivers and subdrivers on the basis of the customer equity framework ([46]; [45]). The database is available to all academic researchers for performing research on the antecedents and consequences of brand reputation changes over time, at a level of granularity not previously available. Table W1 in the Web Appendix compares the proposed brand reputation tracker with existing brand measures to illustrate the gap.
Our brand reputation tracker contributes to brand management by providing actionable data and analytics in a straightforward manner that facilitates strategic brand decisions. Results from our dynamic multivariate panel vector autoregression (VAR) model show that the three drivers have a brand–value reciprocal relationship as well as a brand–relationship–value virtuous circle. A firm thus can leverage either the reciprocal relationship or the virtuous circle, drawing on whichever driver(s) for which the firm has better leverage. We further demonstrate that the three drivers have real-time, short-term, and longer-term impact, collectively and individually, on abnormal stock returns. The findings provide important implications for managing brands on the basis of the dynamics and the tempos of the drivers.
The tracker also builds on the literature connecting social media mining and marketing. Data mining of Twitter feeds and other social media, for example, have been used to measure brand sentiment ([21]) and other marketing-relevant metrics (e.g., [48]; [53], [54]). By applying our methodology to three social media platforms (Twitter, Facebook, and Instagram) that are distinct in data and nature of interaction, we demonstrate that the three sets of trackers converge—evidence that social media mining can be used to track brand reputation in real time and that social media fluctuations reflect important brand events.
Our work also adds to the literature and management of corporate reputation by bridging brand reputation and corporate reputation and by providing actionable drivers for managing corporate reputation. Brand reputation is similar to corporate reputation when a firm uses a branded-house strategy (e.g., Google), and it is a component of corporate reputation when a firm uses a house-of-brands strategy (e.g., Procter & Gamble). While the corporate reputation literature has established the importance of corporate reputation as a strategic asset ([13]; [14]; [28]; [43]) and as a driver of financial performance and firm value ([41]; [42]), that literature has tended not to focus on marketing actions. We show that the time series of brand reputation, drivers, and subdrivers captures important brand and firm events; for example, the fluctuations of Facebook's brand and value drivers time series coincide with its unauthorized account licensing scandal in March 2018, while Google's innovative subdriver time series captures its late 2018 announcement of updating many algorithms. This shows that brand reputation, a key element of, and sometimes equal to, corporate reputation, can be monitored using the tracker and managed by the drivers, which helps drive firm financial performance.
The tracker also contributes to the customer equity literature. To date, most customer equity research has either relied on cross-sectional surveys for data collection (e.g., [17]; [38]; [39]; [45]; [57]) or sacrificed granularity by using aggregate acquisition and retention statistics (e.g., [18]; [27]; [56]; [58]). We show that the tracker correlates significantly positively with other well-known aggregate brand rankings, such as Interbrand, Forbes, and BrandZ, and correlates significantly positively with YouGov's daily brand measures of brand word of mouth (WOM) and brand buzz. This demonstrates that the customer equity literature can be renewed with our new social media tracking methodology that links modern social media marketing actions to customer value and facilitates longitudinal analysis and exploration of customer equity drivers in a more granular way. We also broaden the conceptual nature of customer equity to include all stakeholders, rather than just current and future customers of the brand.
In the following sections, we first conceptualize brand reputation drawing on multiple conceptual sources. Second, we develop our social media tracking method to track and monitor brand reputation. Third, we present empirical evidence based on 130 weeks' tracking data that show that this brand reputation tracker can be used to monitor and manage brand reputation and competition dynamics and is accountable for a firm's abnormal stock returns. Fourth, we generalize the tracker to multiple social media platforms and validate the tracker using other brand measures. Finally, we discuss implications for brand managers and researchers.
We define brand reputation as the overall impression of how stakeholders think, feel, and talk about a brand. This is typically due to brand events that affect firm financial performance. This definition has the following characteristics: ( 1) it is about all stakeholders (current and potential customers, employees, partners, and investors), not just the current or potential customers; ( 2) it has thinking, feeling, and talking components (not just knowledge about brands); ( 3) it can reflect actual brand events (e.g., controllable marketing activities, uncontrollable public events); and ( 4) it connects to firm financial performance.
We highlight "stakeholders" to indicate a broader view of customers, which is an integration of the corporate reputation and the marketing literature. This view is customer-centric, but is broader, considering all stakeholders as customers, including current (current and churning current customers), potential (competitors' customers and noncustomers), internal (employees), and external (investors and partners) customers. For example, [ 4] considers both external customers and internal employees as relevant for building service brand equity. [20] similarly urge the need to take a broader view of the stakeholders of marketing strategy by including the investor community as a customer. [37] demonstrate that social media emotional WOM influences investors' decisions on holding a firm's stocks. This broader view of customers reflects that brand reputation can be perceived by nonrelationship brand stakeholders who can influence the brand's financial performance.
Brand reputation should be based on whatever stakeholders say about a brand, meaning what is explicitly expressed about their thinking and feeling, not what is implicitly inferred. Stakeholders on social media can discuss anything about a brand. It can be brand experience, opinions about brand events, or simply personal sentiments about a brand. It can be positive, neutral, or negative in varying degrees. The overall impression of a brand may be summarized by what stakeholders say about a brand on social media ([21]).
Changes in brand reputation typically result from actual brand events. These brand events can be controllable marketing actions and activities as well as uncontrollable public events about brands. This characteristic emphasizes the actionability of brand reputation, allowing marketers to actively manage a brand's reputation and to track the reputation for risk and crisis management. Such actionability is the focus of models of return on marketing ([45]).
Brand reputation should be value relevant, that is, connecting to a firm's financial performance. This value relevance reflects investors' expectations about the financial value of current, potential, internal, and external customers to the firm. Value-relevant brand reputation thus is a corporate asset and a driver of firm financial performance, as defined in the management literature (e.g., [13]).
Next, we compare brand reputation with other related concepts in the management and the marketing literature streams to illustrate the conceptual nature of brand reputation. Figure 1 illustrates the relationship between brand reputation and other related concepts. It shows that our brand reputation concept lies in the intersection between the concepts of corporate reputation, brand equity, and customer equity. It has consequences for all of the firm's stakeholders (not just customers), focuses on brand thinking and feeling, and emphasizes marketing actions to drive firm value.
Graph: Figure 1. Conceptual sources of brand reputation.
In the management literature, corporate reputation is generally defined as an overall appraisal of a company by its stakeholders, which is the result of the company's past actions and predictions about the company's future ([13]). Such an overall appraisal can be thinking and/or feeling based, pertaining to, for example, the company's quality or capability ([ 6]), people's admiration of the company ([11]), or people's knowledge and emotions about the company ([19]). With its broad scope, taking all stakeholders into consideration ([ 3]), and with the appraisal being at the corporate level, corporate reputation has a broader and higher-level focus on components such as leadership, social responsibility, product and service, workplace/employee, and management/financial performance as well as their consequences on firm performance ([40]; [42]; [44]).
Compared with corporate reputation, which is the overall appraisal of a firm held by stakeholders that has consequences for firm performance, brand reputation is the analogue for companies that use a branded-house strategy (e.g., FedEx, Google, Apple). In addition, it is a component of corporate reputation when the company uses a house-of-brands strategy (i.e., having multiple brands to represent the company; e.g., Procter & Gamble).
The brand equity literature emphasizes customers' overall impression about a brand, even though there is disagreement on the scope of the impression. Some take a broader view. [ 1] brand equity definition includes mainly brand loyalty, brand awareness, perceived quality, and brand associations, constituting various assets and liabilities associated with a brand and value derived from them. [24] customer-based brand equity concept defines brand equity in terms of the individual consumer's brand knowledge, which influences the consumer's reaction to the brand's marketing mix. Brand knowledge is broad, including all knowledge about a brand that the consumer has, such as higher-level brand awareness and brand image, and all lower-levels associations, such as brand recall and brand associations. Alternatively, some take a narrower view. [46] and [45] define brand equity as customers' subjective or emotional appraisal of a brand, above and beyond its objectively perceived value. This inconsistency in the scope of brand equity motivates some studies trying to bridge brand equity with customer equity (e.g., [16]; [29]).
Brand reputation reflects both the knowledge and emotions held by all stakeholders about a brand. Our broader view aims to capture how a broader set of stakeholders thinks, feels, and talks about a brand, not limited to whether they are current or potential customers.
Reflecting a customer lifetime value view, the existing customer equity literature focuses on the contribution of a brand's single stakeholder (i.e., customers) to a firm. [45], p. 110) define customer equity as the total of the discounted lifetime values summed over all of the firm's current and potential customers. [ 5] define customer equity as the discounted profit stream and explore customer equity in terms of the optimal trade-off between acquiring and retaining customers. [18] similarly define the value of the customer base as the expected sum of discounted future earnings.
To enhance customer value, the existing studies focus on either customer equity drivers (mostly perceptual) or marketing investments. For customer equity drivers, research has shown that the three drivers in the customer equity framework—value, brand, and relationship ([46])—can improve customer value, as captured by loyalty intentions ([57]), customer loyalty ([38]; [39]), and customer experience quality ([17]). For marketing investments, the research stream on the marketing–finance interface further links customer equity to shareholder value ([18]; [27]; [47]; [49]; [58]).
Brand reputation is similar to customer equity in that both can be driven by the customer equity drivers and linked to firm financial performance. The focus is on what firms do, in terms of customer equity drivers or marketing investments, to influence firm financial performance.
We use a multistage, theory-data iterative process to develop the tracker, detailed next and shown in Figure 2. We briefly summarize these steps and explain the logic behind them, and describe this process in detail in the next sections.
Graph: Figure 2. A multistage, theory-data iterative process to develop the tracker.
- Theory-driven text mining. Use the customer equity framework to establish a three-driver tracker. This makes the brand reputation outcome theory driven and thus explainable.
- Managerially actionable subdrivers. Develop the subdrivers of brand reputation on the basis of managerial actionability. This makes the drivers managerially relevant and actionable.
- Customized dictionaries. Generate dictionaries for the subdrivers using real-world stakeholders' own words. This extracts what they think and feel about a brand in conversation and in context.
- Real-time brand reputation tracking. Collect data in real time and from multiple social media platforms. This enables firms to respond quickly and allows the methodology to be applied across platforms.
- Leverageable driver synergies. Provide evidence for the dynamics and tempos of the three drivers. This provides managerial and theoretical insight as to the internal relationships between the drivers and how a firm can leverage the synergies of the three drivers.
- Financially accountable brand reputation. Establish the accountability of the tracker for firm abnormal returns. This makes brand reputation and its drivers financially accountable, rather than just time-series of brand reputation fluctuations.
We employ the driver structure from the customer equity framework ([45]; [46]) to develop our tracker. This framework organizes the factors driving customer lifetime value and contribution to the firm into three main drivers, which themselves may be broken down further into subdrivers. Value equity is the rational and objective aspects of a brand, such as quality and price. Brand equity is the subjective feeling that a customer has about the brand, such as brand sentiment and brand image. Relationship equity is the ties between the customer and the brand, above and beyond the value equity and brand equity, such as brand community building and personal connection.
The following considerations contribute to the choice of this framework: First, the conceptual attraction of this framework has been well recognized in the academic community and recognized with several article and book awards. Second, the three customer equity drivers have been validated conceptually and empirically in many subsequent studies, using data from multiple countries; incorporating both perceptual survey and behavioral data; and considering industries, firms, and consumer characteristics; and they have been gauged using various firm performance variables (e.g., loyalty intentions, future sales, customer experience quality) ([17]; [29]; [38]; [39]; [57]). Third, this framework was designed to map to strategic expenditures and thus has high managerial actionability. The drivers and subdrivers have been shown to link to return on marketing, an important characteristic that helps connect brand reputation to firm financial performance. Its managerial relevance is reflected in this framework being applied at many leading companies worldwide. Fourth, the value and brand drivers together capture the thinking and feeling aspects of brand reputation. It is the consensus that a brand metric should include both aspects ([23]; [32]; [57]).
Many different social media platforms are used to discuss brands (e.g., Twitter, Facebook, Instagram). To construct a dynamic tracker of sentiment about brands on social media, we chose Twitter for the following considerations: ( 1) most Twitter accounts are public, meaning that conversations on Twitter have a larger impact on public perception of the brand, whereas many other social media platforms (e.g., Facebook) default to private communications; ( 2) most brands maintain an active presence on Twitter, which means that brand conversations are continuously updated and are available for public access; and ( 3) Twitter provides a publicly available application programming interface that can identify conversations about the brands, for example, using username "@coach," rather than "coach" to identify conversations about the brand ensures precision (which is the number of relevant tweets retrieved divided by the number of all tweets).
The choice of brands to be monitored is based on various prominent industry rankings on brands. These rankings included Forbes's World's Most Valuable Brands, BrandZ's Top 100 Most Valuable Brands, Interbrand's Best Global Brands, CoreBrand's Top 100 Brand Power Rankings, Credit Suisse Research Institute's Great Brands of Tomorrow, Ad Age's Social Media Brand Ranking Top 10, UTA Brand Studio's Brand Dependence Index, and Reputation Institute's Global Reputation Pulse U.S. Top 15.
Once all the brands are tabulated, any brand that appears twice or more across the lists is added to our database as a brand to be included in the tracker. Table W2 in the Web Appendix lists the brands included in the tracker. The database consists of 100 global brands across a broad range of industries such as manufacturing, wholesale trade, retail trade, transportation and warehousing, information, finance and insurance, professional, scientific, and technical services, and accommodation and food services. Both internet brands and traditional brands, and both corporate brands and individual brands, are included.
After the list of brands is established, we identify the top Twitter username (i.e., Twitter handle) associated with each brand. In deciding which handle to use, we apply the rule of the "top returned handle," that is, the brand handle (ignoring subhandles or regional handles) returned in the top search result by searching the brand name on Twitter. If no corporate brand is found, we check @brandname to see if it is a valid handle and double-check if users mention @brandname tweets at least once in the last month. The technical details of the Twitter data collection are shown in the Web Appendix, and Table W2 in the Web Appendix lists the brand names and handles used for collecting the data.
Table 1 summarizes the final 11 subdrivers for the three brand reputation drivers, including their conceptual descriptions and the final positive and negative dictionaries used in the data collection. Specifically, they are ( 1) value driver (price, service quality, and goods quality), ( 2) brand driver (cool, exciting, innovative, and social responsibility), and ( 3) relationship driver (community, friendly, personal relationships, and trustworthy).
Graph
Table 1. Brand Reputation Drivers, Subdrivers, Descriptions, and Dictionaries.
| Driver | Subdriver | Description | Positive Dictionary | Negative Dictionary |
|---|
| Value | Price | Is the brand known for low prices, such as being cheap, affordable, having deals, bargains, discounts, and sales? | Cheap, afford, inexpens, deal, low, bargain, thrifti, reason, econom, frugal, joy, discount, pleas, sale | Expens, pricey, costly, overpr, unfair, rich, excess, extravag, high, exclus, outrag |
| Service quality | Does the brand provide high quality service, such as being competent, helpful, fast, knowledgeable, understanding, with patient and respect? | Help, great, fast, knowledge, attent, understand, easi, polit, patient, respect, prompt, compet | Rude, frustrat, terribl, slow, careless, incompet, disrespect, aw, lazi, irrit, horribl, angri |
| Goods quality | Does the brand create high quality products, such as durable, functional, strong, beautiful, and valuable? | Quality, durabl, function, excel, perfect, use, beauty, strong, valu, sturdi, luxuri, worth, long-last, best, satisfi, impress, uniqu, clean | Junk, bad, poor, wast, ugli, breakabl, worthless, flimsi, useless, disappoint, shoddi, mediocr, garbag, short-liv |
| Brand | Cool | Is the brand known for being trendy, hip, awesome, cool, stylish, and sexy? | Trendi, hip, awesom, cool, modern, stylish, current, sexi | Ordinari, lame, ancient, averag |
| Exciting | Does the brand bring a sense of excitement to its products/services, such as being fun, exciting, inspiring, and stimulating? | Fun, excit, inspir, happi, thrill, stimul, live, interest | Bore, dull, uninspir, tire, bland |
| Innovative | Is the brand new, smart, technologically advanced, intelligent, innovative, creative, novel, and cutting edged? | New, smart, invent, advanc, cut, futurist, intellig, progress, innov, technolog, creative, novel, cutting-edg | Old, old-fashion, tradit, uninterest, outdate |
| Social responsibility | Is the brand caring, benevolent, giving, and beneficial? | Benevol, give, benefici | Greedi, uncar, irrespons, evil, profit |
| Relationship | Community | Does the brand generate a sense of community, such that people are involved, together, and harmonious with the brand, and can communicate and be social with the brand? | Famili, involv, commun, social, togeth, harmoni | Cold, sad, selfish |
| Friendly | Is the brand nice, pleasant, warm, kind, open, and accommodating? | Nice, friendli, pleasant, kind, warm, welcom, trustworthi, open, accommod | Mean, unpleas, unhelp, unfriendli, aloof, nasti, arrog |
| Personal relationships | Does the brand connect personally with its stakeholders by being special, personal, intimate, and close? | Connect, special, person, intim, close, profession, comfort | Cold, distant, imperson, disconnect |
| Trustworthy | Is the brand honest, reliable, and dependable? | Honest, reliable, good, depend, trust, transpar, safe, honesti, principl, honor | Dishonest, unreli, cheat, shadi, untrustwo, deceit, decept, lie |
30022242921995172 Notes: The goods quality subdriver applies to goods brands only, while the service quality subdriver applies to all brands.
The 11 subdrivers and their dictionaries are theoretically derived and empirically validated by multiple rounds of data collection and evaluation. They capture nicely the social media language and technology, while preserving the conceptual nature of the three brand reputation drivers as laid out in [45] framework. For example, for the value driver, they have quality and price as the subdrivers, and we further refine the quality subdriver into service quality and goods quality, a reflection of the service economy and the distinctiveness of service quality from goods quality. The dictionaries of the three subdrivers contain keywords that are unique in a social media setting and use stakeholders' daily language, such as "joy" for the price subdriver, "lazi" for the service quality subdriver, and "beauty" for the goods quality subdriver.
For the brand driver, [45] have corporate citizenship and ethical standards as subdrivers, and we have social responsibility as one of the subdrivers. Our brand driver is the most social media–centric driver, with three of the four subdrivers reflecting stakeholders' usage of social media language in expressing their thinking and feeling about brands, such as "cool," "exciting," and "innovative." The dictionaries of the subdrivers also reveal the language stakeholders use, such as "sexi" for the cool subdriver, "thrill" for the exciting subdriver, "intellig" for the innovative subdriver, and "give" for the social responsibility subdriver.
For the relationship driver, [45] include community as one of the subdrivers, and we add friendly, personal relationships, and trustworthy to capture that the new information and communication technologies connect stakeholders more closely to companies and their brands. These subdrivers include both the interaction and communication process of a relationship as well as the trustworthy outcome of a relationship. The dictionaries of the subdrivers are unique in suggesting new terms when communicating with stakeholders, such as "famili" for the community subdriver, "open" for the friendly subdriver, "intim" for the personal relationships subdriver, and "transpar" for the trustworthy subdriver.
The data set covers the week of July 1, 2016, to the week of December 31, 2018, 130 weeks in total.[ 7] The brand panel data contain 13,000 brand-week observations of 100 unique brands.[ 8] We measure volume and sentiment on the three drivers and subdrivers. Table 2 presents the mean, standardization deviation, minimum, maximum, and correlations among the overall brand reputation, drivers, and subdrivers. All pairs of correlations are significant at the.001 level but are moderate in effect, indicating a balance between representativeness and uniqueness.
Graph
Table 2. Descriptive Statistics: Median, Minimum, Maximum, and Correlations Among Brand Drivers and Subdrivers.
| Variables | Mean | SD | Min | Max | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 |
|---|
| Value driver | | | | | | | | | | | | | | | | | | | |
| 1. Price | 171 | 3,086 | −91,415 | 79,450 | 1.00 | | | | | | | | | | | | | | |
| 2. Service quality | 413 | 1,531 | −5,368 | 81,851 | .12 | 1.00 | | | | | | | | | | | | | |
| 3. Goods quality | 186 | 1,075 | −3,627 | 62,281 | .29 | .43 | 1.00 | | | | | | | | | | | | |
| Brand driver | | | | | | | | | | | | | | | | | | | |
| 4. Cool | 140 | 639 | −1,976 | 36,454 | .10 | .54 | .41 | 1.00 | | | | | | | | | | | |
| 5. Exciting | 347 | 1,433 | −15,620 | 92,119 | .15 | .58 | .42 | .53 | 1.00 | | | | | | | | | | |
| 6. Innovative | 944 | 15,066 | −5,009 | 1.12 million | .09 | .47 | .42 | .52 | .45 | 1.00 | | | | | | | | | |
| 7. Social responsibility | 117 | 592 | −13,730 | 28,238 | .17 | .54 | .45 | .48 | .50 | .36 | 1.00 | | | | | | | | |
| Relationship driver | | | | | | | | | | | | | | | | | | | |
| 8. Community | 121 | 798 | −10,371 | 31,975 | .06 | .43 | .18 | .34 | .42 | .28 | .33 | 1.00 | | | | | | | |
| 9. Friendly | 143 | 637 | −10.989 | 24,851 | .04 | .48 | .37 | .47 | .46 | .43 | .39 | .29 | 1.00 | | | | | | |
| 10. Personal relationships | 182 | 870 | −1,189 | 36,573 | .12 | .57 | .41 | .48 | .49 | .43 | .45 | .40 | .44 | 1.00 | | | | | |
| 11. Trustworthy | 214 | 810 | −9,470 | 33,035 | .18 | .64 | .45 | .54 | .58 | .43 | .59 | .38 | .50 | .54 | 1.00 | | | | |
| 12. Value driver | 303 | 1,994 | −31,303 | 80,651 | .71 | .71 | .87 | .46 | .51 | .41 | .50 | .33 | .38 | .48 | .56 | 1.00 | | | |
| 13. Brand driver | 387 | 3,882 | −3,763 | 284,245 | .16 | .69 | .57 | .81 | .80 | .75 | .75 | .44 | .57 | .60 | .69 | .60 | 1.00 | | |
| 14. Relationship driver | 165 | 584 | −1,534 | 18,910 | .13 | .70 | .52 | .60 | .65 | .52 | .59 | .69 | .74 | .79 | .80 | .58 | .76 | 1.00 | |
| 15. Brand reputation | 285 | 1,648 | −10,175 | 97,036 | .39 | .80 | .73 | .71 | .74 | .64 | .70 | .55 | .64 | .71 | .78 | .83 | .90 | .89 | 1.00 |
40022242921995176 Notes: The unit of analysis is brand-week. Variables 1–11 are subdrivers, variables 12–14 are drivers, and variable 15 is the overall brand reputation. All variables have 13,000 observations, except the goods quality subdriver, which has 8,580 observations. All correlation coefficients are significant at.000. Mean, SD, Min, and Max are calculated using raw scores, and correlations are calculated using the normalized scores, shown in Web Appendix Equation W4. Negative minimums indicate the number of negative keyword mentions is greater than the number of positive keyword mentions.
Brand tweet volume can be viewed as how engaged stakeholders are with a brand. The more discussion a brand can generate, the more engaged it is with its stakeholders. Volume of social media discussion (e.g., tweets and retweets on Twitter) is considered to capture social media engagement ([10]). It has also been shown to impact brand financial results ([26]).
In our data set, the average number of tweets collected per week per brand was 14,102 (SD = 46,310, min = 0, max = 1,660,963), which is substantial, indicating that the brand engagement on social media is high. This varies highly by brand, however. For example, the brands with the largest mean weekly tweet volume were Amazon (mean = 1,660,963, the week of February 23, 2018), T-Mobile (mean = 1,180,385, the week of November 18, 2016), and Google (mean = 824,432, the week of February 23, 2018).[ 9] From the Black Friday and post-Thanksgiving weekend of November 18–20, 2016, T-Mobile rolled out a Magenta Friday promotion, offering two additional lines free to both existing and new customers. T-Mobile has 1.4 million Twitter followers, and this promotion no doubt generated hot discussion. By comparison, examples of brands with the smallest mean weekly tweet volumes were HSBC, Kraft-Heinz, and Canon, the latter of which has zero tweets. This variation is not surprising, given that the decisions to tweet about a given brand will be driven by various factors, including some that are related to the brand, but also factors that are person-related or intrinsic to the individual (e.g., [55]).
Table W3 in the Web Appendix shows that among the three drivers across all brands, in terms of average volume, the brand driver (N = 387) has a higher volume than the value driver (N = 303) and the relationship driver (N = 165). This pattern in engagement volume among the three drivers may imply that the brand driver captures brand events more quickly and closely.
In Table W3 in the Web Appendix, we show the frequencies of net, positive, and negative words, as defined in our dictionaries, and calculate the positive to negative ratio for brand reputation, drivers, and subdrivers. The higher the ratio, the more positive sentiment of the reputation, drivers, and subdrivers. We can see that, in terms of sentiment ratio (proportion of positive to negative volume), the general sentiment is positive (all greater than 1). For brand reputation, the brands in the tracker, in general, have positive sentiment (3.91). This is understandable because they are top brands worldwide. For drivers, the brand driver is the most positive (13.09), followed by the relationship driver (5.58), with the value driver having the least positive sentiment (2.34). For subdrivers, the top three positive subdrivers are exciting (20.28), cool (15.00), and innovative (14.88), whereas the least positive subdrivers are price (1.45), friendly (3.65), and community (4.36).
For subdrivers of the value driver, price (1.45) is relatively ambivalent, and service quality (7.35) and goods quality (5.23) are both quite positive. This indicates that people talk about price for both positive and negative reasons but mainly talk about quality when it is positive.
For subdrivers of the brand driver, the sentiment of all subdrivers is generally positive; especially for exciting, which has the highest ratio (20.28) among all subdrivers, indicating that it is a powerful driving force for brand reputation. The second most influential subdriver is innovative (14.88). It is the most talked-about subdriver and is highly positive.
For subdrivers of the relationship driver, the most positive subdriver is personal relationships (12.38), suggesting the importance of establishing a personal relationship with stakeholders. When people talk about this, it is overwhelmingly positive.
The brand reputation tracker can be used for monitoring brand fluctuations and tracking competitive dynamics at the brand reputation, driver, and subdriver levels. We illustrate this use using two competitive dyads: two technology service brands, Facebook and Google, and two technology goods brands, Apple and Samsung, for 2018. Results and discussions of the Apple and Samsung dyad are presented in the Web Appendix and Web Appendix Figures W1, W2a–W2c, and W3a–W3i.
Figure 3 shows the time series of their brand reputation (blue line for Facebook and red line for Google), Panels A–C of Figure 4 show the time series of the three drivers, and Panels A−D of Figure 5 show the time series of selected subdrivers.
Graph: Figure 3. Brand reputation: Facebook versus Google 2018.
Graph: Figure 4. Time series for the three drivers.
Graph: Figure 5. Time series for the selected subdrivers.
The brand reputation time series show that Facebook appears to have higher brand reputation than Google for the first three quarters of 2018. In the fourth quarter, the gap decreases.
Facebook's brand (Figure 4, Panel A, blue line) and value (Figure 4, Panel B, blue line) drivers time series show that there was a negative spike from the week of March 19, with the brand driver scores plummeting from.575 to.045 to −.218 in two weeks. This coincided with the revelation of the scandal that Facebook permitted the unauthorized licensing of 30 million people's accounts to Cambridge Analytica, a data firm used by Donald Trump's 2016 presidential campaign to target voters. From the week of September 14 to the end of the year, Facebook's brand driver scores reached a long depression with an average of −.332, compared with an average of.029 the month before it. This corresponded with the event that occurred in September, when 50 million Facebook accounts and sensitive personal data were hijacked. The larger scale and more severe data leakage of this negative event are reflected in the more enduring plunge of the brand driver, indicating that stakeholders were quite concerned about the consequences of this significant personal data hijacking. According to Facebook, the company saw that unusual activities began on September 14, and on September 28 the news came out. The brand reputation tracker captures this negative event in real time, as well as its carryover effect.
The relationship driver time series of Google (Figure 4, Panel C, red line) reveals that Google in general underperforms Facebook in this driver. This is understandable because Facebook is a social networking platform for people to establish and maintain their relationships. Nevertheless, in the last week of September and the first week of October, Google had its second-highest relationship driver score (.240) of the year, surpassing Facebook in this driver. September 27th was Google's 20th anniversary, and the company also updated many algorithms of the important services it provides.
We further investigate which subdrivers are most accountable for the ups and downs of the brand reputation, shown in Figure 5, Panels A−D. Panel A shows that Facebook outperformed Google in the exciting subdriver, whereas Panel B shows that Google did a better job in the innovative subdriver, especially in the fourth quarter of 2018, when it updated many algorithms. Both subdrivers capture stakeholders' differential perceptions about the two brands.
Panel C shows that Facebook mainly bested Google in the personal relationships subdriver, but the brand is not considered more trustworthy than Google, as shown in Panel D. This indicates that the trustworthy subdriver can be an opportunity for Google to capitalize on but should be a pain point for Facebook to deal with, especially with all the data leakage negative events.
We have shown that the tracker reflects important brand events and can be used to monitor competition. In this subsection, we further examine the internal relationship of the three drivers for managing brand reputation.
We use a rigorous dynamic multivariate VAR model, estimated with generalized method of moments, to simultaneously estimate the three drivers as a system of equations ([ 2]; [31]). The estimator is dynamic, as the current realization of the endogenous variables (i.e., valuet, brandt, and relationshipt) is influenced by their past values (i.e., valuet − 1, brandt − 1, and relationshipt − 1). The inclusion of the lagged-one-period endogenous variables considers the cumulative effect of the drivers over time.
In the model, the predictors include the lagged-one-period values of the three endogenous drivers (Yit−1). The three drivers are Helmert transformed (i.e., forward orthogonal deviation) to remove the brand-specific fixed effects. Equation 1 shows the dynamic VAR model:
Yit=Yit−1α + ui+eit,1
where
- i = Brand (there are 100 brands),
- t = Week (there are 130 data weeks),
- = A (1 × 3) vector of endogenous variables (i.e., value, brand, and relationship drivers),
- = A (1 × 3) vector of endogenous variable-specific brand fixed effects,
- = A (1 × 3) vector of idiosyncratic errors, and
- _B_α = A (3 × 3) matrix of parameters for endogenous variables to be estimated.
First, we examine whether each of the three drivers is stationary using a Fisher-type test ([ 9]). The test has the null hypothesis that all the brand time series contain a unit root. It assumes the data are generated by a first-order autoregressive process; thus, we specify an augmented Dickey–Fuller unit root test on each brand with one lag of the first-differenced driver to remove the higher-order autoregressive components of the series. To mitigate the impact of cross-sectional dependence, we also follow [30] suggestion to demean the data. All test statistics for the three drivers, respectively, are significant at the.001 level, which rejects the null hypothesis of having a unit root. Thus, the test results support that the three drivers are stationary.
Second, we carried out model selection tests to determine the optimal lag order for the model. The results suggest that the first-order panel VAR minimizes the modified Bayesian information criterion (MBIC = −226.449), the modified Akaike information criterion (MAIC = −26.400), and the Quinn information criterion (MQIC = −93.451), compared with the second-order (MBIC = −158.123; MAIC = −24.758; MQIC = −69.459) and the third-order (MBIC = −78.305; MAIC = −11.622; MQIC = −33.972) models. This is expected, given that we have weekly data and discussion about a brand on Twitter changes rapidly and frequently.
Third, we estimated the first-order panel VAR using generalized method of moments–style instruments as in [22]. Table 3 presents the results. We find that the value driver is influenced by its own lagged value (.214, p =.000), the lagged brand driver (.111, p =.000), and the lagged relationship driver (.122, p =.000). The brand driver is influenced by its own lagged value (.289, p =.000) and is marginally influenced by the lagged value driver (.030, p =.063). The relationship driver is influenced by its own lagged value (.286, p =.000) and the brand driver (.068, p =.013). Hansen's J-statistic is near zero, confirming that the model is not overidentified.
Graph
Table 3. The Mutual Impacts of Brand Reputation Drivers: Multivariate Dynamic Panel Model.
| Endogenous Variables | Valuet | Brandt | Relationshipt |
|---|
| Predictors | α (z-Value) | α (z-Value) | α (z-Value) |
|---|
| Endogenous causality | | | |
| Valuet − 1 | .214 (6.48)*** | .030 (1.86)†** | .023 (1.31)*** |
| Brandt − 1 | .111 (3.46)*** | .289 (8.91)*** | .068 (2.48)*** |
| Relationshipt − 1 | .122 (3.68)*** | .029 (1.19) | .286 (9.90)*** |
| Model statistics | | | |
| No. observations | 12,800 | | |
| No. of brands | 100 | | |
| Avg. no. of weeks | 128 | | |
| Hansen's J | ≅ 0 (d.f. = 0, p = N.A.) | | |
| Granger causality Wald test | χ2(prob.) | χ2(prob.) | χ2(prob.) |
| Valuet | | 3.461 (.063) | n.s. |
| Brandt | 11.951 (.001) | | 6.144 (.013) |
| Relationshipt | 13.560 (.000) | | |
- 50022242921995176 †p <.1.
- 60022242921995176 *p <.05.
- 70022242921995176 **p <.01.
- 80022242921995170 ***p <.001.
- 90022242921995170 Notes: N.A. = not applicable. t denotes the current value, and t − 1 denotes the lagged-one-week value of the respective variables. Instruments include all variables in the equation.
The Granger causality tests[10] confirm that the value driver marginally Granger-causes the brand driver (χ2 = 3.461, p =.063), the brand driver Granger-causes the value driver (χ2 = 11.951, p =.001) and the relationship driver (χ2 = 6.144, p =.013), and the relationship driver Granger-causes the value driver (χ2 = 13.560, p =.000).
We calculate the impulse response function (IRF) confidence intervals using 200 Monte Carlo draws based on the estimated model. Figures W4a–W4d show the relevant IRF figures. The shaded area is a 95% confidence band. The IRF figures show that, in general, the interdriver effects level off in about five to six weeks.
For the impact of the brand driver on the value driver, a shock on the brand driver creates a short-term (lagged-one-week) surge on the value driver, and this surge gradually levels off in five weeks (Figure W4a). For the impact of the value driver on the brand driver, a shock on the value driver has a real-time positive impact on the brand driver. Although it levels off quickly (Figure W4b), its full effect dissipates gradually over four to five weeks.
For the impact of the relationship driver on the value driver, a shock on the relationship driver has a short-term positive impact on the value driver (Figure W4c). Its impact is smaller than the impact of the brand driver (Figure W4a) but is about equally persistent.
For the impact of the brand driver on the relationship driver, a shock on the brand driver has a real-time positive impact on the relationship driver, which levels off more slowly than the impact of the value driver on the brand driver and persists for a longer time period, for about five to six weeks (Figure W4d).
The customer equity framework considers that the three drivers together constitute the bonds that hold the customer to the brand, but it does not specify how the three drivers causally relate to each other. Our empirical examination thus provides original empirical evidence regarding the dynamics of the three drivers. The three rectangular boxes and the white arrows in Figure 6 illustrate the mutual impacts of the drivers. The outer gray arrows depict the financial impact of the three drivers, which we discuss in the next subsection.
Graph: Figure 6. The mutual impacts of brand reputation drivers, and their synergies on abnormal stock returns.Notes: Thicker arrows indicate stronger relationships obtained from the VAR model for all brands. Curved arrows indicate the (dys)synergy obtained from the financial model. The longer-term brand × relationship × value synergy on abnormal returns is obtained from the sentiment model.
Two relationships emerge. First, we find a reciprocal relationship between the brand and the value drivers, with the impact from the brand driver to the value driver being stronger than the reverse. Second, we find a virtuous circle among the three drivers, from the brand driver to the relationship driver, from the relationship driver to the value driver, and finally from the value driver back to the brand driver. The IRF figures further reveal different tempos of the carryover effects among the three drivers.
Together, the two relationships expand our knowledge of how the customer equity drivers relate to each other over time and provide rich implications for managing brand reputation. We discuss their managerial implications in the "Discussion" section.
Previously, we showed that the time series of the brand reputation tracker can capture important brand events. Next, we further demonstrate that the unanticipated components of the tracker provide additional information to stakeholders about the firm's abnormal stock returns.
To do so, we match the brand tracker data with the firm's financial data from the Center for Research in Security Prices (CRSP). After the matching, we obtain 8,710 firm-week observations with 67 single-brand firms that trade in the U.S. stock market. In this matching, individual brands from firms that follow a house-of-brands strategy (i.e., one firm has multiple brands) are excluded (e.g., Pampers) for consistency. Table W4 in the Web Appendix summarizes the industry characteristics of the brands in the tracker, based on their two-digit North American Industry Classification System codes. It shows that we have 40.30% (N = 27) manufacturing brands and 59.70% (N = 40) service brands. Both the manufacturing and the service brands are dominated by technology and information brands, reflecting the nature of the modern information economy.
In calculating abnormal returns, we first estimate a firm's expected stock returns using the Carhart four risk factors ([ 8]; [12]) to adjust stock returns for the risk factors and to demonstrate that the drivers provide additional explanations for abnormal returns. Our calculation is similar to that of [36] and [50], as shown in Equation 2:
Rit−Rft, t=αi+βi(Rmt−Rrf, t)+siSMBt+ hiHMLt+uiUMDt+∊it,2
where Rit is firm i's actual stock return in week t, and Rft, t is the risk-free rate of return in week t. We obtain daily stock returns data from the CRSP and collapse them into weekly stock returns to match our weekly brand data. The three Fama–French factors—the risk-free market return rate (Rmt – Rrf, t), the return difference between small-firm and big-firm stocks (SMBt), and the return difference between high and low book-to-market stocks (HMLt)—are accessed from Fama and French's data library. The momentum factor (UMDt) is the return difference between portfolios of past winners and losers ([12]).
We estimate the impact of the unanticipated component of the drivers on the firm's abnormal return using Equation 3:
ARit=∑d=13∑s2(βd, t−sUXid, t−s)+∑d=12∑g=d+13∑s=02(γdg, t−sUXid, t−sUXig, t−s)+∑s=02(θt−sUXi1, t−sUXi2, t−sUXi3, t−s)+Ci1δ1+Ci2δ2+∊it,3
where
- i = Brand (1 to 67; i.e., single-brand firms),
- t = Week (there are 130 data weeks),
- ARit = Abnormal return for firm i in week t (ARit = [Rit – Rft, t] – Eretit, where expected return, Eretit, is the predicted value of Rit − Rft, t in Equation 2),
- UΔX = Unanticipated component of the drivers,
- d = Driver (1 to 3; i.e., brand, value, and relationship drivers),
- s = No. of week lag (1 to 2),
- g = Index for the other brand in the two-way interaction,
- t-s = Index for time with lag(s) (i.e., the current, lag one week, or lag two weeks),
- = A (4 × 1) vector of coefficients for the industry control variable,
- = A (2 × 1) vector of coefficients for the year control variable,
- C i1 = A (1 × 4) vector of industry dummy variables, with manufacturing industry as the base,
- _B_Ci2 = A (1 × 2) vector of year dummy variables, with year 2016 as the base,
- = The error term, and
- βd, t − s, γdg, t − s, θt − s = Parameters to be estimated
UΔX, the unanticipated component of driver is the standardized residual estimated by a fixed-effect panel model for each driver, using its lagged-one-week value as the predictor. We include lagged-one-week and two-week drivers as predictors to capture the immediate, short-delayed, and longer-delayed effects of the drivers.[11] Industry sector and year dummy are included as control variables. Industry sector is the set of industry dummies, ranging from 1 to 5 (the manufacturing sector as the baseline). Year dummy is the set of year dummies, ranging from 2016 to 2018 (year 2016 as the baseline).[12] Equation 3 is estimated using a feasible generalized least squares panel model, specifying a heteroskedastic error structure and panel-specific autocorrelation. This allows for flexible autocorrelation across brands and a brand-specific first-order autoregressive process for the error in each brand.
The estimation of Equation 2 shows that the market risk factor (Rmt − Rft, t) has a significant positive effect on stock returns (.801, p =.000), whereas the SMB factor has a significant negative effect (−.001, p =.000). The other two factors are not significant. The positive coefficient for the market risk factor shows that each firm's stock returns covary with the risk-free market returns, and the negative coefficient for the SMB factor indicates that big firms have higher returns than small firms. The constant is insignificant (−.000, p =.857), consistent with the efficient-market hypothesis.
We then estimate Equation 3 to check the accountability of the residuals of the drivers for a firm's abnormal return. Table 4 presents the results for the main-effect model and the interaction model, respectively. For the main-effect model, the residual of the brand driver has a real-time positive impact on abnormal returns (.001, p =.034), the residual of the value driver has a short-term positive impact (.001, p =.085), and the residual of the relationship driver has real-time (−.002, p =.028) and short-term (−.002, p =.001) negative impacts but a longer-term positive impact on abnormal returns (.001, p =.079). The information sector has higher abnormal returns than other sectors (.001, p =.013). We find no significant year effect.
Graph
Table 4. Accountability of the Unanticipated Component of Brand Reputation Drivers for Abnormal Stock Returns.
| IMain-Effect Model | IIInteraction Model |
|---|
| Predictor | Estimate | z-Value | Estimate | z-Value |
|---|
| Brand Drivers | | | | |
| UΔBrandt | .001*** | 2.12 | .002*** | 1.96 |
| UΔBrandt − 1 | .001*** | 1.00 | .001*** | 1.05 |
| UΔBrandt − 2 | .000*** | .30 | −.000*** | −.30 |
| UΔValuet | .000*** | .72 | .000*** | .33 |
| UΔValuet − 1 | .001*** | 1.72 | .001*** | 1.13 |
| UΔValuet − 2 | −.001*** | −.95 | −.001*** | −.97 |
| UΔRelationshipt | −.002*** | −2.19 | −.000*** | −.06 |
| UΔRelationshipt − 1 | −.002*** | −3.37 | −.003*** | −3.29 |
| UΔRelationshipt − 2 | .001*** | 1.76 | .001*** | .57 |
| UΔ(Brand × Value)t | | | .001*** | 1.70 |
| UΔ(Brand × Value)t − 1 | | | −.000*** | −.66 |
| UΔ(Brand × Value)t − 2 | | | .000*** | .01 |
| UΔ(Value × Relationship)t | | | −.001*** | −1.47 |
| UΔ(Value × Relationship)t − 1 | | | .001*** | 1.87 |
| UΔ(Value × Relationship)t − 2 | | | .000*** | .61 |
| UΔ(Brand × Relationship)t | | | −.001*** | −1.94 |
| UΔ(Brand × Relationship)t − 1 | | | .000*** | .48 |
| UΔ(Brand × Relationship)t − 2 | | | .001*** | .84 |
| UΔ(Brand × Value × Relationship)t | | | .000*** | 1.17 |
| UΔ(Brand × Value × Relationship)t − 1 | | | −.000*** | −1.43 |
| UΔ(Brand × Value × Relationship)t − 2 | | | −.000*** | −1.13 |
| Industry | | | | |
| Wholesale/retail | .001*** | 1.07 | .001*** | 1.04 |
| Transport/warehouse | −.001*** | −.37 | −.001*** | −.36 |
| Information/finance/professional/scientific | .001*** | 2.47 | .001*** | 2.35 |
| Accommodation/food | .000*** | .20 | .000*** | .19 |
| Year | | | | |
| 2017 | .000*** | .70 | .000*** | .58 |
| 2018 | −.001*** | −1.21 | −.001*** | −1.27 |
| Model Details | | | | |
| Adjusted R-square | | | | |
| Wald χ2(d.f.) | 39.56 (15)*** | 52.66 (27)*** |
- 100022242921995170 *p <.1.
- 110022242921995170 **p <.05.
- 120022242921995170 ***p <.01.
- 130022242921995170 Notes: t denotes the current week value, t − 1 denotes the lagged-one-week value, and t − 2 denotes the lagged-two-week value of the respective variables. UΔ denotes the unanticipated component of brand drivers, estimated as the standardized residual using the lagged-one-week value of the respective variable (i.e., t − 1) as the predictor in a fixed-effect panel model.
Results from the interaction-effect model show that the residual of the brand × value interaction has a real-time positive impact on abnormal returns (.001, p =.089), the residual of the value × relationship interaction has a short-term positive impact on abnormal returns (.001, p =.062), but the residual of the brand × relationship interaction has a real-time negative impact on abnormal returns (−.001, p =.052).
We then take the sentiment of the drivers into consideration by estimating Equation 3, but we separate the residual of the positive and negative sentiments of the drivers into two models.
For the negative sentiment model, we find that the residual of the negative value driver has a short-term negative effect (−.002, p =.041), the residual of the negative relationship driver has a longer-term negative effect (−.003, p =.006), and the residual of the negative brand driver has no impact. The residual of the negative brand × value interaction has a longer-term negative impact (−.001, p =.041), and the residual of the negative relationship × value interaction has a real-time negative impact (−.001, p =.095).
For the positive sentiment model, we find that the residual of the positive value driver has a short-term positive effect (.002, p =.025), the residual of the positive relationship has a short-term negative effect (−.004, p =.000), and the residual of the positive brand driver has no impact. The residual of the positive brand × relationship interaction has a negative effect (−.002, p =.051), but the residual of the positive brand × value × relationship interaction has a marginal positive effect (.000, p =.108).
Together, the results from the two sentiment models show that the negative sentiment of drivers matters more, and more consistently. Furthermore, the marginal positive impact of the three-way interaction from the positive sentiment model confirms the brand–relationship–value virtuous relationship between the drivers found in the VAR model, indicating that this virtuous relationship can be accountable for a firm's abnormal returns.
The analysis demonstrates that the three drivers provide additional information for a firm's risk-adjusted abnormal stock returns in real time, the short term, and the longer term. The results show that the brand driver has a real-time impact and is the dominant driver for abnormal returns, the value driver has a short-term impact and synergizes with the other two drivers, and the relationship driver has a longer-term impact and its positive sentiment synergizes with the other two drivers. This pattern of impact is consistent with the dynamics of the three drivers observed in the previous section.
The impact of the brand driver is more real-time, reflecting that the driver captures stakeholders' immediate sentiments to brand events or activities, as demonstrated by Hewett et al.'s finding (2016) that online WOM echoes fast and wide in an "echoverse" of the brand's communication.
The impact of the value driver is more short-term, reflecting that quality and cost do not fluctuate as frequently as brand feelings. This driver reflects the knowledge aspect of a brand, which, according to the brand equity literature (e.g., [24]), can be expected to be more stable than emotional reactions to brand events. Its positive and negative sentiments provide separate information for a firm's abnormal returns, indicating the need for monitoring both the positive and negative discussions about a brand. Its synergy with the brand driver also indicates that the objective aspects of a brand's reputation (e.g., price, quality) need to be associated with positive feelings with the brand (e.g., cool, exciting) to benefit a firm's abnormal returns.
The impact of the relationship driver is longer-term and hinges more on the positive sentiment, reflecting that relationships take time to play out, but the stock market may be myopic with respect to longer-term marketing impacts (e.g., [23]; [34]). Once a good reputation on this driver is built, it benefits a firm's abnormal returns in the long run, as shown in the longer-term brand–relationship–value synergy for the positive sentiment of the drivers.
Together, the results suggest that the residual of the drivers provides information value for a firm's abnormal returns, immediately or in a delayed manner, individually and collectively. The dark curved arrows in Figure 6 summarize the analysis. Thus, by monitoring the fluctuations of the drivers, stakeholders can have a more accurate picture about a firm's financial performance.
We validate the brand reputation tracker using three approaches. First, we replicate our methodology using two additional social media platforms, Facebook and Instagram, each of which has idiosyncratic features: Facebook focuses on social networking, and Instagram focuses on photo sharing. Second, we establish a nomological relationship with the survey-based YouGov brand data, from which we demonstrate that the tracker is related to YouGov's brand WOM and brand buzz, and leads to YouGov's purchase intention. Third, we demonstrate that the tracker correlates significantly positively with three aggregate annual brand measures, Interbrand, Forbes, and BrandZ, showing that the tracker not only converges with the aggregate annual measures but also provides more granular information (both in terms of time interval and drivers) for brand reputation.
We collected data from Facebook and Instagram, from January 1 to June 30, 2018 (i.e., 26 weeks), for the seven internet brands in the tracker. The data were collected using Crimson Hexagon, with a method similar to what was described in the "Social Media Tracking Method" section. However, the data were more difficult and problematic to collect and analyze, because many brand posts on Facebook are not publicly available, and Instagram concentrates on visual data.
We applied the same dictionaries of the subdrivers to the two social media platforms. For Facebook, we focused on post contents that are available on the firm's own brand pages. For example, for the Amazon brand, its Facebook page is https://www.facebook.com/Amazon, and a sample post content is "Thank you Amazon for excellent customer service and speedy response! Keep up the great work!" For Instagram, we collected captions and comments on photos that mention the brand handles. For example, the Amazon brand's Instagram handle is @amazon, and a sample post is "@amazon...I ordered a waffle iron two days ago and they delivered it to this tiny island 30 miles out to sea so quickly."
In calculating the driver and subdriver scores, an initial screening of the data reveals that data for the Yahoo brand are problematic, because the brand posts a lot of news articles, resulting in 20–40 times more posts than the other internet brands. We thus drop the Yahoo brand from the calculation and subsequent analysis.
We use multiple methods, including descriptive statistics, correlation analysis, and repeated and mixed-measures analysis of variance, to replicate the tracker with the two social media platforms. Table 5 presents the descriptive statistics and correlation analysis for the three social media platforms, along with the overall brand reputation and the three drivers.
Graph
Table 5. Descriptive Statistics and Correlations Across Three Social Media Platforms.
| Variables | Mean | SD | Min | Max | Twitter | Facebook | Instagram |
|---|
| Brand Reputation | | | | | | | |
| Twitter | 3,172 | 4,158 | −166 | 41,078 | 1.00 | | |
| Facebook | 75 | 118 | 0 | 845 | .16 | 1.00 | |
| Instagram | 134 | 111 | 0 | 446 | .36 | .78 | 1.00 |
| Brand Driver | | | | | | | |
| Twitter | 2,854 | 3,346 | −2,325 | 23,678 | 1.00 | | |
| Facebook | 36 | 57 | −2 | 381 | .23 | 1.00 | |
| Instagram | 174 | 146 | 1 | 623 | .24 | .77 | 1.00 |
| Value Driver | | | | | | | |
| Twitter | 4,958 | 8,340 | −3,013 | 80,651 | 1.00 | | |
| Facebook | 143 | 220 | −3 | 1,364 | .29 | 1.00 | |
| Instagram | 144 | 128 | 0 | 473 | .51 | .84 | 1.00 |
| Relationship Driver | | | | | | | |
| Twitter | 1,705 | 2,354 | −1 | 18,910 | 1.00 | | |
| Facebook | 46 | 93 | −1 | 915 | −.02a | 1.00 | |
| Instagram | 86 | 70 | 0 | 354 | .21 | .65 | 1.00 |
- 140022242921995170 a Denotes the only entry in this table that is not significant at the.05 level.
- 150022242921995170 Notes: Mean, SD, Min, and Max are calculated using raw scores, and correlations are calculated using the normalized scores. Negative value means that the negative sentiment is larger than the positive sentiment.
The descriptive statistics show that Twitter has much higher volume than the other two platforms. This is because Twitter has more publicly available content (500 million tweets daily) and has more brand content than other platforms. Although Facebook also has a high volume of posts (comparable with Twitter), most are not on public pages and thus are not readily available. Instagram's posts (95 million posts daily) are mostly not brand related. The descriptive statistics confirm empirically our choice of Twitter as the social media platform of the Tracker, as articulated previously.
The correlation analysis shows that the three social media platforms converge at both the brand reputation and the driver levels. All correlations are significantly positive, except the relationship driver between Twitter and Facebook.
A repeated and mixed-measures analysis of variance, by taking into consideration the time-series nature of the measures and the brand differences, further suggests that our tracker can be replicated using other social media platforms. We treat the brand reputation (and its three drivers), respectively, as within-brand repeated measures that are between platforms. We do not find significant platform differences for brand reputation (F =.96, p =.386, d.f. = 2), but we do find significant brand differences (F = 378.63, p =.000, d.f. = 5) and brand × platform differences (F = 87.31, p =.000, d.f. = 10). The results for the three drivers are the same; all platform differences are insignificant, but brand differences and brand × platform differences are significant. The findings confirm that our tracker can be generalized to other social media platforms and suggest that brands have different social media strategies.
The results of descriptive statistics and correlation analysis provide general support for the robustness of the tracker across social media platforms, which can be text-based or visual-based. Among the three social media platforms, Twitter is more suitable for monitoring and tracking brand reputation, due to its substantially more publicly available brand-related content. Furthermore, the tracker can be replicated using the other two social media platforms at both the brand reputation and the driver levels—evidence for the generalizability of the tracker as a social media–based brand reputation tracker. This replication also supports that our conceptualization and methodology are robust even if the nature of individual social media varies. The results of the repeated and mixed-measures analysis of variance provide stronger evidence by considering the tracker's time-series nature, social media characteristics, and brand differences.
These multiple approaches to replication consistently support the robustness of the Twitter-based brand reputation tracker: the three-driver framework and the methodology we develop here can be generalized to other social media platforms.
We purchased access to YouGov's BrandIndex data and matched 71 noninternet brands[13] that are common between the two data sets for the data period (i.e., 130 weeks, 9,230 brand-week observations). YouGov's BrandIndex interviews a consumer panel about their opinions regarding three broad sets of brand metrics: brand health, media, and purchase funnel metrics. (For a detailed description of YouGov's methodology, visit https://today.yougov.com/solutions/syndicated/brandindex). Essentially, for the three broad sets of brand metrics, the media metrics are conceptually similar to our tracker. It contains brand WOM (i.e., whether the consumer has recently spoken about the brand) and brand buzz (i.e., whether the consumer has heard anything positive or negative about the brand). The purchase funnel metrics, such as purchase intention, are more appropriate as the nonfinancial outcome of the tracker. Thus, we establish the nomological relationship of the tracker with YouGov's BrandIndex by conceptualizing its brand WOM and brand buzz as concurrent variables with the tracker, and its purchase intention as the outcome variable, as shown in Figure 7.
Graph: Figure 7. Nomological relationship between the tracker and YouGov's BrandIndex.Notes: Variables on the left-hand side are from YouGov, and the brand reputation is from the tracker. The BrandIndex measures are based the brand's consumers, whereas our tracker is based all stakeholders.
A simple correlation analysis shows that the tracker's overall brand reputation correlates significantly with BrandIndex's brand WOM (.355, p =.000), brand buzz (.317, p =.000), and purchase intention (.248, p =.000). All three brand reputation drivers of our tracker also all correlate significantly with the three BrandIndex variables.
We then conducted panel regression analysis to establish the causality between the three BrandIndex variables. The results show that brand WOM has a lagged impact (.221, p =.000), while brand buzz has a concurrent impact (.072, p =.027) on purchase intention.
After establishing that the causal chain is likely to be from brand WOM to brand buzz to purchase intention, we ran a panel VAR model to explore the dynamics among BrandIndex's brand WOM and brand buzz, and the tracker's overall brand reputation. Results of the Granger causality test show that it is more likely for brand reputation (χ2 = 18.542, p =.000) and WOM (χ2 = 7.690, p =.006) to Granger-cause brand buzz, indicating that the tracker's brand reputation is likely to be a concept that encompasses BrandIndex's brand WOM and brand buzz.
Last, we ran a panel regression analysis using the current and lagged-one-week values of the tracker's overall brand reputation to predict BrandIndex's purchase intention. Results of the analysis, with standard errors adjusted for brand heterogeneity, show that the lagged brand reputation (not the concurrent one) significantly predicts consumers' intentions to purchase the brand (.019, p =.017).
We establish the nomological relationship between our tracker and YouGov's BrandIndex, with YouGov's conceptually similar metrics of brand WOM and brand buzz correlating significantly with the tracker, and with the lagged tracker being a significant predictor for consumers' intentions to purchase the brand.
There are many other brand-related measures, as listed in Table W1 in the Web Appendix. We compare the brand tracker with three other aggregate brand-related measures for 2018: the Interbrand, Forbes, and BrandZ lists (chosen for their availability) to check whether they converge. The correlational analysis shows that the three lists are highly correlated, with correlation coefficients all greater than.851 (p <.000). This indicates that they are very similar, even if they claim to use different methods of brand evaluation. The correlations of the three lists with the overall brand reputation scores are significant (p <.000) but differentiable, because the correlation coefficients range from.079 (Forbes), to.081 (BrandZ), to.127 (Interbrand), indicating our tracker converges with as well as differentiates from the other aggregate measures.
If we look at the ranking, rather than the brand value, the correlation between the three other brand measures is more discriminable. The Interbrand rank is still highly correlated with the Forbes rank (.802) but is more discriminable from BrandZ's rank (.404). The correlation between Forbes's and BrandZ's ranking also become more discriminable (.581). Meanwhile, their correlations with our tracker become higher (.335 for Interbrand;.356 for Forbes;.179 for BrandZ), suggesting that these measures converge better by using rankings.
We further explore the relationship between the 2018 Interbrand ranking and the three drivers. We regress the Interbrand ranking on the three drivers and find that the brand driver predicts Interbrand ranking most closely (.279, p =.000), followed by the relationship driver (.109, p =.015). The value driver does not predict Interbrand ranking, when all three drivers are considered.
In summary, the analysis using the Interbrand ranking provides evidence that the two ranking systems correspond for the brands included, but our tracker provides more granular measures both in cross-sectional dimension (multiple brand drivers and subdrivers) and longitudinal dimensions (weekly fluctuations).The brand reputation tracker not only provides real-time information about a brand's performance but also is more granular at the driver and subdriver levels, providing additional actionable information about a brand's performance beyond the typical annual, aggregate-level brand ranking system.
We demonstrate the various uses of brand reputation tracker data, using various methods and approaches. Our approach can be used to monitor and manage a brand's reputation over time, both at the driver and subdriver levels. The data can also be used to manage brand competition by tracking the ups and downs of drivers for major competitors, and the subdriver analyses can provide detailed insights about how to enhance or improve brand reputation. The accountability of the residual of the brand reputation drivers for abnormal returns confirms the financial implications of using the tracker to monitor and manage brands. Finally, the validation against other social media platforms and other brand-related measures provides evidence for the generalizability and external validity of the tracker. Next, we discuss implications for managers and practitioners of the tracker and propose a research agenda for researchers to leverage the availability and the methodology of the tracker.
A firm can manage the reciprocal relationship between the brand and the value drivers. Depending on which driver a firm has the comparative advantage, the firm can selectively manage one of the drivers first, and then let the effect carry over to the other driver. For example, Apple has a stronger reputation in the brand driver as being innovative, but a weaker reputation in the value driver, due to its premium price; thus, Apple could prioritize leveraging its stronger brand driver (and the innovative subdriver) and let stakeholders understand that the innovativeness is worthy of a premium price (e.g., with better service quality). Figure W4a in the Web Appendix shows that this brand-to-value carryover effect takes one week to take off but lasts for five to six weeks. Alternatively, Samsung has a stronger reputation in the value driver as being affordable while still innovative; thus, Samsung could prioritize leveraging its stronger value driver (i.e., being affordable with high goods quality) and, by doing so, influence stakeholder perceptions of its innovativeness, due to the reciprocal relationship. Figure W4b shows that this value-to-brand carryover effect occurs in real time and lasts for four to five weeks.
A firm can manage the virtuous circle among the three drivers, from the brand driver to the relationship driver, from the relationship driver to the value driver, and from the value driver back to the brand driver. For example, Apple has a strong foothold on all three drivers (shown in Figures W2a–W2c in the Web Appendix), and thus, it is in a good position to leverage this virtuous circle. For a firm that is good at one or two drivers, it can manage the brand–value reciprocity, as discussed previously. For a firm that is good at the relationship driver (e.g., a mature brand with existing loyal customers but having nothing new to be talked about on social media), the firm can manage the relationship driver to get to the value driver (e.g., generate discussion about its new product, new service, or new price), so that the firm can subsequently leverage the brand–value reciprocity. Figure W4c shows that the relationship-to-value carryover effect takes one week to take off but lasts for five to six weeks, and Figure W4d shows that the brand-to-relationship carryover effect is real-time and lasts for four to five weeks.
We find that the three drivers tend to impact a firm's financial returns at different tempos: The brand driver has a real-time impact, the value driver takes one week, and the relationship driver takes two weeks to play out.
Given that the brand driver reflects brand sentiment that can fluctuate easily with brand events, and given its dominant impact among the three drivers, a firm can boost this driver using brand events and activities, such as a new product launch, and expect a real-time impact on financial returns.
Given that the value driver reflects brand knowledge, which does not change as easily as brand sentiment, a firm can leverage this driver as a relatively more stable foundation of its brand reputation. Its synergy with the brand driver supports this strategy that brand sentiment can be manipulated by brand events, while brand knowledge can settle the sentiment into more enduring brand knowledge that can stabilize the effects of brand sentiment's ups and downs on financial returns. Brand crisis management is one example: in the case of an unexpected negative brand event (e.g., Facebook's account data leakage), given stakeholders' knowledge about Facebook, the impact of the temporarily negative brand sentiment would be settled if a stakeholder has more positive knowledge about Facebook than negative knowledge.
Given that the relationship driver reflects brand relationship, a firm can leverage this driver for long-term returns that are less subject to temporal fluctuations, though the firm needs to be patient with branding efforts for building relationship. Although the observation that relationship takes time to play out is established in the existing customer relationship literature, its long-term synergy with the other two drivers is not yet widely recognized and can be leveraged for a stable brand reputation and its impact on financial returns.
Our theory-data iterative approach to developing the tracker illustrates the importance of theory and the need for data providers to pursue academic collaborations. Most data providers that track data or provide raw or summary data do not have a sound theory to guide them about what data to track and what summary data would be valuable. For example, Crimson Hexagon tracks data but does not provide raw data, whereas YouGov provides data but allows clients to provide guidance on what data to collect. With the black-box machine learning approach continuing to advance for data tracking, making sense of data will be a pressing issue. Our methodology illustrates the importance of theory-based data tracking.
The longitudinal brand data are available for free access to the academic research community. The data can be used in a variety of ways. In this section, we provide a list of possible research agendas along with a sampling of specific research questions for future research that can successfully leverage the data. Table 6 lists the research agendas and specific research questions. We discuss these in detail next.
Graph
Table 6. Agenda of Future Research Questions Using the Tracker.
| Research Agenda | Specific Research Questions |
|---|
| 1. Social media lens to brand reputation | Why does a certain brand "attract" or "offend" consumers on social media? How do people talk about brands on social media and what language do they use in representing what brand reputation means? How to bridge traditional measures to the social media lens of brand reputation? What contributes to the volume and sentiment variations across brands and over time?
|
| 2. Granular investigation of brand reputation | What are the gains and losses of a brand's reputation along the three drivers over time and why? What are the underlying mechanisms (i.e., theories and hypotheses) that can explain why an event having differential impacts on a brand's reputation at the driver and subdriver levels?
|
| 3. Longitudinal research on brand reputation | How do brand reputation drivers and subdrivers vary and covary over time? What are the within-brand (over time) and between-brand temporal volatilities and dynamics? How do brands evolve, in a multitude of ways, between brands, across categories, and over time?
|
| 4. Brand-related events for brand reputation variation | What are the brand events and strategic marketing actions that affect brand reputation, given event, brand, and economy characteristics? How do brand-specific and general events impact brand reputation in both the short and long term? How are brands impacted by negative and positive events such as product recalls, crises and scandals, changes in C-level executives, and product launches?
|
| 5. Brand, customer, and firm characteristics for brand driver variation | How do brand, customer, and firm characteristics, individually and collectively, account for the fluctuations of the brand reputation drivers? How can the success of a brand be predicted (e.g., direct, moderate, or mediate effects) by the brand reputation drivers and subdrivers? Why does a certain brand "attract" or "offend" consumers on social media, as a function of those characteristics? How does product diffusion vary as a function of brand reputation?
|
| 6. Brands in novel classifications | How can brands be classified according to their scores on the three brand reputation drivers or the larger set of subdrivers? What is the best statistical approach for classification to use in subsequent analyses using standard multivariate statistical analysis techniques, such as cluster analysis and multidimensional scaling? What is the machine learning approach for classification pertaining to understanding differences between brands that score high versus low on various drivers of interest?
|
| 7. Brand reputation drivers and marketing/financial outcomes | What are the returns on brand reputation in terms of marketing/financial outcomes, such as customer (re)purchases and short- and long-term financial performance? What are the differential impacts of brand reputation drivers and subdrivers on various stakeholders? What are new approaches to brand valuation? What is the best time-series modeling approach to considering the endogeneity of marketing/financial outcomes and brand reputation drivers?
|
The tracker is built using social data from Twitter and is demonstrated to be generalizable to other social media platforms, which uniquely reflects how stakeholders talk and think about brands on social media. Given the growing importance of social media mentions for brands, one potential use of the data is to understand brand reputation on the basis of social media activities. In the example analyses, we find that stakeholders talk about Samsung as "cool," whereas Pepsi is not "cool." Such wordings are distinct from how brands are portrayed in traditional media but are caught uniquely by the tracker.
The tracker is applicable to multiple social media platforms that are distinct in data type (e.g., text, photo) and interaction pattern (e.g., one to many or one to one). Although different platforms have very different purposes (e.g., Facebook is mostly for social interaction between friends, Twitter is more of a "broadcast" platform), we find that our approach produces meaningful results even for very different platforms.
Given the time-series nature of the data, one area of research that can make use of the tracker is to explore variation in brand drivers and subdrivers over time. This is important because drivers of brand reputation are unlikely to be stable over time for all brands and in fact might vary considerably. The within-brand (over time) and between-brand temporal volatilities and dynamics of these measures are worth exploring as a use of the data. Ultimately, the dynamic richness of the data should open up new empirical possibilities for researchers interested in understanding how brands evolve, in a multitude of ways, between brands, across categories, and over time.
The granularity of the tracker in both the time (weekly, monthly, and quarterly) and driver (drivers, subdrivers, and sentiments) dimensions enables researchers to examine brand reputation in a much more granular way. On this front, the numerous drivers and subdrivers allow for novel theory building and testing opportunities in relation to impacts of exogenous shocks on brands. For example, a theory could predict an impact of a shock or event on one driver but not another, and empirical evidence for this could be sought using our tracker, where there are theoretically meaningful differences between the affected and unaffected drivers that (indirectly at least) shed light on the underlying mechanism for an observed effect.
Another potentially fruitful avenue for researchers is to identify brand-related exogenous events for brand reputation variation over time to see how specific events that are relevant to given brands in the tracker affect brand reputation (overall and for each of the drivers and subdrivers). One obvious application is to examine how brands are impacted by negative and positive exogenous events such as product recalls, crises and scandals, major announcements, changes in C-level executives, product launches, and other potentially significant strategic marketing actions. Although prior research has at times considered such topics, including in the context of social media (e.g., [ 7]), more work is needed to increase our understanding, in the finer granularity provided by the subdrivers.
One theoretically fruitful area is to link subtle and interesting brand, stakeholder, and firm characteristics to the variation in the brand reputation drivers. Using those data to account for the fluctuations of brand reputation drivers allows researchers to develop new models and theories about why a certain brand "attracts" or "offends" stakeholders on social media, as a function of those characteristics. We demonstrate a nomological relationship of the tracker with YouGov data, which can be one of the many approaches to link the driver variations to those characteristics.
One potential way that researchers can benefit from our tracker is to classify brands in line with their scores on the three brand reputation drivers or the larger set of subdrivers, or on the basis of statistically estimated properties such as the extent to which time series within brands are correlated/cointegrated, stationary/evolving, and so on. In addition, machine learning techniques for classification could be used to achieve a similar outcome. This can lead to the identification of new types of classes or groups for brands that might have interesting theoretical and practical implications. This could also lead to new research questions pertaining to understanding differences between brands that score high versus low on various drivers of interest.
We demonstrate some applications of the data by linking brand reputation drivers to firm abnormal stock returns. Our analysis illustrates the potential available for researchers interested in the marketing–finance interface. Given the time-series nature of our data, more complex models could be developed that allow for marketing/financial metrics not only to be considered as being influenced by brand reputation but also to have an effect on changes in the various drivers in our data. This would be interesting, as it would enable us to understand the extent to which brand reputation drives, for example, financial outcomes versus how much reputation is instead driven by firm performance.
As opposed to most aggregate ratings or rankings of brand value, the brand reputation tracker enables a more granular investigation of the components of brand reputation. Unlike survey-based attitudinal brand measures, the tracker is designed to map more directly to competing strategic marketing expenditures.
We tie to the expanding literature on mining ("listening in on") social media to obtain brand reputation insights. By developing a methodology that is applicable across multiple social media platforms and providing a longitudinal database that is granular enough to guide marketing actions, we make it much easier for researchers to tie social media posts to brand reputation.
We provide a new resource for corporate reputation research. To date, most corporate reputation research has been in the management and strategy literature streams and has placed less emphasis on marketing actions. This database enables corporate reputation research to link more naturally to marketing.
Finally, we contribute to the research of brand reputation by making our brand reputation tracker data available to facilitate longitudinal brand research. Further extending over additional time periods or additional brands could be a valuable resource for future brand research and increase our understanding of how brands work.
Our tracker is based on Twitter tweets. This gives the tracker many advantages over other social media. However, the social media environment is not static; for example, Twitter might go out of business, or a change in its terms of service might make it impossible to apply our brand tracker on the platform. In such an eventuality, researchers may need to migrate the tracker to a different social media platform. We have demonstrated the generalizability of the methodology using data from Facebook and Instagram, suggesting that our approach may be usable on other platforms as well.
Historical data are more difficult to collect. We suggest that brands that are not included in the tracker or that want to add more actionable subdrivers should be forward-looking by following our methodology and accessing the underlying data using the public free streaming application programming interface to build their own tracker. Purchasing historical data from data providers is still an option, as we did to backfill the missing data.
Although we can mine millions of tweets automatically, there is still the need to update the usernames manually. Our collection is limited to English tweets. People's use of keywords for talking about brands may also change over time and across contexts, and thus the dictionaries need to be updated periodically. We expect that with more advanced machine learning, usernames and subdriver dictionaries can be updated automatically.
We start from 14 subdrivers and refine them into a smaller set of 11 subdrivers based on the multiple-stage, theory-data iterative process. These subdrivers are shown to be applicable to brands in the data set. Subdrivers are directly actionable, and thus, a firm can explore more potentially actionable subdrivers for further enhancing the marketing relevance of the tracker to its brand.
The brand reputation tracker is a longitudinal data base of brand reputation driver and subdriver data, for 100 top global brands, based on mining Twitter tweets. Our study contributes to the literature by making brand reputation financially accountable and managerially actionable in real time and over time. The tracker is highly time-sensitive and context-specific, allowing firms to respond quickly to market stimuli. The final goal is to provide a database resource that any academic researcher can access and/or extend. We anticipate that this should increase the amount of research done on brand reputation over time, increasing our knowledge of the antecedents and consequences of the components of brand reputation. We also hope and expect that the tracker will give marketing more importance in the broader corporate reputation literature.
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921995173 - Real-Time Brand Reputation Tracking Using Social Media
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921995173 for Real-Time Brand Reputation Tracking Using Social Media by Roland T. Rust, William Rand, Ming-Hui Huang, Andrew T. Stephen, Gillian Brooks and Timur Chabuk in Journal of Marketing
Footnotes 1 Roland T. Rust, William Rand, and Ming-Hui Huang contributed equally to the article.
2 Tomas Hult
3 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
4 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the University of Oxford's Centre for Corporate Reputation, and the Center for Excellence in Service at the University of Maryland, and grants (104-2410-H-002-142-MY3, 106-2410-H-002-056-MY3, 107-2410-H-002-115-MY3) from the Ministry of Science and Technology, Taiwan.
5 Gillian Brooks https://orcid.org/0000-0002-8926-4912
6 Online supplement: https://doi.org/10.1177/0022242921995173
7 The data files are available for weekly, monthly, and quarterly data. We use weekly data for the subsequent analyses. The database may be accessed from the website of the Centre for Corporate Reputation at Oxford University's Saïd School of Business.
8 There are 7 internet brands of Alibaba, Amazon, Facebook, Google, Twitter, Yahoo, and eBay (910 brand-week observations) that are scaled separately from the remaining 93 brands in the data set due to tweets of internet brands being much more numerous, which may distort comparisons between internet and traditional brands.
9 The first day of the week is Friday, following the stock week practice (e.g., the Center for Research in Security Prices (CRSP) data and Fama and French data.
The Granger causality test is referred to more accurately as a prediction test that is a necessary but insufficient condition for causality.
In discussing the results, we always refer the effect of current value of the driver (i.e., t) as the "real-time" effect, lagged-one-week value of the driver (i.e., t − 1) as the "short-term" effect, and lagged-two-week value of the driver (i.e., t − 2) as the "longer-term" effect.
We do not include a constant term for Equation 3, because the constant is estimated in Equation 2. Nevertheless, because the constant term in Equation 2 is insignificant, including a constant term in Equation 3 does not change the results.
Internet brands are scaled separately, as noted previously.
References Aaker David A. (1995), Building Strong Brands. New York: The Free Press.
Abrigo Michael R.M., Love Inessa. (2015), "Estimation of Panel Vector Autoregression in Stata: A Package of Programs," working paper, University of Hawaii.
Argenti Paul A., Druckenmiller Bob. (2004), "Reputation and the Corporate Brand," Corporate Reputation Review, 6 (4), 368–74.
Berry Leonard L. (2000), "Cultivating Service Brand Equity," Journal of the Academy of Marketing Science, 28 (1), 128–37.
Blattberg Robert C., Deighton John. (1996), "Manage Marketing by the Customer Equity Test," Harvard Business Review, 74 (4), 136–44.
Boivie Steven, Graffin Scott D., Gentry Rich J. (2015), "Understanding the Direction, Magnitude, and Joint Effects of Reputation When Multiple Actors' Reputations Collide," Academy of Management Journal, 59 (1), 188–206.
Borah Abhishek, Tellis Gerard J. (2016), "Halo (Spillover) Effects in Social Media: Do Product Recalls of One Brand Hurt or Help Rival Brands?" Journal of Marketing Research, 53 (2), 143–60.
Carhart Mark M. (1997), "On Persistence in Mutual Fund Performance," Journal of Finance, 52 (1), 57–82.
Choi In. (2001), "Unit Root Tests for Panel Data," Journal of International Money and Finance, 20 (2), 249–72.
Colicev Anatoli, Malshe Ashwin, Pauwels Koen, O'Connor Peter. (2018), "Improving Consumer Mind-Set Metrics and Shareholder Value through Social Media: The Different Roles of Owned and Earned," Journal of Marketing, 82 (1), 37–56.
Dowling Grahame R. (2016), "Defining and Measuring Corporate Reputations," European Management Review, 13 (3), 207–23.
Fama Eugene, French Kenneth. (1993), "Common Risk Factors in the Returns on Stocks and Bonds," Journal of Financial Economics, 33 (1), 3–56.
Ferguson Tamela D., Deephouse David L., Ferguson William L. (2000), "Do Strategic Groups Differ in Reputation?" Strategic Management Journal, 21 (12), 1195–214.
Fombrun Charles, Shanley Mark. (1990), "What's in a Name? Reputation Building and Corporate Strategy," Academy of Management Journal, 33 (2), 233–58.
Fossen Beth L., Schweidel David A. (2019), "Measuring the Impact of Product Placement with Brand-Related Social Media Conversations and Website Traffic," Marketing Science, 38 (3), 481–99.
Gani Ariel Nian, Grobler Andreas. (2019), "Linking Brand Equity and Customer Equity: A System Dynamics Perspective," working paper.
Gao Lily, Melero-Polo Iguacel F., Sese Javier. (2020), "Customer Equity Drivers, Customer Experience Quality, and Customer Profitability in Banking Services: The Moderating Role of Social Influence," Journal of Service Research, 23 (2), 174–93.
Gupta Sunil, Lehmann Donald R., Stuart Jennifer Ames. (2004), "Valuing Customers," Journal of Marketing Research, 41 (1), 7–18.
Hall Richard. (1992), "The Strategic Analysis of Intangible Resources," Strategic Management Journal, 13 (2), 135–44.
Hanssens Dominique M., Rust Roland T., Srivastava Rajendra K. (2009), "Marketing Strategy and Wall Street: Nailing Down Marketing's Impact," Journal of Marketing, 73 (6), 115–18.
Hewett Kelly, Rand William, Rust Roland T., van Heerde Harald. (2016), "Brand Buzz in the Echoverse," Journal of Marketing, 80 (3), 1–24.
Holtz-Eakin Douglas, Newey Whitney, Rosen Harvey S. (1988), "Estimating Vector Autoregressions with Panel Data," Econometrica, 56 (6), 1371–95.
Huang Ming-Hui, Dev Chekitan S. (2020), "Growing the Service Brand," International Journal of Research in Marketing, 37 (2), 281–300.
Keller Kevin L. (1993), "Conceptualizing, Measuring and Managing Customer-Based Brand Equity," Journal of Marketing, 57 (1), 1–22.
Kubler Raoul V., Colicev Anatoli, Pauwels Koen H. (2020), "Social Media's Impact on the Consumer Mindset: When to Use Which Sentiment Extraction Tool?" Journal of Interactive Marketing, 50 (May), 136–55.
Kumar V., Bhaskaran Vikram, Mirchandani Rohan, Shah Milap. (2013), "Creating a Measurable Social Media Marketing Strategy: Increasing the Value and ROI of Intangibles and Tangibles for Hokey Pokey," Marketing Science, 32 (2), 194–212.
Kumar V., Shah Denish. (2009), "Expanding the Role of Marketing: From Customer Equity to Market Capitalization," Journal of Marketing, 73 (6), 119–36.
Lange Donald, Lee Peggy M., Dai Ye. (2011), "Organizational Reputation: A Review," Journal of Management, 37 (1), 153–84.
Leone Robert P., Rao Vithala R., Keller Kevin Lane, Luo Anita Man, McAlister Leigh, Srivastava Rajendra. (2006), "Linking Brand Equity to Customer Equity," Journal of Service Research, 9 (2), 125–38.
Levin Andrew, Lin Chien-Fu, Chu Chia-Shang James. (2002), "Unit Root Tests in Panel Data: Asymptotic and Finite-Sample Properties," Journal of Econometrics, 108 (1), 1–24.
Love Inessa, Ziccino Lea. (2006), "Financial Development and Dynamic Investment Behaviour: Evidence from Panel VAR," Quarterly Review of Economics and Finance, 46, 190–210.
Lovett Mitchell J., Peres Renana, Shachar Ron. (2013), "On Brands and Word of Mouth," Journal of Marketing Research, 50 (4), 427–44.
Mizik Natalie. (2014), "Assessing the Total Financial Performance Impact of Brand Equity with Limited Time-Series Data," Journal of Marketing Research, 51 (6), 691–706.
Mizik Natalie, Jacobson Robert. (2007), "Myopic Marketing Management: Evidence of the Phenomenon and Its Long-Term Performance Consequences in the SEO Context," Marketing Science, 26 (3), 361–79.
Mizik Natalie, Jacobson Robert. (2008), "The Financial Value Impact of Perceptual Brand Attributes," Journal of Marketing Research, 45 (1), 15–32.
Nam Hyoryung, Kannan P.K. (2014), "The Informational Value of Social Tagging Networks," Journal of Marketing, 78 (4), 21–40.
Nguyen Hang, Calantone Roger, Krishnan Ranjani. (2020), "Influence of Social Media Emotional Word of Mouth on Institutional Investors' Decisions and Firm Value," Management Science, 66 (2), 887–910.
Ou Yi-Chun, de Vries Lisette, Wiesel Thorsten, Verhoef Peter C. (2014), "The Role of Consumer Confidence in Creating Customer Loyalty," Journal of Service Research, 17 (3), 339–54.
Ou Yi-Chun, Verhoef Peter C., Wiesel Thorsten. (2017), "The Effects of Customer Equity Drivers on Loyalty across Services Industries and Firms," Journal of the Academy of Marketing Science, 45 (3), 336–56.
Parker Owen, Krause Ryan, Devers Cynthia E. (2019), "How Firm Reputation Shapes Managerial Discretion," Academy of Management Review, 44 (2), 254–78.
Pfarrer Michael D., Pollock Timothy G., Rindova Violina P. (2010), "A Tale of Two Assets: The Effects of Firm Reputation and Celebrity on Earnings Surprises and Investors' Reactions," Academy of Management Journal, 53 (5), 1131–52.
Raithel Sascha, Swhwaiger Manfred. (2015), "The Effects of Corporate Reputation Perceptions of the General Public on Shareholder Value," Strategic Management Journal, 36 (6), 945–56.
Rindova Violina P., Williamson Ian O., Petkova Antoaneta P., Sever Joy Marie. (2005), "Being Good or Being Known: An Empirical Examination of the Dimensions, Antecedents, and Consequences of Organizational Reputation," Academy of Management Journal, 48 (6), 1033–49.
Roberts Peter W., Dowling Grahame R. (2002), "Corporate Reputation and Sustained Superior Financial Performance," Strategic Management Journal, 23 (12), 1077–93.
Rust Roland T., Lemon Katherine N., Zeithaml Valarie A. (2004), "Return on Marketing: Using Customer Equity to Focus Marketing Strategy," Journal of Marketing, 68 (1), 109–27.
Rust Roland T., Zeithaml Valarie A., Lemon Katherine N. (2000), Driving Customer Equity: How Customer Lifetime Value is Reshaping Corporate Strategy. New York: The Free Press.
Schulze Christian, Skiera Bernd, Wiesel Thorsten. (2012), "Linking Customer and Financial Metrics to Shareholder Value: The Leverage Effect in Customer-Based Valuation," Journal of Marketing, 76 (2), 17–32.
Schweidel David A., Moe Wendy W. (2014), "Listening In on Social Media: A Joint Model of Sentiment and Venue Format Choice," Journal of Marketing Research, 51 (4), 387–402.
Skiera Bernd, Bermes Manuel, Horn Lutz. (2011), "Customer Equity Sustainability Ratio: A New Metric for Assessing a Firm's Future Orientation," Journal of Marketing, 75 (3), 118–31.
Srinivasan Shuba, Pauwels Koen, Silva-Risso Jorge, Hanssens Dominique M. (2009), "Product Innovations, Advertising, and Stock Returns," Journal of Marketing, 73 (1), 24–43.
Stahl Florian, Heitmann Mark, Lehmann Donald R., Neslin Scott A. (2012), "The Impact of Brand Equity on Customer Acquisition, Retention, and Profit Margin," Journal of Marketing, 76 (4), 44–63.
Tavassoli Nader T., Sorescu Alina, Chandy Rajesh. (2014), "Employee-Based Brand Equity: Why Firms with Strong Brands Pay Their Executives Less," Journal of Marketing Research, 51 (6), 676–90.
Tirunillai Seshadri, Tellis Gerard J. (2012), "Does Chatter Really Matter? Dynamics of User-Generated Content and Stock Performance," Marketing Science, 31 (2), 198–215.
Tirunillai Seshadri, Tellis Gerard J. (2014), "Mining Marketing Meaning from Online Chatter: Strategic Brand Analysis of Big Data Using Latent Dirichlet Allocation," Journal of Marketing Research, 51 (4), 463–79.
Toubia Olivier, Stephen Andrew T. (2013), "Intrinsic vs. Image-Related Utility in Social Media: Why Do People Contribute Content to Twitter?" Marketing Science, 32 (3), 368–92.
Villanueva Julian, Yoo Shijin, Hanssens Dominique M. (2008), "The Impact of Marketing-Induced Word-of-Mouth Customer Acquisition on Customer Equity Growth," Journal of Marketing Research, 45 (1), 48–59.
Vogel Verena, Evanschitzky Heiner, Ramaseshan B. (2008), "Customer Equity Drivers and Future Sales," Journal of Marketing, 72 (6), 98–108.
Wiesel Thorsten, Skiera Bernd, Villanueva Julian. (2008), "Customer Equity: An Integral Part of Financial Reporting," Journal of Marketing, 72 (2), 1–14.
~~~~~~~~
By Roland T. Rust; William Rand; Ming-Hui Huang; Andrew T. Stephen; Gillian Brooks and Timur Chabuk
Reported by Author; Author; Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 106- Regulating Product Recall Compliance in the Digital Age: Evidence from the "Safe Cars Save Lives" Campaign. By: Pagiavlas, Sotires; Kalaignanam, Kartik; Gill, Manpreet; Bliese, Paul D. Journal of Marketing. Sep2022, Vol. 86 Issue 5, p135-152. 18p. 1 Diagram, 5 Charts, 3 Graphs. DOI: 10.1177/00222429211023016.
- Database:
- Business Source Complete
Regulating Product Recall Compliance in the Digital Age: Evidence from the "Safe Cars Save Lives" Campaign
The unprecedented number of product recalls in recent years and subsequent low consumer recall compliance raise questions about the role of regulatory agencies in ensuring safety. In this study, the authors develop a conceptual framework to test the impact of a regulator-initiated digital marketing campaign (DMC) on consumer recall compliance. The empirical context is the launch of a nationwide DMC by the U.S. automobile industry's regulator. The analysis utilizes recall completion data from 296 product recalls active both before and after the DMC's launch. The results show that the DMC improves consumer recall compliance. In the first four quarters after it was introduced, the DMC increased the number of vehicles fixed, on average, by 20,712 per recall campaign over what would be expected without the DMC. Regarding boundary conditions, the study finds that the DMC is more effective for recall campaigns with greater media coverage and for those with older recalled products. However, the DMC's effect weakens as the time needed to repair a defective component increases. The findings should help regulators make compelling cases for greater resource allocation toward digital initiatives to improve recall compliance.
Keywords: digital marketing campaign; product recalls; public policy; regulation
Product recalls and consumer safety are regulated by government agencies in several countries, including the United States, Canada, Germany, the United Kingdom, and Japan. While issuing a recall notice to inform consumers of a potential issue is critical, far too many defective products remain unremedied long after their recall notifications have been sent to consumers. For instance, J.D. Power reports that consumer recall compliance in the U.S. automobile industry is quite low, with approximately 30%–50% of vehicles on the road at any point in time having an unrepaired safety problem (J.D. [54]). Relatedly, according to a report by the Consumer Product Safety Commission (CPSC), only 10% of children's products recalled in 2012 were successfully corrected, replaced, or returned ([44]). For example, since the 2016 recall of millions of IKEA Malm dressers, only 1% of consumers, at best, have had the unstable furniture removed and been issued a refund (Consumers [19]).
The continued use of defective products by consumers is a serious public health concern and raises the specter of injuries and fatalities. The sustained string of casualties in 2018 due to faulty Takata airbags, a safety problem for which a recall was issued in late 2014, underscores the perils of low consumer recall compliance. Unsurprisingly, regulatory agencies have been subjected to intense scrutiny and even rebuke for not taking adequate measures to improve consumer recall compliance. For example, an audit report released in 2018 by the U.S. Department of Transportation faults the National Highway Traffic Safety Administration (NHTSA) for a lack of proper oversight of the recall completion process ([52]).
While numerous factors likely contribute to poor consumer recall compliance, low consumer awareness is often noted as a key issue responsible for inaction. In fact, the NHTSA's former Associate Administrator for Enforcement stated that a lack of public knowledge is "the single greatest weakness" in successfully addressing product recalls ([31], p. 232). Similarly, NHTSA focus group interviews conducted to understand the reasons for low recall compliance found that while over 70% of consumers preferred electronic recall notifications, only 7.4% reported receiving any. Notably, 90% of the respondents mentioned that recall notifications received electronically had a greater chance of being noticed ([28]). In the past, regulatory agencies such as the CPSC have conducted targeted national campaigns to raise public awareness, support industry compliance, and improve safety in specific consumer product categories ([20]). The issue of low consumer recall awareness is so pervasive that in 2014, the CPSC tried to crowdsource solutions, announcing a contest for application developers to create tools to inform the public of consumer product recalls ([26]).
Although regulators introduce such initiatives to raise consumer awareness and reduce accidents, empirical evidence to support or refute these expectations is inconsistent. A potential concern is that campaigns by regulatory agencies might not be effective policy instruments if consumers view them as symbolic acts by government officials and are unresponsive to these efforts ([23]; [32]). In addition, prior research has found that consumers could develop reactance to regulation-related public awareness campaigns because they view such efforts as infringements on their personal freedoms ([15]; [47]). This contention is supported by popular press: a 2018 survey of over 1,500 U.S. consumers revealed that almost two-thirds of them did not believe that government-initiated product recall programs had much to do with increasing consumer safety. Instead, many consumers viewed the recall programs as government exercises in "red tape" ([62]).
The study's objective is to investigate the impact of a regulator-initiated digital marketing campaign (DMC) on consumer recall compliance. The manuscript makes two contributions to marketing literature. First, this study is the only one that we are aware of that examines whether a DMC initiated by a regulator improves consumer recall compliance. The product recall literature has predominantly focused on the stock market and sales consequences of recalls (e.g., [13]; [29]), the ability to learn from and prevent future recalls ([33]; [37]), the drivers and consequences of recall timing decisions ([25]), and how recalls can impact marketing effectiveness ([16]; [65]). Little attention has been devoted to the consumer compliance process. A couple of studies have also investigated how recall attributes, including vehicle country of origin, vehicle age, and the publicity around recalls, influence compliance ([34]; [58]). However, the impact of a regulator's efforts on consumer compliance has, to our knowledge, received virtually no attention. Our study fills this research void and offers valuable insights to policy makers.
Specifically, we examine the effectiveness of a DMC initiated by a regulatory agency. To tackle the problem of low consumer recall compliance, the NHTSA, the agency responsible for regulating consumer safety in the U.S. automobile industry, launched a full-coverage, nationwide DMC, "Safe Cars Save Lives," in January 2016. The DMC featured paid search and online display advertisements that provided consumers with links to check for open recalls and access pertinent recall remedy information online. We exploit this setting to formally test the effectiveness of a regulator-initiated DMC to improve consumer recall compliance. The empirical analysis utilizes recall completion data pertaining to 296 product recalls active both before and after the DMC's launch. Our econometric analyses account for various potential confounds such as the DMC's possible endogenous nature, time trends in recall completion, recall campaign attributes, and unobserved vehicle make and temporal characteristics. The results show that in the first four quarters after it was introduced, the DMC increased the number of vehicles fixed, on average, by 20,712 per recall campaign above what was to be expected without the DMC. This improvement in consumer recall compliance should, in turn, lead to potentially fewer vehicle crashes, casualties, and lower economic costs.
Second, we identify important boundary conditions for the DMC's effectiveness. We find that the DMC is more effective at increasing consumer compliance for recalls with greater media coverage. Although media coverage of recalls could be detrimental to the impacted brand's financial health, our finding implies that it plays a critical role in aiding the DMC to improve compliance. We also find that the DMC is more effective at increasing compliance for older (as opposed to newer) recalled products. This finding is critical for regulators who struggle to reach owners of older products using conventional communication methods. However, the DMC's impact on compliance is lower for recalls in which the defective component takes a longer time to repair. This suggests that the DMC may be unable to fully counteract the barrier of time-related inconvenience for consumers. Collectively, our findings enable regulators to make compelling cases to receive more resources for digital marketing initiatives in the future.
Regulatory agencies introduce interventions with the goal of promoting and protecting consumer welfare. Government interventions supporting public welfare are comprised of pecuniary interventions (e.g., imposing higher taxes/penalties, offering financial incentives) as well as nonpecuniary interventions (e.g., focused on educating, building awareness) ([45]). Prior research has investigated the efficacy of interventions targeted at decreasing the consumption of unhealthy products, stimulating consumer spending, and promoting preventive health screenings ([18]; [59]; [66]). The effectiveness of interventions is often assessed by examining the extent to which consumers comply with or respond to the proposed initiatives.
Prior research provides equivocal evidence regarding the effectiveness of pecuniary interventions. For example, although imposing higher taxes has been found to lower purchases of bottled water, reduce purchases of drinks with sugar additives, and decrease smoking prevalence, their associated unintended consequences may drive consumers toward more dangerous products ([ 6]; [18]; [67]). Relatedly, the "Click It or Ticket" campaign, intended to improve seat belt usage in vehicles and promote consumer safety by ticketing transgressions, has been shown to be successful ([61]). However, programs that offer consumers incentives to stimulate sales through initiatives such as "Cash for Clunkers" ([48]) or drive consumer spending through tax rebates ([59]) have been found to be generally ineffective.
Likewise, evidence from prior research assessing the efficacy of nonpecuniary interventions is also inconsistent. For example, the impact of the Nutrition Labeling and Education Act (NLEA) in altering consumers' behaviors is somewhat nebulous. [49] found that the NLEA significantly increased consumers' nutrition information processing and comprehension, although information-sensitive consumers did not report increased use of nutrition information following its implementation. Relatedly, [ 3] did not find any change in consumers' consumption-related search and recall of nutrition information after the NLEA's enactment. [55] summarizes the effects of government agency-initiated public information campaigns across various contexts and even documents consumer responses that are opposite to those intended in some situations.
The mixed evidence for the efficacy of regulator-initiated interventions (those pecuniary and nonpecuniary) documented in prior research is not satisfactory from a knowledge advancement perspective and raises important questions about the potential effectiveness of a regulator-initiated DMC. More specifically, is a regulator-initiated DMC effective in improving consumer recall compliance? Under what conditions is a DMC more or less effective?
To answer these questions, we develop a conceptual framework (depicted in Figure 1) by drawing on insights from both the health belief model (HBM) and health warning streams of research.[ 6] With conceptual origins in the public health domain, the HBM contends that for individuals to take action to prevent a detrimental outcome, they need ( 1) a cue or trigger (internal and/or external) to overcome their avoidance tendencies, ( 2) to believe that they are susceptible to its risk, ( 3) to perceive that the occurrence of the outcome would have negative consequences for them, and ( 4) to believe that taking preventative action would lead to benefits outweighing the barriers of cost, convenience, pain, and embarrassment ([11]; [36]; [56]). Drawing on these insights, our framework examines the roles of ( 1) awareness cues, ( 2) the perceived susceptibility to risk, ( 3) the perceived threat of noncompliance, and ( 4) the perceived time inconvenience of compliance in eliciting consumer recall compliance following a regulator-initiated DMC.
Graph: Figure 1. A conceptual model of the impact of a regulator-initiated digital marketing campaign on consumer recall compliance.
Our framework also draws on health warning streams of research, specifically, prior work on consumer responses to persuasive public health appeals. This stream of research suggests that increasing the frequency of health warnings might desensitize consumers to the potential hazard, resulting in either inaction or actions that are opposite to the intended warning ([55]; [63]). In our context, government regulators, retailers, manufacturers, and consumer experts are concerned that product recalls have become so frequent in various industries, including food, consumer products, and automobiles, that consumers might suffer from "recall fatigue" ([39]). Therefore, our framework also considers the impact of multiple concurrent health warnings on a consumer's compliance with a regulator's request.
Given that our goal is to investigate a DMC's effectiveness, we conceptualize a regulator-initiated DMC as the primary awareness cue intended to improve consumer recall compliance. The ability of the DMC to improve compliance depends on the extent to which consumer recall awareness is generated and the extent to which consumers are motivated to comply. Accordingly, we expect that the DMC's impact on compliance is contingent on four factors identified in the HBM and the presence of multiple concurrent health warnings for consumers. We propose that a DMC's effectiveness is moderated by ( 1) media coverage, ( 2) the age of the recalled product, ( 3) component hazard, ( 4) time for repair, and ( 5) concurrent recall campaigns.
A DMC improves consumer recall compliance by providing relevant information to consumers impacted by recalls. These consumers could be unaware that their products have been recalled but are actively searching for recall-related information, benefiting from a DMC's paid search advertisements that make them aware of existing recalls and direct them to relevant information. Other consumers may be exposed to a DMC's online display advertisements and then cued to check if their products have been recalled on the regulator's webpage.
In addition to a DMC providing the necessary information to aid consumers in complying with a recall request, media coverage of recalls could heighten a DMC's effectiveness by sensitizing consumers to the fact that their products may have defects and thereby influencing them to comply. Accordingly, we include media coverage (akin to another awareness cue, a factor in the HBM) as a moderator of a DMC's impact on consumer recall compliance.
Because owners of older (vs. newer) products are less likely to receive information about recalls directly from their products' manufacturers, they are likely cognizant of their greater susceptibility to recall risk and more responsive to a DMC that provides relevant information. Accordingly, we include the age of the recalled product (akin to the perceived susceptibility to risk, a factor in the HBM) as a moderator of a DMC's impact on compliance.
Following a DMC, a consumer learns about the hazard of a defective component, the time it will take to repair it, and the existence of concurrent recall campaigns for the product after being made aware that it has been recalled. As such, these factors are pertinent in explaining a consumer's motivation to comply with a recall request after being made aware of it through a DMC. Accordingly, we include component hazard (akin to the perceived threat of noncompliance, a factor in the HBM), time for repair (akin to the perceived time inconvenience of compliance, a factor in the HBM), and the presence of concurrent recall campaigns (akin to multiple concurrent health warnings) as moderators of the DMC's impact on consumer compliance ([36]; [55]; [63]).
Prior research suggests that marketing campaigns can be effective at improving individuals' compliance-related behaviors. For example, [66] showed that the introduction of a mass media campaign was associated with increased consumer information search, displayed by high click-through rates to a website and more referrals to medical centers. Relatedly, [69] found that patients were more likely to obtain a first prescription of a treatment following a direct-to-consumer advertising campaign designed to educate them and provide resources about health-related decisions. The positive impact of advertising initiatives by firms may even spill over to the category level, as [68] noted that advertising by a brand increased consumer drug therapy compliance of its rival brands' products. However, the positive effects of awareness campaigns documented in prior research are primarily the result of initiatives by firms, and not by regulators. One notable exception is work by [32], which found a positive effect from a government-initiated proenvironmental demarketing campaign to reduce water consumption. However, the campaign was more effective on the majority ethnic/religious group than on the minority groups.
There are also reasons to believe that a marketing campaign may not be as effective if initiated by a regulator. For example, [53] found that fewer than half of the antismoking advertisements they showed to adolescents increased their nonsmoking intentions. Similarly, research examining the National Youth Anti-Drug Media Campaign in the United States reported no effects from the initiative ([35]). Although consumers may be skeptical about government-initiated awareness campaigns related to health or safety issues, the preponderance of evidence for these campaigns' positive effects leads us to our baseline hypothesis that a DMC will improve consumer recall compliance.
- H1: A regulator-initiated DMC improves consumer recall compliance.
The effectiveness of a regulator-initiated DMC is likely to vary depending on whether a recall received media coverage. Mass media outlets are a unique voice in the marketplace that provide consumers with information that is different from that of firms or governmental agencies ([14]). Extant research argues that separate media channels could have a synergistic impact on the effectiveness of multichannel marketing campaigns ([12]; [50]). Importantly, in addition to the repetition of the message, the variation of the specificity of the message also likely affects consumers. Although both media coverage and a regulator-initiated DMC generate awareness about a recall, they also vary in the information they provide. In our context, a regulator-initiated DMC provides the specific information consumers need to determine whether they are personally affected by a recall and how to address it, whereas media coverage increases consumer awareness of a recall. We thus argue that consumers are more likely to comply with a recall notice if the information provided by a DMC is complemented by awareness generated through media coverage ([41]). This argument is also consistent with research suggesting that advertising is more effective when there is more associated publicity following a product-harm crisis ([16]; [21]). Drawing on these arguments, we hypothesize:
- H2: Media coverage positively moderates the impact of a regulator-initiated DMC on consumer recall compliance, such that the impact is stronger for recall campaigns with greater media coverage.
Consumers are, in general, more informed of product-related issues (e.g., defects, failures) when they interact with original equipment manufacturers (OEMs) during their products' warranty periods. As products age and their warranties expire, consumers' interactions with OEMs tend to be less frequent. For example, [58] propose that owners of older products are less likely to schedule routine product maintenance. In addition, the likelihood that product ownership could change hands increases as products age, reducing the probability that notifications sent from the OEM will reach the product's present owner ([ 2]; [27]). These decreased interactions with OEMs leave consumers who own older products at greater risk of being unaware that their products are the subject of active recalls. A regulator-initiated DMC can fill this communication gap for owners of older products, allowing them to receive recall-related information through paid search and online display ads. For these reasons, we expect consumers with older products, who likely perceive themselves as more susceptible to recall-related risk, to utilize a DMC's information to a greater extent than consumers owning newer products. Accordingly, we hypothesize that a DMC's impact on consumer recall compliance will be stronger for older (vs. newer) recalled products.
- H3: The age of the recalled product positively moderates the impact of a regulator-initiated DMC on consumer recall compliance, such that the impact is stronger when the recalled product is older.
The impact of a DMC on consumer recall compliance is also likely to vary depending on consumers' perceptions of the threat of not complying with a recall request. For example, [40] finds that a consumer's perception of threat is the most influential determinant of whether they seek medical care services. Similarly, consumers are more likely to comply with treatment regimens when they perceive that noncompliance would threaten their well-being ([11]). Relatedly, vehicle owners are more likely to repair products when the defective component's hazard is higher. [34] provide evidence that consumers remedy more severe defects at higher rates than less severe defects. In general, after a DMC makes consumers aware of a recall and the defective component involved, consumers should be more motivated to comply with a recall request if they perceive a high threat of noncompliance. If a defective component's perceived hazard is low, consumers are less likely to be motivated to comply with a recall request. Conversely, if the perceived hazard of a component is high, consumers are more likely to be motivated to comply. Therefore:
- H4: Component hazard positively moderates the impact of a regulator-initiated DMC on consumer recall compliance, such that the impact is stronger for high hazard components.
The effectiveness of a DMC also depends on the extent to which consumers perceive recall compliance to be convenient. Time-related costs have been shown to be a significant barrier to achieving desired levels of compliance from individuals following a request ([10]; [36]). In addition to providing other relevant information, a DMC informs consumers of the time needed to repair their products. Prior research has found that consumers perceive compliance to be more convenient if the time it takes to complete a task is shorter. For example, [ 4] find that consumers who view recycling as more convenient are more motivated to recycle; consumers weigh the costs of sacrificing time to separate recyclable items and then dispose of them if they are under the threshold of inconvenience. Similarly, [11] find time costs to be better predictors of compliance than monetary costs. Likewise, we argue that while a DMC's value lies in creating awareness that a defective component needs repair, a consumer's motivation to comply with a recall following the DMC will be lower if the amount of time required to repair the product is greater. Therefore, we hypothesize the following:
- H5: Time for repair negatively moderates the impact of a regulator-initiated DMC on consumer recall compliance, such that the impact is weaker for components requiring more time to be repaired.
The impact of a DMC on consumer recall compliance is likely to vary depending on whether there are concurrent recall campaigns for the same product. The rationale for examining concurrent recalls stems from research suggesting that repeated exposure to product warnings might result in consumers getting habituated to or even developing psychological reactance toward warning messages ([30]; [63]). Relatedly, some consumers may be skeptical of multiple instances of health-related product warnings from government information campaigns ([55]). In such situations, consumers might ignore safety warnings as they begin to perceive messages by regulatory agencies seeking compliance as less credible or even coercive. The detrimental effect of warning messages is particularly damaging for frequent users of the products that receive them ([30]). Accordingly, after a DMC makes consumers aware of recalls that have affected them, its impact on recall compliance is likely weaker for consumers facing concurrent recalls for different product issues relative to consumers facing a single recall. If the DMC makes consumers aware that their products are involved in concurrent recalls, they could become desensitized to recalls, leading to recall requests being ignored. Therefore, the impact of a DMC on recall compliance will be weaker for recalls including products involved in concurrent recall campaigns.
- H6: Concurrent recall campaigns negatively moderate the impact of a regulator-initiated DMC on consumer recall compliance, such that the impact is weaker for products involved in concurrent recall campaigns.
The setting for this study is the automobile industry in the United States. The NHTSA regulates product recalls in the U.S. automobile industry, and several of the characteristics of this setting make it an attractive one to explore our research objectives. For instance, safety in this industry is highly regulated, and manufacturers are mandated to report progress on product recall campaign completion rates periodically. As a result, longitudinal data on recall completion is publicly available, thereby facilitating a systematic investigation of recall compliance over time.
In 2014, the U.S. automobile industry experienced the largest product recall in its history. Several automobile manufacturers notified the NHTSA that they were conducting recalls to address a possible safety defect involving Takata airbag inflators. The concern was that the defective airbags were at risk of rupturing violently following a collision, hurling fiery shrapnel into drivers and passengers. Reports indicated that in addition to fatalities, hundreds of drivers had been injured. The scandal's scope escalated sharply in the following months, leading to a record-breaking 63 million vehicles recalled in 2014 and 51 million in 2015.
Against this backdrop, the NHTSA launched a nationwide DMC, "Safe Cars Save Lives," in January 2016. Even though the DMC was motivated by the Takata airbag inflator issue, the campaign was a part of the agency's effort to improve consumer compliance amongst all recall campaigns. Lamenting low recall completion, the U.S. Transportation Secretary noted that informed consumers were "allies" in improving recall compliance ([24]). The DMC was a full-coverage initiative that sought to push consumers to use the NHTSA's recall lookup webpage to check for open recalls and then fix defective vehicles quickly.
The DMC utilized paid search and online display advertisements to influence recall compliance. A key part of the campaign was the use of Google AdWords to target consumers searching for recall-related information online through keywords such as "Vehicle Recalls," "Check for Recalls," and "Is My Vehicle Recalled," among others. A primary goal of sponsored advertisements is to guide consumers searching for information to the right location ([22]). In addition, the NHTSA used online display ads on media platforms such as Facebook that also directed consumers to the NHTSA's recall lookup webpage where they could receive recall-related information. A sample online display advertisement from the DMC is presented in Figure WA1 of the Web Appendix. Between January of 2016 and March of 2017, the NHTSA spent approximately $1 million on sponsored advertisements on Google, Facebook, and other platforms ([28]).
To gain a preliminary understanding of the DMC's effectiveness, we collected data on the paid search traffic to the NHTSA's recall lookup webpage using SEMrush Analytics, a leading vendor providing a competitive research service on online marketing and advertising. Specifically, we collected data on the total number of paid search visits to the NHTSA's recall lookup webpage for every calendar quarter-year in our data period. Figure 2 depicts the paid search traffic to the recall lookup webpage before and after the DMC. The mean quarterly recall-related paid search traffic in the post-DMC period (quarter 1 [Q1] of 2016 onward) is 48,770 visits compared with 650 quarterly visits in the pre-DMC period.
Graph: Figure 2. Model-free evidence of automobile recall–related paid search traffic to the regulator's recall lookup webpage over time.
The unit of analysis for the study is the recall campaign-calendar quarter-year. The NHTSA maintains a database of every vehicle safety recall campaign issued from 1966 onward. A typical recall notice provides information on the vehicle make and models affected, the number of vehicles recalled, the nature of the defect, and the date of the recall's announcement.
The data on the cumulative number of vehicles fixed, the measure for consumer recall compliance, was sourced from the NHTSA's database. The agency mandates that manufacturers provide quarterly updates on the number of vehicles affected by a recall and the number of affected vehicles fixed. These updates allow the NHTSA to monitor recall completion on an ongoing basis. Specifically, manufacturers report the number of vehicles fixed in each calendar quarter-year, and these reports are aggregated by unique recall identification numbers. We also collected data from the NHTSA on individual recall characteristics such as the ages of the recalled vehicle models, the defective component in the vehicles, the estimated time needed to complete the repairs, and if multiple recall notifications were sent to consumers. In addition, we gathered data from LexisNexis on the number of press articles about each recall campaign to capture the media coverage of a recall. Furthermore, we collected make-level advertising and sales data from Kantar's ad$pender database and Ward's Automotive Yearbook, respectively. Table 1 presents the variables and data sources for the study.
Graph
Table 1. Data Sources and Operationalization.
| Variable Name | Operationalization | Data Source(s) | Role of Variable |
|---|
| Consumer Recall Compliance | The cumulative number of vehicles fixed in a recall campaign. | NHTSA | Dependent variable |
| Time from Recall | The number of quarters elapsed from the time a recall campaign began until the calendar quarter directly preceding the start of the DMC. | NHTSA | Explanatory variable |
| Transition to DMC | Coded as 1 if the observation is during or after the first quarter of 2016 (during which the DMC was active), and 0 otherwise. | NHTSA | Explanatory variable |
| Time Since DMC Start | The number of quarters elapsed since the NHTSA's DMC began. | NHTSA | Explanatory variable |
| Media Coverage | The number of media mentions for an individual recall campaign beginning from the day of the recall's announcement until three months afterward. | LexisNexis | Moderator variable |
| Age of Recalled Product | The age, in years, of the oldest model recalled in a recall campaign at each respective recall completion report's calendar quarter-year. | NHTSA | Moderator variable |
| Component Hazard | A binary measure of the hazard of the defective component identified in a recall campaign. Components range from 0 (low hazard) to 1 (high hazard). | NHTSA | Moderator variable |
| Time for Repair | The estimated time, in hours, needed for the dealer to complete the recommended repair. | NHTSA | Moderator variable |
| Concurrent Recall Campaigns | A binary measure coded as 1 for recall campaigns including vehicles involved in concurrent recall campaigns, and 0 otherwise. | NHTSA | Moderator variable |
| Advertising Intensity | Make advertising expenses, in dollars, normalized by make sales, in units. | Kantar ad$pender; Ward's Automotive Yearbook | Control variable |
| Multiple Recall Notifications | A binary measure coded as 1 if the recalling firm sent an additional recall notification to consumers affected by a recall campaign above and beyond the mandatory notification required by the NHTSA, and 0 otherwise. | NHTSA | Control variable |
| Recall Size | The total number of vehicles affected by a recall. | NHTSA | Control variable |
Because the goal of the study is to examine the effectiveness of the DMC launched in January 2016, we only included recall campaigns in the sample if they were active both before and after the DMC began and had recall completion data available. Accordingly, we did not include recalls that ended before the DMC began or were announced after the DMC began. The final sample includes recall completion data from 296 recall campaigns in the U.S. automobile industry from 2015 to 2017, totaling 1,809 recall campaign-calendar quarter-year observations.
We operationalize Consumer Recall Compliance, the dependent variable, as the cumulative number of vehicles fixed for individual recall campaigns on a quarterly basis. For example, if a recall campaign had 10,000 vehicles fixed in its first calendar quarter-year, the dependent variable would equal 10,000. If an additional 4,000 vehicles were remedied in the subsequent quarter, the dependent variable would equal 14,000. Given that the DMC was an ongoing effort by the NHTSA, the cumulative number of vehicles fixed measure allows us to examine the DMC's effectiveness over time.
We operationalize Media Coverage as the count of the number of articles in leading news publications that covered a recall campaign during the three months following its announcement. For example, if a recall began on January 1, 2015, and four news articles were published covering the recall from the date of its announcement to March 31, 2015, the Media Coverage variable for that recall campaign would equal 4.
We operationalize the Age of Recalled Product variable as the number of years elapsed between the manufacturing year of the oldest vehicle model in a recall campaign and the calendar quarter-year in which we observe its cumulative number of vehicles fixed. For example, consider a recall announced in January 2015 featuring two vehicle models manufactured in January 2013 and September 2014. Age of Recalled Product during the first quarter of the recall campaign would be coded as two years, since the January 2013 model is the oldest in the recall campaign. In April 2015 (the next quarter), the variable would equal 2.25 years.
To operationalize Component Hazard, we used the NHTSA's database to identify the defective component in each recall campaign and past literature to determine the component's hazard rating. Consumers learn of the nature of their vehicle's defect in recall notification letters sent by manufacturers. Extant literature has classified recalls into two categories on the basis of the hazard or severity of the issue ([ 1], [ 2]; [34]; [57]; [58]). Consistent with previous research, we operationalize Component Hazard as a dummy variable by classifying defective components related to the driving functionality of a vehicle (i.e., those that could result in a loss of vehicle control due to acceleration, steering or braking, frame corrosion, fire, or repeated stalling) as "high hazard" (coded as 1). All other components not directly related to the drivability of a vehicle are classified as "low hazard" (coded as 0). A nonexhaustive list of examples of components and their hazard ratings is reported in Table WA1 of the Web Appendix.
We operationalize the Time for Repair variable in terms of the estimated time (in hours) needed to repair the defective component provided in recall notification letters.
To operationalize Concurrent Recall Campaigns, we examined active recalls to determine if vehicles of the same make-model-year whose dates of production overlapped were involved in multiple simultaneous recalls during our observation window. For example, we used the NHTSA's database to check if 2015 Toyota Camry vehicles experienced multiple active recalls. We operationalize Concurrent Recall Campaigns as a dummy variable by coding recall campaigns including vehicles involved in concurrent recall campaigns as 1 and all other recall campaigns that did not include vehicles involved in concurrent recalls as 0.
In addition to the DMC, make-level advertising by firms could impact how aware consumers are of active recalls and thereby improve their compliance. Accordingly, we include the make-level Advertising Intensity variable in our specification to control for firm efforts to improve make awareness. We use advertising expenses data from Kantar's ad$pender database at the make level and normalize it by make sales in units using Ward's Automotive Yearbook to account for automaker efforts at improving make awareness.
Although all automakers are mandated to notify consumers about a product recall, in some cases, they issue more than one notification for the same recall. Given that multiple notifications could improve recall completion, we include the dummy variable Multiple Recall Notifications to control for this possibility. The variable takes a value of 1 if a manufacturer notifies consumers affected by a recall more than once and 0 otherwise. Furthermore, we control for the size of a recall, as the total number of vehicles impacted by a recall campaign could positively impact the cumulative number of vehicles fixed. We assemble data on the Recall Size variable from the NHTSA's quarterly recall completion reports. Finally, we control for unobserved automobile make and temporal (i.e., calendar quarter) characteristics with the Make and Quarter dummy variables, respectively.
We ground our empirical specification in the interrupted time series analysis (ITSA) framework, a technique commonly used to evaluate the effectiveness of universally implemented policies. ITSA is a quasiexperimental design typically utilized to assess the longitudinal effects of interventions with observational data where full randomization is not possible. ITSA is implemented using a regression model that includes up to three types of time-related covariates: ( 1) Time, ( 2) Transition, and ( 3) Time Since Intervention. The interpretation of the covariates is sensitive to their coding and the presence or absence of the other covariates; their inclusion in the model is theory driven. We employ an absolute coding approach, which allows the impact of an intervention to be interpreted in absolute terms (i.e., relative to zero) ([ 8]). The key identifying assumption of ITSA is that in the absence of an intervention, the slope in the outcome will remain unchanged in the postintervention period ([ 8]; [38]; [60]).
For a model that includes all three covariates and utilizes absolute coding, Time from Recall is operationalized as the number of calendar quarters elapsed from the time a recall began until the calendar quarter preceding the start of the DMC. The coefficient of Time from Recall captures the slope in the outcome before the DMC begins. Transition to DMC is coded as 1 for all observations during the DMC (Q1 of 2016 and onward) and 0 for observations before the DMC (Q1–Q4 of 2015). In the presence of Time from Recall, the coefficient for Transition to DMC captures the absolute change in the outcome relative to zero in the first time period the intervention is active (i.e., immediately after it is introduced), and its effect remains relevant in the remainder of the observation window thereafter; it is not limited to the first quarter after the DMC. Time Since DMC Start is operationalized as the number of quarters elapsed since the NHTSA's DMC began in Q1 of 2016; it is coded as 0 for all observations before Q2 of 2016. In the presence of Time from Recall and Transition to DMC, the coefficient of Time Since DMC Start captures the slope of the outcome following the DMC relative to zero. Table 2 illustrates the coding of the three ITSA time-related variables for a recall campaign beginning in Q1 of 2015 and ending in Q2 of 2017.
Graph
Table 2. Illustration of Absolute Coding of Time-Related Variables in the ITSA Framework.
| Quarter | Year | Time from Recall | Transition to DMC | Time Since DMC Start |
|---|
| 1 | 2015 | 0 | 0 | 0 |
| 2 | 2015 | 1 | 0 | 0 |
| 3 | 2015 | 2 | 0 | 0 |
| 4 | 2015 | 3 | 0 | 0 |
| 1 | 2016 | 3 | 1 | 0 |
| 2 | 2016 | 3 | 1 | 1 |
| 3 | 2016 | 3 | 1 | 2 |
| 4 | 2016 | 3 | 1 | 3 |
| 1 | 2017 | 3 | 1 | 4 |
| 2 | 2017 | 3 | 1 | 5 |
Consistent with model testing procedures outlined in previous research ([ 8]), we examine whether there are quadratic trends in the dependent variable. Prior research on prescription drug use compliance has found quadratic patterns in some instances ([11]). We compare the model fit for the specification in Equation 1 with and without quadratic pre- and post-DMC slopes. The inclusion of Time Squared and Time Since Intervention Squared decreases the Akaike information criterion for the model from 41,791 to 41,751, indicating better model fit. Accordingly, we augment the specification by including the Time Squared and Time Since Intervention Squared variables.
In our context, we argue that Consumer Recall Compliance is a function of Time from Recall, Time from Recall Squared, Transition to DMC, Time Since DMC Start, and Time Since DMC Start Squared. The Time from Recall and Time from Recall Squared variables capture the quarterly pre-DMC slope in recall compliance. Transition to DMC captures the absolute change, relative to zero, in recall compliance immediately after the DMC was introduced in Q1 of 2016. In our context, the impact of Transition to DMC is reflected over the remainder of the observation window and is not just limited to the first quarter after the DMC. Time Since DMC Start and Time Since DMC Start Squared capture the post-DMC slope in recall compliance (i.e., the impact of the DMC following its introduction).
In summary, the empirical model controls for various recall campaign-level and time-specific factors that could impact consumer recall compliance. We also account for automobile make- and calendar-quarter-specific unobserved heterogeneity using fixed effects. The empirical specification to test the main effect of the DMC on consumer recall compliance is as follows:
Graph
( 1)
where i refers to a recall campaign, t refers to time (by calendar quarter-year), and ɛ refers to the random error.
Table 3 presents the descriptive statistics for the variables. The mean media coverage in our sample is.44 articles. The mean age of the oldest vehicle in each recall is 4.09 years, with vehicles ranging from less than a month old to 19.06 years old. The mean estimated time needed to repair a defective component is 1.88 hours, with a few requiring as many as 13 hours. The mean hazard rating of the components involved in recalls is.19. While there is a wide range for the advertising intensity of different vehicle makes, the mean is.98. That is, approximately $1 is spent on advertising for every unit sold. In the full interaction model sample, the mean number of calendar quarters for a recall campaign is 6.04 (min = 3, max = 10, SD = .70). The mean number of calendar quarters in the pre-DMC period for a recall campaign in the full interaction model sample is 2.47 (min = 1, max = 4, SD = 1.17), whereas in the post-DMC period it is 3.57 calendar quarters (min = 1, max = 8, SD = 1.32).
Graph
Table 3. Descriptive Statistics.
| Variable Name | Mean | SD | Min | Max | # of Observations |
|---|
| Consumer Recall Compliance | 38,503 | 105,586 | 0 | 1,397,612 | 1,809 |
| Media Coverage | .44 | 1.25 | 0 | 10 | 1,809 |
| Age of Recalled Product (in years) | 4.09 | 3.51 | .04 | 19.06 | 1,655 |
| Component Hazard | .19 | .39 | 0 | 1 | 1,809 |
| Time for Repair (in hours) | 1.88 | 2.11 | 0 | 13 | 1,809 |
| Concurrent Recall Campaigns | .35 | .48 | 0 | 1 | 1,809 |
| Advertising Intensity (in $/unit sales) | .98 | 9.51 | 0 | 208.86 | 1,785 |
| Multiple Recall Notifications | .03 | .17 | 0 | 1 | 1,809 |
| Recall Size | 77,366 | 196,468 | 1 | 1,814,284 | 1,809 |
The results of the DMC's main effect on Consumer Recall Compliance are reported in Model 1 of Table 4. We note that the results reported in this analysis pertain to 296 recall campaigns involving issues unrelated to airbag inflators. We exclude airbag inflator–related recall campaigns from the sample to avoid a potential endogeneity issue arising from the NHTSA's unobserved efforts to improve the recall completion of airbag inflator–related recalls specifically. The discussion of endogeneity and the analyses appear in Table WA2 of the Web Appendix.
Graph
Table 4. Results for the Main Effect of the DMC on Consumer Recall Compliance.
| Model 1 |
|---|
| Main-Effects Model |
|---|
| Dependent Variable: Consumer Recall Compliance | Estimate (SE) | Hypothesis Tested |
|---|
| Intercept | −12,560.21 | |
| (27,360.13) | |
| Time from Recall | 14,172.90*** | |
| (3,674.81) | |
| Time from Recall Squared | −2,453.12** | |
| (1,168.53) | |
| Transition to DMC | 11,112.42** | H1 supported (+) |
| (4,393.39) |
| Time Since DMC Start | −1,959.86 | H1 supported (+) |
| (2,894.73) |
| Time Since DMC Start Squared | 1,748.90*** | H1 supported (+) |
| (606.74) |
| Advertising Intensity | −1.32 | |
| (209.64) | |
| Multiple Recall Notifications | 9,011.68 | |
| (20,151.71) | |
| Recall Size | .43*** | |
| (.02) | |
| Make Fixed Effects | Yes | |
| Quarter Fixed Effects | Yes | |
| N (sample size) | 1,785 | |
| R-Squared Within | .09 | |
| R-Squared Between | .74 | |
1 *p < .10.
- 2 **p < .05.
- 3 ***p < .01.
The results in Model 1 of Table 4 reveal that the coefficients for Advertising Intensity and Multiple Recall Notifications are not significant (p > .65). As we expected, the coefficient for Recall Size is positive (.43, p < .01). The coefficient of Time from Recall Squared is negative (2,453.12, p < .05). To understand the linear effect of Time from Recall, we estimate a model specification identical to Equation 1 except that the quadratic terms are dropped, as the average linear effect of a variable is not interpretable in the presence of its quadratic effect ([17], pp. 200–201). The results (not shown) indicate that the coefficient for Time from Recall is positive (8,725.57, p < .01). These results collectively suggest that consumer recall compliance increased at a decreasing rate prior to the start of the DMC.
The coefficient for Transition to DMC is positive (11,112.42, p < .05), indicating that, immediately following the DMC's introduction, the cumulative number of vehicles fixed increased in absolute terms in Q1 of 2016. In our context, the impact of Transition to DMC associated with Q1 of 2016 remains relevant in the remainder of the observation window thereafter and is not just limited to the first quarter after the DMC. The interpretation of Transition to DMC is contingent on the presence of Time from Recall in the model.
In the presence of Time from Recall, Time from Recall Squared, and Transition to DMC, Time Since DMC Start and Time Since DMC Start Squared represent the slope of consumer recall compliance in the post-DMC period. The results in Model 1 of Table 4 indicate that the coefficient for Time Since DMC Start Squared is positive (1,748.90, p < .01). As before, to understand the linear trend in Time Since DMC Start, we estimate Equation 1 without the quadratic terms in the model ([17], pp. 200–201). The results (not shown) indicate that the coefficient for Time Since DMC Start is positive ( 6,036.62, _I_p_i_ < .01). These results collectively suggest that following the DMC's introduction, consumer recall compliance increases at an increasing rate (i.e., a positive accelerating curve), in contrast to the pre-DMC trend. H1 is supported.
The visual plots in Figure 3 provide a representation of the DMC's effectiveness for an average-sized recall using the estimates provided in Model 1 of Table 4. In Figure 3, the solid line ("Number of vehicles fixed with the DMC") denotes the DMC's impact on recall compliance for a recall campaign that begins in Q1 of 2015 and ends in Q4 of 2016. The solid line shows that recall compliance increased at a decreasing rate before the start of the DMC. The DMC begins in Q1 of 2016 after four quarters have elapsed, represented by the vertical dotted line. Between four and five quarters elapsed, the solid line captures the impact of the DMC in Q1 of 2016. Following Q1 of 2016, the solid line captures both the positive linear and positive quadratic trends in consumer recall compliance. Thus, Figure 3 reveals that the DMC interrupted the negative accelerating curve in consumer recall compliance in the pre-DMC period and resulted in a positive acceleration in the cumulative number of vehicles fixed in the post-DMC period.
Graph: Figure 3. A visual representation of the DMC's effect on consumer recall compliance.
To estimate consumer recall compliance in the absence of the DMC, we set the post-DMC time-related variables Transition to DMC, Time Since DMC Start, and Time Since DMC Start Squared to zero in Equation 1. In the pre-DMC period, the dotted line with circular dots ("Number of vehicles fixed without the DMC") is overlapped by the solid line, as both depict the pre-DMC slope. The dashed line ("Predicted number of vehicles fixed without the DMC") depicts the predicted trend in recall compliance in the post-DMC period without the DMC. In line with the negative quadratic trend in compliance before the DMC, the model predicts that the cumulative number of vehicles fixed in Q1 of 2016 would be lower than in Q4 of 2015. However, because the cumulative number of vehicles fixed over time cannot decrease, using the model-based results to predict compliance without the DMC in the post-DMC period would overstate the DMC's positive effect and lead to erroneous conclusions. A reasonable prediction in our context is that compliance would level off from Q4 of 2015 onward without the DMC. Accordingly, the predicted trajectory of consumer recall compliance without the DMC in Figure 3 in the post-DMC period is depicted as a flat line.
To quantify the DMC's positive impact, we compute the difference in the predicted cumulative number of vehicles fixed with and without the DMC for a recall campaign after eight calendar quarters have elapsed using the coefficient estimates in Model 1 of Table 4 (represented by the solid line) and the dashed line in Figure 3. We find that in the first four quarters after it was introduced, the DMC increased the number of vehicles fixed, on average, by 20,712 per recall campaign above what would be expected without the DMC.
To test the moderating effects (H2–H6), we estimate the following specification:
Graph
( 2)
where i refers to a recall campaign, t refers to time (by calendar quarter-year), and ɛ refers to the random error. There are two points worth noting about this specification. First, Equation 2 is identical to the main-effect specification in Equation 1, aside from the addition of the moderators and interaction terms. Second, the interaction terms between Transition to DMC and the moderators test for boundary conditions of the DMC's effectiveness at the moment it is introduced in Q1 of 2016. The interaction terms between Time Since DMC Start and the moderators test for the DMC's boundary conditions following its introduction from Q2 of 2016 onward.[ 7]
The results for the moderating effects are reported in Model 1 of Table 5. None of the interaction terms in Model 1 have variance inflation factors (VIFs) above 5.17. The VIFs for Model 1 are reported in Table WA3 of the Web Appendix. While the lower-order terms of the time-related variables in Model 1 of Table 5 have VIFs above 10, our interest is only in interpreting the coefficients of the interaction terms. Furthermore, the interpretation of an interaction term is not affected by multicollinearity, as this multicollinearity affects neither the interaction term's coefficient nor its standard error ([42], p. 399). The same is true for the coefficients and standard errors of higher-order polynomials, which are essentially interactions of the lower-order terms ([17]).
Graph
Table 5. Results for the Effect of the DMC on Consumer Recall Compliance: Interaction Models.
| Model 1 | Model 2 | Model 3 | Model 4 |
|---|
| Dependent Variable: Consumer Recall Compliance | Interaction Model | Interaction Model Without Make Fixed Effects | Interaction Model with Only Linear Slope Terms | Interaction Model with Only Linear Slope Terms and Without Make Fixed Effects |
|---|
| Estimate (SE) | Hypothesis Tested | Estimate (SE) | Estimate (SE) | Estimate (SE) |
|---|
| Intercept | –6,809.36 | | –3,814.06 | –5,232.02 | 1,761.74 |
| (20,975.05) | | (6,009.95) | (20,958.88) | (5,769.30) |
| Time from Recall | 14,838.04*** | | 13,896.02*** | 8,929.75*** | 8,562.00*** |
| (3,697.30) | (3,603.27) | (1,545.46) | (1,467.90) |
| Time from Recall Squared | –2,165.48* | | –1,988.51* | | |
| (1,194.20) | (1,178.13) | | |
| Transition to DMC | –2,315.27 | | –1,157.12 | –3,212.90 | –2,432.85 |
| (6,250.76) | (6,123.64) | (5,763.04) | (5,668.35) |
| Time Since DMC Start | 1,243.44 | | 652.76 | 2,195.53 | 2,076.94 |
| (3,492.59) | (3,445.95) | (2,136.19) | (2,113.41) |
| Time Since DMC Start Squared | 201.56 | | 314.61 | | |
| (677.47) | (666.82) | | |
| Advertising Intensity | –25.31 | | 48.54 | –49.50 | 31.68 |
| (211.92) | (164.90) | (211.58) | (164.66) |
| Multiple Recall Notifications | 6,328.66 | | 5,699.31 | 7,362.64 | 6,778.29 |
| (15,073.09) | (13,263.56) | (15,078.72) | (13,250.10) |
| Recall Size | .51*** | | .51*** | .51*** | .51*** |
| (.01) | (.01) | (.01) | (.01) |
| Media Coverage | –2,105.46 | | –2,162.93 | –2,263.14 | –2,291.67 |
| (1,614.43) | (1,546.53) | (1,612.49) | (1,544.60) |
| Age of Recalled Producta | –5,434.18*** | | –5,373.96*** | –5,438.46*** | –5,400.57*** |
| (938.50) | (809.64) | (939.49) | (809.57) |
| Component Hazard | 3,789.08 | | 2,573.14 | 4,034.83 | 2,797.70 |
| (7,013.30) | (6,388.82) | (7,018.76) | (6,388.20) |
| Time for Repairb | –518.81 | | 274.31 | –571.21 | 262.58 |
| (1,700.56) | (1,242.39) | (1,702.20) | (1,242.49) |
| Concurrent Recall Campaigns | 4,380.81 | | 4,753.41 | 4,539.60 | 4,777.33 |
| (6,334.25) | (5,499.19) | (6,340.05) | (5,499.64) |
| Transition to DMC × Media Coverage | 3,993.22** | H2 supported (+) | 4,076.06** | 4,288.84** | 4,309.28** |
| (1,770.60) | (1,758.61) | (1,748.75) | (1,737.81) |
| Transition to DMC × Age of Recalled Producta | 2,171.51*** | H3 supported (+) | 2,111.26*** | 2,083.69*** | 2,018.78*** |
| (684.75) | (680.06) | (681.70) | (677.03) |
| Transition to DMC × Component Hazard | –1,538.73 | H4 not supported (+) | –925.12 | –1,336.34 | –807.79 |
| (5,546.05) | (5,506.80) | (5,536.56) | (5,497.76) |
| Transition to DMC × Time for Repairb | 538.61 | H5 partially supportedc (−) | 468.71 | 556.37 | 490.69 |
| (1,077.57) | (1,071.98) | (1,077.58) | (1,072.11) |
| Transition to DMC × Concurrent Recall Campaigns | 3,498.99 | H6 not supported (−) | 3,363.49 | 3,602.48 | 3,495.35 |
| (4,782.83) | (4,759.03) | (4,780.47) | (4,757.08) |
| Time Since DMC Start × Media Coverage | 1,499.02*** | H2 supported (+) | 1,455.20*** | 1,519.93*** | 1,497.27*** |
| (516.48) | (511.32) | (501.47) | (497.09) |
| Time Since DMC Start × Age of Recalled Producta | 884.17*** | H3 supported (+) | 878.30*** | 899.60*** | 900.10*** |
| (246.13) | (244.12) | (241.40) | (239.48) |
| Time Since DMC Start × Component Hazard | –3,080.78 | H4 not supported (+) | –3,350.67 | –3,062.90 | –3,293.08 |
| (2,334.58) | (2,308.70) | (2,324.63) | (2,298.18) |
| Time Since DMC Start × Time for Repairb | –830.94* | H5 partially supportedc (−) | –785.30* | –841.29* | –799.98* |
| (458.47) | (454.04) | (458.16) | (453.75) |
| Time Since DMC Start × Concurrent Recall Campaigns | 2,156.78 | H6 not supported (−) | 2,478.38 | 2,112.02 | 2,390.51 |
| (1,986.83) | (1,971.75) | (1,976.25) | (1,960.38) |
| Make Fixed Effects | Yes | | No | Yes | No |
| Quarter Fixed Effects | Yes | | Yes | Yes | Yes |
| N (sample size) | 1,637 | | 1,637 | 1,637 | 1,637 |
| R-Squared Within | .07 | | .08 | .07 | .08 |
| R-Squared Between | .87 | | .87 | .87 | .87 |
- 4 *p <.10.
- 5 **p <.05.
- 6 ***p <.01.
- 7 a In years.
- 8 b In hours.
- 9 c The interaction between Transition to DMC and Time for Repair is not significant. However, the interaction between Time Since DMC Start and Time for Repair is marginally significant (p =.07). Therefore, H5 is partially supported.
- 10 Notes: The full set of results, with VIFs for the variables, is reported in Table WA3 of the Web Appendix.
We find that the coefficient of the interaction between the Transition to DMC and Media Coverage variables is positive (3,993.22, p < .05). This finding suggests that at the moment the DMC was introduced in Q1 of 2016, its impact on consumer recall compliance was stronger for recall campaigns with greater media coverage. We also find that the coefficient of the interaction between Time Since DMC Start and Media Coverage is positive (1,499.02, p < .01), implying that Media Coverage continues to positively moderate the DMC–consumer recall compliance relationship from Q2 of 2016 onward. H2 is thus supported.
We use visual plots in Figure 4, Panel A, to provide an intuitive understanding of interactions with Media Coverage. We set "low" (0) and "high" ( 1) values for Media Coverage to visually represent its impact on the DMC's effectiveness for a recall campaign of average size that spans from Q1 of 2015 until Q4 of 2016. As Panel A shows, the DMC's impact is stronger when media coverage for a recall is high relative to low.
Graph: Figure 4. The moderating roles of media coverage, age of recalled product, and time for repair during the DMC.
We also find that the coefficient of the interaction between Transition to DMC and Age of Recalled Product is positive (2,171.51, p < .01). That is, the DMC's impact on consumer recall compliance is stronger for older (vs. newer) recalled vehicles at the moment it is introduced in Q1 of 2016. We also find that the coefficient for the interaction between Time Since DMC Start and Age of Recalled Product is positive (884.17, p < .01), which suggests that the DMC also has a stronger impact on consumer recall compliance for older vehicles from Q2 of 2016 onward. H3 is supported. We set "low" (25th percentile) and "high" (75th percentile) values of the Age of Recalled Product variable in Figure 4, Panel B, which demonstrates that the DMC's effectiveness is stronger for older vehicles (i.e., for larger values of the variable) both at the moment of the DMC's introduction and thereafter.
Inconsistent with H4, we find that neither the Transition to DMC × Component Hazard nor the Time Since DMC Start × Component Hazard interaction terms are significant (p > .78 and p > .18, respectively). Perhaps this insignificant finding implies that consumers are unable to actually discern the hazard of the defective components cited in recall notification letters.
We find that the interaction between Transition to DMC and Time for Repair is not significant (p > .61). However, the coefficient for the interaction between Time Since DMC Start and Time for Repair is negative (830.94, p = .07) and marginally significant. Thus, following the DMC's introduction from Q2 of 2016 until Q4 of 2017, the DMC is less effective at improving consumer recall compliance as the time needed to repair a defective component increases. H5 is partially supported. Using "low" (25th percentile) and "high" (75th percentile) values for Time for Repair in Figure 4, Panel C, we demonstrate visually that following the DMC's introduction, the DMC's effectiveness is weaker when a defective component requires more versus less time to repair.
Finally, we find that neither the Transition to DMC × Concurrent Recall Campaigns nor the Time Since DMC Start × Concurrent Recall Campaigns interaction variables are significant (p > .46 and p > .27, respectively). H6 is not supported. The coefficients for the Advertising Intensity and Multiple Recall Notifications variables are also not significant (p > .67), while the coefficient for the Recall Size variable is positive (.51, p < .01).
The moderator results are robust to several alternate model specifications. We estimate three specifications similar to Model 1 in Table 5, removing automobile make–specific fixed effects and/or Time from Recall Squared and Time Since DMC Start Squared. The results of the alternate specifications are reported in Models 2, 3, and 4 of Table 5. Across each model, the estimates of the interaction terms are largely similar to Model 1, and the conclusions regarding the impact of the moderators on the DMC's effectiveness remain the same.
This study addresses the important question of whether a regulator-initiated DMC can improve consumer recall compliance. Further, the study aims to understand the conditions under which a DMC is more or less effective. We were able to exploit a full-coverage national DMC launched in January of 2016 by the NHTSA, the regulator in the U.S. automobile industry, to answer these research questions. The study offers several implications for researchers and policy makers.
The question of whether awareness campaigns are successful in eliciting the required compliance-related behaviors from consumers has been investigated in various empirical settings (e.g., [32]; [66]). The fact that the awareness campaign in our study is introduced by a governmental agency and that it is its first major effort to improve recall compliance using digital media raises questions about its potential efficacy. We find that a regulator-initiated DMC is effective at improving consumer recall compliance, as reflected by the additional number of vehicles fixed above what was to be expected without it. This finding underscores the critical importance of regulators utilizing digital means to provide recall-related information to elicit greater compliance from consumers.
The moderator analyses reveal that the DMC's effectiveness varies across recall campaigns. Specifically, the DMC's impact on consumer recall compliance is stronger for recall campaigns with greater media coverage. This finding is similar in spirit to research that has documented the moderating impact of negative publicity on postcrisis advertising and consumers' brand share and category purchases ([16]). Although brands aim to avoid negative publicity of their recalled products because of its adverse effects on sales, media outlets play an important functional role in reinforcing a regulator's efforts to improve consumer recall compliance. In our context, this effect is likely driven by the varying specificity of information provided by the regulator and the media.
We also find that the DMC is more effective when the recalled products are older. Prior research has found that the utility of paid search campaigns is greater in contexts where consumers have low levels of familiarity with the firms' brands or products ([ 7]). Along similar lines, we find that the DMC is more effective when the recalled products are older. Consumers of older products may not be in contact with their products' OEMs as their warranties expire and thereby be less familiar with safety-related developments for their vehicles. A DMC can provide resources for these consumers who are more likely to seek out recall-related information digitally.
Finally, we find that the DMC is less effective at improving consumer recall compliance for recall campaigns containing products with defects that require more (vs. less) time to repair. Previous work examining tax compliance similarly finds that time-related costs are significant barriers to compliance ([10]; [64]). Although our study finds that the DMC is effective at providing the necessary information to improve compliance, the inconvenience that higher repair times pose to consumers is still a significant impediment to compliance. Future research should examine whether regulators can improve compliance for recalls that require long repair times by using persuasive appeals.
Our study's findings also offer actionable insights for policy makers. In recent years, multiple audit reports submitted to U.S. Congress have questioned the adequacy of the NHTSA's oversight processes in managing consumer recall compliance ([52]). The overarching concern is that the agency has failed to carefully review safety issues, hold automakers accountable, collect safety data, or adequately train its staff, resulting in significant safety concerns being overlooked. Some industry experts go a step further and lament that product recalls may increase the number of crashes on the road, as the extra driving needed to remedy the issues heightens the probability of accidents ([43]). Our analysis reveals that in the first four quarters after it was introduced, the DMC increased the number of vehicles fixed, on average, by 20,712 per recall campaign over what would be expected without the DMC.
The improvement in consumer compliance as a result of the DMC is economically meaningful because it could reduce the number of vehicle accidents and the economic costs associated with vehicle crashes. Research has shown that, on average, a 1% improvement in recall compliance in the automobile industry lowers the number of vehicle accidents in the next three years by.46% (Bae and Benítez-Silva [ 1]). The average economic costs (e.g., fatalities, nonfatal injuries, damaged vehicles) associated with a motor vehicle accident are about $14,000 ([ 9]). Therefore, by improving consumer recall compliance, the DMC reduces the number of automobile accidents on the road and lowers the total economic costs associated with accidents. The last few years have witnessed an average of 953 product recalls in the U.S. automobile industry, suggesting that the total annual economic impact of improved consumer recall compliance across all recall campaigns is not trivial.
Further, our study implies that a lack of available relevant information is a significant contributing factor to low consumer recall compliance. For years, the NHTSA has tried to mandate electronic recall notices to counteract the issue of low consumer recall awareness ([46]). The regulatory agency's foray into the digital domain with "Safe Cars Save Lives" has proven successful, suggesting that improving consumer awareness of recall-related information is a crucial step in improving consumer recall compliance. Even though the automobile industry receives considerable media attention (e.g., GM's faulty ignition issue, Takata's airbag failures), getting consumers to pay attention to recall notifications is challenging.
While extant work notes that manufacturers view greater media coverage of a recall as damaging to their brands' financial health ([ 5]), we find that media coverage helps improve safety outcomes by increasing the effectiveness of the DMC. Furthermore, we find that the DMC's impact on consumer recall compliance is stronger when recalled vehicles are older. Issues of low compliance are particularly prevalent among owners of these older products. According to J.D. [54], just 44% of vehicles manufactured between 2003 and 2007 had their defects remedied, drastically below recall completion percentages of 73% for vehicles of model years 2013–2017. Federal and industry leaders have cited improving compliance among owners of older vehicles as one of four key topic areas to address moving forward ([51]). The DMC's effectiveness on consumers owning older products further suggests that regulators' use of digital tools to facilitate consumer access to relevant information could improve compliance.
Finally, our findings caution regulators to be mindful of perceived time inconvenience as a serious impediment to consumer recall compliance. While the DMC is effective at improving compliance, its impact is lower for recall campaigns with defective components that require longer to repair. Interestingly, consumers often do not cite the time needed to complete the repair as the most important factor in deciding whether to remedy safety defects ([28]). Yet, our findings suggest that perceived time inconvenience is a serious obstacle to achieving consumer compliance. The findings should enable regulatory agencies to make more compelling cases for financial resources to be devoted to DMCs aiming to improve compliance.
The findings of the study are subject to some limitations. First, the study is limited to the U.S. automobile industry and a single DMC, implying that caution is warranted in generalizing our findings to other settings. If systematic recall completion data were available in other contexts, the conceptual and methodological framework employed in this study could be useful in deriving empirical generalizations about the effectiveness of DMCs in other settings. Second, data on the NHTSA's total advertising expenditure on this campaign were not available. As such, the DMC's positive effect should not be interpreted as consumers' responsiveness to advertising dollars. Third, our study was unable to examine if a recall notification message's content impacts compliance because the mandated recall notification letters sent from manufacturers to consumers in our setting were standard, lacking variation. If regulators use assertive versus nonassertive language or fear- versus health- versus norm-based persuasion appeals in their advertisements or messaging, understanding which types improve compliance would be valuable.
sj-pdf-1-jmx-10.1177_00222429211023016 - Supplemental material for Regulating Product Recall Compliance in the Digital Age: Evidence from the "Safe Cars Save Lives" Campaign
Supplemental material, sj-pdf-1-jmx-10.1177_00222429211023016 for Regulating Product Recall Compliance in the Digital Age: Evidence from the "Safe Cars Save Lives" Campaign by Sotires Pagiavlas, Kartik Kalaignanam, Manpreet Gill, and Paul D. Bliese in Journal of Marketing
Footnotes 1 Raj Venkatesan
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The authors acknowledge a research grant by the Darla Moore School of Business that made this research possible.
4 Kartik Kalaignanam https://orcid.org/0000-0003-0821-3687
5 Online supplement:https://doi.org/10.1177/00222429211023016
6 Both the HBM and protection motivation theory consider the impact of risk severity and vulnerability on consumers. Protection motivation theory has been used to study harm reduction behavior among individuals and the nature of fear appeals that bring about a behavioral change. However, the HBM is better suited for understanding consumer noncompliance issues, as it has been used to examine individuals' noncompliance to medical tests, treatment regimens, and preventative behaviors.
7 We do not interact the moderators with Time Since DMC Start Squared because we have no theory to support an interaction with a quadratic post-DMC time trend and find no significant interactions with Time Since DMC Start Squared. However, the inclusion of these nonsignificant terms masked several significant linear interactions.
References Bae Yong-Kyun , Benítez-Silva Hugo. (2011), " Do Vehicle Recalls Reduce the Number of Accidents? The Case of the U.S. Car Market ," Journal of Policy Analysis and Management , 30 (4), 821 – 62.
Bae Yong-Kyun , Benítez-Silva Hugo. (2013), "Information Transmission and Vehicle Recalls: The Role and Regulation of Recall Notification Letters," MPRA Paper No. 50380, Munich Personal RePEc Archive.
Balasubramanian Siva K. , Cole Catherine. (2002), " Consumers' Search and Use of Nutrition Information: The Challenge and Promise of the Nutrition Labeling and Education Act ," Journal of Marketing , 66 (3), 112 – 27.
Barr Stewart , Gilg Andrew W.. (2007), " A Conceptual Framework for Understanding and Analyzing Attitudes Towards Environmental Behaviour ," Geografiska Annaler: Series B, Human Geography , 89 (4), 361 – 79.
Beattie Graham , Durante Ruben , Knight Brian , Sen Ananya. (2021), " Advertising Spending and Media Bias: Evidence from News Coverage of Car Safety Recalls ," Management Science , 67 (2), 698 – 719.
Berck Peter , Moe-Lange Jacob , Stevens Andrew , Villas-Boas Sofia. (2016), " Measuring Consumer Responses to a Bottled Water Tax Policy ," American Journal of Agricultural Economics , 98 (4), 981 – 96.
Blake Thomas , Nosko Chris , Tadelis Steven. (2015), " Consumer Heterogeneity and Paid Search Effectiveness: A Large-Scale Field Experiment ," Econometrica , 83 (1), 155 – 74.
8 Bliese Paul D. , Lang Jonas W.B.. (2016), " Understanding Absolute and Relative Growth in Discontinuous Growth Models: Coding Alternatives and Implications for Hypothesis Testing ," Organizational Research Methods , 19 (4), 562 – 92.
9 Blincoe Lawrence J. , Seay Angela G. , Zaloshnja Eduard , Miller Ted R. , Romano Eduardo O. , Luchter Stephen , et al. (2002), " The Economic Impact of Motor Vehicle Crashes, 2000 ," in National Highway Traffic Safety Administration (NHTSA), DOT HS 809 446. Washington, DC : NHTSA.
Blumenthal Marsha , Slemrod Joel. (1992), " The Compliance Cost of the US Individual Income Tax System: A Second Look after Tax Reform ," National Tax Journal , 45 (2), 185 – 202.
Bowman Douglas , Heilman Carrie M. , Seetharaman P.B.. (2004), " Determinants of Product-Use Compliance Behavior ," Journal of Marketing Research , 41 (3), 324 – 38.
Burmester Alexa B. , Becker Jan U. , van Heerde Harald J. , Clement Michel. (2015), " The Impact of Pre-and Post-Launch Publicity and Advertising on New Product Sales ," International Journal of Research in Marketing , 32 (4), 408 – 17.
Chen Yubo , Ganesan Shankar , Liu Yong. (2009), " Does a Firm's Product-Recall Strategy Affect Its Financial Value? An Examination of Strategic Alternatives During Product-Harm Crises ," Journal of Marketing , 73 (6), 214 – 26.
Chen Yubo , Ghosh Mrinal , Liu Yong , Zhao Liang. (2019), " Media Coverage of Climate Change and Sustainable Product Consumption: Evidence from Hybrid Vehicle Market ," Journal of Marketing Research , 56 (6), 995 – 1101.
Clee Mona A. , Wicklund Robert A.. (1980), " Consumer Behavior and Psychological Reactance ," Journal of Consumer Research , 6 (4), 389 – 405.
Cleeren Kathleen , van Heerde Harald J. , Dekimpe Marnik G.. (2013), " Rising from the Ashes: How Brands and Categories Can Overcome Product-Harm Crises ," Journal of Marketing , 77 (2), 58 – 77.
Cohen Jacob , Cohen Patricia , West Stephen G. , Aiken Leona S.. (2003), Applied Multiple Regression/Correlation Analysis for the Behavioral Sciences , 3rd ed. New York : Routledge.
Colchero M. Arantxa , Rivera-Dommarco Juan , Popkin Barry M. , Ng Shu Wen. (2017), " In Mexico, Evidence of Sustained Consumer Response Two Years After Implementing a Sugar-Sweetened Beverage Tax ," Health Affairs , 36 (3), 564 – 71.
Consumers Union. (2018), "Hazardous IKEA Furniture Remains in Homes Two Years After Recall," press release (July 28) , https://consumersunion.org/wp-content/uploads/2018/06/Safety-advocates-statement-on-IKEA-dresser-recall-anniversary-6-28-2018.pdf.
CPSC (2013), "CPSC Accomplishments from 2009-2012," (April 16), https://www.cpsc.gov/About-CPSC/Agency-Reports/CPSC-Accomplishments-from-2009-2012.
Dawar Niraj. (1998), " Product-Harm Crises and the Signaling Ability of Brands ," International Studies of Management & Organization , 28 (3), 109 – 19.
Dinner Isaac M. , van Heerde Harald J. , Neslin Scott A.. (2014), " Driving Online and Offline Sales: The Cross-Channel Effects of Traditional, Online Display, and Paid Search Advertising ," Journal of Marketing Research , 51 (5), 527 – 45.
Edelman Murray. (1988), Constructing the Political Spectacle. Chicago : University of Chicago Press.
EDriving (2016), "Safe Cars Save Lives Campaign Launches in U.S.," (January 27) , https://www.edriving.com/three60/safe-cars-save-lives-campaign-launches-in-u-s.
Eilert Meike , Jayachandran Satish , Kalaignanam Kartik , Swartz Tracey A.. (2017), " Does It Pay to Recall Your Product Early? An Empirical Investigation in the Automobile Industry ," Journal of Marketing , 81 (3), 111 – 29.
Federal Register (2014), "Announcement of Consumer Product Safety Apps Challenge Under the America Competes Reauthorization Act of 2011," (January 22) , https://www.govinfo.gov/content/pkg/FR-2014-01-22/pdf/2014-01085.pdf.
GAO (2011), "Auto Safety: NHTSA Has Options to Improve the Safety Defect Recall Process," Report to Congressional Requestors (June) , https://www.gao.gov/assets/gao-11-603.pdf.
GAO (2017), "Auto Recalls: NHTSA Should Take Steps to Further Improve the Usability of Its Website," Report to Congressional Committees (December) , https://www.gao.gov/assets/690/688714.pdf.
Gao Haibing , Xie Jinhong , Qi Wang , Wilbur Kenneth C.. (2015), " Should Ad Spending Increase or Decrease Before a Recall Announcement? The Marketing-Finance Interface in Product-Harm Crisis Management ," Journal of Marketing , 79 (5), 80 – 99.
Garretson Judith A. , Burton Scot. (1998), " Alcoholic Beverage Sales Promotion: An Initial Investigation of the Role of Warning Messages and Brand Characters Among Consumers Over and Under the Legal Drinking Age ," Journal of Public Policy & Marketing , 17 (1), 35 – 47.
Gibson Dirk C. (1995), " Public Relations Considerations of Consumer Product Recall ," Public Relations Review , 21 (3), 225 – 40.
Grinstein Amir , Nisan Udi. (2009), " Demarketing, Minorities, and National Attachment ," Journal of Marketing , 73 (2), 105 – 22.
Haunschild Pamela R. , Rhee Mooweon. (2004), " The Role of Volition in Organizational Learning: The Case of Automotive Product Recalls ," Management Science , 50 (11), 1545 – 60.
Hoffer George E. , Pruitt Stephen W. , Reilly Robert J.. (1994), " When Recalls Matter: Factors Affecting Owner Response to Automotive Recalls ," Journal of Consumer Affairs , 28 (1), 96 – 106.
Hornik Robert , Jacobsohn Lela , Orwin Robert , Piesse Andrea , Kalton Graham. (2008), " Effects of the National Youth Anti-Drug Media Campaign on Youths ," American Journal of Public Health , 98 (12), 2229 – 36.
Janz Nancy K. , Becker Marshall H.. (1984), " The Health Belief Model: A Decade Later ," Health Education Quarterly , 11 (1), 1 – 47.
Kalaignanam Kartik , Kushwaha Tarun , Eilert Meike. (2013), " The Impact of Product Recalls on Future Product Reliability and Future Accidents: Evidence from the Automobile Industry ," Journal of Marketing , 77 (2), 41 – 57.
Kontopantelis Evangelos , Doran Tim , Springate David A. , Buchan Iain , Reeves David. (2015), " Regression Based Quasi-Experimental Approach When Randomisation Is not an Option: Interrupted Time Series Analysis ," BMJ , 350 , h2750.
Layton Lyndsey. (2010), "Officials Worry About Consumers Lost Among the Recalls," The Washington Post (July 1) , http://www.washingtonpost.com/wp-dyn/content/article/2010/07/01/AR2010070106504.html.
Leavitt Frank. (1979), " The Health Belief Model and Utilization of Ambulatory Care Services ," Social Science & Medicine. Part A: Medical Psychology & Medical Sociology , 13 , 105 – 12.
Lord Kenneth R. , Putrevu Sanjay. (1998), " Communicating in Print: A Comparison of Consumer Responses to Different Promotional Formats ," Journal of Current Issues & Research in Advertising , 20 (2), 1 – 18.
McClelland Gary H. , Irwin Julie R. , Disatnik David , Sivan Liron. (2017), " Multicollinearity Is a Red Herring in the Search for Moderator Variables: A Guide to Interpreting Moderated Multiple Regression Models and a Critique of Iacobucci, Schneider, Popovich, and Bakamitsos (2016), " Behavior Research Methods , 49 (1), 394 – 402.
McDonald Kevin M. (2009), " Do Auto Recalls Benefit the Public? " Regulation , 32 (2), 12 – 17.
McElhaney Alicia. (2014), "Only 10% of Recalled Kids Products Fixed or Returned," USA Today (February 18) , https://www.usatoday.com/story/money/personalfinance/2014/02/18/child-safety-recall-effectiveness-report/5425555/.
McEvoy David M. , Stranlund John K.. (2010), " Costly Enforcement of Voluntary Environmental Agreements ," Environmental and Resource Economics , 47 (1), 45 – 63.
Meier Fred. (2016), "NHTSA Aims to Mandate Electronic Recall Notices," Cars.com (September 1) , https://www.cars.com/articles/nhtsa-aims-to-mandate-electronic-recall-notices-1420690252650.
Melnyk Vladimir , Carrillat François A. , Melnyk Valentyna. (2021), " The Influence of Social Norms on Consumer Behavior: A Meta-Analysis ," Journal of Marketing (published online June 15), https://doi.org/10.1177/00222429211029199.
Mian Atif , Sufi Amir. (2012), " The Effects of Fiscal Stimulus: Evidence from the 2009 Cash for Clunkers Program ," Quarterly Journal of Economics , 127 (3), 1107 – 42.
Moorman Christine. (1996), " A Quasi Experiment to Assess the Consumer and Informational Determinants of Nutrition Information Processing Activities: The Case of the Nutrition Labeling and Education Act ," Journal of Public Policy & Marketing , 15 (1), 28 – 44.
Naik Prasad A. , Raman Kalyan. (2003), " Understanding the Impact of Synergy in Multimedia Communications ," Journal of Marketing Research , 40 (4), 375 – 88.
NHTSA (2018), "NHTSA Meets with Federal and Industry Leaders to Discuss Boosting Recall Repair Rates," press release (November 8) , https://www.nhtsa.gov/press-releases/nhtsa-meets-federal-and-industry-leaders-discuss-boosting-recall-repair-rates.
Office of Inspector General (2018), "NHTSA's Management of Light Passenger Vehicle Recalls Lacks Adequate Processes and Oversight," Report No. ST2018062 (July 18) , https://www.oig.dot.gov/sites/default/files/NHTSA%20Auto%20Recalls%20Final%20Report%5E07-18-18.pdf.
Pechmann Cornelia , Zhao Guangzhi , Goldberg Marvin E. , Reibling Ellen T.. (2003), " What to Convey in Antismoking Advertisements for Adolescents: The Use of Protection Motivation Theory to Identify Effective Message Themes ," Journal of Marketing , 67 (2), 1 – 18.
Power J.D.. (2016), "Unfixed Recalled Vehicles Pose Risk for Automakers, Dealers and Drivers, Says J.D. Power SafetyIQ," press release (July 25) , http://www.jdpower.com/business/press-releases/2016-safetyiq-july.
Ringold Debra J. (2002), " Boomerang Effects in Response to Public Health Interventions: Some Unintended Consequences in the Alcoholic Beverage Market ," Journal of Consumer Policy , 25 (1), 27 – 63.
Rosenstock Irwin M. (1974), " Historical Origins of the Health Belief Model ," Health Education Monographs , 2 (4), 328 – 35.
Rupp Nicholas G. (2001), " Are Government Initiated Recalls More Damaging for Shareholders? Evidence from Automotive Recalls, 1973–1998 ," Economics Letters , 71 (2), 265 – 70.
Rupp Nicholas G. , Taylor Curtis R.. (2002), " Who Initiates Recalls and Who Cares? Evidence from the Automobile Industry ," Journal of Industrial Economics , 50 (2), 123 – 49.
Shapiro Matthew D. , Slemrod Joel. (2009), " Did the 2008 Tax Rebates Stimulate Spending? " American Economic Review , 99 (2), 374 – 79.
Singer Judith D. , Willett John B.. (2003), Applied Longitudinal Data Analysis: Modeling Change and Event Occurrence. New York : Oxford University Press.
Solomon Mark G. , Ulmer Robert G. , Preusser David F.. (2002), " Evaluation of Click It or Ticket Model Programs ," in National Highway Traffic Safety Administration (NHTSA) , DOT HS 809 498. Washington, DC : NHTSA.
Stericycle Expert Solutions (2018), "Product Recalls: Big Brother or Caring for One Another?" press release (June 12) , https://www.prnewswire.com/news-releases/product-recalls-big-brother-or-caring-for-one-another-300664291.html.
Stewart David W. , Martin Ingrid M.. (1994), " Intended and Unintended Consequences of Warning Messages: A Review and Synthesis of Empirical Research ," Journal of Public Policy & Marketing , 13 (1), 1 – 19.
Tran-Nam Binh , Evans Chris , Walpole Michael , Ritchie Katherine. (2000), " Tax Compliance Costs: Research Methodology and Empirical Evidence from Australia ," National Tax Journal , 53 (2), 229 – 52.
Van Heerde Harald , Helsen Kristiaan , Dekimpe Marnik G.. (2007), " The Impact of a Product-Harm Crisis on Marketing Effectiveness ," Marketing Science , 26 (2), 230 – 45.
Wallington Sherrie F. , Oppong Bridget , Iddirisu Marquita , Adams-Campbell Lucile L.. (2018), " Developing a Mass Media Campaign to Promote Mammography Awareness in African American Women in the Nation's Capital ," Journal of Community Health , 43 (4), 633 – 38.
Wang Yanwen , Smith Michael , Singh Vishal. (2015), " The Unintended Consequences of Countermarketing Strategies: How Particular Antismoking Measures May Shift Consumers to More Dangerous Cigarettes ," Marketing Science , 35 (1), 55 – 72.
Wosińska Marta. (2005), " Direct-to-Consumer Advertising and Drug Therapy Compliance ," Journal of Marketing Research , 42 (3), 323 – 32.
Zaitsu Masayoshi , Yoo Byung-Kwang , Tomio Jun , Nakamura Fumiaki , Toyokawa Satoshi , Kobayashi Yasuki. (2018), " Impact of a Direct-to-Consumer Information Campaign on Prescription Patterns for Overactive Bladder ," BMC Health Services Research , 18 (1), 1 – 9.
~~~~~~~~
By Sotires Pagiavlas; Kartik Kalaignanam; Manpreet Gill and Paul D. Bliese
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 107- Sales and Self: The Noneconomic Value of Selling the Fruits of One's Labor. By: Schnurr, Benedikt; Fuchs, Christoph; Maira, Elisa; Puntoni, Stefano; Schreier, Martin; van Osselaer, Stijn M.J. Journal of Marketing. May2022, Vol. 86 Issue 3, p40-58. 19p. 1 Diagram, 2 Charts, 4 Graphs. DOI: 10.1177/00222429211064263.
- Database:
- Business Source Complete
Sales and Self: The Noneconomic Value of Selling the Fruits of One's Labor
A core assumption across many disciplines is that producers enter market exchange relationships for economic reasons. This research examines an overlooked factor; namely, the socioemotional benefits of selling the fruits of one's labor. Specifically, the authors find that individuals selling their products interpret sales as a signal from the market that serves as a source of self-validation, thus increasing their happiness above and beyond any monetary rewards from those sales. This effect highlights an information asymmetry that is opposite to what is found in traditional signaling theory. That is, the authors find that customers have information about product quality that they signal to the producer, validating the producer's skill level. Furthermore, the sales-as-signal effect is moderated by characteristics of the purchase transaction that determine the signal strength of sales: The effect is attenuated when product choice does not reflect a deliberate decision and is amplified when buyers incur higher monetary costs. In addition, sales have a stronger effect on happiness than alternative, nonmonetary forms of market signals such as likes. Finally, the sales-as-signal effect is more pronounced when individuals sell their self-made (vs. other-made) products and affects individuals' happiness beyond the happiness gained from producing.
Keywords: self-production; signaling; selling; happiness; self-validation
Digital platforms such as Etsy or Amazon Handmade have made it easy for individuals to sell their self-made products to other individuals. Commercial activities that were previously limited to economically marginal contexts such as flea markets have become big business. Although each individual producer's commercial activity might be small, the sheer number of such producers adds an important new source of competitive pressure for traditional firms in many industries, including clothing, food, and home furnishings. For example, in 2020, Etsy had around 4.4 million sellers and 82 million buyers, leading to a total transaction value of around $10 billion ([39]).
Existing research has investigated the psychological and behavioral consequences of engaging in self-production. In short, people like the fruits of their labor and value products they produce themselves more than comparable products made by others ([17]; [31]). The higher valuation of self-made products stems from the sense of accomplishment, pride, and competence that individuals experience when they successfully self-design or assemble a product ([ 9]; [29]). Prior research has focused on individuals engaging in self-production with the objective of consuming the product themselves or giving it as a gift ([30]), but it provides little insight into the increasingly common situation in which individuals make products with the objective of selling them to "the market"; that is, to unknown others.
A common assumption in marketing, economics, and entrepreneurship is that producers participate in market exchanges for economic reasons ([18]; [36]). Despite this disciplinary emphasis on economic motives, it seems possible that individuals produce and subsequently sell products also for noneconomic reasons. Research in organizational behavior ([19]), economics ([ 2]), and entrepreneurship ([37]) has drawn attention to the socioemotional motivations of producing, but little attention has been paid specifically to the noneconomic benefits of selling the fruits of one's labor. Drawing on survey data from the field as well as a series of experiments, we document a sales-as-signal effect: Selling their self-made products increases individual producers' happiness above and beyond any monetary implications from these sales. This effect occurs because individual producers interpret the number of products sold as a market signal that validates their skills and competencies as producers.
Our research makes several contributions. First, in demonstrating the socioemotional consequences of selling one's creations, we introduce a novel perspective on the value of sales. We propose that above and beyond their importance as a source of monetary income, sales can also have socioemotional value by providing a source of self-validation. Economic theory generally conceptualizes supply-side agents as profit-maximizers, such that "managers of a firm make those choices that maximize the sum of current and future profits" ([15], p. 769). Consequently, one would expect that the value individual producers derive from selling their products is a function of the money they make from these sales. However, we demonstrate that the value individual producers gain from sales cannot be solely defined in economic terms. Instead, individual producers also gain considerable happiness via feelings of self-validation from selling their products. We thus provide evidence for the importance of the noneconomic value of participating in market exchanges.
Second, our findings contribute to the literature examining individuals' valuation of their self-made products ([17]; [30]; [31]). Whereas existing research has shown that individual producers feel competent and proud from successfully designing or assembling a product, our work examines the psychological consequences of selling self-made products and not of merely producing those products. We demonstrate that having actual buyers purchase one's self-made products functions as an external confirmation that the individual producer is competent and capable of creating a high-quality, marketable product, which fuels one's happiness beyond the happiness derived from production.
Third, we offer a novel perspective on the role of signals in market exchanges by conceptualizing sales as a signal from the market. Research in economics, marketing, and management has typically conceptualized marketplace signals as actions taken by sellers to convey information about unobservable product qualities to buyers ([ 5]; [ 8]; [23]). Thus, signals are traditionally sent by sellers and interpreted by buyers. In contrast, we propose that sellers interpret the act of purchasing a product as a signal from the buyer that validates the seller's skills and competencies as producer. Furthermore, whereas signals are usually conceptualized as intentional actions meant to benefit the sender ([ 8]), we propose that sending this type of signal to the seller is often incidental to buyers' core motivation to buy, which lies in their consumption goals, and that the signal does not directly benefit the sender but rather the receiver of the signal. Moreover, traditional signaling research assumes that sellers know the quality of their product and that buyers are uncertain. Our work indicates that sellers are, to some extent, uncertain about their own product and thus about their competencies as a producer. Our work indicates that buyers have information about product quality that reduces sellers' uncertainty. Therefore, we propose that individual producers interpret sales as a signal from their buyers that serves as a source of self-validation.
Fourth, our findings broaden the discussion on the societal role of market exchanges, a topic of intense interest among marketing scholars and practitioners ([ 7]). We add to this discussion on how marketing can help create a better world by drawing attention to the way selling might provide a positive source of meaning and happiness for individuals. Just like the social costs of marketing have often been underestimated, we argue that some important benefits have been neglected as well. Specifically, successfully marketing their products is a source of self-validation and happiness for producers.
What benefits do individuals derive from selling their self-made products? Why do they continue to populate online marketplace platforms? The principal and most obvious benefit that people derive from selling their products is money. But can monetary incentives alone explain the increasing popularity of online marketplaces? We propose that learning a customer bought their products increases individual producers' happiness above and beyond the monetary implications of these sales.
Specifically, we argue that sales validate one's skills and competencies as a producer. We found preliminary support for this notion in an exploratory qualitative study conducted as part of a master's thesis supervised by one of the authors ([42]). In ten in-depth interviews with Etsy sellers and nonobtrusive observations of Etsy's online discussion forums (see Web Appendix A [WA-A]), several informants highlighted that selling their products makes them happy; for example, "creating something that I like and others like enough to spend their hard-earned money on, is bliss" (Informant #7 from Etsy forum), and "it's so flattering when people choose to buy your creations" (#61, forum). The narratives suggest that the happiness derived from selling is not necessarily rooted in economic reasons but in one's perceived self-worth as a producer; for example, "Etsy allows me to rediscover my worth" (#31, forum). Describing a producer friend, one of our informants (#4, interview) stated, "for Jeani, I think it is the fact that she is making something someone thinks is worthy of buying. You know, paying some money for and it's like an accolade of her creative talent." Interestingly, the narratives suggest that the increased self-worth derived from selling motivates people to continue producing their own products; for example, "a sale...usually motivates me to make more, since it makes me feel as though my items are appreciated." In summary, our preliminary qualitative insights point to the possibility that sales make individual producers happy not only because of the monetary gains but also because sales more fundamentally validate their skills and competencies as a producer.
We argue that sales function as a signal from the marketplace that boosts individual sellers' self-validation. Signaling theory was developed in information economics to study market interactions under conditions of information asymmetry between sellers and buyers ([38]). It generally assumes that sellers are aware of the quality of their goods but buyers are not. To distinguish low-quality sellers from high-quality sellers, buyers must detect and interpret the signals sent by sellers. Prices, advertising, brands, and different types of firm actions can constitute signals ([ 8]; [22]; [23]). Moreover, signaling theory and its applications in marketing usually presume that the signal originates from a seller and is received by a buyer.[ 7] In our research, we propose that sales constitute a signal that is sent by buyers (with or without buyers' intention to actually signal something) and that validates the seller's competencies as a producer. Thus, in this context, there is an information asymmetry in the opposite direction from traditional signaling theory: It is the buyer who has information that reduces the seller's uncertainty (about the seller's skills and competencies about the producer).
Feelings of competence, which often result from others' validation of one's skills, are a central motivation for people to engage in creative tasks ([ 9]) and greatly affect how satisfied individuals are with their work ([33]). Crucially, feeling competent is a fundamental psychological need among humans, and its fulfillment strongly determines individuals' intrinsic motivation, life satisfaction, and mental health ([34]; [41]). [11], p. 231) even argue that the need to feel competent "must be satisfied for long-term psychological health." We thus propose that being validated as a competent producer through sales increases individual producers' happiness.
In summary, we predict a sales-as-signal effect: Sales increase individual producers' happiness, even when controlling for the effect of monetary gains. This is because producers interpret sales as a positive signal from the market that validates them as competent producers. Formally:
- H1: Sales increase individual producers' happiness above and beyond the monetary rewards from these sales (i.e., the sales-as-signal effect).
- H2: The sales-as-signal effect is mediated by feelings of self-validation as a producer.
The strength of a signal is determined by the extent to which receivers interpret the signal as credible ([ 5]). Signals are perceived as more credible the more they are able to provide information about products' unobservable quality ([ 6]; [ 8]) and the higher the costs and associated risk in sending a signal ([ 3]; [23]). Accordingly, we predict that the strength of the sales signal will depend on at least two characteristics of the purchase transaction: ( 1) the extent to which the product choice reflects a deliberate decision and ( 2) the monetary cost involved in purchasing the product. We decided to focus on these two moderators because they provide an internally valid test of our proposed underlying process (self-validation) and provide actionable implications for the management of online marketplaces.
Sales should more credibly inform individual producers about their competencies the more the sales are a direct consequence of the quality of individual producers' products ([ 6]; [ 8]). Thus, the sales-as-signal effect should be stronger the more the buyer is seen as making a deliberate decision to acquire the product. That is, sales should be self-validating when the buyer intentionally chooses a producer's product, but much less so when the product was selected in a way that does not reflect the buyer's preference for a specific product. Examples of the latter include a chef's choice item on a menu, a surprise wine box subscription, the specific vegetables offered by a community-supported agriculture farm co-op, a "mystery car" car rental option, a sneak preview at a movie theatre, or a sweepstakes in which it is not clear in advance which participant will receive which prize. We hypothesize that, above and beyond the monetary rewards from sales, individual producers will feel greater self-validation, and thus happiness, the more the purchase appears to be the result of a buyer's deliberate decision.
- H3: The sales-as-signal effect is stronger when the product purchase reflects a more (vs. less) deliberate decision.
If signal credibility is a function of signal cost ([ 3]; [23]), varying the cost of buying a product should alter the strength of the sales-as-signal effect. The higher the costs involved in purchasing a product, the more credibly the sales signal should inform individual producers about their competencies as a producer—even if the higher costs do not translate into higher monetary rewards for the individual producer (such as when the buyer bears higher shipping costs). We hypothesize that individual producers will feel higher levels of self-validation, and thus happiness, when the buyer incurs higher monetary costs, even when the higher monetary costs do not lead to higher monetary income for the individual producer.
- H4: The sales-as-signal effect is stronger when buyers incur higher (vs. lower) monetary costs.
Sales are not the only signals consumers might send. In addition to buying products, consumers may also signal quality through noneconomic signals, such as writing a review or liking a product or company on social media. One might argue that such noneconomic signals might have a stronger self-validating effect than sales because the noneconomic signals are sent intentionally (vs. being a usually unintentional by-product of the decision to buy). However, we propose that sales would be a more credible signal because they may be seen as more informative about the product's unobservable quality for several reasons.
First, other forms of signals such as online reviews have been criticized for being unable to reveal a product's actual quality ([12]), and public displays of support for a cause on social media (e.g., Facebook likes) are often unreliable indicators of one's willingness to support the cause when doing so is costly ([26]). Thus, noneconomic signals such as likes may be seen as "cheap talk." In contrast, customers should decide to purchase a product only if they really deem the product to be of high, or at least sufficiently high, quality ([14]). Second, selling products may evoke the specific norms that are associated with an exchange domain rather than a relational domain. In an exchange domain, the normative signal of value may be sales rather than more relational signals such as likes.[ 8] In this domain, having buyers purchase one's product might be the "ultimate" form of appreciation of an individual's competencies as a producer. Thus, sales may more credibly inform individual producers about their skills and competencies compared with other common forms of market signals such as likes, even when the cost to the customer of sales and likes are kept the same. We hypothesize that individual producers will feel greater self-validation, and thus happiness, from sales than from noneconomic signals, even when sales do not lead to higher monetary rewards to the seller than noneconomic signals (and even when monetary cost to the customer is kept constant).
- H5: Sales increase individual producers' happiness more than noneconomic signals above and beyond the monetary rewards from those sales to the producer (and above and beyond the monetary cost of sending those signals).
We further propose that individuals derive greater happiness from sales when selling self-made products as opposed to selling products made by someone else. Our hypothesis is that the effect of sales of self-produced goods on happiness is, to an important extent, driven by validation of the producer's skills and competencies as a producer. Of course, selling self-made products might also validate individual producers' skills as a seller; that is, successful sales may be interpreted as a signal that the seller of self-produced products is a competent marketer rather than a competent producer. However, unlike selling one's self-made products, selling products made by others cannot validate one's skill as a producer. Thus, the effect of sales on happiness should be larger for self-produced than for other-produced products.
- H6: The sales-as-signal effect is stronger among individuals selling self- (vs. other-) made products.
We tested our propositions in eight studies (N = 4,970). Study 1 and a supplementary study provide an initial exploration of our main hypothesis that sales increase individual producers' happiness above and beyond the monetary rewards from these sales (H1). In Study 1, we surveyed actual producers selling their self-made products (e.g., on Etsy). We find a positive relationship between sales and individual producers' happiness, controlling for the monetary implications of sales. In a supplementary study, we replicated this finding experimentally using a recall task among a sample of actual producers.
In Studies 2–6, we provide further evidence for the core sales-as-signal effect, explore the underlying mechanism, examine boundary conditions, and investigate the incremental effects of sales among different samples of producers. Studies 2 and 3 show that the sales-as-signal effect is driven by feelings of self-validation (H2). These studies also show that the strength of the sales-as-signal effect depends on ( 1) the extent to which the product choice reflects a deliberate decision (H3) and ( 2) the monetary cost involved for the consumer in purchasing the product (H4). Study 4 and a supplementary study demonstrate that sales increase individual producers' happiness more than receiving likes (H5). Study 5 shows that the effect of sales on happiness is stronger when individuals sell self-made products than when they sell products made by others (H6). Finally, Study 6 shows that the sales-as-signal effect is different from the mere happiness individuals gain from producing. Study 6 also shows that trying to sell one's self-produced products can backfire when low sales become a signal of low competency.
Study 1 provides an initial exploration of our core prediction that selling their products increases producers' happiness over and above any monetary rewards from those sales. We do so through a field survey of actual producers selling their self-made products online.
To engage this special and difficult-to-recruit population (individual producers), we worked with the administrators of eight Facebook groups of producers of handmade goods to promote our survey. As an incentive to participate, each participant received a $5 Amazon voucher (see WA-B1 for detailed study materials). After agreeing to a data protection disclaimer, we told participants that the aim of the study was to gather knowledge about their life as producers of handmade goods. To increase the comparability of the responses, we asked participants to think about the last four weeks when answering the questions.
The survey first captured our dependent variable, participants' satisfaction and happiness with their life as a producer of handmade products, which we measured with two items (r = .61; see survey scales in Table 1). We measured our key independent variable, one's current sales (i.e., the number of products sold), using two variables. First, we assessed the sales volume by asking participants to indicate the total number of items sold in the past four weeks. Second, we assessed sales growth by capturing how many items a given producer sold in the last four weeks compared to the average number of items sold per month in the last six months. The comparative nature of the measure is important because it is the within-person variance that most strongly predicts one's happiness at any given point in time ([35]).
Graph
Table 1. Survey Scales (Study 1).
| Variable | Measures |
|---|
| Happiness | "If you think about the past four weeks, how satisfied are you with your life as a producer of handmade products?" (1 = "extremely dissatisfied," and 7 = "extremely satisfied") and "if you think about the past four weeks, how happy are you with your life as a producer of handmade products?" (1 = "extremely unhappy," and 7 = "extremely happy") |
| Sales volume | "Over the past four weeks combined, how many handmade items did you sell?" |
| Sales growth | "Now, please compare the number of items that you sold in the past four weeks with the average number of items you sold per month in the past six months. Would you say that you sold more or fewer items in the past four weeks compared to the months before?" (1 = "much fewer," and 7 = "much more") |
| Revenue | "What was the total revenue on these items in the past four weeks? That is, how much money did you make by selling these items in the past four weeks?" |
| Profit | "How much profit did you make in the past four weeks by selling these items? That is, after subtracting all costs, how much money were you able to keep in your pockets?" |
| Future profit expectations | "If you think about the near future, how do you think your profits from selling your handmade products will develop?" (1 = "decrease a lot," and 7 = "increase a lot") |
| Socioeconomic status | "I don't think I'll have to worry about money too much in the future," "I don't need to worry too much about paying my bills," and "I have enough money to buy things I want" (1 = "totally disagree," and 7 = "totally agree") |
We included the following control variables to empirically isolate the sales-as-signal effect from a series of alternative explanations. We captured the direct monetary implications of sales by asking for the respective revenue and profit (in USD) in the said time period.[ 9] Although these measures are important for assessing our hypothesis (i.e., to control for any monetary implications of selling), they do not account for any future profit expectations. For example, one could argue that a positive sales trend in the current period might be (perceived to be) diagnostic of future sales and thus profit developments. Thus, a given happiness level at a given point in time might be due to future profit expectations based on the comparative number of items sold. To account for this alternative explanation, we asked participants how they expected their future profits to develop.
Graph
Table 2. Ordinary Least Squares Regressions on Happiness (Study 1).
| (1a) | (1b) | (2a) | (2b) | (3a) | (3b) | (4a) | (4b) |
|---|
| Ln(sales volume) | .06**(.03) | .05**(.03) | .07***(.02) | .06***(.02) | | | | |
| Sales growth | | | | | .34***(.02) | .34***(.02) | .19***(.03) | .19***(.03) |
| Ln(revenue) | −.01(.02) | | .002(.02) | | .01(.01) | | .01(.02) | |
| Ln(profit) | | −.002(.02) | | .01(.02) | | .01(.01) | | .02(.02) |
| Future profit expectations | | | .23***(.03) | .23***(.03) | | | .15***(.03) | .15***(.03) |
| Ln(experience) | | | .10**(.05) | .10**(.05) | | | .13***(.05) | .13***(.05) |
| Main job | | | .07(.09) | .06(.09) | | | .10(.09) | .10(.09) |
| Proportionate household income | | | −.004**(.002) | −.004**(.002) | | | −.003*(.002) | −.004**(.002) |
| Socioeconomic status | | | .28***(.03) | .28***(.03) | | | .21***(.03) | .21***(.03) |
| Age | | | .01(.004) | .005(.004) | | | .01(.004) | .01(.004) |
| Gender | | | −.07(.08) | −.08(.08) | | | −.02(.08) | −.02(.08) |
| Education controls | No | No | Yes | Yes | No | No | Yes | Yes |
| Product category controls | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| Observations | 828 | 828 | 828 | 828 | 828 | 828 | 828 | 828 |
| R2 | .07 | .07 | .32 | .32 | .25 | .25 | .35 | .35 |
1 Notes: Unstandardized regression coefficients. Standard errors in parentheses. Education controls: bachelor's degree is baseline. Product controls: jewelry is baseline. Main job: 0 = side job, 1 = main job. Gender: 0 = female, 1 = male.
- 2 *p <.10.
- 3 **p <.05.
- 4 ***p <.01.
In addition, we captured a series of control variables with regard to the business type and the producer. We captured the product domain(s) by asking what type(s) of products they sell (see WA-B2). We further asked participants about their experience as a producer (how long they have been selling their products [in months]), whether selling these products is their main or side job (0 = "side job," and 1 = "main job"), and how much the income from selling these products contributes to their total household income (in %). To assess each individual producer's socioeconomic status, we used a three-item scale (α = .82; [20]). Finally, participants indicated their gender (0 = "female," and 1 = "male"), age (in years), country of residence (0 = "United States," and 1 = "other"), relationship status (0 = "single," 1 = "committed relationship," 2 = "married," and 3 = "widowed"), and highest degree of education (see WA-B3). No further measures were taken.
The sample consisted of 828 individual producers (Mage = 35.22 years, SD = 8.43, 61.0% female; 90.6% U.S. residents). We successfully recruited a diverse sample of producers. Participants differed widely in their experience selling their products (M = 56.89 months, SD = 40.59, range: 0 to 360.00), revenue and profit (revenue: Mdn = US$4,100.00, M = US$44,904.99, SD = US$81,115.39, range: US$0 to US$520,000.00; profit: Mdn = US$2,000.00, M = US$14,545.44, SD = US$27,234.03, range: US$0 to US$350,000.00), percentage of household income from selling (M = 54.94%, SD = 34.51%, range: 0% to 100%), and types of products produced (see WA-B4). Because several of our open-ended measures were highly skewed (sales volume, revenue, profit, and experience), we log-transformed those variables (for descriptive statistics and interconstruct correlations, see WA-B5).
We tested the sales-as-signal effect by estimating a series of ordinary least squares (OLS) regression models accounting for different sets of control variables. Columns 1a and 1b in Table 2 report the results of regression models in which one's happiness as a producer, our dependent variable, is regressed on sales volume while controlling for product type(s) and for the economic implications of sales (in terms of either revenue or profit). We find that sales volume is indeed positively and significantly related to one's happiness, controlling for economic implications (Column 1a: b = .06, SE = .03, p = .02; Column 1b: b = .05, SE = .03, p = .04). This effect persists, and even gets a bit stronger, after adding all further control variables (see Column 2a: b = .07, SE = .02, p <.01; Column 2b: b = .06, SE = .02, p <.01).
We ran the same models using sales growth as the independent variable, with consistent results. Columns 3a and 3b in Table 2 show that sales growth is positively and significantly related to producers' happiness when controlling for product domain(s) and producers' revenue (b = .34, SE = .02, p <.01) or profit (b = .34, SE = .02, p <.01). The effect remains positive and significant when adding all further control variables (see Column 4a: b = .19, SE = .03, p <.01; Column 4b: b = .19, SE = .03, p <.01).
Using a survey of actual producers, we find real-world evidence for the sales-as-signal effect (H1): individual producers draw happiness from selling their self-made products above and beyond the money they make from these sales.
In a supplementary study (see Study 1S in WA-C), we examined whether the observed relationship between sales and happiness, controlling for monetary considerations, holds in a more controlled setting with random assignment. We tested this using actual producers from the Australian marketplace madeit.com.au (n = 169) by making them recall a period of time in which they sold more (vs. fewer) products than normal and asking them about their happiness as a producer at that point in time. The results of this experiment replicate the main finding of Study 1. Consistent with our theorizing, we find that individual producers are happier at periods of time in which they sell more (vs. fewer) products even when controlling for the profit they made from these sales. In addition, the effect is robust to another potential happiness driver: future profit expectations.
Study 2 aims to provide a behavioral test of causality for the effect of sales on happiness through feelings of self-validation (H2). We did this by conducting a two-stage behavioral experiment in which we first asked participants to actually produce their own products and then manipulated at a later stage whether their products were sold.
Another aim of Study 2 was to investigate how the extent of deliberateness of product choice moderates the sales-as-signal effect. We theorize that sales credibly inform individuals about their skills and competencies as producers to the extent that they provide information about the quality of the producers' products (H3). When sales do not reflect a deliberate decision, for example when a product is chosen at random, they are less informative about the buyer's product quality perceptions and thus the sales-as-signal effect should be reduced.
We invited 1,700 American workers from Amazon Mechanical Turk (MTurk) in two waves[10] to participate in a two-stage study. We recruited participants interested and skilled in drawing. The experiment employed a 2 (sales: product sold vs. product not sold) × 2 (choice: deliberate vs. random) between-participants design. We informed all participants that, to enrich life in times of crisis, a researcher was delegated by the university's program director to organize an exhibition of comics that symbolize the defeat of the coronavirus (SARS-CoV-2). We told participants that their task would be to draw a picture of a superhero conquering the coronavirus, that all drawings would be shown to potential customers (i.e., faculty and administrative staff at the university) who could purchase one drawing for $1.50, and that all purchased drawings would be exhibited. To ensure that participants only submitted their original work, we told participants to sign their hand-drawn picture with their MTurk ID. Finally, we told participants that they would receive a second survey about one week later that would inform them about whether their drawing was sold. Importantly, to keep the monetary rewards constant, we truthfully told participants that all participants would receive $1.50 as a compensation for their work—irrespective of whether their drawing was sold. We received 417 valid drawings (see Figure 1 for a selection of drawings; invalid submissions included blank submissions, pictures that were unrelated to the task, and unsigned submissions).
Graph: Figure 1. Examples of drawings of superheroes defeating the coronavirus (Study 2).
Approximately one week later, we invited workers who had submitted valid drawings to participate in the second part of this study, which included our manipulations. A total of 347 workers accepted this invitation (Mage = 33.39 years, SD = 11.15, 53.0% female). We first reminded all participants that the researcher showed all drawings to potential customers (i.e., faculty and administrative staff at the university), who could purchase one drawing for $1.50. In addition, participants were reminded that all artists would receive $1.50 as compensation for their work, irrespective of whether their drawing was sold. Then, participants either read that each customer who decided to purchase a drawing selected the one they wanted (deliberate choice) or that each customer who decided to purchase a drawing received one selected at random (random choice). In addition, participants read that their specific drawing either was or was not sold (see WA-D for study materials). Participants then indicated their level of happiness as a producer on two seven-point items ("extremely unhappy/extremely happy," "extremely dissatisfied/extremely satisfied"; r = .87) and their feelings of self-validation on three seven-point items ("not at all skilled/extremely skilled," "not at all competent/extremely competent," and "not at all talented/extremely talented"; α = .95).[11] Finally, participants completed two attention checks ("was your drawing sold?"; "yes, my drawing was sold/no, my drawing was not sold"; "how did the purchase of drawings happen?"; based on customers' deliberate/random choice [the 'random selector' picked the drawing]/I cannot remember)[12] and indicated their age and gender.
A 2 × 2 ANOVA on happiness reveals a significant main effect of sales (Msold = 6.03, SD = 1.47 vs. Mnot_sold = 3.67, SD = 1.27; F( 1, 343) = 263.27, p <.001) and a nonsignificant main effect of product choice (Mdeliberate = 4.81, SD = 1.91 vs. Mrandom = 4.93, SD = 1.70; F( 1, 343) = .65, p = .42). Importantly, we also obtained the expected significant interaction effect (F( 1, 343) = 11.20, p = .001; see Figure 2, Panel A). Planned contrasts reveal that when the product choice was deliberate, the effect of sales on happiness was significantly stronger (Msold = 6.21, SD = 1.17 vs. Mnot_sold = 3.37, SD = 1.38; F( 1, 343) = 194.37, p <.001) than when the product choice was random (Msold = 5.84, SD = 1.34 vs. Mnot_sold = 3.98, SD = 1.50; F( 1, 343) = 81.75, p <.001).
Graph: Figure 2. Happiness and self-validation as a function of product choice and sales (Study 3).
In addition, we find that participants whose products were sold were happier when the focal decision was deliberate versus random (Mdeliberate = 6.21, SD = 1.17 vs. Mrandom = 5.84, SD = 1.34; F( 1, 343) = 8.50, p = .004). Interestingly, for participants who learned that their products were not sold, happiness was marginally higher when the focal decision was random rather than deliberate (Mdeliberate = 3.37, SD = 1.38 vs. Mrandom = 3.98, SD = 1.50; F( 1, 343) = 3.28, p = .07).
A 2 × 2 analysis of variance (ANOVA) on self-validation reveals a significant main effect of sales (Msold = 4.97, SD = 1.32 vs. Mnot_sold = 3.44, SD = 1.55; F( 1, 343) = 99.45, p <.001) and a nonsignificant main effect of product choice (Mdeliberate = 4.17, SD = 1.71 vs. Mrandom = 4.26, SD = 1.53; F( 1, 343) = .34, p = .56). Importantly, we also obtained a significant interaction effect (F( 1, 343) = 12.23, p = .001; see Figure 2, Panel B). Planned contrasts reveal that when the product choice was deliberate, the effect of sales on self-validation was significantly stronger (Msold = 5.19, SD = 1.29 vs. Mnot_sold = 3.14, SD = 1.46; F( 1, 343) = 92.04, p <.001) than when the product choice was random (Msold = 4.74, SD = 1.31 vs. Mnot_sold = 3.76, SD = 1.60; F( 1, 343) = 20.67, p <.001).
In addition, we find evidence that participants whose products were sold reported higher feelings of self-validation when the product choice was deliberate (Mdeliberate = 5.19, SD = 1.29 vs. Mrandom = 4.74, SD = 1.31; F( 1, 343) = 8.18, p = .004), whereas participants whose products were not sold reported higher feelings of self-validation when the product choice was random (Mdeliberate = 3.14, SD = 1.46 vs. Mrandom = 4.74, SD = 1.31; F( 1, 343) = 4.32, p = .04). Although not the focus of our theorizing, the latter difference further validates our signaling framework, as it indicates that a negative sales signal is more detrimental to feelings of self-validation when it is more easily interpreted as a reflection of one's competencies.
A moderated mediation analysis ([21], Model 7, n = 5,000) with sales (0 = not sold, 1 = sold) as the independent variable, product choice (0 = deliberate, 1 = random) as the moderator, self-validation as the mediator, and happiness as the dependent variable produces a significant index of moderated mediation (b = −.48, SE = .16, CI95% = [−.81, −.19]). Supporting our prediction, the effect of sales on happiness through feelings of self-validation was significantly stronger when the product choice was deliberate (b = .93, SE = .15, CI95% = [.66, 1.23]) versus random (b = .45, SE = .11, CI95% = [.23,.67]).
Using a multiwave experimental paradigm involving actual production, Study 2 provides causal evidence in support of our primary prediction that, above and beyond the monetary reward, sales increase producers' happiness via elevated feelings of self-validation (H2). In addition, the results are consistent with our theorizing that sales have a stronger impact on individual producers' self-validation when product choice is more deliberate (H3).
Although significantly smaller (as hypothesized), we also find a residual effect of sales on happiness and, to a lesser extent, self-validation when products were sold but selected at random. This finding is beyond the scope of our hypotheses, so we can only speculate about why even a random sale might make producers happy. One possibility is a process identified by Marx (1844/1993; see also [43]) in his Comments on James Mill. Marx's discussion of what it is like to produce as a human being (rather than being a cog in a machine), suggests that producing something that is used and enjoyed by another person provides important enjoyment of life and, to some extent, affirms the producer's unique competency as a person. Thus, the mere fact that another person has acquired one's product, even if the exact product choice was made at random, may provide some basic sense of self-validation and happiness in turn.
Study 3 investigates another moderator of the sales-as-signal effect: the monetary cost involved in purchasing the product. We theorize that sales credibly inform individuals about their skills and competencies as a producer to the extent that the acquisition of the product is costly to the customer (H4). In this study, we manipulated the monetary cost of sales by varying the shipping cost that a buyer needed to pay to acquire the product. We predicted that, although the financial gain from the sale is constant (the buyer bears the shipping costs), individual producers would be happier when the buyer accepts paying higher (vs. lower) shipping costs. As in the previous study, we again tested whether the increase in happiness can be explained by feelings of self-validation (H2).
With the help of hobbii.com, an online shop that sells knitting kits and supplies, we recruited 1,230 recreational knitters (Mage = 53.38 years, SD = 12.26, 99.4% female). The company promoted a link to our study in their weekly newsletter, which was received by German-speaking customers (and which yielded a response of N = 818) as well as customers from the United States (N = 412). As an incentive to participate, we raffled ten gift cards to the company's online shop worth $30 each.
The experiment employed a between-participants design with three conditions (sales: baseline vs. higher cost vs. lower cost). We asked all participants to imagine marketing their self-made knitted accessories on an online platform. Specifically, as we ran this study in June and participants came from Europe and the United States, we asked participants to assume they currently produce and sell summer beanies (i.e., beanies that are made from thin, lightweight material). Next, participants read that they received an email from a customer in New Zealand asking about winter beanies. Participants then read that they were able to offer their self-made winter beanies for $30.00. Finally, participants either read that the customer from New Zealand did not respond to this offer (baseline), decided to purchase the beanie for $30.00 plus $20.90 shipping costs (higher cost), or decided to purchase the beanie for $30.00 plus $2.90 shipping costs (lower cost; for study materials, see WA-E1).
Participants then indicated their level of happiness (r = .87) and feelings of self-validation (α = .93) on the same scales as in Study 2. To account for alternative mechanisms, participants indicated how much profit they made with the customer from New Zealand (1 = "none," and 7 = "a lot") and how they thought their profit from selling products would develop in the near future (1 = "decrease a lot," and 7 = "increase a lot"). Participants further completed the following attention check: "Did the customer from New Zealand buy your beanie?" (yes/no). Participants in the lower cost and higher cost conditions additionally completed the following attention check: "How much did it cost to ship the beanie to New Zealand?" ($2.90/$20.90).[13] Finally, participants indicated their gender and age.[14]
A one-way ANOVA with happiness as the dependent variable produces a significant effect (F( 2, 1,227) = 406.52, p <.001). Follow-up contrasts reveal that, compared to the baseline condition in which the customer from New Zealand did not purchase the product (M = 3.13, SD = 1.10), participants were happier when the customer from New Zealand decided to purchase the product (higher shipping cost: M = 5.90, SD = 1.38; t( 1,227) = 27.93, p <.001; lower shipping cost: M = 5.03, SD = 1.72; t( 1,227) = 19.03, p <.001). More importantly, participants reported significantly higher levels of happiness when the buyer paid higher versus lower shipping costs (t( 1,227) = 8.79, p <.001).
A one-way ANOVA with feelings of self-validation as the dependent variable produces a significant effect (F( 2, 1,227) = 160.52, p <.001). Follow-up contrasts reveal that, compared with the baseline condition (M = 4.09, SD = 1.27), participants reported greater self-validation when the customer from New Zealand decided to purchase the product (higher shipping cost: M = 5.46, SD = .98; t( 1,227) = 17.43, p <.001; lower shipping cost: M = 5.07, SD = 1.12; t( 1,227) = 12.38, p <.001). In support of our theorizing, we further find significantly higher feelings of self-validation in the case of higher (vs. lower) shipping costs (t( 1,227) = 4.98, p <.001).
We conducted mediation analyses ([21], Model 4, n = 5,000) with our multicategorical independent variable, happiness as the dependent variable, and self-validation as the mediator. We find positive and significant indirect effects on happiness through self-validation when comparing ( 1) the higher shipping cost condition with the baseline condition (b = .78, SE = .07, CI95% = [.65,.91]), ( 2) the lower shipping cost condition with the baseline condition (b = .55, SE = .06, CI95% = [.45,.67]), and ( 3) the higher shipping cost condition with the lower shipping cost condition (b = .22, SE = .05, CI95% = [.14,.31]). These results are robust to the inclusion of current profits and future profit expectations as covariates (for detailed results, see WA-E2).
Study 3 further corroborates our theorizing by showing that the extent to which sales provide self-validation, over and above monetary outcomes for the producer, depends on the monetary cost of the sales signal. In particular, participants reported greater self-validation and thus happiness when the buyer accepted to pay higher (vs. lower) shipping costs (H4).
Study 4 compares the effect of sales with that of a noneconomic signal. We focus on comparing sales with likes, which are arguably the most common form of market signals on electronic platforms such as Etsy. Comparing the effects of sales and likes is also important from a practical point of view because the seller dashboards of prominent online platforms tend to display sales and likes (or related forms of noneconomic signals such as favorites) concurrently, raising the question of how these different forms of signals impact individual producers' happiness. We hypothesize that individual producers experience greater self-validation and thus more happiness from sales than from likes, even above and beyond the monetary rewards from these sales (H5). We tested this by conducting another two-stage behavioral experiment in which we first asked participants to actually produce products and then manipulated at a later stage whether their products were either acquired or liked by customers. To test whether the happiness advantage of sales over likes goes beyond their cost difference to the customer, Study 4 kept cost to the customer (along with the monetary outcomes for the producer) constant across signals.
We invited 1,000 American workers on Prolific to participate in a two-stage study in which we asked them to demonstrate their writing skills. The experiment employed a 2 (signal: sales vs. likes) × 2 (number: high vs. low) between-participants design. All participants were told that their task would be to create a positive slogan for a "post-COVID" event that would be taking place at our university. We told participants that we would print each slogan on a poster and exhibit each poster at the event like a gallery exhibition. Furthermore, participants were told that all guests of the event would pay an entrance fee of $1.50 to cover costs and that, in return, each guest would receive a token. In the sales condition, we told participants that guests could use this token to purchase a poster of their choice and that each poster could only be purchased once. In the likes condition, we told participants that guests could use this token to like a poster of their choice by pinning the token on the poster and that each poster could only be liked once. Next, we asked all participants (those in both the sales and likes conditions) to create their slogan by finishing the following sentence: "When Corona is over...."
Two weeks later, we invited those who had submitted valid slogans (N = 1,000) to participate in the second part of the study, yielding 843 participants (Mage = 34.70 years, SD = 12.63, 48.2% female). A chi-squared test revealed that participation in the second part of the study did not depend on whether participants were assigned to the sales or the likes condition in the first part (χ2( 1) = .11, p = .75). First, we thanked all participants for submitting their slogan and reminded them that all slogans were printed on posters exhibited at the "post-COVID" event, at which guests could either purchase (sales condition) or like (likes condition) a poster of their choice. In addition, we reminded participants that each poster could only be purchased/liked once. Next, participants in the sales condition were either told that one (high sales) or no (low sales) customer(s) bought the poster with their slogan on it, while participants in the likes condition were either told that that one (high likes) or no (low likes) customer(s) liked the poster with their slogan on it. Next, all participants indicated their level of happiness (r = .91) and their feelings of self-validation (α = .96) on the same scales as in the previous studies. Finally, participants responded to an attention check ("how many guests [purchased/liked] the poster with your slogan on it?"; "no (0) guest [bought/liked] the poster with my slogan on it/one ( 1) guest [bought/liked] the poster with my slogan on it")[15] and indicated their gender and age (for study materials, see WA-F).
A 2 × 2 ANOVA on happiness produces significant main effects of signal (Msales = 4.43, SD = 1.64 vs. Mlikes = 4.21, SD = 1.60; F( 1, 839) = 7.28, p = .007) and number (Mhigh = 5.37, SD = 1.27 vs. Mlow = 3.28, SD = 1.20; F( 1, 839) = 610.80, p <.001). More importantly, we obtained the predicted signal by number interaction (F( 1, 839) = 9.23, p = .002; see Figure 3, Panel A). Planned contrasts reveal that the effect of the sales signal was significantly stronger (Mhigh = 5.61, SD = 1.17 vs. Mlow = 3.27, SD = 1.11; F( 1, 839) = 386.45, p <.001) than the effect of the likes signal (Mhigh = 5.13, SD = 1.32 vs. Mlow = 3.30, SD = 1.29; F( 1, 839) = 234.11, p <.001). In addition, we find that participants whose poster was sold were happier than participants whose poster was liked (F( 1, 839) = 16.36, p <.001). We detected no such differences between participants whose poster was not sold versus not liked (F( 1, 839) = .06, p = .81).
Graph: Figure 3. Happiness and self-validation as a function of signal and value (Study 4).
A similar 2 × 2 ANOVA on feelings of self-validation produces a significant main effect of signal (Msales = 3.62, SD = 1.48 vs. Mlikes = 3.41, SD = 1.50; F( 1, 839) = 5.44, p = .02) and a significant main effect of number (Mhigh = 4.05, SD = 1.40 vs. Mlow = 3.00, SD = 1.38; F( 1, 839) = 121.20, p <.001). This main effect was qualified by a significant interaction effect (F( 1, 839) = 7.36, p = .007; see Figure 3, Panel B), demonstrating that the effect of the sales signal was significantly stronger (Mhigh = 4.29, SD = 1.29 vs. Mlow = 2.98, SD = 1.37; F( 1, 839) = 94.48, p <.001) than the effect of the likes signal (Mhigh = 3.81, SD = 1.47 vs. Mlow = 3.02, SD = 1.39; F( 1, 839) = 34.29, p <.001). In addition, participants whose poster was sold felt more validated than participants whose poster was liked (F( 1, 839) = 12.66, p <.001). Feelings of self-validation did not differ between participants whose poster was not sold versus not liked (F( 1, 839) = .07, p = .79).
A moderated mediation analysis ([21], Model 7, n = 5,000 bootstraps) with number (0 = low, 1 = high) as the independent variable, signal (0 = likes, 1 = sales) as the moderator, self-validation as the mediator, and happiness as the dependent variable produces a significant index of moderated mediation (b = .25, SE = .09, CI95% = [.07,.44]). As expected, the mediating effect through self-validation on happiness was stronger when participants' posters were sold versus not sold (b = .63, SE = .08, CI95% = [.48,.78]) than when participants' posters were liked versus not liked (b = .38, SE = .07, CI95% = [.25,.51]).
Study 4 compared the effects of sales and likes regarding their impact on individual producers' feelings of self-validation and happiness as a producer (H5). In support of our theorizing, we find that sales produce stronger effects on happiness and self-validation than likes, even when the associated monetary costs were kept constant between conditions. In a supplementary study (see Study 4S in WA-G), we tested the effects of sales and likes when presenting information about sales and likes simultaneously (mimicking the situation on online platforms such as Etsy) among a sample of recreational knitters (n = 161). Holding the monetary rewards to the producer (but not the costs to the customer) from sales and likes constant, we again find that sales make individual producers happier than likes and that this effect is driven by feelings of self-validation.
Study 5 tests whether the magnitude of the sales-as-signal effect depends on whether individuals sell their self-made products or products that were made by someone else. We tested this by varying sales of products that were self-produced versus produced by someone else. Although selling more versus fewer products should increase individuals' happiness in both situations, we predict an incremental increase in happiness from selling self-made products (H6).
Participants included 1,008 U.S. consumers recruited from MTurk (Mage = 39.81 years, SD = 12.41, 47.6% female). The experiment employed a 2 (sales: high vs. low) × 2 (product: self-made vs. other-made) between-participants design. All participants imagined selling muffins at a local food market. The muffins were either made by themselves (self-made condition) or by someone else (other-made condition). To keep the monetary rewards constant across conditions, we told participants that the organizers of the food market receive all sales revenue from people selling products at the market for the first time (which is how the food market finances itself). We further informed participants that there are no other costs involved in being able to sell products at the food market. Thus, across conditions, the monetary reward from potential sales was zero. Below this description, participants responded to an attention check verifying they understood that revenue from their first-time sales would go to the food market. Next, participants either read that they made (self-made condition) or received (other-made condition) a total of 50 muffins to sell and that they sold either 36 (high sales condition) or six (low sales condition) of them at the food market.
After reading this information, participants indicated their level of happiness on the same scale as in the previous studies (r = .92). Next, participants completed two more attention checks ("which of the following statements is correct?"; "at the food market, I sold muffins that were made by someone else/at the food market, I sold muffins that I made myself"; "how many muffins were sold at the food market?"; 6/36).[16] Finally, participants indicated their gender and age (for complete study materials, see WA-H).
A 2 × 2 ANOVA on happiness reveals the expected significant main effect of sales (Mhigh_sales = 4.53, SD = 1.54 vs. Mlow_sales = 2.59, SD = 1.49; F( 1, 1,004) = 418.73, p <.001) and a nonsignificant main effect of product (Mself-made = 3.58, SD = 1.83 vs. Mother-made = 3.55, SD = 1.78; F( 1, 1,004) = .18, p = .67). Importantly, we also obtained a significant interaction effect (F( 1, 1,004) = 19.65, p <.001; see Figure 4). As hypothesized, planned contrasts reveal that the effect of sales on happiness was significantly stronger when participants sold their self-made muffins (Mhigh_sales = 4.77, SD = 1.42 vs. Mlow_sales = 2.41, SD = 1.37; F( 1, 1,004) = 310.54, p <.001) than when participants sold muffins made by someone else (Mhigh_sales = 4.31, SD = 1.62 vs. Mlow_sales = 2.79, SD = 1.59; F( 1, 1,004) = 128.21, p <.001).
Graph: Figure 4. Happiness as a function of product and sales (Study 5).
Importantly, we find that participants who sold more muffins were happier when they made the muffins themselves versus someone else making the muffins (Mself-made = 4.77, SD = 1.42 vs. Mother-made = 4.31, SD = 1.62; F( 1, 1,004) = 11.85, p = .001). In contrast, participants who sold fewer muffins were happier when the muffins were made by someone else versus by themselves (Mself-made = 2.41, SD = 1.37 vs. Mother-made = 2.79, SD = 1.59; F( 1, 1,004) = 8.00, p = .005).
The results of Study 5 complement the previous studies by demonstrating that sales have a stronger effect on individuals' happiness when individuals sell their self-made products than when they sell products that were made by someone else (H6). This finding provides additional evidence that the effect of selling self-produced products on happiness is driven by the effect of sales on self-validation as a competent producer, and it rules out the alternative explanation that the effect is driven solely by the effect of sales on self-validation as a competent seller or marketer.
Study 6 further tests whether selling their self-made products has any incremental effects on producers' happiness compared with merely producing products. Because sales function as a credible signal regarding producers' skills and competencies, we expect that sales affect individual producers' self-validation and thus happiness beyond individual producers' self-validation and happiness derived from production. We tested this with another two-stage behavioral experiment in which we again asked participants to produce their own products and manipulated at a later stage whether their products were sold. In addition, we added a control condition in which participants' products were not offered for sale, and we assessed participants' feelings of self-validation and happiness derived from both production and selling. Doing so allowed us to assess any increase in self-validation and happiness as a result of a successful sale or any backfiring effect in the case of no sales, compared with the baseline of "not going to the market" to begin with.
We invited 500 U.S. consumers on Prolific to participate in a two-stage study in which we asked them to demonstrate their writing skills. In the study description, all participants were informed that we would ask them to generate a slogan followed by a short survey and that they would receive a second survey in about two weeks. The experiment employed a 2 (stage: production vs. selling) × 3 (sales: control vs. product sold vs. product not sold) mixed design with stage as the within-subjects factor and sales as the between-subjects factor. In the production stage, we told all participants that their task would be to create a positive slogan one could print on a T-shirt that describes what they will do when the COVID crisis is over. Participants in the sales conditions were additionally told that all slogans from this study would be offered to our university community and that, if a given slogan finds a customer, that slogan will be printed on a T-shirt for that customer. We further informed participants in the sales conditions that each slogan will only be printed on a T-shirt once and that the price the customer will pay for the T-shirt will equal the cost incurred in having it produced. We also told participants in the sales conditions that the second survey would inform them about whether the T-shirt with their slogan had been purchased by a customer. Next, we asked all participants (those in the control and in the two sales conditions) to create their slogan by finishing the following sentence: "When Corona is over...." After creating their slogan, all participants indicated their level of happiness (r = .78) and their feelings of self-validation (α = .96) on the same scales used in the previous studies. Finally, we reminded all participants that they would receive a second survey in about two weeks.
Two weeks later, we invited participants who had submitted valid slogans (N = 499) to participate in the second part of the study. A total of 384 participants accepted this invitation (Mage = 31.56, SD = 10.51, 45.6% female).[17] First, we thanked all participants for submitting their T-shirt slogan two weeks earlier. Participants in the sales conditions were additionally reminded that all slogans were offered to our university community and that the T-shirt with their slogan on it could only be purchased once. Next, we informed participants in the sales conditions that their T-shirt was either sold or not. Finally, all participants indicated their level of happiness (r = .85) and their feelings of self-validation (α = .97) on the same scales as in the production stage, as well as their gender and age (for study materials, see WA-I).
A 2 × 3 mixed ANOVA on happiness with stage (production vs. selling) as the within-subject factor and sales (control vs. product sold vs. product not sold) as the between-subjects factor reveals the expected significant stage by sales interaction (F( 2, 381) = 127.47, p <.001; see Figure 5, Panel A). Participants' happiness did not differ across the sales conditions in the production stage (Mcontrol = 5.19, SD = 1.12 vs. Msold = 5.17, SD = 1.21 vs. Mnot_sold = 5.39, SD = 1.15; F( 2, 381) = 1.34, p = .26). In contrast, participants' happiness significantly differed across the sales conditions in the selling stage (F( 2, 381) = 157.63, p <.001). As expected, participants were happier when a T-shirt with their slogan on it was sold (M = 5.87, SD = 1.02) than when a T-shirt with their slogan on it was not sold (M = 3.37, SD = 1.17; t(381) = 17.47, p <.001). Compared with the control condition (M = 4.97, SD = 1.21), participants were happier when a T-shirt with their slogan on it was sold (t(381) = 6.26, p <.001) and less happy when a T-shirt with their slogan on it was not sold (t(381) = −11.40, p <.001).
Graph: Figure 5. Happiness (left) and self-validation (right) as a function of stage and sales (Study 6).
Comparing participants' happiness across the two stages reveals that, in the product sold condition, participants were happier in the selling stage (M = 5.87, SD = 1.02) than in the production stage (M = 5.17, SD = 1.21; F( 1, 381) = 30.93, p <.001). In the product not sold condition, participants were less happy in the selling stage (M = 3.37, SD = 1.17) than in the production stage (M = 5.39, SD = 1.15; F( 1, 381) = 279.11, p <.001). In the control condition, participants were marginally less happy in the selling stage (M = 4.97, SD = 1.21) than in the production stage (M = 5.19, SD = 1.12; F( 1, 381) = 3.38, p = .07). The main effect of stage was also significant (Mselling = 4.72, SD = 1.53 vs. Mproduction = 5.25, SD = 1.58; F( 1, 381) = 53.90, p <.001).
A similar 2 × 3 mixed ANOVA on self-validation reveals a significant stage by sales interaction (F( 2, 381) = 28.33, p <.001; see Figure 5, Panel B). In the production stage, participants' feelings of self-validation did not differ across the sales conditions (Mcontrol = 3.88, SD = 1.36 vs. Msold = 3.85, SD = 1.32 vs. Mnot_sold = 3.92, SD = 1.52; F( 2, 381) = .09, p = .91). In contrast, participants' feelings of self-validation significantly differed across the sales conditions in the selling stage (F( 2, 381) = 23.71, p <.001). As expected, participants felt more validated when a T-shirt with their slogan on it was sold (M = 4.35, SD = 1.27) than when a T-shirt with their slogan on it was not sold (M = 3.14, SD = 1.61; t(381) = 6.80, p <.001). Compared to the control condition (M = 3.89, SD = 1.35), participants felt more validated when a T-shirt with their slogan on it was sold (t(381) = 2.56, p = .01) and less validated when a T-shirt with their slogan on it was not sold (t(381) = −4.31, p <.001).
In the product sold condition, participants felt more validated in the selling stage (M = 4.35, SD = 1.27) than in the production stage (M = 3.85, SD = 1.32; F( 1, 381) = 16.47, p <.001). In the product not sold condition, participants felt less validated in the selling stage (M = 3.14, SD = 1.61) than in the production stage (M = 3.92, SD = 1.52; F( 1, 381) = 42.19, p <.001). In the control condition, participants' feelings of self-validation did not differ between the selling stage (M = 3.89, SD = 1.35) and the production stage (M = 3.88, SD = 1.36; F( 1, 381) = .01, p = .92). The main effect of stage was not significant (F( 1, 381) = 1.61, p = .21).
We conducted two mediation analyses ([21]; Model 4, n = 5,000 bootstraps) to test whether feelings of self-validation can explain the effect of our sales manipulation on happiness. We first looked at participants' ratings of happiness and self-validation obtained in the selling stage. We thus entered sales as the independent variable, self-validation in the selling stage as the mediator, and happiness in the selling stage as the dependent variable into the regression. This analysis produced positive indirect effects when comparing the product sold with the product not sold condition (b = .45, SE = .09, CI95% = [.29,.64]) and the control condition (b = .17, SE = .07, CI95% = [.05,.31]), and it produced a negative indirect effect when comparing the product not sold condition with the control condition (b = −.28, SE = .08, CI95% = [−.45, −.14]).
We next examined whether relative differences in happiness between the selling stage and the production stage can be explained by relative differences in self-validation between the selling stage and the production stage. To do so, we calculated difference scores between happiness and self-validation ratings in the selling stage and the production stage. The respective regression analyses ([21]; Model 4, n = 5,000 bootstraps) produced positive indirect effects when comparing the product sold with the product not sold condition (b = .55, SE = .10, CI95% = [.36,.76]) and the control condition (b = .21, SE = .08, CI95% = [.07,.37]), and it produced a negative indirect effect when comparing the product not sold condition with the control condition (b = −.34, SE = .08, CI95% = [−.51, −.19]).
By including a preselling baseline measure as well as a no-selling control condition, the results of Study 6 confirm there is a positive incremental effect of selling self-produced products on happiness. They also show, however, that marketing self-produced products can have a downside as well. Specifically, there is a happiness penalty to pay when products fail to sell, as the absence of sales leads to negative self-validation; that is, failing to sell products sends a negative signal about one's skills and competencies as a producer, reducing happiness. Thus, deciding to offer their self-made products for sale can enhance but also diminish individuals' happiness.
This research investigates the socioemotional benefits of selling self-made products. Eight studies provide evidence for a sales-as-signal effect: individual producers derive happiness from selling their products above and beyond the money they make from these sales (all studies). This is because sales validate their skills and competencies as a producer (Studies 2–4 and 6). In addition, Study 2 shows that the sales-as-signal effect is more pronounced when the choice mechanism that precedes the purchase is more versus less deliberate. Study 3 demonstrates that the increase in happiness from sales is higher when the buyer incurs higher (vs. lower) monetary costs in purchasing the product, even when this higher cost does not translate into higher financial return for the individual producer. Study 4 demonstrates that individual producers gain more happiness from sales than from receiving likes. Finally, Studies 5 and 6 show that individuals derive greater happiness from high sales of self-made products than from high sales of products that were made by someone else (Study 5) and that individuals show increases in happiness from pre- to post-selling but show decreases in happiness after (vs. before) failing to sell (Study 6). The studies (N = 4,970) span a variety of methodological approaches, designs, and procedures, and they feature different participant populations across three continents and producer communities.
Our research makes a number of theoretical contributions. First, previous research has extensively studied the psychological and behavioral consequences of engaging in self-production (e.g., [17]; [31]). This line of research has, however, focused on studying production for oneself or for gift giving. Our work goes a step further and examines a context in which individuals produce for the market; that is, with the aim of selling their creations to others. Our contribution thus lies in shedding light on the psychological consequences of participating in market exchanges. Our findings suggest that sales provide individual producers with a sense of self-validation and, in turn, happiness. We find that selling self-made products affects individuals' happiness beyond the happiness derived from producing the products.
Second, we introduce a new perspective on the value of sales. Most models of producer behavior assume that producers are driven solely by a profit-maximization motive ([18]; [36]). Our findings caution against taking a reductionist view of sales by demonstrating that selling one's self-made products can also have important socioemotional value, as sales provide individual producers with self-validation regarding their skills and competencies as a producer. Thus, models of producer behavior may benefit from including the self-validation motive documented in this research.
Third, in developing our theory, we conceptualized sales as a signal from buyers. This approach is different from the process investigated by existing research on signaling in marketing and management that conceptualizes signals as actions taken by sellers, who have low uncertainty about the product, to reduce uncertainty about product qualities for buyers ([ 5]; [ 8]; [23]). In our research, we find that sales constitute a signal that is sent by buyers and received by sellers. Moreover, our work suggests that sellers, even of self-produced products, actually do have uncertainty about their own products' qualities that is reduced by quality information from buyers. Thus, the traditional roles about who has information and who is uncertain are to some extent reversed. There may be less information asymmetry than is often assumed, and quality information may be exchanged in both directions. Furthermore, signals are traditionally conceptualized as intentional actions performed by the sender that in turn benefit the sender ([ 8]). Our account, in contrast, suggests that sending a signal can be a more or less incidental by-product (of a purchase) rather than an intentional signal and that the signal benefits the receiver of the signal (i.e., individual producer) by increasing self-validation. In addition, we demonstrate that the strength of a signal depends not only on the signal's costliness but also on its diagnosticity. For example, even when likes and sales are equally costly for the customer, sales may be seen as a more reliable indicator of product quality and producer skill. Likewise, a buyer's deliberate choice to purchase an item makes the sale of that item more diagnostic of producer skill than a sale that involves a buyer randomly picking a specific product. Thus, we believe that the present work offers a novel perspective on signaling.
Finally, our research informs the ongoing debate within the marketing discipline regarding the implications of marketing for society ([ 7]). The backdrop of this debate involves widespread concern about the pernicious aspects of a consumer society ([24]), the negative environmental externalities of market exchanges ([25]), and the potentially exploitative nature of common marketing practices ([40]). We find that marketing one's products can actually have an important positive effect by providing individual producers with a significant happiness benefit through self-validation. We thus demonstrate how studying a classic marketing topic from a "better world" perspective can yield novel insights about marketing's potential to improve people's well-being. Moreover, marketing scholars studying the implications of marketing on society frequently focus on activities undertaken by marketers at large corporations, and more research is needed to uncover the societal impact of marketing beyond these contexts ([ 7]). Answering this call, we study the socioemotional benefits of individual producers participating in market exchanges. Our work also demonstrates the value of taking a behavioral approach to the study of supply-side behaviors. Despite calls for behavioral marketing research to broaden its focus beyond consumers ([27]; [44]), virtually all research by behavioral marketing scholars currently focuses on consumer behavior. We provide an example of how researchers can tackle important marketing phenomena on the supply side with a behavioral lens.
In addition to these theoretical contributions, our work also provides several practical implications, especially for online marketplaces that focus on producers selling self-made products. First, the finding that sales can increase individual producers' happiness could be leveraged in, for example, the recruitment of prospective sellers by highlighting the socioemotional benefits of selling their products (e.g., "Be a pro. Sell your tote bags on Etsy!"). Online marketplaces could also highlight socioemotional benefits to existing sellers to maintain motivation and retention. This could be done, for example, by stressing that customers recognize sellers' expertise by paying good money to buy their products (e.g., "They voted with their wallets to tell you you're a pro").
In addition, our findings yield recommendations for the design of seller dashboards. Study 4 demonstrates that knowing how many people bought their products makes individual producers happier than knowing how many people liked their products. Study 3 suggests that showing how much customers paid might increase motivation, and thus production volume per seller and seller retention, beyond the amount of money the seller made. Therefore, we recommend designing seller dashboards so that the number of people who have made purchases and the average amount paid by customers (including shipping and other fees) are more prominent, rather than highlighting likes or aggregate revenue as is common in such dashboards.
Finally, Study 2 shows that sales make individual producers happier when they know that buyers deliberately chose their products (vs. being chosen at random). This finding could be leveraged, for example, by encouraging buyers to leave a comment indicating why they chose that seller's product over other available product choices or by highlighting that customers decided to buy the focal producer's product despite having many other options.
We hope that our findings will motivate other researchers to explore the under-researched topic of individual producers' motivations, beyond monetary considerations. Specifically, we identify several opportunities for future research. Our findings indicate that the sales-as-signal effect is strongly driven by individual producers' gain in self-validation as a competent producer (e.g., in Studies 2, 4, and 6 sales increased individual producers' happiness even when their self-made products were sold by someone else). This focus on self-validation as a producer was motivated by the fact that sellers on platforms such as Etsy typically spend more than half of their time designing and making their products and only about 10% of their time with marketing and promotion activities ([16]). However, to successfully sell their products, many individual producers not only engage in skilled production activities but also in skilled promotion- or selling-related activities, such as taking pictures of their offerings, pricing their products, maintaining their online appearance, and engaging in advertising on social media platforms. Therefore, sales might, at least for some people and in some contexts, also be an important source of happiness by validating individual producers' skills as a promoter or seller. Study 5 indeed shows that sales also increase the happiness of individuals selling products made by someone else above and beyond the monetary rewards from sales—though to a lesser extent compared to individuals selling their self-made products. Empirically examining the self-validation processes involved in selling would be a worthwhile future research direction. Likewise, while the entrepreneurship literature documents that financial success is related to entrepreneurs' satisfaction ([13]; [32]), future research could investigate whether entrepreneurial success (e.g., firm growth, increase in funding) has a causal effect on entrepreneurs' happiness beyond the related financial gains. In sum, our work provides a fruitful foundation for investigating the socioemotional benefits of participating in market exchanges across business contexts.
Our studies investigated two characteristics of the purchase transaction—the deliberateness of the product choice and the monetary cost of buying—that moderate the strength of the sales signal. However, other important moderators of the sales-as-signal effect likely exist, and future research should expand the nomological network in which the sales-as-signal effect is situated. For example, the strength of the sales signal might depend on the expertise of the buyer. Compared to novices, experts are more capable of assessing products' quality ([ 1]). Having one's products bought by an expert in the respective product category should thus more credibly inform individual producers about their competencies as a producer than having one's products bought by a novice. Similarly, the number of buyers might affect how strongly individual producers perceive sales to correspond to the quality of their products. Would 40 buyers purchasing one product each provide more self-validation than four buyers purchasing ten items each?
Besides the direct monetary cost associated with buying a product (Study 3), the sales-as-signal effect might also depend on the relative cost that buyers incur when purchasing a product. We would expect the effect of sales to be stronger when a buyer has a smaller (vs. larger) budget to spend. Likewise, relative price might also play a role, as the same cost might be a stronger signal of expertise when that cost is high compared to other products in the category than when that same cost is low compared to other products in the category. Finally, although customers accepting a high price should increase the strength of the sales signal, individual producers accepting higher prices might reach a point where they feel that customers are being treated unfairly (such as when customers have to pay extraordinarily high shipping costs) or that the prices might cause customers to refrain from making repeat purchases. Do concerns about fairness and relationship-building reduce the happiness individual producers gain from selling products at exorbitant prices?
One could also argue that the strength of the sales-as-signal effect depends on certain characteristics of individual producers. We used the data obtained in Study 1 to test whether the effect of sales on happiness is moderated by any of the captured control variables. Only one statistically significant interaction emerged for both measures of our independent variable (i.e., sales volume and sales growth): a positive interaction effect between sales and producers' socioeconomic status (ps <.01; see WA-J1). This suggests that the effect of sales on happiness is stronger for producers that have a higher socioeconomic status—in other words, producers who feel financially secure. We also used the data obtained in Study 1S to examine whether the effect of sales on happiness is moderated by any of the captured control variables. Moderation analyses produce nonsignificant interaction effects between ( 1) sales and producers' experience and ( 2) sales and socioeconomic status (ps >.20; see WA-J2). Thus, although the moderation analyses of Study 1 suggest that the sales-as-signal effect is stronger for producers that have a relatively higher (vs. lower) socioeconomic status, we did not observe such an interaction effect in Study 1S. Therefore, future research might more closely look at the role socioeconomic status plays in the sales-as-signal effect.
The strength of the sales-as-signal effect might also depend on individual producers' own evaluations of their products—that is, how competent individuals feel in producing their products might alter how happy they feel as a result of selling those products. We used the data of Study 6 to test this. A moderation analysis produces a nonsignificant interaction between sales (higher vs. lower) and self-validation from production on happiness from sales (p = .98; see WA-J3). This suggests that the positive effect of sales on happiness does not depend on producers' presales perception of their products. We encourage future researchers to look deeper into these, and possibly other, potential moderators of the sales-as-signal effect.
Traditional producer behavior models assume that producers' behavior is driven by the monetary rewards from selling their wares and that there is an information asymmetry that favors producers. In other words, these models assume producers have information about the quality of their products that (prospective) customers do not, leading producers to signal product quality to customers. This research shows that producers derive happiness from selling their wares that goes above and beyond any monetary rewards. It shows that there is also an information asymmetry in the other direction: Customers possess information about the skill level of the producer that they signal by purchasing the producer's products. This provides a feeling of self-validation to the producer, increasing their happiness. To understand the behavior of producers, it is critical to broaden our scope from purely monetary to self-validation benefits and from assuming a one-directional flow of quality information to a two-directional flow, with at least some quality signals being sent from customers to producers.
sj-pdf-1-jmx-10.1177_00222429211064263 - Supplemental material for Sales and Self: The Noneconomic Value of Selling the Fruits of One's Labor
Supplemental material, sj-pdf-1-jmx-10.1177_00222429211064263 for Sales and Self: The Noneconomic Value of Selling the Fruits of One's Labor by Benedikt Schnurr, Christoph Fuchs, Elisa Maira, Stefano Puntoni, Martin Schreier and Stijn M.J. van Osselaer in Journal of Marketing
Footnotes 1 The authors thank Anouk Vandendael for conducting the qualitative study and Karen Petersen, Louise McCauley, and Silvia Cuesta for helping us collect data.
2 The second, third, fourth, fifth, and sixth authors contributed equally to the paper and are listed in alphabetical order.
3 Cait Lamberton
4 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
5 The author(s) received no financial support for the research, authorship and/or publication of this article.
6 Benedikt Schnurr https://orcid.org/0000-0001-7561-5390 Christoph Fuchs https://orcid.org/0000-0002-8457-1294
7 Although the focus of our research is on market interactions in which the existing literature presumes that signals originate from one party (seller) and are received by another party (buyer), we acknowledge that the negotiations literature has demonstrated that both parties in a negotiation send signals to alter their counterparts' perception of power, interest, and concern ([4]; [10]).
8 We thank the associate editor for this suggestion.
9 An initial data inspection revealed that 43 participants indicated higher profit than revenue. As these patterns are implausible and could potentially invalidate our results, we excluded these participants from our main analyses. Core results remained statistically significant when we ran the analyses on the entire sample. Furthermore, we find that revenue and profit were highly correlated (r =.93). To avoid any multicollinearity issues, we only added one of the two measures at once to our regression models (see Table 2).
We aimed for a sample size of at least 240 participants, and we therefore invited 700 MTurk workers in the first wave. However, this resulted in only 184 workers who completed the study. To achieve the prespecified minimum sample size, we invited another 1,000 MTurk workers, which added 163 workers to the sample. The total sample therefore consisted of 347 participants. Participants' gender and age did not differ between the two waves (ps >.80). Moreover, whether participants were in the first or in the second wave had no significant impact on the observed main or interaction effects (ps >.17).
In this and the remaining studies, we find evidence for discriminant validity between happiness and self-validation. Fornell–Larcker tests revealed that the AVEs for the two constructs were higher than the shared variance between them, supporting discriminant validity.
Results are robust when removing 56 participants who failed at least one of the attention checks.
Results are robust when removing 62 participants who failed at least one attention check.
Separate 2 (origin) × 3 (sales) ANOVAs on happiness and self-validation produce nonsignificant interaction effects (happiness: F(1, 1,224) = 1.73, p =.18; self-validation: F(1, 1,224) =.44, p =.64), indicating that the results are unaffected by whether participants were from Europe or the United States.
Results are robust when removing eight participants who failed the attention check.
Results are robust when removing 158 participants who failed at least one of the three attention checks.
A chi-squared test reveals that participation in the second part of the study did not depend on participants' assignment to the conditions in the first part (χ2(2) = 1.20, p =.55). In addition, first-stage happiness and self-validation did not differ between participants who participated in the second part and those who did not (happiness: t(497) =.10, p =.92; self-validation: t(497) = −.85, p =.40).
References Alba Joseph W. , Hutchinson J. Wesley. (1987), " Dimensions of Consumer Expertise ," Journal of Consumer Research , 13 (4), 411 – 54.
Ariely Dan , Kamenica Emir , Prelec Dražen. (2008), " Man's Search for Meaning: The Case of Legos ," Journal of Economic Behavior & Organization , 67 (3/4), 671 – 77.
Basuroy Suman , Desai Kalpesh Kaushik , Talukdar Debabrata. (2006), " An Empirical Investigation of Signaling in the Motion Picture Industry ," Journal of Marketing Research , 43 (2), 287 – 95.
Belkin Liuba Y. , Kurtzberg Terri R. , Naquin Charles E.. (2013), " Signaling Dominance in Online Negotiations: The Role of Affective Tone ," Negotiation and Conflict Management Research , 6 (4), 285 – 304.
Boulding William , Kirmani Amna. (1993), " A Consumer-Side Experimental Examination of Signaling Theory: Do Consumers Perceive Warranties as Signals of Quality? " Journal of Consumer Research , 20 (1), 111 –23.
Busenitz Lowell W. , Fiet James O. , Moesel Douglas D.. (2005), " Signaling in Venture Capitalist–New Venture Team Funding Decisions: Does It Indicate Long-Term Venture Outcomes? " Entrepreneurship Theory and Practice , 29 (1), 1 – 12.
Chandy Rajesh K. , Johar Gita Venkataramani , Moorman Christine , Roberts John H.. (2021), " Better Marketing for a Better World ," Journal of Marketing , 85 (3), 1 – 9.
Connelly Brian L. , Trevis Certo S. , Duane Ireland R. , Reutzel Christopher R.. (2011), " Signaling Theory: A Review and Assessment ," Journal of Management , 37 (1), 39 – 67.
Dahl Darren W. , Moreau C. Page. (2007), " Thinking Inside the Box: Why Consumers Enjoy Constrained Creative Experiences ," Journal of Marketing Research , 44 (3), 357 – 69.
Dawson Greg , Watson Richard , Boudreau Marie-Claude , Pitt Leyland F.. (2016), " A Knowledge-Centric Examination of Signaling and Screening Activities in the Negotiation for Information Systems Consulting Services ," Journal of the Association for Information Systems , 17 (2), 77 – 106.
Deci Edward L. , Ryan Richard M.. (2000), " The 'What' and 'Why' of Goal Pursuits: Human Needs and the Self-Determination of Behavior ," Psychological Inquiry , 11 (4), 227 – 68.
De Langhe Bart , Fernbach Philip M. , Lichtenstein Donald R.. (2016), " Navigating by the Stars: Investigating the Actual and Perceived Validity of Online User Ratings ," Journal of Consumer Research , 42 (6), 817 – 33.
Dijkhuizen Josette , Gorgievski Marjan , van Veldhoven Marc , Schalk René. (2018), " Well-Being, Personal Success, and Business Performance Among Entrepreneurs: A Two-Wave Study ," Journal of Happiness Studies , 19 (8), 2187 – 204.
Dodds William B. , Monroe Kent B. , Grewal Dhruv. (1991), " Effects of Price, Brand, and Store Information on Buyers' Product Evaluations ," Journal of Marketing Research , 28 (3), 307 – 19.
Dutta Prajit K. , Radner Roy. (1999), " Profit Maximization and the Market Selection Hypothesis ," Review of Economic Studies , 66 (4), 769 – 98.
Etsy (2019), "Celebrating Creative Entrepreneurship Around the Globe," https://extfiles.etsy.com/advocacy/Etsy_GlobalSellerCensus_4.2019.pdf.
Franke Nikolaus , Schreier Martin , Kaiser Ulrike. (2010), " The 'I Designed It Myself' Effect in Mass Customization ," Management Science , 56 (1), 125 – 40.
Friedman Milton. (1970), " The Social Responsibility of Business Is to Increase Its Profits ," New York Times Magazine (September 13) , https://www.nytimes.com/1970/09/13/archives/a-friedman-doctrine-the-social-responsibility-of-business-is-to.html.
Grant Adam M.. (2007), " Relational Job Design and the Motivation to Make a Prosocial Difference ," Academy of Management Review , 32 (2), 393 – 417.
Griskevicius Vladas , Tybur Joshua M. , Delton Andrew W. , Robertson Theresa E.. (2011), " The Influence of Mortality and Socioeconomic Status on Risk and Delayed Rewards: A Life History Theory Approach ," Journal of Personality and Social Psychology , 100 (6), 1015 – 26.
Hayes Andrew F.. (2013), Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach. New York : Guilford Press.
Kihlstrom Richard E. , Riordan Michael H.. (1984), " Advertising as a Signal ," Journal of Political Economy , 92 (3), 427 – 50.
Kirmani Amna , Rao Akshay R.. (2000), " No Pain, No Gain: A Critical Review of the Literature on Signaling Unobservable Product Quality ," Journal of Marketing , 64 (2), 66 – 79.
Klein Naomi. (2009), No Logo. New York: Picador.
Kotler Philip. (2011), " Reinventing Marketing to Manage the Environmental Imperative ," Journal of Marketing , 75 (4), 132 – 35.
Kristofferson Kirk , White Katherine , Peloza John. (2014), " The Nature of Slacktivism: How the Social Observability of an Initial Act of Token Support Affects Subsequent Prosocial Action ," Journal of Consumer Research , 40 (6), 1149 – 66.
MacInnis Deborah J. , Morwitz Vicki G. , Botti Simona , Hoffman Donna L. , Kozinets Robert V. , Lehmann Donald R. , et al. (2020), " Creating Boundary-Breaking, Marketing-Relevant Consumer Research ," Journal of Marketing , 84 (2), 1 – 23.
Marx Karl (1844/1993), Economic and Philosophical Manuscripts. Early Writings. Marx/Engels Internet Archive , https://www.marxists.org/archive/marx/works/1844/epm/index.htm.
Mochon Daniel , Norton Michael I. , Ariely Dan. (2012), " Bolstering and Restoring Feelings of Competence via the IKEA Effect ," International Journal of Research in Marketing , 29 (4), 363 – 69.
Moreau C. Page , Bonney Leff , Herd Kelly B.. (2011), " It's the Thought (and the Effort) That Counts: How Customizing for Others Differs from Customizing for Oneself ," Journal of Marketing , 75 (5), 120 – 33.
Norton Michael I. , Mochon Daniel , Ariely Dan. (2012), " The IKEA Effect: When Labor Leads to Love ," Journal of Consumer Psychology , 22 (3), 453 – 60.
Przepiorka Aneta M.. (2017), " Psychological Determinants of Entrepreneurial Success and Life-Satisfaction ," Current Psychology , 36 (2), 304 – 15.
Richer Sylvie F. , Blanchard Cealine , Vallerand Robert J.. (2002), " A Motivational Model of Work Turnover ," Journal of Applied Social Psychology , 32 (10), 2089 – 113.
Ryan Richard M. , Deci Edward L.. (2000), " Self-Determination Theory and the Facilitation of Intrinsic Motivation, Social Development, and Well-Being ," American Psychologist , 55 (1), 68 – 78.
Senik Claudia. (2009), " Direct Evidence on Income Comparisons and Their Welfare Effects ," Journal of Economic Behavior & Organization , 72 (1), 408 – 24.
Shane Scott , Venkataraman Sankaran. (2000), " The Promise of Entrepreneurship as a Field of Research ," Academy of Management Review , 25 (1), 217 – 26.
Shir Nadav , Nikolaev Boris N. , Wincent Joakim. (2019), " Entrepreneurship and Well-Being: The Role of Psychological Autonomy, Competence, and Relatedness ," Journal of Business Venturing , 34 (5), Article 105875.
Spence Andrew Michael. (1974), Market Signaling: Informational Transfer in Hiring and Related Screening Processes. Cambridge, MA: Harvard University Press.
Statista (2021), "Annual Gross Merchandise Sales (GMS) of Etsy Inc. from 2005 to 2020," Statista , (accessed August 31, 2021), https://www.statista.com/statistics/219412/etsys-total-merchandise-sales-per-year/.
Sunstein Cass R.. (2016), " Fifty Shades of Manipulation ," Journal of Marketing Behavior , 1 (3/4), 213 – 44.
Talley Amelia E. , Kocum Lucie , Schlegel Rebecca J. , Molix Lisa , Ann Bettencourt B.. (2012), " Social Roles, Basic Need Satisfaction, and Psychological Health: The Central Role of Competence ," Personality and Social Psychology Bulletin , 38 (2), 155 – 73.
Vandendael Anouk. (2014), "An Exploratory Case Study on the Motivational Factors of Etsy Sellers," (July 10), Erasmus University Thesis Repository, Marketing Management Series, http://hdl.handle.net/2105/20811.
Van Osselaer Stijn M.J. , Fuchs Christoph , Schreier Martin , Puntoni Stefano. (2020), " The Power of Personal ," Journal of Retailing , 96 (1), 88 – 100.
Wertenbroch Klaus. (2015), " From the Editor: An Opportunity for More Relevance from Broadening Behavioral Research in Marketing ," Journal of Marketing Behavior , 1 (1), 1 – 7.
~~~~~~~~
By Benedikt Schnurr; Christoph Fuchs; Elisa Maira; Stefano Puntoni; Martin Schreier and Stijn M.J. van Osselaer
Reported by Author; Author; Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 108- Salesperson Dual Agency in Price Negotiations. By: Lawrence, Justin M.; Scheer, Lisa K.; Crecelius, Andrew T.; Son K. Lam. Journal of Marketing. Mar2021, Vol. 85 Issue 2, p89-109. 21p. 2 Diagrams, 6 Charts. DOI: 10.1177/0022242920974611.
- Database:
- Business Source Complete
Record: 109- Serendipity: Chance Encounters in the Marketplace Enhance Consumer Satisfaction. By: Kim, Aekyoung; Affonso, Felipe M.; Laran, Juliano; Durante, Kristina M. Journal of Marketing. Jul2021, Vol. 85 Issue 4, p141-157. 17p. 1 Diagram, 3 Graphs. DOI: 10.1177/00222429211000344.
- Database:
- Business Source Complete
Serendipity: Chance Encounters in the Marketplace Enhance Consumer Satisfaction
Despite evidence that consumers appreciate freedom of choice, they also enjoy recommendation systems, subscription services, and marketplace encounters that seemingly occur by chance. This article proposes that enjoyment can, in some contexts, be higher than that in contexts involving choice. This occurs as a result of feelings of serendipity that arise when a marketplace encounter is positive, unexpected, and attributed to some degree of chance. A series of studies shows that feelings of serendipity positively influence an array of consumer outcomes, including satisfaction and enjoyment, perceptions of meaningfulness of an experience, likelihood of recommending a company, and likelihood of purchasing additional products from the company. The findings show that strategies based on serendipity are even more effective when consumers perceive that randomness played a role in how an encounter occurred, and not effective when the encounter is negative, the encounter occurs deterministically (i.e., planned by marketers to target consumers), and consumers perceive that they have enough knowledge to make their own choices. Altogether, this research suggests that marketers can influence customer satisfaction by structuring marketplace encounters to appear more serendipitous, as opposed to expected or entirely chosen by the consumer.
Keywords: choice; enjoyment; product recommendations; satisfaction; serendipity; subscription services
Consider two consumer encounters. In one encounter, a consumer is listening to a music streaming service and a song they love comes across the speakers. In a second encounter, the consumer is surfing TV channels on Friday night and arrives at channel 131 to find that The Dark Knight, their favorite movie, is about to begin. These encounters are extremely pleasant when they occur in a consumer's life and may be even more enjoyable than something the consumer personally picks (e.g., choosing a song to listen to or movie to watch). Such occurrences are also increasingly common in the marketplace. For example, subscription product delivery has proliferated in recent years and often involves receiving products (e.g., clothes, wine) periodically without prior knowledge of the items in the shipment. Other occurrences include recommendation systems that make selections for consumers (e.g., songs, videos) and attractions with unstructured experiences that are not previously defined (e.g., seeing works of art in a museum). These encounters may also occur in contexts where the consumer knows what to expect but something unexpected happens, such as a tasting of a consumer's favorite cheese in the grocery store precisely on the day the consumer decided to go shopping.
It is unclear, however, why such marketplace encounters are so enjoyable and whether marketers may be able to create, influence, and enhance these kinds of encounters. We contend that some encounters that do not involve deliberate choice are enjoyable because the way they happen generates feelings of serendipity. Serendipity in the marketplace is the set of feelings resulting from a product, service, or experience that is positive, unexpected, and attributed to some degree of chance. We posit that rather than being random encounters, marketers may have control over consumers' perceptions of how the encounter happened (e.g., "I chose it" vs. "There was some chance involved in how it happened"). By altering certain aspects of consumer encounters to generate feelings of serendipity, marketers can influence an array of outcomes, such as satisfaction, enjoyment, meaningfulness of the experience, and willingness to pay, which we collectively call "consumer outcomes" (see Figure 1). A series of studies in multiple domains (online subscription services, works of art, movies, food consumption, and music) tests the idea that feelings of serendipity positively influence such consumer-relevant outcomes.
Graph: Figure 1. The role of serendipity in the marketplace.
Drawing from research on consumer reactions to unexpected events ([17]), findings on how personal choice may not always generate the highest level of satisfaction ([ 2]), and qualitative work in the area of recommendation systems ([23]), this research contributes to marketing literature and practice in several ways. First, we provide evidence for the characteristics and consequences of serendipity, an underutilized construct in marketing theory and practice. There is research on how surprising consumers may generate positive or negative reactions, but surprise is only one property of encounters that generate feelings of serendipity. By incorporating the role of chance, we can make different recommendations that are not part of the surprise literature, such as how randomness may increase the enjoyment of experiences. Second, this research shows that freedom to choose, which consumers many times desire, does not always lead to the highest consumer satisfaction. This is because choice involves elaboration and the use of limited cognitive resources already taxed by vast amounts of daily information exposure. We show that the absence of choice may actually increase satisfaction with an encounter, and how different properties of such encounters influence satisfaction. Third, this research contributes to practice in several industries. Professionals who work in companies that use subscription services (e.g., Birchbox, Stitch Fix), product recommendations (e.g., Amazon, Spotify), and unstructured experiences (e.g., museums, amusement parks) can better understand why and when creating serendipitous encounters can bring more benefits than encounters that are expected or entirely chosen by the consumer. Feelings of serendipity are akin to the famous adage of being in the right place at the right time, once popularized by Humphrey Bogart in the classic film Casablanca: "Of all the gin joints in all the towns in all the world, she walks into mine." Our conceptualization and findings will provide marketers with insights on how to build some of this magic into marketplace encounters.
The eighteenth-century writer Horace Walpole originally coined the term "serendipity" to describe, in a Persian fairy tale, the idea of people "always making discoveries, by accidents and sagacity, of things they were not in quest of" ([47], p. 407). Today, serendipity is defined as "finding valuable or agreeable things not sought for" (https://www.merriam-webster.com/dictionary/serendipity), looking for something and finding something else that is actually more suitable to one's needs ([39]), and a positive and unexpected discovery ([19]). Despite these definitions, we lack an understanding of serendipity that applies more directly to marketing-relevant phenomena.
Serendipity in the marketplace refers to the set of feelings resulting from an encounter involving a chance finding of a product, service, or experience not directly chosen by the consumer. It happens when a consumer is not looking for anything specific or looking for something and discovers something else ([ 8]; [34]). Thus, serendipity is an unexpected event that occurs when the consumer is in either a passive state, not trying to discover anything, or an active state, trying to find something of value. As such, we propose that feelings of serendipity in the consumer domain result from an encounter that is ( 1) positive, ( 2) unexpected, and ( 3) involving some degree of chance ([32]; [33]; [34]).
Unexpectedness is the cognitive process responsible for the feeling of surprise ([41]), which can be a positive or negative emotional reaction ([11]; [36]). An unexpected surprise that is negative can enhance negative reactions, but when a surprise is positive (i.e., brings value to the consumer and generates positive emotions), it enhances satisfaction ([30]; [48]). For example, consumers experienced greater enjoyment from winning a smaller, unexpected amount of money compared with a larger, but expected, amount of money ([36]). In addition, surprise incentives (e.g., coupons) are viewed positively and lead to increased spending on unplanned purchases ([17]; [45]).
Although serendipitous events are unexpected, surprise is not the only component of serendipity. For an event to generate feelings of serendipity, it must be attributed to some chance or, in the case of chance that leads to positive experiences, luck. This occurs because one consequence of feeling surprised is the search for attribution ([41]; [44]). For example, you may be surprised to receive a free cup of coffee when entering your favorite coffee shop but then see a sign saying that they are giving away coffee to their loyalty club patrons. However, if there is no sign, you may attempt to infer what happened by making an attribution to chance ("They are randomly selecting people to receive a free coffee"), and the search for attribution ends ([12]; [41]). In this view, while receiving a cup of coffee with a clear attribution (you are a loyalty club member) is nice, perceiving that you were lucky to be selected may be more satisfying.
Thus, when marketers deliver a product, service, or experience in a way that is positive, unexpected, and involving chance, this will generate congruent feelings. Consumers will feel that the encounter was a good surprise, make attributions to chance, and feel lucky that it happened. We collectively call these "feelings of serendipity." These feelings, in turn, can influence an array of consumer-relevant outcomes, such as satisfaction with the entire experience with a company. In addition, feelings of serendipity could make such experiences feel more meaningful, as people attach more meaning to events that they perceive to occur by chance or luck. In fact, ascribing meaning to chance events is most prevalent for positive experiences ([22]; [24]; [42]). For example, a positive event attributed to chance may lead people to think about ways in which things could have happened less positively ([25]), making them think the event was "meant to be," and that "there is a reason why it occurred." If feelings of serendipity enhance satisfaction and meaning, serendipity should influence other outcomes, such as likelihood of recommending a product or service, willingness to pay, and willingness to buy additional products or services from a company.
Figure 1 presents a summary of the influence of serendipity on consumer-relevant outcomes. Serendipitous events happen without people choosing them or knowledge that they are going to happen. In our previous example, the person chose to go to the coffee shop that day but did not choose to get a free cup of coffee. A consumer chooses to listen to a music streaming service but does not choose the song that plays once they start listening. These examples raise the question of when serendipity occurs and when marketers can take advantage of it.
Serendipity occurs when, at the time of purchase or consumption, an encounter results in the feelings mentioned previously (i.e., a good surprise, luck, attributions to chance). In such contexts, the product, service, or experience may be judged as quite positive ([33]). Consider, for example, subscription services (e.g., Stitch Fix). The consumer chooses the company and whether they want to receive a box with clothes (or other products) chosen by the company on the basis of a profile they fill out. While the company has information about the consumer's preferences, it is never perfect information, and the consumer is not choosing which products to receive. We propose that a large part of the appeal of such subscription services is that there is an element of surprise and chance (i.e., there is randomness in the process). When the products received are good ("positive"), this generates feelings of serendipity, increasing satisfaction compared with when consumers choose their products.
A similar example is streaming and other recommendation systems, which are the focus of much of the literature on serendipity ([23]). These recommendations are based on consumers' preferences, but the recommendations are unexpected and there is a chance component to them. Thus, when the recommendation is good, the entire experience is more enjoyable because the context generates feelings of serendipity ([29]; [37]; [49]). If a consumer places their favorite songs on a playlist and chooses to play one of them, the listening experience will not generate feelings of serendipity, and enjoyment may not be as high. In support of these predictions, qualitative research on how consumers listen to music suggests that when consumers perceive that they encountered songs and information unexpectedly and by chance, they indicate that their listening experience was better ([ 6]; [29]). The implication is that consumers appreciate choosing ([ 3]; [43]), but serendipitously encountering a product can be more enjoyable as long as the product brings value to the consumer.
As an additional example, consider experiences such as going to a museum. If a consumer knows beforehand which work of art (e.g., a painting) they plan to see, this may be enjoyable but not serendipitous. Alternatively, if the museum places beautiful paintings in locations where consumers may find them by surprise (e.g., immediately upon turning a corner), the experience may become serendipitous and even more enjoyable. Finally, consider sampling at a supermarket. While the consumer has chosen to go grocery shopping, and may even expect that there will be some sampling opportunities, coming across a sampling of a favorite varietal of wine can generate feelings of serendipity, which may lead to heightened enjoyment of the wine and a decision to buy it. The relationship illustrated in these examples, depicted with solid lines in Figure 1, leads to our focal hypotheses:
- H1a: A marketplace encounter that is positive, is unexpected, and involves some degree of chance improves consumer outcomes compared with an encounter that the consumer directly chooses.
- H1b: The effect of a marketplace encounter that is positive, is unexpected, and involves some degree of chance on consumer outcomes is mediated by feelings of serendipity.
There are different ways in which marketers can attenuate or enhance the effects of serendipity (Figure 1). Marketers can manipulate variables that make the properties of serendipity more or less salient or variables that impact how desirable serendipity is. In terms of the properties of serendipity, the current research manipulates the valence of the encounter (i.e., positive vs. negative; Study 2) and how random consumers perceive the encounter to be (i.e., the encounter was the result of random events vs. planned by the marketer; Study 3). Manipulating valence and perceived randomness enables us to investigate the premise that, for an encounter to be serendipitous, it needs to be a good surprise and attributed to some degree of chance, respectively. In terms of how desirable serendipity is, we manipulate the amount of diagnostic information consumers receive about the product option, which can dampen the serendipity effect (Study 4). We show that sometimes all the properties of serendipity are present, but marketers should be careful not to provide information that can make serendipity undesirable, leading consumers to prefer making their own choice. Altogether, these variables directly address what makes serendipitous encounters so enjoyable—the properties must be present and serendipity must be desirable.
First, the encounter must be positive. While surprise can make an experience more positive due to its unexpected nature ([15]), it cannot make all experiences more positive ([27]). In fact, a surprise can amplify negative affect when the event does not have utility to the consumer or imposes a cost ([21]). This means that negative encounters are not serendipitous, even if the element of surprise is present. Thus, we predict that when an unexpected encounter is positive, it will generate feelings of serendipity and improve consumer outcomes compared with when the encounter is chosen by the consumer. When an unexpected encounter is negative, it will not generate feelings of serendipity and may diminish consumer outcomes compared with when the consumer chooses the encounter. Formally:
- H2: Feelings of serendipity and improved consumer outcomes occur when an encounter is positive (i.e., brings value to the consumer), but not when it is negative.
Second, influencing the perceived amount of chance involved in an encounter should result in different perceptions of serendipity and outcomes. This can be done in different ways. For example, consider a consumer who receives a song recommendation. In one approach, the song is described as having been randomly selected from a playlist of 100 great songs, and in another situation, the same playlist is described as having 10 songs. Assuming the song is good, the consumer should have more intense feelings of serendipity when it came from the playlist featuring 100 songs, as there was a lower probability that this specific song would have been selected (the consumer feels "luckier"). In a second approach, consumers know that a marketer is responsible for the product, service, or experience (i.e., most marketplace encounters do not occur completely by chance). However, there can still be a surprise and a chance component to the encounter, as the marketer does not have perfect information about the consumer's preferences. This implies that the more salient it is that the marketer played a role (e.g., "We chose this carefully to match your preferences"), the less attribution to chance there will be. Thus, we predict that the more consumers perceive that an encounter was the result of randomness (vs. selected deterministically), the more feelings of serendipity it will generate, and the more satisfying it will be. Alternatively, if it is salient that an encounter was planned by a marketer to target consumers, the encounter will not generate feelings of serendipity and will not be as satisfying compared with when the presence of the marketer is not salient. We formally hypothesize:
- H3: An increase in the perceived amount of randomness involved in an encounter increases feelings of serendipity and improves consumer outcomes, whereas the perception that an encounter was selected deterministically diminishes feelings of serendipity and consumer outcomes.
Third, sometimes an encounter successfully generates feelings of serendipity, but the effect of having these feelings on satisfaction is moderated by whether serendipity is desirable or not. Consider services that offer recommendations (e.g., music), which can be successful if their recommendations generate feelings of serendipity. This success may depend on consumers' perception that they have enough knowledge to make their own choices. We predict that when consumers receive a high amount of diagnostic information about a recommendation service and the products it offers, consumers will not desire serendipity and will be more satisfied when they make their own choices. When information is diagnostic, it is directly relevant to choice, as it can inform which option(s) is (are) superior to other options in a choice set ([13]). When consumers perceive that they have enough relevant information to make an informed choice themselves, feelings of serendipity should not translate into increased satisfaction, as the consumer should believe that they could have made a better choice based on the knowledge they have. This prediction is consistent with research demonstrating that choice can be quite desirable ([ 3]; [43]), especially when consumers have enough information to make a satisfying choice ([ 2]). Thus,
- H4: Providing a high amount of diagnostic information to consumers makes serendipity less desirable, diminishing consumer outcomes compared with when consumers make their own choices.
Study 1 examines real purchase experiences. We identified four subscription service companies where consumers have the option to choose the products themselves or have the products selected for them. This enabled us to understand the role of serendipity using the natural dichotomy that occurs in subscription services. To do so, we asked participants to describe their experiences with the companies and indicate their satisfaction with the products they purchased, meaningfulness of the consumption experience, willingness to recommend the service, and willingness to extend the subscription. We also measured feelings of serendipity. Consistent with H1a and H1b, we expected that participants who had the products selected for them (vs. chose the products themselves) would be more satisfied with the products and that feelings of serendipity would drive this effect.
We recruited 829 participants from Amazon's Mechanical Turk (MTurk) and paid them a small monetary compensation. After we eliminated 18 outliers on the basis of the overall time spent responding to the survey questions (+3 SDs from the mean; [35]), the final sample size was 811 (43.3% men; age range: 18–71 years, M = 34.90 years, SD = 9.96 years). We used the same exclusion criterion for all studies but also report the main results without employing exclusions in the Web Appendix. The results of all studies are virtually unchanged without excluding participants. This study had a 2 (condition: personal choice vs. serendipity) × 4 (company replicate) between-subjects design.
Participants were told that the survey was about their consumption experiences from specific companies. Participation in the study was contingent on whether the participant indicated that over the last month they had an experience with one of the companies we selected for the study (Birchbox [www.birchbox.com], Stitch Fix [www.stitchfix.com], The Tie Bar [www.thetiebar.com], and FabFitFun [www.fabfitfun.com]). Participants were randomly assigned to one of two conditions. Participants in the serendipity condition were told, "Please examine the companies listed below and indicate a company that recently (within the past month) selected products for you and sent them to you as a box you received in the mail." Participants in the personal choice condition were told, "Please examine the companies listed below and indicate a company where you recently (within the past month) selected products from and received your purchase in the mail." In both conditions, if a participant indicated they had not received products from any of the companies listed, they were redirected out of the survey. We programmed the survey with fixed quotas per condition per company used. This enabled us to collect a similar number of participants per condition per company. Once we achieved a certain number of participants for one company (e.g., 200 Birchbox participants with 100 in the serendipity and 100 in the personal choice condition), the survey automatically hid Birchbox from the company selection list.
After indicating a company, participants were told that we were interested in how people process moments of their life and that we would like to know about the recent consumption experience in more detail: "Please think about the time when you recently received a package from [company name was inserted here]. Take a minute to remember what it felt like to receive the products and then describe the products and how you felt when you opened the box." After this writing task, participants were asked, "How satisfied are you with the products you received?" (1 = "not at all satisfied," and 7 = "very satisfied"). Participants also responded to four items measuring feelings of serendipity: "I feel that the products I received from the company were a good surprise," "I feel lucky to have come across these products," "I feel that these products were an unexpected discovery," and "I feel that there was some element of chance involved in having received these exact products" (1 = "strongly disagree," and 7 = "strongly agree"). We combined the items to form a serendipity index (α =.83). Participants also responded to questions about the meaningfulness of the experience: "The experience with the products I received was meaningful," "The experience with the products was more meaningful than regular consumption experiences," "I felt that the fact that I got these products was "meant to be," and "These products are meaningful to me" (1 = "strongly disagree," and 7 = "strongly agree"). We combined the items to form a meaningfulness index (α =.91). We also asked, "How likely are you to purchase an additional 6-month subscription from [company name]?," and "How likely are you to recommend [company name] subscription service to a friend?" (1 = "not likely at all," and 7 = "very likely").
Participants then responded to two measures related to expectations: "How high were your expectations about the products before you got them?" (1 = "very low," and 7 = "very high") and "How satisfied did you expect to be with the products before you got them?" (1 = "not at all," and 7 = "very satisfied"). We combined these items to form an expectation index (r =.63). This measure allowed us to rule out the alternative explanation that serendipity leads to positive consumer outcomes due to higher expectations when consumers order their own products. We also measured a series of control items designed to check for the robustness of the effects and test the influence of alternative factors on the results. The items measured product cost, product type, perceived quality, number of products, shipping period, time from the purchase, general attitudes toward the company, and satisfaction with the purchase process. The Web Appendix presents the complete procedure and materials of this (Web Appendix A) and all subsequent studies.
A 2 (condition) × 4 (company) analysis of variance (ANOVA) revealed a main effect of condition, such that participants reported greater satisfaction in the serendipity (M = 6.01, SD = 1.10) than in the personal choice condition (M = 5.55, SD = 1.50; F( 1, 803) = 25.58, p <.001). There was no interaction between condition and the company replicates (F( 3, 803) =.12, p =.946), and the effect of condition was significant for each company. There was a main effect of the company replicate (F( 3, 803) = 3.55, p =.014). Because this main effect does not change the interpretation of the results, we report additional details, along with the results for each company, in Web Appendix A. In all studies reported in the main text, there were no interactions with the replicates we used (different companies, paintings, videos, and songs), and therefore we collapsed across replicates for all analyses. We present the analysis of replicates of each study in the Web Appendix. We also tested whether the effect would hold when we included each control variable we measured in the analysis for each outcome. None of the covariates changed the results, which indicates that they cannot explain the influence of condition on satisfaction (see Web Appendix A).
A 2 (condition) × 4 (company) ANOVA revealed a main effect of condition, such that participants reported higher willingness to recommend in the serendipity condition (M = 5.88, SD = 1.29) than in the personal choice condition (M = 5.25, SD = 1.73; F( 1, 803) = 34.30, p <.001).
A 2 (condition) × 4 (company) ANOVA revealed a main effect of condition, such that participants reported higher willingness to extend the subscription in the serendipity condition (M = 5.45, SD = 1.60) than in the personal choice condition (M = 4.83, SD = 1.89; F( 1, 803) = 25.82, p <.001).
A 2 (condition) × 4 (company) ANOVA revealed a main effect of condition, such that participants reported higher meaningfulness in the serendipity condition (M = 5.12, SD = 1.34) than in the personal choice condition (M = 4.72, SD = 1.48; F( 1, 803) = 16.48, p <.001).[ 6]
A 2 (condition) × 4 (company) ANOVA revealed a main effect of condition, such that participants reported greater feelings of serendipity in the serendipity condition (M = 5.50, SD = 1.02) than in the personal choice condition (M = 5.07, SD = 1.36; F( 1, 803) = 25.89, p <.001). We conducted bootstrapping mediation analyses (PROCESS Model 4; [16]) using condition (serendipity vs. personal choice) as the independent variable and serendipity as the mediator for each of the outcomes.
The indirect effect of serendipity was significant for satisfaction (β =.32, SE =.07, 95% confidence interval [CI]: [.19,.47]), meaningfulness (β =.35, SE =.07, 95% CI: [.21,.50]), willingness to recommend (β =.37, SE =.08, 95% CI: [.22,.52]), and willingness to extend the subscription (β =.36, SE =.07, 95% CI: [.22,.50]).
A 2 (condition) × 4 (company replicate) ANOVA did not reveal main effects of condition (Mserendipity = 5.34, SD = 1.10; Mpersonalchoice = 5.27, SD = 1.05; F( 1, 803) =.91, p =.340), showing that the findings cannot be explained by consumers creating higher expectations when they choose the products they will receive.
Using retrospective reports from real purchase experiences, Study 1 found that having products sent by a subscription service without knowing what the specific products are leads to more positive consumer responses than personally choosing products. This effect, which supports H1a and H1b, was due to feelings of serendipity, and not to an increase in expectations about the products. Further, the effect held while controlling for an array of factors that may influence consumer responses (i.e., product cost, product type, perceived quality, number of products, shipping period, time from the purchase, general attitudes toward the company, and satisfaction with the purchase process), meaning that the effect of serendipity is robust and goes beyond any possible effect of such factors.
We have theorized that the experience needs to be positive for serendipity to occur. In Study 2, we examine the role of valence in the context of seeing a painting in a museum. Participants either chose which painting they wanted to see or had a painting randomly chosen for them. We also had a baseline condition in which participants simply saw a painting. The baseline condition allowed us to determine whether the focal effect occurs because of a positive effect in the serendipity condition or a negative effect in the choice condition. Consistent with H2, we predicted that having a painting randomly chosen would increase enjoyment relative to a baseline and a personal choice condition when the painting was attractive (i.e., positive valence), but not when it was unattractive (i.e., negative valence). This context is relevant to practice, as serendipity can also occur when consumers have unstructured experiences such as those that occur in a museum or an amusement park. Thus, findings in this context may help managers configure these experiences in a way that generates feelings of serendipity.
We recruited 462 participants from MTurk and paid them a small monetary compensation. After we eliminated 15 outliers on the basis of the time spent responding to the survey questions (see criterion in Study 1), the final sample size was 447 (46% men; age range: 18–89 years, M = 39.33 years, SD = 13.69 years). This study had a 3 (condition: baseline, personal choice, serendipity) × 2 (valence: positive vs. negative) × 2 (painting replicate) between-subjects design.
Participants were told that we were interested in how people respond in different situations. Participants in the personal choice condition were asked to "Imagine you enter an art gallery. Two of the paintings the gallery features appear below." Participants then saw the titles of two paintings by Gerald Chodak: Moving Around and On the Border. The titles were the same in the positive and negative valence condition, but the paintings were either attractive or unattractive, depending on the condition. The order of presentation of the titles on the screen was counterbalanced. Participants were then asked to "select one of these two paintings to view" and clicked on a continue button to proceed. Once participants proceeded to the next page, they saw the painting they chose (for all the paintings as well as the results of a pretest showing that the positive paintings were indeed perceived as more positive, see Web Appendix B). Participants in the serendipity condition went through the same procedure, but instead of being asked to select one of the paintings when they saw the information about the paintings, they were simply asked to click on a continue button to proceed. Once they proceeded, they were told, "Imagine that you walk down a hallway in the art gallery and turn a corner. Just as you turn the corner, you happen to find this painting on the wall," and one of two paintings, selected randomly, was presented. Participants in the baseline condition were told to "Imagine you enter an art gallery. You will see and rate a painting on the next page." Once participants proceeded to the next page, one of the two paintings was randomly presented to them.
After viewing the painting, participants were asked, "How much did you enjoy the painting?" (0 = "I hated it," and 100 = "I loved it"). Participants also responded to questions about their feelings of serendipity (α =.78): "Getting to see the painting I just saw was a good surprise," "I came across this painting by luck," and "This painting was an unexpected discovery" (1 = "strongly disagree," and 7 = "strongly agree").
We also measured alternative explanations. Participants reported their attachment to the alternative option: "To what extent did you feel attached to the other option?" (1 = "not at all attached," and 7 = "very much attached"). This question examined whether participants enjoyed the chosen option less in the positive valence condition because they were attached to the option they did not choose ([ 5]). Participants also answered a question about regret: "To what extent did you feel regretful about the painting you saw?" (1 = "not at all regretful," and 7 = "very much regretful"). Moreover, they indicated "How much did you scrutinize the painting?" (1 = "not at all," and 7 = "very much"), allowing us to verify whether participants scrutinized the paintings to different degrees across conditions. Participants then answered questions about stress and frustration: "How stressed were you with the painting selection process?" (1 = "not at all stressful," and 7 = "very much stressful"), and "How frustrated were you with the painting selection process?" (1 = "not at all frustrated," and 7 = "very much frustrated"), allowing us to verify whether choosing versus not choosing, and seeing a negative versus a positive painting, generated negative feelings that could explain the results.
A 3 (condition) × 2 (valence) ANOVA revealed a main effect of valence (F( 1, 441) = 32.21, p <.001), such that participants enjoyed the positive paintings (M = 63.42, SD = 26.12) more than the negative paintings (M = 47.68, SD = 33.16). There was no main effect of condition (F( 2, 441) =.209, p =.812). The interaction was significant (F( 2, 441) = 3.92, p =.021; see Figure 2). When participants saw a positive painting, there was a marginally significant effect of condition (F( 2, 441) = 2.83, p =.060). Participants reported higher enjoyment in the serendipity condition (M = 70.08, SD = 23.76) than in the personal choice (M = 59.97, SD = 24.49; F( 1, 441) = 4.32, p =.038) and baseline (M = 60.17, SD = 28.85; F( 1, 441) = 4.17, p =.042) conditions. There was no difference between the baseline and the personal choice conditions (F < 1). When participants saw a negative painting, there was no effect of condition (F( 2, 441) = 1.35, p =.261). We did find that enjoyment was lower in the serendipity condition (M = 42.88, SD = 32.22) than in the baseline condition (M = 50.90, SD = 31.36; F( 1, 441) = 2.59, p =.108). While this difference is not statistically significant, the pattern of results suggests that serendipity can be potentially harmful for negative experiences. None of the results for the measured alternative explanations could explain the effects on enjoyment (for detailed analyses, see Web Appendix B).
Graph: Figure 2. Study 2 results.*p <.05.Notes: Error bars = ±1 SEs. Unbracketed comparisons are not significantly different from each other.
A 3 (condition) × 2 (valence) ANOVA revealed an effect of valence (F( 1, 441) = 6.51, p =.011), such that participants reported greater feelings of serendipity when the painting was positive (M = 4.53, SD = 1.41) than when it was negative (M = 4.17, SD = 1.55). There was also an effect of condition (F( 2, 441) = 5.30, p =.005), such that serendipity was higher in the serendipity condition (M = 4.65, SD = 1.37) than in the baseline condition (M = 4.07, SD = 1.54; F( 1, 441 = 10.60, _I_p_i_ =.001). We observed no difference between the serendipity and personal choice (M = 4.34, SD = 1.51) conditions (F( 1, 441) = 2.69, p =.102). In addition, serendipity was marginally higher in the personal choice than in the baseline condition (F( 1, 441) = 2.75, p =.098). These effects were qualified by a marginally significant interaction (F( 2, 441) = 2.39, p =.093). When participants saw a positive painting, there was an effect of condition (F( 2, 441) = 7.19, p =.001). Serendipity was higher in the serendipity condition (M = 5.02, SD = 1.33) than in the personal choice (M = 4.42, SD = 1.29; F( 1, 441) = 6.31, p =.012) and baseline (M = 4.13, SD = 1.48; F( 1, 441) = 13.80, p <.001) conditions, which did not differ from each other (F( 1, 441) = 1.42, p =.235). When participants saw a negative painting, there was no effect of condition (F < 1).
We conducted a bootstrapping analysis for moderated mediation using the three conditions (baseline, personal choice, and serendipity) as multicategorical independent variables, valence as the moderator, feelings of serendipity as the mediator, and enjoyment as the dependent variable ([16]; PROCESS Model 8). When participants saw a positive painting, the pathway to enjoyment through feelings of serendipity was significant when comparing the serendipity condition with the personal choice (β = 8.85, SE = 3.06, 95% CI: [2.77, 14.87]) and baseline (β = 13.05, SE = 3.36, 95% CI: [6.46, 19.71]) conditions. However, when participants saw a negative painting, the pathway to enjoyment through feelings of serendipity was not significant when comparing serendipity with both the personal choice (β =.62, SE = 3.62, 95% CI: [−7.77, 6.63]) and baseline (β = 3.35, SE = 3.56, 95% CI: [−3.60, 10.47]) conditions.
Study 2 replicates the positive effect of serendipity, showing that the effect is evident not only in online subscription contexts but also in experiences such as those that occur when consumers visit a museum (i.e., art consumption). Importantly, this effect only happens when the experience is positive, which supports H2. The effect was due to feelings of serendipity, and it could not be explained by feelings of attachment to the alternative option or negative feelings during the painting selection process.
Study 3 investigates the role of attributions to chance in determining feelings of serendipity and its outcomes (H3). We propose that when it is salient that a marketer carefully planned a product encounter, there will not be an attribution to chance, and feelings of serendipity will not arise. To test this proposition, we simulated a movie trailer recommendation platform ("Movie Trailer Zone") that ostensibly learned about a consumer's movie preferences to build a profile. Once this profile was built, the platform recommended a movie trailer fitting the consumer's preferences, helping them decide which movie to watch. It was important to investigate serendipity in this context, as recommendation services depend heavily on how satisfied consumers are with products that are chosen for them.
We designed the conditions to generate a high versus low attribution to chance by manipulating the degree of randomness in the ( 1) initial movie selection, by varying whether the movie was selected from a pool of 100 versus 10 curated movie trailers, respectively, and ( 2) final movie selection, by varying whether the movie selection was described as randomly drawn from the movie pool versus carefully selected by a marketer out of the movie pool, respectively. We predicted that when the final movie selection was a random process, an increased amount of randomness (making a selection from 100 vs. 10 options) would increase feelings of serendipity and enjoyment. Alternatively, when the final movie selection was described as being made by a marketer, there would be less attribution to chance (i.e., the selection would be deemed determinist), which would attenuate feelings of serendipity and enjoyment.
We recruited 400 participants from MTurk and paid them a small monetary compensation. After we eliminated 11 outliers on the basis of the criterion outlined in Study 1, the final sample size was 389 (59.6% men; age range: 18–78 years, M = 38.11 years, SD = 11.56 years). This study had a 2 (degree of randomness in the initial selection: high vs. low) × 2 (degree of randomness in the final selection: high vs. low) × 5 (movie trailer replicate) between-subjects design. Participants did not make a personal choice in any of the conditions.
Participants were told that we were interested in consumers' responses to a recently launched platform ("Movie Trailer Zone") that allows members to receive curated movie trailer recommendations, and that they would perform a task that resembled how the platform was used. The procedure was designed to mimic common user experiences on streaming platforms such as Hulu or Netflix. Participants initially saw a short description of the platform, asking them to create an account. To increase immersion, we asked participants to provide a username for their profile and proceed to the profile building task. Once participants proceeded, we showed the username they had chosen and a list with 50 movie posters, asking them to choose 5 movies they liked. We indicated that this would help us (i.e., the platform) find trailers for the movies we thought they might like. As in Hulu or Netflix, participants could choose more than 5 movies if they wanted to. Once participants clicked next, they saw a loading spinner icon so that they would think there was a platform building a customized profile in the background. Participants were told that we were examining their preferences in the background, and that this was a necessary step to make a movie trailer recommendation. We asked them a few filler questions and, once they finished, they saw another loading spinner icon with a "Preparing your recommendation..." message. The page advanced after ten seconds, and the screen showed a recommendation. At this point, we administered different manipulations.
Participants read that, based on their profile, we (i.e., the platform) had curated a selection of movie trailers matching their preferences and that we selected 1 movie trailer out of this curated selection of 100 (randomness in the initial selection: high) or 10 (randomness in the initial selection: low) movie trailers. Here, we also manipulated the perceived degree of randomness in the final selection. Participants in the "high degree of randomness in the final selection" condition read, "We have examined your preferences with the help of our system and have randomly selected one movie trailer that you may enjoy." Participants in the "low degree of randomness in the final selection" condition read, "We have examined your preferences with the help of our system and have carefully selected one movie trailer that you may enjoy." Thus, it was less likely that participants would make attributions to chance in the latter condition, as the instructions made it salient that a marketer planned the experience to target them (i.e., the final selection was deterministic). In addition to the recommendation message, participants saw 60 seconds of the selected movie trailer. Participants saw one of five randomly selected trailers (the movies were Chronicle, Last Stand, Lawless, Priceless, and Wildlife), which we used to ensure that the effect was robust across different content. We used these movies on the basis of a pretest that showed a similar baseline level of enjoyment across them (see Web Appendix C).
After viewing the movie trailer, we asked, "How much did you enjoy the movie trailer?" (0 = "I hated it," and 100 = "I loved it"). We also asked participants whether they wanted to sign up to receive more information about the movie trailer recommendation platform (yes/no; if yes, they had to provide an email address). To ensure that we could assess actual interest in the platform, we used an online email validation tool to clean invalid or nonexistent email addresses. Thus, we had two indicators of interest: answering yes vs. no, and the presence (vs. not) of a valid email.
Participants also responded to items assessing feelings of serendipity (α =.79): "Getting to watch this movie trailer was a good surprise," "The movie trailer was an unexpected discovery," "I came across this movie trailer by luck," and "Based on how the service works, there was a low chance that I would be watching the specific movie trailer that was selected for me." As a manipulation check, we measured "The movie trailer was selected through a random process" (1 = "strongly disagree," and 7 = "strongly agree"). Finally, we measured the following alternative explanations: regret, stress, frustration, and scrutinizing.
A 2 (initial selection randomness) × 2 (final selection randomness) ANOVA revealed a main effect of final selection randomness (F( 1, 385) = 32.51; p <.001), such that participants perceived the movie selection process as more random when the movie was randomly (M = 4.51, SD = 1.85) rather than deterministically (M = 3.42, SD = 1.98) selected. With regard to randomness in the initial selection, there was no main effect (F < 1). This was expected, given that there should be no effect of initial selection randomness within the low-final-selection-randomness condition (F < 1). Thus, the key contrast was the effect of initial selection randomness within the high-final-selection-randomness condition. In this condition, participants perceived the selection process as marginally more random when the degree of randomness in the initial selection was high (M = 4.77, SD = 1.71) than when it was low (M = 4.28, SD = 1.94; F( 1, 385) = 3.40, p =.066).
A 2 (initial selection randomness) × 2 (final selection randomness) ANOVA on enjoyment revealed a main effect of randomness in the final selection (F( 1, 385) = 40.08, p <.001), such that enjoyment was higher when the movie was randomly (M = 73.45, SD = 22.10) rather than deterministically (M = 56.63, SD = 30.76) selected. There was no main effect of randomness in the initial selection (F( 1, 385) = 2.50, p =.114). The interaction was significant (F( 1, 385) = 4.05, p =.045; see Figure 3). When there was a high degree of randomness in the final selection, enjoyment was greater in the high- (M = 78.41, SD = 19.28) than in the low- (M = 68.77, SD = 23.61; F( 1, 385) = 6.80, p =.009) randomness-in-the-initial-selection condition. When there was a low degree of randomness in the final selection, there was no difference between the high- (M = 56.03, SD = 31.61) and low- (M = 57.19, SD = 30.11) randomness-in-the-initial-selection conditions (F < 1). None of the results for the measured alternative explanations could explain the pattern of results on enjoyment and the following dependent variables (for detailed analyses, see Web Appendix C).
Graph: Figure 3. Study 3 results.*p <.01.Notes: Error bars = ±1 SEs. Unbracketed comparisons are not significantly different from each other.
A 2 (initial selection randomness) × 2 (final selection randomness) logistic regression revealed a significant interaction between initial and final selection randomness (β = 1.28, Wald = 5.13, p =.024), such that participants in the high-randomness-in-the-final-selection condition were more likely to provide their email for signup when randomness in initial selection was high (P = 34.3%) than when it was low (P = 15.2%; Wald = 10.05, p =.002). For participants in the low-randomness-in-the-final-selection condition, there was no difference between the high- (P = 11.2%) and low- (vs. P = 13.5%; Wald =.23, p =.64) initial-randomness conditions.
While a 2 (initial selection randomness) × 2 (final selection randomness) logistic regression did not reveal a significant interaction between initial and final selection randomness (β = 1.05, Wald = 2.59, p =.108), the results were consistent with our predictions. Participants in the high-randomness-in-the-final-selection condition were more likely to provide their email for signup when randomness in initial selection was high (P = 21.2%) than when it was low (P = 11.4%; Wald = 3.60, p =.058). For participants in the low-randomness-in-the-final-selection condition, there was no difference between the high- (P = 7.9%) and low- (vs. P = 10.4%; Wald =.36, p =.548) initial-randomness conditions.
A 2 (initial selection randomness) × 2 (final selection randomness) ANOVA revealed a main effect of randomness in the final selection (F( 1, 385) = 64.90, p <.001), such that participants reported greater feelings of serendipity when the movie was randomly (M = 4.99, SD = 1.14) rather than deterministically (M = 3.89, SD = 1.58) selected. There was also an effect of degree of randomness in the initial selection (F( 1, 385) = 4.75, p =.030), such that participants reported greater feelings of serendipity when the movie trailer was selected out of a pool of 100 movies (M = 4.63, SD = 1.53) than when it was selected out of a pool of 10 movies (M = 4.31, SD = 1.39). The interaction was significant (F( 1, 385) = 4.74, p =.030). When there was a high degree of randomness in the final selection, participants reported greater feelings of serendipity in the high- (M = 5.30, SD = 1.09) than in the low- (M = 4.70, SD = 1.11; F( 1, 385) = 9.98, p =.002) randomness-in-the-initial-selection condition. When there was a low degree of randomness in the final selection, there was no difference between the high- (M = 3.89, SD = 1.62) and low- (M = 3.89, SD = 1.55) randomness-in-the-initial-selection conditions (F < 1).
We conducted a bootstrapping moderated mediation analysis using the degree of randomness in the initial selection as the independent variable, degree of randomness in the final selection as the moderator, and serendipity as the mediator for each of the outcomes we measured (PROCESS Model 8; [16]). For the enjoyment measure, the index of moderated mediation was significant (index = 8.23; 95% CI: [.69, 16.08]). When there was a high degree of randomness in the final selection, the pathway to enjoyment through feelings of serendipity was significant (β = 8.23, SE = 2.17, 95% CI: [4.07, 12.57]). When there was a low degree of randomness in the final selection, the pathway to enjoyment through feelings of serendipity was not significant (β =.00, SE = 3.23, 95% CI: [−6.34, 6.33]). A similar pattern emerged for both measures of interest in the platform (see Web Appendix C).
Study 3 supports H3 and makes important contributions to the understanding of serendipity. First, feelings of serendipity only occur when the product encounter does not involve highly deterministic components. When it is salient to the consumer that the marketer controlled the ultimate selection of the experience, the experience cannot be attributed to chance, making it less enjoyable. This means that as long as the presence of the marketer (or other nonrandom component) is not made salient, consumers may attribute the selection to chance, increasing serendipity. In addition, increasing the perceived amount of randomness involved in the initial selection of a product experience had a positive effect, which provides theoretical insight about what makes an experience serendipitous and offers marketers another tool to increase serendipity, enjoyment, and interest. To provide further theoretical and practical insight, Web Appendix C–1 presents an additional study examining the role of chance in encounters involving serendipity in the food domain.
Study 4 examined information as a moderator of the effect of serendipity on consumer satisfaction. We predicted that feelings of serendipity would not translate to higher satisfaction when consumers are presented with enough diagnostic information that makes them perceive they have the knowledge to make their own choices (H4). To test this prediction, we investigated the context of a recommendation service, similar to Study 2, but this time used an existing company that provides a more functional service. We introduced consumers to a service called Brain.fm, which features functional music that can enhance focus. Functional music is used for many specific purposes, including concentration, relaxation, and meditation. We presented consumers with information about what improves a song's ability to increase concentration. In one condition, this information was nondiagnostic to whether a song is functionally effective, whereas in the other condition it was diagnostic. In the nondiagnostic information condition, consumers should be more satisfied with a song when the encounter occurred serendipitously than when they made their own choice, and feelings of serendipity should predict satisfaction. Alternatively, in the diagnostic information condition, they should be more satisfied with a song when they made their own choice than when the encounter occurred serendipitously. In this condition, even when feelings of serendipity are high, this should not translate to satisfaction, and satisfaction should be driven by consumers' perceived knowledge to make their own decisions in the product category.
We recruited 400 participants from MTurk and paid them a small monetary compensation. After we eliminated 7 outliers on the basis of the criterion outlined in Study 1, the final sample size was 393 (55.2% men; 19–79 years, M = 40.66, SD = 13.56). This study had a 2 (condition: personal choice vs. serendipity) × 2 (information: nondiagnostic vs. diagnostic) × 5 (song replicate) between-subjects design.
Participants were told that they would complete a study about Brain.fm, a functional music platform backed by scientific research to help listeners focus and concentrate. Then, participants read a section titled "What is functional music, anyway?" and were informed that Brain.fm develops music to improve concentration. Participants were also told that functional music involves tempo, pitch, neural phase-locking value, induced brain wave, three-dimensional externalized sound, and brain modulation rate. We included this information so participants could understand whether the information they were about to receive was diagnostic or not.
In the nondiagnostic information condition, participants read about three attributes of functional music that were not relevant to the superiority of one song over others (e.g., "initial composition—humans compose the musical content"). In the diagnostic information condition, participants read about three attributes that were relevant to the superiority of one song over others (e.g., "neural phase-locking value—refers to the extent to which populations of neurons engage in various kinds of coordinated activity") and saw the range of values that would make a song highly functional. Key to the manipulation, all participants then saw the title, neural phase-locking value, induced brain wave, and brain modulation rate of five songs, the latter three pieces of information being the three attributes participants in the diagnostic condition had just learned about. Thus, each song had three attributes, and participants in the diagnostic (nondiagnostic) condition had just been presented (not presented) with information about which attribute values made a song highly functional.
Participants in the personal choice condition were asked to "choose one of the five songs available for a listening sample," whereas participants in the serendipity condition were told that we would randomly select one song for them to listen to on the next page. Once participants proceeded to the next page, the song started playing. We fixed the listening page to auto-advance after 60 seconds, so every participant would listen to the same amount of music. To increase immersion and realism, we used existing functional music and imagery associated with the Brain.fm service throughout the survey.
After participants listened to the song, we asked, "How satisfied are you with the song you just listened to?" and "How satisfied are you with the song listening experience in general?" (1 = "not at all satisfied," and 7 = "very satisfied"), which formed a satisfaction index (r =.85). We also asked about interest in the platform ("How interested would you be in subscribing to Brain.fm's platform?" [1 = "not at all," and 7 = "very interested"]) and willingness to recommend ("How likely are you to recommend Brain.fm's subscription service to a friend?" [1 = "not likely at all," and 7 = "very likely"]). In addition, we assessed willingness to pay: "Brain.fm has several subscription options, and such as other platforms (e.g., Spotify, Apple Music), the monthly plan costs between $5 and $15. How much are you willing to pay for a one-month subscription to Brain.fm?," with a slider scale ranging from $5 to $15.
Participants then responded to a serendipity measure. We told them to consider their experience with Brain.fm and how the platform works, and asked, "Getting to experience this one song I just listened to ended up being a good surprise," "Considering the song selection process, I feel lucky to have come across the song I listened to," "From what it could have been, I feel that the song I listened to was an unexpected discovery," and "I feel that there was some element of chance involved in having experienced this specific song I just listened to" (1 = "strongly disagree," and 7 = "strongly agree"). We combined the items to form a serendipity index (α =.89). We also measured participants' perceived knowledge to make a choice using four items (e.g., "From the information provided about functional music, I was knowledgeable enough to choose a song to listen to"), and verified whether perceived knowledge mediated the results when diagnostic information was presented, which it did. These items and their analyses are presented in Web Appendix D. Finally, we measured regret, scrutinizing, and expectations.
Finally, we asked two manipulation check questions: "To which extent did you make your own song choice?" (1 = "not at all," and 7 = "very much") and "How much information was provided about what attributes are necessary for a good functional song?" (1 = "not much," and 7 = "very much").
Participants in the personal choice condition indicated making their own choice (M = 5.51, SD = 1.45) to a greater extent than those in the serendipity condition (M = 1.70, SD = 1.33; F( 1, 389) = 735.08, p <.001). There was no interaction between condition and information (F < 1). Participants in the diagnostic information condition indicated that information about what attributes are necessary for a good functional song were provided to a greater extent (M = 5.49, SD = 1.37) than in the nondiagnostic information condition (M = 4.20, SD = 1.74; F( 1, 389) = 66.34, p <.001).
A 2 (condition) × 2 (information) ANOVA revealed a main effect of information, such that satisfaction was higher when the information was nondiagnostic (M = 5.18, SD = 1.62) rather than diagnostic (M = 4.65, SD = 1.74; F( 1, 389) = 10.23, p =.001). There was no main effect of condition (F( 1, 389) =.028, p =.868). The interaction was significant (F( 1, 389) = 23.01, p <.001; see Figure 4). When the information was nondiagnostic, satisfaction was greater in the serendipity (M = 5.57, SD = 1.29) than in the personal choice condition (M = 4.80, SD = 1.82; F( 1, 389) = 10.64, p =.001). When the information was diagnostic, satisfaction was greater in the personal choice (M = 5.07, SD = 1.61) than in the serendipity condition (M = 4.25, SD = 1.77; F( 1, 389) = 12.41, p <.001). None of the results for the measured alternative explanations could explain the pattern of results on this and the other dependent variables (for detailed analyses, see Web Appendix D).
Graph: Figure 4. Study 4 results.*p <.01.Notes: Error bars = ±1 SEs.
A 2 (condition) × 2 (information) ANOVA on interest did not reveal main effects of condition (F( 1, 389) =.028, p =.867) or information (F( 1, 389) =.110, p =.741). The interaction was significant (F( 1, 389) = 22.56, p <.001). When the information was nondiagnostic, interest was greater in the serendipity condition (M = 4.63, SD = 1.80) than in the personal choice condition (M = 3.65, SD = 2.01; F( 1, 389) = 12.00, p =.001). When the information was diagnostic, interest was greater in the personal choice (M = 4.53, SD = 2.02) than in the serendipity condition (M = 3.61, SD = 2.05; F( 1, 389) = 10.58, p =.001).
A 2 (condition) × 2 (information) ANOVA on willingness to recommend did not reveal main effects of condition (F( 1, 389) =.026, p =.872) or information (F( 1, 389) =.651, p =.420). The interaction was significant (F( 1, 389) = 34.37, p <.001). When the information was nondiagnostic, willingness to recommend was greater in the serendipity condition (M = 4.65, SD = 1.77) than in the personal choice condition (M = 3.59, SD = 1.86; F( 1, 389) = 16.13, p <.001). When the information was diagnostic, willingness to recommend was greater in the personal choice condition (M = 4.53, SD = 1.86) than in the serendipity condition (M = 3.41, SD = 1.87; F( 1, 389) = 18.28, p <.001).
A 2 (condition) × 2 (information) ANOVA on willingness to recommend did not reveal main effects of condition (F( 1, 389) =.001, p =.982) or information (F( 1, 389) =.002, p =.968). The interaction was significant (F( 1, 389) = 8.74, p =.003). When the information was nondiagnostic, willingness to pay was greater in the serendipity condition (M = 8.42, SD = 3.45) than in the personal choice condition (M = 7.39, SD = 3.11; F( 1, 389) = 4.41, p =.036). When the information was diagnostic, willingness to pay was greater in the personal choice condition (M = 8.42, SD = 3.86) than in the serendipity condition (M = 7.42, SD = 3.86; F( 1, 389) = 4.34, p =.038).
A 2 (condition) × 2 (information) ANOVA on feelings of serendipity revealed a main effect of condition, such that participants in the serendipity condition (M = 4.67, SD = 1.57) reported greater feelings of serendipity than those in the personal choice condition (M = 3.96, SD = 1.70; F( 1, 389) = 19.64, p <.001). The interaction (F( 1, 389) =.357, p =.550) and main effect of information (F( 1, 389) = 2.72, p =.100) were not significant.
We conducted a bootstrapping moderated mediation analysis using PROCESS Model 15 ([16]), with the moderator influencing the indirect path postmediator. We used condition (serendipity vs. personal choice) as the independent variable, information as the moderator, and feelings of serendipity as the mediator for each of the outcomes we measured. For the satisfaction measure, the index of moderated mediation was significant (index = –.42; 95% CI: [−.70, −.20]). When the information was nondiagnostic, the pathway to satisfaction through feelings of serendipity was positive (β =.49, SE =.12, 95% CI: [.26,.75]). When the information was diagnostic, the pathway to satisfaction through feelings of serendipity was not significant (β =.08, SE =.06, 95% CI: [−.03,.20]). These results suggest that feelings of serendipity drive satisfaction when consumers do not think they have all the knowledge necessary to make a choice themselves. A similar pattern emerged for the interest, willingness to recommend, and willingness to pay measures (see Web Appendix D).
In support of H4, Study 4 demonstrates that when consumers perceive that they have enough knowledge to make their own choices, they are more satisfied with product encounters that they choose than those that occur serendipitously. Feelings of serendipity were still high, but serendipity was simply not as desirable when consumers perceived they had the information they needed to make a choice. These findings are theoretically and managerially important because they show that consumers may sometimes experience feelings of serendipity, which are not negative, but still prefer to make their own choices. In addition, marketers should be careful to not provide too much diagnostic information that can lead consumers to believe that others should not choose for them. To provide further theoretical and practical insight, Web Appendix D–1 presents an additional study examining the role of information in encounters involving serendipity.
This research developed and tested a conceptualization of the role of serendipity in the marketplace. We proposed that feelings of serendipity arise when a consumer encounter is positive, unexpected, and attributed to some degree of chance. The results of four main studies and two supplemental studies support our conceptualization. In multiple domains (online subscription services, works of art, movies, food consumption, and music), the presence of serendipity (Studies 1–4, C–1, and D–1) positively influenced satisfaction, enjoyment, meaningfulness, willingness to pay, willing to recommend, and interest. This effect was attenuated when the encounter was negative (Study 2), when a product recommendation was deterministic (i.e., carefully controlled by a marketer; Study 3), and when consumers believed they had enough knowledge to make their own choices (Study 4). In contrast, the effect was enhanced when consumers believed there was a high degree of randomness involved in the selection of the experience, which increased attributions of the experience to chance (Study 3).
This research has implications for the literature on serendipity. Some research has examined consumers' appreciation for online recommendations ([14]) but has not provided much evidence for how well recommendations work compared with personal choices and what it is about these recommendations that consumers see positively. We show that feelings of serendipity associated with a recommendation make the consumer experience more positive compared with having a personal choice, and that these feelings can influence a large set of consumer-relevant outcomes. This implies that, instead of simply making recommendations that try to match previous behavior and stated preferences, online recommendation services should design experiences that appear to involve chance, as this will make consumers more satisfied.
The role of chance uncovered in this research informs the literature on surprise in the marketplace. Surprising events can be positive or negative ([ 4]; [31]), and the current research suggests that studies investigating consumer responses to unexpectedness must consider the degree of chance involved. Responses to positive surprises may not be as positive if the consumer is aware that the surprise was carefully planned by a marketer. This implies that the literature on surprise should manipulate or measure the perceived amount of chance that led to something unexpected happening, as this may provide knowledge on why surprises are sometimes so positive (attributed to chance: "this was meant to be") and sometimes not (attributed: to specific events "I know exactly how this happened").
The current work also contributes to the literature on uncertainty and how it influences consumption. Uncertainty and low control are associated with stress ([10]) and lead consumers to engage in behaviors to regain control ([ 9]; [46]). Research investigating how uncertainty and low control influence choice and consumption should be cautious about the role of chance. Leading consumers to believe that something happened as a result of random events could backfire if the encounter turns out to be negative. The result could lead not only to lower consumer satisfaction but also to magnified negative consequences for a consumer's well-being. If an encounter becomes negative (e.g., a movie that leads a consumer to remember traumatic life events, a product the consumer is allergic to), an attribution to chance may generate a strong emotional reaction, which would otherwise be attenuated if the experience was attributed to a specific source. This does not mean that the study of serendipity should be limited because of the possibility that consumers are averse to uncertainty and low control. Instead, serendipity needs to be further explored and be accompanied by a clear understanding of what brings positive value to the consumer.
Moreover, the current work shows that serendipity occurs when the consumer does not choose a specific product or experience, which means that our findings inform work on consumer preference for choice ([ 1]; [ 2]; [ 7]). Prior work has shown that people generally like choosing and having more options, and that personal control over choice can increase satisfaction. Alternatively, the current research aligns with the smaller set of evidence on how choice does not always lead to greater satisfaction. This implies that the assumption that choice is preferred over not having choice must be revised to include considerations about how much serendipity is involved when there is no choice. Consumers may prefer the ability to choose, but the absence of choice can lead to greater satisfaction when the consumption context is positive, unexpected, and involves attributions to chance.
Finally, this research has implications for the literature on how the absence of deliberative choice influences well-being ([20]; [40]). Relatively little marketing research has examined how conditions that lack deliberate action influence consumer outcomes. Here, the lack of deliberative choice made products and experiences seem more meaningful. This is important because part of being happy is the feeling that there is meaning in life ([26]). A lack of meaning is averse and leads people to immediately engage in meaning restoration ([18]). This means that, instead of focusing on understanding how consumers actively seek meaning, researchers may help consumers by putting more emphasis on events that generate meaning without the need for consumers to actively search for it. Our findings show that marketers can structure several contexts, across industries, to imbue experiences with meaning.
These findings have important implications for marketers and consumers. Consider the different domains in which marketers sell based on recommendations. Marketers may want to emphasize the number of options available and how the encounter with a specific option is a result of chance or randomness. Enhancing perceptions of chance engenders the sense that an experience was "meant to be" given the number of alternative outcomes and increases satisfaction with the recommendation. This strategy is most effective, of course, when the recommendations are positive, and it provides an alternative to the view that consumers appreciate knowing that a marketer has specifically tailored an option to them.
In fact, the current research suggests that decreasing the salience that there is a marketer behind the recommendations enhances enjoyment. Thus, marketers should avoid framing an experience with communication suggesting that the firm has "made this selection carefully for you after examining your preferences." This is important because much marketing communication highlights the targeting process by informing consumers that a product was selected for them based on what the company knows about their preferences. The recommendations may be good, but such emphasis decreases the likelihood that consumers will experience feelings of serendipity, as attribution to chance and luck is replaced by an attribution to being watched and targeted by a specific firm. Of course, recommendations are, by definition, made by the company behind a product, but this fact need not be salient at the time a recommendation is made.
In addition, companies may consider enhancing consumer experiences by providing more opportunities for serendipitous encounters. For example, consumers may enjoy some unexpected events more as part of vacation packages relative to events they personally choose to experience. These events could be partially planned, as when consumers know there will be unexpected activities but do not know when or what they are, or completely unplanned, as when consumers are given free time as part of a travel package but are surprised by an activity that feels serendipitous. While vacation packages mostly involve activities chosen by the travel agency (i.e., the marketer), these are typically previously determined and known by consumers, which our results indicate may not always generate the highest level of enjoyment and satisfaction.
Another valuable insight is that distancing the consumer from the controlled act of choice can systematically enhance and sustain enjoyment over longer periods of time. This insight can inform strategies for promotion tactics such as induced trials via sampling, mailers, and event marketing, whereby marketers can take extra steps to imbue such situations with serendipity. For example, when companies send small product samples to consumers via mail, they typically provide a lot of information about the product, its benefits, and why the consumer is receiving the product. Our findings indicate that providing less information, leaving room for thoughts about how there may have been some chance involved in receiving that exact sample product, may increase enjoyment and the likelihood that the consumer will buy the product.
The idea of providing less information about the recommendation mechanism also has implications for online recommendations. These are typically based on consumers' profiles, preferences, and previous behavior ([28]). There are varying levels of satisfaction with these recommendations, and our findings indicate that this variation can be partially explained by how much diagnostic information consumers have about how the services work, how the selections are made, and the options themselves. Making recommendations is a well-advised strategy as long as the information provided to consumers does not make them believe that they know enough to make their own choices. In these cases, consumers still have feelings of serendipity, but these feelings do not translate to higher satisfaction with the recommendation and the experience as a whole. This means that services that use recommendations can still provide important information to consumers but must be careful not to provide too much diagnostic information that will decrease the appeal of an experience that the consumer does not choose.
Moreover, the positive effect of serendipity was mitigated when the product had a negative valence. This suggests that for products that may generate negative affect, from solemn movies to more critical experiences such as medical services, attempts to imbue the experience with serendipity would likely result in a stronger negative appraisal of the experience. Negative experiences do not benefit from serendipity, and serendipity can even exacerbate the negativity, as was the case in Study 2.
An interesting question is whether serendipity will translate to increased satisfaction when the product encounter and consumption do not occur at the same time. There are subscription services, such as those for books or wine, where consumers receive a product and only consume it later. Study 1 provides some insight into this question, as consumers who experienced feelings of serendipity showed higher satisfaction up to a month after receiving the products. However, people wear clothes over and over again, which means that eventually the effect of serendipity should fade away. For products that are consumed once, apart from the initial product encounter, we speculate that feelings of serendipity will still have an effect on satisfaction. The surprise may not be present anymore, but attributions to chance, and the feeling that the consumer was lucky, should still have an influence on the enjoyment of the experience. This is an important extension that future research could explore.
The findings also have implications for when consumers buy gifts. Often, the receiver knows they will receive a gift, and sometimes a person receives a gift by surprise. We all know that receiving a surprise gift is positive, but our findings show a way to make the experience even more positive for the gift receiver. The gift giver could communicate that there were many options to choose from or that one option was selected on the basis of intuition rather than much deliberation. This may generate feelings of serendipity for the receiver of the gift and increase satisfaction. This recommendation is interesting, as we tend to believe that signaling effort and planning behind the choice of a gift makes the receiver more satisfied with the gift. However, this belief ignores the possibility that serendipity can also enhance gift giving, which is an avenue for future research.
The benefits of serendipity emerged across many categories, but it is possible that the effects would not emerge for certain products and experiences. The effects may be less evident for durable goods that are relatively expensive (e.g., cars, appliances), for products or services that require an extended amount of information gathering before being experienced (e.g., surgical procedures, medical treatments), or for products that have a less hedonic orientation than those we investigated. In addition, in the context of ordinary experiences that occur regularly (e.g., an ice cream vendor that visits one's neighborhood nightly), consumers may expect to find them, which decreases the likelihood that consumers will have feelings of serendipity related to these experiences. Future research is needed to examine whether product type or the frequency with which an experience occurs alters the serendipity effect.
Further, it is unclear if the results would hold for products associated with strong, preexisting brand preferences (e.g., colas). It is possible that the effects are specific to the experience of products or services for which preexisting preferences or brand loyalty are not strong ([ 7]). This is a common phenomenon in consumer decision-making research, as strong previous attitudes and preferences may be immune to the effect of marketer-driven manipulations. This is not something we tested in the current investigation, but could be explored in future research.
Despite these limitations, we consistently found that serendipity leads to positive outcomes. Complementing previous research on the positive power of uncontrolled events ([25]; [38]), the effects emerged because people perceived such events to involve surprises and chance. Given that little marketing research has examined consumer outcomes when events are not deliberately orchestrated, future work is poised to build on these findings to further consider the varied effects that can emerge when the absence of choice signals the presence of serendipity.
Supplemental Material, sj-pdf-1-jmx-10.1177_00222429211000344 - Serendipity: Chance Encounters in the Marketplace Enhance Consumer Satisfaction
Supplemental Material, sj-pdf-1-jmx-10.1177_00222429211000344 for Serendipity: Chance Encounters in the Marketplace Enhance Consumer Satisfaction by Aekyoung Kim, Felipe M. Affonso, Juliano Laran and Kristina M. Durante in Journal of Marketing
Footnotes 1 Connie Pechmann
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Aekyoung Kim https://orcid.org/0000-0003-2995-7482 Kristina M. Durante https://orcid.org/0000-0002-2657-7700
5 Online supplement: https://doi.org/10.1177/00222429211000344
6 As indicated in the Web Appendix procedures, we also measured meaningfulness in Studies 2 and 3 and Supplemental Studies C–1 and D–1. This measure follows the same pattern as the main dependent variables and was more central in a previous version of this paper. For this reason, the results for the subsequent studies are discussed only in the Web Appendix.
References Botti Simona, Iyengar Sheena S. (2006), "The Dark Side of Choice: When Choice Impairs Social Welfare," Journal of Public Policy & Marketing, 25 (1), 24–38.
Botti Simona, McGill Ann L. (2006), "When Choosing Is Not Deciding: The Effect of Perceived Responsibility on Satisfaction," Journal of Consumer Research, 33 (2), 211–19.
Brehm Jack W. (1972), Responses to Loss of Freedom: A Theory of Psychological Reactance. Morristown, NJ: General Learning Press.
Calvo Manuel G., Castillo M. Dolores. (2001), "Selective Interpretation in Anxiety: Uncertainty for Threatening Events," Cognition and Emotion, 15 (3), 299–320.
Carmon Ziv, Wertenbroch Klaus, Zeelenberg Marcel. (2003), "Option Attachment: When Deliberating Makes Choosing Feel Like Losing," Journal of Consumer Research, 30 (1), 15–29.
Celma Oscar. (2010), "Music Recommendation," in Music Recommendation and Discovery: The Long Tail, Long Fail, and Long Play in the Digital Music Space, Celma Oscar, ed. Berlin: Springer-Verlag, 43–85.
7 Chernev Alexander. (2003), "When More Is Less and Less Is More: The Role of Ideal Point Availability and Assortment in Consumer Choice," Journal of Consumer Research, 30 (2), 170–83.
8 Cunha Miguel Pina. (2005), "Serendipity: Why Some Organizations Are Luckier than Others," working paper, Universidade Nova de Lisboa, FEUNL Working Paper Series.
9 Cutright Keisha M., Samper Adriana. (2014), "Doing It the Hard Way: How Low Control Drives Preferences for High-Effort Products and Services," Journal of Consumer Research, 41 (3), 730–45.
Durante Kristina, Laran Juliano. (2016), "The Effect of Stress on Consumer Saving and Spending," Journal of Marketing Research, 53 (5), 814–28.
Faraji-Rad Ali, Pham Michel Tuan. (2017), "Uncertainty Increases the Reliance on Affect in Decisions," Journal of Consumer Research, 44 (1), 1–21.
Feather Norman T., Simon Jerrold G. (1971), "Attribution of Responsibility and Valence of Outcome in Relation to Initial Confidence and Success and Failure of Self and Other," Journal of Personality and Social Psychology, 18 (2), 173–88.
Feldman Jack M., Lynch John G. (1988), "Self-Generated Validity and Other Effects of Measurement on Belief, Attitude, Intention, and Behavior," Journal of Applied Psychology, 73 (3), 421–35.
Ge Yong, Xiong Hui, Tuzhilin Alexander, Xiao Keli, Gruteser Marco, Pazzani Michael. (2010), "An Energy-Efficient Mobile Recommender System," in Proceedings of the 16th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Rao Bharat, Krishnapuram Balaji, eds. New York: Association for Computing Machinery, 899–908.
Goldsmith Kelly, Amir On. (2010), "Can Uncertainty Improve Promotions?" Journal of Marketing Research, 47 (6), 1070–77.
Hayes Andrew F. (2018), An Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach, 2nd ed. New York: Guilford.
Heilman Carrie M., Nakamoto Kent, Rao Ambar G. (2002), "Pleasant Surprises: Consumer Response to Unexpected In-Store Coupons," Journal of Marketing Research, 39 (2), 242–52.
Heine Steven J., Proulx Travis, Vohs Kathleen D. (2006), "The Meaning Maintenance Model: On the Coherence of Social Motivations," Personality and Social Psychology Review, 10 (2), 88–110.
Herlocker Jonathan L., Konstan Joseph A., Terveen Loren G., Riedl John T. (2004), "Evaluating Collaborative Filtering Recommender Systems," ACM Transactions on Information Systems (TOIS), 22 (1), 5–53.
Iyengar Sheena S., Lepper Mark R. (1999), "Rethinking the Value of Choice: A Cultural Perspective on Intrinsic Motivation," Journal of Personality and Social Psychology, 76 (3), 349–66.
Kim Min Gyung, Mattila Anna S. (2010), "The Impact of Mood States and Surprise Cues on Satisfaction," International Journal of Hospitality Management, 29 (3), 432–36.
King Laura A., Hicks Joshua A., Krull Jennifer L., Gaiso Amber K. Del. (2006), "Positive Affect and the Experience of Meaning in Life," Journal of Personality and Social Psychology, 90 (1), 179–96.
Kotkov Denis, Wang Shuaiqiang, Veijalainen Jari. (2016), "A Survey of Serendipity in Recommender Systems," Knowledge-Based Systems, 111, 180–92.
Krantz David L. (1998), "Taming Chance: Social Science and Everyday Narratives," Psychological Inquiry9 (2), 87–94.
Kray Laura J., George Linda G., Liljenquist Katie A., Galinsky Adam D., Tetlock Philip E., Roese Neal J. (2010), "From What Might Have Been to What Must Have Been: Counterfactual Thinking Creates Meaning," Journal of Personality and Social Psychology, 98 (1), 106–18.
Lambert Nathaniel M., Stillman Tyler F., Hicks Joshua A., Kamble Shanmukh, Baumeister Roy F., Fincham Frank D. (2013), "To Belong Is to Matter: Sense of Belonging Enhances Meaning in Life," Personality and Social Psychology Bulletin, 39 (11), 1418–27.
Laran Juliano, Tsiros Michael. (2013), "An Investigation of the Effectiveness of Uncertainty in Marketing Promotions Involving Free Gifts," Journal of Marketing, 77 (2), 112–23.
Lee Wei-Po, Liu Chih-Hung, Lu Cheng-Che. (2002), "Intelligent Agent-Based Systems for Personalized Recommendations in Internet Commerce," Expert Systems with Applications, 22 (4), 275–84.
Leong Tuck, Vetere Frank, Howard Steve. (2008), "Abdicating Choice: The Rewards of Letting Go," Digital Creativity, 19 (4), 233–43.
Lindgreen Adam, Vanhamme Joëlle. (2003), "To Surprise or Not to Surprise Your Customers: The Use of Surprise as a Marketing Tool," Journal of Customer Behaviour, 2 (2), 219–42.
Loewenstein George. (1994), "The Psychology of Curiosity: A Review and Reinterpretation," Psychological Bulletin, 116 (1), 75–98.
Makri Stephann, Blandford Ann. (2012), "Coming Across Information Serendipitously–Part 1: A Process Model," Journal of Documentation, 68 (5), 684–705.
Matt Christian, Benlian Alexander, Hess Thomas, Weiß Christian. (2014), "Escaping from the Filter Bubble? The Effects of Novelty and Serendipity on Users' Evaluations of Online Recommendations," in Proceedings of the 35th International Conference on Information Systems (ICIS2014), Myers Michael D., Straub Detmar W., eds. Auckland, New Zealand: Association for Information Systems, 1503–20.
McCay-Peet Lori, Toms Elaine G. (2010), "The Process of Serendipity in Knowledge Work," in Proceedings of the Third Symposium on Information Interaction in Context, Belkin Nicholas J., ed. New York: Association for Computing Machinery, 377–82.
Meade Adam W., Craig S. Bartholomew. (2012), "Identifying Careless Responses in Survey Data," Psychological Methods, 17 (3), 437–55.
Mellers Barbara A., Schwartz Alan, Ho Katty, Ritov Ilana. (1997), "Decision Affect Theory: Emotional Reactions to the Outcomes of Risky Options," Psychological Science, 8 (6), 423–29.
Melo Ricardo, Carvalhais Miguel. (2013), "The Design of Horacle: Inducing Serendipity on the Web," in xCoAx 2013: Proceedings of the First Conference on Computation, Communication, Aesthetics and X, Verdicchio Mario, Carvalhais Miguel, eds. Porto, Portugal: Universidade do Porto, 183–91.
Morewedge Carey K., Giblin Colleen E., Norton Michael I. (2014), "The (Perceived) Meaning of Spontaneous Thoughts," Journal of Experimental Psychology. General, 143 (4), 1742–54.
Parker Mina. (2008), Silver Linings: Meditations on Finding Joy and Beauty in Unexpected Places. Newburyport, MA: Conari Press.
Raghunathan Rajagopal, Irwin Julie R. (2001), "Walking the Hedonic Product Treadmill: Default Contrast and Mood-Based Assimilation in Judgments of Predicted Happiness with a Target Product," Journal of Consumer Research, 28 (3), 355–68.
Reisenzein Rainer, Horstmann Gernot, Schützwohl Achim. (2019), "The Cognitive-Evolutionary Model of Surprise: A Review of the Evidence," Topics in Cognitive Science, 11 (1), 50–74.
Shanahan Michael J., Porfeli Erik J. (2006), "Chance Events in the Life Course," Advances in Life Course Research, 11, 97–119.
Sharot Tali, De Benedetto, Martino, Dolan Raymond J. (2009), "How Choice Reveals and Shapes Expected Hedonic Outcome," Journal of Neuroscience, 29 (12), 3760–65.
Stiensmeier-Pelster Joachim, Martini Alice, Reisenzein Rainer. (1995), "The Role of Surprise in the Attribution Process," Cognition and Emotion, 9 (1), 5–31.
Valenzuela Ana, Mellers Barbara, Strebel Judi. (2010), "Pleasurable Surprises: A Cross-Cultural Study of Consumer Responses to Unexpected Incentives," Journal of Consumer Research, 36 (5), 792–805.
VanBergen Noah, Laran Juliano. (2016), "Loss of Control and Self-Regulation: The Role of Childhood Lessons," Journal of Consumer Research, 43 (4), 534–48.
Walpole Horace. (1754), "Letter to Horace Mann, 28 January 1754," in Horace Walpole's Correspondence: Yale Edition [Electronic Version], Vol. 20, Lewis Wilmarth S., ed. Farmington, CT: The Lewis Walpole Library, Yale University, 407–408. https://images.library.yale.edu/hwcorrespondence/page.asp?vol=20&seq=435.
Westbrook Robert A., Oliver Richard L. (1991), "The Dimensionality of Consumption Emotion Patterns and Consumer Satisfaction," Journal of Consumer Research, 18 (1), 84–91.
Zhang Yuan Cao, Séaghdha Diarmuid Ó., Quercia Daniele, Jambor Tamas. (2012), "Auralist: Introducing Serendipity into Music Recommendation," in Proceedings of the Fifth ACM International Conference on Web Search and Data Mining, Adar Eytan, Teevan Jaime, eds. New York: Association for Computing Machinery, 13–22.
~~~~~~~~
By Aekyoung Kim; Felipe M. Affonso; Juliano Laran and Kristina M. Durante
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 110- Shedding Light on the Dark Side of Firm Lobbying: A Customer Perspective. By: Vadakkepatt, Gautham G.; Arora, Sandeep; Martin, Kelly D.; Paharia, Neeru. Journal of Marketing. May2022, Vol. 86 Issue 3, p79-97. 19p. 1 Diagram, 4 Charts. DOI: 10.1177/00222429211023040.
- Database:
- Business Source Complete
Shedding Light on the Dark Side of Firm Lobbying: A Customer Perspective
Firms spend a substantial amount on lobbying—devoting financial resources on teams of lobbyists to further their interests among regulatory stakeholders. Previous research acknowledges that lobbying positively influences firm value, but no studies have examined the parallel effects for customers. Building on the attention-based view (ABV) of the firm, the authors examine these customer effects. Findings reveal that lobbying negatively affects customer satisfaction such that the positive relationship between lobbying and firm value is mediated by losses to customer satisfaction. These findings suggest a dark side of lobbying and challenge current thinking. However, several customer-focused moderators attenuate the negative effect of lobbying on customer satisfaction, predicted by ABV theory, including the chief executive officer's background (marketing vs. other functional area) and the firm's strategic use of resources (advertising spending, research-and-development spending, or lobbying for product market issues). These moderators ensure consistency between lobbying and customer priorities or direct firm attention toward customers even while firms continue to lobby. Finally, the authors verify that lobbying reduces the firm's customer focus by measuring this focus directly using text analysis of firm communications with shareholders. Collectively, the research provides managerial implications for navigating both lobbying activities and customer priorities, and public policy implications for lobbying disclosure requirements.
Keywords: attention-based view; corporate political activity; customer satisfaction; lobbying; regulation
Lobbying, defined as "expending resources in an attempt to sway government officials to make decisions beneficial to the lobbying firm" ([74], p. 1138), is a primary means for firms to manage their regulatory environment and attain strong returns. Accordingly, lobbying expenditures have increased by more than 130% since 1998 ([15]), and many large firms (e.g., Ford, Cisco, Facebook, Delta) maintain their own government affairs divisions, which retain dozens of lobbyists to represent their interests (opensecrets.org). The strong accounting and financial market returns to lobbying ([89]), estimated by some at 22,000% ([ 3]), can even exceed returns to product market investments such as research and development (R&D) ([11]). Similarly, recent findings reveal that $325 million in lobbying investments by Fortune 100 firms accounted for $338 billion in federal contracts in return ([ 7]).
Lobbying is the most common form of corporate political activity in the United States ([36]). Through lobbying, firms aim to minimize threats and exploit opportunities in their regulatory environment. As regulatory capture theory explains, firms derive competitive advantages from benefits such as subsidies, monopolistic or favorable competitive conditions (e.g., barriers to entry, access to new markets), protective tariffs, and fixed prices ([87]). Because these competitive benefits do not hinge on the firm's ability to satisfy customers, regulatory capture theory hints that lobbying could shift firm attention away from customer priorities ([22]). Anecdotal evidence supports this argument. For example, Oracle lobbies extensively on technology policy matters, to such an extent that some observers have criticized its lack of focus on customers. A former government official accused Oracle of "using government as a weapon to delay, annoy, and extract value from other entities" rather than attending to the marketplace or its customers ([42]). Despite these arguments, the effects of firm lobbying on customer outcomes remain largely unknown.
Regulatory capture theory alludes that lobbying may adversely affect customer outcomes, but no research in marketing directly examines this relationship. Moreover, extant theory does not explain why a focus on the regulatory environment reduces the firm's customer focus. Therefore, our first research objective is to investigate the heretofore unexamined effect of firm lobbying on customer satisfaction—a critical customer outcome that affects firm value. We draw from the attention-based view (ABV) of the firm ([63]), which argues firms have limited attention available to devote to distinct strategic priorities ([50]; [63], [64]; [65]). Because lobbying can produce direct firm advantages by appealing to the regulatory environment, firms may be inclined to focus on specific activities, imperatives, and stakeholders in that environment, rather than on customers, which should lead to diminished customer satisfaction. In line with this reasoning, previous research highlights that firms struggle to maintain focus on multiple distinct priorities, such as when they partner with competitors versus channel members in interfirm alliances ([75]) or attempt to both grow revenues and cut costs ([77]).
Yet the high returns to lobbying make it unlikely that firms will halt this practice. To this end, our second research objective is to identify strategic levers firms can use to minimize the negative effects of their lobbying on customer satisfaction. The ABV of the firm also informs our choice of which strategic levers to study. The theory posits that firm behavior is an outcome of the distribution of decision makers' attention. Decision makers' attention, in turn, is informed by personal values, unique firm resources, and rule configurations ([63], [64]; [66]). Accordingly, we predict that four moderators might effectively channel firm attention toward customers: chief executive officer (CEO) background (marketing vs. other functional area), the firm's spend on advertising and R&D, and lobbying for product market issues (rather than for non–product market issues). These four moderators work as aligning and/or counterbalancing mechanisms. Aligning mechanisms ensure consistency between lobbying and customer priorities. Influential firm decision makers can direct attention and shape focus by aligning the firm's strategic priorities with their own. Counterbalancing mechanisms work to offset firm attention on one strategic priority by redirecting focus to another. We expect that various counterbalancing mechanisms can direct firm attention toward customers even while firms continue to lobby. These attention mechanisms should attenuate negative effects of lobbying on customer satisfaction.[ 6]
Finally, our third research objective is to uncover the mechanism underlying the negative relationship between lobbying and customer satisfaction. In developing our hypotheses, we posit that lobbying reduces customer satisfaction by decreasing the firm's focus on customers. With additional analyses performed on a subset of the data, we confirm this prediction. In support of the ABV, we uncover a loss of customer focus using shareholder communications, a text-based measure that captures firm emphasis on customers as conveyed in shareholder earnings calls.
The tests of our hypotheses rely on an unbalanced panel of 758 observations involving 87 publicly traded firms during the period 2000–2014. We find a significant, negative effect of lobbying on customer satisfaction, providing novel evidence of the dark side of firm lobbying. We also replicate previous findings of a positive effect of lobbying on firm value but identify a negative counteracting effect when we account for customer satisfaction. This insight challenges economic and finance literature that suggests largely positive effects of lobbying on firm value (e.g., [17]; [46]). Consistent with our expectations, we also show that the CEO's background, advertising spend, R&D spend, and product market lobbying each positively moderate the lobbying–customer satisfaction relationship. Finally, a decrease in customer focus helps explain the negative lobbying–customer satisfaction link.
Our findings contribute to marketing theory and practice in three important ways. First, we extend the ABV to reveal that otherwise-beneficial firm actions (lobbying) can simultaneously harm customer outcomes. This view helps augment shortcomings in extant theoretical frameworks (i.e., regulatory capture) for explaining how limits in firm attention reduce the necessary focus on customers, with detrimental effects for satisfaction and firm value. In addition to challenging extant research, identifying this dark side of lobbying represents a warning to firms to be wary of losing customer focus. Second, we offer solutions in the form of a set of theoretically informed, managerially relevant moderators that can align or counterbalance firm attention to customers and thereby attenuate the negative effects of lobbying on customer satisfaction. Third, we detail how the negative effects emerge, by showcasing a key pathway leading to a loss of customer focus conveyed by firms' shareholder communications. Considering firms' increasing strategic attention to lobbying, this research offers timely implications for marketing theory and practice.
We begin by developing the conceptual framework, which blends regulatory capture theory with the ABV of the firm. After we explain our empirical approach, we report the focal study findings. We also present a series of additional analyses. Finally, we conclude with theoretical implications and insights for marketing managers and policy makers.
As the list in Table 1 reveals, concepts highlighted in regulatory capture theory ([22]; [32]; [87]) underpin studies of the relationship between lobbying and firm value, most of which identify positive firm outcomes of lobbying. We draw on these foundations to suggest a positive effect of lobbying on firm value, consistent with previous findings. However, to address our research objectives, we expand our conceptual framework to account for limits to firm attention (Figure 1). We expect that firm lobbying activities decrease the firm's focus on customers, in line with arguments from the ABV ([49]; [63]; [65]). The ABV also informs our investigation of moderators of the lobbying–customer satisfaction relationship.
Graph: Figure 1. Customer focus and the dark side of lobbying: conceptual framework.
Graph
Table 1. Selected Lobbying Research: Past Outcomes and Moderators Studied.
| Study | Firm Outcomes | Customer Outcomes | Moderators |
|---|
| Kee, Olarreaga, and Silva (2004) | Market access, tariffs | | |
| De Figueiredo and Silverman (2006) | Earmarks | | |
| Alexander, Mazza, and Scholz (2009) | Money repatriated | | |
| Richter, Samphantharak, and Timmons (2009) | Effective tax rate | | |
| Yu and Yu (2011) | Fraud lawsuits | | |
| Duchin and Sosyura (2012) | Approval for Troubled Asset Relief Program bailout funds | | |
| Hill et al. (2013) | Firm value | | Political action committee contributions |
| Borisov, Goldman, and Gupta (2016) | Abnormal returns | | Firm unethical behavior |
| Chen, Parsley, and Yang (2015) | Net income, stock market returns | | |
| Gao and Huang (2016) | Returns, trading volume | | |
| Kang (2016) | Policy enactment, lobbying returns | | |
| Unsal, Hassan, and Zirek (2016) | Firm value | | CEO political affiliation |
| Ridge, Ingram, and Hill (2017) | Government contracts | | Connectedness |
| Fidrmuc, Roosenboom, and Zhang (2018) | M&A review outcomes | | Regulator risk |
| Martin et al. (2018) | Firm value, systematic and idiosyncratic risk | | R&D and advertising stock |
| Rayfield and Unsal (2018) | Product approval | Product recall | |
| Diestre, Barber, and Santalo (2019) | Timing of safety alert | Drug-related side effect | |
| Lambert (2019) | Regulatory enforcement, firm performance | | |
| Current study | Firm value | Customer satisfaction | CEO marketing background, advertising spend, R&D spend, product market lobbying |
As noted previously, lobbying has a positive effect on firm value ([12]; [17]; [46]; [60]), which helps explain its growing practice. In particular, lobbying can result in regulatory capture ([87]) or allow the firm to dominate decisions about its regulatory environment ([32]). With regulatory capture, a firm attains disproportionate influence over a regulatory system designed to constrain and temper their behavior, which produces firm-specific benefits ([22]). For example, regulatory capture might create policy advantages or allow firms to establish monopoly power ([56]). The diverse outcomes of regulatory capture might assist firms directly, without benefiting customers, such as reduced regulatory oversight, lower tax rates, preferred government subsidies, or entry into restricted markets (Table 1). It is important to note that financial market responses to lobbying do not necessarily hinge on policy changes. Successful lobbying might preserve a favorable status quo or foster relationships without producing any other immediate outcomes ([29]). As [52] shows, even if policy changes are rare, a firm that lobbies still achieves a positive return on its investment, perhaps because financial markets use lobbying as a signal of firm influence. Accordingly, we hypothesize the following:
- H1: Firm lobbying relates positively to firm value.
Regulatory capture creates competitive advantages for the firm that do not depend on satisfying customers. Therefore, lobbying could make customer-focused efforts seem less necessary ([22]; [23]), though we lack any clear explanation for how this shift occurs or what can be done to attenuate its effect. By turning to the ABV ([49]; [63]; [65]), we seek to address this gap (Figure 1). This theory stipulates that firm actions, adaptations, and performance outcomes result from the distribution of attention—defined as "the noticing, encoding, interpreting, and focusing of time and effort by organizational decision-makers" on behalf of the firm ([63], p. 189)—to various strategic activities ([64]). Firm decision makers confront myriad demands on their time and attention ([18]), and their bounded rationality and information processing constraints limit the activities or strategic imperatives to which they can attend ([21]; [66]; [81]). Consistent with the ABV, prior research suggests that firms and their decision makers generally cannot pursue two strategically opposed foci effectively, such as revenue and cost emphases ([77]), allying with competitors and channel members ([75]), exploration and exploitation strategies ([ 6]), meeting stock market expectations and innovating ([90]), developing international and domestic market knowledge bases (Sapienza, De Clercq, and Sandberg 2005), or pursuing growth through organic and merger and acquisition approaches (Yu, Engleman, and Van de Ven 2005).
We extend this thinking to consider firm lobbying and examine the effect on customer satisfaction. This important performance metric is a function of customer expectations, perceived quality, and perceived value ([ 5]; [35]), and it usually requires an intentional firm focus on customers. When that focus decreases, satisfaction is likely to suffer. Lobbying, or appealing to regulators for direct benefits, does not hinge on appeasing customers ([22]). Because firm attention to both lobbying and customers implies that its focus is spread across diverse priorities (i.e., regulators and customers), ABV suggests that the firm cannot adequately focus on both. We predict that when firms lobby, their attention to customers decreases for four reasons.
First, customer-focused activities such as R&D spending have more uncertain outcomes and relatively lower returns than lobbying ([11]). Therefore, firms might prioritize lobbying to attain direct advantages. Second, lobbying generally entails rent seeking from existing assets, by protecting them and expanding the returns from them to the greatest extent possible, whereas a focus on customers typically implies that firms create new value for or with customers, which is inherently more difficult. Third, lobbying and a customer focus involve conceptually distinct stakeholders and environments: legislators/regulators and customers, respectively. In the latter case, product market competition is intense, as firms jockey for position in heterogeneous customers' consideration sets. Competition in the regulatory sphere instead is less intense because the relatively fewer legislative audiences tend to be more homogeneous, and relationships can be established more quickly ([29]). Furthermore, firms (in contrast to other stakeholders such as special interest groups) have more resources and coordination ability to appeal to legislators ([59]; [67]). Fourth, firms must deploy different resources, skills, and actions to succeed in these disparate stakeholder environments. Because there are fewer, more well-defined, and more accessible regulatory and special interest group stakeholders, firms' resources, skills, and actions can be deployed more effectively than in broad and diverse customer environments. Accordingly, we hypothesize the following:
- H2: Firm lobbying relates negatively to customer satisfaction.
As noted previously, lobbying research reveals positive effects on firm value (Table 1). We examine how lobbying affects firm value when accounting for the mediating role of customer satisfaction (Figure 1). Marketing research highlights the positive influence of customer satisfaction on firm value ([ 4]), perhaps because customer satisfaction increases cash flows ([43]; [61]) and reduces future cash flow volatility ([34]; [43]). Because we expect lobbying to relate negatively to customer satisfaction (H2) but positively to firm value (H1), we predict a competing, mediating role ([96]) of customer satisfaction in the lobbying–firm value relationship. That is, a loss of customer satisfaction is a cost of lobbying, and we believe that negative relationship will detract from the firm value achieved by using lobbying. Lobbying relates positively to firm value, but the negative relationship between lobbying and customer satisfaction should have a counteracting effect. We therefore hypothesize the following:
- H3: Customer satisfaction negatively mediates the positive relationship between lobbying and firm value.
If lobbying produces direct, positive effects on firm value, firms are unlikely to temper the practice, as suggested by real-world examples of its increasing use. Therefore, to address the risk of negative counteracting effects through customer satisfaction, we propose moderators that should attenuate these negative effects. The lobbying–firm value path is well established (Table 1), so we concentrate here on the unexplored relationship of lobbying with customer satisfaction.
According to the ABV, firm behavior is an outcome of the distribution and regulation of decision maker attention and the firm-level attentional structures that support this focus ([63], [64]; [66]). In proposing the ABV, [63], p. 188) established two foundational premises: ( 1) influential firm decision makers determine the firm's attention priorities and shape its focus and ( 2) the firm's attendance to its situation and context depends on "resource and rule configurations" by which the firm allocates resources and channels attention. If lobbying diverts firm attention away from customers, these premises offer direction for refocusing attention back to customers.
Drawing from [63] two premises, we propose that the negative effect of lobbying on customer satisfaction can be attenuated through two attention-directing mechanisms. First, aligning mechanisms ensure consistency between lobbying and customer priorities. Influential firm decision makers can direct attention and shape focus by aligning the firm's strategic priorities with their own. In particular, a CEO with a marketing background can likely better align lobbying with firm attention to customers and customer imperatives than a CEO who does not have a marketing background. Second, counterbalancing mechanisms work to offset firm attention on one strategic priority by redirecting focus to another. We expect that various counterbalancing mechanisms can direct firm attention toward customers even while firms continue to lobby. Firm spending on critical priorities emphasizes their importance, especially when used together with firm efforts that seemingly work to achieve different goals, such as lobbying. Specifically, we propose greater advertising spending and R&D spending, allocations that primarily concern customers, counterbalance firm attention to lobbying, thereby offsetting the negative effects of lobbying on customer satisfaction.
Finally, previous research has not distinguished lobbying for non–product market issues, such as taxes and workplace safety, from product market issues, such as product specifications and patents (see Web Appendix A for a complete list of non–product market and product market issues for which firms can lobby). Consistent with ABV logic, we expect that lobbying for non–product market issues directs attention away from customers, whereas lobbying for product market issues orients firm attention toward customers. We distinguish between these two types of lobbying to posit that customer satisfaction is less adversely affected when firms counterbalance general lobbying efforts with a focus on product market issues. In summary, the ABV leads us to derive four moderators that focus firm attention on customers to offset negative effects, even while lobbying. We discuss each in greater detail in the following subsections.
In the ABV theoretical tradition, CEOs emerge as "the most critical players" in directing the firm's focus because they choose how the firm should channel its attention and which relevant priorities it pursues ([63], p. 197; [93]), even among diverging priorities such as lobbying and customers. Indeed, attention-based perspectives suggest that firm strategies reflect the CEO's values and vision ([19]; [86]) and that the CEO can integrate firm attention or align diverse divisions toward a shared focus on specific priorities ([50]). As we explain, and supported by [63] foundational premises, CEOs shift firm attention to different imperatives through an aligning mechanism. Therefore, CEO background, which is defined as the functional knowledge and skills that CEOs develop throughout their educational and career experiences ([78]), should inform firm strategic priorities and attention because CEOs' background allows them to align firm focus accordingly.
We expect that the negative relationship between lobbying and customer satisfaction is moderated by the firm's CEO background, such that this relationship is less negative for firms with CEOs who have a marketing background, as opposed to firms with CEOs who have other types of functional expertise. A CEO with a marketing background understands the need to monitor customer expectations and create customer value ([13]), and thereby can more likely align disparate firm efforts, such as those focused on customers and regulators, to enhance the synergy between them. Greater CEO attention to customer priorities and alignment of otherwise disparate firm activities can create a shared, customer-centric vision throughout the firm. Taken together, these efforts should lessen the negative effect of lobbying on customer satisfaction. That is, we expect that this lessened negative effect will occur for firms that have CEOs with a marketing background, as opposed to firms that have CEOs with other functional expertise. We hypothesize:
- H4: CEO background moderates the negative effect of lobbying on customer satisfaction, such that the effect is less negative in firms with CEOs with a marketing background.
The negative relationship between lobbying and customer satisfaction may be attenuated among firms that counterbalance lobbying with advertising spend and R&D spend. In their study of strategic attention and firm performance, [31] show that if firms with widely dispersed attention increase their R&D spending, they counterbalance negative effects of attention dispersion. Further, existing research finds that spending on R&D serves as a powerful signal to employees, customers, and other stakeholders that the firm prioritizes customer value creation and maintains a fundamental focus on innovation-related activities ([19]; [72]).
We expect that the negative relationship between lobbying and customer satisfaction will be moderated by the firm's R&D spend. Foundational premises of the ABV state that resource allocation to particular priorities signal their importance to internal stakeholders and channel their attention toward these priorities ([63]; Sapienza, De Clercq, and Sandberg 2005). Firms with higher R&D spend signal greater importance of customer priorities to internal stakeholders as compared with firms that spend less on R&D. Consequently, firms that spend more on R&D experience a stronger counterbalancing effect that attenuates the customer focus loss associated with lobbying. In complement to its attention counterbalancing role, R&D spend may be deployed with lobbying to create value for customers. For example, when R&D spending and lobbying are used in concert, firms can gain access to new markets, thereby providing customers with more and varied product options. Similarly, lobbying may help the firm introduce more innovations, especially if they are subject to regulatory hurdles such as in the drug and medical devices sector (Rayfield and Unsal 2018). Thus, we expect R&D spend to moderate the lobbying–customer satisfaction relationship, lessening its negative effect.
We expect a similar moderating relationship for advertising spend, such that the negative relationship between lobbying and customer satisfaction will be moderated by the firm's advertising spending. Studies have shown that greater spending on advertising signals superior product quality, highlights customer value, and influences stakeholders' perceptions of the firm and its priorities ([38]; [41]; [45]). Typically, advertising activities are customer focused ([ 9]). Therefore, advertising spend together with firm lobbying works in a counterbalancing way: firms that spend more on advertising experience a stronger counterbalancing effect to attenuate the customer focus loss associated with lobbying. In complement to this attention counterbalancing role, when advertising spend and lobbying are integrated, they can lead to increased reach and saliency of a firm's advertising activities. For example, pharmaceutical firms deployed lobbying to help expand their available advertising options to highly profitable direct-to-consumer formats ([76]). Thus, we expect advertising spend to moderate the lobbying–customer satisfaction relationship, lessening its negative effect. For these reasons, we predict the following:
- H5a: Advertising spend moderates the negative effect of lobbying on customer satisfaction., such that the effect is less negative in firms with greater advertising spend.
- H5b: R&D spend moderates the negative effect of lobbying on customer satisfaction, such that the effect is less negative in firms with greater R&D spend.
We integrate the ABV perspective and lobbying research to suggest that the target of a firm's lobbying efforts can attenuate the negative effect of lobbying on customer satisfaction. That is, firms lobby for both non–product market issues (e.g., taxes, beneficial federal budget allocations) and product market issues (e.g., advertising, patents, product safety issues). Lobbying for product market issues should lessen the negative effect of lobbying on customer satisfaction, as product market issues have an inherent customer focus. By lobbying for issues relevant to the product market, the firm conveys its strategic prioritization of customers to its employees, customers, regulators, and other stakeholders. Likewise, lobbying that is focused on product market imperatives counterbalances lobbying that is focused on direct firm benefits, such as those stipulated in regulatory capture.
Lobbying to address product market issues, relative to other issues, should offset otherwise negative customer satisfaction effects from lobbying generally. This occurs by directing firm attention to customer priorities and channeling that attention across functional divisions (i.e., regulatory and marketing firm functions), thus attenuating the customer focus loss associated with lobbying. Conversely, lobbying for non–product market issues does not require attention to customers and resembles the direct path to firm value as stated in regulatory capture theory. Even if such lobbying secures other firm benefits, those efforts are unlikely to produce benefits for customers. We expect firm lobbying for product market issues to moderate the negative effect of lobbying on customer satisfaction; that is, the greater the proportion of lobbying for product market issues relative to total lobbying efforts, the more likely this negative effect is lessened. We predict the following:
- H6: Product market lobbying moderates the negative effect of lobbying on customer satisfaction, such that the effect is less negative in firms with a greater proportion of lobbying for product market issues.
To test our hypothesized relationships, we collected secondary data from a variety of sources. We describe the data, sources, and variable construction approaches next. The sample includes firms of various sizes from a broad range of industries, tracked over time, to provide a thorough test of our research questions.
Our initial sampling frame comprises all firms for which we can obtain customer satisfaction data from the American Customer Satisfaction Index (ACSI; theacsi.org), which provides scores for approximately 200 Fortune 500 firms across multiple industries and is commonly used in marketing strategy research (e.g., [35]; [71]). When a firm had multiple brands for which satisfaction scores are reported, we took the average as our measure of firm-level satisfaction.[ 7] We also obtained industry-level customer satisfaction scores from the ACSI database and then matched firms from this database to their entries in COMPUSTAT. Consistent with our focus and prior research, we dropped firms that failed to report both advertising and R&D expenditures ([71]).[ 8] We manually checked the firms' annual 10K statements if they reported zero R&D expenditures.[ 9] The overlap between ACSI and COMPUSTAT, when accounting for advertising spend and R&D spend, produced a final sample of 87 firms, consistent with prior research that adopts similar approaches ([43]; [71]).
Next, we combined this data set with lobbying data obtained from opensecrets.org, the Senate Office of Public Records (SOPR), and followthemoney.org. If none of these databases provided information, we set the lobbying expenditures for the firm to zero dollars. The Lobbying Disclosure Act mandates that firms report all lobbying expenditures above $5,000 per quarter; a failure to do so incurs penalties (lobbyingdisclosure.house.gov). Thus, we are confident that no firm lobbying is omitted from our data set. Lobbying data include expenditures and the issues on which firms lobby (see also Web Appendix A). To cull the CEO background data, we used disparate sources such as Securities Exchange Commission annual reports, Bloomberg, BoardEx, LexisNexis, popular media, and industry reports. We use publicly available data from Regdata ([ 2]) to identify whether an industry is regulated. Furthermore, to control for the potential effects of customer awareness of firm lobbying (i.e., lobbying visibility) that might affect satisfaction, we count relevant news articles using a Factiva search for Associated Press articles on firm lobbying. Table 2 details the variables, operationalizations, references, and data sources.
Graph
Table 2. Variables, Measures, and Data Sources.
| Construct | Variable Notation | Descriptions (Measures) | Representative Papers | Data Sources |
|---|
| Dependent Variables |
| Customer satisfaction | Cust_Satis | Customer satisfaction index (1–100) | Fornell et al. (2016) | ACSI |
| Firm value | Tobins_q | Tobin's q | Martin et al. (2018) | COMPUSTAT |
| Independent Variable |
| Lobbying | Lobby_Spend | Lobbying expenditure (federal and state)/ Assets | Hill et al. (2013) | Opensecrets.org, SOPR, followthemoney.org |
| Customer-Focused Moderators |
| CEO marketing background | Mkt_CEO | Dummy = 1 if CEO has predominant marketing experience | Saboo et al. (2017) | Annual Reports, Bloomberg, LexisNexis, BoardEx |
| Advertising spend | AD_Spend | Advertising expenditure/Assets | Tuli and Bharadwaj (2009); Gruca and Rego (2005) | COMPUSTAT |
| R&D spend | RD_Spend | R&D expenditure/Assets | Tuli and Bharadwaj (2009); Gruca and Rego (2005) | COMPUSTAT |
| Product market lobbying | Product_Lobby | Number of product market issues/Total number of issues lobbied for | Novel to this research | SOPR, Opensecrets.org |
| Firm-Specific Control Variables |
| Leverage | Leverage | Long-term debt-to-asset | Tuli and Bharadwaj (2009); Jindal and McAlister (2015) | COMPUSTAT |
| Profit | ROA | Return on assets: Ratio of net income to total assets | Rego, Morgan, and Fornell (2013) | COMPUSTAT |
| Firm size | Size | Employee count (in 1,000s) | Otto, Szymanski, and Varadarajan (2020) | COMPUSTAT |
| Lobbying visibility | Visibility | Number of articles (annual) that discuss focal firm lobbying | Novel to this research | Factiva |
| Industry-Specific Control Variables |
| Industry growth rate | Industry_Growth | Change in industry sales in primary two-digit SIC industry/total industry sale | Rego, Morgan, and Fornell (2013); Tuli and Bharadwaj (2009); Gruca and Rego (2005) | COMPUSTAT |
| Industry concentration | Industry_Conc | Herfindahl–Hirschman index (0–1) | Rego, Morgan, and Fornell (2013); Tuli and Bharadwaj (2009); Gruca and Rego (2005) | COMPUSTAT |
| Regulated industry | Industry_Regulated | Dummy = 1 if firms operate in a highly regulated industry | Martin et al. (2018) | REGDATA |
| Industry customer satisfaction | Ind_Cust_Satis | Average customer satisfaction score of industry | Novel to this research | ACSI |
1 Notes: ACSI = American Customer Satisfaction Index; SOPR = Senate Office of Public Records; SIC = Standard Industrial Classification.
Customer satisfaction, advertising spend, and R&D spend are widely used variables, so we do not detail their construction here, beyond the information provided in Table 2. Instead, we focus on the variables that require additional explanation or coding or are unique to our research. First, we calculate Tobin's q using the [20] measure, which is common to marketing research (e.g., [60]). Second, lobbying is the sum of firms' federal- and state-level lobbying spend, obtained from the SOPR, opensecrets.org (both of which report federal data), and followthemoney.org (which reports state data). Consistent with prior research (e.g., [55]; [88]) and to account for size effects, we scale lobbying spend by assets to create our lobbying measure.
Third, firms are not required to report the specific dollar amounts allocated to each lobbying issue. However, they do list the issues for which they lobby (Web Appendix A). We use this information to construct our measure of product market lobbying, according to the coding method described by [16] and [83]. Specifically, two coders independently coded all non–product market issues, defined as "activities that do not directly influence the company's ability to create, deliver, or communicate products or services to customers." A kappa value of.80 indicates high interrater reliability. Any disagreements were resolved by a member of the research team. After we identified non–product market issues, we summed their occurrences in any given year and subtracted that value from the total number of lobbying issues reported by a firm for a given year to arrive at the count of product market issues. We then scaled this count by the total number of lobbying issues reported by the firm in that year to create our measure of product market lobbying.
Fourth, we use Securities Exchange Commission annual reports, Bloomberg, LexisNexis, popular media, and industry reports to construct our CEO background variable. This dummy variable takes a value of 1 if the CEO has dominant marketing experience and 0 if s/he has any other functional background ([78]). Like our coding of product market lobbying, the coding of CEO marketing expertise was independently verified by two coders and validated by a member of the research team. Kappa values, again, exceed.80, providing evidence of interrater reliability.
Table 3 features summary statistics and correlations for the study variables. Among the 155 unique CEOs in our data set, 22% had a marketing background. Their average tenure was 4.66 years. Lobbying expenditures are less than advertising and R&D expenditures, yet firms spend more money lobbying Congress than taxpayers spend to operate the legislative branch ([29]). Each member of Congress is the focus of about $3.7 million annually in attempted influence and persuasion by U.S. firms ([ 7]).
Graph
Table 3. Summary Statistics and Correlations.
| Construct | M | SD | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 |
|---|
| 1. | Firm value | 1.87 | 1.48 | 1.00 | | | | | | | | | | | | | |
| 2. | Customer satisfaction | 78.94 | 5.09 | .16 | 1.00 | | | | | | | | | | | | |
| 3. | Lobbying | .00 | .00 | .11 | −.07 | 1.00 | | | | | | | | | | | |
| 4. | CEO marketing background | .22 | .42 | .00 | .06 | −.06 | 1.00 | | | | | | | | | | |
| 5. | Advertising spend | .04 | .05 | .20 | .21 | .32 | −.04 | 1.00 | | | | | | | | | |
| 6. | R&D spend | .02 | .03 | .27 | .03 | .08 | −.12 | .03 | 1.00 | | | | | | | | |
| 7. | Product market lobbying | .51 | .31 | .18 | .04 | .28 | .02 | .04 | .12 | 1.00 | | | | | | | |
| 8. | Firm size | 164.01 | 271.93 | −.09 | −.18 | −.14 | .12 | −.19 | −.18 | −.03 | 1.00 | | | | | | |
| 9. | Leverage | .24 | .16 | .05 | .06 | −.05 | −.03 | .07 | −.23 | .10 | −.02 | 1.00 | | | | | |
| 10. | Profit | .16 | .09 | .51 | .08 | −.02 | .00 | .28 | .00 | .09 | −.06 | .09 | 1.00 | | | | |
| 11. | Industry concentration | .11 | .10 | −.16 | −.22 | −.08 | −.00 | −.07 | −.36 | −.11 | .31 | .05 | −.10 | 1.00 | | | |
| 12. | Industry growth rate | .04 | .08 | .07 | .05 | .05 | .02 | .02 | .00 | .00 | −.01 | −.02 | .03 | −.06 | 1.00 | | |
| 13. | Regulated industry | .10 | .30 | −.02 | −.34 | .02 | −.06 | −.12 | −.18 | .12 | −.03 | .17 | .10 | .32 | −.05 | 1.00 | |
| 14. | Lobbying visibility | .74 | 2.72 | .10 | −.25 | .05 | −.01 | −.11 | .13 | .09 | −.02 | −.15 | .11 | −.11 | .01 | .12 | 1.00 |
| 15. | Industry customer satisfaction | 78.08 | 4.44 | .03 | .76 | .00 | .11 | .24 | .02 | .03 | −.06 | .01 | .06 | −.27 | .08 | −.34 | −.18 |
2 Notes: Correlations at.06 or greater (absolute value) are significant at p < .10.
We checked for multicollinearity before proceeding to the identification strategy and model specification. The mean variance inflation factor is 2.06, and the highest individual value is 4.32. In addition, to rule out multicollinearity concerns for the interaction terms (with r > .70), we residual-centered the interaction of lobbying with product market lobbying. Residual centering has been shown to reduce multicollinearity between an interaction term and its first-order effect term, to provide stable and unbiased results ([58]), and has been used in recent literature (e.g., [24]; [51]).
Prior to specifying our models, we conducted panel Granger causality tests to examine whether lobbying Granger-causes customer satisfaction or vice versa. They reveal that lobbying Granger-causes customer satisfaction (χ2 = 5.04, p < .10) and not the reverse. Next, we examine independent variable stationarity (lobbying) with panel unit root tests. A lack of stationarity dictates how the variables enter the model. The Fischer-type Phillips–Perron test rejects the null hypothesis that the variables contain unit roots (p < .01). We conclude the variable is mean-stationary and specify it in terms of levels.
To test H1–H3, we specify two equations[10] for any firm i operating in a primary two-digit Standard Industrial Classification (SIC) code industry j at time t[11]:
Graph
( 1)
and
Graph
( 2)
where Tobins_q is Tobin's q, Cust_Satis is customer satisfaction, Lobby_Spend is lobbying spend, Mkt_CEO is CEO marketing background, AD_Spend is advertising spend, RD_Spend is R&D spend, and Product_Lobby is the firm's product market lobbying. With Firm_Controls, we create a vector of firm-specific control variables (including lobbying visibility) for each firm i, whereas Industry_Controls is a vector of industry-specific control variables for firm i operating in the primary two-digit SIC industry j (see Table 2). In Equation 2, we include average industry-level customer satisfaction (Ind_Cust_Satis) as an industry-specific control variable to rule out any covariations in industry customer satisfaction and lobbying that may influence our results. We include time dummies for each observation year to account for macroeconomic factors that influence all firms. Finally, and are the idiosyncratic error terms.
Next, we specify the following equation to test H4–H6, with the same firm- and industry-specific control variables as in Equation 2:
Graph
( 3)
The set of independent variables identified in the previous subsection cover important firm and industry factors that could influence customer satisfaction and Tobin's q. However, for credible identification of the effects it is necessary to consider the potential endogeneity that could arise due to simultaneity and omitted variables ([40]).
Although we use lagged independent variables in Equations 1–3 to account for reverse causality (e.g., [71]), the analyses still might suffer from endogeneity bias due to omitted variables. To alleviate this concern, we include time fixed effects in all our focal equations. Consistent with unobserved effect models ([40]), year fixed effects help control for the omitted variables, and including the average industry-level customer satisfaction (Ind_Cust_Satis) as an industry-specific control variable helps rule out covariation in industry customer satisfaction and lobbying that may influence our results.
Although these efforts reduce concerns about omitted variables, we cannot theoretically claim that lobbying is uncorrelated with the error term in Equation 2 ([ 8]). Thus, we specify a fourth equation with lobbying as the dependent variable, the exogenous variables from Equations 1 and 2 as independent variables, and two variables that meet the criteria for preserving the rank and order conditions of the system of equations. In detail, among industry-based excluded variables, the first instrument that meets the exclusion restriction criterion is average industry lobbying (total industry lobbying expenditures, excluding focal firm lobbying expenditures, divided by number of entities lobbying in that industry as given in opensecrets.org), as an instrument for firm-specific lobbying. Because there are multiple firms in an industry, it is unlikely that average industry lobbying correlates with firm-level omitted variables that influence a focal firm's customer satisfaction. This variable also meets the relevance condition, because peer firm behavior can have a normative effect in that peer firms generally face similar market conditions.
The second instrument is the total number of lobbyists operating in an industry. The industry supply of lobbyists should normatively influence lobbying spend. The greater the total industry expenditure on lobbying, the more lobbyists are likely to be operating. It also is reasonable to assume that the cost of lobbying is lower for firms that operate in industries with more lobbyists. This information, available to all firms within an industry, is unlikely to correlate with unobserved firm-level variables that affect a focal firm's customer satisfaction, so it meets the exclusion criteria. Accordingly, we specify the following equation and add it to our system of equations:
Graph
( 4)
where Industry_Lobby and Num_Lobbyist are the average lobbying expenditure and number of lobbyists, respectively, for firm i operating in the primary two-digit SIC code industry j. The control variables are as described in Equation 2.
Before discussing our results, we note that the results of Hansen's J test reveal that the instruments are valid (p > .10). The Kleibergen–Paap test also shows that our instruments are relevant (p < .00), increasing our confidence in the use of these variables as instruments. Likewise, we examine the instrument effects on firm lobbying (Table 4, Column A). Industry lobbying has a negative, nonsignificant effect on firm lobbying (α4,1 = −.01; p > .10), and industry lobbyist supply has a negative, significant effect on lobbying (α4,2 = −.00; p < .01).[12]
Graph
Table 4. Effect of Lobbying on Customer Satisfaction and Customer Focus Moderation.
| Main Effects Model (H1–H3) | Interaction Effects Model (H4–H6) |
|---|
| Construct | (A)Lobbying | (B)Tobin's q | (C)Customer Satisfaction | (D)Lobbying | (E)Tobin's q | (F)Customer Satisfaction |
|---|
| Lobbying (H1, H2) | | 4.70***(1.67) | −8.37***(1.99) | | 4.52**(1.77) | −12.20***(2.30) |
| Customer satisfaction (H3) | | .03**(.01) | | | .03**(.01) | |
| CEO marketing background | .01(.01) | .08(.11) | −.37(.27) | .01(.01) | .08(.11) | −1.11***(.33) |
| Advertising spend | 1.02***(.36) | −1.68(2.92) | 7.15*(4.08) | 1.03***(.36) | −1.54(2.97) | 2.87(4.09) |
| R&D spend | .27(.23) | 12.26***(2.93) | 4.81(5.30) | .27(.23) | 12.26***(2.92) | −.73(6.02) |
| Product market lobbying | .10***(.02) | −.03(.24) | 1.15**(.48) | .10***(.02) | −.00(.25) | 1.22**(.47) |
| Lobbying × CEO marketing background (H4) | | | | | | 12.33***(3.18) |
| Lobbying × Advertising spend (H5a) | | | | | | 19.51***(4.95) |
| Lobbying × R&D spend (H5b) | | | | | | 57.60*(32.99) |
| Lobbying × Product market lobbying (H6) | | | | | | 8.98***(3.31) |
| Firm size | −.00**(.00) | .00*(.00) | −.00***(.00) | −.00**(.00) | .00(.00) | −.00***(.00) |
| Leverage | −.07(.06) | 1.00**(.44) | 1.44(1.09) | −.07(.06) | .99**(.43) | 1.24(1.01) |
| Profit | −.33***(.12) | 8.18***(1.19) | −1.34(1.96) | −.33***(.12) | 8.12***(1.19) | −.05(2.05) |
| Industry concentration | −.22***(.06) | 1.02**(.47) | 1.66(1.47) | −.22***(.06) | 1.02**(.47) | 1.22(1.47) |
| Industry growth rate | .06(.06) | .93*(.50) | .99(1.76) | .06(.06) | .94*(.50) | .86(1.71) |
| Regulated industry | .03**(.01) | −.05(.12) | −2.16***(.49) | .03**(.01) | −.04(.12) | −2.37***(.49) |
| Lobbying visibility | −.00(.00) | .02(.02) | −.14**(.06) | −.00(.00) | .02(.02) | −.13**(.06) |
| Industry customer satisfaction | −.01***(.00) | | .74***(.04) | −.01***(.00) | | .73***(.04) |
| Industry lobbyists | −.00***(.00) | | | −.00***(.00) | | |
| Industry lobbying spend | −.01(.04) | | | −.00(.04) | | |
| Year fixed effects | Yes | Yes | Yes | Yes | Yes | Yes |
| Constant | .61***(.12) | −2.44**(1.19) | 20.92***(3.10) | .61***(.13) | −2.61**(1.22) | 21.84***(3.08) |
| Log-likelihood | −2,593.35 | −2,582.10 |
| Akaike information criterion | 5,360.71 | 5,346.19 |
| Observations | 758 | 758 |
- 3 *p < .10, **p < .05, ***p < .01.
- 4 Notes: Robust standard errors are in parentheses. We scale lobbying and all its interaction terms by 103 for visual consistency.
We estimate two systems of equations. The first (Equations 1, 2, and 4) tests H1–H3, whereas the second system (Equations 1, 3, and 4) tests H4–H6. We have multiple equations in which the errors across them can be correlated, so we estimate the equations jointly using a structural equation model approach with correlated errors. Joint estimation across multiple equations yields more efficient estimates ([92]), accounts for endogeneity due to common omitted variable bias (Drukker 2014), and has been used to test mediation, moderation, and moderated mediation relationships in the presence of endogenous regressors (e.g., [91]).[13] Although H4–H6 focus on the simple moderation of the lobbying–customer satisfaction relationship, with a system of equations estimation approach, we can examine moderated mediation of the lobbying–Tobin's q relationship as well.
Prior to conducting the formal analyses for hypotheses testing, we describe our data using model-free evidence. By comparing mean customer satisfaction values across firms with high and low lobbying levels, this evidence reveals that lobbying is negatively associated with customer satisfaction. Customer satisfaction scores are 79.55 for low-lobbying-level firms as compared with 78.36 for high-lobbying-level firms, using a median split. When comparing the top and bottom quartile (decile), customer satisfaction is 78.45 (78.71) for low-lobbying-level firms and 77.91 (77.68) for high-lobbying-level firms.
Model-free evidence also shows that higher customer satisfaction is associated with each of our four moderators, including a CEO with a marketing background (in firms with above-median customer satisfaction, 26% have a marketing CEO, whereas this number is 18% for firms with below-median customer satisfaction), higher advertising spend (.51 for above-median customer satisfaction firms and.32 for below-median firms), higher R&D spend (.25 for above-median customer satisfaction firms vs..22 for below-median firms), and a greater proportion of product market lobbying (.54 for above-median customer satisfaction firms and.49 for below-median firms). We hold lobbying constant at a high level and compare customer satisfaction scores of firms that score high on the moderating variables with the scores of firms that score low on these variables. For firms with a marketing CEO, customer satisfaction scores are 80.61, versus 77.80 for firms with a CEO with a different functional area background. High advertising spend produces a customer satisfaction score of 80.01, whereas low advertising spend is 76.45. High R&D spend is 79.73, versus 76.91 for low R&D spend. Finally, customer satisfaction is 76.46 for firms that lobby for product market issues relative to 75.05 for firms that lobby for non–product market issues. Taken together, model-free evidence supports key predictions outlined in our hypotheses, which we test in the following subsection.
Table 4 provides the results of our empirical analyses and hypotheses tests. Recall that Table 4, Column A, displays results of our instrument tests. Column B contains the results of our test of H1. Lobbying has a significant, positive effect on Tobin's q (α1,1 = 4.70; p < .01), in support of H1 and validation of past findings. That is, we confirm a direct effect of lobbying on firm value, even when accounting for customer satisfaction. We also find a positive association between Tobin's q and customer satisfaction (α1,2 = .03; p < .05), R&D spend (α1,5 = 12.26; p < .01), and profit (ω = 8.18; p < .01). The results from Equation 2, shown in Table 4, Column C, indicate a significant negative effect of lobbying on customer satisfaction (α2,1 = −8.37; p < .01), in support of H2. Notably, the negative effect of lobbying on customer satisfaction occurs independent of our lobbying visibility control (ω = −.14; p < .05), explained previously in our sample and data section and described in Table 2. That is, our results establish a negative effect of firm lobbying on customer satisfaction, regardless of whether customers are aware of firm lobbying. Among other controls, we find that operating in a highly regulated industry lowers customer satisfaction ( = −2.16; p < .01).
We use the path modeling framework by [96] to test for mediation. As detailed in Table 4, lobbying has a significant, negative effect on customer satisfaction, which has a significant, positive effect on Tobin's q. The mediation test of the indirect path from lobbying to Tobin's q through customer satisfaction (lobbying → customer satisfaction → Tobin's q) is significant. It is the product of the lobbying to customer satisfaction path (lobbying → customer satisfaction) and the customer satisfaction to Tobin's q path (customer satisfaction → Tobin's q). The indirect effect of lobbying through customer satisfaction on Tobin's q is negative and significant (β = −.22, 95% confidence interval [CI] = [−.40, −.04]). Customer satisfaction partially and negatively mediates (competitive mediation) the effect of lobbying on Tobin's q, as we predicted in H3. The total effect (sum of direct and indirect effects) of lobbying on Tobin's q (β = 4.49, 95% CI = [1.23, 7.74]) is smaller than its direct effect (β = 4.70, 95% CI = [1.43, 7.98]), which indicates competitive mediation ([96]), such that the direct and indirect effects are in opposite directions. The direct benefits of lobbying are larger than previously identified when we consider the negative counteracting effect of customer satisfaction. The lobbying–customer satisfaction path accounts for 4.90% of the total effect of lobbying on Tobin's q.
The results in Table 4, Column F, show that a CEO's marketing background significantly and positively moderates the lobbying–customer satisfaction relationship (α3,6 = 12.33; p < .01), in support of H4. The negative effect of lobbying on customer satisfaction decreases, from significantly negative (β = −6.14, 95% CI = [−11.89, −.40]) to positive, though nonsignificant, when a firm has a CEO with a marketing background (β = 6.19, 95% CI = [−1.52, 13.89]). This result is consistent with our expectation that a marketing-focused CEO can align the firm's focus with customers, which lessens the negative effect of lobbying on customer satisfaction.
Table 4, Column F, also shows that firm advertising spend and R&D spend each positively and significantly moderate the lobbying–customer satisfaction relationship. As we predicted in H5a, the interaction involving advertising spend is positive and significant (α3,7 = 19.51; p < .01), such that the negative effect of lobbying on customer satisfaction decreases with greater advertising spend, and this effect is significantly negative at low (−1 SD) advertising spend (β = −5.43, 90% CI = [−10.26, −.60]) but is nonsignificant at higher (+1 SD) advertising spend (β = −3.22, 90% CI = [−7.93, 1.48]). Similarly, we find a positive and significant interaction between lobbying and R&D spend (α3,8 = 57.60; p <.10), in support of H5b. The negative effect of lobbying on customer satisfaction decreases with an increase in R&D spend, and this effect changes from significantly negative at lower (−1 SD) R&D spend (β = −6.51, 95% CI = [−12.19, −.83]) to nonsignificant at higher (+1 SD) R&D spend (β = −2.14, 95% CI = [−8.76, 4.48]). These results identify two important counterbalancing levers; firms can significantly lessen the negative effect of lobbying on customer satisfaction through advertising spend and R&D spend.
Product market lobbying significantly and positively moderates the lobbying–customer satisfaction relationship (α3,9 = 8.98; p < .01; Table 4, Column F), in support of H6. The negative effect of lobbying on customer satisfaction decreases with an increase in product market lobbying, and this effect changes from being significantly negative at lower (−1 SD) product market lobbying levels (β = −7.16, 95% CI = [−11.60, −2.72]) to nonsignificant at higher (+1 SD) levels (β = −1.49, 95% CI = [−8.75, 5.77]). We thus confirm our assertion that product market lobbying lessens the otherwise negative effect of lobbying generally on customer satisfaction.
Although we did not hypothesize moderating effects of CEO marketing background, advertising spend, R&D spend, or product market lobbying for the indirect effect of lobbying on Tobin's q through customer satisfaction, we conduct additional moderated mediation analyses to estimate these conditional indirect effects. We find that lobbying has a significant, conditional, indirect negative effect on Tobin's q through customer satisfaction, but only in the absence of a marketing CEO (β = −.17, 95% CI = [−.33, −.02]). The effect becomes positive, although nonsignificant, with a marketing CEO present (β = .17, 95% CI = [−.13,.48]). The conditional indirect effect also holds only at lower (−1 SD) advertising spend (β = −.15, 90% CI = [−.28, −.02]) and becomes nonsignificant at higher (+1 SD) advertising spend (β = −.09, 90% CI = [−.20,.02]); it similarly persists only at lower (−1 SD) R&D spend (β = −.18, 95% CI = [−.36, −.01]) and becomes nonsignificant at higher (+1 SD) R&D spend (β = −.06, 95% CI = [−.22,.10]). Finally, lobbying has a significant conditional indirect effect on Tobin's q through customer satisfaction at lower (−1 SD) product market lobbying levels (β = −.20, 95% CI = [−.35, −.05]), but this effect is nonsignificant at higher product market lobbying levels (+1 SD) (β = −.04, 95% CI = [−.23,.15]). That is, customer-focused variables positively moderate the negative effect of lobbying on customer satisfaction and the negative indirect effect of lobbying on Tobin's q through customer satisfaction. Put differently, when each of the moderators is at a low level, the conditional indirect effect of lobbying on Tobin's q through customer satisfaction is negative. When each moderator is at a high level, there is no indirect effect. These findings highlight that firms can use these strategic levers to neutralize the negative indirect effect of lobbying on firm value.
To further shed light on the relationship between lobbying and customer satisfaction, we provide empirical evidence regarding the loss of customer focus due to lobbying, as predicted by ABV theory. Specifically, we employ an accepted proxy for a firm's customer focus using shareholder communications. Such communications (e.g., letters to shareholders, annual reports, conference calls) provide a clear and immediate measure of the firm's priorities (e.g., [74]). These communications are scrutinized by many stakeholders, and therefore, firms are intentional about what they share. Further, these communications should reflect firm priorities and key areas of strategic focus ([31]). For example, prior research has linked such communications with the firm's innovation priorities and outcomes ([93]).
For this study, we use earnings conference call transcripts to create a measure of a firm's customer focus. Firms voluntarily disclose large volumes of information during earnings calls (e.g., [14]). The calls also include question-and-answer sessions that capture information beyond the firm's prepared remarks. Finally, noting their frequency, we believe these transcripts offer good potential for accurately capturing firm attention to (or away from) important priorities. In line with existing literature (e.g., [10]), we create a count of customer-focused words (using the dictionary created by [93]; Web Appendix B) in quarterly earnings call transcripts, as a percentage of the total number of words, and then take the mean value over four quarters in a fiscal year.
The resulting data set refers to earnings conference call transcripts for 75 firms. The mean customer focus value is.78%, and it ranges from.35% to 1.35%. We test whether this measure of customer focus mediates the lobbying–customer satisfaction relationship using the [69] Model 4 PROCESS macro with 10,000 iterations. Consistent with our expectations, lobbying has a negative effect on customer focus (β = −.34, 95% CI = [−.43, −.25]), which has a positive effect on customer satisfaction (β = 1.43, 90% CI = [.18, 2.69]). Customer focus significantly and negatively mediates the lobbying–customer satisfaction relationship (β = −.49, 90% CI = [−.94, −.05]), highlighting the pathway for this negative effect.
We use a counterfactual analysis to estimate the effect of high lobbying on customer satisfaction and its indirect effect on firm value via customer satisfaction. Because we only observe firms in their actual high-frequency or low-frequency lobbying states, we identify their counterfactual matches (i.e., firms similar to them on other variables but with different lobbying) using the nearest-neighbor matching procedure (see Web Appendix C). Using this approach, we find that lobbying has a direct negative effect on customer satisfaction and an indirect negative effect on Tobin's q via customer satisfaction. The results of this counterfactual analysis are consistent with the results of our focal study models.
Along with providing evidence of competitive mediation in the lobbying–firm value relationship through customer satisfaction, we follow prior research guidelines that advise exploring alternative mediation explanations ([96]). According to regulatory capture theory, lobbying can positively affect firm value through other routes, such as when lobbying protects the industry status quo and allows dominant firms to gain power, hinder competitive market entries, and lower competition ([87]). In this case, lobbying should enhance firm value through an increase in its market share.
We empirically evaluate this parallel mediation. While simultaneously accounting for the lobbying → customer satisfaction → firm value relationship, we examine whether lobbying → market share → firm value indicates positive mediation. The formal test uses [69] Model 4 PROCESS macro with 10,000 iterations. We find that market share partially mediates the relationship between lobbying and firm value. Lobbying has a positive effect on market share (β = .03, 90% CI = [.00,.06]), which in turn has a positive effect on firm value (β = 1.32, 95% CI = [.50, 2.15]). Thus, market share significantly and positively mediates the lobbying–firm value relationship (β =.04, 90% CI = [.00,.09]). We acknowledge there may be other routes (e.g., taxes paid, contracts) but limit our approach to providing evidence of one additional path through which lobbying influences firm value.
We examine the effect of lobbying on an alternative customer outcome: brand equity. This variable refers to the outcomes and preferences that accrue to a branded option compared with those that accrue to a similar, nonbranded alternative ([ 1]). Brand equity captures awareness, familiarity, and brand associations, so it drives both new customer acquisition and customer retention ([85]). To test for this effect, we adapt a sales-based brand equity measure ([ 1]; [25]) that reflects the revenue difference between branded and nonbranded alternatives. In our firm-level data, we lack measures of customer satisfaction or sales for nonbranded alternatives, so we proxy for their sales by taking the median of two-digit SIC sales. Thus, our measure of sales-based brand equity is the difference in firm revenue relative to industry median revenue ([44]). The analysis reveals that lobbying has a negative effect on brand equity (β = −17.94; p < .01), thus confirming our findings for another customer outcome (Web Appendix D, Column A).
With a sequence of robustness checks, we ensure the validity of the focal relationship between lobbying and customer satisfaction. Web Appendix D presents complete reporting of these tests, which include analyzing a larger data set of ACSI firms by setting advertising spend and R&D spend to 0 if they are not reported (Column B), constructing a data set of firms not included in our sample due to missing observations for our covariates (Column C), analyzing all data to include firms that do not report both advertising expenditures and R&D expenditures (Column D), testing the effect of lobbying on the difference between a firm's customer satisfaction value and average industry customer satisfaction value to account for industry effects in a different way than controlling for them (Column E), and scaling firm lobbying expenditures by the sum of its advertising expenditures and R&D expenditures to capture the relative strategic emphasis (Column F). Our focal results remain consistent in all cases, further strengthening the confidence in our findings.
Consistent with regulatory capture theory, lobbying is a positive driver of firm performance, and companies are likely to continue using it. However, our findings also reveal that costs of firm lobbying become apparent when accounting for customer effects. Specifically, we augment regulatory capture theory by building arguments using the ABV of the firm to explain how firm lobbying negatively affects customer satisfaction. To our knowledge, this investigation is the first to consider the prevalent, growing practice of firm lobbying in relation to customer outcomes. We advance research in marketing by showing that firm lobbying has a worrisome dark side: it reduces customer satisfaction, a critical customer performance outcome that is foundational to marketing theory and practice.
We also draw from the ABV perspective to suggest a set of moderators (CEO background, advertising spend, R&D spend, and product market lobbying), each of which lessens the negative lobbying–customer satisfaction relationship by preventing customer focus loss. We describe how these moderators, through aligning and counterbalancing means, orient the firm's focus to customers and minimize the negative effects of lobbying on satisfaction. Finally, by testing for customer focus loss through shareholder communications, we provide empirical evidence that the negative effect of lobbying on customer satisfaction is driven by a decrease in a firm's attention to its customers. These findings provide insights into how lobbying hurts customer outcomes. To the best of our knowledge, they offer the first empirical evidence that lobbying reduces the firm's customer focus and thereby the customer satisfaction it achieves.
These findings have important takeaways for managers. Existing research on firm lobbying has not considered customer effects, which is surprising given the critical role of customers for firm growth and survival ([84]). Our research shows that firm lobbying strongly and negatively affects customer satisfaction, and we offer some preliminary evidence that it may negatively affect brand equity–related measures as well. Moreover, we find that lobbying erodes customer satisfaction at a faster rate than advertising spending can build it. Although raw lobbying spend is less than advertising spend, the negative effect of lobbying (β = −8.37) is greater than the influence of advertising spend (β = 7.15) on customer satisfaction (Table 4, Column C). Our mediation analysis (H3) further suggests that if it is not accounted for, the indirect negative effect of lobbying on Tobin's q through customer satisfaction can negate the benefits of a 1,000,000 increase in R&D spend, and the positive effects it would have on Tobin's q, with an increase of only 55,727 in lobbying spend.
Fortunately, our findings reveal managerially relevant strategic levers firms can use to neutralize the negative indirect effect of lobbying on firm value. Taken together, moderation findings suggest that instead of keeping government affairs separate from marketing functions, the two should work together, through aligning and/or counterbalancing means, to enable firms to achieve the highest returns on their lobbying efforts. Indeed, our findings reveal useful synergies that can be derived from these firm attention–directing mechanisms.
As one example, we show that advertising spend and R&D spend counterbalance lobbying spend by reorienting collective firm attention to customers. These moderation findings also suggest that advertising spend and R&D spend serve as important signals of firm focus on customers to internal and external stakeholders and can combine with lobbying to produce firm benefits. Examples we cite from pharmaceuticals and medical device manufacturers highlight how some firms that spend considerably on lobbying use tools from the marketing environment to their advantage. The counterbalancing role of advertising spend and R&D spend is an additional benefit to the already well-known advantages that these expenditures produce for firms.
To the best of our knowledge, our study also is the first investigation to differentiate issues for which a firm can lobby in our model. We find that lobbying for product market issues, relative to non–product market issues, weakens the negative effect of lobbying spend on customer satisfaction. This finding suggests that not all lobbying activities are the same when it comes to their impact on customer satisfaction. Second, although this study just scratches the surface on how lobbying can affect customer outcomes, this moderating relationship suggests that some lobbying issues can reorient firm attention to customers. For example, Apple spent $6.65 million on lobbying in 2020 (opensecrets.org). Of those resources, although some were devoted to non–product market issues, some were devoted to product market issues, such as their music streaming division. Like Spotify, Pandora, and others, Apple lobbies for issues that allow the firm to provide customers greater access to different musical genres, artists, and albums. Although controversial among individual musicians and publishers, customers are the ultimate beneficiaries of this lobbying, via greater access and lower fees.
Additionally, we find that the negative effect of lobbying on customer satisfaction is weaker for firms with CEOs that have a marketing background, which we argue is because these CEOs direct collective attention to customers and align firm lobbying activities to be customer focused as well. This finding has major corporate governance ramifications. By effectively aligning two disparate environments (i.e., customer environment and regulatory environment), a marketing CEO can produce important, positive effects for the firm. These novel benefits provide another compelling reason to boards of directors engaged in top management recruiting for hiring a CEO with a marketing background.
A critical shortcoming of prior research that examines customer satisfaction antecedents is that most studies only note a firm's product market strategies. Satisfaction is a function of customer expectations, perceived quality, and perceived value ([ 5]; [35]), but firm lobbying can significantly influence all these dimensions. With this initial empirical evidence, grounded in compelling theory, we propose that a firm's non–product market strategy (i.e., lobbying) can significantly influence product market performance. Additional non–product market activities and their role on customer outcomes warrant investigation and theoretical refinement to the broad customer satisfaction literature. For example, investigating firm attention to corporate social responsibility initiatives, relative to customer outcomes, also may identify surprising and unintended insights for customer satisfaction theories.
The ABV of the firm is still gaining momentum in marketing theoretical development. We provide one framework for how the ABV can be used in complement with extant theories for refined insights about firm behavior. We contribute to regulatory capture perspectives by identifying a critical shortcoming of this theory regarding firm attention to customers. Our theory development also orients firm lobbying behavior squarely in the marketing literature, representing a novel contribution. Foundational theoretical premises of the ABV suggest additional theoretical applications and extensions in marketing. For example, the manner in which a firm attends to various issues, stakeholders, and environments can have important implications for marketing intelligence dissemination and organizational learning (e.g., [39]). The ABV theoretical emphasis on creating attention structures and channeling that attention suggests that marketing theories about information flows may be ripe for integration. Finally, additional marketing outcomes beyond customer satisfaction may benefit from analysis through an ABV lens. Innovation theories, in particular, may be advanced by incorporating the nature of firm focus. Attention distribution may influence the extent to which key decision makers are able to produce radical versus incremental innovation possibilities for the firm, among other outcomes.
Theoretical premises of the ABV also may shed new light on research findings that reveal the perils of a dual firm focus. Different strategic foci have the potential to pull attention away from benefiting customers. As our findings on lobbying show, it can lead to undesirable customer and firm outcomes. Additionally, in the face of negative effects, it may be useful to determine whether firm attention was spread too thin, or senior managers struggled with strategically opposed priorities. We offer evidence in support of the ABV premises that a firm's attention is even more constrained and limited than the expenditures it can dedicate to various initiatives. Through an ABV lens, we suggest that explicit consideration of firms' limited attention capacity and the influential role of aligning and counterbalancing forces may help clarify prior findings. Our investigation refines theory around these two different mechanisms, building from [63] foundational premises. Although ABV literature has proposed moderators that attenuate negative effects from loss of focus (e.g., [31]), our study is the first to show that different forces can redirect firm attention by using distinct yet complementary means. Future research should continue to disentangle aligning and counterbalancing mechanisms regarding firm attention and the ability to maintain desirable sources of focus.
Finally, in our empirical findings, lobbying is manifest in negative outcomes; customers experience reduced satisfaction regardless of their knowledge of firm lobbying behavior. Yet customers seek out information about such firm activities, and firms also increasingly communicate with customers about politically motivated behaviors ([47]; [62]; [80]). Consumer-side theory, related to perceptions and evaluations of firm attention diversion away from customer priorities, remains underdeveloped. Customers disapprove of strong business–government relationships, but why is that true? Although regulatory capture theory highlights the risks of government–business interaction, it is not clear that this theory's premises about firms' undue influence translates to customer perceptions and evaluations. Theoretical grounding of customers' strong negative reactions to lobbying and the larger family of firm political influence strategies is critical and needed. With the rapid pace of many technological advancements, business–government interactions become increasingly relevant to customers, their consumption experience, and their overall well-being. Theoretical advancement must work to further import these interactions into marketing scholarship.
The potential anticompetitive effects of lobbying, coupled with an erosion of customer satisfaction, suggest public policy implications of our results. Public sentiment suggests a growing distaste for lobbying and close ties between business and government, but few efforts have been made to curb the practice. Our findings suggest that greater limits may be warranted in some areas to promote positive customer outcomes. Although counterintuitive, greater lobbying limits may work to benefit firms, by redirecting focus to customers and by improving the quality of the firm's long-term customer outcomes.
Regardless of whether greater limits on lobbying are imposed, our study supports the need for continued disclosure mandates. The Lobbying Disclosure Act gives customers, special interest groups, advocates, and researchers more information about the role of lobbying in modern business practice. Although this reporting necessarily creates a burden for firm compliance, it may have the unexpected benefit of showcasing when firms lobby for the customer's interests, as in the music streaming example cited previously. When firms lobby for issues that benefit customers, mandatory disclosures can visibly signal a customer focus. Firms could use these disclosures as evidence in support of customer-focused lobbying.
However, there is also compelling evidence that the information contained in lobbying disclosure reports does not go far enough in either detail or metrics. In 2018, shareholder resolutions asking for greater transparency of lobbying activities were presented to 50 prominent U.S. companies ([82]). Currently, firms are required to disclose lobbying expenditures quarterly, but the information is very basic, and reporting can be inconsistent. Identifying specific dollar values devoted to any given issue would further understanding of lobbying's performance outcomes. Greater detail about the direction of firm lobbying (i.e., in support of or against an issue) also would further research objectives and give customers and other stakeholders greater insight into a firm's position on important issues. Indeed, our results show that lobbying issues matter, and clearer communication about them could benefit multiple stakeholders. Added scrutiny may further shift firm attention to customer-focused issues.
Several additional research questions arise from our findings. First, we provide initial evidence that lobbying can lower customer satisfaction, and continued research could examine other firm political behaviors and their effects on customer outcomes. For example, studying customer awareness of lobbying might provide added nuance and reveal a customer-side path that parallels our firm-side focus. Second, although we focus on how lobbying affects customer satisfaction, we provide preliminary evidence that lobbying can influence other customer metrics such as brand equity. Future studies should examine the role of lobbying on customer metrics such as brand equity using additional measures. Third, firms frequently lobby to achieve specific goals. We know of no other studies of differential effects based on the issues being lobbied, so additional research is needed to disentangle the effect of specific lobbying issues, beyond just product market versus non–product market focus, on customer outcomes and firm performance. Perhaps product market lobbying issues lead to different outcomes according to individual areas of emphasis. Alternatively, lobbying effects seemingly might be weaker if the issues are further removed from a firm's focal business domain (e.g., lobbying for environmental issues by a software company, lobbying for guns by an arts organization). Fourth, further research could explore antecedents of firm attention. For example, lobbying might increase market share, to the extent that it even might produce a monopoly. Firms that gain market share through lobbying also might exhibit greater hubris. Thus, in addition to their diverted attention, managers of these firms may have excessive confidence, which could lead them to discount customer priorities.
Firm lobbying has a worrisome dark side when accounting for customer effects. Although our research advances important findings about lobbying outcomes on customer satisfaction, we are only just beginning to realize the many ripple effects from political influence strategies on firm performance and the broader competitive market environment. We hope marketing researchers continue to investigate how marketing strategies and political strategies interface on a variety of firm and societal outcomes.
Footnotes 1 All authors contributed equally.
2 Raj Venkatesan
3 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
4 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The second author is supported in part by funding from the Social Sciences and Humanities Research Council.
5 Online supplement:https://doi.org/10.1177/00222429211023040
6 It is important to note that although we discuss aligning and counterbalancing mechanisms as distinct, they can (and do) operate simultaneously.
7 As a check, we dropped firms with multiple brands and reran the analysis; the results remained consistent.
8 We also conducted a robustness check ([53]) in which we set advertising expenditures and R&D expenditures to 0 when a firm did not report them. We provide these results subsequently.
9 When we dropped these firms from the sample and reran the analysis, the results remained consistent, but slightly stronger, for the subset of remaining firms.
Focal variables of interest in each equation have α as parameters, while ω and ϑ are vectors that include firm and industry controls parameters, respectively. The first subscript that follows a parameter refers to its associated equation.
For parsimony, we do not specify a separate Tobin's q equation without customer satisfaction as a covariate; rather, we calculate the total, direct, and indirect effects of lobbying focus on Tobin's q using a system of equations.
Focal results remain consistent if we drop the nonsignificant industry lobbying instrument from the analysis.
For thoroughness, we estimated the models using the three-stage least squares approach; the results remain consistent.
References Ailawadi Kusum L. , Neslin Scott A. , Lehmann Donald R.. (2003), " Revenue Premium as an Outcome Measure of Brand Equity ," Journal of Marketing , 67 (4), 1 – 17.
Al-Ubaydli Omar , McLaughlin Patrick. (2015), " RegData: A Numerical Database on Industry-Specific Regulations for All United States Industries and Federal Regulations, 1997–2012 ," Regulation & Governance , 11 (1), 109 – 23.
Alexander Raquel , Mazza Stephen W. , Scholz Susan. (2009), " Measuring Rates of Return on Lobbying Expenditures: An Empirical Case Study of Tax Breaks for Multinational Corporations ," Journal of Law & Politics , 25 (4), 401 – 58.
Anderson Eugene W. , Fornell Claes , Mazvancheryl Sanal K.. (2004), " Customer Satisfaction and Shareholder Value ," Journal of Marketing , 68 (4), 172 – 85.
Anderson Eugene W. , Sullivan Mary W.. (1993), " The Antecedents and Consequences of Customer Satisfaction for Firms ," Marketing Science , 12 (2), 125 – 43.
Andriopoulos Constantine , Lewis Marianne W.. (2009), " Exploitation–Exploration Tensions and Organizational Ambidexterity: Managing Paradoxes of Innovation ," Organization Science , 20 (4), 696 – 717.
Andrzejewski Adam. (2019), " How the Fortune 100 Turned $2 Billion in Lobbying Spend Into $400 Billion of Taxpayer Cash," Forbes (May 14), https://www.forbes.com/sites/adamandrzejewski/2019/05/14/how-the-fortune-100-turned-2-billion-in-lobbying-spend-into-400-billion-of-taxpayer-cash/#2132495954ff.
Angrist Joshua , Pischke Jörn-Seffen. (2008), Mostly Harmless Econometrics: An Empiricist's Companion. Princeton, NJ : Princeton University Press.
Bagwell Kyle. (2007), " The Economic Analysis of Advertising, " in Handbook of Industrial Organization , Vol. 3, Mark Armstrong and Robert H. Porter, eds. Amsterdam : Elsevier , 1701 – 1844.
Berger Jonah , Humphreys Ashlee , Ludwig Stephan , Moe Wendy W. , Netzer Oded , Schweidel David A.. (2020), " Uniting the Tribes: Using Text for Marketing Insights ," Journal of Marketing , 84 (1), 1 – 25.
Bessen James. (2016), "Accounting for Rising Corporate Profits: Intangibles or Regulatory Rent," Law & Economics Research Paper No. 16-18, Boston University.
Borisov Alexander , Goldman Eitan , Gupta Nandini. (2016), " The Corporate Value of (Corrupt) Lobbying ," Review of Financial Studies , 29 (4), 1039 – 71.
Boyd Eric D. , Chandy Rajesh K. , Cunha Marcus. (2010), " When Do Chief Marketing Officers Affect Firm Value? A Customer Power Explanation ," Journal of Marketing Research , 47 (6), 1162 – 76.
Brown Stephen , Hillegeist Stephen A. , Lo Kin. (2004), " Conference Calls and Information Asymmetry ," Journal of Accounting and Economics , 37 (3), 343 – 66.
Center for Responsive Politics (2021), "Lobbying Data Summary," (accessed May 11, 2021), https://www.opensecrets.org/federal-lobbying.
Chandy Rajesh K. , Tellis Gerard J.. (2000), " The Incumbent's Curse? Incumbency, Size, and Radical Product Innovation ," Journal of Marketing , 64 (3), 1 – 17.
Chen Hui , Parsley David , Yang Ya-Wen. (2015), " Corporate Lobbying and Firm Performance ," Journal of Business Finance & Accounting , 42 (3/4), 444 – 81.
Child John. (1972), " Organizational Structure, Environment, and Performance: The Role of Strategic Choice ," Sociology , 6 (1), 1 – 22.
Cho Theresa S. , Hambrick Donald C.. (2006), " Attention as the Mediator Between Top Management Team Characteristics and Strategic Change: The Case of Airline Deregulation ," Organization Science , 17 (4), 453 – 69.
Chung Kee H. , Pruitt Stephen W.. (1994), " A Simple Approximation of Tobin's q ," Financial Management , 23 (3), 70 – 74.
Cyert Richard , March James. (1963), A Behavioral Theory of the Firm. Englewood Cliffs, NJ : Prentice-Hall.
Dal Bó Ernesto. (2006), " Regulatory Capture: A Review ," Oxford Review of Economic Policy , 22 (2), 203 – 25.
Dal Bó Ernesto , Rossi Martin A.. (2007), " Corruption and Inefficiency: Theory and Evidence from Electric Utilities ," Journal of Public Economics , 91 (5/6), 939 – 62.
Danneels Erwin , Vestal Alex. (2020), " Normalizing vs. Analyzing: Drawing the Lessons from Failure to Enhance Firm Innovativeness ," Journal of Business Venturing , 35 (1), doi: 10.1016/j.jbusvent.2018.10.001.
Datta Hannes , Ailawadi Kusum L. , van Heerde Harald J.. (2017), " How Well Does Consumer-Based Brand Equity Align with Sales-Based Brand Equity and Marketing-Mix Response? " Journal of Marketing , 81 (3), 1 – 20.
De Figueiredo John , Silverman Brian S.. (2006), " Academic Earmarks and the Returns to Lobbying ," Journal of Law and Economics , 49 (2), 597 – 626.
Diestre Luis , Barber IV Benjamin , Santalo Juan. (2019), " The Friday Effect: Firm Lobbying, the Timing of Drug Safety Alerts, and Drug Side Effects ," Management Science , 66 (8), 3677 – 98.
Drukker David. (2014), "Some Stata Commands for Endogeneity in Nonlinear Panel-Data Models," German Stata Users Group meeting (June 13), https://www.stata.com/meeting/germany14/abstracts/materials/de14%5fdrukker%5fgsem.pdf.
Drutman Lee. (2015), The Business of America Is Lobbying. London : Oxford University Press.
Duchin Ran , Sosyura Denis. (2012), " The Politics of Government Investment ," Journal of Financial Economics , 106 (1), 24 – 48.
Eklund John Charles , Mannor Michael J.. (2020), " Keep Your Eye on the Ball or on the Field? Exploring the Performance Implications of Executive Strategic Attention, " Academy of Management Journal (published online August 18), https://doi.org/10.5465/amj.2019.0156.
Etzioni Amitai. (2009), " The Capture Theory of Regulations-Revisited ," Society , 46 (4), 319 – 23.
Fidrmuc Jana P. , Roosenboom Peter , Zhang Eden Quxian. (2018), " Antitrust Merger Review Costs and Acquirer Lobbying ," Journal of Corporate Finance , 51 , 72 – 97.
Fornell Claes , Mithas Sunil , Morgeson Forrest V. III , Krishnan M.S.. (2006), " Customer Satisfaction and Stock Prices: High Returns, Low Risk ," Journal of Marketing , 70 (1), 3 – 14.
Fornell Claes , Morgeson Forrest V. III , Hult Tomas M.. (2016), " Stock Returns on Customer Satisfaction Do Beat the Market: Gauging the Effect of a Marketing Intangible ," Journal of Marketing , 80 (5), 92 – 107.
Funk Russell J. , Hirschman Daniel. (2017), " Beyond Nonmarket Strategy: Market Actions as Corporate Political Activity ," Academy of Management Review , 42 (1), 32 – 52.
Gao Meng , Huang Jiekun. (2016), " Capitalizing on Capitol Hill: Informed Trading by Hedge Fund Managers ," Journal of Financial Economics , 121 (3), 521 – 45.
Gao Haibing , Xie Jinhong , Qi Wang , Wilbur Kenneth C.. (2015), " Should Ad Spending Increase or Decrease Before a Recall Announcement? The Marketing–Finance Interface in Product-Harm Crisis Management ," Journal of Marketing , 79 (5), 80 – 99.
Gebhardt Gary F. , Farrelly Francis J. , Conduit Jodie. (2019), " Market Intelligence Dissemination Practices ," Journal of Marketing , 83 (3), 72 – 90.
Germann Frank , Ebbes Peter , Grewal Rajdeep. (2015), " The Chief Marketing Officer Matters! " Journal of Marketing , 79 (3) 1 – 22.
Grewal Rajdeep , Chandrashekaran Murali , Citrin Alka. (2010), " Customer Satisfaction Heterogeneity and Shareholder Value ," Journal of Marketing , 47 (4), 612 – 26.
Grimaldi James V. , Mullins Brody , McKinnon John D.. (2020), "Why Are Amazon and Google in Washington's Firing Line? One Answer is Ken Glueck," Wall Street Journal (February 13), https://www.wsj.com/articles/oracles-man-in-washington-fans-the-flames-against-rival-tech-giants-11581615873.
Gruca Thomas S. , Rego Lopo L.. (2005), " Customer Satisfaction, Cash Flow, and Shareholder Value ," Journal of Marketing , 69 (3), 115 – 30.
Han Kyuhong , Mittal Vikas , Zhang Yan. (2017), " Relative Strategic Emphasis and Firm Idiosyncratic Risk: The Moderating Role of Relative Performance and Demand Instability ," Journal of Marketing , 81 (4), 25 – 44.
Haumann Till , Quaiser Benjamin , Wieseke Jan , Rese Mario. (2014), " Footprints in the Sands of Time: A Comparative Analysis of the Effectiveness of Customer Satisfaction and Company–Customer Identification over Time ," Journal of Marketing , 78 (6), 78 – 102.
Hill Matthew D. , Kelly Wayne G. , Lockhart Brandon G. , Van Ness Robert A.. (2013), " Determinants and Effects of Corporate Lobbying ," Financial Management , 42 (4), 931 – 57.
Hydock Chris , Paharia Neeru , Blair Sean. (2020), " Should Your Brand Pick a Side? How Market Share Determines the Impact of Corporate Political Advocacy ," Journal of Marketing Research , 57 (6), 1135 – 51.
Jindal Niket , McAlister Leigh. (2015), " The Impacts of Advertising Assets and R&D Assets on Reducing Bankruptcy Risk ," Marketing Science , 34 (4), 555 – 72.
Joseph John , Ocasio William. (2012), " Architecture, Attention, and Adaptation in the Multibusiness Firm: General Electric from 1951 to 2001 ," Strategic Management Journal , 33 (6), 39 – 61.
Joseph John , Wilson Alex J.. (2018), " The Growth of the Firm: An Attention-Based View ," Strategic Management Journal , 39 (6), 633 – 60.
Josephson Brett W. , Lee Ju-Yeon , Mariadoss Babu John , Johnson Jean L.. (2019), " Uncle Sam Rising: Performance Implications of Business-to-Government Relationships ," Journal of Marketing , 83 (1), 51 – 72.
Kang Karam. (2016), " Policy Influence and Private Returns from Lobbying in the Energy Sector ," Review of Economic Studies , 83 (1), 269 – 305.
Kashmiri Saim , Mahajan Vijay. (2017), "Values that Shape Marketing Decisions: Influence of Chief Executive Officers' Political Ideology on Innovation Propensity, Shareholder Value, and Risk , " Journal of Marketing Research , 54 (2), 260 – 78.
Kee Hiau Looi , Olarreaga Marcelo , Silva Peri. (2004), "Market Access for Sale: Latin America's Lobbying for U.S. Tariff Preferences," World Bank Working Paper , WPS3198.
Kurt Didem , Hulland John. (2013), " Aggressive Marketing Strategy Following Equity Offerings and Firm Value: The Role of Relative Strategic Flexibility ," Journal of Marketing , 77 (5), 57 – 74.
Laffont Jean-Jacque , Tirole Jean. (1993), " Cartelization by Regulation ," Journal of Regulatory Economics , 5 (2), 111 – 30.
Lambert Thomas. (2019), " Lobbying on Regulatory Enforcement Actions: Evidence from U.S. Commercial and Savings Banks ," Management Science , 65 (6), 2545 – 72.
Little Todd D. , Bovaird James A. , Widaman Keith F.. (2006), " On the Merits of Orthogonalizing Powered and Product Terms: Implications for Modeling Interactions Among Latent Variables ," Structural Equation Modeling , 13 (4), 497 – 519.
Lyon Thomas P. , Maxwell John W.. (2004), " Astroturf: Interest Group Lobbying and Corporate Strategy ," Journal of Economics & Management Strategy , 13 (4), 561 – 97.
Martin Kelly D. , Josephson Brett W. , Vadakkepatt Gautham G. , Johnson Jean J.. (2018) " Political Management, R&D, and Advertising Capital in the Pharmaceutical Industry: A Good Prognosis? " Journal of Marketing , 82 (3), 87 – 107.
Mittal Vikas , Anderson Eugene W. , Sayrak Akin , Tadikamalla Pandu. (2005), " Dual Emphasis on the Long-Term Financial Impact of Customer Satisfaction ," Marketing Science , 24 (4), 525 – 48.
Moorman Christine. (2020), " Brand Activism in a Political World ," Journal of Public Policy & Marketing , 39 (4), 388 – 92.
Ocasio William. (1997), " Towards an Attention-Based View of the Firm ," Strategic Management Journal , 18 (S1), 187 – 206.
Ocasio William. (2011), " Attention to Attention ," Organization Science , 22 (5), 1286 – 96.
Ocasio William , Joseph John. (2005), " An Attention-Based Theory of Strategy Formulation: Linking Micro- and Macroperspectives in Strategy Processes, " in Strategy Processes, Advances in Strategic Management , Vol. 22 , G. Cattani , ed. Bingley, UK : Emerald Group Publishing Limited , 39 – 61.
Ocasio William , Laamanen Toni , Vaara Eero. (2018), " Communication and Attention Dynamics: An Attention-Based View of Strategic Change ," Strategic Management Journal , 39 (1), 155 – 67.
Olson Mancur Jr.. (1971), The Logic of Collective Action: Public Goods and the Theory of Groups, Harvard Economic Studies. Cambridge, MA : Harvard University Press.
Otto Ashley S. , Szymanski David M. , Varadarajan Rajan. (2020), " Customer Satisfaction and Firm Performance: Insights from over a Quarter Century of Empirical Research ," Journal of the Academy of Marketing Science , 48 (3), 543 – 64.
Preacher Kristopher , Hayes Andrew F.. (2008), " Asymptotic and Resampling Strategies for Assessing and Comparing Indirect Effects in Multiple Mediator Models ," Behavior Research Methods , 40 (3), 879 – 91.
Rayfield Blake , Unsal Omer. (2018), " Product Recalls, Lobbying, and Firm Value ," Management Decision , 57 (3), 724 – 40.
Rego Lopo L. , Morgan Neil A. , Fornell Claes. (2013), " Reexamining the Market Share–Customer Satisfaction Relationship ," Journal of Marketing , 77 (5), 1 – 20.
Rhee Luke , Leonardi Paul M.. (2018), " Which Pathway to Good Ideas? An Attention-Based View of Innovation in Social Networks ," Strategic Management Journal , 39 (4), 1188 – 1215.
Richter Brian , Krislert Samphantharak , and Jeffrey F. Timmons (2009), " Lobbying and Taxes ," American Journal of Political Science , 53 (4), 893 – 909.
Ridge Jason W. , Ingram Amy , Hill Aaron D.. (2017), " Beyond Lobbying Expenditures: How Lobbying Breadth and Political Connectedness Affect Firm Outcomes ," Academy of Management Journal , 60 (3), 1138 – 63.
Rindfleisch Aric , Moorman Christine. (2003), " Interfirm Cooperation and Customer Orientation ," Journal of Marketing Research , 40 (4), 421 – 36.
Royne Marla B. , Myers Susan D.. (2008), " Recognizing Consumer Issues in DTC Pharmaceutical Advertising ," Journal of Consumer Affairs , 42 (1), 60 – 80.
Rust Roland T. , Christine Moorman , Dickson Peter R.. (2002), " Getting Return on Quality: Revenue Expansion, Cost Reduction, or Both? " Journal of Marketing , 66 (4), 7 – 24.
Saboo Alok R. , Sharma Amalesh , Chakravarty Anindita , Kumar V.. (2017), " Influencing Acquisition Performance in High-Technology Industries: The Role of Innovation and Relational Overlap ," Journal of Marketing Research , 54 (2), 219 – 38.
Sapienza Harry J. , de Clercq Dirk , Sandberg William R.. (2005), " Antecedents of International and Domestic Learning Effort ," Journal of Business Venturing , 20 (4), 437 – 57.
Seiders Kathleen , Flynn Andrea Godfrey , Nenkov Gergana Y.. (2021), " How Industries Use Direct-to-Public Persuasion in Policy Conflicts: Asymmetries in Public Voting Responses ," Journal of Marketing (published online September 30), doi.org/10.1177/00222429211007517.
Simon Herbert A. (1947), Administrative Behavior: A Study of Decision-Making Processes in Administrative Organizations. Chicago : Macmillan.
Smith Timothy , Keenan John. (2018), " Disclosing Corporate Lobbying," Harvard Law School Forum on Corporate Governance (April 2), https://corpgov.law.harvard.edu/2018/04/02/disclosing-corporate-lobbying/.
Srinivasan Raji , Lilien Gary L. , Rangaswamy Arvind. (2006), " The Emergence of Dominant Designs ," Journal of Marketing , 70 (2), 1 – 17.
Srinivasan Shuba , Vanhuele Marc , Pauwels Koen. (2010), " Mind-Set Metrics in Market Response Models: An Integrative Approach ," Journal of Marketing Research , 47 (4), 672 – 84.
Stahl Florian , Heitmann Mark , Lehmann Donald R. , Neslin Scott A.. (2012), " The Impact of Brand Equity on Customer Acquisition, Retentions, and Profit Margin ," Journal of Marketing , 76 (4), 44 – 63.
Stevens Robin , Moray Nathalie , Bruneel Johan , Clarysse Bart. (2015), " Attention Allocation to Multiple Goals: The Case of For-Profit Social Enterprises ," Strategic Management Journal , 36 (7), 1006 – 16.
Stigler Gary. (1971), " The Theory of Economic Regulation ," Bell Journal of Economics and Management Science , 2 (1), 3 – 21.
Tuli Kapil R. , Bharadwaj Sundar. (2009), " Customer Satisfaction and Stock Returns Risk ," Journal of Marketing , 73 (6), 184 – 97.
Unsal Omer , Kabir Hassan M. , Zirek Duygu. (2016), " Corporate Lobbying, CEO Political Ideology, and Firm Performance ," Journal of Corporate Finance , 38 , 126 – 49.
Wies Simone , Moorman Christine. (2015), " Going Public: How Stock Market Listing Changes Firm Innovation Behavior ," Journal of Marketing Research , 52 (5), 694 – 709.
Winterich Karen Page , Gangwar Manish , Grewal Rajdeep. (2018), " When Celebrities Count: Power Distance Beliefs and Celebrity Endorsements ," Journal of Marketing , 82 (3), 70 – 86.
Wooldridge Jeffrey M. (2010), Econometric Analysis of Cross Section and Panel Data , 2nd ed. Cambridge, MA : MIT Press.
Yadav Manjit S. , Prabhu Jaideep C. , Chandy Rajesh K.. (2007), " Managing the Future: CEO Attention and Innovation Outcomes ," Journal of Marketing , 71 (4), 84 – 101.
Yu Jisun , Engleman Rhonda M. , Van de Ven Andrew H.. (2005), " The Integration Journey: An Attention-Based View of the Merger and Acquisition Integration Process ," Organization Studies , 26 (10), 1501 – 28.
Yu Frank , Yu Xiaoyun. (2011), " Corporate Lobbying and Fraud Detection ," Journal of Financial and Quantitative Analysis , 46 (6), 1865 – 91.
Zhao Xinshu , Lynch John G. Jr. , Chen Qimei. (2010), " Reconsidering Baron and Kenny: Myths and Truths About Mediation Analysis ," Journal of Consumer Research , 37 (2), 197 – 206.
~~~~~~~~
By Gautham G. Vadakkepatt; Sandeep Arora; Kelly D. Martin and Neeru Paharia
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 111- Social Media, Influencers, and Adoption of an Eco-Friendly Product: Field Experiment Evidence from Rural China. By: Zhang, Wanqing; Chintagunta, Pradeep K.; Kalwani, Manohar U. Journal of Marketing. May2021, Vol. 85 Issue 3, p10-27. 18p. 1 Diagram, 6 Charts, 1 Graph. DOI: 10.1177/0022242920985784.
- Database:
- Business Source Complete
Social Media, Influencers, and Adoption of an Eco-Friendly Product: Field Experiment Evidence from Rural China
Can low-cost marketing tools that are used to enhance business performance also contribute to creating a better world? The authors investigate the role of online social media tools in alleviating customer (farmer) uncertainty and promoting the adoption of a new eco-friendly pesticide in rural China via a randomized controlled field experiment. The key finding is that even for a new product such as a pesticide, a low-cost social media support platform can effectively promote adoption. A combination of information from peers and from the firm on the platform facilitates learning about product features and alleviates uncertainty associated with product quality and appropriate product usage. Nevertheless, at the trial stage of the funnel, the platform underperforms the firm's customized one-on-one support because available information does not resolve uncertainty in supplier credibility and product authenticity. Having an influencer on the platform, albeit not an expert on this product, vouching for its credibility helps resolve this funnel-holdup problem. From a theoretical perspective, this paper provides suggestive evidence for referent influence and credibility signaling on social media platforms and consequences for new product trial. The authors also provide direct empirical evidence on how information facilitates learning, a phenomenon typically assumed to be present in studies estimating learning models.
Keywords: emerging markets; field experiment; innovation adoption; mobile marketing; social media; eco-friendly; sustainability
For decades, pesticides have been applied to protect crops and livestock from pest infestations, to increase crop yields, and to improve food production ([ 2]). However, pesticide use has also raised serious concerns about food safety, environmental protection, and sprayers' health (see Web Appendix W1). Every year, 200,000 people die because of toxic pesticides ([57]), and the [61] has classified 68 pesticides as potential carcinogens. Promoting the use of safe and green new pesticide technologies is critical to ecological security.
However, "getting a new idea adopted, even when it has obvious advantages, is difficult" ([55], p.1). In this study, we investigate how low-cost, online marketing tools can create a greener and healthier world by promoting the diffusion of a new pesticide technology in rural China. By implementing a field experiment including 34 villages and more than 700 farmers, we seek to understand whether a low-cost approach, based on a widely available social media platform, can be used to alleviate a major deterrent that hinders the adoption of a new technology: customer uncertainty.
Customers, especially those in emerging markets and in rural areas, face several types of uncertainty. These include uncertainty regarding ( 1) the authenticity of the new product and supplier credibility ([37]) given previous experiences with unscrupulous "fly-by-night" operators (e.g., the fake seed problem in China [[20]] and India [[27]]); ( 2) the "objective" quality of the product or the "match value" of the product to the potential user (e.g., [28]); and ( 3) how best to use or apply the technology to get the best outcomes from it ([29]; [38]). The traditional marketing literature has focused on how uncertainty is resolved vis-à-vis the second of these issues, since the first is usually not a concern and the third is usually not an issue in most categories studied.[ 5] One unique feature of our paper is that the technology and context we consider involve all three types of uncertainty.
The previous literature has explored several approaches to providing information to prospective users in rural markets so as to resolve their uncertainties. These approaches include self-experimentation and external information obtained either via social interaction with peers or from firms or governmental organizations ([24]).[ 6] Although recent literature has highlighted the role of online social media in consumer product adoption (e.g., [33]; [60]), its use as a support platform has not been explored in the literature in a business-to-business (B2B) setting in rural markets.[ 7] In such support platforms, consumers interact with each other online, and these interactions are supplemented by "broadcast" information whereby the firm addresses issues raised by consumers on the platform.
In the context of online social media, the literature has examined the roles of influencers in consumer marketing (e.g., [35]; [36]; [44]) and opinion leaders in business marketing ([37]; [42]; [52]). A second unique feature of our study is that we measure the additional impact on adoption, if any, of complementing the social media platform with an influencer.
Quantifying the impact of an influencer, however, is not straightforward when the technology is new and so there are no "expert" users of the technology who can serve as opinion leaders or early adopters to promote the product. Instead, we examine the role of eminent village personalities as influencers whose opinions are valued across a broad set of topics, even if they lack expertise specific to our product. This is a third unique aspect of the current study. To evaluate the cost–benefit trade-offs of using such platforms, we also compare the effects with ( 1) the results of a more traditional, firm-initiated one-on-one support approach ([22]) and ( 2) the results when the consumer tries to resolve uncertainties via self-experimentation.
The diffusion of pesticides involves both trial and ultimate adoption of the product. Further, the nature of the uncertainties facing potential users can be different in different stages. For example, while product authenticity and supplier credibility may be critical to get a user to try a product, ultimate adoption is unlikely unless the user can understand how best to use the product to obtain the greatest benefit from adopting it. A fourth distinguishing feature of our study is that we consider multiple behavioral outcomes along the adoption funnel: trial in the initial stages after introduction, cumulative trial behavior, and ultimate adoption.
We use a randomized controlled trial to measure the causal effects of marketing tools in changing behaviors (for a review, see [25]]; see also [ 7]]). We launched a trial program in three rural areas in two provinces in China, lasting 16 months from April 2017 to August 2018. First, we conducted field research to understand users' production processes with the new technology, the obstacles encountered, and how users make decisions given limited access to information and other constraints. With this knowledge, we designed a field experiment to quantify the effects of alternative information sources and marketing tools in the adoption process.
Our results reveal the following: ( 1) The social media platforms (both with and without an influencer) result in significantly higher adoption rates than the baseline self-experimentation condition. ( 2) However, when the platform is complemented by an influencer, adoption rates are significantly higher than when not using one. ( 3) The source of this difference lies in the differential trial rates across groups rather than in adoption rates conditional on trial. ( 4) The higher trial rates can be attributed to the influencer's early encouragement to try the product. ( 5) Traditional marketing with personalized one-on-one telephone support yields similar cumulative trial and adoption rates as the influencer-complemented social media platform. ( 6) However, personalized telephone support has a 35% lower return on investment (ROI) due to its higher associated costs. Thus, from a cost–benefit perspective, the social media support platform with an influencer is able to deliver comparable performance at a lower cost in our context.
Looking at the volume and nature of posts on the social media platforms, we find that the differential impact of the influencer in the early trial period is consistent with trust building to eliminate uncertainty regarding the product and supplier ([31]; [45]), rather than social learning about product features from noninfluencers. Further, by directly measuring the extent of learning about the various product features by those who tried the product across both social media conditions, we find comparable learning outcomes across the two conditions. Nevertheless, for certain features of the product, learning falls short of that achieved with personalized one-on-one telephone support by the firm. These results suggest that the information on the platforms facilitates learning by potential adopters, thereby providing direct evidence of the learning mechanism ([21]) often assumed in the marketing literature.
Our research contributes to the existing marketing discipline in the following ways. First, we show that low-cost social media tools can indeed facilitate adoption but also have some limitations. While our primary focus is on social media tools, we empirically compare and contrast the causal effects of multiple interventions on influencing adoption behavior in a controlled B2B environment. In contrast, previous literature has typically addressed one specific marketing tool, mostly in business-to-consumer (B2C) categories. Second, we disentangle the effect of a new type of influencer, the eminent village personality, from the effect of the social media platform by using a randomized controlled trial; we also provide suggestive evidence on the mechanism behind its influence. Third, our research is an early attempt to examine the marketing of eco-friendly products in developing areas. By bringing together these contributions, we believe, our research points to ways in which marketing can have a positive impact on the world around us.
From our interviews, we learned that when farmers are first exposed to a new technology, they need to decide whether to try it (we provide more insights in the field study in Web Appendix W2). At this stage, they face ( 1) uncertainty about the authenticity of the product and credibility of the supplier, and ( 2) uncertainty about the product's quality and its match value for their specific situations. These uncertainties are likely to hinder trial. If they decide to try the product, they need to make a decision on how to use the technology, a decision typical in B2B markets (e.g., [37]). At this stage, they face ( 3) uncertainty about how best to use the product most efficiently to get the maximum "bang for the buck." Their decisions on how to use the new technology will also affect their learning regarding product quality. In the final stage, according to the perceived value of the new product, customers decide whether to adopt the new product.
Resolving uncertainties and preventing misuse are therefore key to helping customers navigate the purchase funnel in B2B markets. These aims could be achieved by acquiring useful information. Normally, customers have three ways to obtain information about a new technology: through self-experimentation, from external sources including the innovating firm's support, or through social interactions with peers ([15]; e.g., [24]). Consequently, understanding how these different types of information affect trial, learning, and adoption behavior is critical.
Self-experimentation is, for those users who overcome the perceived risks and try a product, the most common way prospective customers learn about a new technology. Even experienced users, however, are often unable to use a technology appropriately, which, in turn, limits their ability to appreciate a product's true quality. When using a technology, users face a slew of potential factors that might affect production and so cannot attend to all of them: their attention is limited while the number of potentially important variables is large ([43]). Therefore, they can only pay attention to those variables they think are important, and they may ignore variables that are truly important to the production outcome (i.e., selective attention; [38]). In our case, the pesticide solution needs to be of the right consistency (not too much or too little added water), the holes of the sprayer should be as small as possible for better atomization results, and other factors. While self-experimentation is a useful benchmark, without the above knowledge, learning can be incomplete.
In emerging markets, information transmission is usually conveyed by in-person communication, such as discussion with neighbors (e.g., [24]; [62]) or training with agricultural extension agents (e.g., [12]; [13]). Such methods are labor and resource intensive. Information transmission via word of mouth takes time, leading to delayed adoption ([16]). Smartphone-based social media platforms provide a low-cost solution to enable peer effects by moving social interactions online, relaxing restrictions on time and distance required by face-to-face communication. This type of platform can also facilitate firm-customer communication through a "broadcast" function in the sense that every message posted in the online platform can be received by all its members at the same time. In this article, we propose using an online social media support platform to facilitate adoption.
In conjunction with the social media support platform, another marketing intervention we consider is the online influencer. The idea of influencers as catalysts in innovation diffusion has been a key idea in marketing (e.g., [23]; [55]). Empirical studies have provided evidence on the role of influencers (e.g., [34]; Goldenberg et al. 2009; [42]; [47]; [52]). Traditionally, influencers or opinion leaders are functionally defined as people who transmit new information about a product or idea to a group ([17]). For example, physicians who prescribe a new drug share usage experiences of the drug with their colleagues ([23]).
However, the technology in our context is completely new to the market, so none of the prospective users know about its existence, let alone have any experience or knowledge in using it. So, in this article, we explore the role of individuals that we refer to as "eminent village personalities." These influencers have two distinguishing characteristics. First, in the initial stages of the diffusion process, they do not possess any more information about the new product than the other prospective users have. The second characteristic is that notwithstanding their lack of unique knowledge regarding this particular product, their opinions on a variety of topics are nevertheless respected by the prospective users. In our fieldwork, we find that these influencers typically hold some village management responsibilities. This is consistent with observations of village leaders in developing countries who are frequently opinion leaders for a variety of topics, such as health, agriculture, and education ([55]). Eminent village personalities in our context bear some similarity to "market mavens" ([30]), who possess awareness and information on new products not only in a specific category but also across categories.
[ 9], p. 1) note that a consumer's "willingness to try new products and evaluations of these products are related inversely to the amount of perceived risk." With our new technology, farmers face uncertainty regarding the credibility of the supplier or authenticity of the product, the risk of poor performance, and potential crop damage.[ 8] In trying to lower this risk, consumers look at information that is intrinsic to the product, such as its attributes and functions. However, given the newness of our technology, such features are not informative; further, the supplier organization is unknown to the consumer. In such circumstances, the user seeks external information to provide risk reduction ([54]). At the pretrial stage, external information made available through our marketing interventions entirely entails communications from an influencer or an individual in the social network. Therefore, the likelihood of trial will depend on such interpersonal communications.
With influencers, the mechanism underlying the effect on trial, if any, could come from a variety of sources.[ 9] The first of these is referent influence, or, as [31] note, the belief that users want to be like the influencer and will be successful in doing so by behaving or believing as the influencer does. Second, according to cultural evolutionary theory, credibility-enhancing displays ([39]) demonstrate that the action of encouragement itself can enhance product credibility and encourage followers' cooperation. Even if the encouragement is not related to product features, it is credibility enhancing because dissemination of encouragement through the social media platform is costly to influencers: if the new product fails, the reputation of the influencers will be hurt. In the absence of the influencer, it will be more difficult to resolve the uncertainty regarding authenticity and product quality.
Peer effects occur when trial by peers, conveyed on the platform, may affect one's own utility from trial (e.g., [ 5]) as users may gain a sense of belonging and conformity by mimicking others' activities (e.g., [11]). Alternatively, if peers provide information regarding product features, this information might also encourage others to try the product by resolving uncertainty related to product authenticity and quality, and consequently motivate trial.
To adopt the product, a key consideration for potential consumers is the perception of value (e.g., [32]). Since the new technology is priced on par with pesticides currently on the market, price per se is unlikely to hinder adoption. The main route to resolving uncertainty related to quality and usage prior to adoption is learning. In the absence of any marketing interventions, part of the uncertainty might be resolved by learning through self-experimentation (e.g., [28])—if the farmers experience positive outcomes after trial, they might be more inclined to adopt the new technology. Such learning may be incomplete because a negative outcome may stem not from the poor quality or match value of the product but from incorrect usage. This is the third uncertainty discussed earlier.
Learning models (for a review, see [21]]) assume that users pay attention to their key production input variables and to the data from their experiences and those of others. Information obtained from each usage occasion provides a (noisy) signal of the true quality or match value of the product, so users ultimately are able to achieve their "productivity frontier," that is, extract the most from their production inputs (in our case, the new pesticide) once learning is complete ([38]). However, the productivity frontier is not guaranteed with a new technology,[10] as some part of the knowledge associated with applying the technology is tacit, meaning that it is "not feasibly embedded and neither codifiable nor readily transferable" (referred to as technological tacitness; [29]). If prospective customers cannot appreciate the true benefit of the new technology, they will abandon the product even after trial. Users on social media platforms can, however, learn from several sources. First, they can learn from communications from the firm (much like in traditional B2B one-on-one marketing). On the social media platform without an influencer, social learning ([51]) is also possible. The influencer per se, lacking the specific expertise required at this stage, may not be able to provide additional inputs beyond those associated with social learning.
In this study, we focus on a new nanotechnology-based pesticide formulation (for short, the nano-pesticide) invented by scientists in a nanotechnology research lab in China. This new technology has two main advantages over conventional pesticides: ( 1) it is environmentally friendly and safe for users since it does not use toxic organic solvents, and ( 2) its efficiency of application is improved. The new pesticide can be used in the same way as traditional pesticides with no requirement for extra application instruments, such as water barrels and sprayers, lowering the users' switching costs. While the efficacy and safety of the nano-pesticide have been established by many national and international third-party double-blind lab and field tests, the question facing the developers was whether farmers would actually try and then adopt the technology. Thus, while the pesticide awaited approvals from the government, the lab (hereinafter termed the "firm") was interested in studying low-cost ways of reaching its customers—the farmers.
In association with the firm, we launched a trial program that ran from April 2017 to August 2018. The aim was to recruit approximately 1,000 farming households, provide them with free samples of the new pesticide, and get them to try and then adopt (i.e., order at the market price) the new technology. The program included two pilot studies and one field experiment. The first pilot study was conducted from April 2017 to February 2018 in the Wuzhishan area of Hainan province, and the second one was conducted from April to June 2018 in Zhijiang, Hubei province. We recruited 352 farmers from 15 villages for the pilot studies. The pilot studies helped achieve three goals. First, they helped us understand individual farmers' production practices and potential problems encountered while using the new pesticide. From these findings, we designed standardized guidelines for providing customized instructions on using the new technology to address specific questions such as how to adjust important input dimensions if the pest control outcome was not satisfactory, how to customize the application method for certain crop species (e.g., rice, vegetables, cotton), and so forth. Second, as our experimental treatments involved social media support (and personalized telephone assistance for comparison), adequate training for service providers with systematic and standardized protocols was critical. Third, especially for the second pilot study, we replicated our experimental procedure in a place similar to the location of the real experiment but geographically far away. This helped us test for feasibility in the local environment; it also enhanced external validity in terms of repeatability of the program design and implementation. The main field experiment was conducted from June to August 2018 in Zaoyang and involved 34 villages and 702 farmers (one farmer per household).
Figure 1 shows the two levels of marketing interventions. The first level is the communication medium deployed to reach potential adopters: the social media support platform, the firm's traditional one-on-one personalized customer support by telephone, and the self-experimentation control group. The second level involves the deployment of eminent village personalities in the online environment. The sources of information corresponding to each treatment are shown in Table 1.
Graph: Figure 1. Experimental design.
Graph
Table 1. Information Sources by Treatments.
| Treatment | # Villages | # Farmers | Sources of Information |
|---|
| Self- Experimentation | Offline Social Interaction | Customized Expert Instruction | Online Social Interaction | Information from Influencers |
|---|
| Control | 8 | 148 | Y | Y | N | N | N |
| Firm's one-on-one support | 8 | 172 | Y | Y | Y | N | N |
| Social media | 8 | 121 | Y | Y | Y | Y | N |
| Social media with influencers | 10 | 202 | Y | Y | Y | Y | Y |
20022242920985784 Notes: Y = yes; N = no.
For villages in our two social media treatments (one with influencers and one without), we formed an independent online discussion group/platform on WeChat for each village. Only farmers in the same village were invited to the village's discussion group. On the online platform, people can discuss any topic they want, not necessarily only related to the new pesticide. They can raise questions about the new pesticide or agriculture in general, to be answered either by other farmers in the same discussion group or by the firm (represented by the researchers). Any information provided on the platform (i.e., from farmers and the firm) is available to all its members. Information on farmers' trial and adoption decisions were collected via follow-up surveys (described later).
In around half of the social media treatment villages, we introduced eminent village personalities as influencers. Consistent with research in this area (e.g., [50]; [52]), they were nominated by prospective users in the social network rather than appointed by the researchers. Influencers chosen usually had some responsibility related to village management. In Web Appendix W3, we provide a description of the influencers. Of the eight influencers, five are village officers or party secretaries, two are village women's directors, and three are directors of plant protection stations. Those positions hold responsibilities for villagers' daily lives and welfare, such as agricultural production, poverty reduction, birth control, and heath care. Eminent village personalities are respected by farmers because of their positions and professional credentials. However, they do not have expertise with our new product per se.
In the initial week of the experiment, we encouraged the influencers to post messages to motivate other farmers on their platforms to try the new pesticide. Although they were not required to, we expected these influencers to respond (albeit differentially) to our encouragement as they are relatively more advanced in their social networks and their village management duties entail helping farmers achieve better outcomes. Further, they might view this participation as a way of exercising thought leadership in the peer community. Thus, we believed that these eminent village personalities would view their roles as part of providing advice to members of the community on a variety of topics. Consequently, we did not provide them with any monetary incentives.[11]
One challenge facing researchers is how to establish a causal relationship between influencers and the adoption process. Most existent marketing research studying effects of influencers use observational (i.e., nonexperimental) data, where the effect of influencers is confounded with the effect of networks. Therefore, we decided to take an experimental approach instead. Our experimental design is inspired by peer encouragement designs, as in [26], [ 4], and [ 6]. In peer encouragement designs, peers are randomly assigned to an encouragement behavior, which can increase or decrease the chances of those peers engaging in specific behaviors. One can then observe how this encouragement induces endogenous behavior in the network and, consequently, measure how peer effects influence outcomes. Compared with using observational data, running experiments such as peer encouragement designs can effectively avoid the presence of confounding due to homophily and common external causes (e.g., [48]; [58]). In our context, we introduce influencers as an encouragement to induce endogenous online social interactions and, consequently, trial and adoption decisions for the new technology. Importantly, we have two other conditions that help us isolate the effects of the influencers: a condition with the social network but without the influencer and another with neither. Together, these three conditions make our experiment unique while allowing us to isolate the effects of the various interventions.
One-on-one support was provided to farmers by telephone during follow-up surveys starting two weeks after the start of the trial (this group received no interventions till then). In each survey, the support personnel reminded those who had not tried the product to do so and learned how the farmers were using the pesticide, to address any questions or concerns. All the information provided follows standardized instruction guidelines (see Table W4–1 in Web Appendix W4). Only the contacted farmers received customized instructions. This approach is expensive to implement because it involves two-way communication in which each farmer has the opportunity to engage with the service provider. Since the first interaction occurs during the first follow-up survey, we do not expect farmers in this treatment condition to be different from those in the control group (self-experimentation only) at the time of the first survey in terms of trial behavior.
Three specific features of the agricultural environment and farmer behavior have implications for the design and implementation of our field experiment. In our field experiment, we focused on farmers living in the same growing region to mitigate concerns regarding the impact of spatial heterogeneity on agricultural practice (e.g., [19]; [59]). Next, we required that all research tasks and information collection be completed within the same planting season to mitigate the effects of seasonality and unpredictable weather patterns ([25]). Third, our observational period was in line with the pest control cycle since being too late or too early could significantly affect a farmer's willingness to try or adopt a pesticide.
To investigate the impact of social media via a randomized controlled trial and to avoid contamination across treatment units, we needed to use independent, naturally formed, and geographically separated social networks, such as villages, as our observational units. Fortunately, Zaoyang is a large agricultural area with approximately 160 agriculture-based villages. With the endorsement of the local (official) agricultural department, we selected 34 villages that are similar in terms of geographical features, production conditions, income levels, culture, language, and other factors. Farmers in these areas plant rice as their main crop and face the same schedule for seeding, irrigation, pest control, and harvesting.
On the first day, the experiment began with an information session. Since a requirement of the village (government) officers was that all farmers in the village should have the opportunity to participate in the study, an announcement of the information session was made on the village's public address system the day before the session in all villages in this study. Village officers were not privy to any information regarding the specific treatment group that the village was in. Consequently, we do not face an issue of differential selection into the treatment groups. By focusing on those who then showed interest, the experiment helps control for heterogeneity along unobserved dimensions, such as the effort that users are willing to put into the new technology ([25]). Between 14 and 30 farmers showed up in each village to attend the information session conducted by the researchers. During the information session, the researchers gave a 15-minute introduction on the features of the new pesticide technology, including background information and the basic application methods. Participants were required to fill out a baseline survey to collect information on their demographics and farming practices. Extended surveys were then administered to a subset of farmers (described subsequently). See Figure W4-2 in the Web Appendix for a visual illustration of the research process.
After the baseline survey, free samples of the new pesticide, sufficient for 1,333 square meters of crops or vegetables, were distributed. Farmers in villages assigned to the two social media treatments were then invited to join a social media discussion group formed for their specific village by scanning a QR code using WeChat. During the information session of a village in the social media with influencers treatment condition, we asked farmers to nominate one person as the group leader (the eminent village personality) in the discussion group. The next two months were the observation period. We conducted three follow-up surveys every two weeks to collect information on each farmer's production inputs, outcomes if they tried the new pesticide, satisfaction levels, and so forth. During the last follow-up survey, researchers asked farmers whether they were willing to adopt this new technology, offering them an opportunity to order the product at the market price. We asked the farmers who decided to order it to put down 20% of the price of their order as collateral and provide their government-issued personal identification numbers. Since the pesticide could only be used in the next planting season, putting down a partial payment in advance can be seen as a strong commitment toward future use.
Of the 702 farmers, we omitted 59 from our final sample for the following reasons: ( 1) farmers decided to work in cities and did not farm that year, ( 2) farmers left the wrong contact numbers and were untraceable, or ( 3) farmers listed identical contact information. This left us with 643 farmers as individual units in our sample.
We also observe communications on WeChat, the social media platform. During the study period, farmers in the two social media treatments could freely communicate on their villages' social media discussion groups. Messages posted included photos or videos of pesticide application and other types of discussions: asking questions, describing usage experiences, replying to others' questions or comments, and sharing instructions given by the firm (for examples, see Figures W4–3 and W4–4 in the Web Appendix W4). To collect this type of information, we downloaded all messages posted on each village's WeChat group. We then manually categorized those messages into one of the following: ( 1) information format (e.g., video, audio, text, emoji); ( 2) information content reflected in seven different topics (e.g., new pesticide and trial program related, agricultural, nonagricultural); and ( 3) sentiment conveyed in a message: positive (e.g., praise for the product), neutral, negative (e.g., complaints).
Table W5-1 in the Web Appendix provides descriptive statistics of the main characteristics of the farmers. Approximately 66% of farmers in the main sample are men. The average farmer in the study was approximately 51 years old, had a middle school education, and has two family members who farm. The average percentage of farmers who own more than 3.3 acres of arable land is around 40%, in line with the trend of transforming from small-farm planters to larger planters in rural China ([56]). Approximately 20% of farmers are or used to be village officials. In Web Appendix Table W5-2, we provide balance checks across treatments and the control group. As illustrated there, only 4 of the 36 comparisons we consider are significant, which could be due to chance.
For the eminent village personalities, we found that the average age is approximately 46 years, indicating that they are younger than the average farmer in our sample (approximately 51 years). In addition, they are better educated (high school or above) than the average farmer (middle school to high school). In terms of other characteristics, such as the number of family members who farm and size of farmland, the influencers are similar to the entire sample (see Web Appendix W3).
A unique feature of our interventions is the use of social media platforms and the ability to study the nature of online interactions. Before we present our main findings, we first describe the volume, topics, and valence of the online conversations. Figure 2, Panel A, shows the evolution of social interactions on the social media platforms of villages in the two social media treatment platforms. We find that social media with influencers creates more messages than social media alone in terms of both the total number of messages (M = 136.2, SD = 39.0 vs. M = 68.0, SD = 38.8) and messages per person (M = 7.4, SD = 3.2 vs. M = 4.3, SD = 1.1; p <.01). In addition, we check to see whether the influencers are creating the bulk of the comments on the social media platform and find that the average percentage of messages created by them is just 8.8% (SD = 3.8%), which means that most of the discussion is being generated by other prospective users.
Graph: Figure 2. Changes of online social interactions over time.Notes: This figure shows the changes in the cumulative number of messages created by farmers on each village's social media platforms. Villages in the social media with influencers treatment and social media alone treatment are represented by dots and crosses respectively. The mean values of the social media with influencers treatment are denoted by bars and those of the social media alone treatment are shown as dotted bars. Panel A contains messages on all topics, Panel B contains messages on topics related to the new pesticide and trial program, and Panel C contains messages on unrelated topics.
Summary statistics of message topics are in Table 2. We find that farmers in the social media with influencers treatment are more willing to post evidence regarding their application of the pesticides than are farmers in social media alone treatment (see the first row). The former group of farmers is more active in discussing the new technology and topics related to the trial program (see the second through the sixth rows) than their counterparts in the social media alone treatment. Also, in the social media alone treatment, more of the posted messages concern topics unrelated to the new pesticide, such as news and jokes. A similar pattern may also be found over time in Figure 2, Panels B and C. This indicates that the eminent village personalities helped create an online discussion environment that fosters more active and relevant online social interactions. Finally, we also categorize the valence of the content of the posts. We found that the proportion of positive messages created in the social media with influencers treatment is.063, whereas that for the social media alone treatment is.012 (p <.10). Moreover, there is no difference between the two social media treatments in terms of neutral and negative messages.
Graph
Table 2. Summary Statistics for Topics Discussed in the Online Conversations.
| # of Messages per Farmer |
|---|
| Online Message Types | Statistics | Social Media with Influencers | Social Media | Overall |
|---|
| 1. Farmers show application evidence (photos or videos) | Mean | 1.985 | .838 | 1.476 |
| SD | 1.709 | 1.047 | 1.530 |
| 2. Farmers provide descriptions on efficacy of the new pesticide | Mean | .375 | .087 | .247 |
| SD | .411 | .141 | .345 |
| 3. Farmers raise questions or provide answers to inquiries about the new pesticide | Mean | .812 | .440 | .647 |
| SD | .877 | .559 | .756 |
| 4. Farmers raise questions or provide answers to inquiries about the trial program | Mean | 1.017 | .339 | .716 |
| SD | .367 | .407 | .510 |
| 5. Researchers answer farmers' questions related to the new pesticide | Mean | 1.155 | .573 | .896 |
| SD | 1.052 | .363 | .854 |
| 6. Farmers show trial program related photos or videos | Mean | .743 | .053 | .436 |
| SD | .813 | .099 | .692 |
| 7. Topics unrelated to the new pesticide | Mean | 1.305 | 1.981 | 1.605 |
| SD | .767 | 1.373 | 1.099 |
30022242920985784 Notes: We classified the various conversations into seven main topic areas. This table describes each of these topics and also provides descriptive statistics regarding each one.
The focus on social networks as the units of analysis constrained our ability to work with a large number of villages. We recognize that the small sample size (34 villages) makes identifying significant effects difficult. Even so, as we show next, we obtain statistically significant results as reflected in different parametric and nonparametric tests. In this section, we present results on our key behavioral outcomes: trial and adoption behaviors. In the next section we discuss the possible mechanisms behind the influence of different marketing interventions.
Table 3 shows regression results for dependent variables defined as the village-level ( 1) early (during the first two weeks) and final or cumulative (during all eight weeks) trial rates (proportions of sample farmers trying), ( 2) adoption rates (after eight weeks; proportions adopting), and ( 3) conditional adoption rates (ratio of adopters to triers). The independent variables are indicators for the various marketing interventions. The base condition is the self-experimentation control group. The differences across treatment groups are critical in our analysis. However, standard asymptotic tests can over-reject when the number of clusters is small (5 to 30). Hence, we adopt the cluster bootstrap-t procedure (see bottom panel of Table 3), which can provide asymptotic refinement ([18]).
Graph
Table 3. Effects of Marketing Treatments on Trial and Adoption Behaviors (Group-Level Analysis).
| A: Estimation Results (Ordinary Least Squares) |
|---|
| Dependent Variables | Early Trial Rate | Final Trial Rate | Adoption Rate | Conditional Adoption Rate |
|---|
| Social media with influencers | .251*** (.069) | .220*** (.044) | .338*** (.074) | .268*** (.083) |
| Social media | −.076 (.071) | −.055 (.068) | .178*** (.048) | .327*** (.069) |
| Firm's one-on-one support | .010 (.057) | .188*** (.059) | .307*** (.046) | .265*** (.055) |
| Constant | .391*** (.042) | .666*** (.034) | .244*** (.032) | .380*** (.054) |
| Observations | 34 | 34 | 34 | 34 |
| R-squared | .476 | .506 | .503 | .449 |
| Village-level clustered errors | Yes | Yes | Yes | Yes |
| B: Across-Treatment Coefficient Difference Tests (Wald and Wild Cluster Bootstrap t-Test) |
| Social media with influencers = Social media | 17.61*** | 17.75*** | 4.36** | .58 |
| Social media with influencers = Firm's one-on-one support | 13.32*** | .33 | .17 | 0 |
| Social media = Firm's one-on-one support | 1.61 | 1.15*** | 6.84** | 1.81 |
- 40022242920985784 *p <.1.
- 50022242920985784 **p <.05.
- 60022242920985784 ***p <.01.
- 70022242920985784 Notes: This table provides regression results for each of our outcome measures as dependent variables regressed on the treatment dummies. Constant represents the value for the control group. Robust standard errors are in parentheses.
In the second column of Table 3, the dependent variable is the early trial rate. We find that the social media platform that includes an influencer shows the highest mean early trial rate across villages. This indicates that when everyone is unfamiliar with the new technology, an eminent village personality can have a significant positive influence on trial behavior, overcoming uncertainty over authenticity and supplier credibility. We also see that the social media alone treatment does not outperform the control group, the self-experimentation condition. Since we provide the firm's identical online support in the form of "broadcast" messages (e.g., welcome message and reminders) during the first two weeks on every village's social media platform, such information by itself may not be powerful enough to overcome farmers' uncertainties. Note that the firm-initiated one-on-one customer support was only launched right after the second week of the experiment (during and after the first follow-up survey), meaning that there was no difference between the firm's one-on-one support treatment and the control group, as expected without any external information sources.
The third column shows results when the cumulative trial rates are the dependent variables. Interestingly, we find that social media with influencers again outperforms the social media alone treatment and the control group, with the social media alone treatment not showing a statistically significant difference from the control group. This confirms the previous finding that the social media platform alone cannot foster peer effects as expected, shaping our understanding of online social influence in the absence of other ways to overcome uncertainty regarding product authenticity and supplier credibility. The performance of the firm's one-on-one support treatment demonstrates the persuasive role of personalized firm-initiated support in overcoming the uncertainty regarding authenticity.
The fourth column uses the adoption rate as the dependent variable. Marketing interventions that use social media platforms, regardless of the presence of influencers, outperform the self-experimentation (and any offline social interactions) control condition. Further, the social media with influencers treatment shows a significantly higher adoption rate than the social media alone treatment. Nevertheless, these findings indicate potential learning effects from using the social media support platform on the final adoption behaviors of farmers. In addition, since all the marketing interventions involve some external support from the firm, this could also reflect the usefulness of the firm's assistance for B2B customers. Finally, the firm-initiated one-on-one support does as well as social media with influencers in terms of adoption.
From the last column, we see that all marketing interventions show significant effects in improving the adoption rate among farmers who tried the new technology (termed the conditional adoption rate for brevity in the following text), suggesting the presence of some forms of learning. Furthermore, the conditional adoption rates of the three marketing treatments are not significantly different from each other. This implies that once farmers in the social media alone treatment try the pesticide, the additional external information available significantly influences their adoption rate. Given their trial rate similar to that of the control group, this implies that some external information is required even after trial to convince farmers to adopt.[12]
We conducted a number of robustness checks of our findings. The first is that we used as dependent variables the raw numbers of farmers who tried and adopted the pesticide instead of using village-level proportions. The benefit of doing this is that it prevents the potential influence of heavier trial and adoption rates in villages with fewer sample farmers from biasing our results. Table W6–1 in Web Appendix W6 shows the results. We find that all the key differences are significant and consistent with the previous analysis. Next, to further assess statistical significance, we conducted a permutation test, a nonparametric method, as in [14]. Different from the traditional tests, which rely on asymptotic arguments along the cross-sectional dimension (here, the number of villages) to justify the normal approximation, permutation tests do not rely on asymptotic approximations. They are based on the fact that order statistics are sufficient and complete to produce critical values for test statistics. Since the comparisons between treatment groups and the control group are obviously significant even in asymptotic tests, we present only the results of the differences across various treatments from permutation tests in Table W6–2 in Web Appendix W6. We also provide a detailed illustration of this test in Web Appendix W6. We see that all the differences across treatment groups are significant, as in the previous regression analysis.
The literature involving social networks and adoption often leverages individual-level data despite the likely correlation in decisions across members of the network. Such studies include [50], [10], [ 8], and [ 6]. Most of these studies perform individual level analyses based on a conditional independence assumption: that is, conditional on being in each treatment group (and all the factors influencing trial and adoption therein), any unobservable factors influencing the individuals' decisions are independent across individuals. Thus, the treatment dummy encompasses those unobserved factors influencing behavior that might induce correlations across individuals. Under this rather strict assumption, we can run individual-level (logit) analyses by controlling for covariates and clustering standard errors. We present these results, which replicate our findings from the group-level analysis, in Table W6–3 in Web Appendix W6.
At the time of our baseline data collection, in addition to that survey, we were able to collect additional information from about 75% of our sample farmers. It was not possible to collect these data from all of them because the village officers imposed constraints on how long we could talk to them depending on the time of day that the specific farmer was interviewed (the officers did not want the farmers distracted from productive work). Thus participation in the extended survey can be assumed to be at random, and we verified this by comparing their characteristics to the full sample. A list of these variables and the specific questions are in Web Appendix W7.
In this section, we use the additional variables as covariates and moderators to study how customer heterogeneity affects adoption or moderate the effects of different marketing interventions on adoption.[13] Given space constraints, we focus here on the outcome that ultimately is of most interest, that is, adoption (see Web Appendix Table W6–4). Overall, in terms of model fit, including these variables seems to be adding no incremental explanatory power, looking at either the Akaike information criterion or the Bayesian information criterion. The main effects of most of the additional variables are estimated imprecisely and are not statistically different from zero under conventional levels. However, there is one exception: we find that users with larger farms are more likely to adopt the new technology than smaller farmers are. Further, our moderation analyses reveal a few patterns. First, the variable "farmers think the most important factor influencing their pesticide purchase decision is price" has a negative interaction with the social media treatments, suggesting that people who are more price sensitive will benefit less from the social media treatments (than the control group). Further, the interactions between the three treatments and the dummy "farmers think the most important factor influencing their pesticide purchase decision is user safety or health hazard" are positive and statistically significant for all marketing treatments. This, along with the negative main effect (indicating that those for whom health issues are important are least likely to adopt), suggests that all our marketing interventions are able to overcome the baseline lower level of adoption by such farmers. Finally, older farmers benefit significantly more from the firm's one-on-one customized assistance through the telephone than younger farmers; that is, the traditional communication method does better in assisting older customers.
In this section we provide some suggestive evidence for the potential mechanisms that might underlie our findings. Since the evidence is correlational, we cannot make causal claims. Nevertheless, we feel that the information provides insight into what might be going on.
During the first two weeks of the experiment, we find that the social media with influencers treatment outperforms the social media alone treatment. What might be the mechanism underlying this difference? One explanation is that the influencers build trust and enhance credibility through referent influence by providing words of encouragement and mentioning their own usage ([31]; [39]; [49]). An alternative explanation is effective online social learning, whereby peers (noninfluencers) provide information on their own trial and usage experience and directly affect a farmer's knowledge about the new product ([24]).
For suggestive evidence on online social learning, we look at what happened on the social media platform. First, we observe that influencers posted encouraging messages online (see Table W8–1 in Web Appendix W8) in the initial stage of the intervention (e.g., the influencer from a village posted, "Hello, my farmer friends! Recently the weather is good for pest control. Please use the new pesticide from.... Don't forget to post your application photos"). On average, while influencers posted 4.4 encouraging messages in the first two weeks, five influencers posted 2.4 messages regarding their own trial. Further, Table 4 shows a summary of messages generated by farmers (excluding influencers) during the first two weeks (before the first survey) and the other weeks of the experiment. We categorize messages into three types based on their content. The first type contains messages directly related to description of effectiveness of the new pesticide, the second one contains all the other messages related to the new pesticide and the trial program, and the third one contains messages on unrelated topics. We find that during the first two weeks of our experiment, few messages address the new pesticide or the trial program. The total average number of messages per village is less than three, and there is no significant difference between the social media with influencers treatment and the social media alone treatment. Meanwhile, the number of messages related to product efficacy is even smaller. Thus, online social learning ([51]) from peers (noninfluencers) is less likely to drive early trial behavior through enhancing the knowledge base of the new technology.
Graph
Table 4. Descriptive Statistics of Messages Posted by Noninfluencer Farmers on Social Media.
| Related Topics |
|---|
| Treatment | Stats | Product Effecta | All Othersb | Unrelated Topics |
|---|
| Weeks 1 and 2 | Social media with influencers | Mean | 1.70 | .70 | .30 |
| SD | 2.21 | .82 | .67 |
| Social media | Mean | .50 | 1.13 | .75 |
| SD | .76 | 2.80 | 2.12 |
| Weeks 3–8 | Social media with influencers | Mean | 9.60 | 38.20 | 11.10 |
| SD | 8.80 | 22.59 | 6.42 |
| Social media | Mean | 3.50 | 6.63 | 12.25 |
| SD | 4.72 | 6.72 | 17.30 |
80022242920985800 Notes: Product effectiveness and usage related messages posted by farmers. All the other new pesticide and trial program related messages, such as sending application photos. This table shows the number of product-related and other messages posted by the farmers on social media in the first two weeks and subsequent weeks of the intervention. We exclude messages from the eminent village personality influencers in constructing the table.
Taking these findings together, we see that the difference between the two conditions is not in terms of the online behavior of noninfluencers but in the social media with influencers treatment, reflecting encouragement and usage messages by the influencers. We take this as suggestive of the impact of eminent village personalities on initial trial behavior through trust building for the new technology, the supplier, and the trial program, thereby alleviating the first type of uncertainty referred to earlier in the paper. To show the correlation more formally, we estimated a logit model of noninfluencer villagers' early trial decisions (1 when a farmer tries the product, 0 otherwise) with the number of encouraging messages, whether the messages include the influencer's usage experiences, and other controls in the social media with influencers condition (see Table W8–2 in Web Appendix W8). We find that encouragement reflecting usage experience has a strong and significant positive correlation with early trial behavior.
From the third week on, the volume of the new technology-related discussions increased rapidly (see both Table 4 and Figure 2, Panel B). At the same time, the risk mitigation effect of the influencers diminished, as they posted fewer encouraging messages (see Table W8–1 in Web Appendix W8). To show some correlational evidence between trial behaviors and online activities, we look at the decisions of noninfluencer farmers to try the pesticide during the eight-week duration in the two social media support conditions as a function of the number of pesticide-related messages posted by noninfluencers. We find that the number of pesticide-related messages has a positive and statistically significant correlation with cumulative trial, but only in the social media with influencers condition (see Table W8–3 in Web Appendix W8).
These results suggest that the eminent village personalities facilitate diffusion in two ways. First, they can directly motivate the initial use of a new technology by mitigating risk in the early stages by engendering trust. Second, they act as catalysts for online social interactions by others, thereby indirectly influencing the diffusion process. We conjecture that the early trials due to influencer engagement results in these triers' contributing to online word of mouth. As the interactions among prospective users continue to evolve and propagate by themselves, those online interactions motivate more people to try the product. This larger base of users who have tried and experienced the product ultimately results in adoption and diffusion of the new technology. At the same time, for farmers who have already tried the pesticide, the presence of eminent village personalities appears to have no direct effect on final adoption behavior, as information at this stage comes from peers or from the firm's broadcast information. Indicative evidence for this can be seen in the similar conditional adoption rates across the two social media treatment conditions.
For the one-on-one customized telephone support condition, using the logs maintained by the support staff, we categorized the calls' focus as product-related, risk-related (harm to crops or product authenticity), or price- and purchase-related. At the end of the first two weeks (when the first set of calls occurred), a majority of the calls (61%) were related to risk, followed by product (38%). However, in subsequent weeks, the calls shifted to product-related issues (83%). Importantly, a logit analysis of individual trial on call duration (and controlling for demographics; that is, older farmers need longer call durations) shows a strong positive correlation between duration and trial behavior. In this case, using the terminology of [31], it appears that the firm's expert influence facilitates trial by the farmers.
In our conceptual underpinnings section, we noted that adoption requires farmers to resolve their uncertainties regarding quality and usage. In other words, they need to learn about the product's characteristics, such as effectiveness and harm to crops, and about its usage. To this end, in the final survey we asked farmers who tried the new pesticide to evaluate the benefits of the new technology and usefulness of the trial program. We asked the following questions: ( 1) "Comparing the new pesticide with the one you used before, do you think the new technology shows better results in: (i) pest control effectiveness, (ii) harm to crops, and (iii) pesticide usage reduction," and ( 2) "Do you agree/disagree with the following statement: the trial program helped me obtain useful information and knowledge about the new technology." Answers to both questions were measured on five-point scales, where 1 = "Strongly disagree" and 5 = "Strongly agree." This approach of measuring learning outcomes directly is a unique feature of our article, as most learning papers infer that learning occurred from trial/adoption outcomes (see [21], for a discussion).
We first calculate a treatment-level "satisfaction" measure as the proportion of farmers who provided a rating of 4 = "Agree" or 5 = "Strongly agree." Table 5 shows these results. We see that the learning measures are highest for the social media with influencers treatment (and the one-on-one support treatment) for the product features of "effectiveness of pest control" and "pesticide usage reduction." The product attribute "harm to crops" is harder to define, compared with the other two product attributes. Interestingly, we found that the firm's one-on-one support is more effective in promoting understanding and satisfaction for this more opaque attribute, indicating the superior nature of personal interactions between the firm and prospective users in communicating vague product features. The last two columns are related to the overall evaluation of the usefulness of the program. They show that the social media treatments outperform the control group, which is consistent with the results we observed in the previous sections. The traditional marketing approach also performs well.
Graph
Table 5. Farmers' Beliefs on Superiority of The New Pesticide Compared to Traditional Pesticides along Four Attributes: Evidence of Learning.
| b | Effectiveness | Harm to Crops | Usage Reduction | Program Usefulness |
|---|
| Treatment | Mean | SD | Mean | SD | Mean | SD | Mean | SD |
|---|
| Social media with influencers | .470 | .500 | .228 | .420 | .431 | .496 | .698 | .460 |
| Social media | .355 | .481 | .182 | .387 | .273 | .447 | .595 | .493 |
| Firm's one-on-one support | .523 | .501 | .477 | .501 | .552 | .499 | .605 | .490 |
| Control | .284 | .452 | .209 | .408 | .236 | .426 | .372 | .485 |
| Total | .420 | .494 | .281 | .450 | .389 | .488 | .579 | .494 |
| Across-Treatment Difference Tests |
| Social media with influencers = Social media | ** | — | *** | * |
| Social media with influencers = Firm's one-on-one support | — | *** | ** | * |
| Social media = Firm's one-on-one support | *** | *** | *** | — |
| Social media with influencers = Control | *** | — | *** | *** |
| Social media = Control | — | — | — | *** |
| Firm's one-on-one support = Control | *** | *** | *** | *** |
- 90022242920985800 *p <.1.
- 100022242920985800 **p <.05.
- 110022242920985800 ***p <.01.
- 120022242920985800 Notes: This table reports farmers' responses regarding various attributes of the pesticides and their agreement with whether the new pesticide is superior on these attributes. Numbers represent the proportions agreeing or strongly agreeing to the new pesticide's superiority.
To correlate the marketing interventions more formally with measured knowledge and learning about the new pesticide's attributes, Table 6 shows the results of an individual-level ordered logit regression (given the five-point measurement scales described earlier) with village-level clustered standard errors. Our sample focuses on farmers who tried the new technology during the experiment, and so our results should be interpreted with some caution since the farmers who tried it did so as a consequence of receiving different treatments. The dependent variables are the measures on beliefs (learning) regarding the three product benefits. In Models 1 and 3, where the dependent variables are learning measures on product effectiveness and usage reduction, all three marketing interventions have significantly higher results than the control group, meaning that the lower-cost social media tools help improve understanding of product efficacy and usage amount. Harm to crops is the most difficult product feature to learn in our context. Among the three marketing interventions, one-on-one support on the telephone is strongly correlated with improving learning outcomes for all attributes. We also find that individuals who have served as village officials are more likely to have a higher evaluation of product effectiveness and a reduction in usage amount, indicating some heterogeneity in appreciating the new technology.
Graph
Table 6. Effects of Marketing Treatments on Farmers' Beliefs About the New Pesticide (Individual-Level Ordered Logit).
| (1) | (2) | (3) |
|---|
| Dependent Variable | Effectiveness | Harm to Crops | Usage Reduction |
|---|
| Social media with influencers | .680** (.338) | .133 (.441) | .775** (.309) |
| Social media | .775*** (.264) | .161 (.479) | .664** (.301) |
| Firm's one-on-one support | .668** (.282) | 1.464*** (.428) | 1.207*** (.294) |
| Gender | .006 (.164) | .000 (.237) | .212 (.192) |
| Age | −.012 (.010) | .014 (.013) | −.010 (.010) |
| Current or previous village official | .598*** (.212) | .093 (.271) | .439* (.251) |
| Education level | −.110 (.157) | .246 (.169) | −.014 (.128) |
| Number of family members who farm | −.249*** (.083) | −.068 (.136) | −.120 (.125) |
| Owning arable land larger than 3.3 acres | −.016 (.201) | −.230 (.223) | −.126 (.164) |
| Observations | 494 | 494 | 494 |
| Village-level clustered error | Yes | Yes | Yes |
| Akaike information criterion | 1,299.57 | 813.03 | 123.55 |
| Bayesian information criterion | 1,354.20 | 867.66 | 1,285.18 |
- 130022242920985800 *p <.1.
- 140022242920985800 **p <.05.
- 150022242920985800 ***p <.01.
- 160022242920985800 Notes: This table presents results from an ordered logit analysis with farmer-level data where the dependent measure is the agreement on a five-point scale with whether the new pesticide is superior to existing pesticides along the three attributes of interest. Robust standard errors are in parentheses.
We also analyzed whether learning about product features mediated the effects of marketing interventions on adoption. Such an analysis is often used to investigate underlying behavioral mechanisms but is not common with survey data (for an exception, see [16]). A caveat in interpreting such an analysis is that it is not causal in nature. Further, the analysis conditions on trial, which is another potential limitation. By implementing a bootstrapping procedure described by [41], we found that learning about product efficacy mediated the effects of the three interventions on adoption. For the learning about crop damage prevention, we found no significant effects for the social media with influencers treatment and the social media alone treatment, whereas the effect of the firm's one-on-one support on adoption is mediated by this learning measure. Finally, the measure for usage reduction mediated the effects of all the three interventions on adoption, providing further support for our explanation that learning about product features might underlie our effects. These findings provide suggestive empirical evidence that learning was facilitated by our social media interventions, which could then have led to adoption by the farmers. Web Appendix W8 provides a detailed description of the analysis.
In emerging markets, the public sector or nongovernmental organizations play a major role in promoting innovations such as new farming technologies and cures for a variety of illnesses. For these organizations, social welfare is the primary goal rather than the earning of profits. Therefore, sustainability has been hard to achieve with such public programs ([46]). However, for private-sector organizations that strive to promote socially beneficial new products and create a better world, business sustainability and profitability are also paramount. Therefore, adopting the perspective of a company, we compare the costs of the different marketing interventions. To estimate the real-world scenario, we paid our research assistants who served as the firm's representatives in all three marketing interventions more than the market wage that the firm would have paid had it done the implementation. This way, we believe we paint a conservative picture of the costs associated with the various interventions. We calculate the ROI as total revenue (calculated on the basis of the market price) earned from each treatment minus its corresponding costs then divided by those costs. We find that social media with influencers is the most cost-efficient treatment, with the highest ROI value (3.45), followed by the social media alone treatment (2.45). In this instance, we did not need to pay the influencers; however, this may not be true in other contexts. Although the traditional marketing approach is effective in promoting trial behavior and learning performance, it is the most expensive (ROI = 1.91) among the three marketing interventions. In general, marketing interventions brought an averaged increase in adoption rate by 30%, compared with the control group. This may lead to a total increase of productivity by 6% and reduction in production costs of pesticides by 20% (both twice as large as for the control group). In the long run, the potential benefit to the environment and to people's health brought about by the new green pesticide technology is large. A detailed description of the cost analysis can be found in Web Appendix W9.
Many technologies, even those with obvious advantages, have not been widely adopted in developing and emerging markets, where they are urgently needed. Specifically, we investigated how to deploy online social media tools to alleviate customer uncertainty and to promote the adoption of a new nontoxic and eco-friendly pesticide in China. We contribute to the marketing literature in several unique ways. First, we consider three types of uncertainty facing potential adopters. These include ( 1) the authenticity of the new product and the supplier's credibility, ( 2) the "objective" quality or the "match value" of the product, and ( 3) how best to use or apply the technology in order to get the best outcomes from it. Second, we consider multiple behavioral outcomes along the adoption funnel, including trial in the initial stages after introduction, subsequent trial behavior, and ultimate adoption. Finally, we examine the role of a new type of influencer, eminent village personalities, whose opinions, like those of market mavens, are valued across a broad set of topics even if they lack expertise specific to the new product. Together, our research provides new insights on B2B marketing and on ways in which marketers can help create a "better world."
A key insight is that even in a rural setting of an erstwhile emerging market, social media influencers can offer an effective way of promoting the adoption of a "better" new B2B product. Influencers play a key role in dispelling concerns regarding the credibility of the new product early in the adoption cycle, a function critical for the eventual success of that medium. Ultimately, the combination of information sources on the platform promotes learning about the features of the new product and alleviates uncertainty associated with product quality and how best to use the new product in order to achieve the best outcomes from it. At the same time, the study also points to why the social media support platform by itself falls short of the performance of traditional B2B one-on-one marketing support in the purchase funnel.
We highlight three important implications for practitioners. First, social media can provide effective, low-cost means of reaching, communicating with, and convincing potential adopters of new technologies in otherwise difficult-to-reach markets that are nevertheless crucial for long-term success. Second, when promoting a new product in these markets, firms need to take into account the entire purchase funnel rather than focusing on just one specific action, such as trial or final purchase behavior. Indeed, a critical stumbling block is early in the process, where potential consumers may not engage because of concerns about the product's and the firm's credibility. Third, businesses, especially in the technology sector, have embraced the use of field experiments to guide their thinking and decision making about various marketing levers that might be used to grow their businesses. Our study provides evidence that even in rural environments, experiments might be a valuable tool for practitioners. Our experiment, conducted in collaboration with local governments, demonstrates how practitioners seeking better world outcomes can avoid higher-cost marketing interventions in favor of low-cost and readily available tools.
Our findings provide insights for managers and policy makers who aim to leverage marketing for doing good in the world. To do good, marketers need to convince consumers to adopt products that are good. Important barriers to such adoption are the uncertainties associated with the product and the inability to learn about the features and benefits of the product. We addressed these issues by understanding the entire process of adoption. When a product is brand new to the world, encouraging trial behavior among prospective users is key. During this stage, overcoming uncertainty about the new product's authenticity is paramount. We document that an influencer, albeit one not familiar with the new technology, works well in an online social media environment to encourage followers to try the new product. At the same time, traditional firm-initiated customized service and support has a significant effect in motivating trial behavior. Both of these approaches also lead to improved outcomes in the adoption stage but via different routes. On the social media platform, the information exchanged between the triers and the information provided via broadcast by the firm promote learning about specific benefits of the product as well as the best ways to use it. The more traditional marketing approach also accomplishes these objectives but via one-on-one communication between the firm and the potential customer. For marketing to do good, it also needs to do it at scale to have a wider impact. The social media platform with an influencer wins out here because it is more cost-effective than one-on-one marketing by the firm.
Our results also suggest that practitioners should think carefully about how to use social media most efficiently. Although research has documented its use for changing consumer behavior as it is a compelling marketing tool, it is not a panacea, and it requires careful management. Specifically, at the trial stage of the funnel we see the platform underperforming because it cannot, by itself, resolve uncertainty regarding supplier credibility and product authenticity. The lesson to be learned is that creating an online social media platform does not guarantee peer effects as desired. We also offer a solution to this funnel-holdup problem that ultimately propels diffusion of a new product or a new idea: an influencer who can vouch for the credibility of the product, and who tries the product and reports the trial on the platform. The influencers do not need to have expertise specific to the new product. They just need to be eminent persons in the offline world, such as village officers or women's directors in our context, whose opinions are respected and well perceived by others. We find that the presence of an influencer on the platform, relative to not having one, creates an online environment that fosters more product-relevant discussions among participants. Those discussions on products then motivate learning about the new product. In the absence of the online discussion emanating from trial, we are unlikely to find the level of success as in our experiment.[14] Thus, influencers who are well known only in an offline context can nevertheless help promote adoption through online tools. Without the presence of an influencer, a social media platform is only beneficial to people who are intrinsically more interested in trying the new product.
We provide three key implications for researchers. [31] among others, have described the different types of power that influence others, such as legitimate, reward, coercive, referent, expert, and information. From a theoretical perspective, our findings provide suggestive evidence for referent influence as the route through which the influencer plays a role in the adoption process. Different from the traditional view in marketing literature that influencers need to have relevant expertise about the new product in order to exert their influence, our findings point out that personalities who are eminent in offline contexts, although not having expertise or knowledge specific to the new product, can also have influence in promoting adoption through online tools. Such an effect is consistent with credibility signaling on social media and its consequences for new product trial. In situations with a large number of new products entering the market, we view this finding as potentially generalizable beyond our current context. Future research can further endeavor to establish the causal link in a more systematic manner.
A second implication of our findings is that we now have direct empirical evidence on how information on social media platforms facilitates learning and how this learning might potentially be a route to new product adoption. Although previous research has embraced the idea that resolving uncertainty via learning is key to product adoption and use, little direct evidence existed on the mechanism. Going further, our research also underscores the potential limitations of different information mechanisms to resolve the uncertainty. By measuring how learning occurs under each information mechanism for the different attributes or benefits associated with the new product, our research highlights the importance of understanding the linkage between information sources and their ability to resolve uncertainty. Our results point to the social media platform (even with an influencer) as not being able to fully communicate all the product features. Specifically, on the important dimension of crop damage, these interventions performed no better than the control. A key takeaway for researchers is to try to understand the specific barriers to learning associated with the social media platform and approaches to overcoming them. Alternatively, a hybrid approach in which the social media platform identifies specific users who need to receive the firm's one-on-one intervention may be useful to pursue. Understanding the efficacy and cost-effectiveness of such approaches may be a worthwhile future research endeavor.
A third implication is more methodological in nature. While time-consuming and resource heavy, field experiments enable quantitative marketing researchers to obtain mechanism-related insights that are otherwise difficult to obtain using only observational data. Such insights can then feed into building richer theoretical models of behavior. Given the importance of understanding behavior and the role of marketing in it, we encourage researchers to invest effort in the field and conduct more groundwork while engaging in such studies. The real world is too complicated to understand by just digging into existing data. "Through the accumulation of a set of small steps, each well thought out, carefully tested, and judiciously implemented" ([ 7], p. 15), we hope marketing can do better at doing good. We are excited at the possibility that field experiments testing a variety of different marketing tools can help fight poverty, disease, and pollution and contribute to the development of economies the world over.
Supplemental Material, sj-docx-1-jmx-10.1177_0022242920985784 - Social Media, Influencers, and Adoption of an Eco-Friendly Product: Field Experiment Evidence from Rural China
Supplemental Material, sj-docx-1-jmx-10.1177_0022242920985784 for Social Media, Influencers, and Adoption of an Eco-Friendly Product: Field Experiment Evidence from Rural China by Wanqing Zhang, Pradeep K. Chintagunta and Manohar U. Kalwani in Journal of Marketing
Footnotes 1 Jacob Goldenberg
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no external financial support for the research, authorship, and/or publication of this article other than from the institutions they belong to.
4 Online supplement: https://doi.org/10.1177/0022242920985784
5 While one can argue that dosage and compliance are important for pharmaceutical products that have been heavily researched, these can be learned by the physician over time.
6 Note that self-experimentation presupposes trial, implying that it is not relevant for resolving uncertainty regarding authenticity and supplier credibility.
7 We view the pesticide market as a B2B sale because the efficacy of the pesticide influences the farmers' livelihoods and because the uncertainties associated with its adoption are typical of B2B rather than B2C markets. Nevertheless, we acknowledge that some readers may view the product as being more like a consumer product.
8 In our case, since trial involved only the use of free samples, cost considerations are not relevant.
9 Our objective is to highlight possible mechanisms for the effect rather than to test for which of these accounts is supported by the data.
In reality, the productivity frontier is not guaranteed even when experienced users apply existing technologies. For example, [3] document that even experienced teachers do not apply the best teaching practices in secondary school classrooms.
Typically, online influencers are compensated for promoting products; in our case, there was neither the requirement that they post messages nor an incentive if they did so.
We urge caution in interpreting the results for the conditional adoption rates since these are not directly observed outcomes generated by the randomization, unlike the trial and adoption rates.
Because computing and interpreting interactions in nonlinear models are not as straightforward as with linear models ([1]; [40]), we conducted robustness checks as suggested in [53]. The results indicate that our inferences based on the interaction terms are robust.
This finding resembles the evidence documented by [36] in the context of tweeting, where the authors find that influential retweets can increase a show's viewership directly if they are informative, and indirectly by attracting new followers to the show's media company.
References Ai Chunrong, Norton Edward C. (2003), "Interaction Terms in Logit and Probit Models," Economics Letters, 80 (1), 123–29.
Alexandratos Nikos, Bruinsma Jelle. (2012), "World Agriculture: Towards 2030/2050," ESA Working Paper No. 12-03, Agricultural Development Economics Division, Food and Agriculture Organization of the United Nations.
Allen Joseph P., Lun Janetta. (2011), "An Interaction-Based Approach to Enhancing Secondary School Instruction and Student Achievement," Science, 333 (6045), 1034–37.
Aral Sinan, Walker Dylan. (2012), "Identifying Influential and Susceptible Members of Social Networks," Science, 337 (6092), 337–41.
Banerjee Abhijit. (1992), "A Simple Model of Herd Behavior," Quarterly Journal of Economics, 107 (3), 797–817.
Banerjee Abhijit, Chandrasekhar Arun G., Duflo Esther, Jackson Matthew O. (2013), "The Diffusion of Microfinance," Science, 341 (6144), 1236498.
Banerjee Abhijit, Duflo Esther. (2011), Poor Economics: A Radical Rethinking of the Way to Fight Global Poverty. New York: Public Affairs.
Beaman Lori, BenYishay Ariel, Fatch Paul, Magruder Jeremy, Mobarak Ahmed Mushfiq. (2016), "Making Networks Work for Policy," 3ie Impact Evaluation Report 43 (August).
Bearden William O., Shimp Terence A. (1982), "The Use of Extrinsic Cues to Facilitate Product Adoption," Journal of Marketing Research, 19 (2), 229–39.
BenYishay Ariel, Mushfiq Mobarak A. (2014), "Social Learning and Communication," Working Paper No. 20139, National Bureau of Economic Research.
Bikhchandani Sushil, Hirshleifer David, Welch Ivo. (1998), "Learning from the Behavior of Others: Conformity, Fads, and Informational Cascades," Journal of Economic Perspectives, 12 (3), 151–70.
Bindlish Vishva, Evenson Robert E. (1997), "The Impact of T&V Extension in Africa: The Experience of Kenya and Burkina Faso," The World Bank Research Observer, 12 (2), 183–201.
Birkhaeuser Dean, Evenson Robert E., Feder Gershon. (1991), "The Economic Impact of Agricultural Extension: A Review," Economic Development and Cultural Change, 39 (3), 607–50.
Bloom Nicholas, Eifert Benn, Mahajan Aprajit, McKenzie David, Roberts John. (2013), "Does Management Matter? Evidence from India," Quarterly Journal of Economics, 128 (1), 1–51.
Bollinger Bryan, Gillingham Kenneth. (2012), "Peer Effects in the Diffusion of Solar Photovoltaic Panels," Marketing Science, 31 (6), 900–12.
Bollinger Bryan, Gillingham Kenneth, Lamp Stefan, Tsvetanov Tsvetan. (2019), "Promotional Campaign Duration and Word-of-Mouth in Durable Good Adoption," working paper, Stern School of Business, New York University.
Burt Ronald S. (1999), "The Social Capital of Opinion Leaders," Annals of the American Academy of Political and Social Science, 566 (1), 37–54.
Cameron A. Colin, Gelbach Jonah B., Miller Douglas L. (2008), "Bootstrap-Based Improvements for Inference with Clustered Errors," Review of Economics and Statistics, 90 (3), 414–27.
Carter Michael R, Laajaj Rachid, Yang Dean. (2014), "Subsidies and the Persistence of Technology Adoption: Field Experimental Evidence from Mozambique," Working Paper No. 20465, National Bureau of Economic Research.
China Daily (2014), "Du Pont's Pioneer Combating Fake Seeds," http://www.chinadaily.com.cn/business/2014-05/23/content%5f17535514.html.
Ching Andrew, Erdem Tulin, Keane Michael P. (2013), "Learning Models: An Assessment of Progress, Challenges, and New Developments," Marketing Science, 32(6), 913–38.
Cohen Morris A., Agrawal Narendra, Agrawal Vipul. (2006), "Winning in the Aftermarket," Harvard Business Review, 84 (5), 129.
Coleman James, Katz Elihu, Menzel Herbert. (1957), "The Diffusion of an Innovation Among Physicians," Sociometry, 20 (4), 253–70.
Conley Timothy G., Udry Christopher R. (2010), "Learning About a New Technology: Pineapple in Ghana," American Economic Review, 100 (1), 35–69.
De Janvry Alain, Sadoulet Elisabeth, Suri Tavneet. (2017), "Field Experiments in Developing Country Agriculture," in Handbook of Economic Field Experiments, Vol. 2, Banerjee Abhijit Vinayak, Duflo Esther, eds. Amsterdam: Elsevier, 427–66.
Eckles Dean, Kizilcec Rene F., Bakshy Eytan. (2016), "Estimating Peer Effects in Networks with Peer Encouragement Designs," Proceedings of the National Academy of Sciences, 113 (27), 7316–22.
The Economic Times (2017), "Government to Take Action After ICAR Probe on Spurious Cotton Seeds," (October 23), https://economictimes.indiatimes.com/news/economy/agriculture/government-to-take-action-after-icar-probe-on-spurious-cotton-seeds/articleshow/61183480.cms.
Erdem Tulin, Keane Michael P. (1996), "Decision-Making Under Uncertainty: Capturing Dynamic Brand Choice Processes in Turbulent Consumer Goods Markets," Marketing Science, 15 (1), 1–20.
Evenson Robert E., Westphal Larry E. (1995), "Technological Change and Technology Strategy," in Handbook of Development Economics, Vol. 3, Behrman Jere, Srinivansan T. N., eds. Amsterdam: North-Holland, 2209–99.
Feick Lawrence F., Price Linda L. (1987), "The Market Maven: A Diffuser of Marketplace Information," Journal of Marketing, 51 (1), 83–97.
French John R, Raven Bertram, Cartwright Dorwin. (1959), "The Bases of Social Power," Classics of Organization Theory, 7, 311–20.
Gale Bradley T., Wood Robert Chapman. (1994), Managing Customer Value: Creating Quality and Service That Customers Can See. New York: Simon and Schuster.
Godes David, Mayzlin Dina. (2004), "Using Online Conversations to Study Word-of-Mouth Communication," Marketing Science, 23 (4), 545–60.
Godes David, Mayzlin Dina. (2009), "Firm-Created Word-of-Mouth Communication: Evidence from a Field Test," Marketing Science, 28 (4), 721–39.
Goldenberg Jacob, Han Sangman, Lehmann Donald R., Hong Jae Weon. (2009), "The Role of Hubs in the Adoption Process," Journal of Marketing, 73 (2), 1–13.
Gong Shiyang, Zhang Juanjuan, Zhao Ping, Jiang Xuping. (2017), "Tweeting as a Marketing Tool: A Field Experiment in the TV Industry," Journal of Marketing Research, 54 (6), 833–50.
Hada Mahima, Grewal Rajdeep, Lilien Gary L. (2014), "Supplier-Selected Referrals," Journal of Marketing, 78 (2), 34–51.
Hanna Rema, Mullainathan Sendhil, Schwartzstein Joshua. (2014), "Learning Through Noticing: Theory and Evidence from a Field Experiment," The Quarterly Journal of Economics, 129 (3), 1311–53.
Henrich Joseph. (2009), "The Evolution of Costly Displays, Cooperation and Religion: Credibility Enhancing Displays and Their Implications for Cultural Evolution," Evolution and Human Behavior, 30 (4), 244–60.
Hoetker Glenn. (2007), "The Use of Logit and Probit Models in Strategic Management Research: Critical Issues," Strategic Management Journal, 28 (4), 331–43.
Imai Kosuke, Keele Luke, Tingley Dustin. (2010), "A General Approach to Causal Mediation Analysis," Psychological Methods, 15 (4), 309.
Iyengar Raghuram, Bulte Christophe Van den, Valente Thomas W. (2011), "Opinion Leadership and Social Contagion in New Product Diffusion," Marketing Science, 30 (2), 195–212.
Kahneman Daniel. (1973), Attention and Effort. Englewood Cliffs, NJ: Prentice-Hall.
Katona Zsolt, Zubcsek Peter Pal, Sarvary Miklos. (2011), "Network Effects and Personal Influences: The Diffusion of an Online Social Network," Journal of Marketing Research, 48 (3), 425–43.
Kraft-Todd Gordon T., Bollinger Bryan, Gillingham Kenneth, Lamp Stefan, Rand David G. (2018), "Credibility-Enhancing Displays Promote the Provision of Non-normative Public Goods," Nature, 563 (7730), 245–48.
Kremer Michael, Miguel Edward. (2007), "The Illusion of Sustainability," The Quarterly Journal of Economics, 122 (3), 1007–65.
Libai Barak, Muller Eitan, Peres Renana. (2013), "Decomposing the Value of Word-of-Mouth Seeding Programs: Acceleration Versus Expansion," Journal of Marketing Research, 50 (2), 161–76.
Manski Charles F. (1993), "Identification of Endogenous Social Effects: The Reflection Problem," Review of Economic Studies, 60 (3), 531–42.
Merton Robert King, Merton Robert C. (1968), Social Theory and Social Structure. New York: Free Press.
Miller Grant, Mushfiq Mobarak A. (2014), "Learning About New Technologies Through Social Networks: Experimental Evidence on Nontraditional Stoves in Bangladesh," Marketing Science, 34 (4), 480–99.
Mobius Markus, Rosenblat Tanya. (2014), "Social Learning in Economics," Annual Review of Economics, 6 (1), 827–47.
Nair Harikesh S., Manchanda Puneet, Bhatia Tulikaa. (2010), "Asymmetric Social Interactions in Physician Prescription Behavior: The Role of Opinion Leaders," Journal of Marketing Research, 47 (5), 883–95.
Norton Edward C., Wang Hua, Ai Chunrong. (2004), "Computing Interaction Effects and Standard Errors in Logit and Probit Models," Stata Journal, 4 (2), 154–67.
Olson Jerry C., Jacoby Jacob. (1972), "Cue Utilization in the Quality Perception Process," in The Proceedings of the Third Annual Conference of the Association for Consumer Research, Venkatesan M., ed. Iowa City, IA: Association for Consumer Research, 167–79.
Rogers Everett M. (2003), Diffusion of Innovations. New York: The Free Press.
Schuman, Michael (2018), "China's Small Farms Are Fading. The World May Benefit," The New York Times (October 5), https://www.nytimes.com/2018/10/05/business/china-small-farms-urbanization.html.
Science and Technology News (2017), "UN: 200,000 Die Each Year from Pesticide Poisoning," Aljazeera News (March 8), https://www.aljazeera.com/news/2017/03/200000-die-year-pesticide-poisoning-170308140641105.html.
Shalizi Cosma Rohilla, Thomas Andrew C. (2011), "Homophily and Contagion Are Generically Confounded in Observational Social Network Studies," Sociological Methods and Research, 40 (2), 211–39.
Suri Tavneet. (2011), "Selection and Comparative Advantage in Technology Adoption," Econometrica, 79 (1), 159–209.
Trusov Michael, Bucklin Randolph E., Pauwels Koen. (2009), "Effects of Word-of-Mouth Versus Traditional Marketing: Findings from an Internet Social Networking Site," Journal of Marketing, 73 (5), 90–102.
U.S. Environmental Protection Agency (2018), "Chemicals Evaluated for Carcinogenic Potential Annual Cancer Report 2018," (accessed March 1, 2020), http://npic.orst.edu/chemicals%5fevaluated.pdf.
Yamauchi Futoshi. (2007), "Social Learning, Neighborhood Effects, and Investment in Human Capital: Evidence from Green-Revolution India," Journal of Development Economics, 83 (1), 37–62.
~~~~~~~~
By Wanqing Zhang; Pradeep K. Chintagunta and Manohar U. Kalwani
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 112- Tarred with the Same Brush? Advertising Share of Voice and Stock Price Synchronicity. By: Cheong, Chee S.; Hoffmann, Arvid O.I.; Zurbruegg, Ralf. Journal of Marketing. Nov2021, Vol. 85 Issue 6, p118-140. 23p. 1 Diagram, 6 Charts. DOI: 10.1177/00222429211001052.
- Database:
- Business Source Complete
Record: 113- The Concept of Authenticity: What It Means to Consumers. By: Nunes, Joseph C.; Ordanini, Andrea; Giambastiani, Gaia. Journal of Marketing. Jul2021, Vol. 85 Issue 4, p1-20. 20p. 2 Diagrams, 6 Charts. DOI: 10.1177/0022242921997081.
- Database:
- Business Source Complete
The Concept of Authenticity: What It Means to Consumers
The literature is filled with numerous idiosyncratic definitions of what it means for consumption to be authentic. The authors address the resulting conceptual ambiguity by reconceptualizing authenticity, defining it as a holistic consumer assessment determined by six component judgments (accuracy, connectedness, integrity, legitimacy, originality, and proficiency) whereby the role of each component can change according to the consumption context. This definition emerges from a two-stage, multimethod concept reconstruction process leveraging data from more than 3,000 consumers across no fewer than 17 types of consumption experiences. In stage one, the authors take a qualitative approach employing both in-depth interviews and surveys (one conducted on a nationally representative sample) to identify authenticity's six constituent components. The final components are based on themes emerging from consumer data that were integrated and reconciled with existing definitions in the literature. In stage two, quantitative analyses empirically estimate the six components and support the composite formative nature of the construct. The authors document how certain components contribute to assessments of authenticity differently across contexts; in addition, they show that authenticity has consumer-relevant downstream consequences while being conceptually distinct from consumer attitudes. Their findings offer practitioners direction regarding what to emphasize to convey authenticity to consumers.
Keywords: authenticity; consumption; composite construct; concept reconstruction; multimethod
Consumers crave authenticity—so much so that their quest for authenticity is considered "one of the cornerstones of contemporary marketing" ([13], p. 21). This has created an enormous challenge for the field, considering that marketing itself is typically considered inherently inauthentic ([18]). To overcome consumer cynicism, it has been argued that firms must learn to understand, manage, and excel at rendering authenticity ([28]). The critical question is: how? In an effort to better understand the role of authenticity in consumption, academic research on the topic has flourished in marketing and related fields during the past 20 years. An unintended consequence of this proliferation of research, however, has been the creation of numerous idiosyncratic definitions of what it means to be authentic, some loosely connected to one another and often capturing only a part of a complex phenomenon. What is clear is that "despite widespread agreement about authenticity's importance as a concept, no commonly accepted definition exists" ([ 7], p. 25).
The absence of a shared definition of authenticity is due, at least in part, to the fact that historically researchers have tended to develop definitions expressly for the particular context under investigation, be it advertising ([ 7]; [75]), brands ([39]; [53]), tourist sites ([31]), reality television ([64]), classic car ownership ([46]), brand extensions ([74]), employee service encounters ([71]), creative goods ([82]), culture ([83]), alcoholic beverages ([ 8]), social media influencers ([ 3]), or consumer products such as blue jeans and chocolate ([56]).
In addition, most marketing scholars who define what they mean by "authenticity" have ended up introducing a specific subtype of authenticity to the literature (see Tables A and B in the Web Appendix). This includes, but is not limited to, indexical authenticity ([31]), hyperauthenticity ([64]), constructed authenticity ([46]), brand extension authenticity ([74]), employee authenticity ([71]), brand authenticity ([67]; [53]; [25]; [16]), creative authenticity ([82]), cultural authenticity ([83]), and passionate authenticity ([ 3]). In general, the literature provides an insightful yet fragmented picture of what it means for consumption to be authentic. This fragmentation creates problems because when a single term such as "authenticity" acquires a variety of meanings, the inevitable result is conceptual ambiguity ([79]).
Conceptual ambiguity creates challenges for academics because the lack of shared meaning makes it difficult to develop coherent theory ([50]; [78]). What was evident 50 years ago is still true today; progress in consumer behavior research depends on standardized definitions, and this should be a priority for scholars in marketing ([43]). In addition, marketing researchers have made calls for an "increase in the conceptual rigor with which we define our constructs" ([86], p. 245) and for more attention to be given to "construct validity in general and more rigorous assessments of the measurement properties of constructs" ([40], p. 199). Conceptual ambiguity also creates challenges for practitioners who would benefit from clearer guidance regarding ways to enhance consumers' assessments of the authenticity of their offerings.
The goal of this research is to reconceptualize authenticity in marketing such that its definition provides a cohesive, comprehensive understanding of its meaning and specifies the concept's defining characteristics as well as the extent to which it is generalizable, or at least adaptive, across contexts ([62]). Critically, rather than introducing a new construct, the process followed here is one of concept reconstruction, which entails evaluating the state of the existing concept by reviewing its usage in prior research and using evidence from fieldwork to rework and revise how the construct is defined ([85]).
Anticipating our findings, we define authenticity as it pertains to consumption as follows: a holistic consumer assessment determined by six component judgments (accuracy, connectedness, integrity, legitimacy, originality, and proficiency) whereby the role of each component can change according to the consumption context. By placing our definition in the context of consumption, we provide a conceptual understanding of authenticity that is "indigenous" ([65], p. 2) and "organic" ([42], p. 1) to marketing. This definition of authenticity stems from our conceptualization of authenticity as a "composite formative construct," which is an entity defined entirely by its components ([12]). The reasoning for conceptualizing authenticity this way is as follows.
First, the rationale underlying why authenticity is conceived of as a formative rather than reflective construct is straightforward. Consumers make multiple judgments (e.g., Is this original? Is this accurate?) that correspond to the indicators we identify (e.g., originality, accuracy). These judgments are not interchangeable (e.g., originality is not a substitute for accuracy); it is only a combination of these judgments that jointly determines whether consumers consider a consumption experience more or less authentic. When changes in a construct depend on changes in its indicators, as opposed to vice versa, the construct is formative as opposed to reflective ([40]).
Second, authenticity is conceived of as a composite rather than causal formative construct. As a composite construct, authenticity is defined entirely by its components instead of existing on its own as a latent construct. This means "the indicators, as a group, jointly determine the conceptual and empirical meaning of the construct" ([40], p. 201) rather than simply providing a way of gauging the degree to which it is present ([12]). This is consistent with "authenticity" linguistically being a "dimension word," its specific meaning uncertain until one knows which of its dimensions are being discussed ([24]). The critical point is that, for consumers, authenticity derives its full meaning from its dimensions. In fact, when they describe what makes something authentic, they do so only through some combination of the six components we identify. An example of a composite formative construct in marketing is brand equity ([ 1]), a construct made up of brand awareness, brand associations, brand quality, brand loyalty, and other proprietary assets ([37]).
The fact that composite indicators can consist of dissimilar variables that do not need to have "conceptual unity" (i.e., direct correspondence) except in the loosest sense of the word ([11]) has two implications for our conceptualization. First, composite indicators need not covary, allowing for potential trade-offs between them (e.g., originality entails deviating from the mainstream, while legitimacy entails adhering to certain standards or norms). Second, composite indicators can contribute differently in different contexts (e.g., proficiency may be more important when assessing the authenticity of hedonic products than utilitarian products). Thus, our approach helps explain and integrate the fragmented literature in which researchers frequently have selected one or more components of authenticity to investigate separately in different contexts.
Our approach to conceptualizing authenticity does not entail defining a new construct per se, but instead entails concept reconstruction leading to the redefinition of an existing construct ([85]). Concept reconstruction is founded on the idea that, as building blocks of theory, concepts need to be tested, challenged, and revised to ensure clarity and consistency as the number of varied interpretations accumulate ([10]). The reconstruction process commences with a critical review of the concept's use in existing research. Therefore, we begin by identifying how authors have defined authenticity in the past, which reveals a range of meanings and little effort directed at pinpointing commonalities or ensuring consistency.
Next, successful concept reconstruction is predicated on taking a grounded theory approach ([76]). This means the concept is subjected to scrutiny through qualitative fieldwork examining data collected directly from consumers applying the concept within various empirical instances of the phenomenon ([10]). Thus, in stage one, we let consumers themselves describe how they form their assessments of authenticity, doing so across a wide variety of consumption contexts. We collected data across three qualitative studies. In Study 1 we conducted in-depth interviews focusing on a single relevant domain (music). In Study 2a, we conducted a survey of a nationally representative sample of consumers focusing on a variety of theory-driven consumption contexts. These contexts differ in the nature of the offering (product vs. service), its main consumption benefit (hedonic vs. utilitarian), the life cycle of the products (consumable vs. durable), and the extent of consumers' cocreation of value in the services (high vs. low coproduction). In Study 2b, we surveyed a more homogeneous sample whereby respondents self-generated the consumption contexts (included in the Web Appendix). Table 1 presents an overview of how these studies are distinct, yet complementary; the diversity in data collection methods is key to ensuring meaningful and robust results. The final step in stage one involves reconciling the themes drawn from consumer data, used to identify the component indicators that define authenticity, with themes drawn from existing definitions in the literature. Doing so marries evidence from fieldwork with existing theory about what it means for consumption to be authentic and provides the main conceptual outcome of the reconstruction process: the final set of authenticity's constituent components and their definitions (see Table 2).
Graph
Table 1. Overview of Qualitative Studies.
| Study 1Music Consumers | Study 2aRepresentative Sample | Study 2bExecutives |
|---|
| Sampling | Relevant | Representative | Emergent |
| Context(s) | Single, purposeful(music) | Multiple, theory-driven(e.g., craft beer, metro rides) | Multiple, self-generated(e.g., yoga classes, handbags) |
| Sample size | Small (N = 30) | Large (N = 1,011) | Small (N = 73) |
| Method | In-depth interviews | Survey, defined scenarios | Survey, open-ended, discussion |
| Data | Low breadth, high depth | High breadth, low depth | Medium breadth, medium depth |
| Purpose(s) | Preliminary identification of themes | Generalizability of themes across multiple theory-driven contexts and consumers | Generalizability of themes to broader set of consumption contexts |
Graph
Table 2. Components of Authenticity.
| Component | Definition |
|---|
| 1. Accuracy | The extent to which a provider is perceived as transparent in how it represents itself and its products and/or services and, thus, reliable in terms of what it conveys to customers. |
| 2. Connectedness | The extent to which a customer feels engaged, familiar with, and sometimes even transformed by a source and/or its offering. |
| 3. Integrity | The extent to which a provider is perceived as being intrinsically motivated, not acting out of its own financial interest, while acting autonomously and consistently over time. |
| 4. Legitimacy | The extent to which a product or service adheres to shared norms, standards, rules, or traditions present in the market. |
| 5. Originality | The extent to which a product or service stands out from mainstream offerings present in the market and does so without unnecessary embellishments. |
| 6. Proficiency | The extent to which a provider is perceived as properly skilled, exhibiting craftsmanship and/or expertise. |
In stage two, we augment stage one with a quantitative analysis in which we investigate the applicability of the component indicators identified across a variety of consumption contexts. We do so by using a scenario-based study that includes the same theory-driven variation in consumption contexts as Study 2a. Using partial least squares (PLS) modeling ([36]), we estimate authenticity as a composite formative construct, while documenting heterogeneity across contexts among select component indicators in terms of their impact on consumers' overall assessments. The data also allow us to investigate a broader nomological network that documents the effect of consumers' assessments of authenticity on downstream consequences relevant to marketers.
In summary, with very few exceptions (e.g., [86]), little work in marketing of which we are aware engages in a systematic process of concept reconstruction. Against this backdrop, we employ a unique process that takes a two-stage multimethod (qualitative and quantitative) approach, leveraging responses from more than 3,000 consumers across 17 theoretically driven, and nearly 200 self-generated, consumption experiences.
To anticipate the outcome of our endeavor, the fieldwork in stage one facilitates identifying six components of authenticity. As mentioned previously, these include accuracy, connectedness, integrity, legitimacy, originality, and proficiency. Accuracy refers to being transparent and reliable in what is conveyed to consumers. Connectedness describes feelings of engagement and at times a sense of transformation. Integrity means the source is intrinsically motivated, while acting autonomously and consistently. Legitimacy refers to conformity in terms of adhering to norms, standards, rules or traditions, while originality refers to a product or service standing out from the mainstream. Finally, proficiency refers to the display of skills, craftsmanship and/or expertise in the offering. We defined these components when reconciling our consumer data with the existing literature (see Table 2 and Figure 1), and they help clarify and structure the diverse themes collected from definitions in the extant literature (see Table A in the Web Appendix).
Graph: Figure 1. Convergence between definitions based on qualitative studies and core elements of definitions from prior literature.
Stage two provides empirical estimates of the six component indicators identified as contributing to consumers' assessments of authenticity across a wide range of consumption contexts, including some studied previously in the literature (e.g., chocolates, restaurants) and some that, at first glance, seem to be unlikely candidates for studying authenticity (e.g., washing machines, utility services). We find that when assessing authenticity, all six components contribute significantly across contexts, although proficiency routinely appears to be the most important and legitimacy the least important. In addition, judgments of proficiency appear more important for hedonic than utilitarian products, while judgments of legitimacy appear to matter for utilitarian but not hedonic products. It also appears integrity matters more for durable products than for consumable products. Turning to services, originality matters more for low- than for high-consumer-coproduction services, while conversely, legitimacy matters more for high- than for low-coproduction services. The results also show how authenticity relates to individuals' consumption intentions, as it is associated with information search, purchase intentions, and word of mouth, with consumer attitudes partially mediating this association. Notably, tests of discriminant validity support authenticity and attitudes as distinct constructs.
These findings are important for academics as they support conceptualizing authenticity as a composite formative construct with its constituent components contributing differently across different contexts. Conceptualizing authenticity in this way helps connect and make sense of the numerous idiosyncratic definitions in the literature. Knowing how different component judgments apply in different contexts is also important for practitioners who benefit from guidance about what to emphasize for different types of offerings to render authentic experiences.
As a first step in our conceptual reconstruction effort, we conducted a comprehensive review of the literature examining how authenticity has been defined in the past as it relates to consumption. Unlike a typical literature review, our goal here is to identify those characteristics of authenticity that prior researchers have considered essential, something that should be specified in any concept definition ([62]). The search for definitions involved collecting all articles including the words "authenticity" and/or "authentic" in the title, abstract, or as a keyword in the top 25 marketing journals as well as the top journals of related fields including management, organization, psychology, sociology, and anthropology (the Web Appendix provides details of the procedure we followed). This resulted in 436 articles from 61 journals (see Table C in the Web Appendix), including 153 (35%) from marketing. We further expanded our bibliographic search to include articles and books that were well cited in the articles collected. We scoured these articles for explicit and distinct definitions of what the authors meant when using the words "authenticity" or "authentic." Articles with definitions too far removed from consumers or tangential to consumption were excluded (e.g., articles examining authentic leadership, authentic emotions) as were articles that provided definitions taken directly from prior work already identified.
At the end of this process, we identified 63 distinct definitions from 46 different articles (see Table A in the Web Appendix). Forty-five of these definitions (71%) introduce specific subtypes of authenticity (e.g., "constructed authenticity," "moral authenticity") instead of defining authenticity more generally (see Table B in the Web Appendix). Within the marketing literature, 23 of 28 definitions (82%) were of subtypes. Put another way, only five articles in marketing ([ 9]; [39]; [54]; [56]; [67]) attempted to define what it means to be authentic in a general sense.
If we look more closely at these five articles, we find [39], p. 83) work on consumer culture and branding claims that to be authentic, brands must "be perceived as invented and disseminated by parties without an instrumental economic agenda, by people who are intrinsically motivated by their inherent value." [ 9], p. 839) are more general, concluding that "despite the multiplicity of terms and interpretations applied to authenticity, ultimately authenticity encapsulates what is genuine, real, and/or true" with each synonym left to the reader to interpret. [54], p. 99) focus on truth and define authenticity as "the degree to which an entity in one's environment (e.g., object, person, performance) is perceived to be true to or match up with something else." [56], p. 372) consider the act and not the outcome, stating that "authenticity describes a verification process—the evaluation of some truth or fact." In contrast, [67], p. 194) focus on individuals and not artifacts to "define authenticity as the degree to which personal identity is causally linked to individual behavior." It is interesting to note that while all of these are put forth as "general" definitions, each raises distinct features, and only a few loosely align with each other.
Surveying the full set of articles that we compiled, several key insights emerge. First, only about 11% of the 436 articles we identify as expressly involving authenticity provide what could be considered a definition, and fewer still utilize a single definition taken from prior literature. Instead, the vast majority either summarize how prior literature has varied in its interpretation of authenticity's meaning or presume the meaning of authenticity is understood. Second, most authors, when they do provide a definition, do so for a specialized subtype of authenticity that they create. Yet even researchers who label subtypes similarly (e.g., brand authenticity) tend to define them somewhat differently. For example, consider brand authenticity. [16], p. 42) "conceptualize brand authenticity as a judgment about the genuineness of a brand's image" while [67], p. 193) claim "an authentic brand is clear about what it stands for. It is a brand which positions itself from the inside out versus one that panders to the latest trend." [53], p. 203) go further, conceptualizing perceived brand authenticity as a multidimensional construct that includes the extent to which a brand is "true to its consumers" (similar to Cinelli and LeBoeuf) as one of four dimensions, and the extent to which "consumers perceive a brand to be faithful toward itself" (more similar to Schallehn, Burmann, and Riley) as another dimension. Overall, the key takeaway is that the literature presents an important but fragmented picture of what it means for consumption to be authentic. The evidence of conceptual ambiguity makes authenticity ripe for concept reconstruction, prompting the qualitative fieldwork we undertake to understand how consumers themselves interpret the concept.
The fieldwork in Study 1 examines consumers' interpretation of the meaning of authenticity using a critical case sampling method involving in-depth interviews. A critical case reflects a context that is particularly important because it permits logical generalization and maximum application of the information to other cases because "if it happens there, it will happen anywhere" ([59], p. 174). Music is especially well-suited as a critical case study because authenticity is a dominant topic studied extensively in this domain ([32]; [52]; [61]). Moreover, claims of being authentic play a major role in how music is marketed ([ 5]). In an effort to be methodical in capturing how consumers describe what makes the consumption of music authentic, we employed the repertory grid technique (RGT) to collect and analyze the data ([30]). The RGT is a cognitive mapping tool that has been utilized widely in consumer research, including successfully being employed to discover how consumers construct the concept of "customer experience quality" ([47]). Its elicitation procedure is particularly useful for explaining abstract terms from the point of view of respondents.
We selected 30 respondents prescreened for an interest in music to participate as informants. They varied in gender (50% female), age (50% under 30), and nationality (22 European, eight non-European). Informants were assigned to one of two interviewers. Each interview was conducted on an individual basis. During the initial elicitation stage, each informant provided a set of five artists whose music they felt particularly well-informed about. Of the 150 artists provided, 126 were unique (see Table D in the Web Appendix).
At the onset, we asked each informant to identify features important to them when they evaluate the artists and their output. For each informant, authenticity emerged as a relevant feature, reaffirming the concept's importance in the domain under investigation. Next, informants were presented with various combinations of three artists drawn from those they provided and were asked to identify ways in which the music of two is similar yet different from the third in terms of authenticity. The same procedure was applied using a common set of five artists popular at the time (Beyoncé, Pink, Katy Perry, Justin Bieber, and Enrique Iglesias), pretested to ensure heterogeneity in terms of assessments of authenticity (see "Study 1 Artist Pretest" details in the Web Appendix).
The resulting discussions produced a total of 225 meaningful descriptors (e.g., "follow their own style," "lyrics are introspective and personal") used to create opposing construct/contrast poles on a five-point bipolar scale in line with the RGT methodology. One scale, for example, was anchored by "follow their own style" and "follow current trends" (informant #18). Figure A in the Web Appendix includes a sample repertory grid of construct/contrast poles generated by a single informant (informant #10) for illustrative purposes. Using their own self-generated construct/contrast poles (which represent our elementary data), informants rated the work of the five artists they provided as well as the common set of five artists.
From the 225 construct/contrast poles, after a series of thorough coding and refinement efforts, we identified 243 themes associated with the meaning of authenticity in music. No systematic differences in terms of counts or types of themes emerged based on gender, age, or nationality. Using an approach to inductive research rooted in the literature ([29]), after a series of iterations and interpretation efforts, similar clusters of themes emerged that were grouped into six different categories. These six reflect the preliminary list of authenticity component judgments. These components (ultimately labeled accuracy, connectedness, integrity, legitimacy, originality, and proficiency) reflect the conceptual dimensions informants expressed when assessing the authenticity of an artist and their music. Table E in the Web Appendix provides a sample of themes derived from the construct/contrast poles and categorized within the different component judgments. Table F in the Web Appendix presents excerpts of what some respondents said makes music authentic, aligned with each of the six components.
Eliminating redundancies at the individual level, the 243 themes and their associated component judgment were reduced to 117 unique instantiations used to calculate counts at the informant level. While two informants did not raise any relevant themes/components, among the remaining 28 the component judgments raised by the greatest number of informants was integrity (86%), followed by originality (82%), connectedness (71%), proficiency (68%), accuracy (64%), and legitimacy (46%). It is important to highlight a few things about the six components identified.
First, inspecting the results more closely, there appears to be significant breadth in terms of what drives consumers' assessments of authenticity at the individual level. Informants did not simply use descriptors that were synonyms or different ways of representing the same thing. Instead, they point to different themes and ultimately different components of a broader concept. For example, one informant described authentic experiences as those produced by artists who "have the ability to create something new" (originality), "engage with fans" (connectedness), "write their own songs" (proficiency), and "are free to choose what to sing" (integrity). Another informant spoke of artists who "talk about what they have really experienced" (accuracy), have "respect for traditions and styles of a certain genre" (legitimacy), "are unique" (originality), "are consistent" (integrity), and "show a desire to have a dialogue with the listener" (connectedness). This breadth in dimensions that are not substitutes for one another is consistent with authenticity being a formative as opposed to a reflective construct.
Second, informants discussed authenticity as a multidimensional concept. More specifically, no informant described just one type of component judgment ("component" hereinafter). Among the 28 who raised relevant themes, only one person brought up themes associated with just two components, while 21% of respondents raised three, 39% four, 25% five, and 11% discussed all six. Further, we emphasize that informants did not describe what makes music authentic in ways unrelated to its components. No informant presented a single, rudimentary description; instead, they raised specific dimensions (components) of the concept consistent with authenticity being a "dimension word," its meaning depending on those dimensions discussed. This is indicative of how assessments of authenticity depend on a set of distinct judgments in the minds of consumers. This is compelling initial evidence that authenticity is a complex, composite construct that should be defined with "each dimension representing a unique content domain of the broader construct" ([63], p. 1). Taken together, these results are indicative of the formative composite nature of authenticity as a construct.
Recall that we ran a separate pretest that measured the perceived authenticity of the music of five common artists presented to respondents. We also had each informant rate these artists on their own set of five-point bipolar scales. We then ran a multidimensional scaling analysis using the average score at the component level for the music of each artist on the full set (Beyoncé, Pink, etc.) derived from these ratings (see Table G in the Web Appendix). Results reveal that artists' authenticity, measured as a composite score of respondents' evaluations, is in line with the evaluations these artists received in the pretest (see the horizontal positioning of the artists in Figure B in the Web Appendix). This provides initial support for the content validity of our findings. Moreover, some artists reached a similar level of authenticity with different contributions from the six component judgments. For example, Beyoncé and Pink received similarly high authenticity scores—the former due mainly to high levels of connectedness and integrity, the latter due to a high level of originality (see the vertical positioning of artists and components in Web Appendix Figure B). This suggests that the six themes are dissimilar in the way they contribute to consumers' assessments of authenticity, providing further support for the composite nature of the construct.
In summary, in Study 1 we identify six component judgments related to authenticity in music that define it as a concept in the minds of consumers. The results suggest that authenticity is a collection of judgments consumers make that capture different dimensions of the concept, providing preliminary evidence that authenticity is a composite formative construct. Recall that our concept reconstruction process is intended to incorporate consumer data collected using different qualitative methods (see Table 1), involving different populations of consumers, and covering a wide range of consumption contexts. In the next study, we broaden our inquiry substantially on all three fronts.
Study 2a builds on the results of Study 1 in three important ways. First, the method we employ has changed. In lieu of in-depth interviews, we survey consumers. Second, we expand and extend the sample significantly. We enlisted a nationally representative sample of consumers to ensure that we capture the views of a broad cross-section of the population, increasing the generalizability of our findings. Third, whereas Study 1 focuses exclusively on music, in Study 2a we include a wide variety of theoretically driven consumption contexts. This helps address potential limitations inherent in a single case-study approach (see [68]) and allows us to consider the extent to which consumer judgments might change across contexts.
One theory-driven distinction in contexts we draw is between products and services. This dichotomy is important because marketing exchanges that do not result in a transfer of ownership from seller to buyer (services) are fundamentally different from those that do (products) ([48]). Further, services differ from products in terms of other characteristics, including tangibility, perishability, the simultaneity of production and consumption, and the relative uniformity of consumers' experiences ([58]). These differences necessitate distinct consumer evaluation processes, which may lead to differences in what consumers emphasize when assessing authenticity.
Another theory-driven distinction we draw is between hedonic and utilitarian offerings. Hedonic offerings are characterized by an affective and sensory experience of pleasure and fun, whereas utilitarian offerings are more cognitively driven, instrumental, and goal-oriented, helping the consumer accomplish a functional or practical task ([21]). Hedonic and utilitarian motives comprise the "two basic reasons" consumers purchase products and services: ( 1) the affective gratification from sensory attributes and ( 2) instrumental reasons concerned with expectations of consequences ([ 6]). We anticipate that these different motives might also lead to differences in how consumers form their assessments of authenticity. Further, we distinguish between products on the basis of differences in their life cycle (durable vs. consumable). We also distinguish between services in line with the extent to which consumers typically contribute to the coproduction of the service experience (high vs. low coproduction; see [88]).
Leveraging these distinctions enables us to investigate how authenticity is thought of across several meaningful consumption contexts: hedonic products, hedonic services, utilitarian products, utilitarian services, consumable products, durable products, high-coproduction services, and low-coproduction services (see Table 3). We arrived at the specific experiences used in each context drawing on the results of a pretest that ensured that they differed significantly on all dimensions (for details, see "Study 2a Consumption Context Pretest" in the Web Appendix).
Graph
Table 3. Specific Consumption Contexts of Study 2a and Study 3.
| Study 2a |
|---|
| Product | Service |
|---|
| Hedonic | Utilitarian | Hedonic | Utilitarian |
|---|
| Durable | Consumable | Durable | Consumable | High Coproduction | Low Coproduction | High Coproduction | Low Coproduction |
|---|
| Sports car | Craft beer | Vacuum cleaner | Laundry detergent | Toy store | Movie | Banking services | Train/metro ride |
| Study 3 |
| Product | Service |
| Hedonic | Utilitarian | Hedonic | Utilitarian |
| Durable | Consumable | Durable | Consumable | HighCoproduction | LowCoproduction | HighCoproduction | LowCoproduction |
| Gaming console | Chocolates | Washing machine | Toilet paper | Restaurant dining | Sporting event | Health services | Utility services |
We recruited a sample of 1,011 U.S. citizens certified as nationally representative in terms of gender (52% female), age (Mage = 45 years), education, and ethnicity through Qualtrics International. Each respondent was presented with four products and/or services drawn randomly from the set and was asked to express, in their own words, the criteria they would use when forming a judgment of authenticity. Next, they were asked to describe a highly authentic experience of that particular product or service category (e.g., a sports car, a toy store). This resulted in a total of 4,044 statements, of which 3,955 (98%) were interpretable. Of these, 1,513 (37%) provided sound descriptors considered useful in terms of developing themes regarding authenticity. Statements considered nonuseful were categorized as such for multiple reasons. Consider the context of a sports car used in this study. Examples of nonuseful statements include those that are overly broad ("great features"), focus on individual preferences ("they are way too fast"), and identify specific examples instead of general criteria (e.g., "a Dodge Challenger"). Simply put, nonuseful responses fail to provide a solid-enough basis on which authenticity could or would be assessed. By avoiding making inferences based on ambiguous descriptors, we believe our conservative approach leads to results that are more robust.
In coding the data, we developed and refined a coding scheme in line with conventional grounded theory by "breaking down, examining, comparing, conceptualizing, and categorizing" the data ([76]). Each author individually read each response line by line in an attempt to identify key words or phrases belonging to, representing, or being an example of a more general meaning ([73]). The sizable amount of entries required several rounds of coding. Initially, all three authors coded the first 100 entries with the goal of identifying interpretational alignment. After conferring to resolve discrepancies in coding, each author separately coded the remaining entries. As coding progressed, each entry was compared with other entries appearing to belong to the same category to identify similar patterns of responses ([51]).
The coding procedure began initially by drawing from the themes and six components identified in Study 1, which were considered provisional, allowing for new themes and dimensions (components) to emerge. The 1,512 useful responses ultimately included 1,917 themes (23% of descriptions included more than one theme). Table H of the Web Appendix presents illustrative verbiage associated with each component in each context. We then calculated the reliability of our coding using the proportional reduction in loss approach ([66]), which involves the calculation of an overall proportion of interjudge agreement. Across consumption contexts, the interjudge reliability score ranged from.69 to.80, corresponding to an alpha of.92 to.97. After confirming reliability, we resolved all remaining inconsistencies through discussion.
In the end, we were able to categorize all 1,917 themes into a clearer and enriched version of the six components identified in Study 1. We did not find evidence that supported adding new components, despite the diverse sample population and varied contexts, thus corroborating the findings from the critical case approach. We want to highlight that respondents' descriptions were distributed fairly evenly across components as follows: accuracy (20%), connectedness (15%), integrity (12%), legitimacy (20%), originality (16%), and proficiency (17%). Despite the nature of the study, which encouraged short responses, nearly one in four of the useful statements included more than one theme. Notably, as in Study 1, respondents did not provide a single, rudimentary description of what would make a particular product or service category experience authentic. Instead, they raised specific dimensions (components) of the concept consistent with authenticity's meaning depending on the dimension(s) discussed. This, again, is consistent with authenticity emerging as a composite formative construct. Importantly, we find evidence for all six components in each consumption context provided, although some components appear to matter more in some contexts than others (e.g., accuracy was raised in 31% of descriptions for banking services, but only 8% for craft beer, while originality was raised in 36% of descriptions for craft beer, but only 3% for banking services). This type of variability across contexts prompted the additional investigation into heterogeneity we conduct in Study 3. We did not observe systematic differences in components based on gender or age, although integrity was raised somewhat more often by younger respondents.
In a separate study (Study 2b, included in the Web Appendix for brevity), we expanded the set of consumption contexts even further by asking 73 executives working on a master of business administration degree at a major business school to identify those types of experiences in which they see authenticity applying. This encouraged respondents to focus on the most accessible authentic experiences in memory. This was intended to ensure that our reconceptualization effort was not limited to evidence gathered only from predefined experiences in terms of context. Participants were then asked to list specific qualities of those experiences that contribute to making it more or less authentic. Respondents provided 192 distinct consumption experiences, ranging from "do it yourself" items purchased on Etsy to taking a yoga class, which were accompanied by a description of one or more qualities that impacted their perceptions of authenticity. The authors examined each description to determine whether any part aligned with the themes associated with one or more of the six components identified in Studies 1 and 2a. As in Study 2a, each author was also receptive to identifying any new themes or components observed in the data. From the 192 descriptions, we identified 234 descriptors that were categorized into themes and subsequently into one of the six components that emerged in Study 1 and were corroborated in Study 2a. Again, no evidence of new components surfaced from the data.
Integrating themes from Studies 1 and 2a (2b) and reconciling them with the extant literature forms the basis for the definitions of the six constituent components of authenticity (see Table 2). The key takeaway thus far is that, regardless of the method employed, the same six component judgments consistently seem to jointly determine consumers' assessments of authenticity. Next, in the final step in the process of concept reconstruction, we look for convergence among the themes identified in our qualitative studies and themes present in the prior literature.
Reconciling the findings from our consumer data with themes already present in prior literature is the final step in deriving precise definitions for the six components to form a holistic, general concept of authenticity as it pertains to consumption (see Figure 1). Moreover, consistent with the goal of concept reconstruction, it helps make sense of the existing fragmented knowledge on the topic by providing a conceptual scheme (our set of six component indicators) that can accommodate and integrate the disparate perspectives in the authenticity literature.
Here, we briefly summarize our reconciliation effort. Consider the originality component. The definition we propose integrates what consumers reported in stage one and converges with themes drawn from definitions of authenticity from prior literature. Consumers describe authentic experiences as those judged to be "unique," "different from the crowd," and "new and different" (for illustrative consumer quotes associated with each of the six components, see Tables F and H in the Web Appendix). Consumers also point out that distinguishing factors should involve the essence of an experience, free from unnecessary embellishments. These descriptors are consistent with descriptors derived from different definitions in the extant literature, such as "not being a copy or an imitation" ([31], p. 297), "breaking cultural canons" ([19], p. 976), and being "without contamination" ([83], p. 283). A judgment of originality also reflects a common view in the literature that assessments of authenticity involve comparisons with reference points that exist within space and time ([31]; [54]).
Similarly, consumers describe authentic experiences as those that adhere to some shared norms, standards, rules, or traditions present in the marketplace, which constitutes our definition of the component of legitimacy. They said things such as "I would look at safety standards," the extent to which it "meets state law requirements," and uses "all traditional methods." Similar ideas have emerged in prior literature and have been described as "commitments to tradition" ([ 8], p. 1008), being "true to its associated type (or category or genre)" ([15], p. 255), and corresponding "with a socially determined standard" ([54], p. 99). This component judgment of legitimacy, like originality, requires a comparison to external referents. Yet in contrast to originality, the central idea of legitimacy is compliance. As mentioned previously, consumer judgments of authenticity may depend on cues that create tension, for example, between conformity (legitimacy) and nonconformity (originality). This tension is made apparent in research by [17] that highlights how music critics prioritize one of two subtypes of authenticity proposed by [15]. In their analysis of Rolling Stone music reviews, they find that critics with lower cultural capital (domain-specific knowledge) prioritize adherence to expectations (i.e., type authenticity), whereas critics with more cultural capital prioritize the inherent motivation (i.e., moral authenticity) that often results in greater originality. Defining authenticity as a composite formative construct allows for different components (legitimacy, integrity, and originality, in this case) to have independent effects on assessments of authenticity. This reinforces the decision to conceptualize authenticity as a composite construct, as we do.
Another component judgment that emerges from the qualitative data is accuracy, defined as describing the extent to which a source is transparent in how it represents itself, while what is conveyed to consumers is reliable. Consumers describe authentic experiences as "delivering on all its claims," "super truthful, super direct," and getting "what you're expecting with no surprises." This meaning has historical roots in philosophy, with French philosopher Jean-Jacques Rousseau (1712–1778) arguing that to be authentic, one must be transparent ([34], p. 30). In line with this interpretation, prior literature has described something authentic as that which is "true" (e.g., [ 9]; [53]; [80]), "credible and convincing" ([14], p. 399), exhibiting "uncalculated honesty" ([87], p. 475), being verifiable ([55]), "providing fact-based information" ([ 3], p. 565), and communication matching "the actual state of affairs" ([54], p. 99). Like Rousseau, [45], p. 2) discuss a "consistency between an entity's internal values and its external expression." It is critical, we believe, to distinguish between being true to others, what we label "accuracy," and being true to oneself, which is captured in what we call "integrity."
We define integrity as the source of a consumption experience being perceived as intrinsically motivated and not acting out of one's own financial interest, while behaving autonomously and consistently over time. Here consumers describe a source being "free" to make its own choices, "not selling out," and "passionate" about its endeavors. Related meanings in the literature include authenticity being an "assessment of values" ([55], p. 3), and the notions of being "intrinsically motivated" as opposed to "extrinsically motivated" ([ 3], p. 565, [54], p. 100), and acting "without an instrumental economic agenda" ([39], p. 83). [81] characterizes authenticity as staying "true to oneself" to distinguish it from sincerity. These themes are, not surprisingly, raised in the literature frequently (see Figure 1), as they have direct ties to the etymology of the word. The Greek word authenteo is often translated as "to have full power" or "acting on one's own authority" in the sense of autonomy ([60], p. 48).
Another component referencing a quality of the source in assessing authenticity is proficiency, which we define as being properly skilled and exhibiting craftsmanship and/or expertise. Consumers referred to "quality production," employee "know-how," "mastery," "sophistication," and top-quality "employees and ingredients" as criteria they frequently use when assessing authenticity. Similar themes in the extant literature have included "using the appropriate techniques" ([15], p. 255) and "passion for craft and production excellence" ([ 8], p. 1008). Note the latter author may be seen as conflating passion, indicative of the source's values, with excellence in execution. We disentangle these ideas by separating integrity (motives) from proficiency (abilities). There is also a subtle distinction between legitimacy, which addresses whether the source adheres to prevailing standards, and proficiency, which addresses the use of "appropriate techniques" ([15], p. 255). What are "appropriate techniques"? Here, again, one might also observe consistency between the component judgments or tension, depending on the consumption context.
Finally, the component of connectedness is something that emerged quite emphatically among consumers in terms of a sense of intimacy, psychological and physical closeness, and, at times, a feeling of transformation. Consumers referenced feelings of "having a relationship" with the source, and sometimes feelings of "transcendence," as in "being taken to another place." This relates to another specific theme in the literature, albeit from a different vantage point, that of "interpersonal closeness with the customer" ([87], p. 473). Connectedness may also be linked to what has been described as "intrapersonal authenticity" ([46], p. 483), involving "both physical (i.e., relaxation, reinvigoration) and psychological (i.e., self-discovery, self-realization) aspects." In one sense, connectedness would seem to contribute to what has been referred to as "self-authenticating experiences" ([ 2]).
Reconciling themes from the consumer data with prior literature (Figure 1) reinforces the conclusion that the things that make a consumption experience authentic are jointly determined by a variety of distinct and loosely related judgments. These judgments constitute the component indicators of authenticity in our proposed composite construct. By describing how the six constituent components of authenticity integrate several disparate themes currently present in the literature, we are able to take stock of, digest, and synthesize the literature on authenticity in a way that allows consumer researchers to see the forest for the trees ([49]). In the next stage, we provide an empirical investigation of our conceptualization of authenticity as well as a broader framework in which consumer assessments of authenticity have important downstream consequences for marketing.
Study 3 accomplishes three things. First, it validates the set of six components identified in stage one as composite indicators of authenticity, doing so across different consumption contexts. Second, it documents heterogeneity in the roles played by different components. Evidence of heterogeneity is important theoretically, as it helps explain the fragmentation observed in the literature. It is also important practically, as it provides managers an indication of different "routes" to follow in marketing for different products and services that deliver authentic consumption experiences. Finally, Study 3 documents how consumers' assessments of authenticity have important downstream consequences while showing that authenticity is conceptually distinct from, albeit associated with, consumers' attitudes toward an offering.
As in Study 2a, we created different scenarios that varied the nature of the consumption experience (product vs. service), the main consumption benefit (hedonic vs. utilitarian), the life cycle for products (consumable vs. durable), and the extent of consumer cocreation of value for services (high vs. low coproduction). Note that life cycle and value cocreation are nested, and the design is not fully orthogonal. The specific experiences employed were derived from the results of two pretests: the first designed to elicit exemplars of hedonic and utilitarian products and experiences and the second to ensure they differed significantly on all dimensions of interest (details of both pretests are in the Web Appendix). Table 3 provides the specific experiences employed in each context. This study, though exploratory in nature, was preregistered on AsPredicted.org (https://aspredicted.org/blind.php?x=7wa87v).
The proposed model (illustrated in Figure 2) includes the focal construct of Authenticity, measured as a composite of the six formative indicators identified in stage one. Authenticity is expected to affect consumers' Behavioral Intentions, measured as a reflective construct, both directly and indirectly through Attitudes, also measured as a reflective construct. The model is then estimated across different groups of respondents randomly assigned to one of the eight consumption contexts (see Table 3).
Graph: Figure 2. Study 3: the empirical model.
Each respondent read what they believed was a review of a consumption experience in which clear evidence of all six components was present. First, respondents reported their overall assessment of the Authenticity of the offering reviewed using direct measures (two items). These measures were taken for instrumental purposes, to conduct the redundancy analysis prescribed as part of the method for evaluating composite formative measurement models in PLS (see [36]). Second, they rated the extent to which each of the six component judgments (e.g., originality, accuracy) contributed to their overall assessment of Authenticity (six items). The definition of each was provided. This type of measurement instrument was used to directly capture the association between the components and the composite in the minds of respondents. We chose the wording intentionally as a result of the purpose of the study and its concomitant design; the study was intended to validate the structure of the concept and the role of its components, rather than to develop a scale. Moreover, the scenarios provided clear evidence that all six components were present (e.g., in terms of originality each product or service was described as "one-of-a-kind"). We reasoned a priori that asking respondents simply to assess the presence of each component would have resulted in inflated and relatively homogenous ratings. Third, respondents reported their overall Attitude toward the product or service (three items) using conventional measures. Finally, they indicated their Behavioral Intentions with respect to their willingness to seek more information about the offering, purchase the offering, and share information about it via word of mouth (three items). We provide the complete stimuli along with the specific measurement instruments in the Web Appendix.
The target sample was 300 respondents per scenario, which, allowing for exclusions, we predicted to be adequate to ensure a minimum number of quality responses with which to estimate the model. We recruited 2,419 respondents (53% female; Mage = 35 years) on Prolific Academic. We excluded 491 respondents (20.3%) who failed an instructional manipulation check ([57]), leaving 1,928 usable responses. The number of respondents who failed the manipulation check did not vary across conditions (χ2( 7) = 9.44, p =.22).
We performed all of the analyses using PLS structural equation modeling in line with the guidelines proposed by [36] for its use. First and foremost, our modeling approach was motivated by the composite formative nature of our focal construct and the inclusion in the model of both reflective and formative constructs, which PLS can estimate appropriately. Second, our model is exploratory in its investigation of the components of authenticity, which fits with the epistemological premise and underlying features of a model tested using PLS. Third, the data come from scenarios (simulated authentic experiences) in which all cues were intended to signal high levels of authenticity; this led to the data distribution of all six components being left-skewed with limited variance. Being a nonparametric technique, PLS works well with nonnormal data. We estimated our model using the PLS consistent algorithm, which balances the tendency of PLS to magnify measurement loadings while downplaying structural relations ([22]). Notably, the key results and estimates obtained using the consistent PLS alogrithm (see Tables 4, 5 and 6) are similar to those obtained applying the traditional PLS analysis. The Web Appendix details further specifics of the PLS implementation.
Graph
Table 4. Study 3: Outer Weights for Components of Authenticity (Full Sample).
| Outer Weights | Bias-Corrected Boostrap CI |
|---|
| 2.5% | 97.5% |
|---|
| Accuracy → Authenticity | .217 | .143 | .291 |
| Connectedness → Authenticity | .237 | .163 | .311 |
| Integrity → Authenticity | .213 | .142 | .281 |
| Legitimacy → Authenticity | .078 | .019 | .138 |
| Originality → Authenticity | .248 | .172 | .323 |
| Proficiency → Authenticity | .330 | .249 | .403 |
10022242921997080 Notes: CI = confidence interval.
Graph
Table 5. Study 3: Outer Weights for Components of Authenticity.
| Aggregated |
|---|
| Authenticity | Nature of the Experience | Consumption Benefit |
|---|
| Products | Services | Hedonic | Utilitarian |
|---|
| b | p | | b | p | | b | p | | b | p | |
|---|
| Accuracy | .247 | .00 | | .165 | .00 | a | .279 | .00 | B | .187 | .00 | |
| Connectedness | .219 | .00 | A | .232 | .00 | B | .129 | .01 | a | .244 | .00 | A |
| Integrity | .196 | .00 | | .237 | .00 | C | .177 | .00 | a | .275 | .00 | |
| Legitimacy | .085 | .04 | a | .077 | .09 | abc | .049 | .31 | ab | .114 | .00 | abc |
| Originality | .238 | .00 | | .261 | .00 | | .267 | .00 | a | .246 | .00 | B |
| Proficiency | .349 | .00 | | .331 | .00 | A | .407* | .00 | A | .261* | .00 | C |
| Disaggregated |
| Products | Services |
| Hedonic | Utilitarian | Hedonic | Utilitarian |
| Authenticity | b | p | | b | p | | b | p | | b | p | |
| Accuracy | .291 | .00 | A | .247 | .00 | | .220 | .01 | | .129 | .07 | |
| Connectedness | .105 | .15 | b | .220 | .00 | | .157 | .04 | | .238 | .00 | |
| Integrity | .140 | .00 | b | .277 | .00 | | .215 | .00 | | .280 | .00 | A |
| Legitimacy | .028^ | .65 | ab | .139^ | .01 | | .094 | .19 | | .091 | .13 | ab |
| Originality | .239 | .00 | b | .252 | .00 | | .289 | .00 | | .233 | .00 | |
| Proficiency | .489** | .00 | B | .217** | .01 | | .342 | .00 | | .326 | .00 | B |
| Products | Services |
| Consumable | Durable | Low Coproduction | High Coproduction |
| Authenticity | b | p | | b | p | | b | p | | b | p | |
| Accuracy | .245 | .00 | | .257 | .00 | | .097 | .25 | b | .201 | .00 | |
| Connectedness | .212 | .00 | | .248 | .00 | | .236 | .01 | C | .197 | .00 | |
| Integrity | .124 | .06 | | .289 | .00 | | .276 | .00 | | .233 | .00 | |
| Legitimacy | .094 | .12 | | .049 | .37 | | –.033** | .63 | abc | .178** | .00 | |
| Originality | .268 | .00 | | .205 | .01 | | .411** | .00 | B | .116** | .12 | a |
| Proficiency | .392 | .00 | | .268 | .00 | | .271 | .00 | Ab | .385 | .00 | A |
- 30022242921997080 *p <.10.
- 40022242921997080 **p <.05.
- 50022242921997080 ^Indicates that legitimacy matters for utilitarian but not hedonic products (b =.028, p =.65, factor loading <.5).
- 60022242921997080 Notes: Differences across contexts. Within contexts, components with a capital letter differ from components with the same letter that is lowercase at p <.01.
Graph
Table 6. Study 3: Path Coefficients.
| Aggregated |
|---|
| Products | Services | Hedonic | Utilitarian |
|---|
| b | p | b | p | b | p | b | p |
|---|
| Authenticity → Behavioral Intentions | .626 | .00 | .666 | .00 | .584 | .00 | .680 | .00 |
| (Total effect) |
| Authenticity → Behavioral Intentions | .442 | .00 | .381 | .00 | .334 | .00 | .534 | .00 |
| (Direct effect) |
| Authenticity → Attitudes → Behavioral Intentions (Indirect effect) | .184 | .00 | .285 | .00 | .250 | .00 | .146 | .00 |
| Disaggregated |
| | Products | Services |
| | Hedonic | Utilitarian | Hedonic | Utilitarian |
| | b | p | b | p | b | p | b | p |
| Authenticity → Behavioral Intentions | .606 | .00 | .616 | .00 | .567 | .02 | .740 | .00 |
| (Total effect) |
| Authenticity → Behavioral Intentions | .407 | .00 | .479 | .00 | .237 | .00 | .537 | .00 |
| (Direct effect) |
| Authenticity → Attitudes → Behavioral Intentions (Indirect effect) | .199 | .00 | .137 | .01 | .330 | .00 | .202 | .00 |
| | Products | Services |
| | Consumable | Durable | Low Coproduction | High Coproduction |
| | b | p | b | p | b | p | b | p |
| Authenticity → Behavioral Intentions | .644 | .00 | .584 | .00 | .724 | .00 | .599 | .00 |
| (Total effect) |
| Authenticity → Behavioral Intentions | .436 | .00 | .384 | .00 | .410 | .00 | .333 | .00 |
| (Direct effect) |
| Authenticity → Attitudes → Behavioral Intentions | .207 | .00 | .200 | .00 | .314 | .00 | .266 | .00 |
| (Indirect effect) |
Following [36] guidelines, the measurement instruments used for Attitudes and Behavioral Intentions appear good, with outer loadings that are never below.60 (and in the vast majority of cases, >.70). Both constructs always exhibit values of composite reliability and average variance extracted (AVE) greater than threshold levels of.70 and.50, respectively, across the different consumption contexts (see Table I in the Web Appendix).
First, we conducted a redundancy analysis inspecting the correlation between the composite measure of Authenticity based on the six components and two highly correlated general measures of authenticity (reflective items). Results reveal that the six indicators exhibit a fair level of convergent validity. The correlation with the overall measure ranges from.65 to.77 across consumption contexts (see Table J in the Web Appendix) and in the majority of cases is above the critical value of.70. Second, an analysis of the variance influence factors of the six indicators excludes collinearity issues in the estimation of the composite measure of Authenticity. Despite being fairly correlated (average interitem correlation =.45) due in part to study design choices, the greatest variance inflation factor across all contexts is 2.02, which is below the critical value of 3.0 (see Table K in the Web Appendix). Third, given that our composite includes more than four indicators, we can empirically test whether our data fit better with a formative versus reflective measurement model. A confirmatory tetrad analysis ([33]) reveals that five out of the nine nonvanishing tetrads in the full sample are different from zero. Across contexts, the same analysis reveals two nonzero nonvanishing tetrads for products and four for services, supporting the formative nature of our construct.
We ensure discriminant validity between our composite construct of authenticity and the reflective measures of attitude and behavioral intentions in two ways. First, according to the [27] criterion, the square root of AVE values should be greater than the bivariate correlations for all of the constructs under investigation. In all cases (see Table L in the Web Appendix), each set of items loads more strongly with their correspondent construct than with any other construct in the model, suggesting that Attitudes and Behavioral Intentions measures are empirically distinct from those belonging to our composite measure of Authenticity and also distinct from one another. Note, however, that the traditional Fornell and Larcker criterion uses AVE as a baseline for comparison, and for formative constructs the AVE is not relevant. Therefore, we also apply the approach from [41], according to which discriminant validity for a formative construct is established when the average intraconstruct correlation (among the six component indicators for Authenticity in this case) is greater than the interconstruct correlations among its items and those of other constructs involved in the model. We find that the average intraconstruct correlation is greater than all interclass correlations across all contexts (differences range from.03 to.15; see Table M in the Web Appendix), thus ensuring the discriminant validity of the composite Authenticity construct.
The relationships between the component indicators and the composite Authenticity construct are represented by the outer weights estimated by the PLS consistent analysis (see Table 4, full sample). The outer weights are indicative of the relative role played by each of the six proposed indicators for the Authenticity composite construct. Examining the outer weights enables us to determine whether the full set of six indicators is appropriate when conceptualizing authenticity, as well as if and how an individual component's role might differ within and across contexts. What is clear from Table 4 is that all six, when present as they are in this case, can be considered valid indicators of the composite measure of Authenticity.
In most of the consumption contexts (see Table 5), one observes that Proficiency appears to be the most important component while Legitimacy is routinely the least important, and the other four typically fall somewhere in between. When outer weights are compared within each context for statistical differences ([77]), the six components seem to play a similar role in some contexts (e.g., utilitarian products, hedonic services) and a more dissimilar role in others (e.g., hedonic products). In certain cases, such as for low-coproduction services, components seem to inform the perceptions of authenticity in an idiosyncratic way (e.g., originality is more important than proficiency, accuracy, and legitimacy).
Comparing outer weights across contexts, certain components appear as more or less important to consumers' assessment of Authenticity, the specifics of which are worth examining more closely as they are highly informative and particularly relevant for practitioners. First, for products specifically, we find that Proficiency is more important to assessments of authenticity for hedonic than for utilitarian products (b =.489 > b =.217, p =.02). It seems the skill and artisanship of the provider matter more when the primary benefit is affective or sensory pleasure. In addition, Legitimacy matters for utilitarian products but does not matter for hedonic products (b =.028, p =.65, factor loading <.5). It appears that adhering to standards is unimportant when assessing authenticity unless the product is instrumental and intended to help reach a practical goal. It also appears that Integrity plays a more important role for durable than consumable products (b =.289 > b =.124, p =.11). Although only directional, this result suggests the source's motives may matter more for authenticity assessments when a product is expected to have a longer life cycle.
Turning our attention to services, we find that Originality matters more when assessing the authenticity of low- (vs. high-) coproduction services (b =.411 > b =.116, p <.01). It may be that when consumers contribute less in the cocreation of value, and thus perhaps fail to personalize the experience, the distinctiveness of the offering itself contributes more to authenticity assessments. Conversely, Legitimacy matters more when assessing the authenticity of high- (vs. low-) coproduction services (b =.178 > b = −.033, p =.02). Consumers more involved in cocreating value appear to care more about whether an offering adheres to specific standards. We consider the implications of these findings shortly. We should also mention that, at a general level, we find no clear evidence of heterogeneity based on gender or age.
The key takeaway here is these results support the intuition that authenticity should be conceptualized as a composite construct, and all indicators can matter, yet certain indicators matter more or less and sometimes not at all depending on the particular consumption context. Understanding the heterogeneity amongst the indicators forms one of the central managerial contributions of this research, a discernible set of prescriptions for marketers that we discuss in more detail subsequently.
We next turn to the structural part of the model, the relationships involving Authenticity and the other constructs. Recall that the framework in Figure 2 includes the extent to which Authenticity predicts Attitudes and Behavioral Intentions. Table 6 includes the path coefficients associated with the total effect of Authenticity on Behavioral Intentions. This is further decomposed into the direct effect and the indirect effect mediated by Attitudes, presented separately for the different contexts of interest.
Across products and services, and hedonic and utilitarian offerings, Authenticity exhibits a similar, sizable association with Behavioral Intentions (squared coefficients range from 39% to 52%). This implies that when evaluating a consumption experience, there is a positive association between the assessment of authenticity and a consumer's inclination to search for further information, purchase the offering, and spread positive information via word of mouth. Simply put, the composite measure of Authenticity with the six indicators we identify appears to be a predictor of consumers' Behavioral Intentions across various types of consumption experiences. The disaggregated results in Table 6 show that the association of Authenticity with Behavioral Intentions is stronger for utilitarian (vs. hedonic) services (b =.537 > b =.237, p =.02) and is stronger for high- (vs. low-) coproduction services (b =.724 > b =.599, p =.04). The implication is that authenticity matters more to consumers when the consumption goal for a service is instrumental and goal-oriented and when customers are more hands-off in terms of cocreating the service experience. However, some care should be taken when interpreting the latter result as it is not supported by full compositional invariance ([38]).
Focusing on the indirect effect, we observe that Attitudes only partially mediates the effect of Authenticity on Behavioral Intentions. This result further supports the notion that authenticity is conceptually and empirically distinct from attitudes. Indeed, attitudes, as learned predispositions ([26]) typically vary along an evaluative continuum from strongly positive to strongly negative, while authenticity as we define it is neither inherently positive nor negative. Consider, for example, that different individuals may like or dislike highly authentic Thai food (as accurately reflecting how it is made in Thailand). Given the objectives of Study 3, we constructed descriptions of experiences expecting greater authenticity to contribute to more positive attitudes. While we suspect that naive theory would predict that consumers typically prefer the authentic to the inauthentic (e.g., brands accurately reflecting manufacturers, original as opposed to derivative works of art), this need not be the case (e.g., covers of songs that exceed the original in popularity).
The results also suggest that authenticity can affect behavioral intentions through routes other than those associated with positive attitudes. This raises the specter of other reasons Authenticity and Behavioral Intentions might be linked, such as purely economic interests (e.g., as an investment), or a purely emotional response (e.g., nostalgia). Practically speaking, these results imply consumers can be inclined to buy products and services deemed authentic even if they do not especially "like" them, which underscores the importance of considering authenticity as a semi-autonomous driver of consumer decision making.
What does it mean for a consumption experience to be "authentic"? The answer depends on how one conceptualizes authenticity. The marketing literature provides no straightforward answer, as it is replete with varying definitions of what it means to be authentic. Researchers seem to have little trouble generating new definitions related to authenticity. However, until now there has been no attempt to evaluate, deconstruct, or synthesize what is commonly known about what makes consumption authentic. In this research, we engage in a systematic and comprehensive concept reconstruction effort, in which we identify the component indicators that define authenticity as it applies to consumption, reconcile them with the existing literature, and provide the first detailed investigation into the higher-order conceptual structure (i.e., the relationship between authenticity as a construct and its components), setting this work apart from previous research. To illustrate how this work can extend current knowledge, consider [53], who developed and validated scales measuring consumers' perceived brand authenticity along four distinct dimensions. In doing so, they provide initial evidence that these four dimensions do not align into a higher-order reflective construct. Our findings may help explain their results; it may be that their four brand authenticity dimensions are composite indicators as well.
A distinguishing feature of this work is that we take a multimethod approach, utilizing qualitative methods to derive authenticity's composite set of indicators and quantitative methods to empirically investigate how these indicators contribute to authenticity judgments across different consumption contexts. In addition, the reconciliation with existing literature and evidence of heterogeneity help make sense of the disparate ways in which authenticity has been defined in the extant literature. Note that the heterogeneity detected across contexts is consistent with a "family resemblance" concept structure, according to which a concept (authenticity, in this case) may be qualified by different subsets of its dimensions across different contexts, and not always by all of them in the same way ([62]). Again, this is consistent with authenticity being a dimension word, with its meaning remaining uncertain until one knows which of its dimensions are being discussed ([24]). This type of reconceptualization might prove useful to improving the understanding of other marketing constructs that may be conceptually ambiguous.
Conceptualizing authenticity as we have done has important implications for researchers interested in studying the concept in a consumption context. In addition to providing a fuller understanding of what it means for a consumption experience to be authentic, our work helps delineate aspects of an offering that might be manipulated with the intention of changing perceptions of authenticity. Past work in marketing has manipulated certain qualities of an offering presumed to impact perceptions of authenticity (e.g., whether a product is from a company's original manufacturing location; [56]). We suspect being associated with the original manufacturing location reflects a history of consistency, which, in our broad conceptualization, is associated with the component of integrity. In this vein, this research offers guidance on how researchers might systematically identify other specific qualities of an offering that impact assessments of authenticity (e.g., features related to the source's proficiency). For example, one might predict that consumers will consider food cooked by a more recognized chef as more authentic because that chef is perceived as more proficient. Manipulating cues related to these six components could shed further light on how exactly to measure the components we identify and would provide further insight into how marketers can effectively influence consumers' assessments of authenticity.
Given the evidence of heterogeneity we observe among the various components, it would be interesting to examine the relationship between the various components more thoroughly. Consider proficiency, the component that seems to matter most, and legitimacy, the component that seems to matter least. It would seem that skillfulness and artisanship are appreciated without necessarily needing to adhere to tradition ([ 8]), keep to specific genres ([19]), or fit neatly within a category ([15]; [55]). Thinking about how different components relate reinforces the need to reconsider the meaning of concepts periodically and highlights the benefit of concept reconstruction. From our reading of the authenticity literature, historically, far less attention has been paid to qualities of the source (e.g., proficiency, accuracy, integrity) than to qualities of the output (originality, legitimacy). Our holistic conceptualization takes both into account as well as a sense of connectedness between the consumer and the source.
The six components we identify also reveal underlying tensions with respect to how authenticity is appraised (e.g., legitimacy means conforming to standards, whereas originality means standing out). Recall, for example, that legitimacy matters more for high-coproduction services while, conversely, originality matters more for low-coproduction services. It would seem that consumers who are highly involved hedge the risk of reducing authenticity with their participation and idiosyncrasies by weighting the service's adherence to standards (legitimacy) more heavily. Conversely, when consumers are less involved, they put greater weight on the firm's ability to make its offering distinct (originality) when assessing authenticity. This example reveals how much context matters when considering a concept such as authenticity.
This research also contributes to our understanding of authenticity's broader role in consumer decision making. Unlike any prior research of which we are aware, we show how authenticity is connected to consumers' behavioral intentions both directly and indirectly through attitudes. We do so while ensuring discriminant validity between authenticity, attitudes, and behavioral intentions. Thus, this research enhances our understanding of the role of authenticity in consumer decision making and offers a more complete picture of the importance of authenticity to the field.
For practitioners, this work provides valuable insights to marketing managers aiming to enhance the authenticity of their offerings, something that should be of concern to all managers in an environment in which being seen as authentic is increasingly considered table stakes. First, we identify a comprehensive set of judgments consumers make when assessing the authenticity of a consumption experience. Knowing that judgments of accuracy, connectedness, integrity, legitimacy, originality, and proficiency are key components when assessing authenticity, managers can more efficiently and effectively deduce actionable strategies in terms of positioning. How consumers themselves express these judgments is evident in the verbiage in Tables F and H in the Web Appendix. Managers may also identify where there are shortfalls in their offerings on these fronts. In addition, they now also have an initial roadmap with respect to which judgments are more or less important in line with certain characteristics of their offerings. For example, companies selling hedonic products should see relatively large returns, perception-wise, from emphasizing proficiency. The mattress company Tuft & Needle (what is more hedonic that sleep?) leverages on this by making it very clear on its website that it believes in "quality craftsmanship without the gimmicks" (www.tuftandneedle.com/about/story). Knowing the six components enables managers to assess how competitors currently signal authenticity and presents alternative routes to signal the authenticity of their own offerings.
Another example of how managers might leverage the findings presented here includes using them to better understand the role of the consumption context. Consider how originality is more important for low-coproduction services, whereas legitimacy is only important for high-coproduction services. If we consider the services utilized in our studies for illustration, it is important for movies and sporting events to be distinctive (original) to be deemed authentic, while banking and health services should ensure that they are seen as adhering to recognized standards (legitimate). With respect to movies, consider previous research showing that film sequels are more successful if named (e.g., Bridget Jones: The Edge of Reason) than if they are numbered (e.g., Spider-Man 2) precisely because they are perceived as more dissimilar from their predecessor ([72]). The success of emphasizing originality for services such as movies is consistent with our findings, as is the fact [ 4], p. 5) makes it abundantly clear that it maintains a culture committed to ethical behavior and "complying with applicable laws, rules, regulations and policies." Although we do not have direct evidence of the efficacy of these actions in terms of advancing perceptions of authenticity per se, the real-world examples presented here are intended to show how managers can emphasize and act on different components of authenticity. Managers may want to remember the apocryphal yet prescient words of Coco Chanel, who is quoted as saying, "Hard times arouse an instinctive desire for authenticity."
Supplemental Material, sj-docx-1-jmx-10.1177_0022242921997081 - The Concept of Authenticity: What It Means to Consumers
Supplemental Material, sj-docx-1-jmx-10.1177_0022242921997081 for The Concept of Authenticity: What It Means to Consumers by Joseph C. Nunes, Andrea Ordanini and Gaia Giambastiani in Journal of Marketing
Footnotes 1 Karen Winterich
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Joseph C. Nunes https://orcid.org/0000-0002-1812-5042
5 Online Supplement: https://doi.org/10.1177/0022242921997081
References Aaker David A. (1991), Managing Brand Equity. New York: The Free Press.
Arnould Eric J., Price Linda L. (2000), "Authenticating Acts and Authoritative Performances: Questing for Self and Community," in The Why of Consumption: Contemporary Perspectives on Consumer Motives, Goals, and Desires, Ratneshwar S., Mick David Glen, Huffman Cynthia, eds. London: Routledge, 140–63.
Audrezeta Alice, de Kervilerb Gwarlann, Moulard Julie Guidry. (2020), "Authenticity Under Threat: When Social Media Influencers Need to Go Beyond Self-Presentation," Journal of Business Research, 117, 557–69.
Bank of America (2021), "2021 Code of Conduct," (accessed May 5, 2021), https://d1io3yog0oux5.cloudfront.net/bankofamerica/files/pages/corporate-governance/governance-library/code-of-conduct/2021+ADA+Code+of+Conduct+%28English%29.pdf.
Barker Hugh, Taylor Yuval. (2007), Faking It: The Quest for Authenticity in Popular Music. New York: Norton.
6 Batra Rajeev, Ahtola Olli T. (1991), "Measuring the Hedonic and Utilitarian Sources of Consumer Attitudes," Marketing Letters, 2 (2), 159–70.
7 Becker Maren, Wiegand Nico, Reinartz Werner J. (2019), "Does It Pay to Be Real? Understanding Authenticity in TV Advertising," Journal of Marketing, 83 (1), 24–50.
8 Beverland Michael B. (2005), Crafting Brand Authenticity: The Case of Luxury Wines,"Journal of Management Studies, 42 (5), 1003–29.
9 Beverland Michael B., Farrelly Francis J. (2010), "The Quest for Authenticity in Consumption: Consumers' Purposive Choice of Authentic Cues to Shape Experienced Outcomes," Journal of Consumer Research, 36 (5), 838–56.
Blumer Herbert G. (1969), Symbolic Interactionism: Perspective and Method. Upper Saddle River, NJ: Prentice Hall.
Bollen Kenneth A., Bauldry Shawn. (2011), "Three Cs in Measurement Models: Causal Indicators, Composite Indicators, and Covariates," Psychological Methods, 16 (3), 265–84.
Bollen Kenneth A., Diamantopolous Adamantios. (2017), "In Defense of Causal-Formative Indicators: A Minority Report," Psychological Methods, 22 (3), 581–96.
Brown Stephen, Kozinets Robert V., Sherry John F.Jr. (2003), "Teaching Old Brands New Tricks: Retro Branding and the Revival of Brand Meaning," Journal of Marketing, 67 (3), 19–33.
Bruner Edward M. (1994), "Abraham Lincoln as Authentic Reproduction: A Critique of Postmoderism," American Anthropologist, 96 (2), 397–415.
Carroll Glenn R., Wheaton Dennis Ray. (2009), "The Organizational Construction of Authenticity: An Examination of Contemporary Food and Dining in the U.S.," Research in Organizational Behavior, 29, 255–82.
Cinelli Melissa D., LeBoeuf Robyn A. (2019), "Keeping It Real: How Perceived Brand Authenticity Affects Product Perceptions," Journal of Consumer Psychology, 30 (1), 40–59.
Corciolani Matteo, Grayson Kent, Humphreys Ashlee. (2020), "Do More Experienced Critics Review Differently? How Field-Specific Cultural Capital Influences the Judgments of Cultural Intermediaries," European Journal of Marketing, 54 (3), 478–510.
Deibert Ashley. (2017), "Why Authenticity in Marketing Matters Now More than Ever," Forbes(May 26), https://www.forbes.com/sites/forbescommunicationscouncil/2017/05/26/why-authenticity-in-marketing-matters-now-more-than-ever/#28be854d7982.
Delmestri Giuseppe, Montanari Fabrizio, Usai Alessandro. (2005), "Reputation and Strength of Ties in Predicting Commercial Success and Artistic Merit of Independents in the Italian Feature Film Industry," Journal of Management Studies, 42 (5), 975–1002.
Demetry Daphney. (2019), "How Organizations Claim Authenticity: The Coproduction of Illusions in Underground Restaurants," Organization Science, 30 (5), 937–60.
Dhar Ravi, Wertenbroch Klaus. (2000), "Consumer Choice Between Hedonic and Utilitarian Goods," Journal of Marketing Research, 37 (1), 60–71.
Dijkstra Theo K., Henseler Jörg. (2015) "Consistent and Asymptotically Normal PLS Estimators for Linear Structural Equations," Computational Statistics & Data Analysis, 81 (1), 10–23.
Dutton Denis. (1994), "Authenticity in the Art of Traditional Societies," Pacific Arts, (9/10), 1–9.
Dutton Denis. (2003), "Authenticity in Art," in The Oxford Handbook of Aesthetics, Levinson Jerrold, ed. New York: Oxford University Press, 258–74.
Dwivedi Abhishek, McDonald Robert. (2018), "Building Brand Authenticity in Fast-Moving Consumer Goods via Consumer Perceptions of Brand Marketing Communications," European Journal of Marketing, 52 (7/8), 1387–411.
Fishbein Martin. (1967), "A Consideration of Beliefs and Their Role in Attitude Measurement," in Readings in Attitude Theory and Measurement, Fishbein Martin, ed. New York: John Wiley & Sons, 257–66.
Fornell Claes, Larcker David F. (1981), "Evaluating Structural Equation Models with Unobservable Variables and Measurement Error," Journal of Marketing Research, 18 (1), 39–50.
Gilmore James H., Pine B. Joseph. (2007), Authenticity: What Consumers Really Want. Cambridge, MA: Harvard Business Review Press.
Gioia Dennis A., Corley Kevin G., Hamilton Aimee L. (2012), "Seeking Qualitative Rigor in Inductive Research: Notes on the Gioia Methodology," Organizational Research Methods16 (1), 15–31.
Goffin Keith, Koners Ursula. (2011), "Tacit Knowledge, Lessons Learnt, and New Product Development," Journal of Product Innovation Management, 28 (2), 300–318.
Grayson Kent, Martinec Radan. (2004), "Consumer Perceptions of Iconicity and Indexicality and Their Influence on Assessments of Authentic Market Offerings," Journal of Consumer Research, 31 (2), 296–312.
Grazian David. (2004), "The Symbolic Economy of Authenticity in the Chicago Blues Scene," in Music Scenes: Local Translocal, and Virtual, Bennett Andy, Peterson Richard A., eds. Nashville, TN: Vanderbilt University Press, 31–47.
Gudergan Siegfried P., Ringle Christian M., Wende Sven, Will Alexander. (2008), "Confirmatory Tetrad Analysis in PLS Path Modeling," Journal of Business Research, 61 (12), 1238–49.
Guignon Charles. (2004), On Being Authentic. London: Routledge.
Hahl Oliver, Zuckerman Ezra W., Kim Minjae. (2017), "Why Elites Love Authentic Lowbrow Culture: Overcoming High-Status Denigration with Outsider Art," American Sociological Review, 82 (4), 828–56.
Hair Joe F., Howard Matt C., Nitzl Christian. (2020), "Assessing Measurement Model Quality in PLS-SEM Using Confirmatory Composite Analysis," Journal of Business Research, 109, 101–10.
Henseler Jörg. (2017), "Bridging Design and Behavioral Research with Variance-Based Structural Equation Modeling," Journal of Advertising, 46 (1), 178–92.
Henseler Jörg, Hubona Geoffrey, Ray Pauline Ash. (2016), "Using PLS Path Modeling in New Technology Research: Updated Guidelines," Industrial Management & Data Systems, 116 (1), 2–20.
Holt Douglas B. (2002), "Why Do Brands Cause Trouble? A Dialectical Theory of Consumer Culture and Branding," Journal of Consumer Research, 29 (1), 70–90.
Jarvis Cheryl Burke, MacKenzie Scott B., Podsakoff Philip M. (2003), "A Critical Review of Construct Indicators and Measurement Model Misspecification in Marketing and Consumer Research," Journal of Consumer Research, 30 (2), 199–218.
Klein Richard, Rai Arun. (2009), "Interfirm Strategic Information Flows in Logistics Supply Chain Relationships," Management Information Systems Quarterly, 33 (4), 735–62.
Kohli Ajay K. (2009), "From the Editor," Journal of Marketing, 73 (1), 1–2.
Kollat David T., Engel James F., Blackwell Roger D. (1970), "Current Problems in Consumer Behavior Research," Journal of Marketing Research, 7 (3), 327–32.
Kreuzbauer Robert, Keller Joshua. (2017) "The Authenticity of Cultural Products: A Psychological Perspective," Current Directions in Psychological Science, 26 (5), 417–21.
Lehman David W., O'Connor Kieran, Kovács Balázs, Newman George E. (2019), "Authenticity," Academy of Management Annals, 13 (1), 1–42.
Leigh Thomas W., Peters Cara, Shelton Jeremy. (2006), "The Consumer Quest for Authenticity: The Multiplicity of Meanings Within the MG Subculture of Consumption," Journal of the Academy of Marketing Science, 34 (4), 481–93.
Lemke Fred, Clark Moira, Wilson Hugh. (2011), "Customer Experience Quality: An Exploration in Business and Consumer Contexts Using Repertory Grid Technique," Journal of the Academy of Marketing Science, 39 (6), 846–69.
Lovelock Christopher, Gummesson Evert. (2004) "Whither Services Marketing? In Search of a New Paradigm and Fresh Perspectives," Journal of Service Research, 7 (1) 20–41.
MacInnis Deborah J. (2011), "A Framework for Conceptual Contributions in Marketing," Journal of Marketing, 75 (4), 136–54.
MacKenzie Scott B. (2003), "The Dangers of Poor Construct Conceptualization," Journal of the Academy of Marketing Science, 31 (3), 323–26.
Martin Patricia Yancey, Turner Barry A. (1986), "Grounded Theory and Organizational Research," The Journal of Applied Behavioral Science, 22 (2), 141–57.
McLeod Kembrew. (1999), "Authenticity Within Hip-Hop and Other Cultures Threatened with Assimilation," Journal of Communication, 49 (4), 134–50.
Morhart Felicitas, Malär Lucia, Guèvremont Amélie, Girardin Florent, Grohmann Bianca. (2015), "Brand Authenticity: An Integrative Framework and Measurement Scale," Journal of Consumer Psychology, 25 (2), 200–218.
Moulard Julie Guidry, Raggio Randle D., Folse Judith Anne Garretson. (2021), "Disentangling the Meanings of Brand Authenticity: The Entity-Referent Correspondence Framework of Authenticity," Journal of the Academy of Marketing Science, 49 (1), 96–118.
Newman George E. (2019), "The Psychology of Authenticity," Review of General Psychology, 23 (1), 8–18.
Newman George E., Dhar Ravi. (2014), "Authenticity Is Contagious: Brand Essence and the Original Source of Production," Journal of Marketing Research, 51 (3), 371–86.
Oppenheimer Dabier M., Meyvis Tom, Davidenko Nicolas. (2009), "Instructional Manipulation Checks: Detecting Satisficing to Increase Statistical Power," Journal of Experiment Social Psychology, 45 (4), 867–72.
Parasuraman A., Zeithaml Valarie A., Berry Leonard L. (1985), "A Conceptual Model of Service Quality and Its Implication for Future Research," Journal of Marketing, 49 (4), 41–50.
Patton Michael Quinn. (2015), Qualitative Research & Evaluation Methods: Integrating Theory and Practice, 4th ed. Thousand Oaks, CA: SAGE Publications.
Pederson Hans. (2015), "Introduction," in Horizons of Authenticity in Phenomenology, Existentialism, and Moral Psychology: Essays in Honor of Charles Guignon, Pederson Hans, Altman Megan, eds. Amsterdam: Springer, 1–12.
Peterson Richard A. (1997), Creating Country Music: Fabricating Authenticity. Chicago: University of Chicago Press.
Podsakoff Philip M., MacKenzie Scott B., Podsakoff Nathan P. (2016), "Recommendations for Creating Better Concept Definitions in the Organizational, Behavioral, and Social Sciences," Organizational Research Methods, 19 (2), 159–203.
Polites Greta L., Roberts Nicholas, Thatcher Jason. (2011), "Conceptualizing Models Using Multidimensional Constructs: A Review and Guidelines for Their Use," European Journal of Information Systems, 20 (1), 1–27.
Rose Randall L., Wood Stacy L. (2005), "Paradox and the Consumption of Authenticity Through Reality Television," Journal of Consumer Research, 32 (2), 284–96.
Rust Roland T. (2006), "From the Editor: The Maturation of Marketing as an Academic Discipline," Journal of Marketing, 70 (3), 1–2.
Rust Roland T., Cooil Bruce. (1994), "Reliability Measures for Qualitative Data: Theory and Implications," Journal of Marketing Research, 31 (1), 1–14.
Schallehn Mike, Burmann Christoph, Riley Nicola. (2014), "Brand Authenticity: Model Development and Empirical Testing," Journal of Product & Brand Management, 23 (3), 192–99.
Schau Hope Jensen, Muñiz Albert M.Jr, Arnould Eric J. (2009), "How Brand Community Practices Create Value," Journal of Marketing, 73 (5), 30–51.
Schlegel Rebecca J., Hicks Joshua A., Arndt Jamie, King Laura A. (2009), "Thine Own Self: True Self-Concept Accessibility and Meaning in Life," Journal of Personality and Social Psychology, 96 (2), 473–90.
Sheldon Kennon M., Ryan Richard M., Rawsthorne Laird J., Ilardi Barbara. (1997), "Trait Self and True Self: Cross-Role Variation in the Big-Five Personality Traits and Its Relations with Psychological Authenticity and Subjective Well-Being," Journal of Personality and Social Psychology, 73 (6), 1380–93.
Sirianni Nancy J., Bitner Mary Jo, Brown Stephen W., Mandel Naomi. (2013), "Branded Service Encounters: Strategically Aligning Employee Behavior with the Brand Positioning," Journal of Marketing, 77 (6), 108–23.
Sood Sanjay, Drèze Xavier. (2006), "Brand Extensions of Experiential Goods: Movie Sequel Evaluations," Journal of Consumer Research, 33 (3), 352–60.
Spiggle Susan. (1994), "Analysis and Interpretation of Qualitative Data in Consumer Research," Journal of Consumer Research, 21 (3), 491503.
Spiggle Susan, Nguyen Hang T., Caravella Mary. (2012), "More Than Fit: Brand Extension Authenticity," Journal of Marketing Research, 49 (6), 967–83.
Stern Barbara. (1994), "Authenticity and the Textual Persona: Postmodern Paradoxes in Advertising Narrative," International Journal of Research in Marketing, 11 (4), 387–400.
Strauss Anselm, Corbin Juliet M. (1990). Basics of Qualitative Research: Grounded Theory Procedures and Techniques. Newbury Park, CA: SAGE Publications.
Streukens Sandra, Leroi-Werelds Sara. (2016), "Bootstrapping and PLS-SEM: A Step-by-Step Guide to Get More out of Your Bootstrap Results," European Management Journal, 34 (6), 618–32.
Suddaby Roy. (2010), "Construct Clarity in Theories of Organization," Academy of Management Review, 35 (3), 346–357.
Teas R. Kenneth, Palan Kay M. (1997), "The Realms of Scientific Meaning Framework for Constructing Theoretically Meaningful Nominal Definitions of Marketing Concepts," Journal of Marketing, 61 (2), 52–67.
Theodossopoulos Dimitrios. (2013), "Laying Claim to Authenticity: Five Anthropological Dilemmas," Anthropological Quarterly, 86 (2), 337–60.
Trilling Lionel. (1972), Sincerity and Authenticity. Cambridge, MA: Harvard University Press.
Valsesia Francesca, Nunes Joseph C., Ordanini Andrea. (2016), "What Wins Awards Is Not Always What I Buy: How Creative Control Affects Authenticity and Thus Recognition (But Not Liking)," Journal of Consumer Research, 42 (6), 897–914.
Vredeveld Anna J., Coulter Robin. (2018), "Cultural Experiential Goal Pursuit, Cultural Brand Engagement, and Culturally Authentic Experiences: Sojourners in America," Journal of the Academy of Marketing Science, 47 (2), 274–290.
Wang Ning. (1999), "Rethinking Authenticity in Tourism Experience," Annals of Tourism Research, 26(2), 349–70.
Welch Catherine, Rumyantseva Maria, Hewerdine Lisa Jane. (2016), "Using Case Research to Reconstruct Concepts: A Methodology and Illustration," Organizational Research Methods, 19 (1), 111–30.
Williams Lawrence E., Andrew Poehlman T. (2017), "Conceptualizing Consciousness in Consumer Research," Journal of Consumer Research, 44 (2), 231–51.
Yagil Dana Y., Medler-Liraz Hana. (2013), "Moments of Truth: Examining Transient Authenticity and Identity in Service Encounters," Academy of Management Journal, 56 (2), 473–97.
Yim C.K. (Bennett), Chan Kimmy Wa, Lam Simon S.K. (2012), "Do Customers and Employees Enjoy Service Participation? Synergistic Effects of Self- and Other-Efficacy," Journal of Marketing, 76 (6), 121–40.
~~~~~~~~
By Joseph C. Nunes; Andrea Ordanini and Gaia Giambastiani
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 114- The Context (In)Dependence of Low-Fit Brand Extensions. By: Mathur, Pragya; Malika, Malika; Agrawal, Nidhi; Maheswaran, Durairaj. Journal of Marketing. Apr2022, p1. DOI: 10.1177/00222429221076840.
Ahead of Print- Database:
- Business Source Complete
Record: 115- The Control–Effort Trade-Off in Participative Pricing: How Easing Pricing Decisions Enhances Purchase Outcomes. By: Wang, Cindy Xin; Beck, Joshua T.; Yuan, Hong. Journal of Marketing. Sep2021, Vol. 85 Issue 5, p145-160. 16p. 1 Diagram, 1 Chart, 3 Graphs. DOI: 10.1177/0022242921990351.
- Database:
- Business Source Complete
The Control–Effort Trade-Off in Participative Pricing: How Easing Pricing Decisions Enhances Purchase Outcomes
Participative pricing strategies may influence consumer purchase decisions; this research proposes specifically that firms' delegation of pricing decisions to consumers can create a control–effort trade-off. Consumers favor greater pricing control but are deterred by the effort involved in deciding what to pay. Strategies such as pay what you want in turn might reduce purchase intentions due to the effort involved. In contrast, strategies that increase feelings of control but not perceived effort, such as pick your price options that let consumers choose from a limited set of prices, could enhance pricing outcomes. A field study and four laboratory experiments confirm these propositions. The findings demonstrate the mixed effects of participative pricing, identify mediating mechanisms that explain these effects, and specify common moderating conditions that shape the outcomes of participative pricing. These results have notable implications for pricing theory and practice.
Keywords: control; effort; pay what you want; pick your price; pricing
In most purchase situations, the price is fixed, and consumers choose to take it or leave it. But marketers have begun experimenting with pricing strategies that delegate some or all of the price determination task to consumers. Such participative pricing strategies, in which "consumers participate in setting a final price for a product" ([ 8], p. 249), can engage consumers, boost sales, enhance brand loyalty, and contribute to competitive positions ([10]; [51]). Yet many firms—including Priceline ([ 6]), Panera Bread ([27]), and the Metropolitan Museum of Art in New York ([60])—that have experimented with participative pricing also have abandoned it and reverted to fixed prices. We critically examine participative pricing to understand how and in which conditions it might enhance purchase outcomes.
In particular, we theorize about a tension that consumers experience when evaluating a purchase situation that involves participative pricing. Consumers' participation in determining the final price triggers a sense of greater pricing control (a positive purchase driver) but also feelings of more significant pricing effort (a negative purchase driver). Building on this proposition, we predict an overall negative effect of a pay what you want (PWYW) pricing strategy because of the high effort involved in deciding the final price. We also predict an overall positive effect of a novel pick your price (PYP) strategy, which allows consumers to choose a price from a set of options ([28]). Both PWYW and PYP enhance feelings of pricing control, but PYP does not increase pricing effort, because consumers find it relatively easy to make constrained choices ([11]).
Using this control–effort framework, we investigate the effects of PWYW and PYP across five studies in a variety of purchase contexts. Study 1 establishes the strong performance of PYP relative to fixed prices and PWYW in a field quasi-experiment that involves actual purchases and outlines profit implications. Study 2 indicates that compared with a fixed price, PWYW decreases purchase intentions due to the high effort required, even though it confers full pricing control. In contrast, PYP provides a sense of pricing control but does not affect pricing effort, so it increases purchase intentions. These effects hold regardless of the price level. Study 3 demonstrates that shopping motives (to save money or time) moderate the effects of effort and control on purchase intentions. Whereas PWYW is more effective than a fixed price when consumers are motivated to save money, PYP is more effective in general. When consumers are motivated to save time, PYP is just as effective as a fixed price and more effective than PWYW. Study 4 directly manipulates pricing effort and control while holding the pricing strategy (PYP) constant to validate the underlying process. Study 5 identifies single (vs. multiple) purchase decisions as a crucial moderating condition, such that the positive effect of PYP on purchase choice attenuates over multiple purchase decisions, which require a great deal of effort.
The current research thus makes several contributions to pricing literature. First, we explicate the mixed effectiveness of participative pricing strategies by examining psychological mechanisms that drive their effects. Second, we test a novel form of participative pricing (PYP) and show that providing consumers with price options can enhance retail purchase outcomes. Although PYP has been used in various forms in nonprofit and tipping contexts, it has not been tested in a for-profit context or compared with other participative pricing strategies. In exploring PWYW and PYP, we conceptualize pricing control and pricing effort and empirically test our control–effort trade-off framework. Previous research has focused mainly on explaining why consumers would pay more than $0 and identified social factors, such as altruism, fairness, and price consciousness, that determine the price paid ([35]; [47]). We advance this stream of literature by demonstrating a key role of pricing effort, which can deter consumers from making PWYW purchases. Third, we show that participative pricing strategies perform differently in various contexts, such as when consumers are motivated to save time versus money and make single versus multiple purchase decisions. The framework thus provides practical guidance for enhancing the purchase outcomes of participative pricing strategies in the marketplace.
Prices are often fixed before the point of sale, whereas participative pricing delegates some or all of the price determination to the consumer ([35]). Forms of participative pricing have been in use for millennia (e.g., auctions date back to 500 BCE; [39]), and today, they appear in various business models adopted by restaurants, musicians, gaming companies, travel providers, charities, and others ([ 9]; [55]; [57]; [59]).
Prior marketing literature suggests that how participative pricing is presented to consumers matters ([38]). Participative pricing also comes in many forms, including auctions, reverse auctions, exchanges, negotiations, name your own price, and PWYW (for a review, see [35]]). In its unique approach, PWYW involves one buyer and one seller (vs. auctions, which involve many buyers), and the seller must take whatever price is selected (vs. negotiations, where sellers can counter). This full delegation of price determination can be exciting for consumers and risky for firms. In a sense, PWYW is the most participative option; consumers have full say over the final price, and because they might choose to pay $0, firms worry about the performance implications. Yet marketers grew increasingly interested in PWYW after the popular and successful 2007 launch of a PWYW-priced album by the band Radiohead ([57]). Research since then has characterized PWYW as a novel and attractive pricing strategy (e.g., [21]; [35]).
Despite PWYW's initial successes, many firms that adopted it have reverted to fixed-price models: Priceline abandoned PWYW pricing for airfares in 2016 ([ 6]), Panera Bread closed its last remaining PWYW restaurant in Boston ([27]), and the Metropolitan Museum of Art in New York ended its PWYW ticket policy in 2018, when the museum president and chief executive officer Daniel [60], p. 1) described the strategy as "no longer sufficient to meet the Museum's daily operational demands." Lower-than-expected payments may have caused some reversions. For example, PWYW is more effective when consumers are motivated by fairness ([19]), charity ([21]), or relational concerns ([52]). In addition, offering a reference price may cause consumers to pay only that or below that price because paying above it seems irrational. Finally, consumers simply appear reluctant to pay for some PWYW offerings.
However, PWYW is not just challenging for firms; it also can be difficult for consumers. We theorize that two fundamental tensions exist for consumers. First, they are attracted to the economically advantageous pricing terms (i.e., pay anything, including $0), which permits them to maximize their utility. Second, consumers must decide the final price, which takes effort; even though a $0 payment is an option, it often is not in the best interest of the long-term exchange (i.e., consumers understand that sellers need revenue to maintain their ongoing operations). Moreover, paying a price that appears socially inappropriate may have adverse reputational effects (e.g., they may be viewed as cheap). Thus, consumers must weigh the benefits of pricing control against the costs of pricing effort. We refer to these countervailing forces as the control–effort trade-off.
PYP, which asks consumers to choose the price they prefer to pay from multiple price options, has not been studied formally in pricing literature. We regard it as an evolution of PWYW, in that consumers still have the final say over the price, but they do not have to generate the price options themselves. This format is increasingly common in charity and tipping contexts. We also note a commercial application by Everlane, an online clothing retailer. A well-received "choose what you pay" promotion allowed shoppers to pick from among three prices when buying a product ([28]). We contend that PYP offers similar benefits as PWYW (high pricing control) without the accompanying disadvantages (high pricing effort). If this prediction holds, PYP should outperform PWYW in many pricing situations and offer a novel, potentially powerful pricing strategy.
Consumers have a strong economic interest in determining a price. We define pricing control as the perceived level of personal influence over the process of determining the final price paid for a good or service. From a neoclassical economic perspective, consumers should want the highest pricing control possible to pay the lowest price possible (i.e., $0) ([ 1]; [53]). Increased feelings of control also have an array of affective and cognitive psychological benefits ([ 4]; [37]; [41]). Consumers arguably hold favorable attitudes toward brands that grant them a sense of pricing control ([26]; [34]). Pricing control also is linked to economic rewards. Both PWYW and PYP allow consumers to pay custom prices and offer higher levels of pricing control than fixed pricing; we posit that pricing control as a positive purchase driver explains positive outcomes of PWYW and PYP.
Participative pricing strategies also trigger pricing effort, or psychological feelings of mental exertion associated with the price determination. As we have discussed, if consumers consider only economic utility, they always pay the lowest price possible, which is an easy decision. Yet the average prices paid by consumers in PWYW settings are significantly greater than zero ([35]), suggesting that determining a price is a complex process shaped by social and psychological factors ([19]; [22]), including avoiding negative feelings of guilt or shame for overpaying (wasting money) or underpaying (appearing cheap) ([40]). It can be difficult to arrive at an "appropriate" PWYW price, even with a reference price, because it is unclear whether a typical payment is correct ([19]; [34]). Because PWYW involves limitless payment options, it also might prompt busy consumers with limited mental resources to opt out of the decision and avoid purchasing ([29]). Such consumers may be more likely to accept a fixed price from a competing seller rather than paying the same or even a lower price under PWYW ([22]). We present pricing effort as a negative driver that explains the negative impact of PWYW on purchase intentions. To compare PWYW with PYP, we thus distinguish conceptually between control and effort: control involves a situational evaluation (how much control do I have over prices?), whereas effort involves the use of cognitive resources (how effortful is it to think about price?).
Because PYP involves selecting from among several options, it should not demand as much pricing effort. Studies of choice difficulty provide a useful analog: consumers claim they want many choice options but also experience "choice overload" when trying to select from among a vast assortment ([11]; [49]). In a participative pricing context, with maximal pricing control, determining the price is akin to making a selection from a large set of options (e.g., should a cookie cost $.50, $.75, $1.00, $1.25, $1.50, $1.75...?). Rather than leaving the price completely open-ended, a firm could reduce the required pricing effort by asking consumers to pick from a limited number of pricing options. The effort required to make such a decision may be the same or slightly more than the evaluation of a fixed price. In summary, we predict that PWYW generally underperforms fixed pricing and PYP due to the greater pricing effort it demands, which overwhelms the benefits of the greater pricing control it offers; in contrast, PYP should generally outperform fixed pricing and PWYW because it primarily increases pricing control. Formally:
- H1: Relative to fixed price, (a) PWYW generally reduces and (b) PYP generally increases purchase intentions.
- H2: Relative to a fixed price, (a) PWYW and (b) PYP increase consumers' feelings of pricing control, and (c) PWYW increases consumers' feelings of pricing effort.
- H3: (a) The negative effect of PWYW (vs. fixed price) on purchase intentions is mediated by pricing effort, whereas (b) the positive effect of PYP (vs. fixed price) on purchase intentions is mediated by pricing control.
Building on this framework, we consider several moderating factors that may shape the (dis)advantages of PWYW and PYP. Consumers often approach purchase decisions with different motives, such as convenience, greater selection, lower prices, unique offerings, or support for local merchants ([14]; [30]; [31]). In this process, saving time and saving money are common decision motives ([17]).
When consumers are motivated to save money, they focus their attention on their level of control over the price, because higher pricing control allows consumers to save more. The perceptions of pricing effort are minimized, as consumers are already willing to expend effort to save money ([61]). Money-saving promotions, such as coupons, require consumer effort ([48]), yet consumers are enticed by the incentives ([46]). Coupon promotions even increase the number of shopping trips to stores ([58]). Thus, consumers motivated to save money likely focus more on pricing control and less on pricing effort, such that the negative effect of PWYW on consumer purchases should be attenuated. Then, both PWYW and PYP should increase purchase intentions because they elicit high pricing control. Formally:
- H4: The motivation to save money moderates the effect of PWYW and PYP on purchase intentions, such that when consumers are motivated to save money, both (a) PYP and (b) PWYW have positive effects on purchase intentions, compared with a fixed price.
If, instead, consumers are motivated to save time, their focus may shift away from pricing control and toward pricing effort. The process of making a purchase decision, especially if it entails extra effort, can be time consuming and exhausting; it even can reduce consumers' brand loyalty and fairness perceptions ([44]; [61]). Consumers prefer quick, convenient, and less effortful purchase decisions, particularly when they are more motivated to save time ([12]; [32]). Expending effort instead demands time and resources. Therefore, PWYW should be less effective than PYP or fixed pricing when consumers are motivated to save time, as the motive to save time minimizes consumers' focus on pricing control and may enhance their focus on pricing effort, which could nullify some benefits of PYP. We propose the following:
- H5: The motivation to save time moderates the effect of PWYW and PYP on purchase intentions, such that when consumers are motivated to save time, (a) PYP and fixed price have similar effects but (b) PWYW has a negative effect on purchase intentions compared with a fixed price.
Beyond shopping motives, it is important to note that the shopping context also influences the effect of PYP on purchase outcomes. A shopping trip might involve a single or multiple purchase decisions, depending on consumers' shopping goals and purposes ([ 3]; [23]). Multiple purchases in a shopping trip incur more effort because they require multiple decisions, involving a large number of choices. Evaluating perceived risk and uncertainty across multiple items also increases the level of effort required ([ 5]; [49]). Although the pricing effort of PYP remains low for an individual product, the evaluation of multiple price options for multiple products can become more effortful in aggregate. When consumers are involved in multiple purchases, they likely try to preserve mental resources to make all the needed purchase decisions, instead of expending unnecessary effort on each item, even if that effort might lead to more favorable prices and more satisfying experiences. Therefore, we expect that when multiple purchases are involved, the effect of PYP is attenuated. Formally:
- H6: The number of purchase decisions moderates the effect of PYP versus fixed price on purchase intentions, such that the positive effect of PYP on purchase intentions is attenuated after multiple purchase decisions.
To test these hypotheses, we conduct five studies (one field quasi-experiment and four laboratory experiments) that collectively confirm our theoretical framework (see Figure 1) and demonstrate how PWYW and PYP distinctly shape consumer purchases.
Graph: Figure 1. Conceptual model.
We examine the effects of PWYW and PYP in a field experiment, with an emphasis on how PYP may enhance purchases and revenue. Consistent with H1, we expect PYP to outperform both PWYW and fixed pricing. In a retail setting, we manipulated the pricing strategy and measured the number of units sold and revenues. Specifically, we partnered with a bakery chain and varied the pricing strategy for one of its core products: freshly baked cookies.
The study relied on random variations of three pricing strategies (PYP, PWYW, and fixed price) at 50-minute intervals for 15 business hours (5 total hours per strategy) throughout three afternoons. The interval was determined by the store hours, trained staff members' working schedules, and the goal of ensuring even time allocations of all pricing conditions. Consumers self-selected into the study by entering the store.
The popular bakery offers traditional baked goods (e.g., pastries, muffins, cakes), including freshly baked cookies priced at $.60 each. For this study, the bakery agreed to vary its cookie pricing strategy to test the effect on sales. No other promotions occurred during this time. The study ran during regular business hours, from 1:00 p.m. to 6:00 p.m., for three days (Thursday, Friday, and Saturday; 15 hours total). Employees were trained as confederates but were blind to the hypotheses. The pricing strategy was manipulated by signage. In the PWYW condition, consumers were instructed to "pay what you want" (no reference price). In the fixed-price condition, cookies were priced at the usual $.60 each. In the PYP condition, consumers were asked to "pick your price" and provided four prices from which to choose ($.45, $.55, $.65, $.75), which averaged $.60. See study materials and measures for this and following studies in the Web Appendix.
In each condition, the signage was displayed for an equal amount of time. To approximate a random assignment, the staff changed the pricing strategy in randomly determined 50-minute intervals. The staff were careful to ensure that consumers were unaware of the change. The retailer recorded the unit sales and cookie revenue generated in each condition, which served as our primary dependent variable, and kept a log of questions and comments as a source of qualitative data.
The retailer sold 526 units and generated $260 in revenue from cookie sales (Figure 2). To understand the role of pricing strategy, we first examined whether the pricing condition affected units sold by calculating the proportion in each pricing condition as a percent of the total units sold across conditions. A chi-square test of proportions revealed a significant difference in proportion by pricing condition (fixed price = 30.80%, PWYW = 22.43%, PYP = 46.77%; χ2 = 49.72, p <.001). The store sold more units in the PYP condition than the fixed-price condition (χ2 = 17.67, p <.01) or the PWYW condition (χ2 = 46.43, p <.001). In addition, PWYW resulted in fewer units sold than did fixed pricing (χ2 = 7.25, p <.01). Units purchased in the PYP condition generated revenues of $122.10, compared with $40.70 in the PWYW and $97.20 in the fixed-price conditions. The business did not share cost information, but with a conservative assumption of 33% variable costs ($.20 per cookie), the profits in the PYP condition are higher ($73.36) than in the fixed price ($64.80) and PWYW ($16.90) conditions. In this calculation, PYP increased profits by 13% relative to fixed prices and by 334% relative to PWYW.
Graph: Figure 2. Study 1: effect of pricing strategies on sales performance.
The average price paid by participants in the PYP condition was $.50 per unit (vs. $.34 in PWYW and $.60 in fixed price). That is, participants in the PWYW condition purchased fewer units and also paid lower prices (F( 2, 523) = 521.10, p <.001). See price paid details in the Web Appendix. The qualitative data also provide some insights into consumers' reactions to the pricing strategies. First, the PWYW strategy results in greater pricing effort, such that consumers voiced questions such as, "What price should I pay?" and "What price do other people tend to pay?" Second, employees anecdotally reported that consumers spent a longer time thinking about the transaction, whereas they responded more quickly and decisively in the PYP condition.
Because the average price in the PYP condition was $.50 per unit (vs. the usual price of $.60 per unit), the PYP strategy arguably may have generated more unit sales merely due to lower prices. We instead propose that the effects are driven by greater pricing control, paired with an appropriate amount of required effort (choosing rather than forming the price). To explore these two possibilities, we collected additional data in a discount condition in the same setting and priced the cookies at $.50 highlighted by a sign that also indicated the regular price of $.60. The PYP condition in the main study still outperforms the fixed discount price in this study (with the caveat that these data were collected a week apart): A $.50 discounted price generated 203 unit sales (cf. 246 sold with PYP; χ2 = 4.30, p <.05). We also recorded the units sold the next day, during the same experimental window (1:00–6:00 p.m.) but without a prominent pricing sign, as an additional control condition. The bakery generated 161 unit sales, nearly identical to the amount sold in the fixed-price condition in the main study (162 units, χ2 =.003, p =.96).
Study 1 provides initial support for our proposed framework. Consistent with H1, we find that, relative to a conventional fixed price, PWYW has a negative effect on purchase intentions, whereas PYP has a positive effect. Sales records show that PYP outperforms all other pricing strategies, with 108% more units sold than PWYW, 52% more units than the fixed price (as typically used by the store), and 21% more units than a discount strategy. Although consumers paid slightly less on average ($.50 vs. $.60), the significant sales increase in the PYP, relative to the discount, condition demonstrates the strong potential of PYP pricing strategies. Notably, the negative effects of PWYW emerged despite a lack of reference price, suggesting that consumers were not simply paying at or below a reference price.
Although we could not measure consumer control and effort directly in Study 1, the anecdotal evidence suggests that consumers viewed PWYW as more effortful and asked more questions about it. The purchase differences between PWYW and PYP also indicate that PYP increases purchases and profits by enhancing pricing control (relative to fixed price) but alleviates pricing effort (relative to PWYW).
Study 2 has several goals. First, with a lab experiment, we further validate our core predictions that, relative to a fixed-price strategy, consumers are more likely to purchase when they can use PYP and less likely to do so using PWYW. Second, we test the generalizability of these effects across high and low price levels. Third, to gain insights into the underlying process, we include measures of pricing control and pricing effort. We predict that PWYW enhances perceived pricing control (H2a) and effort (H2b), whereas PYP enhances pricing control (H2c) but has no discernible impact on pricing effort. We further predict that pricing effort relates negatively, and pricing control relates positively, to purchase intentions. We test these mediating roles of pricing effort and pricing control on purchase intentions (H3) in Study 2. Fourth, we rule out perceived fairness as an alternative mediator. Bidding is generally perceived as fairer than retailer-set prices ([24]), and consumers might view PYP or PWYW as similar to bidding; we ultimately do not find evidence for this role, though.
This study used a 3 (pricing strategy: PWYW, PYP, fixed price) × 2 (price level: high, low) between-subjects design. We recruited 382 U.S. adults (Mage = 35.62 years; 41.4% female) from Amazo Mechanical Turk (MTurk) for a nominal payment. They read about a fictitious genetic testing service (Zyntech) and rated their purchase intentions. Participants then completed postsurvey measures and were thanked.
The pricing strategy and price-level manipulations were embedded in the service description. Half of the participants were assigned to the high-price condition. Among this group, those assigned to the fixed-price condition read that the price was $199; those in the PWYW condition read that most consumers pay $199, but they could pay any amount they felt was appropriate; and those in the PYP condition read that they could choose a price to pay from a list ($189, $199, or $209). For the other half, assigned to the low-price condition, participants in the fixed-price condition read that the price was $49; those in the PWYW condition read that most consumers pay $49, but they could pay any price they felt was appropriate; and those in the PYP condition read that they could choose a price to pay from a list ($39, $49, and $59). In the PYP conditions, the average price thus was equivalent to the reference price in the PWYW conditions and the retail price in the fixed-price conditions.
Participants rated their purchase intentions on a single seven-point scale item ("How likely are you to purchase genetic testing from Zyntech?"; 1 = "extremely unlikely," and 7 = "extremely likely"). They also reported their pricing effort using three items ("The pricing strategy made me feel like the decision to purchase this genetic test was an effortful one," "The pricing strategy demanded effort for me to reach a purchase decision," and "Compared to other types of pricing, the pricing strategy for this product made it easy for me to reach a purchase decision" [reverse coded]; 1 = "strongly disagree," and 7 = "strongly agree"; α =.76). They rated perceived pricing control using five items ("The way this genetic testing is priced made me feel like I was in control," "When deciding whether or not to get a genetic testing from Zyntech, I felt I was in control of the price," "The price of this genetic testing service was controlled by me," "The pricing strategy would demonstrate how much control I have," and "The pricing strategy made me feel like I am in control of the price"; 1 = "strongly disagree," and 7 = "strongly agree"; α =.95). For perceived fairness, they rated two items ("The pricing strategy was fair" and "The price was fair"; 1 = "strongly disagree," and 7 = "strongly agree"; r =.72) adapted from [15]. Finally, participants provided their demographic information.
Participants completed a pricing strategy attention check (1 = "one price was offered," 2 = "pick-your-price with a list of options," 3 = "pay whatever you want," and 4 = "not sure"), which 348 (91.1%) of 382 participants answered correctly, as well as a price-level attention check (1 = "more than $100," 2 = "less than $100," and 3 = "not sure"), which 353 (92.4%) participants answered correctly. Excluding participants who failed the attention checks did not materially alter the significance of the results, so we report the results with all participants.
We submitted purchase intentions to a 3 (pricing strategy) × 2 (price level) analysis of variance (ANOVA). The results reveal significant main effects of pricing strategy (F( 2, 376) = 17.82, p <.001, η2 =.09) and price level (F( 1, 376) = 7.68, p <.01, η2 =.02) and a nonsignificant two-way interaction (F < 1, n.s.). Participants indicate weaker purchase intentions when the service demands a high price (M = 3.87, SD = 2.04) rather than a low one (M = 4.39, SD = 1.88). Across pricing strategies, we observe that purchase intentions in the PYP condition (M = 4.83, SD = 1.76) are significantly higher than in the fixed-price (M = 4.07, SD = 1.95; t(256) = 3.29, p <.01, d =.41) or PWYW (M = 3.44, SD = 1.98; t(253) = 5.93, p <.001, d =.75) conditions. They are lower in the PWYW than in the fixed-price condition (t(249) = −2.54, p <.01, d =.32) (Table 1). The differences across pricing strategies for the high/low price level conditions reveal similar statistical patterns (i.e., price level does not moderate the effects).
Graph
Table 1. Study 2: Means and Standard Deviations of Purchase Intentions, Pricing Control, and Pricing Effort.
| Purchase Intentions | Pricing Control | Pricing Effort |
|---|
| Mean | SD | Mean | SD | Mean | SD |
|---|
| Fixed price | 4.07 | 1.95 | 3.48 | 1.64 | 3.36 | 1.44 |
| PWYW | 3.44 | 1.98 | 5.26 | 1.15 | 4.35 | 1.62 |
| PYP | 4.83 | 1.76 | 4.78 | 1.28 | 3.64 | 1.24 |
We submitted pricing effort to a 3 (pricing strategy) × 2 (price level) ANOVA. The results reveal a significant main effect of pricing strategy (F( 2, 376) = 16.09, p <.001, η2 =.08), a nonsignificant main effect of price level (F( 1, 376) = 1.45, p =.23, η2 =.004), and a nonsignificant two-way interaction (F < 1, n.s.). Contrasts reveal that, relative to the PYP condition (M = 3.64, SD = 1.24), participants reported similar levels of effort in the fixed-price condition (M = 3.36, SD = 1.44; t(256) = −1.67, p =.12, d =.21) but higher pricing effort in the PWYW condition (M = 4.35, SD = 1.62; t(253) = 3.94, p <.001, d =.50). Pricing effort is also greater in the PWYW condition than in the fixed-price condition (t(249) = 5.12, p <.001, d =.65).
We submitted pricing control to a 3 (pricing strategy) × 2 (price level) ANOVA. The results reveal significant main effects of pricing strategy (F( 2, 376) = 56.98, p <.001, η2 =.23) and price level (F( 1, 376) = 10.04, p <.01, η2 =.03) and a nonsignificant two-way interaction (F < 1, n.s.). Across pricing strategies, pricing control in the PYP condition (M = 4.78, SD = 1.28) is higher than in the fixed-price condition (M = 3.48, SD = 1.64; t(256) = 7.11, p <.001, d =.89) but lower than in the PWYW condition (M = 5.26, SD = 1.15; t(253) = −3.14, p <.01, d =.39). Pricing control also is greater in the PWYW than in the fixed-price condition (t(249) = 9.93, p <.001, d = 1.26). Differences across pricing strategies for the high- and low-price conditions follow similar statistical patterns.
We empirically confirm the discriminant validity of pricing effort and control (r =.177) by calculating their average variances extracted (AVE). The square of the parameter estimate for this pair of constructs (phi =.03) is less than the mean of its AVE estimates (.75), so the two measures are distinct ([18]). We also report the results of an alternate regression analysis for each focal outcome for this and remaining studies in the Web Appendix.
We conducted mediation analyses with [25] PROCESS macro (model 4; 10,000 bootstrapped samples), to test the effect of pricing strategy on purchase intentions, using pricing effort and control as mediators. We coded pricing strategy such that the effects of PWYW and PYP are relative to those of a fixed-price condition. The positive effect of PYP on purchase intentions is mediated by pricing control (indirect effect b =.46, SE =.12; 95% confidence interval [CI] = [.24,.71]) but not by pricing effort (indirect effect b = −.09, SE =.06; 95% CI = [−.21,.02]). Alternatively, the effect of PWYW on purchase intentions was mediated by both pricing control (indirect effect b =.63, SE =.15; 95% CI = [.35,.93]) and pricing effort (indirect effect b = −.33, SE =.09; 95% CI = [−.52, −.16]). We further discuss these mediation patterns in the Web Appendix.
Study 2 provides additional support for our proposed framework. Consistent with H1, we find that, relative to a fixed price, PWYW undermines, whereas PYP increases, purchase intentions, regardless of price level. We also find empirical support for H2, such that both PWYW and PYP increase consumers' perceived pricing control (cf. fixed price), and PWYW increases consumers' perceived pricing effort. In other words, PYP elicits greater pricing control but does not affect pricing effort, relative to a fixed price. The results confirm the mediating roles of pricing control and pricing effort too (H3). Thus, PYP enhances purchase intentions through greater pricing control. For PWYW, we observe mediation by both perceived control and perceived effort, in countervailing pathways. In addition, we rule out the perceived fairness (see the Web Appendix) of the pricing strategy as an alternative explanation for the outcomes in this and remaining studies.
In Study 3, we build on Study 2 findings by examining shopping motives as moderators of the pricing control and effort pathways. It may be possible to shift consumers' perception of pricing control or effort by altering their shopping motives. When consumers are motivated to save money, they are less focused on the expenditure of effort ([17]; [36]), and control over the price paid helps them achieve their goal ([ 9]; [35]). When consumers are motivated to save time, ease and efficiency are critical ([42]; [54]). Thus, we predict that PWYW and PYP (vs. fixed price) both increase purchase intentions among consumers motivated to save money, but PWYW's effect on purchase intentions will remain negative among consumers motivated to save time (H4, H5). To test these effects, we use a different service category (i.e., insurance).
This study used a 3 (pricing strategy: PWYW, PYP, fixed) × 3 (motive: save time, save money, baseline) between-subjects design. We recruited 675 U.S. adults (Mage = 38.7 years; 41.2% female) from MTurk and paid them a nominal fee. Participants were assigned to one of nine conditions.
Participants first completed a priming task that manipulated their shopping motives ([ 7]; [20]). Participants in the save-time (money) condition had to list four to five reasons that saving time (money) is important when making a purchase decision. Participants in the baseline condition did not complete any writing task. All participants read a description of renters' insurance available from American Family Insurance Company. The pricing information appeared on the following page. In the fixed-price condition, participants read that the price was $180/year. In the PYP condition, they could choose one of three prices to pay: $140/year, $180/year, or $220/year. In the PWYW condition, consumers were told they could pay any price they considered appropriate. Participants then rated pricing effort (α =.86), pricing control (α =.95), perceived fairness (r =.79), and purchase intentions. Finally, they provided demographic information.
Participants completed the pricing attention check from Study 2, and 608 (90.1%) of 675 participants passed it. They also completed a motive attention check (1 = "save time," 2 = "save money," and 3 = "no writing task"), which 588 (87.1%) of them passed. The effects did not change when participants who failed the attention checks were excluded, so we report the results obtained from all participants.
We submitted purchase intentions to a 3 (pricing strategy) × 3 (motive) ANOVA. The results revealed significant main effects of pricing strategy (F( 2, 666) = 9.38, p <.001, η2 =.03) and motives (F( 2, 666) = 15.11, p <.001, η2 =.04). Participants in the PYP condition (M = 4.91, SD = 1.77) indicate stronger purchase intentions than those in the fixed-price (M = 4.30, SD = 1.96; t(449) = 3.47, p <.001, d =.33) or PWYW (M = 4.23, SD = 1.97; t(446) = 3.84, p <.001, d =.36) conditions. In the fixed-price and PWYW conditions, we find similar purchase intentions (t < 1, p =.70, d =.04). When motivated to save time (M = 3.94, SD = 2.03), participants indicate weaker purchase intentions than those motivated to save money (M = 4.83, SD = 1.69; t(447) = −5.04, p <.001, d =.48) or those in the baseline condition (M = 4.66, SD = 1.92; t(449) = −3.87, p <.001, d =.37). People motivated to save money and those in the baseline condition indicate similar purchase intentions (t < 1, p =.32, d =.09).
We also find a significant two-way (pricing strategy × motive) interaction (F( 4, 666) = 9.44, p <.001, η2 =.05) (Figure 3). In the baseline condition, participants in the PWYW condition (M = 3.95, SD = 2.04) reveal weaker purchase intentions than those in the fixed-price (M = 4.59, SD = 1.91; t(148) = −1.98, p <.05, d =.33) and PYP (M = 5.42, SD = 1.49; t(148) = −5.05, p <.001, d =.83) conditions. In the PYP condition, we find stronger purchase intentions than in the fixed-price condition (t(150) = 2.99, p <.01, d =.49), replicating the pattern of effects from Study 2 for the baseline condition. Yet in the save-time condition, PWYW (M = 3.36, SD = 2.02) results in weaker purchase intentions than in the fixed-price (M = 4.20, SD = 1.99; t(148) = −2.57, p <.05, d =.42) or PYP (M = 4.27, SD = 1.98; t(148) = −2.79, p <.01, d =.46) conditions. Participants in the PYP and fixed-price conditions indicate similar purchase intentions (t < 1, p =.83, d =.03). Finally, in the save-money condition, PWYW participants (M = 5.39, SD = 1.15) offer stronger purchase intentions than those in the fixed-price condition (M = 4.09, SD = 1.96; t(149) = 4.96, p <.001, d =.81), but they are similar to the PYP condition (M = 5.03, SD = 1.62; t(146) = 1.56, p =.12, d =.26). The PYP condition also leads to stronger purchase intentions than the fixed-price condition (t(147) = 3.18, p <.01, d =.52).
Graph: Figure 3. Study 3: moderating effects of shopping motives on purchase intentions.Notes: Error bars = ±1 SEs.
As we detailed in the "Conceptual Development" section, we predict that consumers' motivation to save time or money will alter perceptions of pricing effort and pricing control. We submitted pricing effort to a 3 (pricing strategy) × 3 (motive) ANOVA. The results reveal a significant main effect of pricing strategy (F( 2, 666) = 13.52, p <.001, η2 =.04), a nonsignificant main effect of motive (F( 2, 666) = 1.05, p =.35, η2 =.003), and a significant two-way interaction (F( 4, 666) = 5.69, p <.01, η2 =.03). Across pricing strategies, participants exert greater effort in PWYW (M = 4.88, SD = 1.52) than in the fixed-price condition (M = 4.17, SD = 1.54; t(449) = 4.93, p <.001, d =.47) and the PYP condition (M = 4.54, SD = 1.39; t(446) = 2.47, p <.05, d =.23). Pricing effort also is greater in the PYP than fixed-price condition (t(449) = 2.67, p <.05, d =.25). Next, we examine the interaction.
In the baseline condition, participants in the PWYW condition (M = 4.99, SD = 1.35) reported greater pricing effort than those in the fixed-price (M = 4.22, SD = 1.52; t(148) = 3.28, p <.01, d =.54) and PYP (M = 4.25, SD = 1.54; t(148) = 3.13, p <.01, d =.51) conditions. In the PYP and fixed-price conditions, there was a similar level of pricing effort (t < 1, p =.94, d =.001), replicating the pattern of effects from Study 2 for the baseline condition. Yet in the save-time condition, PWYW (M = 5.00, SD = 1.74) increased pricing effort relative to the baseline condition (M = 3.64, SD = 1.62; t(148) = 5.38, p <.001, d =.88) but not the PYP condition (M = 4.74, SD = 1.35; t(148) = 1.12, p =.26, d =.18), while participants in the PYP reported higher pricing effort than those exposed to the fixed price (t(148) = 4.52, p <.001, d =.74). Finally, in the save-money condition, pricing effort across all three pricing strategies is similar (MPWYW = 4.66, SD = 1.43; Mfixed price = 4.63, SD = 1.31; MPYP = 4.65, SD = 1.23; ts < 1, n.s.).
We submitted pricing control to a 3 (pricing strategy) × 3 (motive) ANOVA, which shows a significant main effect of pricing strategy (F( 2, 666) = 66.37, p <.001, η2 =.17), a nonsignificant main effect of motive (F( 2, 666) = 1.38, p =.25, η2 =.004), and a significant two-way interaction (F( 4, 666) = 7.07, p <.01, η2 =.04). Pricing control in the PWYW condition (M = 5.48, SD = 1.28) is higher than in the fixed-price condition (M = 4.06, SD = 1.58; t(449) = 10.48, p <.001, d =.99) and the PYP condition (M = 5.06, SD = 1.19; t(446) = 3.60, p <.001, d =.34). Pricing control also is higher in the PYP condition than the fixed-price condition (t(449) = 7.58, p <.001, d =.72).
In the baseline condition, PWYW (M = 5.56, SD = 1.05) increased pricing control relative to the fixed-price (M = 4.06, SD = 1.65; t(148) = 6.62, p <.001, d = 1.09) but not PYP (M = 5.33, SD =.99; t(148) = 1.38, p =.17, d =.23) conditions. PYP significantly increased pricing control relative to the fixed-price condition (t(150) = 5.75, p <.001, d =.94), replicating the pattern of effects from Study 2. The pattern was slightly different in the save-time condition: PWYW (M = 5.11, SD = 1.58) increased control relative to fixed-price (M = 4.51, SD = 1.34; t(148) = 2.51, p <.05, d =.41) but not PYP (M = 4.87, SD = 1.26; t(148) = 1.02, p =.31, d =.17) conditions. Participants in the PYP condition reported marginally greater pricing control than those exposed to a fixed price (t(148) = 1.70, p =.09, d =.28). Finally, participants revealed the largest difference in pricing conditions in the save-money condition: PWYW (M = 5.76, SD = 1.07) increased pricing control relative to fixed price (M = 3.62, SD = 1.61; t(149) = 9.61, p <.001, d = 1.58) and PYP (M = 4.96, SD = 1.29; t(146) = 4.11, p <.001, d =.68). PYP also increased control relative to the fixed-price condition (t(147) = 5.59, p <.01, d =.92).
We again find support for discriminant validity (r =.12); the square of the parameter estimate between the effort and control constructs (phi =.01) is less than the mean of the pair's AVE (AVE =.80; [18]). In addition, we report the detailed significant moderated mediation results in the Web Appendix.
The results from Study 3 provide evidence in support of H4 and H5. In the baseline condition, participants indicated higher purchase intentions in the PYP (vs. fixed-price) condition and lower purchase intentions in the PWYW (vs. fixed-price) condition. Perceived effort and control mediated these relationships, as in Study 2. Alternatively, when the motive to save time was activated, PWYW resulted in lower purchase intentions than in the PYP and fixed-price conditions, and purchase intentions did not differ between PYP and fixed price. In short, the mediation evidence indicates that the motive to save time magnified perceptions of effort and reduced the benefits of PYP, whereas the motive to save money minimized effort and magnified perceptions of control in a way that bolstered PWYW and PYP. The results across Studies 1–3 also indicate that the effects on purchase intentions are robust, regardless of the presence or absence of a reference price, in the PWYW condition.
Thus far, we have provided process evidence through measured mediation (pricing effort and control) and situational moderation (motives to save time vs. money), which confirms that the poor performance of PWYW is due to greater pricing effort, whereas the superior performance of PYP results from increased pricing control without a corresponding increase in effort. In our studies, the PYP conditions include a limited number of price options (three to four), which minimizes perceived pricing effort, and an implicit promise to sell the offering at whatever price the participant selects, which encourages perceived pricing control. In Study 4, to test the control–effort trade-off further, we modify the features of the PYP condition to increase the sense of effort and reduce the sense of control. We thus can examine how variations in PYP designs affect its advantages. Finally, we examine movie ticket sales as the purchase context.
We recruited 260 U.S. adults (Mage = 35.45 years; 43.5% female) from MTurk for a nominal payment. Participants were assigned to one of four pricing conditions (fixed price, PYP, high-effort PYP, low-control PYP) in a between-subjects design.
Participants first read a description about a Regal Premiere Movie eTicket 4-Packs, then saw the pricing information on a subsequent page. In the fixed-price condition, the pack is listed for $37.99 plus tax. In the PYP condition, participants read that consumers can choose a price they want to pay from a list that included $29.99, $37.99, and $45.99 plus tax. In the high-effort PYP condition, participants reviewed a list of ten price options, ranging from $29.99 to 47.99 in $2 increments. In the low-control PYP condition, they saw the price options from the PYP condition but also read, "You can pick your own price from the list below, but the product may not be available at the price you pick." Participants indicated their purchase likelihood and the maximum amount they would be willing to pay for the tickets. Finally, participants provided demographic information.
We used the three-item pricing effort (α =.89) and five-item pricing control (α =.95) measures from Studies 2 and 3. In addition, of the 260 participants, 234 (90.0%) passed the pricing attention check (1 = "One price was offered," 2 = "Pick-your-price with a short list of options [3 options]," 3 = "Pick-your-price with a long list of options [10 options]," and 4 = "Pick-your-price with a short list of options [3 options] but product may not be available"). The effects did not change when we excluded participants who failed the attention checks, so we report the results with all participants.
Verifying the effort and control manipulations, a one-way ANOVA of pricing effort revealed a significant effect of pricing strategy on perceived effort (F( 3, 256) = 4.83, p <.01, η2 =.05). The high-effort PYP condition (M = 4.80, SD = 1.69) significantly increased perceived pricing effort, relative to the fixed-price (M = 3.94, SD = 1.62; t(129) = 2.97, p <.01, d =.52), PYP (M = 3.80, SD = 1.61; t(129) = 3.46, p <.01, d =.61), and low-control PYP (M = 4.08, SD = 1.65; t(130) = 2.45, p <.05, d =.43) conditions. The reported levels of pricing effort were similar across the latter three conditions (ts < 1, n.s.). A one-way ANOVA of pricing control also showed a significant effect (F( 3, 256) = 11.96, p <.001, η2 =.12), such that those in the low-control PYP condition (M = 3.93, SD = 1.73) reported significantly lower pricing control than those in the PYP (M = 4.78, SD = 1.54; t(127) = −2.97, p <.01, d =.53) and high-effort PYP (M = 5.17, SD = 1.51; t(130) = −4.40, p <.001, d =.77) conditions but similar levels of control to those in the fixed-price condition (M = 3.69, SD = 1.74; t < 1, n.s.). In the PYP condition, we find higher pricing control than in the fixed-price condition (t(126) = 3.76, p <.001, d =.67) but directionally lower control than in the high-effort PYP condition (t(129) = −1.45, p =.15, d =.26).
We submitted purchase intentions to a one-way ANOVA. The results reveal a significant main effect of pricing strategy (F( 3, 256) = 4.59, p <.01, η2 =.05; see the Web Appendix). That is, participants express higher purchase intentions in the PYP condition (M = 5.25, SD = 1.46) than in the fixed-price (M = 4.39, SD = 2.02; t(126) = 2.76, p <.01, d =.49), high-effort PYP (M = 4.10, SD = 2.01; t(129) = 3.70, p <.001, d =.65), and low-control PYP (M = 4.34, SD = 1.96; t(127) = 2.99, p <.01, d =.53) conditions. No significant difference arises among the fixed-price, high-effort PYP, and low-control PYP conditions (ts < 1, n.s.). Although we manipulated pricing effort and pricing control, we tested whether the control and effort manipulation checks served as mediators of the effects of pricing conditions on purchase intentions. The results indicated that there was a significant mediation that was consistent with our framework (see the Web Appendix).
This study validates our theoretical model and confirms the role of pricing effort and pricing control in determining the effectiveness of participative pricing strategies, using direct manipulations. When pricing effort increases, consumers are less likely to purchase, even if the pricing strategy offers more pricing control. When pricing control decreases, consumers also show weaker purchase intentions, even when the pricing strategy evokes the same level of pricing effort. Our studies consistently show that PYP is an effective participative pricing form that induces higher purchase intentions in various product categories, both online and offline. But is it effective in all shopping contexts?
The successful manipulation of pricing effort associated with PYP in Study 4 revealed an attenuating effect on purchase intentions, but we do not observe a full reversal (i.e., negative effect). Perhaps the high-effort threshold required to reduce purchase intentions was not reached by ten price choice options. In Study 5, we examine the role of effort differently by manipulating the number of PYP decisions made across products. We predict that having to choose prices for multiple products will increase pricing effort sufficiently to reduce participants' purchase choices, relative to fixed prices (H6). In addition, we sought to examine the revenue implications of PYP more directly. In Study 1, participants paid a specific price, yet we were unable to observe nonpurchases. In Studies 2–4, participants indicated their purchase intentions and maximum willingness to pay, but this permits only an approximation of revenue. Study 5 addresses these limitations by recording purchase choice (yes/no) and the corresponding payment amount in an experiment where participants make selections in a shopping simulation. From this, we are able to assess the revenue implications for a single product paired with PYP (vs. fixed price) and for a shopping basket of products paired with PYP (vs. fixed price). We anticipate a positive effect of PYP for a single product and a negative effect for a basket of products.
This study adopted a 2 (pricing strategy: PYP vs. fixed) × 2 (product evaluation: single vs. multiple) between-subjects design. We recruited 420 U.S. adults (Mage = 36.09 years; 34% female) from MTurk and paid them a nominal fee.
In the single-product-evaluation condition, participants were presented with an electric toothbrush (Quip) and given a brief product description and price. In the fixed-price condition, participants read that the electric toothbrush sells for $30 and were asked to indicate their purchase choice ("Please indicate whether you would purchase the Quip Electric Toothbrush"; 0 = "I would not purchase this product," and 1 = "I would purchase this product for $30"). In the PYP condition, participants were instructed that they could choose the price to pay for the electric toothbrush from a list of $25, $30, and $35. Participants indicated their purchase choice ("Please indicate whether you would purchase the Quip Electric Toothbrush"; 0 = "I would not purchase this product," 1 = "I would purchase this product for $25," 2 = "I would purchase this product for $30," and 3 = "I would purchase this product for $35").
In the multiple-product-evaluation condition, participants were presented an electric toothbrush (Quip), along with three other products: dish soap (Palmolive), laundry detergent (Tide), and a box of storage quart bags (Ziploc), with a brief description and price (presentation order was counterbalanced). In the fixed-price condition, participants read that the electric toothbrush sells for $30, the dish soap sells for $5, the laundry detergent for $11.97, and the storage bags for $7.48. In the PYP condition, participants were told that they may pick their purchase price (toothbrush: $25, $30, $35; dish soap: $4, $5, $6; detergent: $ 9.97, $11.97, $13.97; and bags: $5.98, $7.48, $8.98). Participants indicated their purchase choice for each product (similar to the single evaluation protocol), rated pricing effort (α =.88), pricing control (α =.92), and perceived fairness (r =.81) and provided demographic information.
Of the 420 participants, 412 (98%) passed the attention check. The results are not different when those who failed the attention check are excluded from the sample, so we report the results with all participants.
We conducted a logistic regression to assess the effect of pricing strategy (0 = fixed, 1 = PYP), product evaluation (0 = single, 1 = multiple), and their interaction (pricing strategy × evaluation) on purchase choice for the electric toothbrush (0 = no purchase, 1 = purchase) as the focal dependent variable that was included in all four conditions. We observed a significant pricing strategy × evaluation interaction (B = −2.86, SE =.45, Z = −6.37, Wald = 40.52, p <.001) (Figure 4). As we predicted, in the single-product-evaluation condition, PYP increased purchase choice (fixed price: 54.29%, PYP: 85.71%; B = 1.62, SE =.34, Z = 4.75, p <.001). In the multiple-product-evaluation condition, PYP reduced purchase choice (fixed price: 70.48%, PYP: 40.95%; B = −1.24, SE =.29, Z = −4.23, p <.001).
Graph: Figure 4. Study 5: moderating effect of shopping context on purchase choice.Notes: Data represent the proportion of respondents indicating that they would purchase (yes vs. no) a toothbrush paired with a PYP (vs. fixed price) in a single-evaluation (toothbrush only) or multiple-evaluation (toothbrush and other PYP products) context.
Next, we examined the effects of PYP on purchase choice for the alternate products included only in the multiple-product-evaluation condition. We generally observed a negative (or directionally negative) effect of PYP on purchase choice: dish soap (PYP = 73.33%, fixed price = 89.52%, χ2 = 1.69, p =.19), detergent (PYP = 61.90%, fixed price = 90.48%, χ2 = 5.63, p <.05), and storage bags (PYP = 60.95%, fixed price = 88.57%, χ2 = 5.36, p <.05).
We used the purchase choice and price given/selected[ 5] to estimate the revenue generated in each condition. We first examined only the toothbrush revenue, as it is the only product that was displayed in all four conditions. In the single-evaluation condition, we found that PYP generated a total of $2,595 in toothbrush revenue (vs. $1,710 in the fixed-price condition), an average of $24.71 per participant (vs. $16.29 in the fixed-price condition; t(187.69[ 6]) = 8.43, p <.001). In the multiple evaluation condition, we found that PYP generated a total of $1,215 in toothbrush revenue (vs. $2,220 in the fixed-price condition), an average of $11.57 per participant (vs. $21.14 in the fixed-price condition; t(208) = −4.97, p <.001). Next, we examined only the multiple-evaluation condition to assess the shopping basket revenue generated across all products. We found that the PYP condition generated $2,705 (vs. $4,523 in the fixed-price condition), for an average of $25.75 per participant (vs. $43.07 in the fixed-price condition; t(208) = −7.43, p <.001).
We submitted pricing effort to a 2 (pricing strategy) × 2 (evaluation) ANOVA. The results reveal a nonsignificant main effect of pricing strategy (MPYP = 4.46, SD = 1.52; Mfixed = 4.24, SD = 1.66; F( 1, 416) = 2.06, p =.15, η2 =.005), a significant main effect of evaluation (Mmultiple = 4.53, SD = 1.49; Msingle = 4.17, SD = 1.68; F( 1, 416) = 5.68, p <.05, η2 =.01), and a significant two-way interaction (F( 1, 416) = 3.72, p =.05, η2 =.009). In the single-evaluation condition, participants indicated similar level of pricing effort for PYP and fixed price (MPYP = 4.13, SD = 1.63; Mfixed = 4.20, SD = 1.73; t(208) =.30, p =.76, d =.04). In the multiple-evaluation condition, PYP (M = 4.79, SD = 1.32) increased pricing effort relative to the fixed price (M = 4.27, SD = 1.60; t(208) = 2.57, p =.01, d =.36).
We submitted pricing control to a 2 (pricing strategy) × 2 (product evaluation) ANOVA. The results reveal a significant main effect of pricing strategy (MPYP = 5.15, SD = 1.22; Mfixed = 4.14, SD = 1.59; F( 1, 416) = 53.16, p <.001, η2 =.11), a nonsignificant main effect of product evaluation (Mmultiple = 4.73, SD = 1.37; Msingle = 4.55, SD = 1.62; F( 1, 416) = 1.71, p =.19, η2 =.004), and a nonsignificant two-way interaction (F( 1, 416) = 1.19, p =.28, η2 =.003).
Notably, pricing control and effort exhibited adequate discriminant validity (r =.185); the square of their parameter estimate (phi =.03) is less than the mean of their AVE estimates (AVE =.78; [18]). Consistent with previous studies, pricing control and effort significantly mediated the effect of pricing strategies on purchase choices (see the Web Appendix).
The results from Study 5 confirm our framework in several key ways. First, we conceptually replicate mediation findings from previous studies in the single-product condition where PYP increased purchase choice, and this effect was mediated by increases in pricing control but not effort. Alternatively, in the multiple-product condition, PYP reduced purchase choice for the focal product, and the negative effect of PYP was mediated by both pricing control and effort, which is comparable to the pattern of effects observed for PWYW in Study 2. Furthermore, we found compelling evidence of PYP as a critical strategy for single products that can backfire for multiple products. PYP increased total toothbrush revenue by 52% in the single-product condition, whereas it reduced revenue by 45% in the multiple-product conditions. Notably, these revenue differences account for purchase choice (yes/no) and, conditional on a purchase, the price selected. The choice data suggest that the increase in revenue in the single-product condition is largely driven by an increase in purchases, whereas the reduction in revenue in the multiple-product condition is driven by a decrease in purchases. Interestingly, there was an increase in toothbrush purchases in the multiple- (vs. single-) product condition. We speculate that the increase was caused by the consideration set ([ 5]; [33]; [56]). The toothbrush may have seemed more attractive (e.g., technologically advanced) when compared with dish soap, detergent, and bags. And perhaps this contrast explains the purchase increase, which further manifest the complexity when consumers evaluate products and make purchase decisions. In summary, we demonstrate the robust evidence in favor of the control–effort trade-off for participative pricing, which can be influenced by the pricing strategy (PYP vs. PWYW) or the purchase context where single versus multiple products are paired with a participative pricing strategy.
With this research, we propose a control–effort trade-off framework to understand how participative pricing shapes consumers' purchase decisions. We examine participative pricing by focusing on two forms (PWYW and PYP) that delegate price determinations to consumers. With PWYW, the process of price delegation increases feelings of pricing control and effort, and pricing effort explains its general negative effect on purchase intentions in routine purchase contexts (retail food in Study 1, health services in Study 2, financial services in Study 3, entertainment in Study 4, and household products in Study 5). Because PYP increases only feelings of pricing control, not pricing effort, it generally outperforms PWYW (and fixed prices) across purchase domains. We theorize and present evidence that these effects occur because consumers favor pricing control but also want to avoid the costs associated with engaging in pricing effort.
Several factors shape this control–effort trade-off for participative pricing. In Study 3, when consumers are motivated to save time, PWYW underperforms PYP and fixed price, and PYP exhibits performance similar to that of fixed prices. This outcome occurs because the focus on pricing control diminishes; having control over the final price does not facilitate time savings. It also stems from the magnification of pricing effort (PWYW) because effort undermines time-saving attempts. Alternatively, when consumers are motivated to save money, PWYW and PYP both outperform fixed prices. The money-saving motive enhances the focus on pricing control (PWYW, PYP), such that consumers feel as if they can achieve a satisfactorily low price, and it reduces feelings of effort. In Study 5, multiple PYP product decisions increase pricing effort, so PYP is not as effective as a fixed price. Thus, shopping motives (saving time or money) and shopping contexts serve as key moderators, with practical insights.
In a parallel investigation in donation contexts (see the Web Appendix), both PWYW and PYP perform significantly better than a fixed-price strategy. We thus complement research that demonstrates the efficacy of PWYW in donation contexts by not only conceptually replicating previously findings ([21]) but also providing evidence that PYP offers an appealing and novel pricing strategy. Choosing how much to donate from a list of options ($1, $3, or $5) is a popular solicitation method, but PYP strategies for products or services have not been examined empirically. We find that PYP increases purchase intentions regardless of donation context, but PWYW is more effective in donation (vs. nondonation) contexts, presumably because consumers are willing to expend effort if their purchase benefits others.
By adopting a consumer perspective on participative pricing, this study advances pricing theory in several ways. First, we uncover two key psychological mechanisms that underpin participative pricing: control and effort. Their trade-off helps explicate some mixed findings about the performance of participative pricing ([10]; [19]). It represents a shift in thinking, relative to most pricing literature, which takes a price-paid perspective and prioritizes maximizing profit. Although such research can provide guidance for matching offerings, price points, and consumer heterogeneity to improve pricing efficiency ([13]; [16]), it cannot specify or explain consumers' psychological responses. We fill a research gap related to consumers' decision processes in participative pricing settings, thereby explaining why PWYW underperforms in routine purchase contexts, while also specifying some factors (purchase motives, donation context) that can enhance its performance. The negative effects of PWYW on purchase intentions hold regardless of the presence of reference prices; the positive effects of PYP thus do not appear to be due to merely a reference price effect.
Second, by comparing different forms of participative pricing according to our control–effort trade-off framework, we show that PYP, a relatively newer and untested pricing strategy, can enhance purchase outcomes. By offering price options, PYP balances the trade-off: it provides a high sense of control but does not demand more effort, so it offers an improved form of price participation. In Study 1, the PYP strategy significantly outperforms all other pricing strategies, including a discount with the same economic value, so its performance benefit cannot be explained by mere economics. Rather, these results emphasize the unique impacts of the control–effort trade-off. As an added benefit, PYP is unlikely to erode product or brand quality perceptions, in contrast with frequent discounts, which can have long-term negative consequences for sellers ([ 2]; [45]).
Third, we contribute generally to research on consumer decision making by identifying contexts in which consumers' purchase decisions differ, even with the same pricing strategy. Boundary conditions enhance and attenuate the negative effects of pricing effort. For example, fixed prices can be more effective than PYP for multiple product decisions; PWYW can be as effective as PYP if consumers are motivated to save money or their purchase involves donations to charity. However, if consumers are motivated to save time, the negative effect of PWYW on purchase intentions is amplified. In general, PYP is more effective when used for a single product (vs. multiple products simultaneously).
In addition to explaining why PWYW might decrease purchases, we specify when PWYW will be more or less effective. Managers can selectively implement PWYW to enhance their pricing performance. Our finding that PWYW decreases purchase intentions in specific domains also is consistent with the choices of various companies to abandon PWYW ([27]; [60]). Adopting PWYW strategies requires careful consideration, and even if marketing research were to indicate that consumers favor higher pricing control (Study 2), the execution might fail, due to consumers' reluctance to expend effort to determine their own price (Studies 1–3). Alternatively, firms might benefit more generally from adopting PYP strategies.
Consumers' willingness to exert pricing effort also varies across shopping contexts. For example, our research suggests PWYW is more effective when consumers are motivated to save money, such as during the holiday season ([50]). Yet PWYW likely is less effective for busy consumers just seeking to get their holiday shopping done quickly. For online stores that emphasize monetary savings, both PYP and PWYW could be more effective than conventional fixed prices. However, firms should be cautious with the use of PYP in shopping trips that involve multiple product purchases, as we find that the advantage of PYP is impaired when consumers select prices for multiple products in a single shopping trip, which increases effort. In fact, fixed price would be more effective than PYP in multiple-product decisions; PYP is more effective when paired with a single product. Contexts that involve donations are another scenario for which participative pricing, especially PWYW, promises benefits; consumers already expect to expend effort for the benefit of others. Consistent with research demonstrating PWYW's benefits for donations ([21]), we find that both PYP and PWYW outperform fixed prices in charitable contexts.
Building on previous research (e.g., [22]), we uncover critical roles of pricing control and effort in determining the performance of participative pricing and identify moderating roles of shopping motives and shopping contexts. Continued research could aim to elucidate consumer characteristics and product attributes that also explain the performance advantages and disadvantages of different forms of participative pricing. It also might address the varying dynamics and trade-offs associated with brand attitudes and profitability. Although we observe positive effects on revenue and profits in Study 1 and revenue estimates from Study 5, the willingness-to-pay results in Studies 2–4 indicate inconsistent patterns (see the Web Appendix). Perhaps pricing strategies that offer high pricing control are more effective for building favorable brand attitudes, which could be useful for new firms seeking to acquire consumers.
Various factors influence the outcomes of participative pricing. While this research focuses on pricing effort and its direct effects, future research could explore other relevant factors to expand our knowledge on participative pricing and deepen our understanding of consumers' purchase decisions. Generalizing our control–effort trade-off framework, one might observe similar effects for other marketing practices that demand consumer participation, such as sharing pricing information online or providing feedback in reviews. Limiting effort expenditures in such exchange-extending activities arguably could improve the participation outcomes. Other factors affect consumers' willingness to expend effort, such as when they consider expensive products (e.g., automobiles). When the value, price, or risks of purchase are higher, buyers devote more effort to the purchase decision ([43]).
Our control–effort trade-off framework also might clarify the implications of other pricing strategies. For example, name-your-own-price options might trigger a sense of high effort (because consumers try to predict which low price will be acceptable to the seller) and also strong money-saving motives that offset it. In addition, our research may inspire exploration into other pricing strategies from a control–effort trade-off perspective. In terms of pricing strategies, PYP and PWYW are analogous to a first-price auction. In pricing practice, the minimum starting bid and a maximum "buy it now" pricing options for products listed on eBay offer a structure for the final price to be determined by potential buyers. These structured options are similar to PYP, which strikes a compromise or balance in the control–effort trade-off. Further investigation in other pricing areas such as those will enrich our control–effort trade-off framework.
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921990351 - The Control–Effort Trade-Off in Participative Pricing: How Easing Pricing Decisions Enhances Purchase Outcomes
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921990351 for The Control–Effort Trade-Off in Participative Pricing: How Easing Pricing Decisions Enhances Purchase Outcomes by Cindy Xin Wang, Joshua T. Beck and Hong Yuan in Journal of Marketing
Footnotes 1 Praveen Kopalle
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 https://doi.org/10.1177/0022242921990351
5 Participants who opted not to purchase a given product were assigned a price of $0.
6 We report adjusted degrees of freedom when Levene's test indicated unequal variance between groups.
References Allen Roy G.. (1938), Mathematical Analysis for Economists. New York : MacMillan.
Anderson Eric T. , Simester Duncan I.. (2001), " Price Discrimination as an Adverse Signal: Why an Offer to Spread Payments May Hurt Demand, " Marketing Science , 20 (3), 315 – 27.
Arentze Theo A. , Oppewal Harmen , Timmermans Harry J.. (2005), " A Multipurpose Shopping Trip Model to Assess Retail Agglomeration Effects, " Journal of Marketing Research , 42 (1), 109 – 15.
Beck Joshua T. , Rahinel Ryan , Bleier Alexander. (2020), " Company Worth Keeping: Personal Control and Preferences for Brand Leaders, " Journal of Consumer Research , 46 (5), 871 – 86.
Bettman James R. , Luce Mary Frances , Payne John W.. (1998), " Constructive Consumer Choice Processes, " Journal of Consumer Research , 25 (3), 187 – 217.
Bhattarai Abha. (2016), " Priceline Just Dumped the Features That Made It Famous, " The Washington Post (September 8) , https://www.washingtonpost.com/news/the-switch/wp/2016/09/08/priceline-just-dumped-the-feature-that-made-it-famous-bidding-on-airline-tickets/.
7 Cacioppo John T. , Petty Richard E.. (1981), " Social Psychological Procedures for Cognitive Response Assessment: The Thought-Listing Technique, " Cognitive Assessment (11), 309 – 342.
8 Chandran Sucharita , Morwitz Vicki G.. (2005), " Effects of Participative Pricing on Consumers' Cognitions and Actions: A Goal Theoretic Perspective, " Journal of Consumer Research , 32 (2), 249 – 59.
9 Charness Gary , Cheung Tsz. (2013), " A Restaurant Field Experiment in Charitable Contributions, " Economics Letters , 119 (1), 48 – 9.
Chen Yuxin , Koenigsberg Oded Z. , Zhang John. (2017), " Pay-as-You-Wish Pricing, " Marketing Science , 36 (5), 780 – 91.
Chernev Alexander , Böckenholt Ulf , Goodman Joseph. (2015), " Choice Overload: A Conceptual Review and Meta-Analysis, " Journal of Consumer Psychology , 25 (2), 333 – 58.
Childers Terry , Carr Christopher , Peck Joann , Carson Stephen. (2011), " Hedonic and Utilitarian Motivations for Online Retail Shopping Behavior, " Journal of Retailing , 77 (4), 511 – 35.
Christopher Ranjit M. , Machado Fernando S.. (2019), " Consumer Response to Design Variations in Pay-What-You-Want Pricing, " Journal of the Academy of Marketing Science , 47 (5), 879 – 98.
Darden William R. , Reynolds Fred D.. (1971), " Shopping Orientations and Product Usage Rates, " Journal of Marketing Research , 8 (4), 505 – 08.
Darke Peter R. , Dahl Darren W.. (2003), " Fairness and Discounts: The Subjective Value of a Bargain, " Journal of Consumer Psychology , 13 (3), 328 – 38.
Dhar Sanjay K. , Hoch Stephen J.. (1996), " Price Discrimination Using In-Store Merchandising, " Journal of Marketing , 60 (1), 17 – 30.
Duxbury Darren , Keasey Kevin , Zhang Hao , Chow Shue Loong. (2005), " Mental Accounting and Decision Making: Evidence under Reverse Conditions Where Money Is Spent for Time Saved, " Journal of Economic Psychology , 26 (4), 567 – 80.
Fornell Claes , Larcker David F.. (1981), " Evaluating Structural Equation Models with Unobservable Variables and Measurement Error, " Journal of Marketing Research , 18 (1), 39 – 50.
Gerpott Torsten J. , Schneider Christina. (2016), " Buying Behaviors When Similar Products Are Available Under Pay-What-You-Want and Posted Price Conditions: Field-Experimental Evidence, " Journal of Behavioral and Experimental Economics , 65 , 135 – 45.
Gino Francesca , Mogilner Cassie. (2014), " Time, Money, and Morality, " Psychological Science , 25 (2), 414 – 21.
Gneezy Ayelet , Gneezy Uri , Nelson Leif D. , Brown Amber. (2010), " Shared Social Responsibility: A Field Experiment in Pay-What-You-Want Pricing and Charitable Giving, " Science , 329 (5989), 325 – 27.
Gneezy Ayelet , Gneezy Uri , Riener Gerhard , Nelson Leif D.. (2012), " Pay-What-You-Want, Identity, and Self-Signaling in Markets, " Proceedings of the National Academy of Sciences , 109 (19), 7236 – 40.
Harlam Bari A. , Lodish Leonard M.. (1995), " Modeling Consumers' Choices of Multiple Items, " Journal of Marketing Research , 32 (4), 404 – 18.
Haws Kelly L. , Bearden William O.. (2006), " Dynamic Pricing and Consumer Fairness Perceptions, " Journal of Consumer Research , 33 (3), 304 – 11.
Hayes Andrew F.. (2017), Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach. New York : The Guilford Press.
Henderson Conor M. , Beck Joshua T. , Palmatier Robert W.. (2011), " Review of the Theoretical Underpinnings of Loyalty Programs, " Journal of Consumer Psychology , 21 (3), 256 – 76.
Houck Brenna. (2019), " Panera's Utopic Pay-What-You-Want Restaurant Dream Is Dead, " Eater (Feburary 5) , https://www.eater.com/2019/2/5/18212499/panera-cares-closing-pay-what-you-can-restaurant.
Ismael Amir. (2017), " Everlane Has Brought Back Its 'Choose What You Pay' Sale, " Business Insider (December 27) , https://www.businessinsider.com/everlane-choose-what-you-pay-sale-2017-12.
Iyengar Sheena , Lepper Mark. (2000), " When Choice Is Demotivating: Can One Desire Too Much of a Good Thing? " Journal of Personality and Social Psychology , 79 (6), 995 – 1006.
Jasper Cynthia R. , Lan Pi-Nan Rosa. (1992), " Apparel Catalog Patronage: Demographic, Lifestyle and Motivational Factors, " Psychology & Marketing , 9 (4), 275 – 96.
Jin Liyin , He Yanqun , Zhang Ying. (2014), " How Power States Influence Consumers' Perceptions of Price Unfairness, " Journal of Consumer Research , 40 (5), 818 – 33.
Johnson Eric J. , Goldstein Daniel. (2003), " Do Defaults Save Lives? " Science , 302 (5649), 1338 – 39.
Kahneman Daniel , Miller Dale T.. (1986), " Norm Theory: Comparing Reality to Its Alternatives, " Psychological Review , 93 (2), 136 – 53.
Kim Ju-Young , Kaufmann Katharina , Stegemann Manuel. (2014), " The Impact of Buyer–Seller Relationships and Reference Prices on the Effectiveness of the Pay What You Want Pricing Mechanism, " Marketing Letters , 25 (4), 409 – 23.
Kim Ju-Young , Natter Martin , Spann Martin. (2009), " Pay What You Want: A New Participative Pricing Mechanism, " Journal of Marketing , 73 (1) , 44 – 58.
Kivetz Ran. (2003), " The Effects of Effort and Intrinsic Motivation on Risky Choice, " Marketing Science , 22 (4), 477 – 502.
Koriat Asher , Lichtenstein Sarah , Fischhoff Baruch. (1980), " Reasons for Confidence, " Journal of Experimental Psychology: Human Learning and Memory , 6 (2), 107 – 18.
Krishna Aradhna , Briesch Richard , Lehmann Donald R. , Yuan Hong. (2002), " A Meta-Analysis of the Impact of Price Presentation on Perceived Savings, " Journal of Retailing , 78 (2), 101 – 18.
Krishna Vijay. (2002), Auction Theory. San Diego : Academic Press.
Kunter Marcus. (2015), " Exploring the Pay-What-You-Want Payment Motivation, " Journal of Business Research , 68 (11), 2347 – 57.
Langer Ellen J. , Rodin Judith. (1976), " The Effects of Choice and Enhanced Personal Responsibility for the Aged: A Field Experiment in an Institutional Setting, " Journal of Personality and Social Psychology , 34 (2), 191 – 98.
Leclerc France , Schmitt Bernd H. , Dubé Laurette. (1995), " Waiting Time and Decision Making: Is Time Like Money? " Journal of Consumer Research , 22 (1) , 110 – 19.
List John A. , Lucking-Reiley David. (2000), " Demand Reduction in Multiunit Auctions: Evidence from a Sportscard Field Experiment, " American Economic Review , 90 (4), 961 – 72.
Mehta Nitin , Rajiv Surendra , Srinivasan Kannan. (2003), " Price Uncertainty and Consumer Search: A Structural Model of Consideration Set Formation, " Marketing Science , 22 (1), 58 – 84.
Mela Carl F. , Gupta Sunil , Lehmann Donald R.. (1997), " The Long-Term Impact of Promotion and Advertising on Consumer Brand Choice, " Journal of Marketing Research , 34 (2), 248 – 61.
Mulhern Francis J. , Padgett Daniel T.. (1995), " The Relationship Between Retail Price Promotions and Regular Price Purchases, " Journal of Marketing , 59 (4), 83 – 90.
Schmidt Klaus M. , Spann Martin , Zeithammer Robert. (2015), " Pay What You Want as a Marketing Strategy in Monopolistic and Competitive Markets, " Management Science , 61 (6), 1217 – 36.
Schneider Linda G. , Currim Imran S.. (1991), " Consumer Purchase Behaviors Associated with Active and Passive Deal-Proneness, " International Journal of Research in Marketing , 8 (3), 205 – 22.
Schwartz Barry. (2004), The Paradox of Choice: Why More Is Less. New York : Ecco.
Shively Holly. (2018), " Shoppers Find Deals on Black Friday, " Dayton Daily News (accessed March 10, 2019) , https://www.daytondailynews.com/news/shoppers-find-deals-black-friday-but-some-say-not-fun/eXMsY3MLD4mOlC4im7GsSO/.
Spann Martin , Tellis Gerard J.. (2006), " Does the Internet Promote Better Consumer Decisions? The Case of Name-Your-Own-Price Auctions, " Journal of Marketing , 70 (1), 65 – 78.
Stangl Brigitte , Kastner Margit , Prayag Girish. (2017), " Pay-What-You-Want for High-Value Priced Services: Differences Between Potential, New, and Repeat Customers, " Journal of Business Research , 74 (May), 168 – 74.
Strotz Robert H.. (1955), " Myopia and Inconsistency in Dynamic Utility Maximization, " Review of Economic Studies , 23 (3), 165 – 80.
Thaler Richard. (1985), " Mental Accounting and Consumer Choice, " Marketing Science , 4 (3), 199 – 214.
Thomaselli Rich. (2012), " Priceline Kills the Messenger Because Ads Worked Too Well, " Advertising Age (January 30) , http://adage.com/article/news/priceline-kills-messenger-ads-worked/232409/.
Trijp Hans , Hoyer Wayne , Inman J. Jeffrey. (1996), " Why Switch? Product Category–Level Explanations for True Variety-Seeking Behavior, " Journal of Marketing Research , 33 (3), 281 – 92.
Tyrangiel Josh. (2007), " Radiohead Says: Pay What You Want, " Time (October 1) , http://content.time.com/time/arts/article/0,8599,1666973,00.html.
Walters Rockney G. , Rinne Heikki J.. (1986), " An Empirical-Investigation into the Impact of Price Promotions on Retail Store Performance, " Journal of Retailing , 62 (3), 237 – 66.
Warren Nathan , Hanson Sara , Yuan Hong. (2021), " Feeling Manipulated: How Tip Request Sequence Impacts Customers and Service Providers " Journal of Service Research , 24 (1), 66 – 83.
Welss Daniel H.. (2018), " The Met's Updated Admissions Policy, " The Met (January 4) , https://www.metmuseum.org/blogs/now-at-the-met/2018/updated-admissions-policy-daniel-weiss.
Xia Lan , Kukar-Kinney Monika , Monroe Kent B.. (2010), " Effects of Consumers' Efforts on Price and Promotion Fairness Perceptions, " Journal of Retailing , 86 (1), 1 – 10.
~~~~~~~~
By Cindy Xin Wang; Joshua T. Beck and Hong Yuan
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 116- The Double-Edged Effects of E-Commerce Cart Retargeting: Does Retargeting Too Early Backfire? By: Li, Jing; Luo, Xueming; Lu, Xianghua; Moriguchi, Takeshi. Journal of Marketing. Jul2021, Vol. 85 Issue 4, p123-140. 18p. 5 Charts, 5 Graphs. DOI: 10.1177/0022242920959043.
- Database:
- Business Source Complete
The Double-Edged Effects of E-Commerce Cart Retargeting: Does Retargeting Too Early Backfire?
Consumers often abandon e-commerce carts, so companies are shifting their online advertising budgets to immediate e-commerce cart retargeting (ECR). They presume that early reminder ads, relative to late ones, generate more click-throughs and web revisits. The authors develop a conceptual framework of the double-edged effects of ECR ads and empirically support it with a multistudy, multisetting design. Study 1 involves two field experiments on over 40,500 customers who are randomized to either receive an ECR ad via email and app channels (treatment) or not receive it (control) across different hourly blocks after cart abandonment. The authors find that customers who received an early ECR ad within 30 minutes to one hour after cart abandonment are less likely to make a purchase compared with the control. These findings reveal a causal negative incremental impact of immediate retargeting. In other words, delivering ECR ads too early can engender worse purchase rates than without delivering them, thus wasting online advertising budgets. By contrast, a late ECR ad received one to three days after cart abandonment has a positive incremental impact on customer purchases. In Study 2, another field experiment on 23,900 customers not only replicates the double-edged impact of ECR ads delivered by mobile short message service but also explores cart characteristics that amplify both the negative impact of early ECR ads and positive impact of late ECR ads. These findings offer novel insights into customer responses to online retargeted ads for researchers and managers alike.
Keywords: e-commerce; e-commerce cart retargeting; field experiment; online shopping cart abandonment; recency bump; timing
Empowered by modern e-commerce technologies, many companies shift their online ad budgets to immediate retargeting. That is, companies actively engage in e-commerce cart retargeting (ECR), defined as a form of digital behavioral retargeting wherein online reminder ads are delivered to consumers who had carted products but left without purchasing. For instance, Amazon sends emails to inform customers of their carted products as a call-to-action reminder. Macy's regularly sends short mobile messages to remind customers who inspected and shortlisted products but did not buy ([11]). Indeed, the rate of cart abandonment in e-commerce is high: over 69% customers abandon carts online, and the lost sales amounted to over $4.6 trillion in 2019.[ 6] These statistics suggest a colossal opportunity for firms to deploy ECR ads. It is no wonder that Booking.com, Taobao, and Target deliver app notifications within minutes after customers abandon their shopping carts ([37]).
Such prevalent industry practices of immediate retargeting are fueled by the "recency bump," wherein early reminder ads, relative to late ones, are premised to generate more click-throughs and web revisits ([34]). At first glance, this recency bump makes sense because timing is critical. Ads may have immediate recency: immediately retargeted reminders can be still relevant to user intent, consistent with the common wisdom of "striking while the iron is hot" ([28]).
However, this recency bump can be misleading because it does not measure the causal impact of immediate retargeting. It simply measures the different consumer responses between early and late ECR ads (among the treatment group with all retargeted ads), yet the latter is not a valid comparison baseline for the former due to many alternative explanations (i.e., missing the control group without retargeted ads). For example, consumers who have recently filled an e-commerce cart may be intrinsically more likely to purchase than a consumer for whom it is longer ago because the latter consumer's revealed hesitance may indicate a lack of purchase intent. If so, it is customer self-selection or other confounds, rather than ad recency, that causes the purchase bump. Thus, the valid comparison baseline of a randomized immediate control (without early ECR ads) with similar consumers is required to scientifically quantify the causal impact of immediate retargeting (with early ECR ads).[ 7]
Worse, immediate retargeting, relative to the control, may annoy customers and backfire. That is, as consumers' memory has not faded yet, an early ECR ad sent within minutes after cart abandonment may be too pushy and may seem like the retailer is telling them what to do so that it can make more profits, which can trigger ad annoyance and thus lead consumers to purchase less (e.g., [ 1]; [14]; [41]; [45]).
Against this backdrop, we develop a conceptual framework of the double-edged effects of ECR and empirically support it with a multistudy, multisetting design. Study 1 involves two field experiments on over 40,500 customers who are randomized to either receive an ECR ad via email and app channels (treatment) or not receive it (control) across different hourly blocks after cart abandonment. Results show that in the absence of ECR ads, customer purchases in the control group decayed over time, in line with the memory decay literature. However, relative to the early control, early ECR ads sent within 30 minutes to one hour after cart abandonment have a significantly negative incremental impact on customer purchases. That is, the immediate retargeting is less effective than the randomized early control. In other words, the purchase rate with early ECR ads is even worse than that without them. By contrast, a late ECR ad sent 24–72 hours after cart abandonment has a positive incremental effect: late retargeting ads are more effective than the randomized late control. While the early retargeting treatment generates higher purchase rates than the late retargeting treatment, the early control has even higher purchase rates than the late control. Thus, the causal incremental impact is negative for early ECR but positive for late ECR, in support of the double-edged effects of ECR ads on customer purchases.
Study 2 involves another field experiment on over 23,900 customers from a different company with ECR ads delivered by mobile short message service (SMS). The results first replicate the double-edged impact of ECR ads. Furthermore, because customers have different reasons for cart abandonment, they may have quite different purchase responses to ECR ads. Leveraging the detailed clickstream data on shopping cart characteristics such as product quantity and product prices, we find that the double-edged effects of ECR ads are significantly moderated by these cart features. That is, both the negative impact of early ECR ads and the positive impact of late ECR ads are amplified when the products in the retargeted carts are of a larger quantity and at higher average prices.
Our findings contribute to the literature in three key ways. ( 1) Substantively, as Table 1 shows, we are among the first to reveal a causal adverse incremental impact of immediate retargeting on customer purchases. Advancing prior research on retargeting ([ 7]; [17]; [21]; [36]), we not only conceptually differentiate early ECR from late ECR but also empirically demonstrate the double-edged effects of ECR ads and explore the moderated effects. ( 2) Methodologically speaking, we leverage a multistudy, multisetting research design with three large-scale randomized field experiments based on a fine-grained hourly level of retargeted ads and over 64,000 customers from different companies, which can rigorously test the causal incremental effects of early and late ECR ads and attain a higher generalizability of our findings. ( 3) Managerially, companies should not blindly follow the recency bump and shift their online ad budgets to immediate retargeting. Delivering the ECR ads too early can engender worse purchase rates than without delivering them, thus wasting online advertising budgets. Prudent advertisers ought to match the timing of ECR ads with the retargeted cart features (for detailed research and managerial implications, see the "Discussion and Implications" section).
Graph
Table 1. Literature Gap.
| Negative Effects of Ads | Positive Effects of Ads |
|---|
| Retargeting | Our article | Lambrecht and Tucker (2013); Bleier and Eisenbeiss (2015); Hoban and Bucklin (2015); Johnson, Lewis, and Nubbemeyer (2017); Sahni et al. (2019) |
| Nonretargeting | Aaker and Bruzzone (1985); Yoo and Kim (2005); Goldstein et al. (2014); Jenkins et al. (2016) | Bettman (1979); Alba and Chattopadhyay (1985); Tellis (1988); Lewis and Reiley (2014); Van Heerde et al. (2004, 2013) |
A recent stream of research in marketing has examined retargeting ads ([ 7]; [17]; [21]; [36]). As Table 2 shows, researchers investigated with days or weeks after abandonment. That is, retargeted ads in prior studies were sent in the first few days or weeks after consumers left the focal website. In contrast, we examine with hours: our immediate retargeted ads are sent within 30 minutes to one hour after cart abandonment. Because the retargeting technology aims to "reduce the time lag between the consumers leaving the website and the beginning of the campaign to almost zero" ([36], p. 401), a finer-grained time interval with hours (relative to days or weeks) after abandonment can more accurately capture the immediacy in retargeted reminder ads.
Graph
Table 2. A Comparison of Prior Studies on Retargeted Ads.
| Article | Timing Interval | IV | Channel | DV | Moderators | Key Findings |
|---|
| Lambrecht and Tucker (2013) | Days after abandonment | Ad content | Banner | Customer purchase | Browsing review website | Dynamic retargeted ads (relative to generic ads) have a positive effect for consumers who browsed a review website. |
| Bleier and Eisenbeiss (2015) | Weeks after abandonment | Ad content | Banner | Click-throughs | — | Retargeted personalization ads (relative to nonpersonalization ads) have a positive effect, but quickly lose effectiveness as time (in days) passes since that last visit. |
| Hoban and Bucklin (2015) | Days after abandonment | Ad content | Banner | Web revisit | Pre-experiment stages | Retargeted ads (relative to charity ads) have a positive effect for visitors except those without creating the account. |
| Johnson, Lewis, and Nubbemeyer (2017) | Days after abandonment | Ad content | Banner | Website visit | — | Retargeted ads (relative to ghost ads) have a positive effect. |
| Sahni, Narayanan, and Kalyanam (2019) | Days after abandonment | Ad frequency and timing | Banner | Web revisit | — | Retargeted ads (relative to control) have a positive effect and are most effective for the first day of the first week. |
| Our study | Hours after abandonment | Early and late ECR ads | Email, app, mobile SMS | Customer purchase | Cart features such as product quantity and prices | Retargeted ads have double-edged effects: early ECR ads within 30 minutes to one hour (relative to the early control) have a negative effect, while late ECR ads in one to three days (relative to the late control) have a positive effect. |
50022242920959040 Notes: IV = independent variable, DV = dependent variable.
In addition, prior studies focused on ad personalization and compared different ad copies ([ 7]; [21]), whereas we put the spotlight on the causal effects of early and late retargeted ads. For both early and late ECR ads, we have the randomized early and late controls to scientifically identify the incremental effects. Recently, [36] examined the frequency and timing of retargeted ads at daily level. They found that the effect of frequent retargeting ads is positive and largest for the first day within the first week. We extend their study by examining retargeted ads at hourly level, uncovering the potential annoyance effect of retargeted ads when delivered too early and exploring the moderating role of cart characteristics.
Furthermore, prior works rely on one channel—namely, internet banners—to deliver the retargeted ads ([ 7]; [17]; [21]; [36]). By contrast, we use multiple channels: email, app, and mobile SMS, which enhances the generalizability of the findings across different customer touchpoints. Indeed, companies are now retargeting their customers via emails, app notifications, and SMS in an omnichannel manner ([11]; [37]).
In addition, whereas most prior studies rely on website revisits and click-throughs (cf. [21]), we use customer purchases to measure the outcome of retargeting. While web clicks and visits are important, they are upper-funnel metrics heralding sales revenues. By contrast, customer purchases are lower-funnel outcomes directly related to sales revenues for companies. Furthermore, advancing prior studies on retargeting consumers who abandoning websites in general (some of them just browse around, while others inspect product details), we take a deeper dive into the consumer decision-making journey by focusing on retargeting consumers who have placed products in their carts but then left the online store. Extending [21] and other studies that documented the positive impact of retargeting sent in days after abandonment, we uncover the negative impact of immediate retargeted ads delivered within the first hour after cart abandonment.
Figure 1 presents our conceptual framework of the negative incremental impact of early ECR ads and positive incremental impact of late ECR ads on customer purchases. In our framework, the timing of retargeted ads refers to the time lag (e.g., hours, days) between a consumer abandoning the online shopping cart without buying and the start of retargeting ad campaigns. Specifically, early ECR ads are delivered to customers within the first hour after cart abandonment,[ 8] whereas late ECR ads are delivered at least one day after cart abandonment.
Graph: Figure 1. Conceptual framework.Notes: Early = within the first hour after cart abandonment; Late = one to three days after cart abandonment. This framework is empirically supported by our multistudy, multisetting research design with three large-scale randomized field experiments on over 64,000 customers retargeted by email, app, and SMS ads of different companies.
As Figure 1 illustrates, in the absence of ECR ads, customer purchase rates decrease over time with a downward trend in the control group. This is because, according to the memory decay theory ([ 8]; [30]; [40]), after consumers abandon the shopping carts, their memory of the products fades over time; thus, their purchase probability of the carted products dwindles as the time elapses after abandonment. Ad reminders then can be leveraged to rekindle this memory, as the ability of ads to remind consumers is fairly well established ([ 2]; [ 6]).
However, the incremental effects of early and late ECRS ads, over the early and late control, are not straightforward. Specifically, our conceptual framework posits that early ECR ads, relative to the early control, have a negative incremental impact on customer purchases, whereas late ECR ads, relative to the late control, have a positive incremental impact. This contrasting pattern results from the two driving forces: negative ad annoyance and positive ad reminder.
On the one hand, ads may annoy consumers. Prior studies have pointed out some adverse effects of ads. For example, [45] note that fast animation banner ads can annoy customers and result in negative attitudes toward the advertisers. Others find that consumers are irritated when exposed to commercial ads that are too strident and frequent ([ 1]; [ 9]; [33]). By and large, the literature suggests that ad repetition may annoy consumers and negatively affect the purchase funnel ([41]) because it interrupts consumer goals, such as surfing the internet ([14]) and accomplishing a task online ([16]). Extending this stream of research that frequent ads engender consumer annoyance, we note that the one-time ECR ad may also annoy consumers when it is delivered too early.[ 9]
On the other hand, ads may remind consumers. Advertising can allow brands to signal superior quality over rivals and commendably remind consumers about their products ([ 5]; [23]; [31]; [44]). Viewing reminder ads can rekindle memories[10] associated with the advertised products and thus help consumers recall the focal brands ([ 2]; [ 6]; [42]). In other words, advertising can persuade consumers and enable advertisers to win in the marketplace (i.e., through output interference and displacement of other ads; see [22]; [36]; [43]). These two competing forces lead to the differential effects between early and late ECR ads, as we elaborate next.
When retargeted ads are deployed as soon as customers abandon their shopping carts, their memories have not faded yet, so there is little benefit from rekindling memory (i.e., low positive ad reminder effect; [39]; [44], [43]). However, consumers may feel a high level of the negative ad annoyance effect. This is because as consumers' memories have not wilted yet, early ECR ads (relative to a control without early ECR ads) sent within minutes after cart abandonment may be too pushy and seem like the retailer is telling them what to do so that it can make more profits. This can trigger ad annoyance and thus negatively influence customer purchases. In other words, very early retargeting comes across as too pushy, almost like a too-insistent salesperson who desperately wants customers to buy but actually annoys them and ends up with fewer sales ([ 3]; [13]; [27]). By contrast, a control group without early ECR ads has the same time elapse after cart abandonment but no such negative ad annoyance because it has no reminder ads served. Thus, to the extent that early ECR ads (relative to a control without early ECR ads) lead to a high level of negative ad annoyance but low positive ad reminder, early ECR ads likely backfire with a negative incremental impact on customer purchases.
- H1: Relative to the randomized early control, early ECR ads backfire with a negative incremental impact on customer purchases.
In the case of late ECR ads, consumer memory has faded extensively, and the reminders help overcome this. That is, as the memory wanes, late ECR ads (relative to a control group without late ECR ads) can rekindle the rusty memory of the carted products, thus leading to a high positive ad reminder effect ([ 2]; [ 6]; [22]). Furthermore, because of the extensive memory loss, late ECR ads may not be too pushy to consumers and thus trigger little ad annoyance. The late control without late ECR ads also has the same time elapse after cart abandonment but no such positive ad reminder effect, because no ads are served. Thus, to the extent that late ECR ads (relative to a control group without late ECR ads) lead to a high positive ad reminder effect but low negative ad annoyance, late ECR ads likely have a positive incremental impact on customer purchases.
- H2: Relative to the randomized late control, late ECR ads have a positive incremental impact on customer purchases.
A major Japanese online fashion retailer (that wishes to remain anonymous) cooperated with us to conduct a set of field experiments. The retailer sells fashion products such as clothing, shoes, and handbags, in addition to household items. The retailer targets a wide variety of customers, ranging from children to older adults, and its core customers are men and women aged 20–45 years. The retailer provided us data on customer demographics, such as gender, age, area of residence, and customer tenure, in addition to purchase history, clickstream browsing, and shopping cart data. The time window of the data collected covers three periods: six months before the experiments, during the experiments, and one month after the experiments. From September 21 to October 25, 2016, the retailer conducted two randomized field experiments.
The retailer has two major communication channels: email and a messaging app called Line (similar to WhatsApp's dominance in the United States, Line is the most popular mobile messaging app in Japan). Thus, in Experiment 1, the retargeting message was sent via email to a random sample of 33,234 customers. It is worth noting that email is also the most popular retargeting channel in the United States. In Experiment 2, the retargeting message was sent via Line to a different random sample of 7,314 customers. This smaller sample size reflects the fact that the retailer has many fewer users using its mobile app, which registers users for receiving updates from the retailer on Line. Because customers self-select the email or mobile app channel, customers updated through Line and those updated through email differed in their patterns of shopping behavior. To account for this difference, we conducted two separate experiments to ensure the generalizability of our results.
The research design is similar in the two experiments: the company randomly assigned its customers into 16 groups (8 hour blocks × 2 retargeting conditions). After extensive consultation among the research team who ensured experimental rigor and company executives who oversaw the experimental execution, the retargeted customers in the treatment groups were sent reminder messages in the eight blocks:.5, 1, 3, 6, 9, 12, 24, and 72 hours after cart abandonment. The retailer also had randomized control groups—customers who were not retargeted and did not receive such messages—for each of the eight blocks. Thus, each of the eight retargeting timings had a unique pair of treatment and control groups, and each pair has the same amount of time elapsed after cart abandonment. This is a crucial feature of our experimental design because it enables us to identify the causal incremental effect of the specific hour block while estimating the whole data set simultaneously. In other words, the randomized control conditions empower us to reveal the causal effects of ECR while accounting for many alternative explanations such as the general loss of interests in the carted product over time (e.g., customers have bought that or a different product at another store), seasonality, and competition effects in the marketplace.
The retailer's retargeting ads include product information (brand name, category name, and price). It sent the retargeting message to customers who had abandoned only one product in their shopping cart. For these customers, the retargeted product in the message is the same as the abandoned product; this allows them to more precisely identify the effects of the product-specific retargeting message.[11] Web Appendix B, Panel A, presents some examples of the retargeting message, which contains no new information or price incentives; they are simply reminders about the carted product that was not purchased prior to the experiment ([36]).
As for the experimental execution, if the retailer observed customers to have abandoned the cart and forgone purchasing the product for half an hour, for example, these customers were randomly assigned into either the retargeting treatment or control group. Thus, customers were randomly assigned into all other experimental cells in both Experiments 1 and 2, allowing us to estimate the causal effects. To avoid customer complaints of receiving messages late in the night, the retailer has the policy of not sending messages to customers between 10:00 p.m. and 9:00 a.m. Therefore, some messages could not be sent to subjects in the treatment group who had abandoned the cart late in the day. Our results are robust to additional analyses accounting for bias from this messaging policy.[12] The analyses included propensity score matching, which was used to balance the subjects of the treatment and control groups for each of the eight timings. The variables in propensity score matching were age, gender, area of residence, customer tenure, total money spent (in JPY) in the past six months, number of products purchased in the past six months, and dummy variables corresponding to the time at which the carts were abandoned.
The characteristics of final subjects in the treatment and control groups are summarized in Web Appendixes C and D. According to the data presented in these appendices, the treatment and control groups did not significantly differ with respect to demographics and past purchases for each of the eight hour blocks. Web Appendix B, Panel B indicates that the distribution of the product categories was highly similar across the treatment and control groups. Therefore, the data passed the randomization checks.
Figure 2, Panel A, illustrates the comparison of the purchase rates between the retargeted group (marked by dark bars) and control group (marked by light bars) for the various timings in Experiment 1, along with their 95% confidence intervals. The purchases are measured within one month after sending the retargeting ads to the subjects. In other words, the purchase window is not constrained to the hour block of the retargeting message, but rather one month after because it can take a while before the purchase happens after the message has been received.
Graph: Figure 2. Model-free evidence for the purchase rates of treatment and control (retargeting channels: email and app).Notes: Error bars here represent ±1 standard error.
Consistent with the theory of organic decaying memory (e.g., [ 4]; [ 8]; [40]), the control group exhibited a generally downward trend in purchase rates, attributable to the fading memory of the carted products over time when not retargeted. In other words, for the control group without retargeting ads, the organic purchase rate decreases over time, with the highest at.5 hours and 1 hour after cart abandonment, and the lowest at 72 hours after cart abandonment.
For the treatment groups, the ad for early ECR (.5 hour or 1 hour) had the highest purchase rate of 13.1%, greater than the other hour blocks ads for late ECR. At first glance, in the absence of a comparison with the control group, one might erroneously conclude that early ECR is more effective and beneficial than late ECR, as was done in the recency bump ([28]; [34]). However, purchase rate comparisons between immediate and late retargeting yield invalid comparisons because user self-selection is not controlled for—users receiving the later message are likely to have not purchased the product for a longer time after abandoning the cart. In other words, consumers who receive an early ECR ad (relative to a late ECR ad) may have higher purchase intent and buy more even without the ad. Luckily, we have the randomized controls (where no retargeting was involved but with the same amount of time elapsed after abandonment), which account for user self-selection such as consumer memory decay over time, lost interests in the products, competition, seasonality, or any other observed or unobserved confounds. Thus, comparisons with these randomized controls can effectively parse these confounds from the retargeting timing effects. That is, we can use the incremental effectiveness, by comparing retargeted users with the randomized control within each of the specific hour block groups, to account for this self-selection bias. Such comparison reveals that in the control group, the purchase rate was also highest for the half-hour and one-hour blocks, even higher than that in their respective treatment group. Thus, early ECR within half an hour and one hour had significantly negative effects on purchase rates relative to their respective control groups (.5 hour: 13.1% vs. 15.2%, p =.012; 1 hour: 13.4% vs. 16.7%, p <.01). That is, purchase rates with the early ECR ad are even significantly lower than those without it, thus wasting online ad budgets. Consequently, simple absolute purchases are not causal and can be misleading when used as a measure of retargeting success. By using the relative purchases incremental to the early control, we reveal that early ECR ads can actually backfire. That is, immediate retargeting after cart abandonment has a causal adverse impact on customer purchases. Therefore, H1 is initially supported by such model-free evidence.
According to Figure 2, Panel A, messages sent in the hour blocks of 3, 6, and 9 hours after cart abandonment have no significant effects relative to the control baseline (3 hours: 12.6% vs. 14.3%, p =.114; 6 hours: 11.2% vs. 10.4%, p =.523; 9 hours: 11.7% vs. 9.7%, p =.142). Such zero incremental effect of middle hour blocks makes sense because of the trade-off between the negative ad annoyance and positive ad reminder effect (i.e., these two forces may cancel each other, thus leading to insignificant effects in the middle hour blocks). However, late ECR ads at 24 hours or 72 hours had significantly positive effects on incremental purchases (24 hours: 8.3% vs. 6.0%, p <.01; 72 hours: 4.8% vs. 1.8%, p <.01) over the late control baseline. Thus, H2 is initially supported as well.
Figure 2, Panel B, presents the purchase rates across the hour blocks in the app channel–based retargeting message. Again, early ECR had negative effects on incremental purchase rates (half-hour: 18.0% vs. 21.1%, p =.098; one hour: 14.4% vs. 18.8%, p =.074). The treatments and controls in the 3-, 6-, 9-, and 12-hour blocks did not significantly differ (3 hours: 13.3% vs. 13.7%, p =.842; 6 hours: 12.9% vs. 14.0%, p =.673; 9 hours: 12.2% vs. 8.4%, p =.060; 12 hours: 14.4% vs. 10.2%, p =.099). Nevertheless, late ECR at 24 and 72 hours had significantly positive effects on incremental purchase rates (24 hours: 8.0% vs. 4.0%, p <.01; 72 hours: 6.2% vs. 2.8%, p =.037), replicating the pattern observed for the email channel retargeting message. Thus, these initial model-free results support the double-edged effects of ECR: whereas early ECR has a negative incremental impact, late ECR has a positive incremental impact on customer purchases.
We formally test H1 and H2 by using a moderated logit regression model as follows.
dij=1,if user i in hour block j makes a purchase within the next month0,otherwise.1
This purchasing decision between 1 and 0 is based on a latent-utility function U. Specifically, the differences in purchase decision between the retargeting (treatment) and control groups are moderated by the various hour blocks.
Uij=γo+∑j∈Jγ1jRetargetingij×Hourij+γ2Retargetingij+∑j∈Jγ3jHourij+γkWij+ϵijj = 1, 2,..., J hour blocks,2
where Retargetingij is the treatment variable (1 and 0 represent the retargeting treatment and control, respectively), and Hourij denotes the hour blocks (.5, 1, 3, 6, 9, 12, 24 and 72 hours, and the middle-range block of 6 hours is the baseline). Wij is a vector of covariates (including the customer's gender, age, membership, area of residence, tenure, past message received, past shopping frequency, past shopping expenditure, day fixed effects, and time-of-day fixed effects).
Table 3 reports the results. Compared with the middle hour block, the early and late hour blocks had significantly positive and negative effects, respectively, on purchase rates (all ps <.01). That is, similar to previous model-free results and consistent with the memory decay theory (e.g., [ 4]; [ 8]; [40]), the purchase rate generally had an organic downward trend over time if there were no retargeting messages.
Graph
Table 3. Regression Results on Incremental Retargeting Effects with Hourly Block Interactions.
| Email Channel | Email Channel | App Channel | App Channel |
|---|
| .5 h × Retargeting (H1: −) | −.321** | −.324** | −.307** | −.315** |
| (.136) | (.136) | (.142) | (.142) |
| 1 h × Retargeting (H1: −) | −.336** | −.339** | −.232* | −.248* |
| (.132) | (.132) | (.147) | (.148) |
| 3 h × Retargeting | −.025 | −.027 | .0490 | .0380 |
| (.049) | (.069) | (.290) | (.291) |
| 9 h × Retargeting | .132 | .125 | .243 | .225 |
| (.182) | (.183) | (.465) | (.466) |
| 12 h × Retargeting | .166 | .167 | .482 | .472 |
| (.159) | (.159) | (.319) | (.319) |
| 24 h × Retargeting (H2: +) | .451** | .454** | .824** | .808** |
| | (.191) | (.199) | (.341) | (.342) |
| 72 h × Retargeting (H2: +) | .950*** | .954*** | .923** | .912** |
| (.350) | (.352) | (.461) | (.461) |
| .5 h | .527*** | .506*** | .502*** | .497*** |
| (.0963) | (.0965) | (.168) | (.168) |
| 1 h | .441*** | .405*** | .356** | .369** |
| (.0934) | (.0936) | (.170) | (.180) |
| 3 h | .358*** | .341*** | −.0195 | .000327 |
| (.105) | (.105) | (.203) | (.203) |
| 9 h | −.0801 | −.0991 | −.211 | −.203 |
| (.133) | (.133) | (.275) | (.276) |
| 12 h | −.110 | −.131 | −.358 | −.348 |
| (.116) | (.116) | (.233) | (.233) |
| 24 h | −.778*** | −.767*** | −1.363*** | −1.349*** |
| (.137) | (.137) | (.262) | (.263) |
| 72 h | −1.868*** | −1.839*** | −1.730*** | −1.698*** |
| (.241) | (.241) | (.369) | (.369) |
| (Baseline: 6 h) | | | | |
| Retargeting | .0741 | .0723 | −.0889 | −.0782 |
| (Baseline: control) | (.116) | (.116) | (.211) | (.211) |
| Covariates | Yes | Yes | Yes | Yes |
| Product category effects | No | Yes | No | Yes |
| Time effects | No | Yes | No | Yes |
| Constant | −2.146*** | −2.136*** | −1.819*** | −1.950*** |
| (.0833) | (.105) | (.147) | (.219) |
| Pseudo R2 | .0211 | .0269 | .0377 | .0407 |
| N | 33,234 | 33,234 | 7,314 | 7,314 |
- 60022242920959040 *p <.1.
- 70022242920959040 **p <.05.
- 80022242920959040 ***p <.01.
- 90022242920959040 Notes: Robust standard errors in parentheses.
Our hypotheses pertain to the interactions between hour block and the retargeting treatment. The results in Table 3 consistently suggest that the interaction effects between early hour blocks (.5 and 1 hours) and retargeting on incremental purchase rate are significantly negative in both the email and app channels (most p <.05). As such, these results support the negative effects of early ECR in H1. In addition, the interaction effects between late hour blocks (24 and 72 hours) and retargeting on incremental purchase rate are significantly positive in both email and app channels (at least p <.05), thus supporting the positive effects of late ECR in H2.[13]
Moreover, Figure 3 plots the model-based incremental impact of retargeting (coefficients in Table 3), which visualizes that early ECR ads (in the.5- and 1-hour blocks) have a significantly negative incremental impact, while late ECR ads (in the 24- and 72-hour blocks) have a significantly positive incremental impact on customer purchases for both the email channel in Experiment 1 and app channel in Experiment 2.
Graph: Figure 3. Model-based evidence for the incremental purchases of treatment over control (retargeting channels: email and app).Notes: These figures plot the coefficients and the robust standard errors in Columns 2 and 4 of Table 3. Baseline is six hours after cart abandonment and without retargeted ads.
Furthermore, to more directly test the effects of early and late ECR ads, we combine the.5- and 1-hour blocks into the "Early" group; the 3-, 6-, 9-, and 12-hour blocks into the "Middle" group; and the 24- and 72-hour blocks in the "Late" group. (Web Appendix E visualizes the model-free evidence.) Then, we run the regression models and report the results in Table 4. Again, the interaction effects between Early and Retargeting on incremental purchase rates are significantly negative in both the email and app channels (at least p <.05), in support of H1. Furthermore, the interaction effects between Late and Retargeting on incremental purchase rates are significantly positive in both the email and app channels (at least p <.05), in support of H2.
Graph
Table 4. Regression Results on Incremental Retargeting Effects with Hourly Block Interactions.
| Email Channel | Email Channel | App Channel | App Channel |
|---|
| Early × Retargeting (H1: −) | −.284*** | −.285*** | −.369** | −.370** |
| (.0725) | (.0727) | (.151) | (.151) |
| Late × Retargeting H2: + | .577*** | .582*** | .630** | .624** |
| (.134) | (.134) | (.251) | (.252) |
| Early | .408*** | .393*** | .653*** | .643*** |
| (.0508) | (.0509) | (.106) | (.107) |
| Late | −1.167*** | −1.134*** | −1.286*** | −1.276*** |
| (.105) | (.105) | (.201) | (.201) |
| Retargeting | .0596 | .0555 | .135 | .137 |
| (.0553) | (.0554) | (.114) | (.114) |
| (Baseline: Middle) | | | | |
| Covariates | YES | YES | YES | YES |
| Product Category Effects | NO | YES | NO | YES |
| Time Effects | NO | YES | NO | YES |
| Constant | −2.063*** | −2.073*** | −2.017*** | −2.146*** |
| (.0395) | (.0753) | (.0824) | (.180) |
| Pseudo R2 | .0182 | .0242 | .0348 | .0379 |
| N | 33,234 | 33,218 | 7,314 | 7,313 |
- 100022242920959040 *p <.1.
- 110022242920959040 **p <.05.
- 120022242920959040 ***p <.01.
- 130022242920959040 Notes: Robust standard errors in parentheses. Early = the.5- and 1-hour blocks; Middle = the 3-, 6-, 9-, and 12-hour blocks; Late = the 24- and 72-hour blocks.
Overall, these model-free and model-based results provide consistent empirical evidence for H1 and H2 and thus strongly support the double-edged effects of ECR ads (the negative incremental impact of early ECR and positive incremental impact of late ECR) on customer purchases across Experiments 1 and 2.
The aim of Study 2 is twofold. First, it aims to replicate the double-edged effects of H1 and H2 with a different company to improve the generalizability of the findings. Here we engaged a different retailer and used a different channel of SMS to deliver the ECR ads. Our anonymous corporate partner in Study 2 is a category killer (focusing on maternal and baby products) in China. Considering that our partner in Study 1 was a fashion retailer in Japan, our research settings cover more than one country and two different companies with multiple product lines. Second, Study 2 empirically explores the moderated effects for the double-edged effects of ECR. Cart abandonment in Study 2 involves multiple products left without purchasing. This setting enables us to effectively identify cart characteristics such as the quantity and prices of products left in the retargeted carts to explore the moderated effects, in addition to replicating the double-edged effects of ECR.
Study 2 involves a retargeting message delivered via SMS, thus complementing Study 1's focus on the email and mobile app channels. Compared with email and the mobile messaging app, SMS delivery is displayed as a banner on personal devices ([20]), and the probability that people receive and read the SMS message might be higher ([24]). In addition, SMS promotions are gaining popularity among companies such as Macy's and Target in the United States. Our retail partner in Study 2 is similar to Babies R Us in the United States. The retailer sells a wide variety of maternal and infant supplies, including diapers, infant formula, equipment, toys, baby clothes, and household items. Its customers are primarily young parents with children under four years old. Our retailer partner sent targeted message through SMS to its customers after they had abandoned their carts online (this constituted the triggering event). The experiment involved a random sample of 23,914 customers and was conducted from March 6 to March 9, 2017. To ensure generalizability, the experimental designs of Study 2 were similar to those in Study 1. During our experiment window, if consumers left a new product in the shopping cart, they entered our sample pool and would receive the retargeting message treatment in hours after cart abandonment (or not receive any message if he or she was in the control group). In addition to consumers' demographics and past purchase information, we collected the shopping cart characteristics based on the clickstream data.
The company randomly assigned the customers into eight experiment groups (4 hour blocks × 2 retargeting conditions). The hour blocks were 1, 3, 9, and 24 hours after the first shopping cart abandonment during our time window. Given time and resource limitations, other hour blocks could not be tested. The company also determined these four hour blocks to be the most common in the local market. As per the standard practice in Chinese e-commerce, the retailer had the mobile numbers of its customers. Their customers are required to provide their mobile numbers when registering as a member on the retailer's website, and this number is used to authenticate their membership. As in Study 1, the experiment in Study 2 had a between-subjects design, where consumers neither were in multiple experimental conditions nor received more than one SMS message. All subjects were customers who made at least one purchase in the six-month period prior to the experiment. Web Appendix B, Panel C, presents an example of the SMS retargeting treatment message.
An extension in Study 2 is the execution of experiment randomization. In Study 1, customers were randomly assigned to either the retargeting treatment or control groups within each hour block (e.g., 1 hour or 24 hours after cart abandonment), thus allowing for estimating causal effects within each hour block for early and late ECR ads. However, Study 1 did not randomize the hour blocks ex ante by using an intent-to-treat (ITT) approach ([12]; [17]). Thus, across the hour blocks, customers might be different due to a self-selection bias (i.e., customers who received ECR ads 24 hours later may be intrinsically less likely to buy the product than those 1 hour later). To further account for this potential bias, Study 2 also randomizes the hour blocks, besides the random assignment of treatment or control groups. More specifically, customers are randomly assigned into the treatment and control across all hour blocks ex ante by using the ITT approach ([12]; [17]; [21]).[14] This ITT execution ensures that all individuals are the same ex ante, regardless whether they received the early or late ECR in a specific hour block. In other words, such randomization ensures the unbiasedness of the incremental effects of retargeting (i.e., differences between the ECR and control groups) across all hour blocks. Thus, such ITT estimates allow for not only identifying the causal effects of the early and late ECR ads relative to early and late controls but also directly comparing the causal incremental effect of early ECR with that of late ECR.
The data time window here was six months before the experiment and one week after the experiment. Summary statistics of all variables and randomization check results are reported in the Web Appendix F. The treatment and control groups did not significantly differ with respect to demographics and past purchase characteristics; therefore, the data set passed randomization checks.
Figure 4, Panel A, presents the purchase rates for all hour blocks. We use the purchase rate within a week after sending the retargeting ads to the subjects as our outcome variable. Similar to Study 1, for early ECR (1 hour), the treatment group (marked by dark bars) had a significantly smaller purchase rate than its control group counterpart (marked by gray bars). Furthermore, Figure 4, Panel B, presents the differences in the purchase rates between treatment and control groups for all hour blocks. The results suggest that the difference is negative for the early ECR. That is, relative to the early control, retargeting in the 1-hour block had a significantly lower purchase rate (p <.01). Thus, the incremental effect of early ECR was negative for this SMS channel data, too. Furthermore, the results indicate that the difference is around zero for retargeting in the 3- and 9-hour blocks. Thus, relative to the control, retargeting in the 3- and 9-hour blocks had similar purchase rates. However, the difference is positive for the late ECR. That is, relative to the late control, retargeting in the 24-hour block had a significantly higher purchase rate (p <.01). As such, similar to Study 1 with email and app channels data, the incremental effect of late ECR was positive for this study when using SMS channel data, too.
Graph: Figure 4. Model-free evidence for the differences in purchase rates between treatment and control (retargeting channel: SMS).Notes: Error bars here represent ±1 standard error.
The moderated regression models in Equation 2 were also fitted to the data set of Study 2. The results are reported in Table 5, and the middle range 9 hours block is the baseline. We report effects, as measured by various metrics: purchase incidence (in the logit model) and purchase amount (Tobit model) one week after sending the ECR ads. The results suggest that across all these effectiveness metrics, compared with the middle hour block, the main effects of early hour blocks had significantly positive effects on purchase rates (p <.05), thus supporting the organic decaying memory theory ([ 4]; [ 8]) in the absence of ECR.
Graph
Table 5. Regression Results on Incremental Retargeting Effects with Hourly Block Interactions (SMS Channel).
| Purchase Incidence | Purchase Amount |
|---|
| Logit | Tobit |
|---|
| 1 h × Retargeting (H1: −) | −.545*** | −11.01*** |
| (.130) | (2.921) |
| 3 h × Retargeting | −.269 | −8.794 |
| (.278) | (6.047) |
| 24 h × Retargeting (H2: +) | .104** | 1.237** |
| (.045) | (.613) |
| 1 h | .351** | 3.360*** |
| (.079) | (.866) |
| 3 h | .203*** | 2.342*** |
| (.073) | (.762) |
| 24 h | .0037 | .159 |
| (.089) | (2.045) |
| (Baseline: 9 h) | | |
| Treatment | −.0743 | 4.588** |
| (Baseline: control) | (.0923) | (2.159) |
| Covariates | Yes | Yes |
| Product category effects | Yes | Yes |
| Time effects | Yes | Yes |
| Constant | −2.180*** | 16.78*** |
| (.0570) | (1.310) |
| Pseudo R2/R2 | .039 | .016 |
| N | 23,914 | 23,914 |
- 140022242920959040 *p <.1.
- 150022242920959040 **p <.05.
- 160022242920959040 ***p <.01.
- 170022242920959040 Notes: Robust standard errors in parentheses.
More importantly, the interaction between retargeting treatment and the hour blocks is significantly negative for early ECR (Treatment × 1 hour) on incremental purchase incidence (logit model) and incremental purchase amount (Tobit model) (all p <.01), thereby revealing additional support for H1 with SMS channel data.
In addition, we observed a significantly positive interaction effect for late ECR (Treatment × 24 hour) on incremental purchase incidence (logit model) and incremental purchase amount (Tobit model) (all p <.05), thus again supporting H2.
Because customers have different reasons (e.g., high prices, low budget, multiple products to inspect) for cart abandonment ([11]; [19]; [25]), they may have quite different probabilities of purchasing after viewing ECR ads. Since we have detailed clickstream data on cart characteristics such as product quantity and product prices, we can further extend prior literature ([ 7]; [21]; [36]) by exploring whether these cart characteristics may moderate the double-edged effects of ECR ads.
It is plausible that when the products in retargeted carts are of a larger quantity and at higher average prices, the early ECR may induce even more ad annoyance among shoppers because they may feel the retailer is pushing them to buy more expensive products in a larger amount to make more profits and hence react more negatively ([14]; [45]). Meanwhile, when the products in retargeted carts are of a larger quantity and at higher average prices, these customers tend to be more serious shoppers (who may buy more with higher interest in the carted products), so their memory of the carted products is less likely to fade quickly ([22]; [39]). Then, the immediate ECR ads sent too early are more likely to engender low ad reminder effect. Thus, by inducing even more ad annoyance but less ad reminder effect, the negative impact of early ECR ads may be enlarged when the products in the retargeted carts are of a larger quantity and at higher average prices. On the other hand, as time elapses after cart abandonment, and consumer memory of the more expensive and larger quantity of products fades extensively, the serious shoppers will be more likely to appreciate the ad reminders rekindling their rusty memory of those inspected products ([35]; [38]; [44], [43]), with even higher ad reminder effects and lower ad annoyance, hence likely strengthening the positive incremental impact of late ECR on customer purchases. As such, the double-edged effects of ECR ads can be moderated by these cart features: not only the negative impact of early ECR ads but also the positive impact of late ECR ads are amplified when the products in the retargeted carts are of a larger quantity and at higher average prices.[15]
To test these moderating effects of the cart characteristics of Product Quantity (Pnum) and Product Price (Pprice), we specify the interaction model in Equation 3.
Uij2=ξo+ξ1Retargetingij×Earlyij×Pnumii+ξ2Retargetingij×Earlyij×Ppriceii+ξ3Retargetingij×Lateij×Pnumii+ξ4Retargetingij×Lateij×Ppriceii+ξ5Retargetingij×Earlyij+ξ6Retargetingij×Lateij+ξ7Retargetingij×Pnumij+ξ8Retargetingij×Ppriceij+ξ9Retargetingij+ξ10Earlyij+ξ11Lateij+ξ12Pnumii+ξ13Ppriceii+ξkWij+εij2.3
where Early is the 1-hour block, and Late is the 24-hour block (baseline is Middle with the 3- and 9-hour blocks). As shown in Table 6, the results suggest that the three-way interaction between Early, Retargeting, and Product Quantity is significantly negative (p <.01) for both purchase incidence and amount. Thus, when the products in retargeted carts are of a larger quantity, the negative effect of early ECR is stronger. In addition, the three-way interaction between Late, Retargeting, and Product Quantity is significantly positive (p <.05). Thus, the positive effect of late ECR is also amplified, when the products in retargeted carts are of a larger quantity. Furthermore, results show that the three-way interaction between Early, Retargeting, and Product Price is significantly negative (p <.01). As such, when the products have higher average prices, the negative effect of early ECR is also amplified. However, the coefficient of the three-way interaction term between Late, Retargeting, and Product Prices is insignificant. Thus, these explorative results provide some evidence that the double-edged effects of ECR ads are moderated by cart characteristics. By and large, both the negative impact of early ECR ads and the positive impact of late ECR ads are amplified when the products in the retargeted carts are of a larger quantity and at higher average prices.
Graph
Table 6. Explorative Results on the Moderating Effects of Cart Characteristics.
| Purchase Incidence | Purchase Amount |
|---|
| Early × Retargeting × Pnum | −.264*** | −1.953*** |
| (.0221) | (.430) |
| Late × Retargeting × Pnum | .0388** | 6.806** |
| (.0196) | (3.369) |
| Early × Retargeting × Pprice | −.214*** | −1.010** |
| (.0636) | (.482) |
| Late × Retargeting × Pprice | .0195 | 4.970 |
| (.0181) | (3.491) |
| Early × Retargeting | −.872*** | −2.198*** |
| (.249) | (.616) |
| Late × Retargeting | .853*** | 2.100*** |
| (.203) | (.622) |
| Retargeting ×Pnum | −.202 | −3.557 |
| (.182) | (2.709) |
| Retargeting × Pprice | −.0398 | −.291 |
| (.0351) | (.908) |
| Early ×Pnum | −.142 | −1.120 |
| (.0982) | (2.362) |
| Early × Pprice | −.129*** | −2.043*** |
| (.0371) | (.568) |
| Late × Pnum | −.425 | −7.225 |
| (.435) | (7.022) |
| Late × Pprice | .0273 | 1.195 |
| (.0431) | (1.071) |
| Early | .286*** | 5.526*** |
| (.089) | (1.449) |
| Late | .0854 | −1.700 |
| (.367) | (1.382) |
| Retargeting | .824 | 1.939 |
| (.602) | (1.108) |
| Pnum | .0649 | 1.528 |
| (.0571) | (1.392) |
| Pprice | −.183*** | −3.632*** |
| (.0195) | (.431) |
| Covariates | Yes | Yes |
| Constant | .553** | 75.27*** |
| (.265) | (8.121) |
| Pseudo R2 | .157 | .086 |
| Observations | 23,914 | 23,914 |
- 180022242920959040 *p <.1.
- 190022242920959040 **p <.05.
- 200022242920959040 ***p <.01.
- 210022242920959040 Notes: Robust standard errors in parentheses. Early = the 1-hour block; Middle = the 3- and 9-hour blocks; and Late = the 24-hour block. Pnum = product quantity (in natural log); Pprice = average product prices (in natural log) of the retargeted carts.
On the basis of multistudy, multisetting data from randomized field experiments, our research reveals that ECR ads have double-edged incremental effects on customer purchases. In particular, an early ECR ad has a negative incremental effect, whereas a late ECR ad has a positive incremental effect. Explorative analyses suggest that such double-edged effects of ECR ads are moderated. Both the negative impact of early ECR and positive impact of late ECR can be amplified when the products in the retargeted carts are of a larger quantity and at higher average prices. Our findings have broad research and managerial implications.
Our findings offer several research implications. We are among the first to reveal a causal adverse incremental impact of immediate retargeting on customer purchases in e-commerce. Extending prior research on retargeting ([ 7]; [21]; [36]), we not only conceptually differentiate early ECR from late ECR but also empirically show the double-edged effects of ECR ads. Our novel insight here is that early ECR ads within the first hour after cart abandonment may backfire with significantly negative incremental effects on customer purchases. This insight is nontrivial for two key reasons. First, it may rectify the wrong one-sided view of the effectiveness of immediate retargeting. By simply comparing purchase responses with early versus late ECR ads in the treatment (as done in the recent bump view) without valid early and late controls, researchers may erroneously conclude that immediate retargeting has a positive impact and is more effective than late retargeting. However, with scientific randomized controls, the opposite is true: the former has a causal negative impact, while the latter has a causal positive impact and is more effective in reality. Thus, research that documents only the positive impact of retargeting ads could overestimate the effect of early ECR ads and should reckon that immediate retargeting within minutes after cart abandonment (ads served too early) might engender harmful impacts on consumer behavior. Second, it may change our vision for the technology–consumer interface. [36], p. 401) note that "retargeting technology aims to reduce the time lag between the consumers leaving the website and the beginning of the campaign to almost zero." We agree and add that the modern retargeting technology is one thing, but consumer response is another. Although technologies can immediately retarget customers based on the fine-grained shopping cart data, doing so too early may actually annoy customers and adversely impact their purchases. Thus, research on the technology–consumer interface should account for the double-edged (both positive and negative) consumer responses to the innovative retargeting technologies.
Furthermore, we leverage a multistudy, multisetting research design with three large-scale randomized field experiments on over 64,000 customers from different companies via a fine-grained hourly level of retargeted ads, which can rigorously test the causal incremental effects of early and late ECR with a higher generalizability in findings. Prior research examined retargeting ads at the daily level and found a generally positive effect on clicks and web revisits ([ 7]; [17]; [36]). In support of this line of research, we find that late ECR ads delivered one to three days after cart abandonment lift customer purchases. Furthermore, extending this stream of research, we are among the first to operationalize immediate retargeting at the hourly level within the first day after cart abandonment and uncover the negative impact on customer purchases of immediate retargeted ads. A finer-grained time interval with hours (vs. days) can more accurately capture the immediacy in retargeting.
In addition, advancing prior studies that focused on retargeted ads' content, placement, and frequency ([ 7]; [17]; [21]; [36]), we are among the first to put the spotlight on the timing of retargeted ads, a crucial but underresearched factor influencing the conversions in e-commerce. Even with perfectly crafted and placed ad content with the appropriate frequency, retargeting campaigns may bypass the opportunity to earn higher purchase responses by not taking into account the timing (early or late) of ECR ads.
Moreover, our explorative findings enrich the understanding of the moderated double-edged effects of ECR ads. Extending the literature ([ 7]; [21]; [36]), we reveal another new insight that both the negative impact of early ECR ads and positive impact of late ECR ads can be amplified when retargeting carts with a larger quantity and higher average price of products. These findings on the moderated double-edged effects are nontrivial because research might over- or underestimate the impact of early and late ECR ads if ignoring the moderating roles of carted product features. Because customers have different reasons for cart abandonment, they will have different probabilities of purchasing after viewing ECR ads ([19]; [25]). This is different from the cross-sectional consumer heterogeneity, because, over time, even the same individual may have different reasons to abandon the shopping cart. Thus, a comprehensive understanding of consumer responses to early and late ECR ads should consider the contextual factors such as cart characteristics. Matching early and late ECR ads with such contextual factors is crucial for the efficacy of retargeted ads in e-commerce.
Relatedly, our findings have implications for the advertising literature. Prior literature has well documented a myriad of ad effects: provide informative content, offer output interference, and displace other ads ([36]; [39]; [44], [43]). We contribute to this literature by uncovering the nuanced timing (early vs. late) effects of ECR ads. Indeed, prior studies on ad annoyance are largely based on the frequency (i.e., ad repetition; [ 1]; [14]; [41]). Advancing these studies, we uncover that the one-time ECR ad may also annoy consumers when it is delivered too early.
Furthermore, the negative effects of early ECR ads yet positive effects of late ECR ads help account for the mixed effects of digital advertising in the literature ([ 1]; [14]; [21]; [26]). In this sense, we extend the literature by suggesting the importance of implementing contextual ads in retargeting (i.e., deliver the ads in the right time, not within the first hour after cart abandonment, and for the right shopping carts). This is critical because marketers might stall if they blindly advertise to customers without accounting for when, how many, and how expensive the carted products are.
Given the prevalence of retargeting ads in practice, our findings provide managers with specific guidance on implementing ECR ads to boost return on investment on retargeting campaigns. First, companies should not heedlessly follow the recency bump and shift all their online ad budgets to immediate retargeting. Delivering the ECR ads too early can engender worse purchase rates than without delivering them. That is, reminder ads sent too soon may annoy consumers and backfire, thus not only squandering ad budgets and but also likely hurting customers' long-term satisfaction. Prudent marketers should resist the temptation of using the immediate retargeting, even though advanced digital e-commerce technologies can deliver retargeting ads within minutes after consumers abandon carts online. Nevertheless, early ECR with price discounts or scarcity framing may allow managers to engender more purchase responses ([25]). However, price discounts are not a panacea: when repeatedly used, they may train strategic customers who purposefully cart products and then wait for price discounts before purchasing.
Second, it is pivotal to scientifically gauge the causal impact of ECR ads. Firms should not rely on the absolute purchases as a measure of success but rather adopt the relative purchase, (i.e., incremental to the control without retargeting). Naively, if not comparing the retargeting with the control, managers may mistakenly conclude that the early ECR is most effective: our data indeed show that if simply observing the absolute effect, the early ECR within one hour induces the highest absolute purchases. Yet, compared with the early control, the early ECR actually backfires with negative incremental purchase responses. Thus, we underscore the importance of scientific experimental methodology for managers to avoid the erroneous conclusion on the true effects of ECR ads.
Furthermore, we find that a late ECR ad can be effective and win back potential customers with an increase in return on investment on advertising. Thus, firms can better deploy ECR ad campaigns with a delay after consumers abandon carts to minimize the negative ad annoyance, as well as maximize the positive ad reminder effects on customer purchases. Indeed, retargeting carts in e-commerce has enormous business potential because over 69% consumers abandon carts online, which amounts to over $4.6 trillion ([11]; [37]). An interesting point is that the right timing of ECR does not incur additional financial costs in retargeting but can significantly lift customer purchases.
Finally, managerial actions call for an appropriate match between the timing of ECR ads and retargeted products. It is necessary to use ECR to cater to different types of cart abandonment; different cases would include carts with a high quantity of products versus carts with only one item, or carts with an expensive product versus those with a cheap one ([19]; [25]). Thus, we reveal the tactic e-commerce retailers can use to more accurately retarget customers with different digital carts. Strategically, firms can decide the time to turn on ECR, depending on its suitability for different types of carts, to maximize conversions. For example, managers win back more customers by implementing late ECR ads for carts with a larger quantity of products abandoned.
Our research has several limitations, which serve as avenues for future studies. First, although our findings are drawn from two countries and different companies, they may not be generalizable to other cultures and products. Thus, more empirical evidence from other settings can be provided in the future. For example, an early reminder might work well for impulse purchases about which customers do not have to ponder much, while products that require a lot of deliberation before purchasing (e.g., cars) might benefit from late ECR. Incorporating the idea of the length of the purchase decision process could also be pertinent, because early versus late in retargeted ads can be a relative concept. Moreover, the strength of our field experiment is about documenting the causal impact of early and late ECR ads on customer purchases, rather than the underlying mechanisms. Future research could investigate the related psychological mechanisms in the lab and explore how privacy concerns, seasonality, and ad competition in retargeting messages regulate the effects of ECR. Furthermore, our results on the moderating role of cart characteristics are exploratory in nature. Future research could use survey data to pinpoint consumers' specific reasons for cart abandonment first and then retarget them with different ad framing and incentives to enhance the efficacy of ECR ads.
Finally, our data are limited to the hourly level within the first several days after cart abandonment. Thus, the effect of even later ECR ads (weeks or months later) is not tested here. Future research might investigate the impact of much later ECR at the weekly or monthly level. Nevertheless, [ 7] find that if the time since last online store visit is over 48 days, the incremental effect of retargeted ads on click-through rates is close to zero. Likewise, [29] show that retargeted ads sent one week after cart abandonment are ineffective in generating incremental customer purchases. [36] find that the effect of retargeted ads on web revisits is positive but becomes negligible by the end of the first week. These findings suggest that too late ECR ads may turn out to be ineffective. Indeed, once too long a time has elapsed (e.g., after several months or years) since cart abandonment, consumer memory of carted products might be totally lost given the large amount of information in social media people are exposed daily. Then, the reminder function of ECR ads will not work anymore because the memory trace is too weak for a reminder ad to be successful: it is difficult to restore or activate the lost memory ([18]). However, if designed with price incentives, too late retargeted ads might still be effective ([25]). Taking a broad perspective of the literature on ad personalization based on customers' preferences and demographic profiles ([21]) and browsing content ([ 7]), future research may also consider how these characteristics regulate consumer purchase responses to too late ECR ads.
In conclusion, this study represents an initial effort in examining the double-edged effects of ECR ads on customer purchases. We hope that our study stimulates future research on ECR, an increasingly important topic in digital marketing.
Supplemental Material, JM.19.0765_Web_Appendix_PDF - The Double-Edged Effects of E-Commerce Cart Retargeting: Does Retargeting Too Early Backfire?
Supplemental Material, JM.19.0765_Web_Appendix_PDF for The Double-Edged Effects of E-Commerce Cart Retargeting: Does Retargeting Too Early Backfire? by Jing Li, Xueming Luo, Xianghua Lu and Takeshi Moriguchi in Journal of Marketing
Footnotes 1 S. Sriram
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Xianghua Lu acknowledges support from the National Natural Science Foundation of China [Grants 71872050 and 91746302]. Takeshi Moriguchi acknowledges support from the JSPS KAKENHI Grant Number JP16H03675.
4 Jing Li https://orcid.org/0000-0002-2242-1987 Xueming Luo https://orcid.org/0000-0002-5009-7854
5 Online supplement: https://doi.org/10.1177/0022242920959043
6 Baymard Institute reports that the highest abandonment rate is on travel sites (81.1%) and the lowest on fashion websites (69.1%) (https://baymard.com/lists/cart-abandonment-rate, accessed September 29, 2020).
7 For ease of exposition, we use "immediate retargeting" and "early ECR ads" interchangeably.
8 Our definition of early ECR ads within one hour after abandonment is in line with industry practices, where immediate retargeting means sending ads within one hour after consumers leave the website in retailing, fashion, travel, and other industries ([11]; [37]). We do not consider time over a week in our late ECR ads because if a very long time has elapsed after cart abandonment, consumer memory can be totally lost and is notoriously difficult to restore ([18]). Indeed, research has found that retargeted ads sent one week after abandonment are ineffective in generating incremental purchases ([29]). We return to this point in the "Discussion and Implications" section. Furthermore, because consumers rarely put cars in shopping carts online (most people would still need to test drive the cars in the physical world offline), immediate retargeting is more applicable to online purchases in business sectors such as retailing and fashion.
9 As we discuss subsequently, the one-time exposure to an immediate ECR ad can stimulate annoyance because very early retargeting, when consumer memory has not faded yet, comes across as too pushy, akin to a too-insistent salesperson. Indeed, for preliminary evidence that early ECR ads lead to ad annoyance among consumers, which then reduces their purchase intention, see Web Appendix A.
Prior psychology literature has noted that human memory decays over time (i.e., forgetting) ([40]). Forgetting is a function of age, perceptual speed, and central executive functioning ([10]), and different people have different memory decay patterns. While some still have a fresh memory after a long time, others forget quickly; thus, unobserved heterogeneity exists across consumers. Consequently, we conducted field experiments to account for such unobserved heterogeneity by randomizing consumers who have the same time elapse after cart abandonment into treatment and control groups.
Among the retailer's customers who abandoned carts, approximately 90% abandoned just one item in their shopping carts.
This policy might bias our results, as it would not affect the control groups but would affect the treatment groups across the hour blocks. Thus, it may be informative to conduct additional analyses with subjects that abandoned carts between 9 a.m. and 12:59 p.m. because they could be assigned to all hour blocks except the 12 hours. We checked the robustness with these subjects and found consistent results (results available on request).
We have also estimated the marginal effects for the logit model (where we hold all other variables at the mean level; [32]) and found consistent results. Results are available on request.
Note that ITT design may have an issue of compliance, where treatment is only administrated to individuals who have not dropped out ([12]).
There could be other arguments for the effects. For example, more expensive products tend to have longer purchase decision processes, so customers may simply need more time to decide. Another aspect might be that ads retargeting a larger basket of products could be less annoying, as consumers might more quickly forget about the specific items in a large assembly. These arguments can be fruitful for future research.
References Aaker David A., Bruzzone Donald E. (1985), "Causes of Irritation in Advertising," Journal of Marketing, 49 (2), 47–57.
Alba Joseph W., Chattopadhyay Amitava. (1985), "Effects of Context and Part-Category Cues on Recall of Competing Brands," Journal of Marketing Research, 22 (3), 340–49.
Babin Barry J., Boles James S., Darden William R. (1995), "Salesperson Stereotypes, Consumer Emotions, and Their Impact on Information Processing," Journal of the Academy of Marketing Science, 23 (2), 94–105.
Baddeley Alan D., Thomson Neil, Buchanan Mary. (1975), "Word Length and the Structure of Short-Term Memory," Journal of Verbal Learning and Verbal Behavior, 14 (6), 575–89.
Bagwell K. (2007), "The Economic Analysis of Advertising," in Handbook of Industrial Organization, Vol. 3, Armstrong Mark, Porter Robert, eds. Amsterdam: Elsevier, 1701–844.
Bettman James R. (1979), "Memory Factors in Consumer Choice: A Review," Journal of Marketing, 43 (2), 37–53.
Bleier Alexander, Eisenbeiss Maik. (2015), "Personalized Online Advertising Effectiveness: The Interplay of What, When, and Where," Marketing Science, 34 (5), 669–88.
Brown John. (1958), "Some Tests of the Decay Theory of Immediate Memory," Quarterly Journal of Experimental Psychology, 10 (1), 12–21.
Burke Marian C., Edell Julie A. (1986), "Ad Reactions Over Time: Capturing Changes in the Real World," Journal of Consumer Research, 13 (1), 114–18.
Fisk J.E., Warr P.B. (1998), "Associative Learning and Short-Term Forgetting As a Function of Age, Perceptual Speed, and Central Executive Functioning," Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 53 (2), P112–21.
Garcia Krista. (2018), "Brands Know They're Being Creepy: Personalization Comes with Known Risks," eMarketer(February 21), https://www.emarketer.com/content/brands-know-they-re-being-creepy.
Gerber Alan, Green Donald. (2012), Field Experiments: Design, Analysis, and Interpretation. New York: W. Norton and Company.
Gillis Claire, Pitt Leyland, Robson Matthew J., Berthon Pierre. (1998), "Communication in the Salesperson/Customer Dyad: An Empirical Investigation," Marketing Intelligence and Planning, 16 (2), 100–106.
Goldstein Daniel G., Suri Siddharth, McAfee R. Preston, Ekstrand-Abueg Matthew, Diaz Fernando. (2014), "The Economic and Cognitive Costs of Bothersome Display Advertisements," Journal of Marketing Research, 51 (6), 742–52.
Hoban Paul R., Bucklin Randolph E. (2015), "Effects of Internet Display Advertising in the Purchase Funnel: Model-Based Insights from a Randomized Field Experiment," Journal of Marketing Research, 52 (3), 375–93.
Jenkins Jeffrey L., Anderson Bonnie B., Vance Anthony, Kirwan C. Brock, Eargle David. (2016), "More Harm Than Good? How Messages That Interrupt Can Make Us Vulnerable," Information Systems Research, 27(4), 880–96.
Johnson Garrett A., Lewis Randall A., Nubbemeyer Elmar I. (2017), "Ghost Ads: Improving the Economics of Measuring Online Ad Effectiveness," Journal of Marketing Research, 54 (6), 867–84.
Kelley Derek H., Gorham Joan. (1988), "Effects of Immediacy on Recall of Information," Communication Education, 37 (3), 198–207.
Kukar-Kinney Monika, Close Angeline G. (2010), "The Determinants of Consumers' Online Shopping Cart Abandonment," Journal of the Academy of Marketing Science, 38 (2), 240–50.
Lai Tung L. (2004), "Service Quality and Perceived Value's Impact on Satisfaction, Intention and Usage of Short Message Service (SMS)," Information Systems Frontiers, 6 (4), 353–68.
Lambrecht Anya, Tucker Catherine. (2013), "When Does Retargeting Work? Information Specificity in Online Advertising," Journal of Marketing Research, 50 (5), 561–76.
Leenheer Jorna, van Heerde Harald J., Bijmolt Tammo H., Smidts Ale. (2007), "Do Loyalty Programs Really Enhance Behavioral Loyalty? An Empirical Analysis Accounting for Self-Selecting Members," International Journal of Research in Marketing, 24 (1), 31–47.
Lewis Randall A., Reiley David H. (2014), "Online Ads and Offline Sales: Measuring the Effect of Retail Advertising via a Controlled Experiment on Yahoo!"Quantitative Marketing and Economics, 12 (3), 235–66.
Luo Xueming, Andrews Michelle, Fang Zheng, Phang Chee W. (2014), "Mobile Targeting," Management Science, 60 (7), 1738–56.
Luo Xueming, Lu Xianghua, Li Jing. (2019), "When and How to Leverage E-Commerce Cart Targeting (ECT): The Relative and Moderated Effects of Scarcity and Price Incentives," Information Systems Research, 30 (4), 1203–27.
Manchanda Puneet, Dubé Jean-Pierre, Goh Khim Y., Chintagunta Pradeep K. (2006), "The Effect of Banner Advertising on Internet Purchasing," Journal of Marketing Research, 43 (1), 98–108.
Martin Steve W. (2017), "Reasons Salespeople Don't Close the Deal," Harvard Business Review, (August 2), https://hbr.org/2017/08/7-reasons-salespeople-dont-close-the-deal.
Moore James. (2013), "Time Means Everything in Programmatic Display," Marketing Land(February 25), https://marketingland.com/the-element-of-time-means-everything-in-programmatic-display-33928.
Moriguchi Takeshi, Xiong Guiyang, Luo Xueming. (2016), "Retargeting Ads for Shopping Cart Recovery: Evidence from Online Field Experiments," working paper.
Mueller Shane T., Seymour Travis L., Kieras David E., Meyer David E. (2003), "Theoretical Implications of Articulatory Duration, Phonological Similarity, and Phonological Complexity in Verbal Working Memory," Journal of Experimental Psychology: Learning, Memory, and Cognition, 29 (6), 1353–80.
Nelson Phillip. (1974), "Advertising as Information," Journal of Political Economy, 82 (4), 729–54.
Norton Edward C., Wang Hua, Ai Chunrong. (2004), "Computing Interaction Effects and Standard Errors in Logit and Probit Models," Stata Journal, 4 (2), 154–67.
Pokrywczynski James, Crowley John H. (1993). "The Influence of Irritating Commercials on Radio Listening Habits," Journal of Radio Studies, 2 (1), 51–68.
Prioleau Frost. (2013), "The Recency Bump: In Retargeting Timing Is Everything," Search Engine Land(March 14), https://searchengineland.com/the-recency-bump-in-retargeting-timing-is-everything-151099.
Raj S.P. (1982), "The Effects of Advertising on High and Low Loyalty Consumer Segments," Journal of Consumer Research, 9 (1), 77–89.
Sahni N.S., Narayanan S., Kalyanam K. (2019), "An Experimental Investigation of the Effects of Retargeted Advertising: The Role of Frequency and Timing," Journal of Marketing Research, 56 (3), 401–18.
Statista (2020), "Primary Reason for Digital Shoppers in the United States to Abandon Their Carts as of November 2018," (accessed September 30), https://www.statista.com/statistics/379508/primary-reason-for-digital-shoppers-to-abandon-carts/.
Suri Rajneesh, Monroe Kent B. (2003), "The Effects of Time Constraints on Consumers' Judgments of Prices and Products," Journal of Consumer Research, 30 (1), 92–104.
Tellis Gerard J. (1988), "Advertising Exposure, Loyalty, and Brand Purchase: A Two-Stage Model of Choice," Journal of Marketing Research, 25 (2), 134–44.
Thorndike Edward L. (1914), The Psychology of Learning. New York: Teachers College.
Todri Vilma, Ghose Anindya, Singh Param V. (2020), "Trade-Offs in Online Advertising: Advertising Effectiveness and Annoyance Dynamics Across the Purchase Funnel," Information Systems Research, 31 (1), 102–25.
Van der Lans Ralf, Pieters Rik, Wedel Michel. (2008), "Research Note—Competitive Brand Salience," Marketing Science, 27 (5), 922–31.
Van Heerde Harald J., Gijsenberg Maarten J., Dekimpe Marnik G., Steenkamp Jan-Benedict E.M. (2013), "Price and Advertising Effectiveness Over the Business Cycle," Journal of Marketing Research, 50 (2), 177–93.
Van Heerde Harald J., Leeflang Peter S.H., Wittink Dick R. (2004), "Decomposing the Sales Promotion Bump with Store Data," Marketing Science, 23 (3), 317–34.
Yoo Chan Y., Kim Kihan. (2005), "Processing of Animation in Online Banner Advertising: The Roles of Cognitive and Emotional Responses," Journal of Interactive Marketing, 19 (4), 18–34.
~~~~~~~~
By Jing Li; Xueming Luo; Xianghua Lu and Takeshi Moriguchi
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 117- The Economic and Social Impacts of Migration on Brand Expenditure: Evidence from Rural India. By: Narayan, Vishal; Kankanhalli, Shreya. Journal of Marketing. Nov2021, Vol. 85 Issue 6, p63-82. 20p. 2 Diagrams, 8 Charts. DOI: 10.1177/00222429211021992.
- Database:
- Business Source Complete
Record: 118- The Effect of Content on Zapping in TV Advertising. By: Becker, Maren; Scholdra, Thomas P.; Berkmann, Manuel; Reinartz, Werner J. Journal of Marketing. Sep2022, p1. DOI: 10.1177/00222429221105818.
Ahead of Print- Database:
- Business Source Complete
Record: 119- The Impact of Advertising Creative Strategy on Advertising Elasticity. By: Dall’Olio, Filippo; Vakratsas, Demetrios. Journal of Marketing. Apr2022, p1. DOI: 10.1177/00222429221074960.
Ahead of Print- Database:
- Business Source Complete
Record: 120- The Impact of Corporate Social Responsibility on Brand Sales: An Accountability Perspective. By: Nickerson, Dionne; Lowe, Michael; Pattabhiramaiah, Adithya; Sorescu, Alina. Journal of Marketing. Mar2022, Vol. 86 Issue 2, p5-28. 24p. 2 Diagrams, 7 Charts, 1 Graph. DOI: 10.1177/00222429211044155.
- Database:
- Business Source Complete
The Impact of Corporate Social Responsibility on Brand Sales: An Accountability Perspective
Consumers are increasingly mindful of corporate social responsibility (CSR) when making purchase and consumption decisions, but evidence of the impact of CSR initiatives on actual purchase decisions is lacking. This article introduces a novel brand accountability–based framework of consumer response to CSR initiatives, which categorizes CSR efforts as "corrective," "compensating," or "cultivating goodwill." Leveraging a database of CSR press releases by leading consumer packaged goods brands, the authors examine the effect of the different types of CSR announcements on brand sales. The findings suggest that CSR initiatives that genuinely aim to reduce a brand's negative externalities ("corrective" and "compensating") lift sales, whereas CSR actions focused on philanthropy ("cultivating goodwill") can hurt sales. The authors propose two moderators—CSR reputation and CSR focus on environmental or social causes—and a mechanism for these effects, which they examine under controlled experimental settings. The experimental results show that, conditional on CSR reputation, consumers perceive varying degrees of sincerity in the different CSR types and that sincerity mediates the effect of CSR type on purchase intentions. Overall, the results suggest that consumers are more inclined to reward firms that directly reduce the negative by-products of their own business practices than to be impressed by public goodwill gestures.
Keywords: corporate social responsibility; sustainability; CSR reputation; brand sincerity; environmental initiatives; social initiatives
Corporate social responsibility (CSR)—defined as discretionary business practices and contributions of corporate resources intended to improve societal well-being ([51]) —is increasingly present in consumer consciousness. As more consumers support brands that contribute to the greater societal good, companies have incentives to engage in some form of CSR ([21]; [44]). Effective CSR can enhance corporate perceptions, differentiate products, and reduce the impact of public relations miscues ([18]; [49]; [62]). However, despite a stream of research that has documented various positive effects of CSR, a causal link between a firm's CSR activities and actual consumer purchase decisions has not been established. Thus, one goal of this research is to provide evidence of the effect of CSR initiatives on brand sales.
Another goal of this article is to provide a categorization of CSR and an examination of the contingent effects of different types of CSR on brand sales. CSR typically spans a wide array of potential activities, including philanthropic community support, environmental initiatives, diversity promotion, employee support, changes to products and supply chains, and corporate governance issues. These activities have been classified in extant literature into categories such as philanthropic versus business practice ([42]), environmental versus product-focused ([47]), or proactive versus reactive ([80]) CSR. The breadth of these categorization schemes, however, can complicate both the study and the efficient managerial deployment of CSR initiatives. We aim to provide structure to this variety by using an important, but understudied, characteristic of CSR: the extent to which CSR addresses a brand's liability and thereby demonstrates accountability in consumers' minds.
Anecdotal evidence suggests that consumers care about brands being accountable for their actions. For instance, hoping to better understand what types of CSR activities consumers would prefer, Coca-Cola recently tested a battery of potential initiatives using a series of consumer focus groups. These initiatives ranged from social to environmental, and from the purely philanthropic (women's economic empowerment) to the seemingly apologetic (helping address obesity). In the end, the initiative that most clearly addressed and preempted the brand's own potential negative social and environmental impact (reduced water consumption) elicited the most favorable consumer response ([19]). Likewise, a recent public survey about CSR found that a majority of respondents preferred that firms adopt business operations aimed at minimizing their own societal and environmental harm ([21]). The degree to which a brand's CSR efforts actually address any of its own negative externalities (i.e., harmful effects on society and the environment) may thus help predict consumer response and guide the management of CSR decisions ([40]).
Using accountability as a basis to address negative externalities, we distinguish between three types of CSR engagement: correcting for the potential negative societal or environmental impact of a brand's business operations by making changes to those operations, compensating for the negative impact of a brand's business operations without making changes to those operations, and cultivating goodwill[ 6] through prosocial acts that are not directly related to any negative impact of a brand's business operations. We argue that this conceptualization captures important and fundamentally distinct CSR-related concerns and expectations among consumers and covers the vast majority of CSR activities. Using this typology, we develop a framework and conduct one of the first large-scale examinations of the effect of different types of CSR on brand sales. We extend this framework and complement the secondary data analysis with experimental evidence that both replicates these results and explores the mechanism underlying the differential effect of CSR type on consumer response.
Our work contributes to the literature in three important ways. First, we provide a novel typology of CSR based on the underexplored concept of firm accountability, whereby firms take responsibility for the consequences of their operations ([28]). We develop our framework based on this typology in a consumer-centric manner, drawing on sociopsychological theory, invoking the role of responsibility and restitution in attitude change ([16]). This typology encourages greater integrity in the practice of CSR by highlighting the alignment of societal and business interests. It also provides more actionable managerial insights because it directly links CSR initiatives to firm actions and is more granular than previous categorizations (e.g., dual categorizations such as CSR focused on primary vs. secondary stakeholders, CSR focused on business practice vs. philanthropy).
Second, to our knowledge, this article represents one of the first attempts at leveraging field data to offer direct empirical support for the existence of an economically significant effect of CSR on brand sales (see Table 1). Although prior work has drawn valuable insights from CSR case studies, work that involves actual purchase behavior has been rare (for exceptions, see [ 4]] and [46]]). Findings from experimental studies suggest that CSR may lead to greater purchase intent and increased brand loyalty ([26]), though a few behavioral studies also suggest that under certain conditions CSR can lead to negative attitudinal outcomes (e.g., [31]; [74]). Moreover, an attitude–behavior gap caused by social desirability bias may exist, particularly in contexts that involve social and ethical issues ([ 7]; [70]). Our focus on brand sales offers the benefit of performing a real-world, decision-based examination of how CSR shapes actual consumer response. Our findings are further nuanced by the inclusion of two moderators of the relationship between types of CSR initiatives and brand sales: ( 1) the role of the CSR reputation of the firm and ( 2) the environmental versus social focus of CSR efforts.
Graph
Table 1. Representative CSR Literature: Measurement, Type, and Classification of CSR Initiatives in Extant Research.
| Research | Experimental Data | Survey Data | Secondary Data (Firm Level) | Secondary Data (Brand Level) | Effect | Dependent Variable | Type of CSR Analyzed | Classification of CSR |
|---|
| Ailawadi et al. (2014) | | ✓ | | | Positive (negative) for behavioral loyalty for CSR (not) related to customer's direct exchange with the firm | Attitude, behavioral loyalty | Environmental and social responsibility | None |
| Anselmsson and Johansson (2007) | | ✓ | | | Positive | Attitude/purchase intent | Product, social, and environmental responsibility | None |
| Becker-Olsen, Cudmore, and Hill (2006) | ✓ | | | | Positive (negative) for purchase intent for high- (low-) fit CSR | Attitude/purchase intent | Environmental and social responsibility | None |
| Bhardwaj et al. (2018) | ✓ | | | | Positive | Attitude/purchase intent | Social responsibility | Company-ability-relevant CSR (positively impacting performance) and company-ability-irrelevant CSR (no impact on performance) |
| Buell and Kalkanci (2021) | ✓ | | | | Positive | Bookstore sales from a field experiment | Environmental and social responsibility | Internally or externally focused on the value chain |
| Chernev and Blair (2015) | ✓ | | | | Positive | Product evaluations | Philanthropy | None |
| Du, Bhattacharya, and Sen (2007) | | ✓ | | | Positive | Loyalty, consumer advocacy for the brand | Environmental and social responsibility | None |
| Groza, Pronschinske, and Walker (2011) | ✓ | | | | Positive | Attitude, purchase intent | Environmental responsibility | Reactive versus proactive CSR |
| Homburg, Stierl, and Bornemann (2013) | | ✓ | | | Positive | Customer loyalty | Corporate CSR measure | Business practice CSR engagement and philanthropic CSR engagement |
| Inoue, Funk, and McDonald (2017) | | ✓ | ✓ | | Positive | Attendance at football games | Survey-based perceived CSR | None |
| Kang, Germann, and Grewal (2016) | | | ✓ | | Positive (no effect) when CSR is a good management (penance) mechanism | Firm performance (Tobin's q) | Corporate CSR measure | None |
| Luchs et al. (2010) | ✓ | | | | Positive (negative) when gentleness (strength) product attributes are valued | Consumer preference | Sustainability | None |
| Luo and Bhattacharya (2006) | | | ✓ | | Positive; negative for firms with low innovativeness capability | Stock return, firm performance (Tobin's q) | Corporate CSR measure | None |
| Mishra and Modi (2016) | | | ✓ | | Positive for stock returns; negative for risk (community CSR, n.s.) | Stock returns, Idiosyncratic risk | Corporate CSR measure | None |
| Newman, Gorlin, and Dhar (2014) | ✓ | | | | Positive effect greater for unintentional (vs. intentional) product changes | Purchase intent | Environmental responsibility | None |
| Sen and Bhattacharya (2001) | ✓ | | | | Positive | Company evaluation/purchase intention | Corporate CSR measure | None |
| Servaes and Tamayo (2013) | | | ✓ | | Positive (negative/n.s.) for firms with high (low) customer awareness; effects reversed for firms with poor reputations as corporate citizens | Firm performance (Tobin's q) | Corporate CSR measure | None |
| Wagner et al. (2009) | ✓ | | | | Negative | Attitude | Corporate CSR measure | Reactive versus proactive CSR, abstract versus concrete CSR policy, inoculation strategy or not |
| Yoon, Gurhan-Canli, and Schwarz (2006) | ✓ | | | | Positive (negative) when consumers learn about low-benefit-salience CSR through a neutral (company) source | Company evaluations | Corporate CSR measure | High- versus low-benefit-salience CSR |
| Our study | ✓ | | | ✓ | Positive (negative) when firms take (do not take) accountability for negative externalities | (1) Brand sales, (2) purchase intentions | CSR measure encompassing product, social, and environmental responsibility as well as philanthropy | Accountability based: corrective, compensating, and cultivating CSR |
Third, we use laboratory experiments to provide process evidence regarding the perceived sincerity of brand motives. Perceived sincerity serves as a mechanism that underlies the changes in consumer purchase behavior associated with CSR initiatives ([ 8]). Specifically, we examine the effects of CSR type on consumer purchase intention and the mediating role of perceived brand sincerity. Results from experiments largely support our findings obtained with brand sales data for corrective and compensating CSR and show that perceived brand sincerity mediates the effect of CSR type on purchase intention and CSR reputation moderates the mediation chain. Our framework is depicted in Figure 1.
Graph: Figure 1. Conceptual framework of the effect of CSR initiative on purchase intentions and sales.
To investigate the effect of CSR on brand sales, we collect CSR press releases issued by a comprehensive set of prominent consumer packaged goods (CPG) brands, documented in the CSRwire database as well as on brand websites between the years 2002–2011. These data contain the announcement date as well as the textual content of all CSR announcements made by these brands in this time window. We then collect detailed sales data from the Information Resources Inc. (IRI) consumer panel data set for the brands that announced CSR initiatives as well as a set of close substitute brands that did not engage in CSR. After merging the two databases, our sample includes a total of 55 brands that announced CSR initiatives and 194 brands that did not, spanning 21 CPG product categories.[ 7] Our CSR event list contains 80 actual CSR initiatives (27 corrective actions, 19 compensating actions, and 34 cultivating goodwill actions) that were announced by the corporate parents of the 55 focal brands.
We specify an empirical model estimated on the sales of the focal brands as well as those of peer brands from the relevant product categories, measured one year before and one year after the focal brands implemented CSR events. The results from our empirical analyses indicate that the type of CSR effort undertaken has distinct implications for brands engaging in CSR. While, on average, corrective and compensating CSR actions provide a boost to the sales of participating brands, cultivating CSR actions lead to a slight drop in sales. This negative effect of cultivating goodwill actions is in line with the behavioral literature that has documented, under certain conditions, a reduction in purchase intentions for firms that engage in CSR ([ 8]; [74]). Cultivating goodwill may reduce sales because it detracts resources that could be used to support the brand's primary stakeholders, such as customers and channel partners, and redirects them to external constituencies that may not respond by purchasing the brand's products. At the same time, and consistent with these findings, the results obtained from our experiments suggest that cultivating goodwill CSR actions are viewed as less sincere than the other two types of CSR.
In summary, findings from our analysis of brand sales, in conjunction with results from laboratory experiments, suggest that taking an accountability-based view of CSR may offer useful insights to managers looking to enhance the consumer impact of CSR actions. We next present our conceptual framework and hypotheses, followed by the description of the data, methods, and results for the secondary data empirical analysis. We conclude with a description of the experiments, followed by a discussion of implications from our research.
The literature focused on the impact of CSR is vast and has evolved primarily in two separate streams: one focused on the financial consequences of corporate CSR (e.g., [49]; [59]) and one focused on how CSR impacts antecedents to consumer purchase behavior, including consumer attitudes and purchase intentions (e.g., [26]; [56]; [74]). Most studies show a positive effect of CSR, though some authors identify conditions under which CSR has null or negative effects (Table 1).
There is less consistency in the type of CSR analyzed. CSR, whether measured at the brand or corporate level, encompasses actions that can pertain not only to products, employees, or business partners but also to the community or special groups of stakeholders as well as more general environmental or philanthropical initiatives ([64]). For instance, [42] distinguish between business practice CSR, which targets the firm's primary stakeholders, and philanthropic CSR engagement, which targets the firm's secondary stakeholders. In turn, [47] examine product-focused CSR actions and environmentally focused CSR actions. Other researchers differentiate between proactive CSR, in which firms engage in CSR before consumers receive potentially negative firm information, and reactive CSR, in which firms conduct CSR to protect their image after reports of an irresponsible action ([37]; [80]). Perhaps because of the large variety of CSR initiatives that firms can undertake, very few articles, as illustrated in Table 1, attempt to comparatively assess the effect of different types of CSR.
An underresearched aspect of CSR that is under increasing public scrutiny is the extent to which the costs of a firm's quotidian operations are passed on to the general population. Such costs include, for instance, waste, pollution, or use of labor from developing countries with weak labor laws ([40]). A categorization scheme that focuses on the firm's responsibility for varied externalized costs would help managers choose the appropriate type of CSR action from among a cornucopia of available options. This is all the more critical as managers face increased scrutiny of their firms' externalities from an environmental and social perspective.
The most useful categorization would be one that is actionable, in the sense that it both readily translates to specific actions and wields distinct effects on metrics helpful to managers. The extant literature is lacking on this latter dimension as well. Research on corporate CSR is often conducted using complex CSR indices aggregated at the corporate level (e.g., [58]), while behavioral studies typically leverage metrics with lower external validity, such as laboratory participants' evaluations of fictional CSR initiatives (e.g., [31]). We intend to tackle both shortcomings in the literature by proposing a categorization based on the notion of firm accountability and by examining the impact of this categorization on brand sales. We do so after carefully surveying the literature and noticing that in the few instances where a negative effect of CSR was documented, it was because consumers did not find the CSR actions to be meaningful. For instance, [ 8] show that companies hoping that their sales would improve as a result of their efforts to combat homelessness or domestic violence find instead that purchase intentions for the companies' products decrease. The authors attribute this effect to consumers being skeptical that the firms sincerely want to make a positive change.
Accountability in CSR represents firms' acknowledgment that their business operations create negative externalities, which may include pollution, waste generation, or downstream consumer health issues. Although these externalities may vary, they do constitute a liability that consumers may expect firms to correct by taking specific actions ([ 5]).
First, to directly reduce its negative impact on society or the environment, a firm may adopt changes to its business operations. Examples include product or packaging modification, responsible ingredient sourcing, ethical labor practices, or expansions to the existing product line to cater to consumers at the bottom of the pyramid. Second, a firm may choose to make philanthropic or service contributions aimed at offsetting its negative externalities without changing its business operations (e.g., donations to a cause benefiting stakeholders negatively affected by the brand, cleanup efforts, in-kind donations). Finally, a firm may engage in philanthropic activities unassociated with its negative externalities. Such activities may be intended to engender consumer goodwill (e.g., public relations campaigns, scholarships, endowments). Drawing from these three possibilities, we propose a typology of corrective, compensating, and cultivating goodwill CSR.
Corrective CSR is a form of CSR whereby a brand attempts to minimize its negative impact on society or the environment via actual changes to its products or business operations. For example, a bottled water brand may reduce the amount of plastic used in its bottles, or a retail brand may work on providing more favorable working conditions for its labor force. We argue that explicit changes to a brand's products/operations targeting reduced societal harm represent, in consumers' minds, an acceptance of accountability along with restorative action ([30]). Corrective CSR actions share similarities with Porter's "Shared Value" concept wherein companies find business opportunities in social problems ([71]). Corrective CSR, however, has the distinct goal of minimizing one's harm to society, rather than the deliberate search for a business opportunity within an existing social problem (generally not of the firm's making).
In contrast, compensating CSR involves initiatives whereby a brand addresses its negative externalities "indirectly" (i.e., no actual changes to its products or business practices occur). Compensating CSR initiatives thus represent an implicit acceptance of accountability with attempted redress (e.g., charitable giving, volunteering). For example, a bottled water brand may donate money to plastic recycling programs. While corrective and compensating CSR actions are similar in their implicit acceptance of firm accountability for the negative externality, a key difference is that in the latter case there is no direct restitution offered in the form of actual product or business practice changes. Nonetheless, research in interpersonal relationships demonstrates that an apology without restitution is more effective than no apology at all ([16]).
Finally, when engaging in cultivating goodwill CSR, brands do not address their negative externalities but, instead, offer support for one of an endless variety of unrelated good causes. In this case, brands make no strides toward the acknowledgment of responsibility for any negative externality. Many philanthropic efforts may fall into this category. For example, a bottled water brand may donate money to literacy programs. Although the benefiting cause may be worthwhile, consumers may see these CSR initiatives as a failure by firms to acknowledge any liability arising from their operations. We expect that consumers may view such initiatives as insincere and potentially as a waste of corporate resources ([53]).
The three types of CSR that we study can target both primary and secondary stakeholders and encompass both business practice and philanthropic CSR engagement ([42]). At the same time, there are clear theoretical differences between the three types of CSR; for instance, corrective and compensating initiatives are rooted in separate strategies outlined in the theory of image restoration ([ 9]). Categorizing a CSR initiative into one of the three types involves answering two questions: First, does the CSR initiative address the brand's own social or environmental harm by making changes directly to the company's business operations (i.e., product, supply/distribution network, labor practices)? If so, it is a corrective action. If the initiative does not, then the next question is, does the CSR initiative address a social or environmental harm for which the brand's business operations are perceived as bearing responsibility? If so, then it is a compensating action. However, if it is addressing a social or environmental issue for which the brand bears no clear responsibility and involves no changes to its product nor business operations, then it is a cultivating goodwill action.
Consumers often assess a brand and its actions as they would other members of society ([ 3]). As with interpersonal relationships, consumers often evaluate brands positively if they conform to accepted behavioral norms and negatively if they violate these norms. Irresponsible brand behavior toward society or the environment represents one form of social norm violation that consumers are likely to disapprove of ([ 1]; [47]).
Just as consumers can punish brands for violations of social norms through negative evaluations, attitude, or behavior toward a brand, they can also forgive brands that take accountability for the harm that they may cause. The psychology literature explains the link between accountability and forgiveness. Apology, which incorporates an acknowledgment of violated norms, particularly if coupled with restitution, which involves restorative action and remediation, has been shown to promote forgiveness ([16]). In the same vein, consumers are more likely to favorably evaluate brands that acknowledge their own shortcomings and perform restorative actions. The literature on brand crises also provides support for this assertion. Restorative actions taken during brand crises that involve both an acknowledgment of the problem and plans for remedial actions can be effective at repairing brand attitudes ([30]). Moreover, recent research suggests that consumers understand if the CSR efforts of the firm are focused on its own value chain, and they are more likely to purchase from such firms than from their peers whose CSR is external to the value chain ([15]).
These arguments suggest that corrective CSR actions, which convey the highest level of accountability to the firm stakeholders, will be received most positively by consumers. Likewise, there is also some level of implicit accountability in compensating CSR, though not as direct as that advanced by corrective actions. While compensating CSR may provide a weaker form of restitutive action than corrective CSR, we still expect compensating CSR to increase sales.
In contrast, consumers may view cultivating goodwill CSR as disingenuous or wasteful even if the cause supported by the firm is worthwhile. They may perceive this type of initiative as an attempt to "check a box," and thereby as an insincere approach to CSR that fails to acknowledge the potentially negative consequences that the firm's products or operations may have on society. Prior literature has found some support for this assertion, as CSR initiatives can lead to reduced purchase intent or other negative attributions when consumers believe that the initiatives come at the expense of investments that could improve corporate abilities (e.g., [74]). In addition to consumers, retailers are also important stakeholders who may view CSR initiatives as redirecting resources that could have been used to more directly support an increase in sales, such as price or display promotions. Moreover, both consumers and retailers may believe that corporate philanthropy is driven more by managers' desire to enhance their personal reputations than by stakeholders' interests ([79]). In summary, stakeholders may feel, at the margin, disenfranchised by initiatives meant to cultivate the goodwill of unrelated groups, resulting in a negative impact on sales. Formally, we expect the following:
- H1: Corrective and compensating (cultivating goodwill) CSR initiatives have an overall positive (negative) effect on consumer purchase intentions and, consequently, on brand sales.
Extant CSR literature can guide us in establishing boundary conditions for the effects hypothesized in H1. We focus on two factors that emerge from the literature as likely to have a role in determining the effectiveness of CSR initiatives. First, firms' CSR reputation is particularly important in shaping consumers' reactions to a firm's CSR activities. In addition to establishing expertise and increasing the credibility of CSR initiatives, firms' CSR reputation can influence product evaluations and, in instances of product harm, can temper consumers' negative evaluations of the brand ([50]). Second, the focus of CSR (environmental or social) is one of the fundamental characteristics of CSR highlighted in the literature ([22]). While initiatives in both domains have been found to have a positive impact (e.g., [ 6]; [26]), their relative contribution to the success of a CSR initiative has not been clearly established. In addition, by focusing on the interaction between CSR focus and type, we provide managers with a 2 × 3 matrix of possible CSR initiatives that can help them adopt a CSR outreach that is most appropriate for their firm.
The expectation disconfirmation paradigm suggests that consumers' responses to CSR initiatives will be contingent on their assessment of a firm's CSR reputation, defined as stakeholders' assessment of the past performance and success of the firm's CSR activities ([63]). Favorable firm reputations can influence the actions of firms' stakeholders, including consumers ([33]). Moreover, brand activities that are congruent with prior CSR reputation are less likely to change consumer brand perceptions and may have little effect on consumer response ([68]). Consequently, we expect the effects of all three types of CSR initiatives on sales, whether positive or negative, to be smaller in magnitude for firms with high CSR reputations, as these initiatives confirm what consumers already believe.
Conversely, CSR actions from firms with lower CSR reputations may come as a surprise to consumers. When firm actions are inconsistent with existing knowledge, consumers engage in deeper processing of the new information, which may make them question the sincerity of brands' motives for engaging in these efforts ([85]). Consumers may be particularly suspicious of the sincerity of lower-reputation brands that engage in cultivating CSR actions. Such actions may be perceived as perfunctory (e.g., [ 4]), which can magnify their negative effect on sales. In contrast, corrective and compensating CSR actions, which invoke a certain level of accountability, may provide a positive disconfirmation of consumers' initial perceptions of low-CSR-reputation firms and further enhance their brand sales. Indeed, research on brand crises and service failures suggests that demonstrating accountability after negative incidents is particularly effective at improving a consumer's brand perceptions ([30]). We thus expect that, all else being equal, CSR initiatives announced by brands with more favorable CSR reputations are relatively less likely to impact brand sales than initiatives from brands with less favorable CSR reputations.
- H2: CSR reputation mitigates the effect of CSR initiatives on sales as follows:
- Higher CSR reputation reduces the positive effect of corrective and compensating CSR initiatives on brand sales.
- Higher CSR reputation reduces the negative effect of cultivating CSR initiatives on brand sales.
Our proposed typology describes the actions taken by a brand, which may allow it to take accountability for its negative externalities. However, a brand may go about corrective, compensating, and cultivating CSR in a myriad of ways. One of the most frequently discussed dimensions of CSR is the domain in which CSR is implemented: environmental or social (e.g., [ 4]; [ 8]; [26]). Prior literature has not offered a clear comparison between the efficacies of CSR in these two domains, but it would be helpful for managers to know whether one of the two demonstrates accountability for the brand's negative externalities more effectively, particularly in conjunction with the three types of CSR.
We argue that environmentally focused CSR initiatives will have an enhanced effect on brand sales for two reasons. First, consumers tend to place greater relative importance on environmental concerns than on social issues ([67]). As the media regularly highlights the liability of firm operations for harm done to the environment, consumers have become increasingly aware of these issues. For instance, the 2017 Carbon Majors Report found that a mere 100 companies generate 70% of global greenhouse gas emissions, and this received significant media coverage ([24]). As a result, firms are finding that it is increasingly necessary to address such liabilities ([23]). Academics agree; for instance, [52], writing on the occasion of the 75th anniversary of the Journal of Marketing, centers his article on the growing importance of the environmental imperative to marketing theory and practice. Thus, environmental CSR is important to stakeholders, relatively objective, and typically noncontroversial.
Second, while social CSR initiatives have the potential to create a favorable image among subsets of stakeholders, these initiatives are perceived as being less focused, less verifiable, and thus more prone to agency costs ([64]). Moreover, recent research shows that consumers and shareholders do not always agree with the direction of social CSR, with some viewing such actions as an alienating form of activism ([17]; [45]; [11]). Therefore:
- H3: Having an environmental rather than a social focus in a CSR initiative enhances the positive effect of corrective and compensating CSR initiatives and mitigates the negative effect of cultivating CSR initiatives.
In this subsection, we propose a mechanism for the predicted effects and explore the mediating role of perceived sincerity, a process we test subsequently in controlled laboratory experiments. Perceived sincerity is the extent to which consumers perceive a brand as caring and genuine in its actions ([41]). Greater perceived sincerity in CSR can lead to higher brand evaluations, purchase intent, and brand loyalty ([ 5]; [85]). In the domain of service failure, demonstrating accountability and taking reparative action are viewed as sincere gestures needed to improve customer satisfaction and repurchase intention ([77]). In our typology, corrective and compensating CSR actions signal a brand's willingness to take responsibility for its impact on society and the environment. As discussed previously, this may entail making changes to products, the supply chain, or manufacturing operations, or contributing time, money, or other resources. Such efforts directly acknowledge fault, may be costly and difficult to implement, and are thus unlikely to be taken lightly by consumers.
In contrast, research suggests that when a brand does not sufficiently redress the harm caused by its actions, consumers are likely to perceive its CSR actions as insincere ([ 5]). In the absence of an acknowledgment of accountability, consumers may discount the good deeds associated with cultivating CSR activities or even be cynical of them. The CSR activities may backfire, leading to negative evaluations of the company and reduced purchase intentions or behavior. We thus predict that corrective and (to a lesser degree) compensating actions will be perceived as relatively more sincere than cultivating actions, and that this greater perceived sincerity will result in a more favorable consumer response, mediating the effect of CSR on purchase intentions.
We also expect that brand CSR reputation will moderate the aforementioned mediation chain. In line with our previous arguments, brands with higher CSR reputations are likely to be viewed as simply fulfilling expectations by engaging in CSR and acting relatively sincerely regardless of CSR type ([34]). Thus, attitudes toward high-CSR-reputation brands are ultimately less likely to be affected when these firms engage in any new CSR activities. In contrast, when brands with weaker reputations engage in CSR, consumers are likely to think more deeply regarding their motives, leading to greater relative differences in perceived sincerity across CSR types ([85]). Thus, for lower-reputation brands in particular, we expect corrective and compensating CSR to be seen as more sincere than cultivating CSR, a difference that should influence purchase intentions accordingly. More formally:
- H4a: Perceptions of brand sincerity mediate the effect of CSR initiatives on purchase intentions.
- H4b: CSR reputation mitigates the mediation mechanism that indirectly links CSR initiatives to purchase intentions through perceptions of brand sincerity.
We use two approaches to test our hypotheses. First, we use a regression model estimated with panel data with fixed brand effects to examine the impact of CSR initiatives on brand sales. Second, we use experiments to document the process that underlies these effects and to show that brand sincerity mediates the effects of CSR actions on intentions of purchase. We begin by presenting the data, our method, and the results of the brand sales model, and we follow with a summary of the experiments that demonstrate the mediating role of brand sincerity.
We leverage two main sources of data to examine the impact of CSR on brand sales: ( 1) 3BL CSRwire service (CSRwire.com), to extract the CSR announcements, and ( 2) the IRI data set, to obtain brand sales before and after these announcements. CSRwire contains a searchable CSR news archive of more than 20,000 news items including corporate- and brand-level CSR-related press releases, CSR reports, and other event announcements dating back to 1999. Through CSRwire, companies disseminate CSR information to a diverse global audience via a myriad of websites and portals including Google, Reuters, LexisNexis, and Bloomberg ([23] Corporation 2009; [36]). Data from CSRwire have been previously used to study the impact of CSR (e.g., [27]; [35]).
We obtain data on brand sales from the IRI academic data set ([13]] provide a detailed description of the data set). The IRI data set comprises weekly aggregate store-level product sales as well as consumer panel data for 30 CPG categories. The data set provides a rich time series of sales information at the Universal Product Code level for various brands and across markets (designated market areas [DMAs]). A vast body of research has employed the IRI data set to study the impact of marketing actions on brand sales (e.g., [ 5]; [14]). We begin by tracking CSR initiative announcements from brands in the IRI data set in the time period of our data, from 2001 to 2011. We first record the date of the CSR press releases drawn from CSRwire and the CSR/sustainability initiative press announcements from brand websites in this time period. Our CSR announcements were made between January 2002 and December 2011 and are listed in Web Appendix A. Archival searches for these announcements revealed that about 95% of them were prominently featured and discussed in major local and national newspaper outlets on the same date as the one reported on CSRwire. This suggests that there was a reasonable level of awareness for the events in our sample; at the same time, having events with lower coverage would work against the effects we hypothesize, making our tests more conservative.
Our analysis sample includes 80 CSR initiatives across 55 brands, 21 product categories, and 48 DMAs. The classification of CSR announcements as corrective, compensating, or cultivating CSR was done by a panel of independent judges (N = 378), who each categorized a small random subset of these announcements by applying our definitions to the text of the press releases, with a high degree of interrater reliability (intraclass correlation coefficient = .80).
For each brand, we use weekly brand sales aggregated (across stores) at the DMA level as our outcome of interest. We extract sales information from the IRI data set for the brands that have announced a CSR initiative for the 12-month period before and the 12-month period after the CSR announcement. We also wanted to obtain data on a set of appropriate control brands. Among the brands in the IRI database that belong to the same category as the focal brands, we kept all brands that, in descending order of market share, made up for 70% of the focal brand's market share. From this control group of brands, we exclude the ones that announced a CSR initiative in the year before and the year after the focal brand announced a CSR action.[ 8] Thus, our control group size ranges from 3 brands (in the facial tissue category) to 18 brands (in the cereal category), with an average size of 5.15 control brands (across all categories).
Next, for each of the 80 CSR announcements, we choose an observation window of 104 weeks of weekly sales activities (52 pre- and 52 postannouncement weeks) for both focal and control brands in the product category. While 52 weeks is sufficient time for the sales effect of brands' CSR announcements to have manifested, the focus on a relatively tight window helps mitigate the influence of unobserved time-varying drivers of sales changes for both focal and control brands. We find that, on average, 65% of brands that form our control group also ended up announcing CSR initiatives at a later date (i.e., at least 12 months after our postannouncement observation window ends). This pattern is perhaps intuitive and, to some extent, also showcases the increasing extent to which the relatively prominent CPG brands that are part of our data set opt into engaging in CSR.
Our framework includes two moderators: CSR reputation and CSR focus. The CSR focus on social versus environmental issues can be easily categorized from the text of each announcement. The focus of CSR is distinct from CSR type, and our sample includes observations for each combination of focus and type. Specifically, for environmentally focused CSR, we observe 14 corrective, 13 compensating, and 8 cultivating initiatives, and for socially focused CSR we observe 13 corrective, 6 compensating, and 26 cultivating actions.
To measure CSR reputation, we return to CSRwire and construct an index based on the recorded total number of instances over the one-year pre-CSR announcement window during which each of the brands in our analysis sample either ( 1) relayed sustainability-related information—but not new CSR efforts/engagements—on CSRwire (e.g., "Seventh Generation Releases Annual Corporate Consciousness Report") or ( 2) was featured in a sustainability-related report showcased by its corporate parent on CSRwire (e.g., "Kimberly-Clark Receives Perfect Score on 2011 Corporate Equality Index," "Miller Coors Launches Corporate Responsibility Web Site"). We find that this index offers sufficient variability across the focal brands in our sample, having a mean of.98 announcements and a standard deviation of 1.87.
Our research design exploits two useful sources of variation: ( 1) while some brands within a product category announce CSR initiatives, others do not, and ( 2) CSR announcements in our data are spread over a wide time horizon (vs. being clustered over a narrow time window). The variation in ( 1) helps us account for possible differences between brands that announce CSR initiatives and ones that do not. Alternatively, the sizable spread offered by ( 2) helps us partially mitigate the influence of broader macroeconomic trends (such as the recession of 2008) that may have otherwise played a role in influencing brand sales pre- and postintervention over a few specific years ([76]; [78]).
We use a host of other data sources to construct control variables and instruments to account for endogeneity. For each of the brands in our data set, we collect information on ( 1) product prices (from IRI); ( 2) whether the product is on display—categorized into "none," "minor," or "major" displays (from IRI); ( 3) distribution intensity (number of stores carrying the brand, from IRI); ( 4) monthly advertising spending (from Kantar Media's Ad$pender database); and ( 5) press coverage (from RavenPack). To construct the press coverage control variable, we identify the corporate parent of each brand and download all the press releases of this corporate parent available in RavenPack for the same period for which we collect brand sales data. As RavenPack provides a sentiment score for each press release, we separate them into positively and negatively valenced announcements. Further, we include both the count of positively and negatively valenced announcements to proxy not only for the extent of press coverage during our sample period but also for the sentiment that underlies that coverage (for a description of RavenPack and the sentiment scores associated with the press releases, see [81]]). Finally, we use multiple data sources to construct instruments that account for ( 1) the endogeneity of the type of the CSR decision and ( 2) the endogeneity of the marketing instruments used as controls in our main brand sales model. We describe the instruments and their operationalization next.
We first describe how we address the potential endogeneity associated with the choice of the type of CSR, followed by a description of our controls for the endogeneity of the marketing-mix instruments used in the sales model. In Web Appendix B, we present the results of an additional robustness step that assesses the potential importance of unobserved confounders in explaining our effects, by following the approach proposed by [69].
To accurately assess the impact of CSR initiatives on brand sales, we need to control for the endogeneity of the type of CSR initiative undertaken by brands. Specifically, brands choose which type of CSR initiative to implement, and this choice could be driven by unobservable characteristics, leading to biased estimates for the effects of CSR on sales. To account for this choice, we estimate a multinomial logit model where the dependent variable has four levels—one for each type of CSR initiative and one for the choice to not do any CSR. The dependent variable takes a value of 0 for both the brands that did not announce CSR initiatives at all, as well as for focal CSR-announcing brands but only during the weeks preceding the CSR announcement. In the post-CSR announcement window for the focal brands, the dependent variable is coded as a categorical variable designating the type of CSR action undertaken (corrective, compensating, or cultivating). We use this model to obtain a set of three generalized inverse Mills ratios that will be included in the brand sales model as controls for this particular type of endogeneity. This follows the approach outlined in Wooldridge (1995) and Bourguignon, Fournier, and Gurgand (2007), which has been used in marketing applications by [25]; [32], and [43].
To estimate this model in a manner that does not exclusively rely on the functional form of the chosen selection equation, we need exclusion restrictions, in the form of one or more variables that significantly impact the choice of conducting CSR but do not directly impact sales. We identified three such variables: the Product Responsibility Score (PR_Score) and the Innovation Score (Innovation Score) from Refinitiv's EIKON database as well as a variable that denotes the proportion of new products introduced by the brand that contain CSR claims, but in product categories other than the focal one (Prop_CSR claims) from Product Analytics. We describe in Table 2 the construction of these variables and explain their validity.
Graph
Table 2. Description of Instruments Used to Control for the Endogeneity of the Type of CSR Action.
| Instrument | Definition | Instrument Relevance | Exclusion Restriction | Data |
|---|
| Product Responsibility Score (PR_Score) | The Product Responsibility Score is a weighted combination of scores that captures the extent to which a company has structures and processes in place dedicated to producing quality goods and services, ensuring the customer's health and safety, and protecting customers' data privacy. This variable is different from CSR reputation: the policies of a firm with high PR score are not necessarily visible to the public; alternatively, a firm with high CSR reputation could owe this reputation to philanthropic efforts or to CSR initiatives not directly related to policies that impact consumers This is a firm-level variable, and data vary annually. | PR_Score captures firms' CSR emphasis on ensuring that as little harm as possible is done to consumers, it should be positively correlated with the firms' propensity to conduct corrective actions. | PR_Score is calculated at the corporate level and refers to the existence of corporate-level policies and processes that are meant to reinforce a positioning focused on responsibility. The products associated with this type of positioning may be niche or may elicit a price premium, but their sales are not necessarily higher than those of more conventional alternatives, as many consumers continue to prefer the latter (e.g., Wilcox et al. 2009). | Refinitiv's EIKON. The EIKON database includes ESG scores for over 9,000 global firms, which are computed from a variety of public sources including annual reports, company websites, news sources, nongovernmental organization websites, and others. |
| Innovation Score (Innovation Score) | The Innovation Score within the Environmental pillar in EIKON reflects the brand's inclination to use new environmental technologies and processes or to manufacture ecodesigned products. This is a firm-level variable, and data vary annually. | Firms that score high on this pillar are focused on innovation and, therefore, are less likely to be focused on their old, existing products. Moreover, because their new products are already likely to be more sustainable and incorporate more responsible practices, these firms are more likely to engage in cultivating CSR than in the other two types. | Innovation Score captures mostly corporate processes that reinforce a manufacturing and positioning strategy focused on new environmental technologies. As previously argued, such products do not necessarily surpass conventional alternatives in sales, but are rather more likely to target unique consumer segments. |
| Proportion of new products introduced by the brand that contain CSR claims, in product categories other than the focal one (Prop_CSR claims) | For each brand in our sample, we obtain all new products introduced in the two years before each CSR initiative was introduced, from which we exclude the products introduced in the categories we study in our article—we call this resulting measure "NP." We then classify the package claims for these products into CSR (NPCSR claims) and non-CSR-related (NPother claims). We use this classification to compute NPCSR claims/NP, which we use as an instrument for the propensity to engage in CSR initiatives. This is a brand-level variable computed using two years of data preceding each CSR initiative. | This variable reflects a brand's commitment to incorporate CSR practices in its products and should be positively associated with the general propensity of the firm to engage in CSR but negatively associated with corrective CSR actions because firms may have less remaining to correct for, or less that can easily be corrected. | Because the variable was constructed using new products from all categories in which brands operate, except for the focal one, it should not directly impact brand sales in the focal category, ensuring that the exclusion \restriction is verified. | GlobalData Product Launch Analytics database, a database that provides extensive information on CPG products (e.g., Moorman et al. 2012). |
The utility of choosing a CSR initiative of type j by brand i at time t is given by Uijt = Vijt + εijt, where Vijt is a deterministic component and εijt is a random error. Using the multinomial logit model and assuming that the random error is independently and identically Gumbel-distributed, the probability that the CSR initiative of type j is chosen by brand i at time t is given by
Graph
( 1)
where Vijt = α0j + α1jPR_Scoreit + α2jInnovation_Scoreit + α3jProp_CSRclaimsit + α4jAdvertisingit + α5jBrand_assetit, j = 1, 2, 3 refers to the three types of CSR (corrective, compensate, and cultivating goodwill), i refers to the brand, and t to the month of measurement, which spans 12 months before a CSR initiative was announced and 12 months after. Brand_asset is the Brand Asset Valuator Y&R overall measure of brand equity (measured at the annual level), and Advertising denotes brand-level advertising expenditures (measured at the monthly level). Thus, for each brand that has undertaken a CSR initiative and for each peer brand in its product category we have 24 months of advertising data and at least two years of brand asset and CSR score data (contingent on each brand reporting sales in each particular DMA), resulting in an unbalanced panel over which the model is estimated. Using the choice probabilities predicted from Equation 1, we compute a set of three generalized inverse Mills ratios (one for each CSR type j) of the form[ 9] to include in the outcome equation governing the sales response of CSR (which we discuss subsequently).
In addition to accounting for the endogeneity of the decision to implement a CSR initiative, we also account for the endogeneity of the marketing-mix variables included as controls in the model of brand sales. To do so, we use a two-stage least squares approach. Specifically, we specify an additional equation for each marketing-mix variable and model these variables as a function of all fixed effects and exogenous variables from the sales equation and an instrument for the brand's marketing-mix variable. We follow [78] in using as instruments weighted averages of the marketing mix of brands that do not have products in the same narrow product category but belong to the same industry. The marketing-mix variable of the focal brand is likely to be correlated with that of these brands, because the same underlying cost structures apply and may lead to similar movements in these variables. We use this approach for advertising, display, and distribution intensity. We use a different set of instruments to account for the endogeneity of product prices. For prices, we use measures that commonly govern the factor costs of production/packaging in the CPG industry, such as the producer price indices for plastic (North American Industry Classification System code 326160) and wood pulp (North American Industry Classification System code 322110), gathered from the Bureau of Labor Statistics website.
We specify the following model to estimate the effect of CSR on brand sales:
Graph
( 2)
The model is estimated at the brand (i ), DMA (d ), and week (t ) levels. For each brand that introduced a CSR initiative and all the peer brands from the same product category, the sample includes data for 52 weeks before the date of that brand's CSR initiative announcement and 52 weeks postannouncement. The term CSR_PostAnnounceijt takes a value of 1 in the 52 postannouncement weeks t if/after brand i made a CSR announcement of type j and 0 otherwise. CSR_Focusijt takes the value 1 if the CSR initiative has an environmental focus and 0 if the focus is social. For each of the N brands and D DMAs in our data, we include separate fixed effects (θ1id) to account for heterogeneous brand preferences at the local market level. Including brand × DMA fixed effects obviates the need to separately account for whether brand i implemented CSR (i.e., via a dummy variable for having implemented a CSR initiative) and whether the CSR was environmentally or socially focused. We also include week fixed effects θ2t to control for seasonality, which obviates the need to separately account for a common main effect for the postannouncement period in the data. The term lnPriceit is the logarithm of price for brand i, lnAdvertisingit reflects the logarithm of advertising spending for brand i, lnDistribit captures the logarithm of the number of stores carrying brand i, and lnDisplayit reflects the logarithm of brand i's in-store display intensity, all in week t.[10] Following the two-stage least squares approach, we replace all marketing-mix controls with their predicted values from the respective first-stage equation used to account for their plausibly endogenous nature.
The terms lnPositive Pressit and lnNegative Pressit are, respectively, the logarithm of the number of positively and negatively valenced mentions of brand i in the news. The lagged value of log sales of brand i in week t (lnSalesidt − 1) is also included to account for the carryover effect of marketing events on brand i. The IMR measures are the inverse Mills ratios incorporated to account for the endogeneity of brands' CSR choices. We compute bootstrapped cluster-robust standard errors using 50 replications to account for any within-unit serial correlation and sampling error inherent in the predicted probabilities generated from Equation 1 used in the computation of the IMR measures and of the endogenous marketing-mix controls.
The terms β1j, β2j, and β3j denote our three coefficients of interest. The coefficients β1j capture the main effects of CSR of type j on brand sales, while β2j and β3j capture the moderating role of CSR reputation and CSR focus, respectively. They correspond to the effect that different types of CSR efforts announced by brands have on their sales, after controlling for the influence of heterogeneity in consumers' brand preferences, changes in brands' marketing-mix strategies, and seasonality. Next, we discuss the results from these models.
We start by presenting the results of the auxiliary equations used to account for endogeneity and selection, and we then present the main model results. Descriptive statistics for the variables used in Equations 1 and 2 are presented in Tables 3 and 4, respectively. We do not observe any concerning correlations that could suggest multicollinearity.
Graph
Table 3. Correlation Matrix and Summary Statistics of Determinants of CSR Type.
| PR Score | Ad Expenditure | Prop CSR Claims | Innovation Score | Brand Asset | Mean | SD |
|---|
| PR Score | 1 | | | | | 3.35 | .96 |
| Ad expenditure ($M) | .14 | 1 | | | | 1.67 | 3.52 |
| Prop CSR claims | −.02 | −.03 | 1 | | | 1.04 | .77 |
| Innovation Score | .33 | −.05 | −.08 | 1 | | 2.49 | 1.15 |
| Brand asset | .03 | .11 | .16 | .29 | 1 | 73.50 | 22.01 |
Graph
Table 4. Correlation Matrix and Summary Statistics of Determinants of Brand Sales.
| CSR Reputation | CSR Focus | Log of Price | Log of Ad Expenditure | Log of Display Intensity | Log of Distribution Intensity | Log of Positive Press | Log of Negative Press | Log of Salest−1 | Mean | SD |
|---|
| CSR reputation | 1 | | | | | | | | | .30 | 1.11 |
| CSR focus | .17 | 1 | | | | | | | | .08 | .27 |
| Log of price | .08 | .06 | 1 | | | | | | | 1.56 | .48 |
| Log of ad expenditure | .10 | .10 | .05 | 1 | | | | | | 8.20 | 6.69 |
| Log of display intensity | .04 | .06 | −.02 | .16 | 1 | | | | | −5.73 | 3.67 |
| Log of distribution intensity | −.01 | −.02 | −.09 | .24 | .15 | 1 | | | | 7.02 | .45 |
| Log of positive press | −.15 | −.12 | .022 | .16 | .01 | .14 | 1 | | | 9.10 | .71 |
| Log of negative press | −.02 | −.11 | −.12 | .18 | .09 | .25 | −.16 | 1 | | 5.88 | 3.55 |
| Log of salest − 1 | .04 | .02 | −.31 | .23 | .38 | .36 | .08 | .07 | 1 | 6.51 | 1.76 |
The results from the multinomial logit model are presented in Table 5 and suggest that the instruments for the CSR initiatives have strong explanatory power for the propensity of firms to conduct CSR, confirming their validity. The first instrument, PR_Score, is positively associated with the propensity to conduct corrective CSR, in line with this type of CSR being focused on the firm's existing products (β = .031, p < .01). Interestingly, a high PR_Score is also positively associated with cultivating CSR (β = .371, p < .01) but negatively associated with firms' propensity to engage in compensating CSR efforts (β = −.149, p < .01). We expect that the latter result is due to the singular focus that firms with high PR_Score place on improving their previously introduced products, rather than compensating for other negative externalities. The direction of the Innovation Score instrument, which refers to the extent to which the firm's processes and new products incorporate sustainable technologies, is as expected. We find a negative association for the two types of CSR focused on addressing the negative externalities associated with the firm's existing products and operations, which are already likely to be designed using high-CSR standards (corrective: β = −.073, p < .01; compensating: β = −.188, p < .01). In contrast, the association with cultivating CSR is positive (β = .618, p < .01). The third instrument, PropCSR claims, is also negatively related to corrective CSR (β = −.609, p < .01), consistent with the argument that firms whose new products already include CSR claims are less likely to need to engage in corrective efforts. At the same time, Prop_CSR claims is positively related to compensating CSR (β = .112, p < .01) and marginally positively related to cultivating CSR (β = .012, p < .10). Finally, we find that brand equity is positively related to the propensity to conduct corrective and compensating CSR, while advertising expenditures are positively related to all three types of CSR, suggesting that strong brands view CSR as an additional avenue to maintain their brand equity.
Graph
Table 5. Determinants of the Type of CSR Initiative.
| DV = CSR Type Choice | Corrective | Compensating | Cultivating |
|---|
| Est. | SE | Est. | SE | Est. | SE |
|---|
| PR Score | .031** | .006 | −.149** | .005 | .371** | .013 |
| Ad expenditure | .180** | .012 | .776** | .009 | .113** | .009 |
| Prop CSR claims | −.609** | .007 | .112** | .006 | .012† | .007 |
| Innovation Score | −.073** | .005 | −.188** | .006 | .618** | .007 |
| Brand asset | .008** | .000 | .004** | .000 | .001 | .000 |
| N | 1,196,057 |
| Log-pseudolikelihood | −456,251.55 |
1 *p < .05. **p < .01. †p < .1.
2 Notes: Heteroskedasticity-robust SEs are reported alongside estimates.
To determine the effect of CSR initiatives on sales, we use a regression model estimated with panel data with fixed brand effects. We first conduct augmented Dickey–Fuller tests to assess whether the brand sales series is stationary or possesses a unit root. We also conducted Perron tests, which extend the Dickey–Fuller methodology to structural breaks in the model. Both augmented Dickey–Fuller (p <.01) and Perron (p <.01) tests of the null hypothesis of all panels with a unit root are significant, suggesting that the series are stationary. In addition, our analyses are conducted using 104 weekly sales observations, alleviating concerns of "dynamic panel bias" salient in studies utilizing few time periods ([48], p.13; [73], p. 103).
We present our model results in Table 6, where we include the results from various specifications, including with and without controlling for the endogeneity of the marketing-mix variables. In all the specifications presented, we account for the endogeneity of firms' CSR choices as described previously. To begin, column 1 shows the results from a model that includes main effects but without any marketing-mix controls. Column 2 shows estimates from a model that includes interactions with CSR reputation and CSR focus, but again without marketing-mix controls. In column 3, we control for firms' marketing-mix strategies without (yet) correcting for their plausibly endogenous nature. Column 4 presents results from models that employ instruments to address the endogeneity in the marketing-mix controls and where the continuous moderator CSR reputation is standardized.
Graph
Table 6. Effect of CSR Initiatives on Brand Sales.
| Main Effects, Excluding Marketing-Mix Controls | Full Model, Excluding Marketing-Mix Controls | Marketing-Mix Controls Included (Without Their Endogeneity Correction) | Marketing-Mix Controls Included (with Endogeneity Correction), CSR Rep Standardized |
|---|
| DV = Log (Brand Sales) | (1)Est. | SE | (2) Est. | SE | (3)Est. | SE | (4)Est. | SE |
| Corrective | .020** | .002 | .012** | .005 | .016** | .005 | .010* | .005 |
| Compensating | .011** | .003 | .029* | .011 | .020† | .013 | .030* | .014 |
| Cultivating | −.012** | .003 | −.013* | .006 | −.036** | .009 | −.035** | .008 |
| Interactions with CSR Focus |
| Incremental effect for environmental—Corrective | | | .039*** | .011 | .029* | .013 | .045** | .009 |
| Incremental effect for environmental—Compensating | | | −.013 | .013 | .001 | .014 | −.006 | .007 |
| Incremental effect for environmental—Cultivating goodwill | | | .024* | .011 | .057** | .015 | .051** | .015 |
| Interactions with CSR Reputation |
| Corrective × CSR Rep | | | −.007* | .003 | −.011** | .004 | −.012** | .001 |
| Compensating × CSR Rep | | | −.007* | .002 | −.011** | .002 | −.011** | .002 |
| Cultivating × CSR Rep | | | −.005 | .003 | −.008* | .003 | −.004 | .003 |
| Control Variables |
| Log of price | | | | | .007 | .555 | −.142 | .742 |
| Log of ad expenditure | | | | | .004** | .000 | −.001 | .005 |
| Log of display | | | | | .987** | .113 | .641** | .093 |
| Log of distribution intensity | | | | | .359** | .084 | .015 | .270 |
| Log of positive press | | | | | 75.96** | 5.439 | 66.88** | 9.811 |
| Log of negative press | | | | | −68.26** | 4.952 | −59.47** | 9.615 |
| Lag of log (brand sales) | .375** | .001 | .375** | .005 | .343** | .029 | .334** | .037 |
| Endogeneity Correction |
| IMR—Compensating | −.035** | .005 | −.036** | .006 | 2.346** | .173 | 2.099** | .266 |
| IMR—Corrective | .101** | .004 | .101** | .007 | −.865** | .065 | −.746** | .129 |
| IMR—Cultivating | .078** | .003 | .078** | .004 | 1.437** | .109 | −1.276** | .176 |
| N | 1,157,530 | 1,157,530 | 1,141,022 | 1,141,022 |
- 3 *p < .05. **p < .01. †p < .1.
- 4 Notes: DV = dependent variable. Brand × DMA fixed effects and week fixed effects are included in all specifications. Bootstrapped heteroskedasticity-cluster-robust SEs are reported. Main effects of CSR focus of an announcement and the brand's CSR reputation are not identified separately from the brand × DMA fixed effects. Employs one-week lags for the DV. Results are similar with four- or eight-week lags.
Our results reveal that the effect of CSR on brand sales varies, and in material ways, with the type of CSR action undertaken by brands. Consistent with H1, the direction of the change in brand sales is positive for corrective and compensating CSR announcements and negative for cultivating CSR announcements. Because our dependent variable is specified in logarithms, we can compute the percentage change in sales for the focal brand on account of CSR as (exp(βit) − 1). The change in sales for brands engaging in corrective (compensating) CSR appears to be in the order of 1.0% (3.05%), whereas for cultivating CSR it is around −3.45%. The resulting equilibrium sales levels over the long term are about = 1/(1 −.33) = 1.5 times the size of the short-term sales changes (i.e., approximately 1.5%, 4.6%, and −5.2%, on average, for corrective, compensating, and cultivating CSR, respectively).
The estimates of the coefficients corresponding to these controls are all in line with expectation—for example, product price and advertising are respectively negatively and positively related to sales. The auxiliary equations that link the marketing-mix variables to the instruments that help us account for the endogeneity of these marketing-mix variables yield results that are all in the expected direction and are significant at p < .01. We also formally verified the strength of the instruments employed to rule out weak identification concerns. The Sanderson–Windmeijer multivariate F-statistics linked to each of our endogenous marketing-mix variables range between 23.88 for price and 61.48 for distribution intensity (the p-value associated with each case was <.001). Finally, the coefficients corresponding to the inverse Mills ratios for corrective, compensating, and cultivating CSR actions are significantly different from zero, highlighting the importance of accounting for such an endogenous influence on our estimates.
In terms of the moderators, we find that higher CSR reputation brands experience a lower increase in brand sales compared with lower-scoring brands for corrective (β = −.012, p <.01) and compensating (β = −.011, p < .01) CSR actions. This is in line with our arguments about consumers expecting less from brands with low CSR reputations and therefore being more pleasantly surprised when they undertake CSR initiatives. In contrast, for cultivating CSR action, we do not find a significant moderating influence of CSR reputation. Thus, H2a is supported, whereas H2b is not.
The focus of CSR on environmental (vs. social) issues also impacts the effect of CSR on brand sales. We find that the incremental effect of environmental CSR focus (β = .05, p <.01) appears to help reduce the negative effects of cultivating CSR on sales. Similarly, environmentally focused corrective CSR actions appear to contribute a positive boost to sales (β = .045, p <.01), while socially focused corrective CSR actions also have a significant but slightly more modest positive effect on brand sales (β = .014, p <.01). In contrast, brands that announce environmentally focused compensating CSR actions experience benefits that are statistically indistinguishable from those announcing socially focused compensating CSR actions. We provide a more detailed discussion of these effects next and an illustration in the Web Appendix D.
While similar, two important differences between corrective and compensating CSR arise in the results. First, the effect of compensating CSR does not differ significantly between environmental or social initiatives. Second, there is a more pronounced moderating effect of a brand's CSR reputation on corrective CSR. Finally, while, on average, cultivating CSR initiatives result in a slight decrease in sales, this effect was not significantly moderated by CSR reputation. Rather, the effect on cultivating CSR was more pronounced among social initiatives.
In summary, the results are consistent with the effects hypothesized in H3 for corrective and cultivating CSR, but an environmentally focused compensating CSR effort does not appear to further enhance the effect of this type of CSR on sales.
Next, we briefly examine our predictions under experimental settings, with the primary purpose of understanding the process underlying the effects observed in the model. Specifically, the experiments serve to replicate the pattern of results observed in the model, moderated by CSR reputation (H1 and H2), while also demonstrating the mediating effect of perceived brand sincerity (H4a and H4b). To reiterate, H4 predicts that the different effects of CSR actions on brand sales postulated in the model are attributable in part to differences in consumers' inferences regarding the sincerity of the brand's actions. Consumer inferences regarding a brand's motives for CSR actions are known to influence consumer responses to those actions ([26]), as consumers are reluctant to reward CSR when they distrust the company's motivations ([18]). We expect that perceptions of sincerity will be relatively high across CSR types for high-CSR-reputation brands and generally lower for brands with lower CSR reputations ([85]). However, when a relatively lower-reputation brand demonstrates greater accountability through corrective actions, we predict that consumers will be relatively less skeptical of ulterior motives and, thus, more likely to attribute these actions to the character of the brand ([85]). In contrast, actions that recognize a problem without solving it (compensating), or do not relate to or address the brand's negative externalities (cultivating), are likely to be met with greater skepticism due to the increased salience of possible ulterior, selfish motives ([31]). Thus, although we predict that the main effects of CSR type on sales will be mediated by perceived sincerity (H4a), we also predict a moderated mediation effect (H4b). Specifically, we expect that CSR reputation will moderate the effect of CSR action on perceptions of brand motive sincerity, which will, in turn, affect purchase intentions.
We conducted three studies using similar study designs and representing different products, CSR initiatives (social and environmental), stimuli, and participant pools (total N = 507; for details, see Table 7). The three studies are each individually reported in detail in Web Appendix B. Here, however, in the interest of parsimony, we summarize all experimental results in the form of a single-paper meta-analysis, specifically utilizing an independent participant data (IPD) meta-analysis to allow for the test for mediating effects ([72]).[11] This enables us to provide a concise summary of our experimental results and process evidence based on all of the available data, while providing greater generalizability (i.e., a lower risk of idiosyncratic stimuli effects).
Graph
Table 7. Road Map of Experimental Studies That Show the Effect of CSR on Purchase Intentions.
| Study 1 | Study 2 | Study 3 |
|---|
| Distinguishing Feature | Packaging Stimuli/Laboratory Study | Alternative Process Variables/Retail Context | Tests Alternative Process Variables and Individual Differences |
|---|
| CSR domain | Social (health) | Social (labor) | Environmental |
| DV | Purchase intentions | Purchase intentions | Purchase intentions |
| Product category | Cold cereal | Coffee (chain) | Bottled water |
| Design | 3 × 2 | 3 × 3 | 3 × 2 |
| Participants | Students | Online panel | Online panel |
| Interactions | p < .001 | p < .001 | p = .05 |
| N | 181 | 176 | 150 |
5 Notes: DV = dependent variable.
All three studies employed the same base 2 (brand CSR reputation: favorable vs. unfavorable) × 3 (CSR type: corrective vs. compensating vs. cultivating) study design. To begin each study, participants (N = 507; 41.6% female; average age = 29.9 years) were given background information regarding a CPG brand in a particular category. All participants were told that these were actual brands but were either marketed exclusively in another country (Study 1; cold cereal) or were not identified for privacy reasons (Studies 2 and 3; coffee and bottled water, respectively). These brands were then described as being either relatively socially responsible (positive-CSR-reputation condition) or socially irresponsible (negative-CSR-reputation condition) compared with peer brands within their categories.[12]
After reading this background information about the focal brand, participants rated the likelihood that they would consider purchasing the brand (purchase intent) using a seven-point scale anchored on "not likely at all" ( 1) and "very likely" ( 7). Following this introduction and initial measurement, participants then read a description of a recent CSR initiative announced by the brand. This initiative represented an action that reduced the brand's own negative social or environmental impact (corrective CSR), addressed the brand's impact without actually reducing it (compensating CSR), or was an unrelated philanthropic gesture (cultivating CSR). After exposure to the CSR initiative, participants were then asked a second time about their purchase intentions. In addition, participants responded to measures about their perceptions of the brand's motives for the CSR initiative. Specifically, participants completed two seven-point items about how "sincere" and "genuine" they believed the brand's interest in the cause to be, and Study 3 included an additional item asking how much the brand "truly cares" about the initiative ([85]). Participants then rated the subjective fit of the CSR initiative with the brand ([10]) as an alternative process measure. Finally, participants provided demographic information including age and gender before the studies concluded. Thus, while stimuli details and CSR contexts differed to better generalize results, the basic study designs were highly consistent, enabling a very straightforward single-paper (IPD) meta-analysis (for a summary of results, see Table 8).
Graph
Table 8. Results from a Meta-Analysis of Three Experimental Studies: Main Effects, Moderating Effects of CSR Reputation, and Indirect (Mediating) Effects of Perceived Sincerity for Each CSR Type in the Experimental Data.
| Main Effects of CSR Type on Purchase Intenta | Moderating Effect of CSR Reputation on the Relationship Between CSR Type and Purchase Intenta | Main Effects of CSR Type on Perceived Brand Sincerityb | Relative Indirect Effect of CSR Type on Purchase Intent Mediated by Perceived Sincerityc | Index of Moderated Mediationd |
|---|
| Correct | .869, p <.05 | −1.13, p <.001 | .210, p <.01 | (Reference category) | (Reference category) |
| Compensate | .582, p <.05 | −.275, p = .12 | −.024, n.s. | −.0473, p <.05 | −.075, p <.05 |
| Cultivate | .275, p <.05 | .124, n.s. | −.280, p <.01 | −.0992, p <.05 | −.063, p <.1 |
| Omnibus | p <.001 | p <.001 | p <.001 | — | — |
- 6 aMarginal means (column 1) and parameter estimates (column 2) from ANOVA results on purchase intent.
- 7 bMarginal means from ANOVA results on perceived sincerity.
- 8 cIndirect effects resulting from PROCESS Model 4, controlling for CSR reputation ([38]). The indirect effect represents the product of ( 1) the effect of CSR type on perceived sincerity and ( 2) the effect of perceived sincerity on purchase intentions.
- 9 dIndirect effects resulting from PROCESS Model 8 ([38]). The indirect effect represents the product of ( 1) the effect of CSR type on perceived sincerity, moderated by CSR reputation, and ( 2) the effect of perceived sincerity on purchase intentions.
The primary dependent variable of interest was the change (Δ) in purchase intentions from before and after the CSR information was presented to participants. Individual analyses of variance (ANOVAs) interacting CSR reputation and CSR type as factors were significant for all three studies (see Table 8). For the meta-analysis, we use an aggregated data set of all observations enabling our test for process (PROCESS Model 8; [38]), using indicator coding and controlling for study-level effects ([39]; [55]; [72]).
Consistent with the model, the studies individually and collectively show a significant interaction between CSR type and CSR reputation (F( 2, 442) = 13.67, p <.001), with significant main effects for CSR reputation (F( 1, 442) = 17.94, p <.001) and CSR type (F( 2, 442) = 11.88, p <.001). Corrective CSR again produced the most positive consumer response (M = .869, SE = .087, p <.05), followed by compensating CSR (M = .582, SE = .089, p <.05). Cultivating CSR again proved the least effective (M = .275, SE = .086, p <.05; see Figure 2 and Table 8). All contrasts between CSR types were significant. Although cultivating CSR was again significantly less effective than the other two CSR types, the net effect of cultivating CSR across the studies was positive overall, which itself was not consistent with model results. We expect that this was due to simple anchoring effects given the laboratory setting and study procedures. It is also relevant that while the effect was positive within the high-CSR-reputation condition, the effects of cultivating CSR were not significant within the low-CSR-reputation condition (M = .213, SE = .115, n.s.). Thus, the overall pattern of results was largely consistent with the pattern obtained from our empirical analysis using brand sales. The exception was cultivating CSR, which, while obtaining lower evaluation than the other two types, was nevertheless not negatively viewed.
Graph: Figure 2. Results from a meta-analysis of three experimental studies: change in purchase intentions across CSR types and CSR reputation conditions.
The moderating effect of CSR reputation was also consistent with model results, such that a high CSR reputation attenuated the positive effects of corrective and compensating CSR but improved the effect of cultivating CSR on purchase intentions (see Figure 2). Overall, the main-effect differences between CSR types were primarily driven by differences in the unfavorable-CSR-reputation conditions. Thus, H2 is supported in the studies, while H1 is partially supported. The pattern of results largely mimics those found with the brand sales model.
The effects of CSR type on purchase intentions were mediated by perceived sincerity (H4a). An ANOVA on perceived brand sincerity[13] showed main effects of CSR reputation (F( 1, 442) = 104.31, p <.001) and of CSR type (F( 2, 442) = 11.94, p <.001) with a marginally significant moderating interaction (F( 2, 442) = 2.43, p = .089). Not surprisingly, brands with better CSR reputations were viewed as having more sincere motives for their CSR actions (M = .411) than brands with lower reputations (M = −.446). More importantly, among CSR types, corrective CSR actions were perceived as the most sincere (M = 2.55, t(150) = 3.50, p <.001), followed by compensating CSR actions (M = −.035, t(142) = −.414, p = .680) and cultivating CSR actions, which was viewed as significantly insincere (M = −.313, t(156) = −3.79, p <.001).[14] In turn, perceived sincerity positively affected purchase intentions (β = .203, t = 3.52, p <.001). The mediating effects were significant, as perceived sincerity mediated the effects of compensating CSR (a1 × b1 = −.0473, 95% confidence interval [CI] = [−.1084, −.0044], p <.05) and cultivating CSR (a2 × b1 = −.0992, 95% CI = [−.1875, −.0311], p <.05) relative to corrective CSR, in support of H4a.
Beyond the mediation of the main effects, we also found evidence for moderated mediation (H4b). The moderating effect of CSR reputation on perceived sincerity was similar to that observed with purchase intentions, such that differences in perceived sincerity were more pronounced within brands of lower CSR reputations (F( 2, 442) = 11.21, p <.001) than brands with higher CSR reputations (F( 2, 442) = 2.86, p = .056). In a test of the entire model, a moderated mediation analysis (PROCESS Model 8) returned significant indices of moderated mediation through perceived sincerity for both compensating CSR (a1 × b1 = −.075, 95% CI = [−.1921, −.0032], p <.05) and cultivating CSR (a2 × b1 = −.063, 90% CI = [−.1493, −.0014], p <.10), relative to the corrective CSR condition. The latter moderated mediation effect was only marginally significant, as the indirect effects for cultivating actions by brands with favorable and unfavorable CSR reputations were both negative and significant (favorable: a2 × b1 = −.1194, 95% CI = [−.2364, −.0046], p <.05; unfavorable: a2 × b1 = −.0562, 95% CI = [−.1294, −.0046], p <.05). Overall, the moderated mediation results predominantly support H4b.
Perceived fit of the CSR initiative with the brand, which was tested as an alternative mediator, did mediate the main effects of CSR type on purchase intentions (compensating: a1 × b1 = −.0331, 95% CI = [−.0743, −.0079], p <.05; cultivating: a2 × b1 = −.0753, 95% CI = [−.1538, −.0172], p <.05). However, perceived fit was not significant in the moderated-mediation model (PROCESS Model 8), as the index of moderated mediation was not significant for either compensating (a1 × b1 = −.0378, 90% CI = [−.0912,.0049], p >.10) or cultivating (a2 × b1 = .0315, 90% CI = [−.0164,.0919], p >.10) CSR actions, relative to corrective CSR. Thus, while perceived fit may thus help explain the main effects of CSR on sales response, fit does not appear to fully explain the effects observed in the model and studies.
The results from three laboratory experiments lend support to the results documented with secondary data, but more importantly, they provide process evidence for the underlying effect. As was the case in the brand sales model, participants rewarded corrective CSR actions with increased purchase intentions. The moderating effect of brand CSR reputation also shows a similar pattern, attenuating the positive effects for both corrective and compensating CSR, and improving purchase intention outcomes for cultivating CSR. These effects were driven in part by subjects' inferences regarding the sincerity of the brand's motives behind the CSR initiatives. Overall, the results of these studies are consistent with our findings from the sales data, with the exception of cultivating CSR not being negatively viewed by participants. Finally, while we did not explicitly manipulate the environmental versus social focus of CSR action in these studies, we note that the CSR actions used in Studies 1 and 2 are socially focused, the CSR actions used in Study 3 are environmentally focused, and the direction of the effects across these two sets of studies is consistent with the moderating effect hypothesized in H3.
This article proposes a typology of CSR activities that is based on demonstrating accountability for the impact that brands have on consumers and the environment. Using both brand-level sales data as well as data from lab experiments, we show that consumers respond positively to brands that undertake corrective and compensating actions, but not to those that engage in cultivating goodwill actions. We also show that the effects of these actions on sales and purchase intentions are mitigated by high CSR reputation. Further, we show that an environmental CSR focus, relative to a social one, strengthens the positive effect of corrective CSR and weakens the negative effect of cultivating CSR. Finally, we find that perceptions of brand sincerity mediate the effect of CSR actions on purchase intentions: corrective and compensating CSR, which suggest a higher desire to correct brand liabilities, are perceived as more sincere and increase purchase intentions, whereas cultivating goodwill CSR is viewed as less sincere and does not appear to lead to any changes in purchase intentions.
The CSR literature is sizable and already includes classifications of CSR activities into business practice versus philanthropic, reactive versus proactive, or environmental versus social. However, such categorizations are broad and do not provide a direct link to what would be the most suitable type of CSR for each firm. By focusing on CSR activities that address the negative externalities associated with firms' operations, our typology establishes this link, while also offering an umbrella large enough to encompass the full spectrum of CSR efforts typically undertaken by firms. Our work connects the CSR stream of literature with the one on brand harm crises and with image restoration theory. This work suggests an underlying mechanism for the success of corrective and compensating CSR initiatives, which starts with firms acknowledging their externalized costs and selecting appropriate compensatory CSR actions and results in firms enjoying positive consumer outcomes including increased perceptions of firm sincerity and stronger purchase behavior. In contrast, our results suggest that cultivating goodwill is a special type of CSR and that stakeholders' reactions to it require further scholarly and practitioner inquiry. This finding is in line with recent research that has documented negative employee outcomes to cash donations to a nonprofit, which can be classified as cultivating CSR ([54]).
Moreover, our framework goes beyond this categorization and also allows for a comparison of two broad classes of CSR—environmentally versus socially focused—which can both materialize within each of the three types of CSR included in our categorization ([22]). Thus, assessing the intersection of environmental and social focus with our typology deepens our understanding of CSR as a complex and multidimensional construct. In line with research finding that consumers view environmental issues as relatively more important than social ones ([67]), our results suggest that an environmental focus further increases the positive effect of corrective CSR initiatives on brand sales, while it mitigates the negative effect of cultivating CSR. The intriguing finding that an environmental focus does not appear to help compensating CSR initiatives could be due to a heightened awareness of the cost that the firm's operations have on the environment, which may not be adequately addressed by firms' compensation efforts.
We also highlight the moderating effect of CSR reputation, as prior work finds that consumers may react differently to the same CSR actions depending on whether the perception is that the actions are isolated endeavors or typical of the brand ([26]). CSR reputation is especially informative for consumers given the potential for greenwashing in the CSR space. We add to the literature by outlining some of the reasons for the differences in consumer reactions to the same CSR actions. High CSR reputation provides a ceiling effect for potential rewards associated with such initiatives, while lower-CSR-reputation brands' corrective and compensating CSR engagement leads to increased perceptions of brand sincerity and ultimately brand sales. Moreover, our research adds a new facet to the stream of literature that has examined CSR reputation in the context of tempering negative evaluations associated with brand harm crises ([26]; [50]). The CSR typology proposed in this article—by focusing on firms taking accountability for "everyday" harm, rather than reacting to distinct crises—expands the role of CSR as a mitigating factor of a broader set of negative externalities associated with the activities of firms.
Our theory also allows us to reconcile some of the negative effects of CSR previously documented in the literature, such as that of profit-oriented CSR in [ 8], activities focused on community support in [ 4], and agency costs in [79]. Our accountability-based framework offers overarching insight into these negative effects. We propose that these negative effects are more likely to occur when the CSR actions appear to be disconnected from the brand's own footprint, are not perceived as sincere and do not suggest that the firm wants to counteract some of the negative externalities associated with its operations. Consumers may find CSR actions aimed at cultivating goodwill to be less sincere. They may attribute this lack of sincerity to profit orientation, support of issues that they do not agree with, or engagement in corporate philanthropy that enhances managers' personal reputations. The experiments that we present in our article not only establish brand sincerity as a mediator of the relationship between CSR and purchase intentions in three very different scenarios but also highlight the fact that cultivating CSR, perceived as being the least sincere of the three types of initiative, fails to sway consumers toward the brand.
Our results also carry implications for managers. With the caveat that our (observational) data are not particularly suited for enabling strong normative claims about the type of CSR actions firms should undertake, we propose that one of the key managerial takeaways of our study is that brands can benefit from emphasizing the accountability of their CSR efforts, particularly if these efforts address environmental issues. Furthermore, our results suggest that managers should reconsider engaging in CSR actions that cannot be clearly linked to the brand's perceived negative externalities on either society or the environment. This is critical, as consumers are becoming ever more aware of the potentially harmful effects of brands' business operations on both societal and environmental dimensions. Brands in the CPG category, in particular, are increasingly being taken to task on these issues. For example, Coca-Cola was named top plastic polluter for the third year in a row, beating out other top polluters Pepsi and Nestlé. Coca-Cola was accused of making zero progress on plastic waste reduction, with its beverage bottles found littered on beaches, rivers, and parks ([61]). At the same time, Coca-Cola recently celebrated its having awarded over $73 million in college scholarships over the last 25 years ([20]). Our results suggest that, as a general guiding philosophy, brands should strive to "clean up their own mess" before engaging in general charitable efforts that may otherwise be seen as insincere efforts aimed squarely at garnering consumer goodwill. While college scholarship donations are a worthy endeavor, our study suggests that a focus on plastic waste reduction may resonate more strongly with Coca-Cola's consumers. It is thus not surprising that in response to being named "top plastic polluter," Coca-Cola has highlighted its commitment to recycling every one of its beverage bottles by 2030 ([61]).
In addition to helping managers better understand the consequences of various CSR activities, our findings also suggest what aspects of these initiatives have to be clearly communicated to the public. Accountability and efforts to reduce negative externalities should be highlighted in companies' press releases about CSR initiatives, as these appear to lead to positive consumer outcomes.
There may be unaccounted factors that could impact the effect of CSR actions on brand sales. For instance, the extent of resource investment (in terms of both financial resources and effort) into the CSR rollout process may play a role in influencing sales returns and the success of the CSR initiatives. However, information on brand investments into CSR is proprietary, hard to quantify, or both. We partially account for the support given to CSR actions in two ways—( 1) by including time-varying advertising spending at the brand level as a covariate in our regressions and ( 2) by making the brand's identity constant in our experimental analyses—but there could be other strategic considerations that drive CSR that we were not able to incorporate in our analysis.
Our research represents a first step in better understanding the effects of CSR initiatives on brand sales. We have examined brand sales up to a year after an initiative was announced, but CSR activities may also strengthen brand loyalty and satisfaction, which could in turn lead to strong longer-term outcomes. Identifying additional boundary conditions for the effects we present in this article can also help managers make better CSR choices. Likewise, although we test and find evidence for one proposed mediator, we recognize that this process may involve further unexplored nuance. For instance, might CSR affect other outcome variables (e.g., consumer identification with the brand, product attractiveness, strength of partnership with retailers) that help mediate and explain the differences in sales? Systematically understanding when and why consumers respond in particular ways to these CSR types may be a promising area of future inquiry.
An interesting question that stems from our findings is the sequencing of the types of CSR at the firm level. While our results indicate that cultivating initiatives lead to negative outcomes, it is plausible that this effect could be softened or even reversed if a firm has already established a strong history of accountability, having already corrected and compensated for its negative externalities through CSR initiatives conducted in the past. With more companies engaging in CSR, researchers may soon have access to data that allow them to expand on our research to examine the effects of a firm's rich history of CSR actions on brand sales, beyond merely controlling for corporate CSR reputation. Moreover, might it be wise for firms to deploy multiple CSR strategies at once? Could conducting cultivating and corrective CSR at the same time counteract the negative effects of cultivating on sales, or could it confuse consumers and hurt the brand's CSR reputation? Concurrent CSR actions may increase awareness and be more likely to be noticed by investors ([83]), but estimating the direction of their net effect may not be straightforward.
Finally, while our justification for the negative effects of cultivating initiatives is rooted in theory (lower accountability) and supported by data from experiments showing that consumers view brands that engage in this type of CSR as less sincere, the question of why firms pursue such activities, beyond the obvious tax benefits, remains open. Fully answering this question is beyond the scope of this article, but one possibility is that firms may target other stakeholders, such as employees or shareholders, with these efforts. For instance, it could be that employee engagement in some cultivating actions may lead to positive employee outcomes, such as higher productivity and retention, which may help offset some of the negative effects on sales. Moreover, cultivating CSR is directed at a set of stakeholders that is in many cases distinct from the consumer base. Consequently, it would be valuable to take a closer look at which external group of stakeholders can be targeted with cultivating CSR and how these actions should be framed and communicated so that consumers view them more favorably.
We show that brand decisions to engage in CSR present both opportunities and challenges. By reducing their negative impact or footprint first, brands appear to win consumers' approval to the greatest degree. However, stepping beyond its natural purview may be met with cynicism if a brand has not yet met a certain standard for its own behavior. For brands with already sterling reputations, the prospect of further impressing consumers can be more challenging, though doing good outside their footprint in the form of philanthropic efforts becomes an option. These results provide practical guidance for managers making decisions about their own CSR. Overall, it is both encouraging and promising to note that business, consumer, social, and environmental interests can align in the form of businesses genuinely reducing their adverse impact for global betterment.
sj-pdf-1-jmx-10.1177_00222429211044155 - Supplemental material for The Impact of Corporate Social Responsibility on Brand Sales: An Accountability Perspective
Supplemental material, sj-pdf-1-jmx-10.1177_00222429211044155 for The Impact of Corporate Social Responsibility on Brand Sales: An Accountability Perspective by Dionne Nickerson, Michael Lowe, Adithya Pattabhiramaiah and Alina Sorescu in Journal of Marketing
Footnotes 1 This article is based on the first essay of the first author's doctoral dissertation at the Georgia Institute of Technology. All authors contributed equally to this research and are listed in random order. The standard disclaimer applies.
2 Els Gijsbrechts
3 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
4 The author(s) received no financial support for the research, authorship, and/or publication of this article.
5 Adithya Pattabhiramaiah https://orcid.org/0000-0002-0501-9937 Alina Sorescu https://orcid.org/0000-0002-3625-0656
6 Throughout the article, we use the labels "cultivating goodwill CSR" and "cultivating CSR" interchangeably, for ease of exposition.
7 The ability to track weekly sales information at the local market level for each of these brands allows for more granular causal inference (i.e., it enables us to adopt detailed controls for brand-, market-, and time-specific influences driving brand sales) that may otherwise interfere with the inferred effect of CSR engagement.
8 Our results are also robust to including control brands that did not engage in CSR in the two years immediately preceding the focal brand's CSR announcement date.
9 To account for the self-selection of firms' CSR choices, which can take four unique values (j=1, 2, 3, or 0), we follow [29], p. 356) and include three inverse Mills ratio terms, under their assumption that where r denotes the correlation between the error terms corresponding to the multinomial logit selection Equation 1 and the outcome Equation 2 (see [29], p. 352] and [2], p. 29]).
We add a small integer before taking logs to get around instances of zeroes in our data.
Results were consistent when we followed the procedure outlined by [60], though we utilize the IPD approach herein to allow for tests of mediation.
Study 2 also included a "low salience" or control condition in which no background information regarding CSR reputation was given. This was not used in the meta-analysis and did not meaningfully impact results.
Standardized due to the different items in Study 3.
These results are compared with the scale midpoint. Results remain consistent when using unstandardized values and scale midpoint for perceived sincerity.
References Aaker Jennifer , Fournier Susan , Brasel S. Adam. (2004), " When Good Brands Do Bad ," Journal of Consumer Research , 31 (1), 1 – 16.
Adams Richard H. Jr. , Cuecuecha Alfredo. (2013), " The Impact of Remittances on Investment and Poverty in Ghana ," World Development , 50 (October) , 24 – 40.
Aggarwal Pankaj. (2004), " The Effects of Brand Relationship Norms on Consumer Attitudes and Behavior ," Journal of Consumer Research , 31 (1), 87 – 101.
Ailawadi Kusum , Ma Y. , Grewal Dhruv. (2018), " The Club Store Effect: Impact of Shopping in Warehouse Club Stores on Consumers' Packaged Food Purchases ," Journal of Marketing Research , 55 (2), 193 – 207.
Ailawadi Kusum L. , Neslin Scott A. , Luan Y. Jackie , Taylor Gail Ayala. (2014), " Does Retailer CSR Enhance Behavioral Loyalty? A Case for Benefit Segmentation ," International Journal of Research in Marketing , 31 (2), 156 – 67.
Alhouti Sarah , Johnson Catherine M. , Holloway Betsy Bugg. (2016), " Corporate Social Responsibility Authenticity: Investigating Its Antecedents and Outcomes ," Journal of Business Research , 69 (3), 1242 –4 9.
Anselmsson Johan , Johansson Ulf. (2007), " Corporate Social Responsibility and the Positioning of Grocery Brands: An Exploratory Study of Retailer and Manufacturer Brands at Point of Purchase ," International Journal of Retail & Distribution Management , 35 (10), 835 – 56.
Atefi Yashar , Ahearne Michael , Hohenberg Sebastian , Hall Zachary , Zettelmeyer Florian. (2020), " Open Negotiation: The Back-End Benefits of Salespeople's Transparency in the Front End ," Journal of Marketing Research, 57 (6), 1076 – 94.
Auger Pat , Devinney Timothy M. , Louviere Jordan J. , Burke Paul F.. (2008), " Do Social Product Features Have Value to Consumers? " International Journal of Research in Marketing , 25 (3), 183 – 91.
Becker-Olsen Karen L. , Cudmore B. Andrew , Hill Ronald Paul. (2006), " The Impact of Perceived Corporate Social Responsibility on Consumer Behavior ," Journal of Business Research , 59 (1), 46 – 53.
Benoit William L. (1997), " Image Repair Discourse and Crisis Communication ," Public Relations Review , 23 (2), 177 – 86.
Berens Guido , van Riel Cees B.M. , van Bruggen Gerrit H.. (2005), " Corporate Associations and Consumer Product Responses: The Moderating Role of Corporate Brand Dominance ," Journal of Marketing , 69 (3), 35 – 48.
Bhagwat Yashoda , Warren Nooshin L. , Beck Joshua T. , Watson George F. IV. (2020), " Corporate Sociopolitical Activism and Firm Value ," Journal of Marketing , 84 (5), 1 – 21.
Bhardwaj Pradeep , Chatterjee Prabirendra , Dogerlioglu-Demir Kivilcim , Turut Ozge. (2018), " When and How Is Corporate Social Responsibility Profitable? " Journal of Business Research , 84 (March), 206 – 19.
Bourguignon François , Fournier Martin , Gurgand Marc. (2007), " Selection Bias Corrections Based on the Multinomial Logit Model: Monte Carlo Comparisons ," Journal of Economic Surveys , 21 (1), 174 – 205.
Bronnenberg Bart J. , Dubé Jean-Pierre , Mela Carl. (2010), " Do Digital Video Recorders Influence Sales? " Journal of Marketing Research , 47 (6), 998 – 1010.
Bronnenberg Bart J. , Kruger Michael W. , Mela Carl F.. (2008), " Database Paper—The IRI Marketing Data Set ," Marketing Science , 27 (4), 745 –4 8.
Buell Ryan W. , Kalkanci Basak. (2021), " How Transparency into Internal and External Responsibility Initiatives Influences Consumer Choice ," Management Science , 67 (2), 932 – 50.
Burbano Vanessa C. (2021), " The Demotivating Effects of Communicating a Social-Political Stance: Field Experimental Evidence From an Online Labor Market Platform ," Management Science , 67 (2), 1004 – 25.
Carlisle Robert D. , Tsang Jo-Ann , Ahmad Nadia Y. , Worthington Everett L. , van Oyen Witvliet Charlotte , Wade Nathaniel. (2012), " Do Actions Speak Louder Than Words? Differential Effects of Apology and Restitution on Behavioral and Self-Report Measures of Forgiveness ," Journal of Positive Psychology , 7 (4), 294 – 305.
Chernev Alexander , Blair Sean. (2015), " Doing Well by Doing Good: The Benevolent Halo of Corporate Social Responsibility ," Journal of Consumer Research , 41 (6), 1412 – 25.
Coca-Cola (2019), " Stakeholder Engagement ," (accessed May 22, 2019) , https://www.coca-colacompany.com/stories/stakeholder-engagement.
Coca-Cola (2020), " 2021 Semifinalists ," (accessed September 25, 2020) , http://www.coca-colascholarsfoundation.com/research-blog/2017-csr-study.
Cone Communications (2017), " 2017 Cone Communications CSR Study ," (accessed October 19, 2021), http://www.conecomm.com/research-blog/2017-csr-study.
Allstate Corporation (2009), " Allstate Recognized as One of America's Best Corporate Citizens ," CSRwire, press release (March 6) , https://www.csrwire.com/press%5freleases/26529-allstate-recognized-as-one-of-america-s-best-corporate-citizens.
Dahlsrud Alexander. (2008), " How Corporate Social Responsibility Is Defined: An Analysis of 37 Definitions ," Corporate Social Responsibility and Environmental Management , 15 (1), 1 – 13.
Dans Enrique. (2018), " Corporate Social Responsibility Is Turning Green, and That's a Good Thing ," Forbes (September 14) , https://www.forbes.com/sites/enriquedans/2018/09/14/corporate-social-responsibility-is-turning-green-and-thats-a-good-thing.
Del Valle Gaby. (2018), " Can Consumer Choices Ward Off the Worst Effects of Climate Change? An Expert Explains ," Vox (October 12) , https://www.vox.com/the-goods/2018/10/12/17967738/climate-change-consumer-choices-green-renewable-energy.
Du Shuili , Bhattacharya C.B. , Sen Sankar. (2007), " Reaping Relational Rewards from Corporate Social Responsibility: The Role of Competitive Positioning ," International Journal of Research in Marketing , 24 (3), 224 – 41.
Du Shuili , Bhattacharya C.B. , Sen Sankar. (2017), " The Business Case for Sustainability Reporting: Evidence from Stock Market Reactions ," Journal of Public Policy & Marketing , 36 (2), 313 – 30.
Dubbink Wim , Graafland Johan , van Liedekerke Luc. (2008), " CSR, Transparency and the Role of Intermediate Organisations ," Journal of Business Ethics , 82 (2), 391 – 406.
Dubin Jeffrey , McFadden Daniel. (1984) " An Econometric Analysis of Residential Electric Appliance Holdings and Consumption ," Econometrica , 52 (2), 345 – 62.
Dutta Sujay , Pullig Chris. (2011), " Effectiveness of Corporate Responses to Brand Crises: The Role of Crisis Types and Response Strategies ," Journal of Business Research , 64 (12), 1281 –8 7.
Ellen Pam Scholder , Webb Deborah J. , Mohr Lois A.. (2006), " Building Corporate Associations: Consumer Attributions for Corporate Socially Responsible Programs ," Journal of the Academy of Marketing Science , 34 (2), 147 – 57.
Fang Eric , Lee Jongkuk , Palmatier Robert , Guo Zhaoyang. (2016), " Understanding the Effects of Plural Marketing Structures on Alliance Performance ," Journal of Marketing Research , 53 (4), 628 – 45.
Fombrun Charles , Shanley Mark. (1990), " What's in a Name? Reputation Building and Corporate Strategy ," Academy of Management Journal , 33 (2), 233 – 58.
Gilbert Daniel T. , Malone Patrick S.. (1995), " The Correspondence Bias ," Psychological Bulletin , 117 (1), 21 – 38.
Gopaldas Ahir. (2014), " Marketplace Sentiments ," Journal of Consumer Research , 41 (4), 995 – 1014.
Griffin Paul A. , Sun Estelle. (2013), " Going Green: Market Reaction to CSRwire News Releases ," Journal of Accounting and Public Policy , 32 (2), 93 – 113.
Groza Mark D. , Pronschinske Mya R. , Walker Matthew. (2011), " Perceived Organizational Motives and Consumer Responses to Proactive and Reactive CSR ," Journal of Business Ethics , 102 (4), 639 – 52.
Hayes Andrew F. (2017), Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach. New York : Guilford Press.
Hayes Andrew F. , Preacher Kristopher J.. (2014), " Statistical Mediation Analysis with a Multicategorical Independent Variable ," British Journal of Mathematical and Statistical Psychology , 67 (3), 451 – 70.
Heal Geoffrey. (2005), " Corporate Social Responsibility: An Economic and Financial Framework ," The Geneva Papers on Risk and Insurance-Issues and Practice , 30 (3), 387 – 409.
Hoeffler Steve , Keller Kevin Lane. (2002), " Building Brand Equity Through Corporate Societal Marketing ," Journal of Public Policy & Marketing , 21 (1), 78 – 89.
Homburg Christian , Stierl Marcel , Bornemann Torsten. (2013), " Corporate Social Responsibility in Business-to-Business Markets: How Organizational Customers Account for Supplier Corporate Social Responsibility Engagement ," Journal of Marketing , 77 (6), 54 – 72.
Homburg Christian , Vollmayr Josef , Hahn Alexander. (2014), " Firm Value Creation Through Major Channel Expansions: Evidence from an Event Study in the United States, Germany, and China ," Journal of Marketing , 78 (3), 38 – 61.
Hughes Brian. (2016), " Why Corporate Social Responsibility Is Essential for Brand Strategy ," Huffington Post (February 22) , https://www.huffpost.com/entry/why-corporate-social-resp%5fb%5f9282246.
Hydock Chris , Paharia Neeru , Blair Sean. (2020), " Should Your Brand Pick a Side? How Market Share Determines the Impact of Corporate Political Advocacy ," Journal of Marketing Research , 57 (6), 1135 – 51.
Inoue Yuhei , Funk Daniel C. , McDonald Heath. (2017), " Predicting Behavioral Loyalty Through Corporate Social Responsibility: The Mediating Role of Involvement and Commitment ," Journal of Business Research , 75 (June), 46 – 56.
Jayachandran Satish , Kalaignanam Kartik , Eilert Meike. (2013), " Product and Environmental Social Performance: Varying Effect on Firm Performance: Research Notes and Commentaries ," Strategic Management Journal , 34 (10), 1255 – 64.
Judson Ruth A. , Owen Ann L.. (1999), " Estimating Dynamic Panel Data Models: A Guide for Macroeconomists ," Economics Letters, 65 (1), 9 – 15.
Kang Charles , Germann Frank , Grewal Rajdeep. (2016), " Washing Away Your Sins? Corporate Social Responsibility, Corporate Social Irresponsibility, and Firm Performance ," Journal of Marketing , 80 (2), 59 – 79.
Klein Jill , Dawar Niraj. (2004), " Corporate Social Responsibility and Consumers' Attributions and Brand Evaluations in a Product–Harm Crisis ," International Journal of Research in Marketing , 21 (3), 203 – 17.
Korschun Daniel , Bhattacharya C.B. , Swain Scott D.. (2014), " Corporate Social Responsibility, Customer Orientation, and the Job Performance of Frontline Employees ," Journal of Marketing , 78 (3), 20 – 37.
Kotler Philip. (2011), " Reinventing Marketing to Manage the Environmental Imperative ," Journal of Marketing , 75 (4), 132 –3 5.
Lantos Geoffrey P. (2001), " The Boundaries of Strategic Corporate Social Responsibility ," Journal of Consumer Marketing , 18 (7), 595 – 632.
List John A. , Momeni Fatemeh. (2021), " When Corporate Social Responsibility Backfires: Evidence From a Natural Field Experiment ," Management Science , 67 (1), 8 – 21.
Lowe Michael L. , Haws Kelly L.. (2019), " Confession and Self-Control: A Prelude to Repentance or Relapse? " Journal of Personality and Social Psychology , 116 (4), 563 – 81.
Luchs Michael G. , Naylor Rebecca Walker , Irwin Julie R. , Raghunathan Rajagopal. (2010), " The Sustainability Liability: Potential Negative Effects of Ethicality on Product Preference ," Journal of Marketing , 74 (5), 18 – 31.
Luo Xueming , Bhattacharya C.B.. (2006), " Corporate Social Responsibility, Customer Satisfaction, and Market Value ," Journal of Marketing , 70 (4), 1 – 18.
Luo Xueming , Bhattacharya C.B.. (2009), " The Debate over Doing Good: Corporate Social Performance, Strategic Marketing Levers, and Firm-Idiosyncratic Risk ," Journal of Marketing , 73 (6), 198 – 213.
Margolis Joshua D. , Elfenbein Hillary Anger , Walsh James P.. (2009), " Does It Pay to Be Good…and Does It Matter? A Meta-Analysis of the Relationship Between Corporate Social and Financial Performance ," SSRN , http://www.ssrn.com/abstract=1866371.
McShane Blakeley B. , Böckenholt Ulf. (2017), " Single Paper Meta-Analysis: Benefits for Study Summary, Theory-Testing, And Replicability ," Journal of Consumer Research , 43 (6), 1048 – 63.
McVeigh Karen. (2020), " Coca-Cola, Pepsi, and Nestle Named Top Plastic Polluters for Third in a Row ," The Guardian (December 7), https://www.theguardian.com/environment/2020/dec/07/coca-cola-pepsi-and-nestle-named-top-plastic-polluters-for-third-year-in-a-row.
McWilliams Abagail , Siegel Donald. (2001), " Corporate Social Responsibility: A Theory of the Firm Perspective ," Academy of Management Review , 26 (1), 117 – 27.
Miller Stewart R. , Eden Lorraine , Li Dan. (2020). " CSR Reputation and Firm Performance: A Dynamic Approach ," Journal of Business Ethics , 163 (3), 619 – 36.
Mishra Saurabh , Modi Sachin B.. (2016), " Corporate Social Responsibility and Shareholder Wealth: The Role of Marketing Capability ," Journal of Marketing , 80 (1), 26 – 46.
Moorman Christine , Wies Simone , Mizik Natalie , Spencer Fredrika J.. (2012), " Firm Innovation and the Ratchet Effect Among Consumer-Packaged Goods Firms ," Marketing Science , 31 (6), 934 – 51.
Newman George E. , Gorlin Margarita , Dhar Ravi. (2014), " When Going Green Backfires: How Firm Intentions Shape the Evaluation of Socially Beneficial Product Enhancements ," Journal of Consumer Research , 41 (3), 823 – 39.
Oberseder Magdalena , Schlegelmilch Bodo B. , Murphy Patrick E.. (2013), " CSR Practices and Consumer Perceptions ," Journal of Business Research , 66 (10), 1839 – 51.
Olsen Mitchell C. , Slotegraaf Rebecca J. , Chandukala Sandeep R.. (2014), " Green Claims and Message Frames: How Green New Products Change Brand Attitude ," Journal of Marketing , 78 (5), 119 – 37.
Oster Emily. (2019), " Unobservable Selection and Coefficient Stability: Theory and Evidence ," Journal of Business & Economic Statistics , 37 (2), 187 – 204.
Peloza John , White Katherine , Shang Jingzhi. (2013), " Good and Guilt-Free: The Role of Self-Accountability in Influencing Preferences for Products with Ethical Attributes ," Journal of Marketing , 77 (1), 104 – 19.
Porter Michael E. and Kramer Mark R.. (2019), " Creating Shared Value: How to Reinvent Capitalism—and Unleash a Wave of Innovation and Growth ," in Managing Sustainable Business , Lenssen G.G. , Smith N.C. , eds. Dordrecht, Netherlands : Springer , 323 – 46.
Riley Richard D. , Lambert Paul C. , Abo-Zaid Ghada. (2010), " Meta-Analysis of Individual Participant Data: Rationale, Conduct, and Reporting ," BMJ , 340 , c221.
Roodman David. (2009), " How to Do Xtabond2: An Introduction to Difference and System GMM in Stata ," The Stata Journal , 9 (1), 86 – 136.
Sen Sankar , Bhattacharya C.B.. (2001), " Does Doing Good Always Lead to Doing Better? Consumer Reactions to Corporate Social Responsibility ," Journal of Marketing Research , 38 (2), 225 – 43.
Servaes Henri , Tamayo Ane. (2013), " The Impact of Corporate Social Responsibility on Firm Value: The Role of Customer Awareness ," Management Science , 59 (5), 1045 – 61.
Srinivasan Raji , Lilien Gary L. , Sridhar Shrihari. (2011), " Should Firms Spend More on Research and Development and Advertising During Recessions? " Journal of Marketing , 75 (3), 49 – 65.
Tarofder Arun Kumar , Nikhashemi Seyed Rajab , Ferdous Azam S.M. , Selvantharan Prashantini , Haque Ahasanul. (2016). " The Mediating Influence of Service Failure Explanation on Customer Repurchase Intention Through Customers Satisfaction ," International Journal of Quality and Service Sciences , 8 (4), 516 – 35.
Van Heerde Harald J. , Gijsenberg Maarten J. , Dekimpe Marnik G. , Steenkamp Jan-Benedict E.M.. (2013), " Price and Advertising Effectiveness over the Business Cycle ," Journal of Marketing Research , 50 (2), 177 – 93.
Wagner Tillmann , Lutz Richard J. , Weitz Barton A.. (2009), " Corporate Hypocrisy: Overcoming the Threat of Inconsistent Corporate Social Responsibility Perceptions ," Journal of Marketing , 73 (6), 77 – 91.
Wang Heli , Choi Jaepil , Li Jiatao. (2008), " Too Little or Too Much? Untangling the Relationship Between Corporate Philanthropy and Firm Financial Performance ," Organization Science , 19 (1), 143 – 59.
Warren Nooshin L. , Alina Sorescu (2017a), " Interpreting the Stock Returns to New Product Announcements: How the Past Shapes Investors' Expectations of the Future, " Journal of Marketing Research , 54 (5), 799 – 815.
Warren Nooshin L. , Sorescu Alina (2017b), " When 1 + 1 > 2: How Investors React to New Product Releases Announced Concurrently with Other Corporate News ," Journal of Marketing , 81 (2), 64 – 82.
Wilcox Keith , Vallen Beth , Block Lauren , Fitzsimons Gavan J.. (2009), " Vicarious Goal Fulfillment: When the Mere Presence of a Healthy Option Leads to an Ironically Indulgent Decision ," Journal of Consumer Research , 36 (3), 380 – 93.
Wooldridge Jeffrey M.. (1995), " Selection Corrections for Panel Data Models under Conditional Mean Independence Assumptions ," Journal of Econometrics , 68 (1): 115 – 132.
Yoon Yeosun , Gurhan-Canli Zeynep , Schwarz Norbert. (2006), " The Effect of Corporate Social Responsibility (CSR) Activities on Companies with Bad Reputations ," Journal of Consumer Psychology , 16 (4), 377 – 90.
~~~~~~~~
By Dionne Nickerson; Michael Lowe; Adithya Pattabhiramaiah and Alina Sorescu
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 121- The Impact of Organic Specialist Store Entry on Category Performance at Incumbent Stores. By: Maesen, Stijn; Lamey, Lien. Journal of Marketing. May2022, p1. DOI: 10.1177/00222429221090983.
Ahead of Print- Database:
- Business Source Complete
Record: 122- The Impact of Platform Protection Insurance on Buyers and Sellers in the Sharing Economy: A Natural Experiment. By: Xueming Luo; Siliang Tong; Zhijie Lin; Cheng Zhang. Journal of Marketing. Mar2021, Vol. 85 Issue 2, p50-69. 20p. 1 Diagram, 10 Charts, 2 Graphs. DOI: 10.1177/0022242920962510.
- Database:
- Business Source Complete
Record: 123- The Influence of Social Norms on Consumer Behavior: A Meta-Analysis. By: Melnyk, Vladimir; Carrillat, François A.; Melnyk, Valentyna. Journal of Marketing. May2022, Vol. 86 Issue 3, p98-120. 23p. 1 Diagram, 5 Charts, 3 Graphs. DOI: 10.1177/00222429211029199.
- Database:
- Business Source Complete
The Influence of Social Norms on Consumer Behavior: A Meta-Analysis
Social norms shape consumer behavior. However, it is not clear under what circumstances social norms are more versus less effective in doing so. This gap is addressed through an interdisciplinary meta-analysis examining the impact of social norms on consumer behavior across a wide array of contexts involving the purchase, consumption, use, and disposal of products and services, including socially approved (e.g., fruit consumption, donations) and disapproved (e.g., smoking, gambling) behaviors. Drawing from reactance theory and based on a cross-disciplinary data set of 250 effect sizes from research spanning 1978–2019 representing 112,478 respondents from 22 countries, the authors examine the effects of five categories of moderators of the effectiveness of social norms on consumer behavior: ( 1) target behavior characteristics, ( 2) communication factors, ( 3) consumer costs, ( 4) environmental factors, and ( 5) methodological characteristics. The findings suggest that while the effect of social norms on approved behavior is stable across time and cultures, their effect on disapproved behavior has grown over time and is stronger in survival and traditional cultures. Communications identifying specific organizations or close group members enhance compliance with social norms, as does the presence of monetary costs. The authors leverage their findings to offer managerial implications and a future research agenda for the field.
Keywords: cultural influence; meta-analysis; reactance; social approval; social influence; social marketing; social norms marketing; social norm
Social norms shape consumer behavior. Defined as "rules and standards that are understood by members of a group, and that guide and/or constrain social behavior without the force of laws" ([26], p. 152), social norms influence various forms of everyday consumption, including food choices ([87]), responses to new products ([51]), and loyalty ([63]). For example, signs in a hotel stating that other hotel guests reuse their towels increase towel reuse ([38]). Social norms are often leveraged by marketers and policy makers to encourage various socially approved behaviors, such as conserving energy ([95], [96]), complying with product recalls ([85]), and making tax payments (Cabinet Office UK [22]). They are also used to discourage socially disapproved behaviors, such as polluting the environment ([109]) and smoking or excessive alcohol or drug use ([108]).
The academic literature examining social norms has produced conflicting findings ([61]; [95], [96]). Some studies report large-scale favorable results for using social norms to curb socially disapproved behaviors ([21]). [90], for example, report a significant reduction (13%) in the prevalence of impaired driving among students. However, some campaigns encouraging socially approved behaviors have backfired. For example, [95], [96]) find that social norms for energy preservation can increase energy consumption. These mixed findings suggest contingent effects of social norms on behavior. A second reason for mixed findings is that some research studies actual behavior, while other research examines behavioral intentions. A final reason for mixed findings may lie in the fact that the country context introduces cultural factors into the study of norms that are important to their impact.
Our article looks across a wide range of research on social norms across behaviors, time, and cultures to resolve these conflicting findings and to synthesize the extant literature on social norms. Specifically, we investigate the effects of social norms on actual consumer behavior and identify moderators of these effects, using reactance theory as a theoretical lens ([18]; [92]). We contend that the effectiveness of social norms varies with the level of consumer reactance they trigger ([18]); norms that are less likely to trigger reactance are more likely to be effective.
We conduct a meta-analysis that examines the effects of five categories of moderators of the effectiveness of social norms on consumer behavior, matching central factors that may induce reactance. Specifically, we examine how the relationship between social norms and behaviors depends on ( 1) social approval or disapproval of behavior and other target behavior characteristics, ( 2) communication factors, ( 3) consumer costs, ( 4) environmental factors (e.g., culture, time), and ( 5) methodological characteristics (e.g., type of sample, study).
We collected 250 effect sizes from 136 articles published between 1978 and 2019 across different fields (e.g., marketing, psychology, health, environmental studies), representing 112,478 respondents from 22 countries. In conducting this research, we encountered several meta-analyses related to social norms. However, most prior meta-analyses focus on a single behavior, such as condom usage ([98]), or else investigate limited set of communication factors, such as whether the norm is descriptive or injunctive ([76]; [91]). Moreover, most include consideration of behavioral intentions rather than actual behavior, which is our focus. Finally, some prior meta-analysis focus on studies that use a specific theoretical framework, such as the theory of planned behavior ([ 2]; [70]), which limits their generalizability.
We aim to go beyond these insights by investigating critical moderators that have not been addressed in prior research. We not only study the new moderator of socially approved versus disapproved behavior but also examine new and managerially actionable moderators, such as target behaviors, communication factors, and consumer costs. Importantly, this study investigates behaviors (observed or reported) rather than intentions and covers consumer behaviors across domains, regardless of the theoretical framework used in primary studies. With this comprehensive approach, we establish that social norms have significant impacts on behavior, but the effect varies systematically according to the influence of a wide range of moderators.
This research makes several contributions across domains. First, we go beyond previous meta-analyses and contribute to theories of reactance and social influence by uncovering previously overlooked moderators and establishing several new empirical generalizations. Second, for social norms marketing ([38]; [110]), we specify the effects of social norms for a broad spectrum of consumer behaviors and detail how practitioners and government officials can utilize actionable moderators, such as using appropriate communication elements for certain behaviors, countries, and consumers. This should improve their success rate which has been mixed to date.
Third, we contribute to the literature on cross-cultural marketing ([88]; [93]; [106]) by establishing how cultural differences can determine the effects of social norms on both socially approved and disapproved behaviors. Finally, we develop a comprehensive research agenda, based on insights from our meta-analysis.
Social norms are a shared understanding among members of a society about which behaviors are permitted, forbidden, or obligatory ([29]). They result from exposure to and observations of others' behavior and act as a "social proof," whereby consumers follow the actions or opinions of others, in the belief that "If everyone is doing it, it must be a sensible thing to do" ([25], p. 1015). Social norms serve as decision shortcuts for choosing how to behave in a given situation ([24]).
One of their distinctive features is that social norms are shared, which implies the existence of some group through which they spread ([26]). Humans maintain social harmony by complying with the social order and developing coping strategies to "fit in" ([66]) or "copy the successful" ([46]). Consequently, humans have an almost automatic propensity to learn social norms ([81]). Yet, this propensity does not necessarily result in compliance with them ([84]).
The reason is that unlike laws, social norms are informal—they regulate behaviors without formal enforcement ([43]), so consumers have the freedom to follow or violate social norms ([26]). Accordingly, their impact on behavior stems from two evolutionary desires: ( 1) for social acceptance or affiliation and ( 2) for avoiding negative social outcomes such as social exclusion ([ 8]; [66]). As people are free to comply, and are inclined to do so, understanding why they do not comply is key for identifying systematic differences in reactions to social norms.
Despite consumers' natural inclination to comply with social norms, research consistently shows that attempts at influence that cite social norms can evoke psychological reactance ([18]; [92]). Reactance stems from individuals cherishing their autonomy and freedom of choice. As [19], p. 420) explain, "For a given individual at a given time, there is a set of behaviors in which he believes he is free to engage. Any reduction or threat of reduction in that set of free behaviors arouses a motivational state, 'reactance,' which is directed toward reestablishment of the lost or threatened freedom." If consumers believe their freedom to engage in a specific behavior is threatened, this evokes reactance, which enhances the attractiveness of the threatened behavior.
Reactance theory is useful for approaching the vast, heterogeneous literature on social norms because it provides a broad theoretical lens for investigating the influence of diverse factors, including behavioral, communication, consumer, and environmental factors ([92]). Thus, we build on the key antecedents of reactance: consumer expectations of freedom and the extent of the freedom threat ([92]) to understand the drivers of systematic differences in the effects of social norms on behavior.
Consumers do not perceive all their behaviors as freedoms ([20]), so reactance is contingent on an expectation that the person can freely choose among different behavioral alternatives ([27]). Thus, consumers likely exhibit reactance to social norms that appear to undermine their freedom ([19]). If they lack expectations of freedom in the first place, social norms should trigger less reactance ([92]). Several studies affirm that consumer reactance to attempts to influence decreases if an experimental manipulation lowers their perceptions of choice freedom ([36]; [58]).
A social influence attempt that implies that someone is trying to reduce freedom represents a threat ([18]). This threat of social norms is exacerbated if the norms exert greater pressure for change ([20]). The threat level tends to reflect the way a social norm is communicated ([92]), so more forceful messages prompt more reactance ([59]; [111]). For example, research suggests reduced compliance with messages that advocate teetotalling rather than limited drinking ([ 9]). Yet, freedom threats may also stem from barriers to performing a behavior, such as consumer costs. When costs are a barrier to free choice, the aroused reactance is directed at maintaining the threatened behavior and therefore increasing its desirability ([27]).
Building on reactance theory, we identify different groups of moderators driving behavioral compliance with social norms, as shown in Figure 1. These moderators include target behavior characteristics, communication factors, consumer costs, and environmental factors.
Graph: Figure 1. Conceptual framework: Factors that influence the effect of social norms on behavior.
The effects of social norms may vary across behaviors because characteristics inherent to the behavior influence perceptions of the freedom to perform it and threat to that freedom.
Societies have developed social reinforcement mechanisms that encourage some behaviors and discourage others ([48]; [72]). We define socially approved behaviors as being explicitly encouraged by society (e.g., recycling, volunteering), socially acceptable (e.g., carpooling), and/or perceived as appropriate by society (e.g., using condoms). We define socially disapproved behaviors as being explicitly discouraged (e.g., smoking), socially unacceptable (e.g., littering), and/or perceived as inappropriate (e.g., binge drinking). A socially approved behavior evokes positive reinforcement via social outcomes, such as inclusion, acceptance, and affiliation ([24]). A socially disapproved behavior instead induces negative reinforcement via social consequences, such as social exclusion, alienation, or ridicule ([66]). Social approval versus disapproval of behaviors is thus a crucial factor that has implications for consumers' expectations of their freedoms to perform them.
Performing a socially disapproved behavior is potentially more damaging to society as a whole than failing to perform an approved behavior ([47]). Thus, to maintain social order, societies tend to be more punitive of disapproved behaviors ([33]; [46]). In contrast, not adopting an approved behavior is less harshly punished and sometimes can even bring positive benefits, such as elevation in inferred social status ([ 8]). Thus, consumers are less likely to perceive social norms regulating socially disapproved behaviors as limitations to their freedoms, which diminishes reactance and increases compliance with social norms discouraging these behaviors. Yet, social norms targeting socially approved behaviors are seen as freedom limitations, causing more reactance and reduced compliance. Thus, we expect social norms pertaining to socially disapproved behaviors to be more effective than those pertaining to socially approved behaviors (H1).
Existing behaviors are already performed by consumers, at least sometimes, in contrast to entirely new behaviors. Consumers already have exercised their freedom to perform the existing behaviors, so they may feel less threatened when encouraging existing behaviors and their reactance to social norms that target existing behaviors may be relatively low. In contrast, targeting a new behavior may represent a stronger threat to freedom and, thus, induce reactance and decrease compliance. Consistently, for example, compliance with hand-washing advice has been higher than compliance with mask-wearing advice during COVID-19 and other infection outbreaks ([32]; [103]). Hand washing is an existing behavior and engrained into daily routines, whereas mask wearing was new for most consumers and generated more reactance. Therefore, we expect social norms pertaining to existing behaviors to be more effective than those pertaining to new behaviors (H2).
Hedonic behaviors are those driven by pleasure-related goals and are evaluated primarily on the benefits related to enjoyment, taste, aesthetics, and symbolic meaning. Utilitarian behaviors, instead, are driven by functionality goals and are performed and evaluated primarily on the basis of functional, instrumental, and practical benefits ([23]). Social norms can pertain to both types, including utilitarian behaviors such as banking (e.g., "Most millennials use online banking") and hedonic ones such as buying cosmetics (e.g., "12 makeup bag must-haves"). But their effectiveness is not clear a priori. On the one hand, reactance to social norms might be higher for hedonic behaviors because consumers have a stronger desire to perform those behaviors as part of their sense of freedom ([79]). Consumers can leverage social norms to justify a desirable behavior for themselves and enhance their perceptions of freedom to perform it. For example, the justification that "everyone's doing it" is common for hedonic behaviors ([39]) and can increase perceived freedom for engaging in these behaviors. On the other hand, indulging in a hedonic behavior often prompts a sense of guilt, making it harder to justify ([75]; [83]), which may reduce consumers' perception of freedom. With these opposing predictions, we treat the effects of social norms on hedonic versus utilitarian behaviors as an empirical question.
Some behaviors benefit other people directly (e.g., donating to charity), whereas others have indirect benefits (e.g., recycling). Social norms stem from group considerations, so consumers' willingness to enact their freedom may decrease if they realize that others will be negatively affected by their social norm violations ([101]). Correspondingly, behaviors that have negative implications for others yield lower reactance levels ([34]). We therefore expect that when others benefit from the behavior, this will enhance the effect of social norms on that behavior (H3).
We define public behaviors as those that are performed in public or can be observed by others (e.g., using public transport), in contrast to private behaviors (e.g., reducing energy consumption at home). Private behaviors are not subject to others' scrutiny, so consumers' perception of freedom threat to perform them may be relatively low, which should decrease reactance ([92]). This argument would imply that compliance with social norms regulating private (vs. public) behaviors should be higher. Yet, for public behaviors, reactance may also be reduced but for a different reason. Specifically, consumers are often concerned with how others perceive them ([65]), which reduces their willingness to enact their freedom and, in turn, reduces reactance. This would suggest higher compliance with social norms regulating public (vs. private) behaviors. Given these two opposing predictions, we refrain from making a directional hypothesis about this variable.
The use of social norms can trigger reactance because communication factors influence the perceived threat to behavioral freedom ([ 9]; [92]). We consider several communication factors, such as how the norm is formulated as well as whether it benefits an organization, references specific groups, and includes explicit sanctions or rewards.
Social norms can be formulated as descriptive or injunctive. Descriptive norms describe typical behaviors of some relevant group and signal which behaviors are most popular ([25]; [95]). Injunctive norms instead prescribe certain behaviors and indicate what the target consumer should or should not do. For example, a list of "bestsellers" represents descriptive norms, but "ten must-read books" lists communicate injunctive norms. As injunctive norms convey explicit demands, which consumers likely perceive as forceful threats to their freedom, they should generate more reactance than descriptive norms ([69]; [109]). Consumers exposed to descriptive norms instead may come up with reasons for the behavior of the majority and adjust their own behavior accordingly, without much reactance ([95]). Therefore, we expect a stronger impact of descriptive (vs. injunctive) social norms on behaviors (H4).
We define social norm communications that reveal a specific entity, such as a firm or government body, which would benefit from compliance with the social norm, as organization-benefiting social norms. For example, "my friends subscribed to the university's gym program" would benefit the gym if the target consumer complied with this behavior. While specific entity matters, overall, social norms that refer to organizations tend to be more concrete and specific because they activate situational factors (i.e., where and when the norm applies) ([ 1]). Such specificity and concreteness diminish the general threat to freedom for consumers by limiting it to the particular situation, which lowers their reactance ([38]). We thus expect organization-benefiting social norms to be more effective than those that do not mention organizations (H5).
Communications about social norms often specify close group members—that is, people who are genetically related (e.g., family) or similar (e.g., close friends)—rather than refer to an abstract group (e.g., fellow citizens, people). Evolutionary predictions of social cooperation highlight kinship mechanisms. Namely, a request that activates a kin care motive reduces reactance and promotes compliance without expectations of reciprocation ([41]; [46]). The closer the relationship is, the less reactance consumers are likely to experience, enhancing norm compliance ([79]; [99]). Thus, we expect social norms referring to a close group member to be more effective than social norms referring to abstract or distant groups (H6).
Communications around social norms often refer to authority figures, or individuals who can exercise power over others, formally or informally (e.g., superiors, experts, government officials, teachers), to enhance compliance. [78] famous studies show that formal orders from an authority figure (real or perceived) increase obedience. Yet, because social norms are informal, being required to do something by an authoritative source may make the threat to freedom more salient and trigger reactance ([ 4]). For example, expert recommendations may lead to reactance and diminish compliance ([36]). For these reasons, we expect social norms referring to authority figures to be less effective than those that do not refer to authority figures (H7).
The sanctions and rewards associated with noncompliance and compliance with social norms might be either implicit, meaning they are indirect and left for consumers to infer, or explicit, meaning they are clearly stated. If sanctions and rewards are explicit, they might diminish behavioral compliance because they make the persuasive nature of the social norm message salient ([86]) and threaten freedom expressly ([58]). Both aspects increase the perceived threat to freedom to perform the behaviors and reactance ([ 4]). Thus, we expect social norms that specify potential sanctions (for failing to comply) (H8) or potential rewards (for complying) (H9) to be less effective than social norms that do not make those consequences explicit.
The costs incurred to perform a behavior can create barriers. For example, social norms may direct consumers to buy an electric car, which is considerably more expensive than a regular car. We believe such costs will increase consumer reactance ([92]). However, the direction of the cost effect is not clear a priori (see [27]). On the one hand, a high cost may signal the desirability or status of the behavior, thereby motivating compliance. On the other hand, a high cost may dissuade consumers from attempting the behavior. Given these two opposing forces, we refrain from making directional hypotheses about costs, including costs associated with effort, money, and time.
We define "effort costs" as the amount of physical or mental resources consumers must invest to comply with social norms. Some behaviors require more effort (e.g., exercising), others less (e.g., not littering). Social norms that require more effort demand greater behavioral change. They may either increase compliance by increasing the attractiveness of the effortful behavior or decrease compliance by decreasing the attractiveness of the effortful behavior (by derogating it because of reduced attainability) ([27]). Which of these two forces is stronger is an empirical question.
Consumers may incur additional monetary costs to comply with social norms. For example, buying organic rather than conventional food requires more monetary resources. However, reusing a hotel towel does not result in monetary costs. Monetary costs constitute a direct barrier to free choice because consumers must sacrifice extra resources to comply. When social norms regulate costly behaviors, the monetary costs may imply a greater threat to the freedom to engage in this behavior, enhancing the attractiveness of this option and increasing compliance with such social norms ([27]). Yet, monetary costs may also emphasize the unattainability of the option, which would reduce compliance. Thus, we treat the effect of monetary costs on compliance with social norms as an empirical question.
Compliance with social norms may require long-term (e.g., adhering to a healthy eating program) or temporary (e.g., reusing hotel towels) commitment. The temporal costs barrier is greater for behaviors with long-term commitments because these social norms impose more behavioral constraints than those that require only temporary commitments. Thus, on the one hand, consumers may also have stronger resistance to losing an option with potential longer-term consequences ([58]), which would increase compliance with social norms regulating longer-term behaviors. On the other hand, perceived unattainability of behaviors is also greater if they persist, now and into the future, rather than if they involve a single instance, which could decrease compliance with social norms involving longer-term commitment. Thus, we treat the effect of temporal costs on compliance with social norms as an empirical question.
Consumers form freedom expectations through socialization in a specific cultural environment at a particular time ([80]). Culture shapes expectations by providing a logic for acting both housed in members' knowledge and beliefs and observed in members' behaviors ([102]). In some cultures, the range of approved behaviors is wide, and behavioral transgressions of social norms are tolerated. Other cultures allow a narrower range of behaviors and exhibit lower tolerance for deviations from social norms ([74]; [106]). Consumer reactance and compliance to social norms should thus differ systematically across cultures ([94]).
To account for cultural differences, we adopt Inglehart's cultural framework ([52]) with two bipolar dimensions: traditional versus secular-rational and survival versus self-expression values. These dimensions have clear implications for reactance to social norms because they influence tolerance for transgressions (traditional–secular-relational) and the range of approved behaviors (survival–self-expression). Moreover, these dimensions are measured regularly, which enables us to account for cultural dynamics ([104]).[ 6]
This dimension contrasts traditional societies, in which religion is very important, and secular-rational societies, in which it is not ([52]). Traditional societies also emphasize deference to authority, absolute standards, cultural protectionism, and national pride, and they generally exhibit less tolerance for transgressions of social norms. Secular-rational societies reflect opposing values. We expect the effect of social norms on behavior to be stronger in cultures closer to the traditional (vs. the secular-rational) pole (H10) because they effectively restrict consumers' awareness and expectations of freedom, which should decrease reactance.
This dimension reflects transitions from industrial to postindustrial societies ([52]). Survival societies emphasize economic and physical security and familiar norms to maximize the predictability of others' behaviors, which results in a relatively narrow range of behaviors that may be perceived as freedoms. Consumers in survival societies have low expectations of personal freedoms and identify less freedom to be threatened ([54]). In contrast, self-expression values emphasize variety, imagination, and tolerance of outgroups. As societies move toward self-expression, people generally become freer to make choices for themselves ([106]), which enhances their reactance to social norms and decreases compliance. In cultures that value self-expression, noncompliance with social norms may even signal the person's freedom to be unique, which is valued by consumers of these societies ([40]; [106]). In contrast, in survival cultures, violation of social norms is more likely to jeopardize economic or physical security ([49]), diminishing perception of these behaviors as freedoms. Thus, we expect the effect of social norms on behavior to be stronger in cultures close to the survival (vs. the self-expression) pole (H11).
The human propensity to comply with social norms has resulted from evolutionary processes ([41]). Therefore, the effect of social norms on behaviors should be stable throughout the short time (in evolutionary terms) marketers have been using them as a persuasion strategy. Yet research into conformity to social pressures also indicates some changes over time, including studies that document that conformity in the United States has declined ([16]), increased ([60]), or fluctuated ([62]) due to changes in social media and the cohesiveness of society, among other things. Thus, the effectiveness of social norms over time is an empirical question.
Cultures also vary in moral freedom, which reflects the extent to which people make their own moral choices rather than being influenced by state intervention ([ 3]). We expect the effect of social norms on behavior to be stronger in countries lacking moral freedom (H12), because of the lower expectations of freedom in those countries.
Thus far, our discussion has focused on main effects. However, consumers do not learn social norms in isolation; instead, they become aware of freedoms to perform certain behaviors through socialization in a particular culture and by observing different behaviors over time ([80]). To the extent that different societies shape consumer awareness of social norms, we expect the effects of the behavior being socially approved versus disapproved to be moderated by environmental characteristics (i.e., culture and time).
Specifically, with respect to the traditional versus secular-rational cultural dimension, we note that participation in a world religion makes punishments for socially disapproved behaviors more salient to people ([47]). In Christianity, seven of the Ten Commandments start with the phrase "you shall not." In Judaism, of 613 mitzvot in the Torah, 365 (60%) forbid bad behaviors. Islam explicitly specifies an extended list of behaviors that are haram, or forbidden ([71]). Thus, in traditional cultures, where religion is more important, reactance to social norms that target socially disapproved (vs. approved) behaviors might be lower, because many of these behaviors already have been forbidden by religions. Thus, we expect the stronger effect of social norms on behaviors in traditional cultures to be especially pertinent for socially disapproved (vs. approved) behaviors (H13).
With respect to the survival versus self-expression cultural dimension, engaging in socially disapproved behaviors in survival cultures is more likely to jeopardize economic or physical security ([49]), diminishing perception of these behaviors as freedoms. In contrast, in societies leaning toward the self-expression pole, engaging in disapproved behaviors would be more tolerated ([52]). Thus, we expect that the effect of social norms on behaviors in survival cultures is especially strong for socially disapproved (vs. approved) behaviors (H14).
As to the effect of time, we expect that social media might enhance the effects of social norms by exposing consumers to more regular reinforcements pertaining to a wider range of socially approved and disapproved behaviors ([10]). Social media enables consumers to share content and feedback in real time, much of which remains available indefinitely and can be tracked by other parties ([45]). Exposure to norm violations in such settings triggers exhibitions of moral outrage, as manifested in the notion of a "cancel culture," whereby social media users shame and punish perpetrators of bad behaviors, signaling that such behaviors are not tolerated ([30]). These developments imply that, over time, engaging in socially disapproved behaviors is stigmatized more severely than not engaging in socially approved behaviors ([33]), which reduces expected freedom to perform socially disapproved behaviors. Thus, we expect a stronger, more positive impact of social norms on socially disapproved (vs. approved) behaviors over time (H15).
In addition to the aforementioned hypothesized interaction effects, given the importance of the fundamental distinction between social norms regulating socially approved versus disapproved behaviors, we also explore additional interactions. Specifically, we investigate the interaction between social approval/disapproval and the target behavior characteristics, communication factors, and consumer costs. When these interactions are significant, we return to them in the discussion and highlight theoretical and managerial insights.
Systematic differences in the methodologies used by studies may cause variation in the reported effects ([14]). We control for ( 1) type of data ([quasi]experiment vs. other), ( 2) whether a sample involves students or regular consumers, ( 3) whether participants were exposed to (vs. indicated their perceptions of) social norms, and ( 4) whether participants' behavior was self-reported (vs. observed). Finally, to account for publication bias, which arises when the effect sizes in published studies are not representative of the entire population of effect sizes ([17]), we control for ( 5) the association between the strength and the precision of the effect sizes ([100]). More details follow.
To identify relevant studies of the impact of social norms on consumer behaviors, we retrieved references from Google Scholar, Online Contents National, PsycINFO, and the Web of Science up to March 2019. We searched for keywords such as "norm," "social norms," and "social pressure" (for the full list of keywords, see Web Appendix A). We also checked the websites of the Social Science Research Network, the National Social Norms Resource Center, and Higher Education Center for Alcohol and Other Drug Abuse and Violence Prevention for relevant studies. We posted requests for unpublished manuscripts and working papers on the online academic platform ELMAR. Finally, we examined all cross-references from applicable documents. The procedure resulted in articles from five research domains: psychology (35.2%), health (34.4%), marketing (10.4%), food and nutrition (10.8%), and the environment (9.2%).
Our dependent variable is the strength of the relationship between social norms and consumer behavior in the studies, which constitutes their observed effect sizes. We selected Pearson's product-moment correlation coefficient to measure effect sizes, because most studies operationalize both social norms and the target behavior as continuous variables. The consumer behaviors investigated in these eligible studies refer to the purchase, consumption, use, or disposal of products, services, material objects, or consumption experiences (e.g., buying organic products, subscribing to a gym, adopting mobile banking, donating). We exclude studies that focus on ( 1) aggregate entities (e.g., countries, societies) rather than individual consumers; ( 2) behaviors unrelated to consumption, such as social perceptions or interpersonal relations (e.g., stereotypes); ( 3) criminal behaviors, because the influence of the law would be confounded with the influence of social norms; and ( 4) consumers with impaired autonomy, such as workers making job-related decisions who must follow organizational policy, patients who rely extensively on others to make medical decisions ([77]), or people whose addictions limit their decision-making ability ([64]). Furthermore, to be included an eligible study must ( 1) examine actual behaviors, reported or observed (rather than intentions); ( 2) contain enough information to calculate the correlations between social norms and behaviors; and ( 3) support computations of the unconfounded effects of social norms. To illustrate ( 3), we excluded studies that collapsed the impacts of social norms and marketing promotion (e.g., [112]) or injunctive and descriptive social norms (e.g., [56]).
The final sample thus consists of 252 effect sizes extracted from 137 articles, comprising 177 studies over the period 1978–2019. Web Appendix B lists the articles, effect sizes, and moderator values. The sample sizes of the primary studies range from 28 to 44,108 (median = 269), so that they produce a total of 112,929 unique respondents from 22 countries. Three of the 252 effect sizes have studentized residuals that are greater than 2.57 ([107]); two of them (r = −.19, n = 353; r = .71, n = 451) are influential, in that they lie outside the prediction interval (i.e., range of plausible values for any individual effect size) and cannot be explained by small sample sizes ([17]). We remove them from the subsequent analyses, leaving 250 effect sizes from 136 articles, based on 112,478 unique respondents. (As we detail in Web Appendix C, the primary studies provide an explanation for the extreme values of these outliers; the results are robust for including them in the analysis.)
Two independent coders (blind to the hypotheses) coded the moderators and cataloged the technical information (e.g., sample size). The intercoder agreement was 94.8%, and any disagreements were resolved through discussion.
We retrieved zero-order correlations, measuring the association between social norms and the target behavior, from the studies' correlation matrices or else converted statistics (e.g., F-value, t-value, p-value, χ2) into r (see [17]; [67]). If partial correlations were available, we also retrieved them from the studies. (The results are robust whether we use partial or zero-order correlations as measures of effect sizes.) We transformed the correlations into Fisher's z-scores ([17]) to satisfy the assumptions of normal distributions and known sampling variance of the effect sizes to estimate the model (for details, see the "Model" section).[ 7] In turn, we estimate the meta-analytic regression model with Fisher's z-scores as the dependent variable. We obtain the mean effect sizes, confidence intervals, predicted values, and plots by back-transforming Fisher's z-scores into correlation coefficients to facilitate interpretation (for details, see Web Appendix D). For robustness, we perform the analyses also by using the correlation coefficients. The effect sizes are coded such that a positive sign indicates a positive change in behavior (i.e., increase in socially approved or decrease in disapproved behaviors).[ 8]
Table 1 shows the coding scheme for all the moderators. We mean-centered all continuous moderators and all dummy variables involved in interactions ([89]). We retrieve scores for the cultural dimensions from the World Values Survey ([53]) for each effect size, using the country and year of publication of each study. For the time variable, the code reflects the year of publication.[ 9] The precision of the effect size estimate is measured as the inverse of its standard deviation ([100]). If a publication bias is present, retrieved small sample studies are more likely to yield stronger effect sizes than those that are not retrieved, which implies a negative relationship between precision and effect size. By controlling for precision, the effects can be estimated more accurately ([35]).
Graph
Table 1. Coding Scheme for the Moderators of Social Norms–Behavior Effects.
| Variable | Code |
|---|
| Target Behavior Characteristics | |
| Social approval | Dummy = 1 if the behavior is socially approved (i.e., discussed positively by the authors) and 0 if the behavior is socially disapproved (i.e., noted as problematic by the authors). Mean-centered. |
| Existing | Dummy = 1 if the behavior exists (i.e., consumers already engage in it at least sometimes) and 0 if the behavior is new (i.e., consumers have not adopted the behavior yet). |
| Hedonic | Dummy = 1 if the behavior is hedonic (i.e., driven by pleasure-related goals) and 0 if the behavior is utilitarian (i.e., driven by functionality-related goals). |
| Benefits to other people | Dummy = 1 if the behavior brings about social benefits and 0 otherwise. |
| Public behavior | Dummy = 1 if the behavior is public (i.e., is visible to others) and 0 if the behavior is private (i.e., is invisible to others). |
| Communication Factors | |
| Social norm formulation | Dummy = 1 if the social norm is descriptive (i.e., describes behaviors of others) and 0 if the social norm is injunctive (i.e., suggests what should be done). |
| Organization-benefiting | Dummy = 1 if the social norm benefits a specific organization and 0 otherwise. |
| Close group | Dummy = 1 if the norm refers to a person close to the individual and 0 otherwise. |
| Authority figure | Dummy = 1 if the norm refers to a person in a position of authority and 0 otherwise. |
| Explicit sanctions | Dummy = 1 if the negative consequences of not abiding by the norms are made explicit and 0 otherwise. |
| Explicit rewards | Dummy = 1 if the positive consequences of abiding by the norm are made explicit and 0 otherwise. |
| Consumer Costs | |
| Effort | Dummy = 1 if complying with the social norm entails much physical or mental effort and 0 if compliance entails little effort. |
| Monetary costs | Dummy = 1 if complying with the social norm entails additional monetary costs and 0 otherwise. |
| Temporal costs | Dummy = 1 if complying with the social norm entails a long-term investment and 0 if it entails a temporary investment. |
| Environmental Factors | |
| Traditional–Secular-rational | Continuous: scores for the Inglehart dimension in the year of publication minus 2 and country of data collection. Mean-centered. |
| Survival–Self-expression | Continuous: scores for the Inglehart dimension in the year of publication minus 2 and country of data collection. Mean-centered. |
| Time | Continuous: year of publication of the paper from which the effect sizes are extracted minus 2. Mean-centered. |
| Moral freedom | Continuous: World Index of Moral Freedom (i.e., extent to which individuals make their own moral choices rather than being influenced by state intervention; Álvarez, Kotera, and Pina 2020). Mean-centered. |
| Methodological Controls | |
| Type of data | Dummy = 1 if the study is an experiment or a quasiexperiment and 0 otherwise. |
| Sample | Dummy = 1 when a student sample was used and 0 otherwise. |
| Effect size precision | Continuous: inverse of the standard error of the effect sizes (Fisher's z-transformed). Mean-centered. |
| Behavior operationalization | Dummy = 1 when participants self-report the behavior and 0 when the behavior is observed. |
| Social norm operationalization | Dummy = 1 when participants are exposed to the social norms and 0 when the social norms are perceived. |
1 Notes: We control for the operationalizations of behaviors and social norms in a robustness check (see Web Appendix G).
To test the conceptual framework in Figure 1, the model should account for the structure of the data, because the effect sizes are nested within samples that are nested within articles, which could lead to correlated errors. We specify a mixed-effects meta-regression model using a multilevel parameterization ([105]) in which ( 1) observed effect sizes are assumed to be a normally distributed random sample from the population of true effect sizes; and ( 2) the variance distribution of true effect sizes can be explained by random effects at the effect size, sample, and article levels, to account for data nesting, and by the fixed effects of the moderators. Thus, the effect size i extracted from sample j in article u is modeled as follows:
Effect Sizeiju = β0 + β1 Social approvaliju + β2 Existing behavioriju + β3 Hedoniciju + β4 Benefit peopleiju + β5 Public behavioriju + β6 Norm formulationiju + β7 Organization-benefitingiju + β8 Close groupiju + β9 Authority figureiju + β10 Explicit sanctionsju + β11 Explicit rewardsiju + β12 Effortiju + β13 Monetary costsiju + b14 Temporal costsiju + β15 Traditional–secular-rationalu + β16 Survival–self-expressionu + β17 Timeu + β18 Moral freedomu + β19 Social approvaliju × Traditional–secular-rationalu + β20 Social approvaliju × Survival–self-expressionu + β21 Social approvaliju × Timeu + β22 Type datau + β23 Sampleju + β24 Effect size precisioniju + δu + εi + γju + φiju, where δu N(0, ) is a random effect that reflects the variance among articles, εi ∼ N(0, ) is the sampling variance of the observed effect sizes, γju N(0, ) is a random effect estimating the variance across samples nested within articles, φiju N(0, ) is a random effect that indicates the variance among effect sizes nested within samples and within articles, β0 is the intercept, and β1–24 are the parameter estimates for the moderators defined in Table 1. We perform all the analyses with the Metafor package for R ([107]).
We first present the grand mean effect size and the distribution of individual effect sizes. Next, we present the results of the meta-regressions testing for the moderators. To provide evidence without the multilevel specification, Model 1 does not account for the nested data structure. The hypothesized moderators (H1–H15) are then tested with Model 2, which accounts for the nested structure. We note that the findings of Model 1 versus Model 2 are very similar (see Table 3), despite their different approaches to nesting. Drawing on Model 2, Table 3 also presents predicted effect sizes for each level of the categorical moderators with all the other moderators set at their sample average ([15]) as well as simple mean correlations. The predicted values and simple mean correlations do not differ substantially, suggesting a good balance of moderator conditions across studies ([57]). Model 3 adopts an iterative approach to identify additional significant interactions between socially approved versus disapproved behaviors and target behavior characteristics, consumer costs, and communication factors, and we discuss its results where relevant.
Overall, social norms have a positive, small to medium impact on behaviors (i.e., the grand mean effect size is positive and significant, with = .254 and a 95% confidence interval [CI95%] ranging from.232 to.277; [28]). The Q-statistic, which represents the total weighted deviation of each individual effect size from the mean, is significant (Q = 4,360, p < .001). Most observed effect size variance thus is systematic rather than due to sampling error and can be explained by moderators ([17]). Other heterogeneity indicators (Tau2 and I2) lead to the same conclusion (see Web Appendix E).
The distribution of individual effect sizes, shown in Figure 2, reveals that they range from r = −.22 to r = .63 (M = .249, Mdn = .240, SD = .152), and 67% of them fall within a.10–.40 interval. Multicollinearity is not a concern. The largest bivariate correlation, r = .54, is between monetary costs and socially approved behaviors. The variance inflation factors are 3.73 or less for all variables. Table 2 provides the correlation matrix and descriptive statistics.
Graph: Figure 2. Social norm–behavior effect size frequency distribution (k = 250).
Graph
Table 2. Bivariate Correlations and Descriptive Statistics for the Social Norm–Behavior Effect and Moderators.
| Mean | SD | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 |
|---|
| 1. Effect size (r) | .25 | .152 | 1 | | | | | | | | | | | | | | | | | | | | | |
| 2. Social approval | .79 | | −.121 | 1 | | | | | | | | | | | | | | | | | | | | |
| 3. Existing | .26 | | .004 | .032 | 1 | | | | | | | | | | | | | | | | | | | |
| 4. Hedonic | .32 | | .141 | −.525 | .068 | 1 | | | | | | | | | | | | | | | | | | |
| 5. Benefits to people | .32 | | .095 | −.237 | −.074 | .013 | 1 | | | | | | | | | | | | | | | | | |
| 6. Public behavior | .52 | | .066 | .050 | .023 | .127 | −.105 | 1 | | | | | | | | | | | | | | | | |
| 7. Norm formulation | .30 | | .287 | −.152 | −.056 | .150 | .118 | .035 | 1 | | | | | | | | | | | | | | | |
| 8. Org. benefiting | .52 | | .319 | −.028 | −.087 | .161 | −.208 | .038 | .140 | 1 | | | | | | | | | | | | | | |
| 9. Close group | .66 | | .182 | −.207 | .005 | .112 | .125 | −.030 | .009 | −.014 | 1 | | | | | | | | | | | | | |
| 10. Authority figure | .10 | | −.120 | .048 | −.040 | −.121 | .078 | −.092 | −.137 | −.145 | −.004 | 1 | | | | | | | | | | | | |
| 11. Explicit sanctions | .15 | | .014 | −.038 | .155 | .078 | −.127 | .029 | −.057 | .084 | .141 | −.018 | 1 | | | | | | | | | | | |
| 12. Explicit rewards | .41 | | −.078 | .106 | .064 | −.093 | −.162 | −.005 | −.076 | .085 | .109 | .116 | .101 | 1 | | | | | | | | | | |
| 13. Effort | .30 | | −.215 | −.019 | −.150 | −.155 | −.169 | −.026 | −.129 | −.218 | .034 | .003 | .000 | .092 | 1 | | | | | | | | | |
| 14. Monetary costs | .57 | | −.005 | .540 | −.007 | −.429 | −.490 | −.006 | −.175 | .107 | −.058 | −.050 | .080 | .201 | .185 | 1 | | | | | | | | |
| 15. Temporal costs | .69 | | −.038 | −.159 | −.039 | .073 | −.434 | .027 | .008 | .182 | .027 | −.138 | .005 | .135 | .389 | .150 | 1 | | | | | | | |
| 16. Trad.–Sec-rational | .12 | .406 | .090 | .156 | −.193 | −.110 | −.074 | .007 | .104 | .067 | −.062 | .022 | .135 | −.111 | .129 | .109 | .049 | 1 | | | | | | |
| 17. Survival–Self-exp. | .80 | .330 | −.054 | .202 | .034 | −.200 | −.035 | .034 | −.082 | .001 | −.106 | −.058 | −.016 | −.066 | .160 | .051 | −.040 | .134 | 1 | | | | | |
| 18. Time (year) | 2004 | 9.219 | .164 | .026 | .087 | .009 | −.101 | .266 | .324 | .049 | −.231 | .071 | −.052 | −.073 | .025 | .106 | .133 | .207 | −.018 | 1 | | | | |
| 19. Moral freedom | 74.91 | 8.71 | −.096 | .108 | .130 | −.072 | −.016 | −.041 | .007 | .030 | −.085 | .036 | .024 | .018 | .110 | −.012 | −.004 | .154 | .436 | .059 | 1 | | | |
| 20. Type of data | .16 | | −.065 | −.194 | −.249 | −.022 | .333 | .122 | −.133 | −.060 | .094 | −.126 | −.165 | −.009 | −.270 | −.211 | −.086 | −.240 | −.027 | −.195 | .006 | 1 | | |
| 21. Student sample | .41 | | .034 | −.346 | .170 | .285 | −.057 | .167 | .048 | .098 | −.016 | .046 | −.048 | .053 | −.137 | .173 | .081 | −.253 | −.070 | .084 | −.041 | −.044 | 1 | |
| 22. Effect size precision | 20.47 | 19.77 | .144 | .029 | −.183 | .117 | −.023 | .155 | .168 | −.033 | −.103 | −.031 | −.011 | −.063 | −.091 | −.101 | .055 | .130 | −.009 | .169 | .076 | −.166 | .229 | 1 |
2 Notes: Boldfaced correlations are significant at α = .05. For the dummy variables, only the mean is provided as a descriptive statistic.
Contrary to H1, social norms are not more effective for disapproved (vs. approved) behavior (b = −.002, p = .957). Although the insignificant main effect suggests that social norms are equally effective for approved and disapproved behaviors overall, this variable is involved in several interactions with other moderators, as we discuss next.
The effect of existing (vs. new) behavior is not significant (b = .017, p = .514); thus, H2 is not supported. However, according to Model 3, this variable interacts with the social approval of behavior at marginal significance (b = .138, p = .072). Figure 3, Panel A, shows that when targeting socially approved behaviors, social norms tend to be more effective in encouraging existing (predicted = .253) versus new (predicted = .217) behaviors. For socially disapproved behaviors, the opposite pattern emerges, as social norms tend to be more effective for discouraging new (predicted = .298) versus existing (predicted = .207) behaviors.
Graph: Figure 3. Social norm–behavior effect sizes for socially approved and disapproved behaviors as a function of focal moderators.
The impact of social norms does not differ across hedonic and utilitarian behaviors (b = .016, p = .599). This finding is not surprising given the two opposing forces (desire and guilt) driving hedonic behaviors.
Social norms are more effective when behaviors benefit others (predicted = .289) than when they do not (predicted = .229, b = .064, p = .035), in support of H3.
The effect of public behavior is not significant (b = .013, p = .500), suggesting that social norms are equally effective for public and private behaviors.
In support of H4, descriptive norms have stronger effects on behavior (predicted = .305) than injunctive norms (predicted = .223, b = .088, p < .001). Further, the interaction between norm formulation and socially approved behaviors in Model 3 is also significant (b = −.110, p = .018). Figure 3, Panel B, shows that descriptive (vs. injunctive) norms are more effective when targeting disapproved behaviors (predicted descriptive = .337 vs. predicted injunctive = .183, Δ = .154) than when targeting approved behaviors (predicted descriptive = .281 vs. predicted injunctive = .228, Δ = .053). Thus, descriptive social norms that describe how the majority behaves are especially effective in curbing disapproved behaviors.
In support of H5, social norms are more effective when they benefit an organization (predicted = .279) than otherwise (predicted = .214, b = .069, p = .004).
In support of H6, social norms are more effective if they refer to a close group member (predicted = .266) versus an abstract group (predicted = .213, b = .057, p = .008).
The effectiveness of social norms is not impeded by references to an authority figure (b = −.012, p = .729), in contrast to H7.
The effect of social norms does not depend on the presence of explicit sanctions (b = −.010, p = .767), disconfirming H8. However, social norms with explicit mentions of rewards (predicted = .208) are marginally less effective than those where rewards are not mentioned (predicted = .255, b = −.050, p = .078), in line with H9.
The amount of effort required to comply with social norms does not have a significant effect on compliance (b = −.031, p = .255). This finding is consistent with the idea that there are two opposing forces behind this effect that counteract each other.
Social norms exert stronger effects on behavior when compliance entails monetary costs (predicted = .276) than when it does not (predicted =.211, b = .069, p = .017). This finding is consistent with the idea that barriers such as monetary costs can make the behavior more desirable for consumers (e.g., via status signaling; [27]).
Social norms seem to be equally effective for behaviors requiring a long-term or a temporary investment, as temporal costs are nonsignificant (b = −.011, p = .706).
The traditional–secular-rational cultural dimension does not have a main effect on the effectiveness of social norm (b = −.017, p = .643), which fails to support H10.[10] However, we find some support for H13 because the interaction with the social approval of behaviors is positive and marginally significant (b = .211, p = .104). Figure 3, Panel C, shows that the impact of social norms on socially disapproved behaviors is somewhat weaker in more secular-rational cultures, whereas their impact on socially approved behaviors remains stable across the traditional–secular-rational dimension.
The effect of this cultural dimension is negative, as we expected, but it is not significant (b = −.027, p = .448), which does not support H11. However, in support of H14, its interaction with the social approval of behaviors is positive and significant (b = .180, p = .033). Thus, consistent with our expectations, social norms are less effective for socially disapproved behaviors in cultures closer to the self-expression pole, whereas their effectiveness for socially approved behaviors is stable across the survival–self-expression dimension (Figure 3, Panel D).
The effect of time is not significant (b = .001, p = .956), but its interaction with the social approval of behaviors is negative and significant (b = −.006, p = .036), in support of H15. Social norms have become more effective over time at curbing socially disapproved behaviors, while their influence on socially approved behaviors remains stable (Figure 3, Panel E).
The national level of moral freedom lowers behavioral compliance at a marginal level of significance (b = −.002, p = .094), in support of H12.
The studies in our meta-analysis yield the same results regardless of the type of data (b = −.045, p = .189) and whether they rely on student samples (b = −.001, p = .986). We obtain similar results if we control for whether the studies operationalize social norms using exposure or perception (note that we exclude this control variable in the main model because of multicollinearity but address it in supplementary analyses). Precision has a positive, marginally significant effect (b = .001, p = .082); therefore, effect sizes greater than the mean might be missing, which suggests that publication bias is not an issue.
To confirm that our results are robust, we first perform a series of diagnostic tests (Web Appendices E and F) to rule out publication bias; they reconfirm the positive influence of effect size precision in the meta-regression model. Next, we performed analyses based on 16 alternative model specifications. In Tables A1 and A2 in Web Appendix G, we present the results when we adopt alternative methodological choices: ( 1) including theoretical moderators only; ( 2) adding two demographic controls—the primary study participants' mean age and the percentage of men; ( 3) accounting for effect sizes coming from marketing journals; ( 4) estimating the meta-regression parameters with t-tests instead of z-values; ( 5) estimating the model including the two outliers (k = 252); ( 6) using the raw effect sizes—r and the variance of r—rather than Fisher's z transforms; ( 7) relying on partial instead of the zero-order correlation, if both measures could be retrieved from the primary study; and ( 8) controlling for the operationalization of the behavior (self-reported vs. observed) and for the operationalization of social norms (exposed vs. perceived) instead of controlling for the type of data. Across these methodological choices, the results remain consistent with the findings of the main models (Table 3).
Graph
Table 3. Meta-Regression Results.
| Model 1: | Model 2: | Model 3: | Predicted Values | Simple Mean Correlations |
|---|
| Predictors | b (SE) | b (SE) | b (SE) | [CI95%] | [CI95%] |
|---|
| Intercept | .097 (.047)* | .104 (.049)* | .127 (.048)** | — | — |
| Social disapproval (k = 53) | — | — | — | .250 [.181,.317] | .299 [.248,.348] |
| Social approval (k = 197) | .009 (.040) | −.002 (.042) | .015 (.044) | .248 [.225,.271] | .250 [.230,.269] |
| New behavior (k = 84) | — | — | — | .236 [.196,.276] | .263 [.221,.303] |
| Existing behavior (k = 186) | .016 (.025) | .017 (.026) | .010 (.027) | .253 [.228,.277] | .259 [.237,.280] |
| Utilitarian behavior (k = 170) | — | — | — | .244 [.216,.270] | .244 [.222,.265] |
| Hedonic behavior (k = 80) | −.012 (.028) | .016 (.030) | .001 (.031) | .258 [.216,.300] | .295 [.257,.331] |
| Benefits people: no (k = 171) | — | — | — | .229 [.200,.258] | .251 [.228,.274] |
| Benefits people: yes (k = 79) | .064 (.029)* | .064 (.030)* | .066 (.032)* | .289 [.249,.328] | .279 [.244,.314] |
| Private behavior (k = 120) | — | — | — | .242 [.214,.270] | .250 [.224,.275] |
| Public behavior (k = 130) | .016 (.019) | .013 (.019) | .015 (.019) | .254 [.228,.280] | .269 [.241,.297] |
| Norm formulation: injunctive (k = 175) | — | — | — | .223 [.200,.247] | .229 [.208,.251] |
| Norm formulation: descriptive (k = 75) | .086 (.022)*** | .088 (.021)*** | .080 (.021)*** | .305 [.272,.338] | .329 [.293,.363] |
| Organization-benefiting: no (k = 120) | — | — | — | .214 [.183,.245] | .208 [.181,.235] |
| Organization-benefiting: yes (k = 130) | .077 (.022)*** | .069 (.024)** | .078 (.025)** | .279 [.250,.308] | .304 [.280.329] |
| Close group: no (k = 85) | — | — | — | .213 [.179,.246] | .215 [.185,.244] |
| Close group: yes (k = 165) | .066 (.021)** | .057 (.022)** | .051 (.022)* | .266 [.242,.290] | .283 [.259,.306] |
| Authority figure: no (k = 224) | — | — | — | .249 [.228,.271] | .266 [.246,.286] |
| Authority figure: yes (k = 26) | −.001 (.033) | −.012 (.034) | −.004 (.034) | .238 [.179,.297] | .202 [.149,.253] |
| Explicit sanctions: no (k = 227) | — | — | — | .249 [.228,.270] | .259 [.239,.279] |
| Explicit sanctions: yes (k = 23) | −.019 (.032) | −.010 (.035) | .001 (.035) | .239 [.177,.300] | .266 [.204,.325] |
| Explicit rewards: no (k = 213) | — | — | — | .255 [.234,.277] | .265 [.244,.286] |
| Explicit rewards: yes (k = 37) | −.056 (.027)* | −.050 (.028)† | −.041 (.029) | .208 [.158,.257] | .229 [.184,.272] |
| Effort: small (k = 174) | — | — | — | .257 [.231,.282] | .281 [.259,.303] |
| Effort: large (k = 76) | −.034 (.026) | −.031 (.027) | −.044 (.028) | .228 [.187,.268] | .210 [.174,.246] |
| Monetary costs: no (k = 107) | — | — | — | .211 [.173,.249] | .262 [.230,.294] |
| Monetary costs: yes (k = 143) | .071 (.027)** | .069 (.029)* | .072 (.029)* | .276 [.246,.305] | .259 [.235,.282] |
| Temporal costs: temporary (k = 78) | — | — | — | .256 [.211,.299] | .267 [.237,.297] |
| Temporal costs: long-term (k = 172) | −.012 (.028) | −.011 (.029) | −.010 (.031) | .245 [.220,.270] | .256 [.232,.281] |
| Traditional–Secular-rational | −.014 (.035) | −.017 (.038) | −.031 (.039) | — | — |
| Survival–Self-expression | −.026 (.033) | −.027 (.036) | −.021 (.037) | — | — |
| Time | −.001 (.001) | .001 (.001) | .001 (.001) | — | — |
| Moral freedom | −.002 (.001)† | −.002 (.001)† | −.003 (.001)* | — | — |
| Social approval × Traditional–Secular-rational | .175 (.121) | .211 (.130)† | .298 (.145)* | — | — |
| Social approval × Survival–Self-expression | .159 (.079)* | .180 (.084)* | .206 (.087)* | — | — |
| Social approval × Time | −.005 (.003)* | −.006 (.003)* | −.008 (.004)* | — | — |
| Social approval × Existing behavior | — | — | .138 (.077)† | — | — |
| Social approval × Norm formulation | — | — | −.110 (.046)* | — | — |
| Type of data: other (k = 210) | — | — | — | .255 [.233,.277] | .265 [.244,.286] |
| Type of data: (quasi)experiment (k = 40) | −.048 (.033) | −.045 (.034) | −.014 (.036) | .213 [.155,.270] | .229 [.182,.274] |
| Sample: nonstudent (k = 148) | — | — | — | .249 [.222,.275] | .254 [.230,.278] |
| Sample: student (k = 102) | −.001 (.023) | −.001 (.024) | .011 (.025) | .248 [.214,.282] | .269 [.237,.301] |
| Effect size precision | .001 (.0005)† | .001 (.0005)† | .001 (.0006)† | — | — |
| Pseudo R2 | 30% | 29% | 30% | | |
| Variances components | Tau2 = .016; = .001 | = .0001; = .001; = .004; = .012 | = .0006; = .001; = .005; = .011 | | |
- 3 *p < .05.
- 4 **p < .01.
- 5 ***p < .001.
- 6 †p < .10.
- 7 Notes: There is no regression coefficient for the reference category of the moderators. The pseudo R2 is the amount of between-effect size variance explained by the moderators. In Model 1, the nested structure of the data is not modeled and Tau2 represents the residual between-effect size variance. In Models 2 and 3, the nested structure of the data is accounted for and the variance components are the following: = variance across articles, = sampling error variance, = variance between samples within articles, and = variance between effect sizes within samples within articles. Predicted values are based on Model 2 parameter estimates, with the levels of all the other moderators set at the sample average. The slight discrepancy between the predicted values for the categories of the moderators (e.g., monetary costs: no = .211; yes = .276, Δ = .065) and the corresponding regression coefficient (b = .069 > .065) arises because the parameter estimates are based on Fisher's z-scores, whereas the predicted values are expressed as correlation coefficients. The simple mean correlations are obtained using a random-effect model with restricted maximum likelihood as a between-effect size variance estimator. Predicted values and simple mean correlations are not provided for the continuous moderators, because their effect is linear (see Figure 3).
In Table A3 (Web Appendix G), we also rule out the effects of four alternative moderators: ( 9) whether the target behavior entails environmental benefits, (10) whether the target behavior has social and physical consequences for the individual, and (11) whether the norm features a promotion or prevention frame. None of these alternative variables is significant.
Finally, Table A4 (Web Appendix G) presents the results when we include additional country controls based on (12) country-level gross domestic product and population density, as well as (13) Hofstede's (vs. Inglehart's) cultural dimensions. The results again remain similar to those from the main models. Further, to account for a potential cultural invariance bias, we control for (14) the language spoken in the country of the data collection and (15) whether the study was conducted in an Asian country; (16) we also model the nesting of articles within countries. The stable regression coefficients suggest that cultural invariance does not affect the results. Across all 16 alternative models, the magnitudes, signs, and significance of the parameter estimates are consistent and aligned with our main findings, which strengthens our conclusions and affirms the robustness of the effects obtained in the main model.
By meta-analyzing the extant empirical evidence, our research provides new evidence for the effects of social norms on consumer behavior. On average, social norms significantly influence behavior and their effect ( = .254) is small to medium in size ([28]). Importantly, several moderators can explain substantial variation among these effects. Table 4 summarizes our findings. We next detail the theoretical and practical implications of our study.
Graph
Table 4. Hypotheses and Empirical Questions.
| Moderator and Predictions (Hypotheses) or Competing Explanations (Empirical Questions) | Reactance Explanation | Evidence |
|---|
| Target Behavior Characteristics | | |
| Social approval (vs. disapproval) (H1): Social norms for socially disapproved behaviors are more effective than those pertaining to approved behaviors. | Expectations of freedom are lower for socially disapproved vs. socially approved behaviors | × |
| Existing (vs. new) (H2): Social norms are more effective when targeting existing (vs. new) behaviors. | Freedom threat is lower for existing than for new behaviors | × |
| Hedonic (vs. utilitarian) (Empirical question): Greater desire or more guilt for hedonic benefits may either increase or decrease freedom threat from social norms | N.A. | n.s. |
| Benefits to other people (H3): The presence of benefits to other people of a behavior enhances the effect of social norms on that behavior. | Expectations of freedom are lower for behaviors that are beneficial to others | ✓ |
| Public behavior (Empirical question): Freedom threat from social norms may be lower for private behaviors due to a lack of scrutiny from others, or it may be lower for public behaviors due to concerns about perceptions from others. | N.A. | n.s. |
| Communication Factors |
| Norm formulation (H4): Descriptive social norms have a stronger impact on behaviors than injunctive norms. | Descriptive norms imply less freedom threat than injunctive norms | ✓ |
| Organization-benefiting (H5): Organization-benefiting social norms are more effective than social norms that do not mention organizations. | Organization-benefiting social norms imply less freedom threat than social norms that do not benefit organizations | ✓ |
| Close group members (H6): Social norms referring to a close group member are more effective than social norms referring to abstract or distant groups. | Close group members imply less freedom threat than abstract or distant groups | ✓ |
| Authority figures (H7): Social norms referring to authority figures are less effective than social norms that do not refer to them. | Authority figures imply more freedom threat | × |
| Explicit sanctions (H8): Social norms that specify potential sanctions are less effective than social norms that do not make these consequences explicit. | Explicit sanctions and rewards imply more freedom threat | × |
| Explicit rewards (H9): Social norms that specify potential rewards are less effective than social norms that do not make these consequences explicit. | ✓ |
| Consumer Costs | |
| Effort, monetary costs, and temporal costs (Empirical questions): The presence of costs barriers arouses reactance, which either makes the behaviors more attractive, or dissuades consumers from attempting the behaviors. | Costs create barriers to free choice and increase reactance | Monetary costs lift social norm effect; n.s. for effort and temporal costs |
| Environmental Factors |
| Traditional–Secular-rational (H10): The effect of social norms on behavior is stronger in cultures closer to the traditional (vs. secular-rational) pole. | Expectations of freedom are lower in cultures toward the traditional pole and toward the survival pole | × |
| Survival–Self-expression (H11): The effect of social norms on behavior is stronger in cultures near the survival pole versus the self-expression pole. | × |
| Time (Empirical question): Evolutionary forces driving social norms effectiveness should be stable over time but fluctuations of conformism over time is observed. | N.A. | n.s. |
| Moral freedom (H12): The effect of social norms on behavior is stronger in countries lacking moral freedom. | Expectations of freedom are lower in low moral freedom countries | ✓ |
| Interactions |
| Traditional–Secular-rational × Approved behavior (H13): The stronger effect of social norms on behaviors in traditional cultures is driven by socially disapproved (vs. socially approved) behaviors. | Expectations of freedom for socially disapproved behaviors are especially low in traditional and survival cultures | ✓ |
| Survival–Self-expression × Approved behavior (H14): The effect of social norms on behaviors in survival cultures is especially strong for socially disapproved (vs. approved) behaviors. | ✓ |
| Time × Approved behavior (H15): There is a more positive impact of social norms for socially disapproved than for socially approved behaviors over time. | Expectations of freedom for socially disapproved behaviors decrease over time | ✓ |
| Existing (vs. new) × Approved behavior (Exploratory interaction) | N.A. | Social norms are effective for existing, approved behavior |
| Norm formulation × Approved behavior (Exploratory interaction) | N.A. | Descriptive norms are effective for disapproved behavior |
8 Notes: N.A. = not applicable; n.s. = not significant; x = not supported.
Our results go beyond previous meta-analyses by uncovering previously overlooked boundary conditions of the effects of social norms on consumer behavior using reactance theory as our theoretical lens. While our meta-analysis provides important insights about whether and how social norms can influence behavior, any meta-analysis is limited to the factors included in available primary studies. Those gaps also open avenues for future research, as discussed next.
We reveal that the effects of social norms on behavior differ systematically for behaviors that are socially approved versus disapproved in the presence of certain environmental factors. We find that the effects of social norms on socially disapproved behaviors have increased over time and are particularly strong in survival (vs. secular-rational) and traditional (vs. self-expression) cultures. In contrast, the effects of social norms on socially approved behaviors are more stable across cultures and time.
The critical difference in social norm effectiveness for regulating socially approved versus disapproved behaviors establishes the need to investigate drivers of this difference. In the process of adopting socially approved behaviors, consumers might glean benefits beyond the direct consequences of complying with norms. For example, does compliance promote positive emotions, due to an enhanced sense of belonging, acceptance, or well-being, independent of the benefits of adopting the behavior? And do consumers suffer negative emotions (e.g., guilt) when they engage in socially disapproved behaviors, and do social norms reinforce these emotions? Another issue is how legal mandates affect behaviors, both when they contradict social norms of disapproved behaviors (e.g., using cell phones while driving) and when they reinforce approved behaviors (e.g., driving below the speed limit).
We shed new light on several unexplored moderators of the effects of social norms on consumer behavior and thereby contribute to the literature on social influence ([24]; [109]; [110]). For example, social norms have stronger effects on behaviors that benefit others, but they are weaker for already existing socially disapproved behaviors. Social norms are equally effective for private and public behaviors, for hedonic and utilitarian behaviors, for behaviors requiring much versus little effort, and for behaviors requiring long-term versus temporary commitment. Importantly, monetary costs do not deter consumers; on the contrary, they make them more likely to comply with social norms regulating costly behaviors, in line with the reactance theory explanation that barriers enhance behaviors' desirability.
The intriguing finding that private behaviors are just as likely to be influenced by social norms as public ones suggests new research avenues. Perhaps consumers overestimate the extent to which they are monitored by others because of the spotlight effect ([37]). If so, social norms could be effective in nonsocial settings, where they traditionally have been perceived as less influential. Further, finding that social norms are equally effective for hedonic and utilitarian behaviors highlights the need to clarify the roles of desire and guilt as underlying processes. Teasing out these effects could also help enhance communication strategies (e.g., downplaying or emphasizing desire and guilt). Yet, the lack of effect of some behavioral characteristics might also be due to selection; for example, researchers may be inclined study situations in which they expect social norms to matter. Thus, testing the effects of social norms in situations where they a priori seem less relevant is insightful. Studying the effectiveness of social norms when behavioral autonomy is impaired (e.g., employees making decisions on behalf of firms) is also intriguing. Finally, research could address behaviors consumers adopt to distinguish themselves from the group (e.g., to enhance authenticity; [82]) and the leadership behaviors they display to encourage others to go against current social norms (e.g., boycotting, brand sabotaging; [55]).
We contribute to marketing communication research by identifying communication strategies that enhance the effectiveness of social norms in several ways. First, several commonly used social norm communication factors appear to be ineffective, such as referring to authority figures or specifying explicit sanctions; the same is true for rewards, which even seem to hinder social norms' effectiveness. Second, formulating norms as descriptive (vs. injunctive), organization-benefiting, and/or referring to close others enhance the effect of social norms on behaviors. Further, our exploratory results suggest that injunctive norms are especially weak when targeting disapproved behaviors, which is consistent with reactance theory. To discourage socially disapproved behaviors, injunctive norms tend to be proscriptive ("you should not") rather than prescriptive ("you should"). Proscription represents a greater threat to freedom than prescription, so it prompts more reactance to injunctive norms ([11]). Instead, socially disapproved behaviors can be better curbed by descriptive norms.
The finding that organization-benefiting social norms enhance compliance contributes to the emerging stream of research that examines how situational factors influence the effectiveness of social norms ([ 1]; [38]). While which type of organization benefits should matter as well, on a broader level, our results suggest that organization-benefiting social norms, because they are situation-specific, enhance compliance.
More research is needed for boundary conditions for the effectiveness of different communication strategies. For example, there are multiple avenues for future research on the impacts of sanctions and rewards. First, reward size and reward type (e.g., material vs. nonmaterial; [73]) might be important. Second, rewards or sanctions may not exert effects past a ceiling level of the impact of social norms when many consumers are "uninfluenceable." Third, research should identify ways to account for consumer needs and boost perceptions of the benefits of following and the costs of not following social norms without triggering reactance.
Boundary conditions under which injunctive norms are more effective than descriptive ones (e.g., the level of ambiguity, the source of the norm: "Why fight city hall?") should be addressed. Further, because norms are group-specific, research could explore how group exclusivity should be communicated. Social norms linked to homogeneous exclusive groups ("Harvard students recycle") might evoke less reactance because they distinguish group members from outsiders while also satisfying the need to belong ([13]). The effect of group exclusivity on behavior might be stronger if the communication contrasts ingroup versus outgroup norms ("Harvard students, unlike [vs. similar to] MIT students, recycle"). Relatedly, communicating the size of the group that shares social norms could also enhance social norms' influence ([46]).
Finally, most of the primary studies in our meta-analysis predated social media, but our results suggest that social media may have disrupted the influence of social norms on behavior. Online interactions enable consumers to develop friendships with people they have never met in real life, so further research might investigate how social norms evolve on social media. For example, how does the immediate, often permanent feedback available through social media shape the effects of social norms on online behaviors? Consumers might not just comply with social norms but also try to become agents of influence by spreading the norm further on social media. Future research should investigate which factors facilitate such efforts ([68]), and the role of social norms in interpersonal relationships in general. Finally, future research should address the interaction between social norms and marketing-mix instruments, particularly promotion.
Our results inform discussions on reactance ([18]; [92]) and the role of intrinsic versus extrinsic behavioral motivations ([31]). The nonsignificant effect of public behavior implies that consumers follow social norms even if their behavior is not observable to others. Further, while effort may impede social norms' effectiveness, the monetary costs of behavior enhance it, suggesting that consumer effort and temporal costs are a greater barrier to compliance than monetary costs. These results, together with the finding that explicit sanctions and rewards do not help, suggest that complying with social norms may be a more intrinsically motivated activity than previously believed.
Follow-up research should investigate other barriers that may incite reactance and ways to circumvent them. Noting our finding that monetary costs enhance social norm compliance, we call for research that specifies the boundary conditions of this effect. More research is needed to clarify what makes costly behavior attractive to consumers. For example, costly behaviors may help signal status ([42]).
Research should address the relative explanatory power of alternative mechanisms parallel to reactance that might influence the effect of social norms on behaviors. For example, the literature identifies other potential mechanisms, such as self-efficacy ([ 5]), sense of belonging ([ 6]), internalization ([97]), and justification of behaviors ([39]).
The findings offer insights for marketers and public policy makers by identifying effective (and some commonly used but ineffective) strategies for enhancing the impact of social norms on consumer behavior. In contrast to conventional wisdom ([ 7]), our results suggest that the influence of social norms can prompt private acceptance. Thus, marketers and policy makers can leverage social norms to encourage both private and public behaviors.
The content of the communication should feature descriptive rather than injunctive forms of social norms (i.e., describe what [most] people actually do rather than what they should do[11]). Further, we recommend that marketers should avoid specifying explicit sanctions and rewards associated with social norms. Instead, strategies that highlight benefits to others or consumer freedom (e.g., a communication with a postscript "The choice is yours"; [12]) may mitigate reactance and thus be more effective for inducing the target behavior.
Practitioners might worry about highlighting a specific organization when communicating about social norms, but our results suggest that referring to a specific firm, governmental body, or nongovernmental organization can make communications about social norms more influential. Social norms are also more powerful when they cite people who are perceived as close to the target consumers. Thus, practitioners should target social norm communication toward nano-influencers on social media, with their smaller, more engaged audiences. In contrast, our results indicate that references to authority figures when using social norms do not affect consumer behavior.
In communicating norms, marketers can acknowledge the monetary costs associated with the targeted behaviors. Although monetary costs are a financial barrier, they seem to also increase the desirability of the behavior, so social norms can be particularly effective for promoting costly behaviors like donations or buying (more expensive) organic food. Further, social norms are equally effective irrespective of required effort, and the time investment in complying.
The impact of social norms on socially disapproved behaviors varies significantly depending on the country of implementation, but it is more stable for socially approved behaviors. Social norms have stronger influences on socially approved than disapproved behaviors in secular-rational and self-expression cultures. These findings have important public health implications when group behavior is essential. To encourage mask wearing in most Western countries, for example, public officials should communicate that wearing a mask is a socially approved behavior that close others adopt. In most survival countries, the communications should highlight that not wearing a mask is socially disapproved.
To specify the net effect of culture at the country level, we estimate the impact of social norms in countries with different cultural profiles. We calculate the predicted effect sizes for socially disapproved and approved behaviors in Figure 4 for eight countries that represent different regions of the world. We use descriptive norms as a base category and country scores on Inglehart dimensions from the latest wave of the World Values Survey.
Graph: Figure 4. Predicted current effect sizes (Pearson's product moment correlations) of descriptive norms for socially approved and disapproved behaviors, by global regions.
In Scandinavian countries such as Denmark (eight effect sizes), which score high on secular-rational values (85th percentile) and high on self-expression values (98th percentile), the mean effect size for socially approved behaviors is = .365 (CI95% = [.282,.442]), whereas it is not significant for disapproved behaviors ( = .176, CI95% = [−.077,.407]). Campaigns using social norms thus may be effective for encouraging healthy eating, for example, but are likely not the best choice to curb excess drinking in such countries.
In Western European (e.g., France) and Commonwealth (e.g., Canada, United Kingdom) countries with medium to high scores on both dimensions (but lower than Scandinavia), social norms are equally effective regardless of the social approval of behavior. For example, in Australia (20 effect sizes), a country with average secular-rational values (55th percentile) and high self-expression values (96th percentile), the impact of social norms on approved behaviors ( = .344, CI95% = [.273,.411]) is the same as the impact on disapproved behaviors ( = .355, CI95% = [.249,.452], Wald-type test = −.175, p = .431).
In contrast, in the United States (104 effect sizes), which is more traditional (43rd percentile for secular-rational) and has high self-expression values (85th percentile), the effect of social norms on socially disapproved behaviors ( = .460, CI95% = [.374,.538]) is stronger than their impact on approved behaviors ( = .330, CI95% = [.262,.395], Wald-type test = −2.406, p = .008).
Interestingly, we find the same pattern in Southern Europe (e.g., Italy) and China, even though these areas represent relatively high secular-rational (65th percentile) and low self-expression values (28th percentile). In these regions, social norms' effectiveness is greater for disapproved behaviors ( = .530, CI95% = [.397,.642]) than for approved behaviors ( = .319, CI95% = [.213,.417], Wald-type test = −2.611, p = .005).
Finally, in countries with strong traditional and survival values, such as most African and Muslim-majority countries, social norms' impact on disapproved behaviors is much stronger than on approved behaviors. Consider Ethiopia (25th and 30th percentiles), where the mean effect size for disapproved behaviors ( = .660, CI95% = [.455,.799]) is much greater than that for approved behaviors ( = .297, CI95% = [.188,.398]). In these countries, social norms are especially effective for discouraging disapproved behaviors.
This extensive meta-analysis shows that social norms significantly impact behavior and uncovers novel contingencies of this effect. We hope the proposed research agenda, which reflects our comprehensive investigation of the extant literature, sparks additional research in the fascinating ways in which social norms shape (or do not shape) consumer behavior.
Footnotes 1 Don Lehmann
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no special funding for the research, authorship, and/or publication of this article.
4 Valentyna Melnyk https://orcid.org/0000-0003-4963-4584
5 Online supplement:https://doi.org/10.1177/00222429211029199
6 We also consider [50] dimensions for robustness in supplementary analyses.
7 As noted by Hedges and Piggott (2001), correlation coefficient effect sizes are often nonnormally distributed, and determining their variance requires knowing the underlying population correlation ρ. As a solution, we perform the analysis with Fisher's z-scores, which are normally distributed with variance determined directly by the sample size.
8 For all but two effect sizes, the study design was cross-sectional. In the two studies that observe behavior at several points in time, we include only effect sizes at t0 to ensure compatibility with the cross-sectional effects.
9 We relied on the date of publication because most of the primary studies do not report data collection dates; we applied the t − 2 rule for the related variables, such as culture and time.
Because the variables included in interaction terms are mean-centered, their simple effects in the model represent the average effect across socially disapproved and approved behaviors ([89]).
With the caveat that what most people do should be beneficial for society.
References Aarts Henk , Dijksterhuis Ap. (2003), " The Silence of the Library: Environment, Situational Norm, and Social Behavior ," Journal of Personality and Social Psychology , 84 (1), 18 – 28.
Albarracin Dolores , Johnson Blair T. , Fishbein Martin , Muellerleile Paige A.. (2001), " Theories of Reasoned Action and Planned Behavior as Models of Condom Use: A Meta-Analysis ," Psychological Bulletin , 127 (1), 142 – 61.
Álvarez Gloria , Kotera Yasuhiro , Pina Juan. (2020), World Index of Moral Freedom: WIMF 2020. Madrid : Foundation for the Advancement of Liberty.
Balliet Daniel , Mulder Laetitia B. , Van Lange Paul A.M.. (2011), " Reward, Punishment, and Cooperation: A Meta-Analysis ," Psychological Bulletin , 137 (4), 594 – 615.
Bandura Albert. (1982), " Self-Efficacy Mechanism in Human Agency ," American Psychologist , 37 (2), 122 – 47.
Baumeister Roy F. , Leary Mark R.. (1995), " The Need to Belong: Desire for Interpersonal Attachments as a Fundamental Human Motivation ," Psychological Bulletin , 117 (3), 497 – 529.
Bearden William O. , Etzel Michael J.. (1982), " Reference Group Influence on Product and Brand Purchase Decisions ," Journal of Consumer Research , 9 (2), 183 – 94.
Bellezza Silvia , Gino Francesca , Keinan Anat. (2013), " The Red Sneakers Effect: Inferring Status and Competence from Signals of Nonconformity ," Journal of Consumer Research , 41 (1), 35 – 54.
Bensley Lillian Southwick , Wu Rui. (1991), " The Role of Psychological Reactance in Drinking Following Alcohol Prevention Messages ," Journal of Applied Social Psychology , 21 (13), 1111 – 24.
Berger Jonah , Milkman Katherine L.. (2012), " What Makes Online Content Viral? " Journal of Marketing Research , 49 (2), 192 – 205.
Bergquist Magnus , Nilsson Andreas. (2016), " I Saw the Sign: Promoting Energy Conservation via Normative Prompts ," Journal of Environmental Psychology , 46 , 23 – 31.
Bessarabova Elena , Fink Edward L. , Turner Monique. (2013), " Reactance, Restoration, and Cognitive Structure: Comparative Statics ," Human Communication Research , 39 (3), 339 – 64.
Bhattacharya Chitrabhan B. , Rao Hayagreeva , Glynn Mary Ann. (1995), " Understanding the Bond of Identification: An Investigation of Its Correlates Among Art Museum Members ," Journal of Marketing , 59 (4), 46 – 57.
Bijmolt Tammo H.A. , Pieters Rik G.M.. (2001), " Meta-Analysis in Marketing when Studies Contain Multiple Measurements ," Marketing Letters , 12 (2), 157 – 69.
Bijmolt Tammo H.A. , van Heerde Harald J. , Pieters Rik G.M.. (2005), " New Empirical Generalizations on the Determinants of Price Elasticity ," Journal of Marketing Research , 42 (2), 141 – 56.
Bond Rod , Smith Peter B.. (1996), " Culture and Conformity: A Meta-Analysis of Studies Using Asch's (1952b, 1956) Line Judgment Task ," Psychological Bulletin , 119 (1), 111 – 37.
Borenstein Michael , Hedges Larry V. , Higgins Julian P.T. , Rothstein Hannah R.. (2009), Introduction to Meta-Analysis. Chichester, UK : John Wiley & Sons.
Brehm Jack W. (1966), A Theory of Psychological Reactance. New York : Academic Press.
Brehm Jack W. , Cole Ann H.. (1966), " Effect of a Favor Which Reduces Freedom ," Journal of Personality and Social Psychology , 3 (4), 420.
Brehm Sharon S. , Brehm Jack W.. (1981), Psychological Reactance: A Theory of Freedom and Control. New York : Academic Press.
Burchell Kevin , Rettie Ruth , Patel Kavita. (2017), " Marketing Social Norms: Social Marketing and the 'Social Norm Approach ,'" Journal of Consumer Behavior , 12 (1), 1 – 9.
Cabinet Office UK , Behavioural Insights Team (2012), "Applying Behavioural Insights to Reduce Fraud, Error and Debt," (accessed September 1, 2019), https://www.gov.uk/government/uploads/system/uploads/attachment%5fdata/file/60539/BIT%5fFraudErrorDebt%5faccessible.pdf.
Chitturi Ravindra , Raghunathan Rajagopal , Mahajan Vijay. (2008), " Delight by Design: The Role of Hedonic Versus Utilitarian Benefits ," Journal of Marketing , 72 (3), 48 – 63.
Cialdini Robert B. , Goldstein Noah J.. (2004), " Social Influence: Compliance and Conformity ," Annual Review of Psychology , 55 , 591 – 621.
Cialdini Robert B. , Reno Raymond R. , Kallgren Carl A.. (1990), " A Focus Theory of Normative Conduct: Recycling the Concept of Norms to Reduce Littering in Public Places ," Journal of Personality and Social Psychology , 58 (6), 1015 – 26.
Cialdini Robert B. , Trost Melanie R.. (1998), " Social Influence: Social Norms, Conformity and Compliance, " in Handbook of Social Psychology , Gilbert Daniel T. , Fiske Susan T. , Lindzey Gardner , eds. Boston : McGraw-Hill , 151 – 92.
Clee Mona A. , Wicklund Robert A.. (1980), " Consumer Behavior and Psychological Reactance ," Journal of Consumer Research , 6 (4), 389 – 405.
Cohen Jacob. (1992), " A Power Primer ," Psychological Bulletin , 112 (1), 155 – 59.
Crawford Sue E.S. , Ostrom Elinor. (1995), " A Grammar of Institutions ," American Political Science Review , 89 (3), 582 – 600.
Crockett Molly J. (2017), " Moral Outrage in the Digital Age ," Nature Human Behaviour , 1 (11), 769 – 71.
Deci Edward L. , Koestner Richard , Ryan Richard M.. (1999), " A Meta-Analytic Review of Experiments Examining the Effects of Extrinsic Rewards on Intrinsic Motivation ," Psychological Bulletin , 125 (6), 627 – 68.
Dzisi Emmanuel Komla Junior , Dei Oscar Akunor. (2020), " Adherence to Social Distancing and Wearing of Masks Within Public Transportation During the COVID 19 Pandemic ," Transportation Research Interdisciplinary Perspectives , 7 , 1 – 6.
Fehr Ernst , Gächter Simon. (2002), " Altruistic Punishment in Humans ," Nature , 415 (6868), 137 – 40.
Feldman-Summers Shirley. (1977), " Implications of the Buck-Passing Phenomenon for Reactance Theory ," Journal of Personality , 45 (4), 543 – 53.
Ferguson Christopher J. , Brannick Michael T.. (2012), " Publication Bias in Psychological Science: Prevalence, Methods for Identifying and Controlling, and Implications for the Use of Meta-Analyses ," Psychological Methods , 17 (1), 120 – 28.
Fitzsimons Gavan J. , Lehmann Donald R.. (2004), " Reactance to recommendations: When Unsolicited Advice Yields Contrary Responses ," Marketing Science , 23 (1), 82 – 94.
Gilovich Thomas , Medvec Victoria Husted , Savitsky Kenneth. (2000), " The Spotlight Effect in Social Judgment: An Egocentric Bias in Estimates of the Salience of One's Own Actions and Appearance ," Journal of Personality and Social Psychology , 78 (2), 211 – 22.
Goldstein Noah J. , Cialdini Robert B. , Griskevicius Vladas. (2008), " A Room with a Viewpoint: Using Social Norms to Motivate Environmental Conservation in Hotels ," Journal of Consumer Research , 35 (3), 472 – 82.
Green Ronald M. (1991), " When Is 'Everyone's Doing It' a Moral Justification? " Business Ethics Quarterly , 1 (1), 75 – 93.
Griskevicius Vladas , Goldstein Noah J. , Mortensen Chad R. , Cialdini Robert B. , Kenrick Douglas T.. (2006), " Going Along Versus Going Alone: When Fundamental Motives Facilitate Strategic (Non) Conformity ," Journal of Personality and Social Psychology , 91 (2), 281 – 94.
Griskevicius Vladas , Kenrick Douglas T.. (2013), " Fundamental Motives: How Evolutionary Needs Influence Consumer Behavior ," Journal of Consumer Psychology , 23 (3), 372 – 86.
Griskevicius Vladas , Tybur Joshua M. , Van den Bergh Bram. (2010), " Going Green to Be Seen: Status, Reputation, and Conspicuous Conservation ," Journal of Personality and Social Psychology , 98 (3), 392 – 404.
Hechter Michael , Opp Karl-Dieter , eds. (2005), Social Norms. New York : Russell Sage.
Hedges Larry V. , Pigott Therese D.. (2001), " The Power of Statistical Tests in Meta-Analysis ," Psychological Methods , 6 (3), 203 – 17.
Hennig-Thurau Thorsten , Malthouse Edward C. , Friege Christian , Gensler Sonja , Lobschat Lara , Rangaswamy Arvind , et al. (2010), " The Impact of New Media on Customer Relationships ," Journal of Service Research , 13 (3), 311 – 30.
Henrich Joseph , Boyd Robert. (2001), " Why People Punish Defectors: Weak Conformist Transmission Can Stabilize Costly Enforcement of Norms in Cooperative Dilemmas ," Journal of Theoretical Biology , 208 (1), 79 – 89.
Henrich Joseph , Ensminger Jean , McElreath Richard , Barr Abigail , Barrett Clark , Bolyanatz Alexander , et al. (2010), " Markets, Religion, Community Size, and the Evolution of Fairness and Punishment ," Science , 327 (5972), 1480 – 84.
Herman C. Peter , Roth Deborah A. , Polivy Janet. (2003), " Effects of the Presence of Others on Food Intake: A Normative Interpretation ," Psychological Bulletin , 129 (6), 873 – 86.
Herrmann Benedikt , Thöni Christian , Gächter Simon. (2008), " Antisocial Punishment Across Societies ," Science , 319 (5868), 1362 – 67.
Hofstede Geert. (2001), Culture's Consequences: Comparing Values, Behaviors, Institutions and Organizations Across Nations. Thousand Oaks, CA : SAGE Publications.
Homburg Christian , Wieseke Jan , Kuehnl Christina. (2010), " Social Influence on Salespeople's Adoption of Sales Technology: A Multilevel Analysis ," Journal of the Academy of Marketing Science , 38 (2), 159 – 68.
Inglehart Ronald , Baker Wayne E.. (2000), " Modernization, Cultural Change, and the Persistence of Traditional Values ," American Sociological Review , 65 (1), 19 – 51.
Inglehart Robert , Haerpfer C. , Moreno Alejandro , Welzel Chris , Kizilova Kseniya , Diez-Medrano Jaime , et al. (2014), World Values Survey: All Rounds-Country-Pooled Datafile Version. Madrid : JD Systems Institute , http://www.worldvaluessurvey.org/WVSDocumentationWVL.jsp.
Iyengar Sheena S. , Lepper Mark R.. (1999), " Rethinking the Value of Choice: A Cultural Perspective on Intrinsic Motivation ," Journal of Personality and Social Psychology , 76 (3), 349 – 66.
Kähr Andrea , Nyffenegger Bettina , Krohmer Harley , Hoyer Wayne D.. (2016), " When Hostile Consumers Wreak Havoc on Your Brand: The Phenomenon of Consumer Brand Sabotage ," Journal of Marketing , 80 (3), 25 – 41.
Keizer Kees , Lindenberg Siegwart , Steg Linda. (2008), " The Spreading of Disorder ," Science , 322 (5908), 1681 – 85.
Keller Punam Anand , Lehmann Donald R.. (2008), " Designing Effective Health Communications: A Meta-Analysis ," Journal of Public Policy & Marketing , 27 (2), 117 – 30.
Kivetz Ran. (2005), " Promotion Reactance: The Role of Effort-Reward Congruity ," Journal of Consumer Research , 31 (4), 725 – 36.
Kronrod Ann , Grinstein Amir , Wathieu Luc. (2012), " Go Green! Should Environmental Messages Be So Assertive? " Journal of Marketing , 76 (1), 95 – 102.
Lamb Theodore A. , Alsikafi Majeed. (1980), " Conformity in the Asch Experiment: Inner-Other Directedness and the Defiant Subject ," Social Behavior and Personality: An International Journal , 8 (1), 13 – 16.
Lapinski Maria Knight , Rimal Rajiv N.. (2005), " An Explication of Social Norms ," Communication Theory , 15 (2), 127 – 47.
Larsen Knud S. (1990), " The Asch Conformity Experiment: Replication and Transhistorical Comparison ," Journal of Social Behavior and Personality , 5 (4), 163 – 68.
Lee Richard , Murphy Jamie , Neale Larry. (2009), " The Interactions of Consumption Characteristics on Social Norms ," Journal of Consumer Marketing , 26 (4), 277 – 85.
Leshner Alan I. (1997), " Addiction Is a Brain Disease, and It Matters ," Science , 278 (5335), 45 – 47.
Lewis Nehama. (2013), " Priming Effects of Perceived Norms on Behavioral Intention Through Observability ," Journal of Applied Social Psychology , 43 (Special Issue), E97–E108.
Lin Lily , Dahl Darren W. , Argo Jennifer J.. (2013), " Do the Crime, Always Do the Time? Insights into Consumer-to-Consumer Punishment Decisions ," Journal of Consumer Research , 40 (1), 64 – 77.
Lipsey Mark W. , Wilson David B.. (2001), Practical Meta-Analysis. London : SAGE Publications.
Lisjak Monika , Bonezzi Andrea , Rucker Derek. (2021), " How Marketing Perks Influence Word of Mouth ," Journal of Marketing , 85 (5), 128 – 44.
Mann Millard F. , Hill Thomas. (1984), " Persuasive Communications and the Boomerang Effect: Some Limiting Conditions to the Effectiveness of Positive Influence Attempts ," in Advances in Consumer Research , Vol. 11 , Thomas C. Kinnear, ed. Provo, UT: Association for Consumer Research, 66 – 70.
Manning Mark. (2009), " The Effects of Subjective Norms on Behaviour in the Theory of Planned Behaviour: A Meta-Analysis ," British Journal of Social Psychology , 48 (40), 649 – 705.
Mathewes Charles. (2010), Understanding Religious Ethics. New York : John Wiley & Sons.
Mead Nicole L. , Baumeister Roy F. , Stillman Tyler F. , Rawn Catherine D. , Vohs Kathleen D.. (2010), " Social Exclusion Causes People to Spend and Consume Strategically in the Service of Affiliation ," Journal of Consumer Research , 37 (5), 902 – 19.
Melnyk Valentyna , Bijmolt Tammo. (2015), " The Effects of Introducing and Terminating Loyalty Programs ," European Journal of Marketing , 49 (3/4), 398 – 419.
Melnyk Valentyna , Giarratana Marco , Torres Anna. (2014), " Marking Your Trade: Cultural Factors in the Prolongation of Trademarks ," Journal of Business Research , 67 (4), 478 – 85.
Melnyk Valentyna , Klein Kristina , Völckner Franziska. (2012), " The Double-Edged Sword of Foreign Brand Names for Companies from Emerging Countries ," Journal of Marketing , 76 (6), 21 – 37.
Melnyk Vladimir , Van Herpen Erica , Jak Suzanne , Van Trijp Hans C.M.. (2019), " The Mechanisms of Social Norms' Influence on Consumer Decision Making ," Zeitschrift für Psychologie , 227 (1), 4 – 17.
Meyers Christopher. (2004), " Cruel Choices: Autonomy and Critical Care Decision-Making ," Bioethics , 18 (2), 104 – 19.
Milgram Stanley. (1974), Obedience to Authority. New York : Harper & Row.
Miller Claude H. , Burgoon Michael , Grandpre Joseph R. , Alvaro Eusebio M.. (2006), " Identifying Principal Risk Factors for the Initiation of Adolescent Smoking Behaviors: The Significance of Psychological Reactance ," Health Communication , 19 (3), 241 – 52.
Miron Anca M. , Brehm Jack W.. (2006), " Reactance Theory: 40 Years Later ," Zeitschrift für Sozialpsychologie , 37 (1), 9 – 18.
Nolan Jessica M. , Wesley Schultz P. , Cialdini Robert B. , Goldstein Noah J. , Griskevicius Vladas. (2008), " Normative Social Influence Is Underdetected ," Personality and Social Psychology Bulletin , 34 (7), 913 – 23.
Nunes Joseph C. , Ordanini Andrea , Giambastiani Gaia. (2021), " The Concept of Authenticity: What It Means to Consumers ," Journal of Marketing , 85 (4), 1 – 20.
Okada Erica Mina. (2005), " Justification Effects on Consumer Choice of Hedonic and Utilitarian Goods ," Journal of Marketing Research , 42 (1), 43 – 53.
Ostrom Elinor. (2000), " Collective Action and the Evolution of Social Norms ," Journal of Economic Perspectives , 14 (3), 137 – 58.
Pagiavlas Sotires , Kalaignanam Kartik , Gill Manpreet , Bliese Paul D.. (2021), " Regulating Product Recall Compliance in the Digital Age: Evidence from the 'Safe Cars Save Lives' Campaign ," Journal of Marketing (published online May 20), https://doi.org/10.1177/00222429211023016.
Petty Richard E. , Cacioppo John T.. (1979), " Effects of Forewarning of Persuasive Intent and Involvement on Cognitive Responses and Persuasion ," Personality and Social Psychology Bulletin , 5 (2), 173 – 76.
Pliner Patricia , Mann Nikki. (2004), " Influence of Social Norms and Palatability on Amount Consumed and Food Choice ," Appetite , 42 (2), 227 – 37.
Rajavi Koushyar , Kushwaha Tarun , Steenkamp Jan-Benedict E.M.. (2019), " A Multicategory, Multicountry Investigation of Sensitivity of Consumers' Trust in Brands to Marketing-Mix Activities ," Journal of Consumer Research , 46 (4), 651 – 70.
Raudenbush Stephen W. , Bryk Anthony S.. (2002), Hierarchical Linear Models: Applications and Data Analysis Methods , 2nd ed. Thousand Oaks, CA : SAGE Publications.
Rice Richard , Haines Michael P.. (2003), " Social Norms: A Publicly Funded, Cost-Effective Approach," The National Social Norms Research Center (January), http://www.socialnormsresources.org/pdf/funding.pdf.
Rivis Amanda , Sheeran Paschal. (2003), " Descriptive Norms as an Additional Predictor in the Theory of Planned Behaviour: A Meta-Analysis ," Current Psychology , 22 (3), 218 – 33.
Rosenberg Benjamin D. , Siegel Jason T.. (2018), " A 50-Year Review of Psychological Reactance Theory: Do Not Read This Article ," Motivation Science , 4 (4), 281 – 300.
Samaha Stephen A. , Beck Joshua T. , Palmatier Robert W.. (2014), " The Role of Culture in International Relationship Marketing ," Journal of Marketing , 78 (5), 78 – 98.
Savani Krishna , Wadhwa Monica , Uchida Yukiko , Ding Yu , Naidu N.V.R.. (2015), " When Norms Loom Larger Than the Self: Susceptibility of Preference–Choice Consistency to Normative Influence Across Cultures ," Organizational Behavior and Human Decision Processes , 129 (July), 70 – 79.
Schultz P. Wesley , Nolan Jessica M. , Cialdini Robert B. , Goldstein Noah J. , Griskevicius Vladas. (2007), " The Constructive, Destructive, and Reconstructive Power of Social Norms ," Psychological Science , 18 (5), 429 – 34.
Schultz P. Wesley , Nolan Jessica M. , Cialdini Robert B. , Goldstein Noah J. , Griskevicius Vladas. (2018), " The Constructive, Destructive, and Reconstructive Power of Social Norms: Reprise ," Perspectives on Psychological Science , 13 (2), 249 – 54.
Scott John Finley. (1971), Internalization of Norms: A Sociological Theory of Moral Commitment. Englewood Cliffs, NJ : Prentice-Hall.
Sheeran Pascal , Abraham Charles , Orbell Sheina. (1999), " Psychosocial Correlates of Heterosexual Condom Use: A Meta-Analysis ," Psychological Bulletin , 125 (1), 90 – 132.
Silvia Paul J. (2005), " Deflecting Reactance: The Role of Similarity in Increasing Compliance and Reducing Resistance ," Basic and Applied Social Psychology , 27 (3), 277 – 284.
Stanley Tom D. , Doucouliagos Hristos. (2012), Meta-Regression Analysis in Economics and Business. London : Routledge.
Staunton Mina , Louis Winnifred R. , Smith Joanne R. , Terry Deborah J. , McDonald Rachel I.. (2014), " How Negative Descriptive Norms for Healthy Eating Undermine the Effects of Positive Injunctive Norms ," Journal of Applied Social Psychology , 44 (4), 319 – 30.
Stephan Ute , Uhlaner Lorraine M.. (2010), " Performance-Based vs. Socially Supportive Culture: A Cross-National Study of Descriptive Norms and Entrepreneurship ," Journal of International Business Studies , 41 (8), 1347 – 64.
Teasdale Emma , Santer Miriam , Geraghty Adam W.A. , Little Paul , Yardley Lucy. (2014), " Public Perceptions of Non-Pharmaceutical Interventions for Reducing Transmission of Respiratory Infection: Systematic Review and Synthesis of Qualitative Studies ," BMC Public Health , 14 (1), 589.
Tung Rosalie L. , Verbeke Alain. (2010), " Beyond Hofstede and GLOBE: Improving the Quality of Cross-Cultural Research ," Journal of International Business Studies , 41 , 1259 – 74.
Van den Noortgate Wim , López-López José Antonio , Marín-Martínez Fulgencio , Sánchez-Meca Julio. (2015), " Meta-Analysis of Multiple Outcomes: A Multilevel Approach ," Behavior Research Methods , 47 (4), 1274 – 94.
Van der Lans Ralf , van Everdingen Yvonne , Melnyk Valentyna. (2016), " What to Stress, to Whom and Where? A Cross-Country Investigation of The Effects of Perceived Brand Benefits on Buying Intentions ," International Journal of Research in Marketing , 33 (4), 924 – 43.
Viechtbauer Wolfgang. (2010), " Conducting Meta-Analyses in R with the Metafor Package ," Journal of Statistical Software , 36 (3), 1 – 48.
Wechsler Henry , Nelson Toben E. , Lee Jae Eun , Seibring Mark , Lewis Catherine , Keeling Richard P.. (2003), " Perception and Reality: A National Evaluation of Social Norms Marketing Interventions to Reduce College Students' Heavy Alcohol Use ," Journal of Studies on Alcohol , 64 (4), 484 – 94.
White Katherine , Habib Rishad , Hardisty David J.. (2019), " How to SHIFT Consumer Behaviors to Be More Sustainable: A Literature Review and Guiding Framework ," Journal of Marketing , 83 (3), 22 – 49.
White Katherine , Simpson Bonnie. (2013), " When Do (and Don't) Normative Appeals Influence Sustainable Consumer Behaviors? " Journal of Marketing , 77 (2), 78 – 95.
Zemack-Rugar Yael , Moore Sarah G. , Fitzsimons Gavan J.. (2017), " Just Do It! Why Committed Consumers React Negatively to Assertive Ads ," Journal of Consumer Psychology , 27 (3), 287 – 301.
Zhang Xiaoling , Li Shibo , Burke Raymond R. , Leykin Alex. (2014), " An Examination of Social Influence on Shopper Behavior Using Video Tracking Data ," Journal of Marketing , 78 (5), 24 – 41.
~~~~~~~~
By Vladimir Melnyk; François A. Carrillat and Valentyna Melnyk
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 124- The Pet Exposure Effect: Exploring the Differential Impact of Dogs Versus Cats on Consumer Mindsets. By: Jia, Lei; Yang, Xiaojing; Jiang, Yuwei. Journal of Marketing. Sep2022, Vol. 86 Issue 5, p42-57. 16p. 2 Charts. DOI: 10.1177/00222429221078036.
- Database:
- Business Source Complete
The Pet Exposure Effect: Exploring the Differential Impact of Dogs Versus Cats on Consumer Mindsets
Despite the ubiquity of pets in consumers' lives, scant research has examined how exposure to them (e.g., recalling past interactions with dogs and cats, viewing ads featuring a dog or a cat) influences consumer behavior. The authors demonstrate that exposure to dogs (cats) reminds consumers of the stereotypical temperaments and behaviors of the pet species, which activates a promotion- (prevention-) focused motivational mindset among consumers. Using secondary data, Study 1 shows that people in states with a higher percentage of dog (cat) owners Google more promotion- (prevention-) focused words and report a higher COVID-19 transmission rate. Using multiple products, Studies 2 and 3 demonstrate that these regulatory mindsets, when activated by pet exposure, carry over to influence downstream consumer judgments, purchase intentions, and behaviors, even in pet-unrelated consumption contexts. Study 4 shows that pet stereotypicality moderates the proposed effect such that the relationship between pet exposure and regulatory orientations persists to the extent consumers are reminded of the stereotypical temperaments and behaviors of the pet species. Studies 5–7 examine the role of regulatory fit and evince that exposure to dogs (cats) leads to more favorable responses toward advertising messages featuring promotion- (prevention-) focused appeals.
Keywords: pets; regulatory orientation; advertising; COVID-19
Pets are prevalent and play important roles in consumers' daily lives ([ 2]; Cavanaugh, Leonard, and Scammon 2008; Hirschman 1994; Holbrook and Woodside 2008; Serpell and Paul 2011). According to the survey of the American Pet Products Association ([ 4]), 68% of U.S. households, or 84.6 million homes, own a pet. Dogs and cats are the most popular pets, with 48% of U.S. households (60 million homes) owning at least one dog and 37% of U.S. households (47 million homes) owning at least one cat. Pet adoption rates have climbed significantly, with about one in five households having acquired a dog or cat since the outbreak of the COVID-19 pandemic (American Society for the Prevention of Cruelty to Animals [ASPCA] 2021). Pets also frequently appear in popular culture, mass media, and marketing communications. For example, Target uses a dog as its brand mascot, Microsoft features dogs in its 2020 holiday commercial to inspire people to find joy, and Wells Fargo uses a cat in its commercial to advertise its suspicious card activity alert services.
Despite the significance of pets in people's lives and in mass media, popular culture, and marketing communications, scant research has examined how pets may influence consumers' judgments, decisions, and behaviors. Existing research on pet–human relationships largely revolves around examining how owning a pet influences the owner's pet-related judgments and behaviors. For example, attesting to a strong tie between owners and their pets ([ 2]; Cavanaugh, Leonard, and Scammon 2008), this stream of literature suggests that pets provide not only companionship but also a sense of safety and belongingness for their owners ([79]). Pet owners have significantly greater physical and psychological well-being than non-pet owners ([ 2]) and are more likely to endorse causes protecting animal rights (Kidd, Kidd, and Zasloff 1995).
Research that examines how pets may influence consumer behavior beyond the immediate context of pet ownership is lacking, however. Such knowledge would provide novel and important insights to marketers and allow them to develop marketing strategies based on pet exposure situations. For example, marketers might choose to recommend more fitting products or services or craft appropriate communication messages to effectively target consumers depending on the type of pets to which they are exposed. Consider the following scenario: A newly opened massage center is pondering the language to use in direct mail to potential customers and whether to focus on how its therapies help reduce fatigue and stress or how its therapies promote metabolism and energy levels. Which strategy might be more effective if the company features a cat (dog) figure in its advertisement? This article provides a theory-based answer to this question.
In our research, we focus on the effects of exposure to pets (e.g., recalling consumers' past interactions with dogs or cats, viewing ads featuring a dog or a cat) and examine how such experience influences consumers' judgments and decision making through the lens of regulatory focus theory. Drawing from literature on human–animal relationships, research on regulatory orientation, and work on mindset, we suggest that exposure to pets will remind consumers of the stereotypical personality traits, temperaments, and behaviors of the pets and thus will evoke different regulatory mindsets among consumers. Specifically, we predict and find that exposure to dogs fosters a more promotion-focused motivational mindset whereas exposure to cats activates a more prevention-focused motivational mindset. We further identify pet stereotypicality as a moderator for our findings, such that our results on the relationship between pet exposure and regulatory orientations persist only when consumers are reminded of the stereotypical temperaments and behaviors of the pet species, and that our main proposed effects dissipate when consumers are reminded of information inconsistent with the pet stereotypes. Moreover, we show that these regulatory mindsets, when activated by pet exposure, carry over to influence downstream consumer judgments, purchase intentions, and behaviors, even in pet-unrelated consumption contexts.
Regulatory focus theory suggests that consumer judgments, decisions, and behaviors are motivated by two regulatory orientations: promotion and prevention focus (Higgins 1997; Lee, Aaker, and Gardner 2000; Pham and Avnet 2004; Wang and Lee 2006). A promotion focus in self-regulation reflects consumers' motivations to attain growth and nurturance in an effort to align their actual selves with their ideal selves (achieving accomplishments and fulfilling aspirations; Higgins 1987, [28]). Promotion-focused consumers are characterized by an eagerness regulatory system during behavioral regulation (Crowe and Higgins 1997; Lee, Keller, and Sternthal 2010; Pham and Chang 2010; Wang and Lee 2006). For example, they are sensitive to the presence or absence of positive outcomes (gains and successes; Higgins 1997), concerned about reducing errors of omission (Croe and Higgins 1997), and more risk seeking when processing information and rendering decisions (Zhou and Pham 2004). By contrast, self-regulation with a prevention focus reflects consumers' motivations to attain safety and security in an attempt to bring their actual selves into alignment with their "ought" selves (fulfilling duties and obligations; Higgins 1987, [28]). Thus, prevention-focused consumers are more vigilant and cautious during behavior regulation (Crowe and Higgins 1997; Lee, Keller, and Sternthal 2010; Pham and Chang 2010; Wang and Lee 2006). In this system, consumers are sensitive to the presence or absence of negative outcomes (losses and failures; Higgins 1997), concerned about reducing errors of commission (Crowe and Higgins 1997), and more risk averse (Zhou and Pham 2004) when processing information and making decisions.
More germane to our research is the finding that social influences play a pivotal role in shaping people's regulatory orientations. For example, interactions with childhood caretakers and parents' parenting style can influence the formulation of consumers' regulatory orientations during the socialization process (Crowe and Higgins 1997; Higgins 1996). Social exclusion causes a shift toward prevention motivation (Park and Baumeister 2015), whereas making choices for other people instigates a promotion focus ([59]). In addition, distinct social relationships can activate alternative regulatory orientations, such that reminders of friends activate a promotion focus while reminders of family members engender a prevention focus ([22]). Similarly, positive role models induce a promotion focus, whereas negative role models instigate a prevention focus (Lockwood, Jordan, and Kunda 2002).
Given pets' prevalence in consumers' daily lives, we posit that consumers' interactions with pets are also an important part of socialization that can influence their regulatory orientations. These socialization activities can involve direct or indirect interactions with pets (e.g., observing pets' interactions with other people). Indeed, research on human–animal relationships evinces that pets play an important part in people's socialization process, influencing the development of various cognitive and social abilities (e.g., worldviews, empathy; [ 2]; Myers 1999; Purewal et al. 2017).
As we have mentioned, dogs and cats are two primary types of pets ([ 4]). Despite within-species breed differences, research on animal behavior has identified systematic cross-species differences between domesticated dogs and cats ([11], [12]; Jardim-Messeder et al. 2017). This stream of research suggests that a promotion-oriented eagerness system better captures dogs' temperaments and behavioral characteristics, whereas a prevention-focused cautious system better describes cats' temperaments and behavioral characteristics. For example, on a temperament level, dogs tend to be open and expressive, while cats are elusive and cautious ([11], [12]; Potter and Mills 2015). Consistent with the promotion orientation's receptivity to change (Boldero and Higgins 2011; [43]), dogs (vs. cats) cope better with and adapt quicker to changes in the environment, such as moving into a new house or having a new person in the household ([11], [12]; Langenfeld 2020). In line with the prevention orientation's preference for the status quo ([10]; Chernev 2004), cats (vs. dogs) appear more concerned with the protection their owners provide and the consistency and stability of their social and physical surroundings ([11], [12]). Similarly, consistent with the eagerness prediction of a promotion regulatory system ([15]), dogs are more responsive to rewards (e.g., food, praise, petting) than cats and thus are easier to train ([49]).
When interacting with human beings and other pets, dogs are more eager to please their owners and socialize with other dogs, whereas cats are more cautious, suspicious, boundary setting, and anxious when surrounded by unfamiliar people or other cats ([11], [12]; Potter and Mills 2015). Indeed, research has shown that dogs are more attentive and responsive to human's social cues (e.g., gestures) than cats ([49]; Wynne, Udell, and Lord 2008). Dogs' eagerness can be exemplified by the spike in oxytocin (a hormone mammals release when they feel love or affection for someone) when their owners are around ([53]). A study conducted by scientists at BBC shows that dogs produce five times more oxytocin than cats upon seeing their owners ([21]).
We further predict that through repeated socialization episodes with pets (through either direct or indirect interaction with pets), the traits and motivational characteristics of dogs (cats) are gradually associated with a promotion-focused (prevention-focused) eagerness (cautiousness) system. These learned associations are brought to mind and thus accessible when consumers interact with pets or encounter stimuli featuring pets (e.g., ads) in their daily lives. To confirm the associations of dogs and cats with promotion and prevention orientations, we conducted a pilot study, which found that participants indeed associated promotion-focused words with dogs and prevention-focused words with cats (for details of the pilot study, see Web Appendix A).
Drawing from research on motivational mindset, which we review next, we further predict that exposure to dogs (cats) or stimuli featuring them (e.g., ads) will remind consumers of the temperaments and behaviors of the dogs (cats), which will in turn activate a promotion-focused (prevention-focused) mindset among consumers and guide their subsequent judgment and decision making. Mindset reflects "the activation and use of a procedure that is stored in memory as part of declarative knowledge" ([76], p. 110). That is, engaging in a particular operation when pursuing a goal in a prior task may give rise to a mindset (e.g., a promotion-focused mindset) that remains accessible in consumers' memory and, in turn, guides their pursuit of a different goal in a subsequent, unrelated context.
A growing body of literature has found considerable evidence of the role of mindset across a wide range of information-processing activities, from comprehension, to judgment, to decision making (Ma and Roese 2014; Wyer and Xu 2010; Xu et al. 2020). In some situations, mindsets involve cognitive procedures induced by engaging in a prior task that spills over to influence a subsequent, unrelated context. For example, Xu et al. (2020) show that managers during election years are more likely to adopt a comparative mindset due to the omnipresence of comparative political advertisements. Accordingly, they spend more money on their managerial decisions because the comparative mindset accentuates "which option to spend money on" and forgoes the "whether or not to spend" consideration. More germane to our theorizing, mindsets may also be based on motivation ([76]), such that the motivational mindset induced by pursuing a goal in a prior task will guide consumers' subsequent behavior in an unrelated context (e.g., pursuit of a different goal). For example, Wyer and Xu (2010) assert that the promotion (prevention) regulatory mindset can be induced procedurally by, for example, making salient participants' desire to achieve their ideal (ought) self. When activated, the promotion (prevention) regulatory mindset produces a cross-domain effect, making consumers, for example, more likely to approach positive (avoid negative) consequences in their decision making.
Drawing on this stream of literature, we posit that exposure to pets (e.g., recalling an interaction with a pet, viewing ads featuring a pet as the spokescharacter) in a prior task may render different regulatory mindsets salient. Specifically, because the stereotypical personality traits, temperaments, and behaviors of dogs (cats) brought to mind by the pet exposure are associated with eagerness (cautiousness) strategies commonly employed by a promotion (prevention) orientation, consumers' different regulatory orientations (promotion vs. prevention) will be activated. When evoked, these motivational regulatory mindsets will carry over to influence consumers' subsequent, unrelated judgments and decision making, rendering them more eager (cautious) during behavioral regulation, leading them to pursue promotion- (prevention-) focused goals such as growth and advancement (safety and stability), and making them more risk seeking (more risk averse) in decision making. Thus,
H1: Pet exposure activates different regulatory motivational mindsets among consumers, such that (a) exposure to dogs or dog-featuring stimuli (vs. cats or cat-featuring stimuli) activates a more promotion-focused mindset and (b) exposure to cats or cat-featuring (vs. dogs or dog-featuring) stimuli activates a more prevention-focused mindset.
A key premise of our theorizing that exposure to pets activates different regulatory mindsets is that such exposure will remind consumers of the stereotypical temperaments and behavioral characteristics of dogs (cats), giving rise to a promotion-focused (prevention-focused) mindset. In other words, through repeated socialization, consumers have developed preestablished mental connections between dogs' (cats') typical temperaments and behaviors and the promotion (prevention) focus, and exposure to pets or pet-featuring stimuli can render these stereotypical associations accessible, thus activating the corresponding regulatory-focus mindset among consumers. Prior research has shown that established mental associations are likely to be temporarily weakened, nullified, or even reversed when presented with information inconsistent with the original associations. For example, Gorn, Jiang, and Johar (2008) reversed the association between baby-faceness and unintentionality by presenting counterassociation information about a baby-faced person intentionally harming others. Thus, if our reasoning that mental associations between dogs (cats) and promotion- (prevention-) focused mindsets is right, our proposed effects should persist to the extent consumers are reminded of the stereotypical behaviors and temperaments of the pet species; when consumers are reminded of pet information inconsistent with the stereotypes of the species (e.g., dogs [cats] unlike a stereotypical dog [cat]), we are likely to show that our results are attenuated. More formally,
H2: Pet stereotypicality moderates the effect of pet exposure on regulatory mindsets, such that the effect dissipates when consumers are exposed to pets that are inconsistent with the stereotypes of the species.
We expect that the impact of pet exposure on consumers' motivational mindsets will carry over to influence downstream variables, including ad evaluation, purchase intention, and real purchasing behavior, even in pet-unrelated consumption contexts. We anticipate that the effects of pet exposure on these variables will stem from the activation of a regulatory mindset and regulatory fit. Regulatory fit occurs when the regulatory strategies individuals employ during goal pursuit are compatible with their regulatory orientations (Higgins 2000; Hong and Lee 2007); it usually results in favorable effects on downstream consumer responses, such as enhanced value of the product ([ 7]), brand attitudes ([37]), self-regulation (Hong and Lee 2007), and decision making (Zhou and Pham 2004).
Therefore, in accordance with this literature, we anticipate that consumers who are exposed to dogs or dog-featuring stimuli will experience higher regulatory fit and develop more favorable product evaluations when presented with ad messages featuring promotion-focused claims. By contrast, consumers who are exposed to cats or cat-featuring stimuli will experience higher regulatory fit and develop more favorable product evaluations when presented with ad messages featuring prevention-focused claims. Thus,
H3: There is an interaction between pet exposure and the regulatory focus of an ad on consumers' evaluations of the advertised product, such that (a) when exposed to ads with promotion-focused claims, exposure to dogs or dog-featuring stimuli (vs. cats or cat-featuring stimuli) leads consumers to form more favorable product evaluations, and (b) when exposed to ads with prevention-focused claims, exposure to cats or cat-featuring stimuli (vs. dogs or dog-featuring stimuli) leads consumers to form more favorable product evaluations.
H4: Regulatory fit mediates the interaction between pet exposure and ads' regulatory focus proposed in H3 on product evaluations.
Studies 1 and 2 provide initial evidence for our prediction by showing that long-term exposure to dogs (cats) is associated with a promotion (prevention) focus. Specifically, using secondary data gathered from the American Veterinary Medical Association, Google Trends, and Centers for Disease Control and Prevention (CDC), Study 1 finds that people in states with a higher percentage of dog (cat) owners search more promotion- (prevention-) focused words (Study 1a) and show a higher COVID-19 transmission rate (Study 1b). Study 2 shows that dog (vs. cat) owners are more likely to invest in stocks and are less likely to invest in mutual funds in financial decision making. Studies 3a–3d establish the basic effect that exposure to dogs (cats) activates a promotion- (prevention-) focused motivational mindset by employing multiple experimental manipulations of pet exposure and different measures of regulatory orientation in both pet-related and pet-unrelated contexts, including incentive-compatible choices. Study 4 explores a moderating effect for our findings, showing that our hypothesized effects will dissipate when consumers are exposed to pet information inconsistent with the stereotypes of the pet species. Study 5 examines the downstream effects of pet exposure on consumers' incentive-compatible behaviors, showing that consumers exposed to dogs (cats) bid higher for products framed with a promotion (prevention) focus. Studies 6 and 7 provide additional support for our theorizing by examining the mediation effect of regulatory fit. Table 1 in Web Appendix B provides a summary of all studies.
Graph
Table 1. Study 1a: Results from the Multiple Regression Model.
| b | t | p |
|---|
| Pet ownership indexa | .36 | 2.62 | .012 |
| Median household income | −.15 | −.63 | .534 |
| Per capita GDP | −.25 | −1.26 | .214 |
| Political orientationb | −.05 | −.29 | .774 |
1 a Higher scores indicate more dog (vs. cat) owners.
- 2 b 1 = Democratic, 2 = Republican.
- 3 Notes: Dependent variable = Regulatory-focus index: high scores indicate more promotion- (vs. prevention-) focused.
Relying on secondary data and operationalizing pet exposure as pet ownership, Study 1 aims to provide preliminary evidence for our prediction that exposure to dogs and cats is associated with different regulatory mindsets. We collected aggregated state-level data on pet owner statistics, public interest in promotion- versus prevention-oriented behaviors, and per capita COVID-19 cases during the pandemic. We expect that at the state level, having a relatively higher dog-owning (cat-owning) population will be related to more search interests in promotion-oriented (prevention-oriented) behaviors in general (Study 1a) and more per capita COVID-19 cases during the pandemic (Study 1b).
For pet ownership, we obtained the latest (2016) state-level pet ownership data set (n = 49) from the U.S. Pet Ownership and Demographics Sourcebook released by the [ 1]. This data set provides the most complete data on pet population demographics, covering the 48 U.S. continental states and the District of Columbia (excluding Alaska and Hawaii). For each state, we divided the percentage of dog-owning households by the percentage of cat-owning households to obtain a pet-owning index, with a higher score indicating more dog-owning (vs. cat-owning) households in the state.
To obtain a proxy for citizens' public interest in regulatory-oriented behaviors in each state, we examined the search interest scores data from Google Trends (Du, Hu, and Damangir 2015; Kozinets, Patterson, and Ashman 2017). Google is the most often-used internet search engine in the United States (accounting for 88% of the market share; Schultheiß and Lewandowski 2021), and Google Trends counts how often a particular search term is entered relative to the total search volume across various geographic regions. After a search term, period, and interested geographic area are entered, Google Trends displays how often that search term appears on Google in that geographic area and in that period relative to the total search volume on a standardized scale ranging from 0 (lowest search volume) to 100 (highest search volume). Given its viable role in monitoring public interests, Google Trends has become an increasingly used data source for research in psychology ([46]), political sciences (Mellon 2013; Weeks and Southwell 2010), and marketing (Du, Hu, and Damangir 2015; Kozinets, Patterson, and Ashman 2017).
To build the state-level regulatory orientation index, we first selected ten representative promotion-focused words (i.e., "growth," "gain," "achievement," "aspiration," "pleasure," "proud," "hope," "earn," "win," and "spontaneous") and ten representative prevention-focused words (i.e., "privacy," "safety," "loss," "prevention," "pain," "stable," "saving," "frugal," "rules," and "risky"), in line with literature on regulatory focus (Higgins 1997, [29], [30]; Scholer, Cornwell, and Higgins 2019). We then obtained search interest scores of these words on Google Trends from January 1, 2016, to December 31, 2020, across the 48 continental states plus the District of Columbia. For each state, we calculated the average search interest score for the ten promotion-focused words (α = .85) and the ten prevention-focused words (α = .74). Finally, we built a regulatory orientation index for each state by dividing the promotion search interest score by the prevention search interest score (i.e., higher numbers indicate a higher promotion focus).
To demonstrate the ecological validity of our findings, we also controlled for state-level microeconomic influence (income), macroeconomic influence (gross domestic product [GDP]), and political orientations. Specifically, we included the (state-level) covariates median household income in 2016 (U.S. Bureau of the Census 2017), per capita GDP in 2016 (U.S. Bureau of Economic Analysis 2019), and political orientation based on the 2016 presidential election results (The New York Times 2017).
We conducted a multiple linear regression with the pet ownership index as the independent variable, regulatory-orientation index as the dependent variable, and median household income, per capita GDP, and political orientation as covariates. Table 1 shows the results. The results reveal that our regression model was significant (F( 4, 44) = 4.86, p = .002), suggesting that the independent variables significantly explained the variance in regulatory orientation. More importantly, after controlling for the covariates, the pet ownership index (b = .36, t = 2.62, p = .012) significantly predicted the regulatory-orientation index, showing that at the state level, a relatively higher dog-owning (cat-owning) population is associated with more search interests in promotion-oriented (prevention-oriented) behaviors in general (for additional analyses, see Web Appendix C).
Study 1b uses the same pet-ownership data from Study 1a but focuses on per capita COVID-19 cases (CDC 2020) as a proxy for regulation-related behaviors. Considering the findings that promotion- (prevention-) focused people are more risk seeking (risk averse; Zhou and Pham 2004), we expect that dog owners will have an increased probability to engage in promotion-focused, relatively risky behaviors that may result in COVID-19 transmission (e.g., more willing to dine in restaurants, letting their guard down when following social distancing); by contrast, cat owners will have an increased probability to be more cautious and engage in less risky, prevention behaviors (e.g., behaving extra cautiously, practicing social distancing, wearing face masks). Accordingly, we predict that states with more dog (cat) owners will report a higher (lower) number of per capita COVID-19 cases.
To obtain a state-level proxy for regulatory-oriented behavior, we examined each state's COVID-19 cases per 100,000 people reported to the CDC from January 21, 2020 (the earliest available date) to November 1, 2020 (the date Study 1b was conducted). As of November 1, 2020, the 48 continental states and the District of Columbia had reported 2,819 COVID-19 cases per 100,000 people, on average, with Vermont being the lowest (348 per 100,000) and North Dakota the highest ( 6,054 per 100,000).
We performed a linear regression on COVID-19 cases (per 100,000), with the pet ownership index as the independent variable, and controlled for the same covariates as in Study 1a. The results reveal that our regression model was significant (F( 4, 44) = 8.93, p <.001), suggesting that the independent variables significantly explained the variance in COVID-19 cases. As Table 2 shows, the pet ownership index was significantly related to an increase in COVID-19 cases per 100,000 people (b = .38, t = 3.15, p = .003), suggesting that, at the state level, a relatively higher dog-owning (cat-owning) population was associated with more reported per capita COVID-19 cases. Consistent with this finding, our ancillary analyses (for detailed analyses and results, see Web Appendix D) also suggest that the pet ownership index (higher scores indicating more dog [vs. cat] owners) significantly increased search interests (per Google Trends during the same period as the COVID-19 data) in promotion-focused behaviors, such as dining in, but significantly reduced search interest in prevention-focused behaviors, such as face mask and social distancing.
Graph
Table 2. Study 1b: Results from the Multiple Regression Model.
| b | t | p |
|---|
| Pet ownership indexa | .38 | 3.15 | .003 |
| Median household income | .09 | .56 | .579 |
| Per capita GDP | .13 | .97 | .336 |
| Political orientationb | .53 | 3.71 | .001 |
- 4 a Higher scores indicate more dog (vs. cat) owners.
- 5 b 1 = Democratic, 2 = Republican.
- 6 Notes: Dependent variable = number of COVID-19 cases per 100,000 people.
Using aggregated state-level data across different data sources, Study 1 provides support for our prediction of a significant association between long-term pet exposure and people's regulatory orientations, such that dog (cat) exposure is associated with a promotion (prevention) focus. Specifically, we find that, at the state level, a relatively higher dog-owning (cat-owning) population is associated with more search interests in promotion- (prevention-) focused behaviors in general (Study 1a) and more reported per capita COVID-19 cases (Study 1b). In subsequent studies, we use individual-level data to provide additional support for our prediction.
Study 2 also operationalizes pet exposure as pet ownership and examines whether consumers' pet-owning situations are associated with different regulatory mindsets. Unlike Study 1, which used aggregate, state-level data, Study 2 relies on individual-level pet ownership data. In addition, we used an established measure of regulatory orientation ([65]), which involved participants in a financial decision-making task choosing between two investment options: stock (a proxy for promotion focus) and mutual fund (a proxy for prevention focus).
We recruited 145 pet owners from Amazon Mechanical Turk (MTurk) (Mage = 35.3 years; 53.1% female; 53% dog owners and 47% cat owners). We recruited only participants who own dogs only or cats only; owners of both dogs and cats were excluded (for the screening criteria, see Web Appendix E). We asked participants to partake in a financial decision-making task, which served as our measure of regulatory orientation (Zhou and Pham 2004). We first gave them basic definitions of stocks and mutual funds and told them that stock investments were typically associated with a higher level of risk, whereas mutual fund investments were typically associated with a lower level of risk and therefore more conservative. Next, we asked participants to imagine that they had $2,000 and were considering investing in two assets: a stock and a mutual fund. Afterward, we asked them to indicate which asset they would invest in if they could choose only one asset and then to indicate the amount of money they would invest. Given that prior research has shown that the activation of a promotion- (vs. prevention-) focused mindset entails greater risk taking (vs. risk aversion; Zhou and Pham 2004), we expected that dog (vs. cat) owners would be more willing to take risks in their financial investments and choose the stock option. After the financial decision-making task, participants completed measures of their mood using PANAS (Positive Affect Negative Affect Schedule; Watson, Clark, and Tellegen 1988) and a few demographic measures, including their age, gender, ethnicity, and income level (Web Appendix F presents the measures used).
A logistic regression showed a significant effect of pet ownership (cat owners = 0, dog owners = 1) on investment choice, such that dog owners (36.4%) were more likely to choose to invest in the riskier stock option than cat owners (20.6%; b = .79, SE = .38, χ2 = 4.73, p = .039, Exp (B) = 2.20). Similarly, a one-way analysis of variance (ANOVA) revealed a significant effect of pet ownership on money allocations. As we expected, dog owners allocated more money to the stock option (Mdog = $796.10, SD = $524.45) than cat owners (Mcat = $603.69, SD = $474.90; F( 1, 143) = 5.31, p = .023, = .04).
To rule out possible alternative explanations that participants' mood, gender, age, ethnicity, or income level accounted for our findings, we controlled for these variables simultaneously. Our results for both choice (χ2 = 5.47, p = .019) and money allocations (F = 5.74, p = .018) remained significant even after we controlled for these covariates.
Study 2's findings show that dog (vs. cat) owners were more likely to take risks in their financial decisions, showing more preference for stock investment. Importantly, incorporating the demographic variables age, ethnicity, gender, and income as covariates did not change the results. Taken together, using pet ownership as an operationalization, Studies 1 and 2 provide initial support for our prediction that exposure to pets is associated with different regulatory mindsets. However, despite the extra steps taken, such as controlling for demographic variables (e.g., income) to rule out alternative explanations, Studies 1 and 2 were correlational in nature. To provide stronger causal evidence for our prediction, in the subsequent studies, we manipulate exposure to pets in various ways.
The purpose of Study 3 is twofold. First, the study aims to establish causality between pet exposure and the formation of regulatory motivation mindsets by using multiple manipulations of pet exposure. Second, the study operationalizes regulatory orientations in various ways and across different (pet-related and pet-unrelated) contexts.
In a pet-related domain, Study 3a shows that participants exposed to dogs (vs. cats) will be more likely to prefer a pet toothpaste ad with promotion- (vs. prevention-) focused benefits. Studies 3b–3d test the effect in pet-unrelated domains. Consistent with prior research showing that the activation of a promotion- (prevention) focused mindset entails greater risk taking (risk aversion) (Zhou and Pham 2004), participants who are exposed to dogs (vs. cats) will be more willing to take risks to participate in a lottery (an incentive-compatible behavior; Study 3b) and in their financial investment decisions (Study 3c). In a health product context, Study 3d demonstrates that participants who are exposed to dogs (vs. cats) will be more likely to prefer a vitamin product with promotion- (vs. prevention-) focused benefits.
In this and subsequent studies, participants completed mood measures and demographic measures, including pet ownership, gender, age, income level, and ethnicity, at the end of study. Incorporating these variables as covariates did not influence our results (for the exact measures used, see Web Appendix F; for results pertaining to the impact of pet ownership across studies, see Web Appendix G), and thus we do not discuss them further.
In Study 3a, we examine our prediction in the context of pet-related decisions: pet toothpaste choice. One hundred eighty-three participants recruited from MTurk completed the study, which featured a two-cell (pet exposure: dog vs. cat) between-subjects design, for a small financial compensation (Mage = 37.4 years; 54.6% female). To manipulate pet exposure (dog vs. cat), under the cover story that we wanted to examine consumers' day-to-day experiences, participants were asked to recall and write down a past experience in which they interacted with a dog or cat (for details of the recall instructions, see Web Appendix H).
Afterward, participants were told that a pet toothpaste brand was testing advertisements for its new product and needed their opinions on two ad versions (adapted from Wang and Lee [2006]). Corresponding to their assigned pet exposure condition, participants in the dog (cat) condition viewed dog (cat) toothpaste ads. Ad A, which featured a promotion-focused claim, read, "Our product helps your dog [cat] freshen breath and strengthen tooth enamel!" Ad B, which emphasized the prevention-focused benefits of the product, read, "Our product helps your dog [cat] prevent gingivitis and fight plaque buildup!" A separate pretest confirmed that Ad A (B) was indeed perceived as more promotion (prevention) focused (Web Appendix H).
After viewing the two ads, participants indicated their preference for one of the two ads on three seven-point scales (1 = "definitely/for sure/certainly Ad A," and 7 = "definitely/for sure/certainly Ad B"). We created a preference index by averaging participants' responses to the three items (α = .99), with higher scores indicating a preference for Ad B, the prevention-focused version.
As we expected, a one-way ANOVA revealed a significant effect of exposure to pets on ad preference. Specifically, participants in the dog condition indicated a stronger preference for the promotion-focused ad (Mdog = 4.10, SD = 2.06) than those in the cat condition (Mcat = 4.85, SD = 1.83; F( 1, 181) = 6.78, p = .010, = .04).
Study 3b aims to examine our prediction using an incentive-compatible behavior in a pet-unrelated domain. One hundred eighty MTurk workers completed the study, which featured a two-cell (pet exposure: dog vs. cat) between-subjects design, in exchange for a small financial compensation (Mage = 39.5 years; 60% female). We told participants that the study was about people's general knowledge about pets and their past experiences with pets. We randomly assigned them to one of the two conditions (dog vs. cat). Participants first answered five quiz questions about dogs (cats; see Web Appendix I) and then recalled a past experience interacting with a dog (cat) and wrote it down (following the same instructions as in Study 3a).
We next told participants that they could participate in a lottery and explained the options they had as follows. If they chose not to participate in the lottery, they would still get paid the initial amount ($.40) as described in the study, so there was nothing to lose. If they chose to participate in the lottery, they had a 50% chance to receive a bonus ($.20) in addition to the base pay; however, they also had a 50% chance to lose half the base pay ($.20). A separate pretest confirmed that the lottery participation (nonparticipation) option was indeed perceived as more promotion- (prevention-) focused (Web Appendix I).
Because the promotion (prevention) focus prompts people to focus more on gains (losses) and thus be more risk seeking and open to change (risk averse and status quo oriented) (Liberman et al. 1999; Zhou and Pham 2004), we expected participants who were exposed to dogs to be more likely to participate in the lottery than their counterparts who were exposed to cats. A logistic regression showed a significant effect of pet exposure (cat = 0, dog = 1) on lottery participation, such that participants in the dog condition showed a higher likelihood to take part in the lottery (63.4%) than participants in the cat condition (44.8%; b = .76, SE = .31, χ2 = 6.20, p = .013, Exp (B) = 2.14).
Two hundred twenty-five MTurk workers completed Study 3c in exchange for a small financial compensation (Mage = 38 years; 49% female). The study featured the same two-cell (pet exposure: dog vs. cat) between-subjects design and manipulated pet exposure by asking participants to view a series of four print ads, one per screen, that featured either dogs or cats as the spokescharacter (see Web Appendix J) and to provide their thoughts and feelings after viewing the ads. We then measured participants' regulatory orientation using the same financial decision-making task ([65]) as in Study 2.
A logistic regression showed a marginally significant effect of exposure to pets (cat = 0, dog = 1) on investment choice, such that participants in the dog condition were more likely to choose to invest in the riskier stock option (29.8%) than participants in the cat condition (18.9%; b = .60, SE = .32, χ2 = 3.57, p = .059, Exp(B) = 1.82). Similarly, a one-way ANOVA revealed a significant effect of exposure to pets on money allocation. Participants in the dog condition allocated more money to the stock option (Mdog = $790.35, SD = 514) than those in the cat condition (Mcat = $613.06, SD = 429; F( 1, 223) = 7.86, p = .005, = .03).
One hundred fifty-seven MTurk workers completed Study 3d in exchange for a small financial compensation (Mage = 42 years; 61% female). Study 3d employed the same two-cell (pet exposure: dog vs. cat) between-subjects design. To manipulate pet exposure, participants watched a short video featuring either dogs or cats. Both videos had the same theme—pets "shopping" around in a store (see Web Appendix K).
After participants watched the video, we presented them with a choice scenario. Specifically, we asked them to imagine that they were buying vitamins and that two brands were available (Zhou and Pham 2004). Brand A was rich in vitamin C and iron and could promote high energy. Brand B was rich in antioxidants and could reduce the risk of cancer and heart disease. A separate pretest confirmed that Brand A (Brand B) was perceived as more promotion- (prevention-) focused (see Web Appendix K).
After viewing the two brands, participants then indicated their preference for one of the two brands on three seven-point scales (1 = "definitely/certainly/for sure Brand A," and 7 = "definitely/certainly/for sure Brand B"). We created a preference index by averaging and reverse coding participants' responses to the three items (α = .99; a higher rating indicating a stronger preference for Brand A, the promotion-focused brand).
A one-way ANOVA revealed a significant effect of exposure to pets on brand preference. As expected, participants in the dog condition indicated a stronger preference for the promotion-focused brand (Mdog = 3.96, SD = 2.34) than those in the cat condition (Mcat = 3.19, SD = 2.02; F( 1, 155) = 4.93, p = .028, = .03).
Using a variety of methods to manipulate exposure to pets (i.e., pet knowledge, viewing print ads featuring pets, watching a short pet video, and recalling the experience of interacting with a pet), Studies 3a–3d provide converging support for H1 and show that exposure to dogs can lead to behaviors consistent with a promotion-focused mindset, whereas exposure to cats can prompt behavior patterns more aligned with a prevention-focused mindset. Specifically, in Study 3a, consumers preferred the ad with promotion-focused (prevention-focused) benefits for a dog (cat) toothpaste product. In Studies 3b–3d, we extended this finding to pet-unrelated domains. In Studies 3b and 3c, consumers exposed to dogs (vs. cats) were more willing to take risks in their decisions. In Study 3d, exposure to dogs (vs. cats) prompted consumers to prefer a vitamin brand with promotion-focused benefits. These findings provide converging support for our basic prediction that in both pet-related and pet-unrelated contexts, exposure to dogs can activate more of a promotion-focused mindset, whereas exposure to cats can activate more of a prevention-focused mindset.
Importantly, in Studies 3a–3d, we found no systematic differences between the dog- and cat-exposure conditions in terms of mood, age, gender, ethnicity, income, and pet ownership. In addition, including these variables as control variables does not change our results anyway; thus, we do not discuss these variables further. Although some stimuli used in these studies may not be completely balanced (e.g., the pet pictures used in Study 3c may differ on certain dimensions, and the energy-boosting benefits of the vitamin seem less consequential than the cancer-risk-reducing benefits of the product in Study 3d), these studies taken together show convergent evidence for our main hypothesis and suggest that the proposed effect is robust across different contexts.
The primary purpose of Study 4 is to examine the moderating effect of pet stereotypicality on the activation of regulatory-focus mindsets (H2). We predict that making nonstereotypical information (i.e., pets that do not possess the stereotypical characteristics of their species) available to consumers will attenuate the effect of pet exposure on regulatory-focus mindsets.
Three hundred eighty MTurk participants (57% female; Mage = 40.3 years) completed the study for a small monetary compensation. Study 4 featured a 2 (exposure to pets: dog vs. cat) × 2 (pet stereotypicality: stereotypical vs. nonstereotypical) between-subjects factorial design.
We used a recall task similar to Study 3a. Specifically, we told participants that we were interested in consumers' experience with a pet. Participants then read that some dogs (cats) possess stereotypical characteristics of a dog (cat) and some of them do not. To manipulate pet stereotypicality, in the stereotypical conditions, participants were asked to describe an experience interacting with a dog (cat) that reminds them of the stereotypical characteristics of a dog (cat) (i.e., with personality, temperament, and behavior like a stereotypical dog [cat]). In the nonstereotypical conditions, participants were asked to describe an experience interacting with a dog (cat) that does not have the stereotypical characteristics of a dog (cat) (i.e., with personality, temperament, and behavior unlike a stereotypical dog [cat]). (See Web Appendix L for the detailed manipulation).
After the recall task, participants completed the financial decision scenario used in Study 2 and Study 3c. Specifically, participants imagined they had $2,000 and considered investing in two assets: a stock and a mutual fund. They indicated their preference between the two options on a nine-point scale (1 = "stock," and 9 = "mutual fund"; reverse-coded with a higher rating indicating a stronger preference for the promotion-focused option [i.e., stock]) and then indicated the amount of money they would invest.
A 2 × 2 ANOVA revealed a significant two-way interaction of pet exposure with pet stereotypicality on investment preference (F( 1, 376) = 8.11, p = .005, = .021). Planned contrasts showed that, after pets were described as consistent with their stereotypes, the prior findings were replicated. That is, participants in the dog condition demonstrated higher preference for the stock (Mdog = 4.21, SD = 2.69) than those in the cat condition (Mcat = 3.06, SD = 2.14; F( 1, 376) = 11.72, p <.001, = .03). However, after pets were described as inconsistent with their stereotypes, the prior findings of pet exposure disappeared in that participant did not show preferences for the stock (Mdog = 3.40, SD = 2.47; Mcat = 3.66, SD = 2.36; F( 1, 376) = .53, p = .466).
A 2 × 2 ANOVA on money allocation to the stock also revealed a significant two-way interaction (F( 1, 376) = 4.72, p = .031, = .012). Planned contrasts showed that, after pets were described as consistent with their stereotypes, the prior findings were again replicated such that participants in the dog condition allocated more money to the stock option (Mdog = $765.82, SD = 523.89) than those in the cat condition (Mcat = $623.39, SD = 466.25; F( 1, 376) = 4.20, p = .041, = .011). However, after pets were described as inconsistent with their stereotypes, the prior findings of pet exposure disappeared in that the amount of money allocated to the stock option was not statistically different between the dog and cat conditions (Mdog = $606.80, SD = $477.79; Mcat = $688.04, SD = $527.19; F( 1, 376) = 1.14, p = .286). Thus, the results support H2.
Providing support for our theorizing that associations triggered by pet exposure evoke different regulatory motivational mindsets, Study 4 shows that information related to pet stereotypicality moderates the effect of pet exposure on the activation of regulatory-focus mindsets. Specifically, Study 4 demonstrates that exposing participants to pet information consistent with their stereotype replicated the findings in the previous studies; by contrast, exposure to pets inconsistent with their stereotype nullified the effect of pet exposure on the activation of regulatory-focus mindsets.
Having established the basic effect of pet exposure on consumers' regulatory motivational mindsets, in subsequent studies we aim to further examine the downstream effects of pet exposure on consumer behavior, including product evaluations, purchase intentions, and real incentive-compatible behaviors. Specifically, as we predict in H3, which is based on the regulatory fit between pet exposure and ad frames, because exposure to dogs (cats) activates a promotion (prevention) regulatory mindset among consumers, they should form more favorable evaluations and show more purchase intentions of products framed with promotion-focused (prevention-focused) benefits.
Study 5 aims to provide evidence for H3, which predicts that there is an interaction between pet exposure and the regulatory focus of an ad on consumers' evaluations of the advertised product by using incentive-compatible behaviors. The study also uses a different ad to further augment the robustness of our findings. Two hundred eighty-three undergraduate students from a large midwestern U.S. university participated in the study for partial course credit (45.9% female; Mage = 20.5 years). Study 5 employed a 2 (pet exposure: dog vs. cat) × 2 (regulatory focus: promotion vs. prevention) between-subject factorial design.
Similar to the previous studies, to manipulate pet exposure, under the cover story that we wanted to understand consumers' day-to-day experiences, we first asked participants to recall a past experience in which they interacted with a dog or a cat and to write it down. We then told participants that they would read a message from a local massage center. We varied the message to accentuate either a promotion or a prevention focus (see Web Appendix M). The promotion-focused message emphasized that massages performed by therapists help people increase metabolism, boost immunity, and build a rejuvenated body. The prevention-focused message indicated that massages performed by therapists help soothe body aches, relieve tensions, and reduce stress from school and work. We conducted a separate pretest to confirm the success of our regulatory focus manipulation (see Web Appendix M).
Next, we told participants that the local massage center would offer $50 gift cards to several survey participants. They were asked to bid on the gift cards and were told that the top bidders would be contacted later and offered the gift cards at the bidding price (though later the top bidders received the gift cards for free). Participants were then instructed to write down the dollar amount they were willing to bid on a $50 gift card.
A 2 × 2 ANOVA on participants' bidding amount revealed only a significant interaction between regulatory focus and pet exposure (F( 1, 279) = 8.91, p = .003, = .03). Planned contrasts showed that after exposure to the promotion-focused version of the ad message, participants in the dog condition bid significantly higher (Mdog = $20.31, SD = $14.57) than those in the cat condition (Mcat = $14.98, SD = $13.20; F( 1, 279) = 4.77, p = .030, = .017). By contrast, after exposure to the prevention-focused version of the ad message, participants in the cat condition placed significantly higher bids (Mcat = $20.51, SD = $15.35) than those in the dog condition (Mdog = $15.67, SD = $13.68; F( 1, 279) = 4.14, p = .043, = .02).
Using a behavioral study with an incentive-compatible measure, Study 5 confirmed the robustness of our findings that exposure to pets activates different regulatory mindsets among consumers. After viewing the promotion-focused version of an ad promoting a local massage center, participants who recalled exposure to a dog placed higher bids on the gift card; by contrast, after viewing the prevention-focused version, participants who recalled exposure to a cat placed higher bids. These results provide support for H3.
The goal of Study 6 is threefold. First, Study 6 aims to augment robustness for H3 by conceptually replicating the findings of Study 5 using a different context (bidding for a product). Second, Study 6 aims to test H4, which predicts that regulatory fit will mediate the interaction of exposure to pets with ads' regulatory focus on consumer behavior. Third, it uses a new method to manipulate exposure to pets, such that dogs (cats) are directly incorporated into the stimuli as an integral part of the ad message.
Two hundred sixty-four undergraduate students from a large southeastern U.S. university participated in the study in exchange for course credit (Mage = 20.2 years; 52% female). Study 6 featured a 2 (pet exposure: dog vs. cat) × 2 (regulatory focus: promotion vs. prevention) between-subjects factorial design.
The experimental procedure was similar to Study 5 except that a new ad with a new product (sneaker) was employed and that pets (dogs or cats) were referenced in the ad. We told participants that they would review a message from a sneaker brand. Dependent on the assigned condition, participants next viewed one of the four versions of the ad (adapted from [22]]). The promotion-focused version of the ad read, "Be a dog (cat) person! Reach your health goal with eagerness. Our sneakers feature H-Ergy synthetic material, which improves breathability of the shoes and promotes strong support for your feet." The prevention-focused version of the ad read, "Be a dog (cat) person! Reach your health goal with caution. Our sneakers feature N-Ergy synthetic material, which is anti-skid and reduces the possibility of foot pain." We conducted a separate pretest to confirm the success of our regulatory focus manipulation and the believability of the stimuli.
Two hundred sixty undergraduate students (Mage = 20.3 years; 56% female) were randomly assigned to one of the four conditions. Participants first indicated the extent to which the advertised sneakers had benefits that could help people attain something positive and the extent to which the advertised sneakers had benefits that could help people avoid something negative (adapted from Mogilner, Aaker, and Pennington [2008]; 1 = "strongly disagree," and 9 = "strongly agree"). Participants then rated the extent to which the message was reasonable/appropriate/believable as an ad (1 = "strongly disagree," and 9 = "strongly agree"; α = .91; averaged to form a believability index). The ANOVA on manipulation check measures revealed only main effects, such that the promotion-focused message (Mpromotion = 7.53, SD = 1.19) was deemed as having benefits that helped people attain something positive to a greater extent than the prevention-focused message (Mprevention = 6.71, SD = 1.51; F( 1, 256) = 23.76, p <.001, = .09), and that the prevention-focused message (Mprevention = 6.46, SD = 1.93) was perceived as having benefits that helped people avoid something negative to a greater extent than the promotion-focused message (Mpromotion = 5.56, SD = 2.51; F( 1, 256) = 10.58, p = .001, = .04). A one-sample t-test on the believability index revealed a significant difference against the mid-point of the scale (5; M = 5.40; t(259) = 3.10, p = .002; d = .19) such that participants perceived the ad message they viewed as believable. There was no significant difference on the believability index across conditions (F = .03, n.s.).
Next, after revealing that the suggested retail price for the sneakers was $50, we asked participants to bid on the advertised sneakers and told them that the top bidders would be offered the sneakers based on the price they bid. After entering the bidding amount, participants then responded to regulatory fit measures (Lee and Aaker 2004) on two nine-point scales ("It was easy to process the message" and "It was difficult to understand the message (reverse coded)"; r = .84)
A 2 × 2 ANOVA on participants' bidding amount revealed only a significant interaction between regulatory focus and pet exposure (F( 1, 260) = 8.38, p = .004, = .03). Planned contrasts showed that after exposure to the promotion-focused version of the ad message, participants in the dog condition bid significantly higher (Mdog= $33.74, SD = $11.81) than those in the cat condition (Mcat = $28.23, SD = $15.61; F( 1, 260) = 5.49, p = .020, = .02). By contrast, after exposure to the prevention-focused version of the ad message, participants in the cat condition placed significantly higher bids (Mcat = $32.45, SD = $13.61) than those in the dog condition (Mdog = $28.40, SD = $12.66; F( 1, 260) = 3.05, p = .082, = .012).
A 2 × 2 ANOVA on regulatory fit revealed only a significant interaction between exposure to pets and regulatory focus (F( 1, 260) = 9.80, p = .002, = .036). Specifically, planned contrasts showed that for the promotion-focused ad, the dog version elicited higher regulatory fit (Mdog = 5.21, SD = 2.53) than the cat version (Mcat = 4.19, SD = 2.37; F( 1, 260) = 5.37, p = .021, = .02). By contrast, for the prevention-focused ad, the cat version elicited higher regulatory fit (Mcat = 5.15, SD = 2.40) than the dog version (Mdog = 4.24, SD = 2.62; F( 1, 260) = 4.45, p = .036, = .017).
To test the mediation prediction in H4, we conducted a moderated mediation analysis using 5,000 bootstrapped samples (PROCESS Model 8; [24]), with exposure to pets as the independent variable, regulatory fit as the mediator, regulatory focus as the moderator, and bidding amount as the dependent variable. The index of moderated mediation was significant (b = 1.68, 95% confidence interval [CI]: [.30, 3.70]). Specifically, for the promotion-focused ad, the conditional indirect effect of pet exposure on bidding amount through regulatory fit was positive and significant (b = .89, 95% CI: [.05, 2.11]). By contrast, for the prevention-focused ad, the conditional indirect effect of pet exposure on bidding amount through regulatory fit was negative and significant (b = −.80, 95% CI: [−2.05, −.003]). Thus, the data support H4.
Using a new method (featuring a dog/cat as an integral part of the ad) to manipulate exposure to pets, we provide further evidence for our theorizing of an interactive effect between exposure to pets and regulatory focus of an ad on consumer responses. For sneaker ads featuring promotion-focused (prevention-focused) claims, consumers exposed to dogs (cats) formed higher bidding amount than those exposed to cats (dogs). Importantly, the findings of Study 6 also provide support for H4, such that the influence of exposure to pets on bidding amount was mediated by the regulatory fit between the activated regulatory mindset and the regulatory focus of the ad claim. We also conducted an additional study to conceptually replicate this study in the financial decision-making context (for details, see Web Appendix P).
Study 7 has two objectives. First, it aims to lend additional support to H3 and H4 using a within-subject design. Second, to augment the robustness of our findings, we employ a different product category (toothpaste) and a new manipulation of pet exposure (pet pictures).
Two hundred thirty-seven undergraduate students from a large southeastern U.S. university completed the study in exchange for partial course credit (Mage = 20 years; 45% female). The study featured a 2 (pet exposure: dog vs. cat) × 2 (products' regulatory focus: promotion- vs. prevention-focused) mixed ANOVA design, with pet exposure a between-subjects factor and products' regulatory focus a within-subjects factor.
Participants were randomly assigned to one of two conditions (pet exposure: dog vs. cat). We told participants that an online calendar company was interested in people's feedback on several dog (cat) pictures that it planned to incorporate into a dog- (cat-) themed calendar. Participants then were shown a series of dog (cat) pictures, one on each screen (order counterbalanced). Afterward, participants were again shown all the pictures they had seen on one screen and were instructed to pick one of the pets to imagine interacting with. (For the stimuli used, see Web Appendix N.)
Next, in an ostensibly different task, we presented participants with the descriptions of two toothpaste products (Wang and Lee 2006): Toothpaste A had strong promotion but weak prevention product claims, and Toothpaste B had strong prevention but weak promotion product claims (see Web Appendix O). We counterbalanced the order of the two toothpaste products across all participants. We then asked participants to evaluate each of the two products on four nine-point scales (1 = "dislike very much/very unfavorable/very unattractive/very bad," and 9 = "like very much/very favorable/very attractive/very good").
Afterward, we presented participants with all the strong feature claims and asked them to evaluate each of the features on a nine-point scale (1 = "not at all attractive," and 9 = "very attractive"). The features were evaluated as generic features of the product category, rather than as the features of a specific brand, and served as measures of regulatory fit (Wang and Lee 2006). That is, if participants relied more on (promotion or prevention) features that fit their regulatory orientations in their evaluation, they should find strong feature claims consistent with their regulatory orientations more attractive than claims that do not fit their orientations.
We averaged participants' evaluations of the toothpaste products on the four items to form a brand attitude index for each product (αA = .94, αB = .95). We expected that participants assigned to the dog exposure condition would evaluate Toothpaste A (with the strong promotion claims) more favorably than those assigned to the cat exposure condition. By contrast, participants assigned to the cat exposure condition would evaluate Toothpaste B (with the strong prevention claims) more favorably than those assigned to the dog exposure condition. A mixed ANOVA with toothpaste attitudes as the within-subject variable and pet exposure as the between-subjects variable first revealed a significant main effect of toothpaste attitudes (F( 1, 235) = 4.47, p = .036, = .02), such that, overall, participants evaluated Toothpaste A (with the strong promotion claims; M = 6.94) more positively than Toothpaste B (with the strong prevention claims; M = 6.61). Importantly, the predicted interaction also emerged (F( 1, 235) = 29.76, p <.001, = .11). We found that participants who were exposed to dogs evaluated Toothpaste A more positively (M = 7.41, SD = 1.25) than Toothpaste B (M = 6.29, SD = 1.74, t(235) = 5.74, p <.001; d = .52). By contrast, participants who were exposed to cats formed more positive attitudes toward Toothpaste B (M = 6.94, SD = 1.63) than Toothpaste A (M = 6.45, SD = 1.73, t(235) = −2.21, p = .029; d = –.21).
We then analyzed participants' feature attractiveness ratings, which served as our regulatory fit measure. Because both feature type and toothpaste were measured within subject, we calculated a relative feature attractiveness index by dividing participants' attractiveness ratings of promotion features by their ratings of prevention features. With the index as the dependent variable, a one-way ANOVA revealed that participants in the dog exposure condition (Mdog = 1.27, SD = .83) perceived the promotion features as more attractive than participants in the cat exposure condition (Mcat = 1.09, SD = .35; F( 1, 235) = 4.66, p = .032, = .02).
To further investigate the mediating role of regulatory fit, we conducted mediation analyses to examine whether perceived attractiveness of the product features mediated the effect of pet exposure on product evaluation. Given the within-subjects design, we used MEMORE macro developed by Montoya and Hayes (2017). With 5,000 bootstrapped samples, the analysis revealed a significant indirect effect of feature attractiveness (b = .41, SE = .09, 95% CI: [.23,.59]), indicating that regulatory fit mediated the effect of pet exposure on product evaluation.
Using a within-subject design and a new product category for the dependent variable, Study 7 conceptually replicates the previous studies. Moreover, it provides additional support for H3 in that exposure to pets interacted with an ad's regulatory focus to influence consumer responses and for H4 in that such an effect is driven by regulatory fit.
Across 11 studies, employing different methods (secondary data, lab experiments, and real behavior), different operationalizations of pet exposure (pet ownership; viewing pet-featuring pictures, videos, or ads; and recalled experience with pets), and different measures and contexts of regulatory focus (e.g., financial, service, consumer products), we find converging evidence that pet exposure influences consumer judgments and behaviors through regulatory focus and regulatory fit. Specifically, we show that pet exposure fosters divergent regulatory orientations among consumers, such that exposure to dogs activates a promotion-focused motivational mindset while exposure to cats activates a prevention-focused motivational mindset (Studies 1–3). In Study 4, we further show that this effect is moderated by pet stereotypicality, such that the effects of pet exposure on regulatory mindsets dissipate when consumers are reminded of a pet that is inconsistent with the stereotypes of the species. These regulatory mindsets, when activated by pet exposure, carry over to influence downstream pet-unrelated consumer judgments, purchase decisions, and behaviors through regulatory fit (Studies 5–7), even in pet-unrelated consumption contexts.
Our research contributes to the literature in several ways. First, it recognizes pets as an important source of social influence on consumers' judgments and behaviors (Cavanaugh, Leonard, and Scammon 2008; Hirschman 1994; Holbrook and Woodside 2008). Prior research on marketplace social influence has examined the contexts of consumer-to-consumer and marketer-to-consumer interactions (e.g., Argo, White, and Dahl 2006; Chan and Sengupta 2010; Duclos, Wan, and Jiang 2013; Lee and Shrum 2012; Mead et al. 2011; White and Argo 2011; White and Dahl 2006, [75]). As examples of consumer-to-consumer interactions, Duclos, Wan, and Jiang (2013) show that social exclusion prompts riskier financial decisions because interpersonal rejection heightens the instrumentality of money to obtain benefits in life; Lee and Shrum (2012) find that social exclusion can lead to conspicuous consumption or prosocial behavior. As an example of marketer-to-consumer interactions, Chan and Sengupta (2010) find that consumers who receive insincere flattery from marketers still form favorable implicit evaluations of the marketer, but their explicit evaluations of the marketer are negative. Beyond these human-to-human contexts, our research suggests that the context of pet–human interactions as a source of social influence can similarly affect consumers' motivations, judgments, and behaviors.
Second, our research diverges from extant literature on human–animal relationships by going beyond the immediate context of pet ownership and investigating how pet exposure affects the way people render subsequent pet-unrelated judgments and decisions. Prior research on the effects of human–animal relationships has mainly focused on the immediate context of pets and their owners, such as pet owners' health and psychological well-being ([ 2]), their awareness and protection of animal rights (Kidd, Kidd, and Zasloff 1995), and their relationship satisfaction with pets (Cavanaugh, Leonard, and Scammon 2008). Going beyond this immediate context between pets and their owners, we posit and find that exposure to pets or pet-featuring stimuli fosters divergent regulatory orientations, which in turn influence downstream pet-unrelated judgments and decisions in the consumption domain. We hope that this research will stimulate further research to gain a more nuanced understanding of the impact of human–animal relationships.
Third, prior research identifies several antecedent variables of regulatory focus, such as one's cultural background (Lee, Aaker, and Gardner 2000), impulse purchase ([65]), and choosing for oneself versus others ([59]). Most of these factors, however, are intrapersonal. Only recently have researchers begun exploring the interpersonal drivers of regulatory focus (e.g., social exclusion [Park and Baumeister 2015], reminders of friends vs. family members [Fei, You, and Yang 2020]). Contributing to this stream of research, we uncover a novel social source of regulatory orientation: exposure to pets. We find that dog exposure is associated with a promotion orientation, whereas cat exposure is associated with a prevention orientation. Additional research is necessary to further explore the social and interpersonal impact on people's regulatory orientation.
Our findings also offer novel implications to marketers. First, marketers should consider crafting their advertising messages differently or recommending different products and services when they target consumers depending on their pet exposure situations. For example, to enhance the effectiveness of their advertising appeals or communication messages, marketers should emphasize promotion-focused goals such as gains and nongains if they are targeting dog owners or after consumers are exposed to dogs or dog-featuring stimuli (e.g., after just watching an ad about dogs). Conversely, they should focus on prevention-focused goals such as losses and nonlosses if they are pursuing cat owners or after consumers who are exposed to cats or cat-featuring stimuli. Importantly, our findings show that this advice holds even when the advertised product or service has nothing to do with pets or pet products.
Second, our findings offer important insights into how to incorporate pets into marketing communications. Dogs and cats frequently appear in advertisements and marketing campaigns. For example, Subaru has been running the "Dog Tested, Dog Approved" campaign since 2009 (Subaru 2020), and Sainsbury's "Mog the Cat" campaign raised a great amount of attention during the 2015 Christmas season ([25]). One consideration factor, according to our findings, is the type of products or services being advertised. For products or services mainly perceived as promotion-focused (e.g., stock investment, sports cars), featuring dogs in the ad is likely to increase the ad's persuasiveness. For products or services deemed more prevention-focused (e.g., mutual fund investment, insurance), featuring cats may increase the ad's appeal. According to the findings of the pet stereotypicality study, a caveat is that marketers should ensure that stereotypical pet temperaments are made salient in the message (e.g., the eagerness [cautiousness] aspect of the dog [cat] should be highlighted). Otherwise, the intended effects of featuring pets in the ad may not be achieved.
Third, pet-related marketing strategies are especially relevant in today's big-data era, in which marketers likely know more about consumers, including which types of pets they own or interact with. For example, marketers could obtain consumers' pet exposure information from the pet-related products purchased, the YouTube pet videos watched, or the Instagram pet photos posted and use this data to decide what type of marketing information to highlight. Alternatively, census information can provide marketers with an aggregated level of information about pets, as certain states or zip codes may have a higher concentration of dog (cat) owners. Marketers could use this information to determine the design of regional marketing campaigns.
Lastly, our findings that pets and pet ownership are potentially related to COVID-19 transmission rate and prevention behaviors could shed new light on policies related to prevention of COVID-19 and potentially other infectious diseases. For example, policy makers in states with more dog owners could design more customized educational programs and materials related to the diseases. Alternatively, when designing ads to prevent the transmission of COVID-19 and other infectious diseases, cats could be incorporated as a spokesperson and/or their temperaments can be referenced in the message to enhance the effectiveness of the ad.
This research indicates that pets can exert a strong social influence on consumers. Future research could examine the strength of the influence of pets versus people on consumer behavior. On the one hand, one may argue that pets' influence on consumers is weaker than other human influences, because people are more influenced by similar others ([ 3]; Bandura 2002). On the other hand, it is also possible that pets may occasionally exert a stronger influence, especially considering that many pet owners admit that they prefer spending time with their pets over other people ([56]).
Another possible research direction is to examine other differences between dog and cat exposure. For example, exposure to dogs may be more likely to lead to conspicuous consumption than cat exposure. This follows because many people likely perceive dogs (vs. cats) as less seclusive and more social. Along similar lines, research could examine the different effects of dog versus cat stimuli. For example, using a dog (cat) as a spokescharacter can increase a brand's perceived excitement. Although our research focuses on examining the effect of pet exposure on consumers' regulatory orientations, future research could investigate the reverse relationship of whether promotion- (prevention-) focused consumers are more likely to adopt a dog (cat) or prefer dog- (cat-) related stimuli.
Research could also examine additional moderators for our findings. One example is pet anthropomorphism, or people's tendency to assign human characteristics to pets ([ 2]). One prediction is that our findings that dog (cat) exposure evokes a promotion (prevention) orientation would be more salient among consumers who engage in pet anthropomorphism, because these consumers are more likely to be influenced by the pets. Another possible moderator is the role of culture. We conducted our studies primarily with American participants, and the United States is a culture in which the majority of people treat their pets as friends and family members (Sanders and Hirschman 1996). In many other cultures however, pets are not elevated to the same level, and thus consumers may treat their pets as possessions or servants ([ 9]). Future research could examine whether our identified findings still hold in such cultures.
sj-pdf-1-jmx-10.1177_00222429221078036 - Supplemental material for The Pet Exposure Effect: Exploring the Differential Impact of Dogs Versus Cats on Consumer Mindsets
Supplemental material, sj-pdf-1-jmx-10.1177_00222429221078036 for The Pet Exposure Effect: Exploring the Differential Impact of Dogs Versus Cats on Consumer Mindsets by Lei Jia, Xiaojing Yang and Yuwei Jiang in Journal of Marketing
Footnotes 1 Karen Winterich
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by a research grant from the Darla Moore School of Business at University of South Carolina awarded to the second author, and research grants from the Hong Kong Research Grants Council (PolyU 155008/21B) and the Asian Centre for Branding and Marketing to the third author.
4 Lei Jia https://orcid.org/0000-0002-7727-9316 Xiaojing Yang https://orcid.org/0000-0001-9337-907X
5 Supplemental material for this article is available online.
References American Veterinary Medical Association (2018), US Pet Ownership & Demographics Sourcebook. Schaumburg , IL : American Veterinary Medical Association.
Amiot Catherine E. , Bastian Brock. (2015), " Toward a Psychology of Human–Animal Relations ," Psychological Bulletin , 141 (1), 6 – 47.
Andsager Julie L. , Bemker Victoria , Choi Hong-Lim , Torwel Vitalis. (2006), " Perceived Similarity of Exemplar Traits and Behavior: Effects on Message Evaluation ," Communication Research , 33 (1), 3 – 18.
APPA (2018), APPA National Pet Owners Survey 2017-2018. Greenwich , CT : APPA.
Argo Jennifer J. , White Katherine , Dahl Darren W.. (2006), " Social Comparison Theory and Deception in the Interpersonal Exchange of Consumption Information ," Journal of Consumer Research , 33 (1), 99 – 108.
6 ASPCA (2021), "New ASPCA Survey Shows Overwhelming Majority of Dogs and Cats Acquired During the Pandemic Are Still in Their Homes," press release (May 26), https://www.aspca.org/about-us/press-releases/new-aspca-survey-shows-overwhelming-majority-dogs-and-cats-acquired-during.
7 Avnet Tamar , Higgins E. Tory. (2006), " How Regulatory Fit Affects Value in Consumer Choices and Opinions ," Journal of Marketing Research , 43 (1), 1 – 10.
8 Bandura Albert. (2002), " Social Cognitive Theory in Cultural Context ," Applied Psychology , 51 (2), 269 – 90.
9 Beverland Michael B. , Farrelly Francis , Lim Elison Ai Ching. (2008), " Exploring the Dark Side of Pet Ownership: Status- and Control-Based Pet Consumption ," Journal of Business Research , 61 (5), 490 – 96.
Boldero Jennifer M. , Higgins E. Tory. (2011), " Regulatory Focus and Political Decision Making: When People Favor Reform Over the Status Quo ," Political Psychology , 32 (3), 399 – 418.
Bradshaw John. (2012), Dog Sense: How the New Science of Dog Behavior Can Make You a Better Friend to Your Pet. New York : Basic Books.
Bradshaw John. (2013), Cat Sense: How the New Feline Science Can Make You a Better Friend to Your Pet. New York : Basic Books.
Cavanaugh Lisa A. , Leonard Hillary A. , Scammon Debra L.. (2008), " A Tail of Two Personalities: How Canine Companions Shape Relationships and Well-Being ," Journal of Business Research , 61 (5), 469 – 79.
CDC (2020), "CDC COVID Data Tracker," (accessed November 1, 2020), https://covid.cdc.gov/covid-data-tracker/#cases_casesper100k.
Cesario Joseph , Grant Heidi , Higgins E. Tory. (2004), " Regulatory Fit and Persuasion: Transfer from 'Feeling Right,' " Journal of Personality and Social Psychology , 86 (3), 388 – 404.
Chan Elaine , Sengupta Jaideep. (2010), " Insincere Flattery Actually Works: A Dual Attitudes Perspective ," Journal of Marketing Research , 47 (1), 122 – 33.
Chernev Alexander. (2004), " Goal Orientation and Consumer Preference for the Status Quo ," Journal of Consumer Research , 31 (3), 557 – 65.
Crowe Ellen , Higgins E. Tory. (1997), " Regulatory Focus and Strategic Inclinations: Promotion and Prevention in Decision-Making ," Organizational Behavior and Human Decision Processes , 69 (2), 117 – 32.
Du Rex Yuxing , Hu Ye , Damangir Sina. (2015), " Leveraging Trends in Online Searches for Product Features in Market Response Modeling ," Journal of Marketing , 79 (1), 29 – 43.
Duclos Rod , Wan Echo Wen , Jiang Yuwei. (2013), " Show Me the Honey! Effects of Social Exclusion on Financial Risk-Taking ," Journal of Consumer Research , 40 (1), 122 – 35.
Farand Chloe. (2016), "Which Love Us More – Cats or Dogs?" The Independent (January 31), https://www.independent.co.uk/news/uk/bbc-documentary-answers-age-old-question-which-love-us-more-cats-or-dogs-a6844846.html.
Fei Xianzheng , You Yanfen , Yang Xiaojing. (2020), " 'We' Are Different: Exploring the Diverse Effects of Friend and Family Accessibility on Consumers' Product Preferences ," Journal of Consumer Psychology , 39 (3), 543 – 50.
Gorn Gerald J. , Jiang Yuwei , Johar Gita Venkataramani. (2008), " Babyfaces, Trait Inferences, and Company Evaluations in a Public Relations Crisis ," Journal of Consumer Research , 35 (1), 36 – 49.
Hayes Andrew F.. (2018), Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach. New York : Guilford Press.
Hendriksz Vivian. (2015), "Sainsbury's 'Mog the Cat' Wins the Christmas 2015 Advert Battle," FashionUnited (December 23), https://fashionunited.uk/news/fashion/sainsbury-s-mog-the-cat-wins-the-christmas-2015-advert-battle/2015122318828.
Higgins E. Tory. (1987), " Self-Discrepancy: A Theory Relating Self and Affect ," Psychological Review , 94 (3), 319 – 40.
Higgins E. Tory. (1996), " The 'Self Digest': Self-Knowledge Serving Self-Regulatory Functions ," Journal of Personality and Social Psychology , 71 (6), 1062 – 83.
Higgins E. Tory. (1997), " Beyond Pleasure and Pain ," American Psychologist , 52 (12), 1280 –1 300.
Higgins E. Tory. (2000), " Making a Good Decision: Value from Fit ," American Psychologist , 55 (11), 1217 – 30.
Higgins E. Tory. (2002), " How Self-Regulation Creates Distinct Values: The Case of Promotion and Prevention Decision Making ," Journal of Consumer Psychology , 12 (3), 177 – 91.
Hirschman Elizabeth C.. (1994), " Consumers and Their Animal Companions ," Journal of Consumer Research , 20 (4), 616 – 32.
Holbrook Morris B. , Woodside Arch G.. (2008), " Animal Companions, Consumption Experiences, and the Marketing of Pets: Transcending Boundaries in the Animal-Human Distinction ," Journal of Business Research , 61 (5), 377 – 81.
Hong Jiewen , Lee Angela Y.. (2007), " Be Fit and Be Strong: Mastering Self-Regulation Through Regulatory Fit ," Journal of Consumer Research , 34 (5), 682 – 95.
Jardim-Messeder Débora , Lambert Kelly , Noctor Stephen , Pestana Fernanda M. , de Castro Leal Maria E. , Bertelsen Mads F. , et al. (2017), " Dogs Have the Most Neurons, Though Not the Largest Brain: Trade-Off Between Body Mass and Number of Neurons in the Cerebral Cortex of Large Carnivoran Species ," Frontiers in Neuroanatomy , 11 , https://doi.org/10.3389/fnana.2017.00118.
Kidd Aline H. , Kidd Robert M. , Zasloff R. Lee. (1995), " Developmental Factors in Positive Attitudes Toward Zoo Animals ," Psychological Reports , 76 (1), 71 – 81.
Kozinets Robert , Patterson Anthony , Ashman Rachel. (2017), " Networks of Desire: How Technology Increases Our Passion to Consume ," Journal of Consumer Research , 43 (5), 659 – 83.
Labroo Aparna A. , Lee Angela Y.. (2006), " Between Two Brands: A Goal Fluency Account of Brand Evaluation ," Journal of Marketing Research , 43 (3), 374 – 85.
Langenfeld Shane. (2020), "Why Dogs Adapt to Their Environment," Wag! (accessed December 15, 2020), https://wagwalking.com/behavior/why-dogs-adapt-to-their-environment.
Lee Angela Y. , Aaker Jennifer L.. (2004), " Bringing the Frame into Focus: The Influence of Regulatory Fit on Processing Fluency and Persuasion ," Journal of Personality and Social Psychology , 86 (2), 205 – 18.
Lee Angela Y. , Aaker Jennifer L. , Gardner Wendi L.. (2000), " The Pleasures and Pains of Distinct Self-Construals: The Role of Interdependence in Regulatory Focus ," Journal of Personality and Social Psychology , 78 (6), 1122 – 34.
Lee Angela Y. , Keller Punam Anand , Sternthal Brian. (2010), " Value from Regulatory Construal Fit: The Persuasive Impact of Fit Between Consumer Goals and Message Concreteness ," Journal of Consumer Research , 36 (5), 735 – 47.
Lee Jaehoon , Shrum L.J.. (2012), " Conspicuous Consumption Versus Charitable Behavior in Response to Social Exclusion: A Differential Needs Explanation ," Journal of Consumer Research , 39 (3), 530 – 44.
Liberman Nira , Idson Lorraine Chen , Camacho Christopher J. , Higgins E. Tory. (1999), " Promotion and Prevention Choices Between Stability and Change ," Journal of Personality and Social Psychology , 77 (6), 1135 – 45.
Lockwood Penelope , Jordan Christian H. , Kunda Ziva. (2002), " Motivation by Positive or Negative Role Models: Regulatory Focus Determines Who Will Best Inspire Us ," Journal of Personality and Social Psychology , 83 (4), 854 – 64.
Ma Jingjing , Roese Neal J.. (2014), " The Maximizing Mind-Set ," Journal of Consumer Research , 41 (1), 71 – 92.
MacInnis Cara C. , Hodson Gordon. (2015), " Do American States with More Religious or Conservative Populations Search More for Sexual Content on Google? " Archives of Sexual Behavior , 44 (1), 137 – 47.
Mead Nicole L. , Baumeister Roy F. , Stillman Tyler F. , Rawn Catherine D. , Vohs Kathleen D.. (2011), " Social Exclusion Causes People to Spend and Consume Strategically in the Service of Affiliation ," Journal of Consumer Research , 37 (5), 902 – 19.
Mellon Jonathan. (2013), " Where and When Can We Use Google Trends to Measure Issue Salience? " Political Science & Politics , 46 (2), 280 – 90.
Miklósi Ádám , Pongrácz Péter , Lakatos Gabriella , Topál József , Csányi Vilmos. (2005), " A Comparative Study of the Use of Visual Communicative Signals in Interactions Between Dogs (Canis familiaris) and Humans and Cats (Felis catus) and Humans ," Journal of Comparative Psychology , 119 (2), 179 – 86.
Mogilner Cassie , Aaker Jennifer L. , Pennington Ginger L.. (2008), " Time Will Tell: The Distant Appeal of Promotion and Imminent Appeal of Prevention ," Journal of Consumer Research , 34 (5), 670 – 81.
Montoya Amanda K. , Hayes Andrew F.. (2017), " Two-Condition Within-Participant Statistical Mediation Analysis: A Path-Analytic Framework ," Psychological Methods , 22 (1), 6 – 27.
Myers Eugene Olin. (1999), " Human Development as Transcendence of the Animal Body and the Child-Animal Association in Psychological Thought ," Society & Animals , 7 (2), 121 – 40.
Nagasawa Miho , Mitsui Shouhei , En Shiori , Ohtani Nobuyo , Ohta Mitsuaki , Sakuma Yasuo , et al. (2015), " Oxytocin-Gaze Positive Loop and the Coevolution of Human-Dog Bonds ," Science , 348 (6232), 333 – 36.
The New York Times (2017), "2016 Presidential Election Results," (accessed November 1, 2020), https://www.nytimes.com/elections/2016/results/president.
Park Jina , Baumeister Roy F.. (2015), " Social Exclusion Causes a Shift Toward Prevention Motivation ," Journal of Experimental Social Psychology , 56 , 153 – 9.
Pasquini Maria. (2019), "Over 70 Percent of Dog Owners Admit They Prefer Spending Time with Their Pets over Other People," People (August 26), https://people.com/pets/over-70-percent-of-dog-owners-admit-they-prefer-spending-time-with-their-pets-over-other-people/.
Pham Michel Tuan , Avnet Tamar. (2004), " Ideals and Oughts and the Reliance on Affect Versus Substance in Persuasion ," Journal of Consumer Research , 30 (4), 503 – 18.
Pham Michel Tuan , Chang Hannah H.. (2010), " Regulatory Focus, Regulatory Fit, and the Search and Consideration of Choice Alternatives ," Journal of Consumer Research , 37 (4), 626 – 40.
Polman Evan. (2012), " Self–Other Decision Making and Loss Aversion ," Organizational Behavior and Human Decision Processes , 119 (2), 141 – 50.
Potter Alice , Mills Daniel Simon. (2015), " Domestic Cats (Felis silvestris catus) Do Not Show Signs of Secure Attachment to Their Owners ," PLoS One , 10 (9), e0135109.
Purewal Rebecca , Christley Robert , Kordas Katarzyna , Joinson Carol , Meints Kerstin , Gee Nancy , et al. (2017), " Companion Animals and Child/Adolescent Development: A Systematic Review of the Evidence ," International Journal of Environmental Research and Public Health , 14 (3), 234.
Sanders Clinton R. , Hirschman Elizabeth C.. (1996), " Guest Editors' Introduction: Involvement with Animals as Consumer Experience ," Society & Animals , 4 (2), 111 – 19.
Scholer Abigail A. , Cornwell James F.M. , Higgins E. Tory. (2019), " Regulatory Focus Theory and Research: Catching up and Looking Forward After 20 Years ," in Oxford Handbook of Human Motivation , Vol. 2 , Ryan Richard M. , ed. New York : Oxford University Press.
Schultheiß Sebastian , Lewandowski Dirk. (2021), " How Users' Knowledge of Advertisements Influences Their Viewing and Selection Behavior in Search Engines ," Journal of the Association for Information Science and Technology , 72 (3), 285 – 301.
Sengupta Jaideep , Zhou Rongrong. (2007), " Understanding Impulsive Eaters' Choice Behaviors: The Motivational Influences of Regulatory Focus ," Journal of Marketing Research , 44 (2), 297 – 308.
Serpell James A. , Paul Elizabeth S.. (2011), " Pets in the Family: An Evolutionary Perspective ," in The Oxford Handbook of Evolutionary Family Psychology , Salmon C.A. , Shackelford T.K. , eds. New York : Oxford University Press , 298 – 309.
Subaru (2020), "Dog Tested. Dog Approved," Subaru (accessed December 10, 2020), https://www.subaru.com/pets/our-videos.html.
U.S. Bureau of Economic Analysis (2019), "Per Capita Real GDP by State," U.S. Department of Commerce (accessed November 1, 2020), https://apps.bea.gov/iTable/index_nipa.cfm.
U.S. Bureau of the Census (2017), "Household Income: 2016," American Community Survey Briefs.
Wang Jing , Lee Angela Y.. (2006), " The Role of Regulatory Focus in Preference Construction ," Journal of Marketing Research , 43 (1), 28 – 38.
Watson David , Clark Lee Anna , Tellegen Auke. (1988), " Development and Validation of Brief Measures of Positive and Negative Affect: The PANAS Scales ," Journal of Personality and Social Psychology , 54 (6), 1063 – 70.
Weeks Brian , Southwell Brian. (2010), " The Symbiosis of News Coverage and Aggregate Online Search Behavior: Obama, Rumors, and Presidential Politics ," Mass Communication and Society , 13 (4), 341 – 60.
White Katherine , Argo Jennifer J.. (2011), " When Imitation Doesn't Flatter: The Role of Consumer Distinctiveness in Responses to Mimicry ," Journal of Consumer Research , 38 (4), 667 – 80.
White Katherine , Dahl Darren W.. (2006), " To Be or Not Be? The Influence of Dissociative Reference Groups on Consumer Preferences ," Journal of Consumer Psychology , 16 (4), 404 – 14.
White Katherine , Dahl Darren W.. (2007), " Are All Out-Groups Created Equal? Consumer Identity and Dissociative Influence ," Journal of Consumer Research , 34 (4), 525 – 36.
Wyer Robert S. Jr., , Xu Alison Jing. (2010), " The Role of Behavioral Mind-Sets in Goal-Directed Activity: Conceptual Underpinnings and Empirical Evidence ," Journal of Consumer Psychology , 20 (2), 107 – 25.
Wynne Clive D.L. , Udell Monique A.R. , Lord Kathryn A.. (2008), " Ontogeny's Impacts on Human–Dog Communication ," Animal Behavior , 76 (4), e1 – 4.
Xu Alison Jing , Moorman Christine , Qin Vivian Yue , Rao Akshay R.. (2020), " Four More Years: Presidential Elections, Comparative Mindset, and Managerial Decisions ," Academy of Management Journal , 63 (5), 1370 – 94.
Zeifman Debra , Hazan Cindy. (1997), " Attachment: The Bond in Pair-Bonds ," in Evolutionary Psychology , Simpson Jeffry A. , Kenrick Douglas T. , eds. Mahwah , NJ : Lawrence Erlbaum Associates , 237 – 62.
Zhou Rongrong , Pham Michel Tuan. (2004), " Promotion and Prevention Across Mental Accounts: When Financial Products Dictate Consumers' Investment Goals ," Journal of Consumer Research , 31 (1), 125 – 35.
~~~~~~~~
By Lei Jia; Xiaojing Yang and Yuwei Jiang
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 125- The Platformization of Brands. By: Wichmann, Julian R.K.; Wiegand, Nico; Reinartz, Werner J. Journal of Marketing. Jan2022, Vol. 86 Issue 1, p109-131. 23p. 4 Diagrams, 3 Charts. DOI: 10.1177/00222429211054073.
- Database:
- Business Source Complete
The Platformization of Brands
Digital platforms that aggregate products and services, such as Google Shopping or Amazon, have emerged as powerful intermediaries to brand offerings, challenging traditional product brands that have largely lost direct access to consumers. As a countermeasure, several long-established brands have built their own flagship platforms to resume control and foster consumer loyalty. For example, sports brands such as Nike, Adidas, or Asics launched tracking and training platforms that allow for ongoing versatile interactions among participants beyond product purchase. The authors analyze these emerging platform offerings, whose potential brands struggle to exploit, and provide guidance for brands that aim to platformize their business. This guidance comprises the conceptualization of digital platforms as places of consumer crowdsourcing (i.e., consumers drawing value from platform participants such as the brand, other consumers, or third-party businesses) and crowdsending (i.e., consumers providing value to platform participants) of products, services, and content along with a well-defined framework that brands can apply to assemble different types of flagship platforms. Evaluating the consequences of crowdsourcing and crowdsending for consumer–platform relationships, the authors derive a typology of archetypical relationship states and develop a set of propositions to help offline-born product brands thrive through platformization.
Keywords: digital platforms; platform assemblage; building blocks; relationship states; product brands; digital transformation; relationship marketing
Digitization promised established product brands nothing short of their emancipation from the traditional retailing value chain. Online channels made cutting out intermediaries and establishing direct consumer access and relationships both simple and inexpensive ([25]), and many brands willingly embraced this opportunity. However, when new digital aggregators of products and services emerged—in particular, online marketplaces and search engines—they rapidly usurped the interface to consumers, forcing many offline-born product brands back to second rank. For example, Amazon, Google Shopping, and JD.com have become important access points for products from household tools to sports equipment (which we use as a recurring example), leaving brands such as Adidas, Nike, or Asics to resume their places as suppliers to this new group of digital intermediaries.
These intermediaries, which we call "brand aggregation platforms," differ from conventional retailers by relying on a platform business model, where they provide the infrastructure and governance to enable commercial transactions between external suppliers and consumers of branded products while not offering these products themselves ([16]; [77]). For example, searching for Adidas running shoes on Google brings up Google Shopping pages from numerous retailers as well as Adidas's own shop. Google aggregates these offerings while the consumer transacts directly with the brand or an online retailer. As a result, these platforms provide value by granting users access to a vast variety of products and services, enabling them to organize their consumption around a few powerful interfaces ([64]).
This reintermediation through brand aggregation platforms arguably leaves many brands worse off than in the pre–digital retailing era because these platforms diminish brand differentiation and foster price competition by featuring many similar or even identical offerings at different prices from competing suppliers ([25]). For example, a search for sports jerseys on Zalando, a former online apparel retailer turned brand aggregation platform, provides an exhaustive overview of products from brand-owned shops and myriad retailers.
As a countermeasure to this development, some established product brands have started to venture into the platform business themselves ([20]), either by extending their operations organically (e.g., Nike Run Club) or by acquiring existing platforms (e.g., Adidas Runtastic, Asics Runkeeper). This "platformization" of brands creates offerings that transcend the specific product brand by including third-party complementary products, services, and content to occupy the broader category space and address consumer needs more holistically ([89]). For example, Nike features events, expert guidance, exclusive products, motivational music playlists, and even personal training as part of its Run Club and Training Club platforms. These emerging offerings, which we call "brand flagship platforms," may be a potent means for product brands to fend off brand aggregation platforms and establish a direct interface to consumers.
However, brands have a hard time building competitive flagship platforms. While retailers have long embraced platformization (e.g., Amazon, Zalando, Douglas) as a natural evolution of their consumer-focused aggregator approach, product brands lack cross-product expertise, treat the platform as yet another sales channel, or fear the inclusion of competitor offerings, to name just a few impediments. Thus, the platformization of brands is still in its infancy, though some encouraging examples have emerged in such markets as athletics (e.g., Nike, Garmin), do it yourself (DIY; e.g., Bosch), mobility (e.g., BMW and Daimler), nutrition (e.g., Maggi), or gaming (e.g., Epic). For example, Bosch's DIY & Garden allows consumers to register their tools, engage with a versatile community, and receive advice for DIY projects. In mobility, BMW and Daimler have extended their Share Now platform to include ride hailing; renting scooters, cars, and bikes; and finding parking and charging spots. And in nutrition, Maggi offers online cooking classes and events, inspirational content, and recipes that link directly to supermarkets. In other industries, including business-to-business industries, brands gradually acknowledge the potential of building their own digital platforms, but many have difficulty getting started ([19]; [89]).
We address this difficulty by proposing paths for brands to build brand flagship platforms. To this end, we develop a novel conceptualization of digital platforms as places of consumer crowdsourcing and crowdsending, which lie at the core of platformized value creation and which the brand can foster or restrict to shape platform interactions. Consumer crowdsourcing in this context denotes the consumer's assignment of a task (e.g., finding the perfect running shoe, learning how to train for a marathon) to a network of diverse platform participants (i.e., the brand, other consumers, or third-party businesses) that supply products, services, and content, allowing them to draw value from these participants ([33]). Consumer crowdsending denotes the consumer's own contribution of products, services, and content, allowing them to provide value to these platform participants.
Drawing on this conceptualization, this article provides brands with a step-by-step guide to venture into platform business. First, we put forth five key goals that consumers pursue when using digital platforms, enhancing the understanding of why platform offerings are often preferred over retail or brand offerings. These insights help design flagship platforms that align consumer and brand goals. Second, we develop a clear framework that disassembles the platform concept into building blocks that facilitate consumer crowdsourcing and crowdsending. Depending on the assemblage of these building blocks, different platform types emerge that may foster or impede the achievement of consumer and brand goals. Third, we propose the intensities of consumer crowdsourcing and crowdsending as typology dimensions that give rise to distinct consumer–platform relationship states. The typology allows brands to evaluate how their selected assemblage affects consumers' use of and attachment to the brand flagship platform. A set of propositions details the benefits and risks of the emergent relationship states. Finally, we offer several suggestions to rethink managerial practices for brands to master the platform transition.
Our analysis liaises with an existing stream of literature on (digital) platforms and two-sided markets, extending related work in several directions. Above all, prior studies focus on businesses that are "born" as platforms or retailers that expand on their natural aggregation function through platformization. However, how product brands—especially those originating from a nondigital, nonplatformized marketspace—build and use their own platform has been neglected. As a notable exception, [64] raise the idea that platformized brands may pose a threat to traditional retailers in their attempt to reclaim direct consumer access. Yet they neither flesh out the nature of those brand platforms nor analyze how brands can successfully transition to that stage. Furthermore, literature predominantly examines platforms that facilitate commercial exchange. [57] develop a typology of peer-to-peer markets, classifying platforms into discrete types that offer distinct benefits. However, they do not focus on value creation beyond buyer–seller transactions, nor does their model accommodate the possibility that versatile platform assemblages may mix and match functionalities to offer combined benefits. Finally, this study is the first to examine the relational consequences of platformization. While prior work stresses platforms' ability to lower transaction costs, we show that brands can use flagship platforms to build intimate, mutually beneficial relationships based on high levels of self-relevance and affective commitment.
Research and practice have developed a multifaceted understanding of (digital) platforms. However, broad applicability of the platform concept and the myriad platform types and markets that emerged also add complexity, especially for traditional brands, which were largely founded before the advent of internet technologies. To develop the concept of brand flagship platforms, we first briefly define digital platforms in general. We then introduce brand flagship platforms—a specific type of digital platform—by highlighting the opportunities they present to counter and differentiate from brand aggregation platforms.
Platforms provide the infrastructure and governance to facilitate interactions between autonomous agents ([16]; [54]; [77]). That is, platforms link agents on two or more market sides that determine the conditions of their interaction directly (i.e., what to exchange, how to exchange it), maintaining residual control rights over their assets ([29]). If Nike or a designated retailer offers the brand's merchandise to consumers on Amazon or an online comparison site, the seller can typically set prices, control the items and quantities offered, and decide sales promotions and delivery conditions. Platforms therefore assume a mediator role, aiming to provide optimal matches between agents on both market sides often through digital technologies ([57]).
Platform interactions can range from the commercial selling of products and services (e.g., [57]; [63]) to engaging in forums or social communities (e.g., [67]), to posting and consuming media content on video platforms (e.g., [63]). Accordingly, examples are abundant, and these include social media (e.g., Instagram, TikTok), knowledge exchange forums (e.g., StackOverflow, Quora), communities (e.g., Reddit, brand communities), video sharing platforms (e.g., YouTube, Vimeo), advertising platforms (e.g., Google AdWords), service platforms (e.g., Uber, Airbnb), hardware–software platforms (e.g., game consoles), and software–software platforms (e.g., operating systems, app store). The starting point of this analysis is the digital reintermediation of established brands by brand aggregation platforms (e.g., Amazon, Google Shopping, Booking.com), which emerged as powerful interfaces to consumers' product search and purchase activities. Brand flagship platforms may offer a way out of this renewed dependence for product brands that aim to reestablish direct consumer access and foster brand loyalty. We describe these two types of platforms next.
Research has developed a detailed understanding of brand aggregation platforms (e.g., [29]; [54]), on which the interaction between market sides is, at its core, a commercial exchange between buyers and sellers of branded products. Brand aggregation platforms thus serve as intermediaries to discrete transactions that lower search costs and efficiently match customers with product or service offerings ([38]). The transactions are discrete in that purchases do not trigger an ongoing feedback process (other than, for example, dialogues on social media) and customers typically return to enter a new purchase cycle ([71]). For instance, a consumer who needs running shoes might compare different models on JD.com or Wish, purchase on the platform, and return when they need a new pair or other equipment.
A set of distinct features of brand aggregation platforms (see Table 1) poses numerous challenges for product brands, of which we highlight a few. First, their broad, cross-category scope induces consumers to increasingly start their purchase journeys on brand aggregation platforms, often searching for product classes rather than specific brands ([82]). This development puts the platform brand (e.g., Amazon) in the center, threatening to degrade product brands to "white-label back offices" ([ 9], p. 13). Second, contributing to this degradation, brands have difficulty differentiating from competitors because aggregation platforms standardize product presentation, limit the use of branding elements, and encourage simple comparison on a few key features (e.g., price, rating, delivery). Third, product competition occurs not only between brands (e.g., Adidas vs. Asics) but also within brands, as different sellers will offer the identical product at different prices. This aspect represents a strong benefit for customers, making the platform itself agnostic to legitimate brand interests (e.g., preventing price erosion and brand-building efforts) because its primary goal is to provide optimal matches.
Graph
Table 1. Properties of Brand Aggregation and Brand Flagship Platforms.
| Property | Brand Aggregation Platform | Brand Flagship Platform |
|---|
| Definition | Platform-brand-owned digital platform that mediates discrete transactions between buyers and commercial sellers of branded products | Product-brand-owned digital platform that mediates versatile interactions between participants within the brand-related category space |
| Key goals and activities | Facilitating direct selling of products and services | Multiple goals and activities; for example, creating brand awareness, strengthening consumer relationships, building community, fostering consumer learning, selling products and services |
| Commercial target function | Facilitating transactions; largely agnostic to product branding | Directly or indirectly driving brand sales and fostering brand loyalty |
| Category scope | Broad: cross-category generalist | Deep: within-category specialist |
| Activity scope | Narrow: Product-focused aggregator of commercial offerings with ancillary services | Broad: consumer-focused, versatile, and multisided network of products, services, and content |
| Owner | Online-born platform brand Platformized retailer
| Offline-born product brand |
| Role of platform owner | Mediator of commercial transactions | Mediator and supplier of products, services, and content |
| Roles of platform participants | Largely well defined (buyers and sellers) | Loosely defined (participants can assume many different roles) |
| Role of product brand | Competes for limited platform space and addresses similar consumption needs as competing offerings | Orchestrates value-creating activities on the platform |
| Competition | Encourages within-brand and between-brand competition | Aims to avoid within-brand and between-brand competition; focus on complements |
| Inventory | Internal: none or comparatively limited External: extensive assortment of (competing) third-party offerings
| Internal: offers own-branded products Possibly external: offers complements, possibly direct competitor products
|
| User experience | Standardized (category-agnostic); aims for simple product comparison | Individualized (category-specific); aims for optimized consumer category experience |
| Examples | Amazon Marketplace, Google Shopping, Wish, Idealo, JD.com, Alibaba, Zalando | Nike Training Club, Adidas Runtastic, Asics Runkeeper, Bosch DIY & Garden |
| Related research | Two-sided markets (e.g., Hagiu and Wright 2015; Reinartz, Wiegand, and Imschloss 2019; Rochet and Tirole 2003) | Community building and cocreation (e.g., Kozinets, Hemetsberger, and Schau 2008; Ramaswamy and Ozcan 2018; Siebert et al. 2020) |
Despite the widespread success of brand aggregation platforms, their focus on facilitating transactions across a broad product range also makes them vulnerable to competitive attacks. They typically lack the resources, expertise, reputation, and infrastructure to individualize customer experiences with respect to any single category and thus fail to occupy a specific category space ([64]). By contrast, product brands can approach value creation through platformization from a more specialized angle (e.g., Nike's Run Club and Training Club platforms are built around athletics).
Critically, brand flagship platforms, as mediators of versatile interactions between participants, are destined to become much more than an own-brand sales channel. They offer vast opportunities to cocreate value within the brand's category space through a plethora of commercial and noncommercial activities ([62]). Versatile interactions may include anything from buying products to providing and consuming content (e.g., product reviews, creative videos, educative podcasts) or services (e.g., participating in brand- or community-organized events). For example, participants on Bosch's DIY & Garden can access and upload home projects, discuss tools and techniques in forums, receive expert advice, participate in DIY challenges, collect points and badges, and engage in many other activities. The platform thus provides a rich array of products, services, and content around consumers' idiosyncratic DIY needs, offering them a potentially superior, more specialized yet more comprehensive experience than brand aggregation platforms.
Owing to their versatility, brand flagship platforms may pursue numerous goals such as increasing brand awareness and loyalty, offering complementary products and services, or triggering consumer feedback processes. Although the platform may include direct selling of brand inventory, this need not be its sole or central purpose. Runtastic started out as a tracking application and, despite its acquisition by Adidas in 2015, has preserved and enhanced these features, which remain at the core of the platform's value proposition. Table 1 summarizes our conceptions of brand aggregation and brand flagship platforms.
Assembling brand flagship platforms is not trivial, as they require strategic and operational choices that may substantially enhance or limit their value to consumers. In both practice and theory, brand platformization is still evolving, and guidance on how to approach the transformation is scarce. In response, we next present a novel conceptual framework to systemize the assemblage and management of brand flagship platforms.
Our framework is rooted in the idea for platforms as communities that leverage the wisdom and addressability of the crowd—that is, "an undefined (and generally large) network" of actors ([46], p. 346). Brands can use this framework as a conceptual basis to map the possible paths to their own flagship platform. We detail the underlying concepts and their relationships and tie them to brands' platform assemblage decisions and consumer relationship outcomes.
Digital platforms address consumer goals by relegating them to the open community, or the crowd ([33]; [74]). Underlying the platform concept is the idea that standing in a crowd and shouting out why you are there is often more beneficial than reaching out to agents in this crowd individually ([ 1]; [33]). The crowd's assets and capabilities collectively exceed those of any subset of agents, making scarce resources more abundant and increasing the number of unique offerings ([34], p. 186). Furthermore, the search for solutions in crowds become less resource-intensive because one market side (e.g., the consumer) articulates its goal to all participants on the other side (e.g., suppliers) simultaneously. As a result, drawing from the crowd is likely to increase effectiveness and efficiency of problem solving ([74]; [83][83]).
Crowds have been shown to play an important part in new idea generation and selection ([24]), product development ([ 4]), and solving of micro tasks ([26]). However, most studies take a managerial perspective, focusing on firms' decision to relegate business problems to consumers ([87]). We extend this notion to the digital platform context, where the "shouting out" of platform participants' goals translates to, for example, consumers browsing through content, using search queries, or sorting and filtering categories.
We view these activities as forms of consumer crowdsourcing—that is, consumers' open assignment of a task to a network of people or other entities ([87]). On platforms, task assignment pertains not only to consumers but to all platform participants, including third-party businesses and product brands, which assume similar positions as contributors to the whole. Consumers harness the power of this crowd—rather than individual retailers or brands—whose product, service, and content offerings are bundled and made digestible through technology. Notably, it is irrelevant if the consumer decides to source from one platform participant only (e.g., follows one training plan posted by a single person), as this choice is the outcome of the crowdsourcing activity, akin to the winner of a crowdsourcing competition. The consumer selects the offering that suits them best out of all crowd-supplied offerings, like strolling through the weekly market to fill their bag.
On digital platforms, recipients can also add to crowd solutions themselves by, for example, rating products; engaging in discussions; and uploading pictures, videos, or music playlists. We term such behavior consumer "crowdsending," which we define as consumers' activity of contributing to the network through the supply of products, services, or content. Engaging in crowdsending may, for example, strengthen consumers' social identity and status ([40]), bestow a sense of purpose and belonging ([73]), and allow them to gain (monetary) rewards ([44]). Akin to crowdsourcing, crowdsending on digital platforms supersedes piecemeal activities by ensuring that the input resonates with a large and engaged crowd of recipients, who in turn use the crowdsent material for further value creation, dramatically extending the scope of these activities compared with a nonplatform environment.[ 6]
Building on these ideas, we structure our conceptual development along a decision–process–outcome framework. Assembling infrastructure and governance mechanisms represent upstream decisions the brand makes. The resulting platform assemblage, as the totality of infrastructure and governance mechanisms, determines which crowdsourcing and -sending activities the platform facilitates and how these activities are designed.
Consumer crowdsourcing and crowdsending represent value-creating processes that aim to fulfill consumer goals. While brands may foster or restrict specific processes through their assemblage decisions (Kozinets, Ferreira, and Chimenti 2021), they manifest only in the interplay with consumers. That is, a platform's capacity to affect (through crowdsourcing) or to be affected (through crowdsending) needs to be met with the capacity to affect or be affected on the part of the consumer ([21]). Accordingly, which platform offerings are exploited hinges on each consumer's idiosyncratic engagement with the platform's crowdsourcing and -sending capacities, which they mix and match to create their own solution ([22]). For example, learning how to consistently lose weight requires the exploration of parts of the flagship platform that differ from those required when training for a marathon.[ 7]
Finally, crowdsourcing and -sending experiences on the brand flagship platform have important downstream consequences (outcome) for the development and maintenance of consumer relationships, which ultimately define the long-term success of the brand's platform business. As we show in a subsequent section, different relationship states manifest from consumer crowdsourcing and -sending, which can range from transaction-oriented, discrete exchanges to ongoing interactions that lead to a profound embeddedness of the platform in consumers' lives ([13]; [90]). Brands need to assess the level of alignment between the relationship states they seek with their flagship platform (i.e., their brand goals) and the achieved outcome and dynamically reassemble platform parts to spur new crowdsourcing and -sending processes or redirect existing ones.Figure 1 summarizes this framework.
Graph: Figure 1. Decision–process–outcome framework for the creation of brand flagship platforms. Notes: The different shapes/colors indicate whether a topic area relates to decisions, processes, or outcomes. Platform building blocks are a subset of the decision-related topic areas and consumer-platform relationship states are a subset of the outcome-related topic areas.
The following sections tie together the depicted processes of crowdsourcing and crowdsending with brands' platform assemblage activities (decision), on the one hand, and with the emerging consumer–platform relationships (outcome), on the other hand.
We posit that the crowd, that is, the collective of platform participants, is particularly good at addressing important consumer goals, which renders digital platforms a superior alternative to retailers and direct interaction. The question thus arises as to when crowdsourcing and -sending activities become particularly rewarding for consumers. We have synthesized prior research from major marketing journals (see Web Appendix A) and industry observations to derive five key goals that consumers pursue when sourcing from or sending to the crowd: commercial exchange, social exchange, self-improvement, epistemic empowerment, and creative empowerment. We put forth these goals, summarized in Table 2 and explained next, because they follow directly from the interpretation of platforms as crowd-based providers of value.
Graph
Table 2. Key Consumer Goals and Platform Building Blocks.
| Consumer | Platform |
|---|
| Nature of the Consumer Goal | Definition of the Consumer Goal | Literature Example | Building Block | Related Features | Literature Example |
|---|
| Commercial exchange | Finding the best- matching offering or exchange partner | Perren and Kozinets (2018) | Transaction block | Product/service marketplace, complaint handling | Hagiu and Wright (2015) |
| Social exchange | Engaging in social interaction | Schau, Muñiz, and Arnould (2009) | Community block | Community forums, social sharing | Ramaswamy and Ozcan (2016) |
| Self-improvement | Competing and comparing to live up to one's full potential | Ramaswamy and Ozcan (2018) | Benchmarking block | Tracking, measurement, benchmarking | Labrecque et al. (2013) |
| Epistemic empowerment | Spreading and gaining knowledge to make informed decisions | Kozinets, Ferreira, and Chimenti (2021) | Guidance block | Peer-to-peer route planning, "how-to" videos, customer feedback | Kozinets, Ferreira, and Chimenti (2021) |
| Creative empowerment | Inspiring or being inspired by something new or curious | Albuquerque et al. (2012) | Inspiration block | Decoration videos, tools for exploration and experimentation | Füller, Matzler, and Hoppe (2008) |
Crowdsourcing may pertain to efficiently finding an offer for products or services from the network of suppliers ([29]), but also extends to, for instance, gathering knowledge or inspiration about products, services, or activities. Likewise, crowdsending may pertain to, for example, the goals of engaging in social interaction (e.g., [67]), self-actualization ([72]), or self-expression by providing information or inspiration ([47]). Consumers strive to become active contributors on the platform and are rewarded by the crowd's engagement with their contributions ([46]). Naturally, consumers may pursue more than one goal with the same platform (and the same platform feature), and platforms may serve different goals for different consumers.
Digital platforms cater to these important goals in the form of five building blocks, depicted on the right-hand side of Table 2. Importantly, following straight from the consumer-goal perspective, these building blocks do not map exactly to technical platform features. For instance, the sharing of a training exercise could reflect a social desire (engaging with the crowd) but also serve as guidance, inspiration, or a way of benchmarking for users. Thus, the same technical implementation may reflect several building blocks and thus cater to several consumer goals. We explain each block and corresponding goal in the following.
Catering to the goal of commercial exchange with a transaction block focuses on providing matches between demand and supply to ensure that consumers find the offering that best suits their needs within the assortment of products or services ([29]; [85]). Although facilitating commercial exchange is not unique to digital platforms, the introduction of virtually endless shelf space coupled with advanced matching algorithms and rapid scalability on both market sides (buyers and sellers) through network effects has laid the foundation for platforms providing superior customer value relative to traditional pipeline businesses (Van Alstyne, Parker, and Choudary 2016). Digital platforms allow for almost frictionless participation, provide higher price and quality transparency, and allow buyers and sellers to find the perfect exchange partner easily and quickly, thereby vastly diminishing transaction costs ([38]).
For many commercial platforms (including brand aggregation platforms) the exchange block is at the core of their operations, with additional infrastructure assuming supporting functions (e.g., Amazon Marketplace, eBay, Uber). However, some platforms integrate an exchange block as part of a more balanced approach so that the commercial-exchange focus does not dominate other platform benefits and thereby prevent further user growth.
Social exchange relates to the goal of engaging in interactions with other platform participants such as consumers, third parties, or employees. These interactions are often part of an ongoing and evolving dialogue between participants aimed at advancing an array of social practices ([67]). Research on cocreation, brand communities, and website use shows that these interactions lead to several social benefits for consumers (e.g., [51]). They create a sense of belonging, identity, and even friendship ([86]), and consumers enjoy the status, reputation, and esteem they acquire within a community of peers as well as their expression of a unique self-image, giving them a sense of self-efficacy ([32]; [51]).
Consumers may achieve these goals on the platform through a variety of means, such as by sharing experiences and personal beliefs ([48]). Depending on which features are part of the platform assemblage and how well they are integrated, broadly positioned brand flagship platforms are able to cater to social exchange goals much better than individual brand communities, which are limited by their commercial and brand focus, relatively small group of active participants, and lack of diversity.
Consumers have the fundamental urge to live up to their full potential, "the desire to become more and more what one is, to become everything that one is capable of becoming" ([50], p. 383). Prior literature has incorporated individual aspects of self-improvement in terms of acquiring excellence ([32]). We extend this concept to include self-respect and accomplishment ([32]; [86]). These aspects underlie the quantification and competition trends that many digital platforms address, especially in the contexts of fitness, health, nutrition, and sports ([66]). We posit two points of reference that are most relevant for the consumer in terms of benchmarking and drawing motivation to self-improve ([43]; [84]): the platform crowd and the consumer's own past performance. This development is fueled by the widespread use of connected devices such as mobile phones and wearables that can track and quantify consumers' activities, workouts, sleep quality, heart rate, and more (James, Deane, and Wallace 2019). Digital platforms often combine self-tracking and measurement technology with crowd-based motivational features and gamified experiences such as user rankings, virtual badges, and opportunities to share accomplishments ([47]).
Consumer empowerment is defined as "the strengthening of consumers' abilities, rights, or authority to consume or otherwise fulfil their objectives as a marketplace actor" (Kozinets, Ferreira, and Chimenti 2021, p. 7). Digital platforms support epistemic consumer empowerment through information on the products and services they offer—but also do more. Some platforms focus on participants' exchange of information, as in the case of Skillshare, where thousands of teachers can upload educational videos, or Komoot, where participants can record hiking or biking routes and recommend them to others. Other platforms build a strong guidance block around their main commercial operations by allowing participants to share instructional videos, blog posts, or structured product feedback and evaluation ([45]).
Guidance blocks are typically standardized, tailored to a specific purpose, one-directional, and aim to provide clear and concise results compared with the more open and socially oriented community block. Owing to the quantity and diversity of content available through crowd-based guidance blocks, epistemic empowerment of participants on all market sides easily supersedes what retailers or single brands can offer.
With creative empowerment, we denote satisfying consumers' curiosity and longing for new experiences ([70]; [76]). Whereas epistemic empowerment refers to objective and functional knowledge such as that for comparing product alternatives for purchase, creative empowerment refers to knowledge and stimulation that inspires consumers ([15]). Consumers can use digital platforms to gain or provide elements of creative empowerment by consuming or creating content (including entertaining content) and products. Platforms such as Pinterest and Wattpad leverage the crowd to provide a richer set of ideas for consumers seeking inspiration and a large and engaged group of potential recipients for consumers sharing their ideas. Nike's Run Club platform induces consumers to inspire and be inspired by the crowd through music playlists or workout videos that they can add for others to use in their own routine.
The five building blocks can be (re)assembled to represent different types of platforms observable in the market (Figure 2). Thinking about the heterogeneous platform landscape as different assemblages of these building blocks presents a powerful tool for brands in their creation of their own flagship platforms because it helps them select the optimal assemblage to achieve specific brand goals. We suggest that the existence and the relative scope of each building block jointly determine which platform type emerges. For example, an emphasis on the transaction block tends to produce marketplaces that cater primarily to the goal of commercial exchange. A strong benchmarking block yields tracking and coaching platforms that cater primarily to the goal of self-improvement. Other platforms combine blocks to fulfill several consumer goals at the same time and allow each consumer to pursue their own goal choosing a specific consumer–platform subassemblage ([21]; [31]). For instance, Nike's Run Club offers tracking and ranking features to support self-improvement and sharing features to socially engage with other participants. While the consumer's idiosyncratic crowdsourcing and -sending activities yield individual subassemblages, the brand sets the general frame through its assemblage of building blocks.
Graph: Figure 2. Using the building blocks to assemble different platform types.
Our conceptualization of digital platforms as places of consumer crowdsourcing and crowdsending and corresponding assemblages of building blocks moves the focus of analysis from the platform architecture, which most prior studies address, to the consumer goals that platforms aim to fulfill. It also reduces the complexity and ambivalence of the platform concept reflected in prior literature. This view implies that digital platforms are revolutionary not only because of their ability to scale quickly and leverage network effects ([65]) but also because of their versatility and integration ability: they create greater value for consumers by allowing them to draw from and provide to the crowd using assemblages of building blocks that facilitate idiosyncratic goal pursuit. This recent development is important for brands because it allows them to counter the intermediation through brand aggregation platforms by evading direct competition on their home turf and instead establish a direct interface with consumers, developing relationships that go beyond discrete transactions. Next, we discuss what these relationships look like and how crowdsourcing and -sending activities affect their emergence.
The relationships that emerge on flagship platforms reflect focal outcomes of brands' platformization efforts because they determine how brands can leverage their platforms strategically. By "relationship," we denote the possible states that consumers may enter in their interactions with brand flagship platforms and that vary in terms of their self-relevance, commitment, and durability (e.g., [23]). As consumers interact with platform participants through crowdsourcing and -sending activities, the nature of these interactions shapes the relationship states that emerge ([36]).
The following subsections develop this effect in two steps. In the first step, we establish the general link between the consumer processes of crowdsourcing and -sending, on the one hand, and the consumer–platform relationship, on the other. To this end, we argue that the nature of interactions on brand flagship platforms varies by the intensity of consumer crowdsourcing and -sending, and that this intensity increases as consumers use the platform to pursue higher-level goals. We then show that this intensity in turn affects consumer–platform relationships on a spectrum from discrete, exchange-based transactions to versatile, ongoing interactions. More intense crowdsourcing and crowdsending move the needle toward the versatile end of the spectrum, that is, increasingly profound relationships that culminate in a deep embeddedness of the platform in consumers' lives ([13]; [90]).
In the second step, we detail this general link by separating out crowdsourcing and crowdsending intensities to constitute key dimensions of a typology of archetypical relationship states. This step allows us to enter a more fine-grained discussion of consumer–platform relationships that helps brands analyze specific outcomes of their platform assemblage efforts. We first describe the typology and then develop propositions that clarify which relationship state brands should foster on their flagship platform to realize particular brand goals, as well as which risks are involved.
Consumers' crowdsourcing and -sending activities on brand flagship platforms can vary greatly in terms of their intensity—that is, how extensive and/or intimate they are ([ 5]; [27]). Drawing from prior literature, we refer to extensiveness as the frequency, duration, and variety of activities, whereas intimacy in crowdsourcing (crowdsending) denotes how sensitive or personal the consumed (disclosed) information is or the degree to which participants are influenced by (influence) other platform participants with respect to their attitudes and behaviors ([ 5]; [ 6]). Intensity is therefore a continuous dimension of crowdsourcing and -sending that increases with growing extensiveness and intimacy. For example, rating a running shoe constitutes lower-intensity crowdsending than writing a review, which in turn reflects a lower crowdsending intensity than sharing a series of videos that explain different running postures.
When does intense consumer crowdsourcing and crowdsending occur on brand flagship platforms? Brands provide the infrastructure that enables consumers to engage in crowdsourcing and -sending through the platform's assemblage of building blocks and governance mechanisms. However, intense interactions only occur if consumers are also motivated to draw from and contribute to the crowd. As a key factor of this motivation, we put forth whether and how comprehensively the platform addresses consumers' higher-level goals and thus allows the platform to transcend discrete, product-centric transactions, as we explain next.
According to means-end theory, consumer goals are hierarchically organized (Huffman, Ratneshwar, and Mick 2003; [58]). On the highest level, consumers formulate abstract goals that describe why they perform certain actions in the pursuit of their personal values and ideal self-identity, such as following a healthy and active lifestyle ([58]). Midlevel goals constitute projects with concrete targets such as being able to run ten kilometers in under 50 minutes ([58]). On the lowest level, consumers define actions and behaviors that allow them to achieve their superordinate goals, including purchase decisions such as buying a particular pair of running shoes ([13]; [58]).
Brand flagship platforms that address higher-level goals should be more central to consumers' sense of self ([ 8]; [13]) and thus increase the platform's self-relevance—that is, the alignment of the platform with consumers' self-concept and self-image ([14]). As self-relevance grows, consumers' satisfaction, trust, and commitment tend to increase ([41]) along with their willingness to engage in more intense interactions. A platform with an exchange-based, product-centric focus serves lower-level goals (Huffman, Ratneshwar, and Mick 2003), implying low self-relevance. In contrast, flagship platforms with a higher-level goal focus should achieve greater self-relevance to consumers and motivate them to engage in more intense crowdsourcing and -sending.
[11] findings on the car-sharing provider Zipcar show how fostering intense crowdsourcing and -sending may fail when higher-level consumer goals are not addressed. Zipcar tried to create more intense interactions by introducing a community block to the platform. However, the platform lacked self-relevance, as consumers used the platform mainly for utilitarian motives. Thus, they remained highly focused on the lower-level consumption-related goals, did not engage in intense crowdsourcing or -sending, and even actively rejected the community features. By contrast, Adidas Runtastic and Nike Run Club explicitly and consistently address higher-level, self-relevant consumer goals pertaining to self-improvement and an active lifestyle, which foster intimate crowdsourcing (e.g., following advice) as well as crowdsending (e.g., contributing to the community) activities.
- P1: Brand flagship platforms that address higher-level rather than lower-level consumer goals (a) are more self-relevant to consumers and, therefore, achieve (b) higher crowdsourcing and (c) higher crowdsending intensities.
By motivating consumers to engage in intense crowdsourcing and -sending, a platform's locus of value creation shifts from ephemeral value-in-exchange toward ongoing value-in-use ([62]; [82]). Accordingly, product-centric exchanges evolve to project-centric interactions, as consumers use the platform to pursue discrete (e.g., running a marathon, remodeling one's garden) or even continuous projects based on their higher-level goals (e.g., living an active and healthy life, living in a comfortable and safe home). These projects are inherently long-term, involving multiple consumer crowdsourcing and -sending activities on different levels. The platform thus offers a crowd-based, integrated solution that goes far beyond selling products or services ([22]; [80]).
To illustrate, brand aggregation platforms typically revolve around discrete transactions, including buying and rating products, or answering product-related questions. These activities may occur repeatedly but remain superficial and short-lived, lacking intimacy and extensiveness. Platforms such as Google Shopping provide outstanding transactional value, but crowdsourcing and -sending beyond that are limited and often discouraged or even prevented ([18]). In contrast, a brand flagship platform like Adidas Runtastic supports extensive and intimate crowdsourcing and -sending. Consumers upload self-created content, publicize their performance, or follow workout routines from peers. As the value-creation process is inherently interactional, greater intensity in these activities leads to greater and perpetuated value for consumers ([62]).[ 8]
The extensiveness and intimacy of intense crowdsourcing and -sending and the interdependence that follows from the joint, project-centric value creation elevate consumer–platform relationships from being ephemeral and exchange-focused to becoming more committed and durable ([23]; [56]). As a result, brand flagship platforms can transcend the traditional customer journey stages and achieve an increasingly deep embeddedness in consumers' lives ([13]; [90]) with increasing intensities of crowdsourcing and crowdsending.
- P2: Brand flagship platforms that support more intense rather than less intense crowdsourcing and crowdsending foster consumer relationships that are (a) less exchange-focused, (b) more committed, and (c) more durable.
The general assertion that crowdsourcing and crowdsending intensities are directly related to relationship outcomes can be further detailed by separating out the activities and examining their distinct effects on consumer–platform relationships. Accordingly, we propose consumer crowdsourcing and -sending intensities as the two core dimensions that determine the emergence of archetypal relationship states on brand flagship platforms (Figure 3). We briefly describe these states and illustrate their differences using our athletics example before discussing their implications for product brands as platform owners.
Graph: Figure 3. Consumer–platform relationships and brand outcomes.
When consumer crowdsourcing and -sending intensities are low, value creation is typically transaction- or product-centric, addressing consumers' lower-level goals. Value creation may occur through purchase-related activities (e.g., commercial exchange, product reviews) or functional extensions to the core product (e.g., "smart" features such as reporting the wear on a running shoe). Importantly, the crowd is not uninvolved in the platform's value creation, as consumers may still crowdsource from and crowdsend to the platform by, for example, consulting or providing product ratings or responding to product-related questions ([10]). However, these activities remain superficial because they lack extensiveness and intimacy. As a result, self-relevance is low, and the emerging relationship is ephemeral and exchange-focused. This outcome is not necessarily negative, because consumers do not always seek highly committed relationships but value instrumentality in low self-relevance consumption contexts ([12]).
When consumer crowdsourcing intensity is high but crowdsending intensity is low, the consumer primarily capitalizes unidirectionally on value provided by the brand flagship platform while not being highly engaged in the creation of value themselves. For example, the consumer may follow a workout routine or extend their knowledge base by watching instructional coaching videos. But as crowdsending intensity is low, the consumer does not share their progress with the crowd or engage in community discussions. Intense crowdsourcing implies that the platform addresses self-relevant higher-level consumer goals, creating a willingness to integrate the platform assemblage into their own self ([ 8]). The platform becomes part of the consumer and their identity, endowing them with new capacities that result in self-expansion experiences ([31]). However, as consumer crowdsending intensity is low, the platform misses out on the input that "lurking" consumers fail to provide, neither creating value for the consumer by actively contributing to the crowd, nor being able to collect data from these crowdsending activities to further improve crowdsourcing opportunities down the line ([68]).
When consumer crowdsourcing intensity is low but crowdsending intensity high, the brand flagship platform catalyzes consumers' value creation, providing a canvas on which they can express and affirm their self ([ 7]; [79]). For example, consumers may share and annotate their favorite running routes or design sportswear.[ 9] This intense crowdsending implies that the platform addresses self-relevant higher-level goals and leads to self-extension experiences ([13]) as the consumer integrates part of their self into the platform ([31]). As a result, the platform assemblage gains new capacities from the participant. Thus, in contrast to the relationship states with low crowdsending intensity, consumers are more deeply integrated into the value cocreation process ([61]).
Nonetheless, value creation in these relationships is innately truncated. The consumer self-extends into the platform while their self remains stuck in the status quo, owing to a lack of self-expansion. Thus, the consumer does not gain new capacities such as skills or knowledge which limits the relationship to what the consumer is instead of what they could be.
When consumers engage in intense crowdsourcing and crowdsending, a nurturing partnership between consumer and platform emerges. Contrary to other relationship states, value creation is virtually unrestricted, as both parties can grow mutually. When the consumer extends their self into the platform (crowdsending), the platform's capacities expand, allowing it to provide value through which the consumer self-expands (crowdsourcing). For example, while the consumer engages in crowdsourcing by following a series of video tutorials and live events, they may also engage in crowdsending by contributing to discussions, sharing their training diary, and creating workout playlists. These activities may create consumer value as others engage with the consumer's input by upvoting their content, cheering them on, or inviting them to challenges. Furthermore, the input can also spawn new opportunities for crowdsourcing, for example by others offering advice on the consumer's newly acquired skills or the platform improving and personalizing their crowdsourcing experience, reaching ever higher levels of value drawn from the crowd. With these ongoing interactions, the platform and consumer can keep nurturing each other's capacities through the interplay of extending into and being expanded by the respective other. Mutual value creation and codependence in a nurturing partnership occur to perpetuate the relationship and its intense interactions ([49]), bearing the potential for the platform to become a close companion on the consumer's path toward their higher-level goals.
Brands can use the depicted typology to assemble and dynamically revise their flagship platform to foster the desired relationship state with consumers. Which relationship state is desirable depends on the goal(s) a brand pursues, as not every state lends itself well to each brand goal ([11]). Thus, brands need to identify and prioritize their goals and select building blocks and governance mechanisms accordingly. We first present brands' main goals when venturing into platform business and subsequently discuss suitable relationship states to achieve specific goals, as well as the pitfalls that may arise.
Combining insights from academic literature, the business press, and a practitioner workshop (see Web Appendix B), we identified five primary goals that brands pursue when creating their own flagship platform. First, brands expect the platform to support sales of their core offering, either by selling on the platform directly as an additional distribution channel or by indirectly generating leads through recommendations and ads. This activity may also include after-sales services and customer care. Second, brands may aim to extend their core offering by sourcing functional enhancements through the platform, such as connected software applications or inspiration for alternative or improved product use. Third, brands aim to extend their core offering through complementary products and services, typically reaching into adjacent product categories. To stick with our sports example, a brand that sells running equipment might use the platform to offer sports nutrition, personal coaching, active travel booking, and a sports events ticketing service with the help of third-party suppliers. Fourth, brands aim to increase brand equity and deepen consumer relationships through frequent interactions and ways to nudge engagement behavior (e.g., by facilitating social interaction or cocreation activities). For example, a platform may allow runners to co-design their own sports gear, thereby increasing engagement with the core offering and the brand ([59]). Fifth, brand flagship platforms promise access to vast amounts of data that brands can use to better understand consumers and competitors as well as complement providers.
In ad hoc relationships, the transaction block often is core to interactions. If other building blocks exist, they either serve the exchange (e.g., community features for product-related questions) or are limited to functional product enhancements (e.g., after-sales service, tracking features). Means for intense crowdsourcing or -sending are usually not provided, yielding an instrumental consumption experience with low self-relevance for consumers and a focus on utilitarian benefits rather than intimate relationships ([11]). As interactions are limited, brand flagship platforms can employ a restrictive governance style in which the brand retains control over interactions ([18]).
Despite its lack of self-relevance, the platform can still achieve high consumer loyalty if consumers seek instrumental consumption experiences and the platform provides substantial utilitarian benefits ([11]; [41]), for example, through efficiency gains from a broad range of offerings or high matchmaking quality ([38]; [57]). This loyalty built on utilitarian benefits such as convenience or cost savings constitutes what we term "cold loyalty." If a flagship platform fails to deliver on these utilitarian benefits or if a competitor delivers greater utilitarian benefits, consumers are likely to abandon the ad hoc relationship quickly.
- P3: If the brand's main objective is to support its core offering (through direct selling and functional enhancements), it is optimal to foster ad hoc relationships on the brand flagship platform.
- P4: Lower consumer crowdsourcing and crowdsending intensities on the brand flagship platform (a) foster instrumental consumption experiences and, therefore, (b) increase the risk of consumers switching to platforms that offer superior utilitarian benefits.
In capitalizing relationships, platforms offer opportunities for intense crowdsourcing under higher-level consumer goals. These goals spur concrete needs that the platform may address to increase self-relevance and foster intense crowdsourcing. On an athletics platform, for example, the higher-level goal of an active and healthy lifestyle may create needs for adhering to a training schedule, learning about nutrition, or participating in sports events. The more comprehensively a platform covers the many needs associated with a higher-level goal—that is, the further it expands into the broader category space—the more self-relevant the platform can become and the more intensely consumers can engage in crowdsourcing. Thus, capitalizing relationships allow—and usually require—brands to extend their core operations into a variety of adjacent fields connected to higher-level goals.
The low degree of consumer crowdsending intensity implies that consumers are not deeply involved in the value creation such that value is usually provided by the brand itself or by third parties. In addition, it allows for a stricter governance style through which brands can retain control over the value-creation process and, by extension, monetization and customer satisfaction ([18]). Typically, the guidance block is most pronounced as consumers focus on expanding their selves rather than engaging with others. Discrete interactions in capitalizing relationships thus become part of a longer journey to self-expansion, during which the consumer builds knowledge and skills. The result is a "warm loyalty" that goes far beyond rational motives and that features a high level of self-relevance, affective commitment, and attachment ([41]).
Despite consumers' strong attachment, this relationship state faces risks of destabilization ([21]). On the one hand, as the platform extends further into a category space supported through third-party complementors, the brand risks being diluted as its positioning is stretched across the category space and it vies with third parties for consumers' attention (Swaminathan et al. 2020). On the other hand, consumers may disengage from the platform if it lacks alignment with their higher-level goals ([52]). This response may result from the lack of intense crowdsending, which challenges the brand's ability to evaluate the alignment. As consumers are not deeply involved in the value creation, they cannot steer the platform toward their own higher-level goals.
- P5: If the brand's main objective is an extension of its offering into the broader category space, it is optimal to foster capitalizing relationships on the brand flagship platform.
- P6: Higher consumer crowdsourcing intensity on the brand flagship platform increases the risk of brand dilution.
- P7: Lower consumer crowdsending intensity on the brand flagship platform increases the risk of misalignment with consumer goals.
In catalyzing relationships, consumers engage in intense crowdsending through which they self-extend into the platform and cocreate its value. The platform thus becomes a canvas that consumers can project their selves and higher-level goals onto. It enables this activity through its assemblage of building blocks, typically by providing pronounced community and inspiration blocks, and a relaxed governance style. The resulting cocreation experiences are associated with a variety of psychological benefits that increase involvement, attachment, and engagement of consumers, fostering another form of "warm loyalty" ([17]).
Given the low degree of consumer crowdsourcing, interactions typically center on the brand's core products or a few key platform functionalities. Thus, the brand focus is stronger in these relationships, decreasing the risk of brand dilution compared with states with high crowdsourcing intensity. Accordingly, brands can leverage catalyzing relationships to increase consumers' engagement with the brand's core products. Relationship states with high crowdsending intensity also allow brands to gather intelligence from the information shared and value provided by consumers, such as for product development ([35]).
While intense crowdsending requires a relaxed governance style that enables consumers' creative empowerment ([57]), a catalyzing relationship may degrade when consumers exploit these conditions ([52]). For example, consumers may hijack the platform by taking it in a direction that is not aligned with the product brand's values, such as by engaging in antibrand behavior (e.g., [55]) or hostile behavior against others (e.g., stalking, bullying). Accordingly, product brands may be tempted to inhibit the expressive capacities granted to consumers. However, this reaction may cause the platform to become unresponsive to consumers' intense crowdsending ([52]), thereby reverting to an ad hoc relationship.
Because the low crowdsourcing intensity increases consumers' focus on the brand and its core offerings, the potential to expand into the broader category space, as in capitalizing relationships, is limited. This limitation bears the risk that consumers feel constrained by the brand corset, and the platform may attract (only) product and brand enthusiasts rather than a more diverse crowd. Take Lego Ideas as an example, which allows consumers to create and share own designs, but which is firmly tied to the core product of Lego bricks and its buyers. Thus, an open and inspiring interest community may revert to a more specialized (brand) community ([28]). This issue may further aggravate as in-group members tend to seclude from out-group members ([78]), which could result in the latter leaving the platform or refraining from joining it in the first place. The focus on the brand's core offering and its existing customers may also render the platform less attractive for third-party businesses and reduce positive network effects ([53]). The platform crowd may thus become smaller and less diverse.
- P8: If the brand's main objective is to strengthen engagement with its core offering, it is optimal to foster catalyzing relationships on the brand flagship platform.
- P9: Higher consumer crowdsending intensity on the brand flagship platform increases the risk of platform hijacking.
- P10: Lower consumer crowdsourcing intensity increases the risk of attracting a (a) smaller and (b) less diverse crowd.
In nurturing partnerships, platforms enable high-intensity crowdsourcing and -sending by assembling a variety of building blocks and by extending value creation into the category space spurred by consumers' higher-level goals, thereby fostering self-relevance. The resulting combination of simultaneous self-expansions and self-extension experiences has profound implications for the relationship. The platform becomes an integral part in the consumer's identity project by being not only a resource to be used ([ 7]) but also an actual partner cocreating the consumer's self just as much as the consumer cocreates the platform offering. In this way, brands may achieve what was beyond their reach in traditional nonplatformized business models: a profound embeddedness in consumers' identity and lives ([ 7]; [13]).
Take a user of an athletics platform who is spurred by the higher-level goal of living a healthy and active life. By following a chosen training schedule and participating in weekly challenges (crowdsourcing), their self expands, and they increasingly identify as an active and healthy person. By sharing their performance and achievements, forming a running group, and giving recommendations to others (crowdsending), the user then also proclaims and exerts their new, expanded self. This means that by extending their expanded self into the platform, the consumer is affirming and reinforcing it ([79]). Therefore, self-expansion combined with self-extension is likely to be more profound and lasting than self-expansion alone. As the platform becomes part of the consumer and vice versa, the relationship can achieve a new level of intimacy, attachment, and perpetuation of value (co)creation, leading to what we term "hot loyalty." Thus, brands can harness nurturing partnerships to build consumer relationships that are highly self-relevant, committed, and durable and that benefit multiple downstream brand goals relating to the extension of the offering, sales, engagement, and intelligence.
However, nurturing partnerships also combine the risks of high-intensity crowdsourcing and -sending in terms of brand dilution and platform hijacking. In addition, highly self-relevant relationships have been shown to elicit strong adverse consumer reactions when a critical incident occurs, even provoking antibrand behavior such as spreading negative word of mouth ([41]). Thus, if goals in nurturing partnerships should become misaligned or perceptions of exploitation emerge, consumers' "hot loyalty" may boil over, causing a sudden and lasting dissolution of the relationship.
Nurturing partnerships also tend to be demanding and costly to maintain because brands need to manage third-party businesses as well as high-intensity consumer crowdsourcing and -sending. Their interactions may create new challenges such as complementors defecting with consumers ([88]), and consequently, the value captured from the platform—monetary or otherwise—may be insufficient to justify its continuation. A case in point is Under Armour, which recently discontinued parts of its brand flagship platform Connected Fitness to focus on a more concentrated set of functionalities ([61]).
- P11: If a product brand's main objective is to establish consumer relationships that are highly (a) self-relevant, (b) committed, and (c) durable, it is optimal to foster nurturing partnerships on the brand flagship platform.
- P12: Higher consumer crowdsourcing and crowdsending intensities on the brand flagship platform increase the risk of (a) conflicts causing a lasting relationship dissolution and (b) the incurrence of high operational costs.
Our discussion of relationship states clarifies that although it may seem intuitively desirable for brands to reap the full benefits of intense consumer crowdsourcing and -sending activities, aiming for nurturing partnerships poses specific risks and requires high conviction to overturn the traditional product-centric business model. This strategy is not ideal for all brands, and not every product category is suitable to elicit intense crowdsourcing and -sending. We discuss these aspects next and propose directions of necessary managerial rethinking for brands aiming to create their own flagship platform.
Brand flagship platforms are flexible vessels that harness the latest evolutions in platform-based business models. Drawing on our analysis of these platforms, we offer three sets of extensions to conceptual thinking. First, we extend the notion of sourcing from consumer crowds, as prevalent in the crowdsourcing and community literature streams ([30]; [46]), to a platform context, where consumers crowdsource value from a pool of offerings supplied by other consumers, third-party businesses, and the platform owner. We integrate this notion, with its flipside of crowdsending, to an overarching theory of consumer value creation in platform-based business models.
Second, our approach spans the boundaries between different types of digital platforms discussed in literature and observable in the marketplace. Our assemblage model generalizes across diverse platforms such as exchange markets or online communities and accommodates a wide range of consumer benefits derived from platform offerings. In extension to the typology developed in [57], which classifies exchange markets into discrete types (forums, matchmakers, enabler, and hubs), the assemblage model suggests that brand flagship platforms may unify several or even all these types under one roof, while also integrating other, noncommercial interactions. Our conceptualization thereby also expands the usage of assemblage theory to study consumption and market phenomena (e.g., [60]) by applying it to the context of digital platforms.
Third, we introduce a relational perspective to the theory of platform intermediation. Prior research emphasizes transaction cost advantages and efficiency gains of digital platforms based on network effects ([42]). Our analysis reveals that this rational view, which culminates in relationships of "cold loyalty" and the pursuit of immediate consumption goals, paints an incomplete picture. Consumer–platform relationships may evolve toward ongoing and versatile interactions that target consumers' higher-level goals and elicit self-expansion and self-extension experiences. While product brands are in a good position to build these kinds of relationships, the mechanisms carved out in this article generally also hold for other platform owners across myriad industries.
The platformization of the digital space has put substantial pressure on traditional product brands as competition intensifies and the role of the brand diminishes ([25]). However, our analysis shows that the poison can also be the antidote: product brands can carefully assemble their own flagship platforms to loosen aggregation platforms' grip and reclaim direct consumer access—that is, if decision makers successfully rethink long-established viewpoints and behaviors, which we discuss next.
With brand flagship platforms, product brands can expand beyond their core offering and into the broader category space. Operating a flagship platform, especially with ample consumer crowdsourcing opportunities, will thus shift brands' locus of value creation from selling (somewhat standardized) merchandise to offering a virtually endless variety of project solutions that consumers can mix and match to integrate in their discrete or ongoing goal pursuit.
Three steps are important to manage this shift: First, brands must decide on the ground they want to cover. While consumers define the boundaries of the category space, brands fix how far and how fast they want to move away from their home turf. Following literature on brand extensions and resource-based advantage, we suggest that brands address those higher-level goals that fit their positioning and resource configuration. A conservative approach would be to advance in concentric circles by gradually expanding the brand's core offering (e.g., an athletics brand may initially provide guided runs instead of leaping into nutrition). This approach would also give the brand time to build the required infrastructure and additional capabilities, while bearing the risk of competitors occupying larger parts of the category space more quickly through bolder expansion strategies.
Second, brands need to decide which brand goal(s) to pursue and align their platform assemblage accordingly. Not any goal can be achieved with any combination of building blocks, and forcing brand goals on a specific assemblage will likely have dire consequences. For example, aggressive selling on platforms with a strong benchmarking block forces the source of value for consumers to reduce or even collapse. Furthermore, if brands do not consider how each building block creates crowd-based value, their goal attainment on the platform is inhibited, and consumers may look for alternatives. For example, to achieve a strong brand presence, product brands may be inclined to stack building blocks closely around their core business. In this regard, turning a versatile platform community into a narrow brand community could be just as detrimental as curating inspirational content to ban competitors. Such excessive control would limit the platform's ability to expand into the broader category space.
Third, the brand needs to decide what parts of the offering it intends to "make," "buy" (from third-party businesses), or "earn" (from consumers). While own production is the default for product brands, occupying entire category spaces implies that offerings will increasingly lie outside of their core business. To expand effectively and efficiently, managers must integrate third-party businesses and stimulate supply from consumers. Again, brands should not be afraid to give up (some) control, as creating large parts of the offering in-house is costly and leads the crowdsourcing and -sending concepts ad absurdum.
Just as focus needs to shift away from the core offering, key performance indicators (KPIs) need to evolve beyond product sales. Instead, the value created and appropriated through crowdsourcing and -sending should be measured holistically, with a broader system of KPIs. This system should capture at least five facets of value: ( 1) value to the brand from its products, ( 2) value to the brand from intermediation, ( 3) value to third parties, ( 4) value to consumers from crowdsourcing activities, and ( 5) value to consumers from crowdsending activities.
Brand flagship platforms are unique in that they are built around a product-based business core and owned by a brand that ultimately aims to strengthen this core ([89]). Thus, a suitable measurement system would augment rather than replace traditional product KPIs such as sales and market share. Additional indicators that reflect value to the brand from intermediation may comprise direct earnings from operating the platform (commissions, subscription fees, ad revenues) and extend to engagement metrics such as feedback and word-of-mouth activities by third-party businesses and consumers. Brand managers could also learn from Netflix or YouTube, which largely focus on time spent with their services as a key success measure ([69]). Finally, brands should integrate measures of interaction quality through short, automated surveys, reciprocal feedback systems (such as those used by Airbnb or Uber), and tracking of interactions such as the rate of completion of content or service consumption (e.g., training session dropout).
To measure value to third parties and consumers, individual-level or segment indicators can inform effective lifetime management. Brands need to track, for example, customer growth; new product, service, and content listings by third-party businesses; consumer usage and ratings of platform offerings; and platform access frequency and recency. One idea to structure all these measurement activities would be to classify them into platform building blocks such that brands can better identify white spots in their offering.
With the expansion of the brand's role, optimization objectives expand as well. Brands need to steer away from their sales maximization logic and toward keeping all platform participants happy (Van Alstyne, Parker, and Choudary 2016). Accordingly, managers may have to balance conflicting interests between different value objectives (e.g., Do I integrate an open feedback system to increase crowdsending intensity or would that counter my core brand objectives?). This stretch is uniquely difficult for traditional brands owing to the double burden of being offline born and having to entertain a profitable product business "on the side." Our relationship typology can point managers in the right direction, but brands need to be aware that they cannot have their cake and eat it too: brand flagship platforms are doomed to fail if immediate value to the core products always trumps all other measures.
Contrary to common wisdom, brands aiming to build their own flagship platform need to commit to substantial investments and organizational restructuring. Manufacturing brands must, for example, develop market research expertise to elicit consumers' higher-level goals, set up a partner management to integrate third-party solutions, and develop user experience and engagement skills. Contrary to pure players, brands must juggle both industrial production and digital intermediation, which require different skillsets and infrastructure.
This latter point pertains also to resources that leverage the exciting dynamics for which flagship platforms provide an ideal playground. Owing to the shift from sequential purchase journeys (characterized by so-called "loyalty loops" that describe a predictable cyclical purchase pattern) to an ever-evolving and highly individual platform experience, brands have the unique chance to establish so-called "involvement spirals" ([71], p. 45). Perpetuating consumer crowdsourcing and/or crowdsending activities can create unpredictable experiences at every interaction and thereby motivate "increasing experiential involvement over time" ([71], p. 46). To this end, brand managers who are used to sequential product life cycles will need to provide the resources, staff power, and organizational setup to adopt an agile approach.
Despite these challenges, product brands possess numerous existing assets and capabilities that level the seemingly skewed playing field. Contrary to pure platform players, brands can often rely on an engaged customer base, strong industry ties, expertise in their category, and an assortment of fresh and enticing products to offer. Managers ought to capitalize on these resources, for example, by using their brand power and partnerships to negotiate exclusive supply deals, to jumpstart crowdsourcing and -sending, and to propel strong network effects (Van Alstyne, Parker, and Choudary 2016). While brand aggregation platforms may be unsurmountable in their ability to realize efficiency gains, their Achilles' heel is the lack of tailored consumer experiences and credible expertise within a specific category space. And this is exactly where brand flagship platforms excel.
To derive future research avenues in the emerging field of brand platformization, we build on open issues that arise from our theorizing and classify these using a survey of 72 marketing managers in the consumer goods industry. We asked participants to evaluate important topic areas relating to key platform decisions, processes, and outcomes. Managers rated each area in terms of its novelty and current relevance for the success of their brand (for details, see Web Appendix C). We then mapped their answers onto a two-by-two matrix (Figure 4).
Graph: Figure 4. Managerial novelty and current relevance of topic areas (manager survey). Notes: The different shapes/colors indicate whether a topic area relates to decisions, processes, or outcomes. Platform building blocks are a subset of the decision-related topic areas and consumer-platform relationship states are a subset of the outcome-related topic areas.
In niche areas, researchers must identify specific market conditions under which topics are important and interesting to craft a contribution. Evergreens promise profound impact, as they represent lasting conundrums that practice and research have not or not sufficiently tackled thus far. In emerging areas, research opportunities are plenty, and although results may not be of immediate practical relevance, they can shape future business practices that put brands at the forefront of innovation. New key challenges promise researchers strong and immediate impact due to their high degree of novelty and managerial relevance. We add this classification to our extensive research agenda presented in Table 3. Next, we highlight a few key challenges and emerging areas that may inspire future inquiries, structured along our decision–process–outcome framework.
Graph
Table 3. Research Agenda.
| Topic Area | Research Questions | Classification |
|---|
| Decision-Related | |
| Individual building blocks | How can brands incorporate different types, sources, and presentation modes of inspirational content? How should brands elicit and leverage consumer inspiration? And to what degree does inspirational content foster more intense crowdsending? How can brands mitigate possible negative platform experiences when consumers are not progressing or even regressing as part of self-tracking through the benchmarking block? How can brands provide guidance on the platform while mitigating the risk of harmful outcomes (e.g., eating disorders from meal recommendations, injuries from a mismatched workout plan)? How can consumers be steered from a brand community to an interest community?
| Emerging area, new key challenge, evergreen |
| Assembling building blocks | How do building blocks interact with each other? How can products, services, and content be made searchable and engaging so that building blocks become better connected and provide networked value? How do existing brand associations and attitudes affect the optimal assemblage of building blocks? Which building blocks are required to elicit a specific relationship state and which are optional?
| New key challenge |
| Third-party integration | How do insights from brand extensions, cobranding, and coopetition transfer to value-creation extensions through third-party integration? Which parts of the value creation should brands "make" versus "buy"? Which circumstances (e.g., brand strength, market concentration) promote opening up to competitors? How much competition between suppliers, and between suppliers and the brand, is optimal? How to identify suitable third-party partners? Should brands enter exclusive contracts only? And how can the brand ensure that brand values and standards are reflected by third parties?
| Niche area |
| Consumer integration | How relevant is the size and makeup of the existing customer base for brands that platformize? Which parts of the value creation should brands "make" versus "earn"? How can brands acquire new consumers beyond its existing customer base? Which consumer characteristics determine their willingness to engage in intense crowdsending? How can these consumers be identified and targeted? How does compensation of consumer-created value affect the platform and its consumer relationships?
| Emerging area |
| Branding | Does the joint creation of consumer experiences on flagship platforms diminish brands' importance and dilute their image? And does the shift away from product-centricity harm especially traditional manufacturing brands? Is this development moderated by the intensity of crowdsourcing and -sending increasing the relevance of the brand in high intensity contexts? How does our knowledge from brand extensions and cobranding transfer to platformized brands? Specifically, how is brand positioning in this context being impacted? In which direction and how far can brands expand into a category space without harming their reputation? And how can brands mitigate this through careful platform rebranding? How prominently should the brand be featured in interactions with consumers, among consumers, and between consumers and third parties?
| New key challenge |
| Process-Related | |
| Consumer governance | How much empowerment versus restriction is optimal under which relationship state? How does this vary by building block? Which individual consumer–supplier relationships occur as part of the platforms subassemblages? How do they affect the platform at large? How do brand flagship platforms affect collectives (e.g., families) rather than individual consumers? Should the platform account for collectives in their value-creation process? Should they focus on providing solutions for individuals or collectives? How open does the platform need to be to engage consumers but not hurt its own brand goals?
| Emerging area |
| Third-party governance | How much openness versus restriction is optimal under which relationship state? How does this vary by building block? How should platforms be shielded against "hostile takeovers" or third parties acquiring new customers through the platform and then defecting with them? Is the threat of hostile takeovers less pronounced in more intimate consumer–platform relationships? How should the platform brand mitigate adverse effects through service failures by third-party brands?
| Emerging area |
| Fostering consumer interactions | How can game design inform the creation of mechanisms to foster interactions? How effective are gamified mechanisms in fostering consumer interactions depending on the relationship state? How can consumers be transformed from passive crowdsourcers to active crowdsenders? Which crowdsending activities are more and less rewarding for consumers? Are there consumer segments motivated by distinct factors that increase their engagement in crowdsending? Which boundaries should the platform obey to avoid feelings of intrusiveness or exploitation? What detrimental consequences could fostering consumer crowdsourcing and -sending activities have on consumer behavior and well-being?
| New key challenge |
| Outcome-Related | |
| Individual relationship states | Which brands specifically are able to evolve to providing nurturing partnerships? And how far can brands venture into the broader category space before they are stretched too thin? How far can a platform advance the relationship before consumers respond with reactance? How to identify consumers that are receptive to a nurturing partnership? And which consumer characteristics predict this receptiveness? How can brands intervene to elicit the desired relationship state? Which brand factors (e.g., brand strength, positioning) influence which relationship state(s) manifest? Can brands develop a market to evolve instrumental ad hoc relationships toward relational states?
| Emerging area, evergreen, niche area |
| Relational outcome measurement | How can we measure relational outcomes and the role of the platform in consumers' goal pursuit? How do we need to extend existing brand attitude scales? How can we use behavioral big data to infer consumers' goal pursuit and relationship state? How can brands measure and attribute customer engagement value?
| Evergreen |
| Monetization | How can platforms monetize relationships without scaring off consumers? How does monetization affect relationships with consumers? When should brands employ subscription models over pay per use? What is the monetary value of the data and insights generated through the platform? How do consumer privacy rules play out? How should privacy be balanced with monetization goals? How does this vary by relationship state?
| Evergreen |
| Business model transformation | Does the conceptualization extend to born-digital brands? Are they born with a platform mindset or do they revert to a pipeline approach? Which brand, company, and market characteristics determine a successful brand platformization? Which brands should platformize, and to what degree? And which brands should assume a platform complementor role? How to transform: with a "big bang" or gradually over time? And what are contingency factors?
| Evergreen |
How to assemble elements of consumer inspiration garners strong managerial interest but is only sparsely researched ([15]). Further studies could examine different types (e.g., integrated narratives vs. open sequences, professional vs. improvised), sources (e.g., consumers, designers, product brands), and presentation modes (e.g., music, picture, video, text) of inspiring content and study how to optimally integrate these in the platform assemblage. Research might also expand our conceptualization of building blocks by exploring whether and how the weaving together of building blocks gives rise to positive or negative interactions. For example, creating a strong inspiration block alone may elicit less intense crowdsourcing and -sending than if paired with community features that allow for collaborative idea development. Lastly, managers view the role of branding on flagship platforms as a key challenge to address. Important questions that arise are whether the joint creation of consumer experience on flagship platforms diminishes brands' importance and dilutes their image, and whether the shift away from product-centricity could harm traditional manufacturing brands in particular. However, increasing crowdsourcing and -sending intensities could also reinforce brands' importance, as they may become the connective tissue that embodies and communicates the platform's higher-level goal. These perspectives complement the work by Swaminathan et al. (2020), who argue that the importance of brands in access-based, instrumental consumption contexts—characteristic of the ad hoc relationships we describe—may diminish. Research could examine these issues, using relationship states as relevant moderators to branding outcomes.
With platformization being uncharted territory for most brands, several issues emerge that concern the governance of flagship platforms. For example, how fast and to what degree should the platform open up to consumers and third-party businesses? And how does the optimal degree of openness vary by building block? Governance issues may also pertain to the individual subassemblages that emerge on the platform. That is, while our discussion focused on consumer–platform interactions, it would be fruitful to explore the individual subassemblages emerging from consumer–supplier interactions and their consequences for the platform. For example, consumers may build close relationships with third parties that may even culminate in defection from the platform ([88]). Thus, future research should explore how brands can monitor consumer–supplier relationships and deploy governance mechanisms that prevent such exploitative behaviors. Researchers should also consider the moderating role of relationship states, as exploitative behaviors may be more common in the ad hoc consumer–platform relationships studied by [88] but less likely in intimate consumer–platform relationships. Future studies might also extend our conceptualization beyond the individual consumer to collectives such as families and groups of friends as relevant actors on the platform ([60]). Increasing embeddedness in consumers' lives makes the platform more likely to affect their immediate social environment, raising the question of how interactions and interventions may affect such collectives rather than individuals.
Fostering consumer interactions reflects a key challenge that researchers can explore on different levels. For one, it would be important to garner insights about coherent incentive systems, for example, based on gamification approaches, to spur consumer crowdsourcing and -sending. Gamified mechanisms may prove especially useful for intimate relationship states in which monetary incentives are ineffective or even detrimental as consumers tend to apply less exchange-focused norms ([ 2]). In addition, gamified mechanisms can create unpredictable experiences that dynamically adapt to consumers' evolving skill levels and keep them engaged over extended periods of time ([71]), making them well suited for more intimate relationship states. Thus, future research could analyze the effectiveness of gamified mechanisms conditional on relationship states. The optimal degree of brand intermediation of these efforts also warrants further investigation.
The emerging interest in catalyzing relationships and nurturing partnerships suggests that platform thinking in terms of "buying" and "earning" products, services, and content is still new to most managers. Accordingly, many important questions have yet to be answered. For example, which kinds of brands will be able to create nurturing partnerships, and how far can brands venture into the broader category space before they are stretched too thin? And in turn, up to which point do consumers allow brand flagship platforms to conquer their daily lives and cocreate their identities before responding with reactance to regain sovereignty? Such reactance is one of many (de)stabilizing processes that may cause relationship states to change over time. While we describe some of these processes, scholars could analyze in detail which specific interventions can transform a relationship into an (un)desired state.
Finally, although our discussion focuses on offline-born product brands, future businesses will be born into a digital world. Research should explore how our conceptualization extends to these brands and whether they internalize platformized thinking, seek their salvation in the role of platform complementors, or even revert to the more traditional linear approach. Because the digital platformization of industries is here to stay, brands will be required to take a stance.
sj-pdf-1-jmx-10.1177_00222429211054073 - Supplemental material for The Platformization of Brands
Supplemental material, sj-pdf-1-jmx-10.1177_00222429211054073 for The Platformization of Brands by Julian R.K. Wichmann, Nico Wiegand and Werner J. Reinartz in Journal of Marketing
Footnotes 1 The first two authors contributed equally to this work. The article is based on the first author's dissertation.
2 Markus Giesler
3 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
4 The author(s) received no financial support for the research, authorship and/or publication of this article.
5 Julian R.K. Wichmann https://orcid.org/0000-0003-0205-5243 Nico Wiegand https://orcid.org/0000-0002-7348-7716 Werner J. Reinartz https://orcid.org/0000-0002-2440-3117
6 Naturally, not all activities on digital platforms involve the crowd (e.g., covertly sharing private data to feed recommendation algorithms). However, crowd-based activities are key to distinguish platform from pipeline offerings, thus chiefly contributing to the conceptual uniqueness and novelty of brand flagship platforms.
7 We thus distinguish between assemblages on two levels: the brand's infrastructure and governance decisions giving rise to the platform assemblage and the consumer's idiosyncratic interactions with the platform in the context of consumer–platform subassemblages ([21]; [31]).
8 Sharing intimate data may also constitute intense crowdsending if it is directed at the crowd and creates value for it (e.g., uploading performance data as part of a leaderboard or competition on athletics platforms).
9 Intense crowdsending in the form of sharing intimate data (e.g., as part of the benchmarking block) is less common in this relationship state, as consumers receive little in return because of low crowdsourcing intensity ([49]).
References Afuah Allan , Tucci Christopher L.. (2012), " Crowdsourcing as a Solution to Distant Search ," Academy of Management Review , 37 (3), 355 – 75.
Aggarwal Pankaj. (2004), " The Effects of Brand Relationship Norms on Consumer Attitudes and Behavior ," Journal of Consumer Research , 31 (1), 87 – 101.
Albuquerque Paulo , Pavlidis Polykarpos , Chatow Udi , Chen Kay-Yut , Jamal Zainab. (2012), " Evaluating Promotional Activities in an Online Two-Sided Market of User-Generated Content ," Marketing Science , 31 (3), 406 – 32.
Allen B.J. , Chandrasekaran Deepa , Basuroy Suman. (2018), " Design Crowdsourcing: The Impact on New Product Performance of Sourcing Design Solutions from the 'Crowd,' " Journal of Marketing , 82 (2), 106 – 23.
Altman Irwin , Taylor Dalmas A.. (1973), Social Penetration: The Development of Interpersonal Relationships. New York : Holt, Rinehart and Winston.
Aral Sinan , Walker Dylan. (2014), " Tie Strength, Embeddedness, and Social Influence: A Large-Scale Networked Experiment ," Management Science , 60 (6), 1352 – 70.
Arnould Eric J. , Thompson Craig J.. (2005), " Consumer Culture Theory (CCT): Twenty Years of Research ," Journal of Consumer Research , 31 (4), 868 – 82.
Aron Arthur , Aron Elaine N. , Tudor Michael , Nelson Greg. (1991), " Close Relationships as Including Other in the Self ," Journal of Personality and Social Psychology , 60 (2), 241 – 53.
Atluri Venkat , Dietz Miklós , Henke Nicolaus. (2017), " Competing in a World of Sectors Without Borders ," McKinsey Quarterly , 54 (3), 1 – 14.
Banerjee Shrabastee , Dellarocas Chrysanthos , Zervas Georgios. (2021), " Interacting User-Generated Content Technologies: How Questions and Answers Affect Consumer Reviews ," Journal of Marketing Research , 58 (4), 742 – 61.
Bardhi Fleura , Eckhardt Giana M.. (2012), " Access-Based Consumption: The Case of Car Sharing ," Journal of Consumer Research , 39 (4), 881 – 98.
Bardhi Fleura , Eckhardt Giana M.. (2017), " Liquid Consumption ," Journal of Consumer Research , 44 (3), 582 – 97.
Belk Russell W. (1988), " Possessions and the Extended Self ," Journal of Consumer Research , 15 (2), 139 – 68.
Bhattacharya C.B. , Sen Sankar. (2003), " Consumer–Company Identification: A Framework for Understanding Consumers' Relationships with Companies ," Journal of Marketing , 67 (2), 76 – 88.
Böttger Tim , Rudolph Thomas , Evanschitzky Heiner , Pfrang Thilo. (2017), " Customer Inspiration: Conceptualization, Scale Development, and Validation ," Journal of Marketing , 81 (6), 116 – 31.
Boudreau Kevin. (2017), " Platform Boundary Choices & Governance: Opening-Up While Still Coordinating and Orchestrating ," in Advances in Strategic Management , Furman J. , Gawer A. , Silverman B.S. , Stern S. , eds. Bingley, UK: Emerald Publishing Limited , 227 – 97.
Brodie Roderick J. , Hollebeek Linda D. , Jurić Biljana , Ilić Ana. (2011), " Customer Engagement: Conceptual Domain, Fundamental Propositions, and Implications for Research ," Journal of Service Research , 14 (3), 252 – 71.
Broekhuizen Thijs L.J. , Emrich Oliver , Gijsenberg Maarten J. , Broekhuis Manda , Donkers Bas , Sloot Laurens M.. (2021), " Digital Platform Openness: Drivers, Dimensions and Outcomes ," Journal of Business Research , 122 , 902 – 14.
Bughin Jaques , Catlin Tanguy , Dietz Miklós. (2019), " The Right Digital-Platform Strategy ," McKinsey Quarterly (May 7), https://www.mckinsey.com/business-functions/mckinsey-digital/our-insights/the-right-digital-platform-strategy.
Constantinides Panos , Henfridsson Ola , Parker Geoffrey G.. (2018), " Introduction—Platforms and Infrastructures in the Digital Age ," Information Systems Research , 29 (2), 381 – 400.
DeLanda Manuel. (2016), Assemblage Theory. Edinburgh : Edinburgh University Press.
Epp Amber M. , Price Linda L.. (2011), " Designing Solutions Around Customer Network Identity Goals ," Journal of Marketing , 75 (2), 36 – 54.
Fournier Susan. (1998), " Consumers and Their Brands: Developing Relationship Theory in Consumer Research ," Journal of Consumer Research , 24 (4), 343 – 73.
Füller Johann , Matzler Kurt , Hoppe Melanie. (2008), " Brand Community Members as a Source of Innovation ," Journal of Product Innovation Management , 25 (6), 608 – 19.
Gielens Katrijn , Steenkamp Jan-Benedict E.M.. (2019), " Branding in the Era of Digital (Dis)Intermediation ," International Journal of Research in Marketing , 36 (3), 367 – 84.
Gino Francesca , Staats Bradley R.. (2012), " The Microwork Solution: A New Approach to Outsourcing Can Support Economic Development—and Add to Your Bottom Line ," Harvard Business Review , 90 (12), 92 –9 6.
Granovetter Mark S. (1973), " The Strength of Weak Ties ," American Journal of Sociology , 78 (6), 1360 – 80.
Gruner Richard L. , Homburg Christian , Lukas Bryan A.. (2014), " Firm-Hosted Online Brand Communities and New Product Success ," Journal of the Academy of Marketing Science , 42 (1), 29 – 48.
Hagiu Andrei , Wright Julian. (2015), " Multi-Sided Platforms ," International Journal of Industrial Organization , 43 , 162 – 74.
Hemetsberger Andrea. (2012), " Crowdsourcing ," in The Routledge Companion to Digital Consumption , Belk R.W. , Llamas R. , eds. London; New York : Routledge , 159 – 70.
Hoffman Donna L. , Novak Thomas P.. (2018), " Consumer and Object Experience in the Internet of Things: An Assemblage Theory Approach ," Journal of Consumer Research , 44 (6), 1178 –1 204.
Holbrook Morris B. (1999), Consumer Value: A Framework for Analysis and Research. London : Routledge.
Howe Jeff. (2006), " The Rise of Crowdsourcing ," Wired , 14 (6), 1 – 4.
Howe Jeff. (2008), Crowdsourcing: How the Power of the Crowd Is Driving the Future of Business. London : Random House.
Hoyer Wayne D. , Chandy Rajesh , Dorotic Matilda , Krafft Manfred , Singh Siddharth S.. (2010), " Consumer Cocreation in New Product Development ," Journal of Service Research , 13 (3), 283 – 96.
Hudson Simon , Huang Li , Roth Martin S. , Madden Thomas J.. (2016), " The Influence of Social Media Interactions on Consumer–Brand Relationships: A Three-Country Study of Brand Perceptions and Marketing Behaviors ," International Journal of Research in Marketing , 33 (1), 27 – 41.
Huffman Cynthia , Ratneshwar Srinivasan , Glen Mick David. (2003), " Consumer Goal Structures and Goal-Determination Processes: An Integrative Framework ," in The Why of Consumption , Huffman C. , Mick D.G. , Ratneshwar S. , eds. London; New York : Routledge , 29 – 55.
Iansiti Marco , Levien Roy. (2004), " Strategy as Ecology ," Harvard Business Review , 82 (3), 68 – 78.
James Tabitha L. , Deane Jason K. , Wallace Linda. (2019), " Using Organismic Integration Theory to Explore the Associations Between Users' Exercise Motivations and Fitness Technology Feature Set Use ," MIS Quarterly , 43 (1), 287 – 312.
Jeppesen Lars Bo , Frederiksen Lars. (2006), " Why Do Users Contribute to Firm-Hosted User Communities? The Case of Computer-Controlled Music Instruments ," Organization Science , 17 (1), 45 – 63.
Johnson Allison R. , Matear Maggie , Thomson Matthew. (2011), " A Coal in the Heart: Self-Relevance as a Post-Exit Predictor of Consumer Anti-Brand Actions ," Journal of Consumer Research , 38 (1), 108 – 25.
Katz Michael L. , Shapiro Carl. (1985), " Network Externalities, Competition, and Compatibility ," American Economic Review , 75 (3), 424 – 40.
Kelly Kevin. (2016), The Inevitable: Understanding the 12 Technological Forces that Will Shape our Future. New York : Viking.
Kozinets Robert V. , De Valck Kristine , Wojnicki Andrea C. , Wilner Sarah J.S.. (2010), " Networked Narratives: Understanding Word-of-Mouth Marketing in Online Communities ," Journal of Marketing , 74 (2), 71 – 89.
Kozinets Robert V , Ferreira Daniela Abrantes , Chimenti Paula. (2021), " How Do Platforms Empower Consumers? Insights from the Affordances and Constraints of Reclame Aqui ," Journal of Consumer Research (published online March 15), https://doi.org/10.1093/jcr/ucab014.
Kozinets Robert V. , Hemetsberger Andrea , Schau Hope Jensen. (2008), " The Wisdom of Consumer Crowds: Collective Innovation in the Age of Networked Marketing ," Journal of Macromarketing , 28 (4), 339 – 54.
Labrecque Lauren I., vor dem Esche Jonas , Mathwick Charla , Novak Thomas P. , Hofacker Charles F.. (2013), " Consumer Power: Evolution in the Digital Age ," Journal of Interactive Marketing , 27 (4), 257 – 69.
Marder Ben , Joinson Adam , Shankar Avi , Thirlaway Kate. (2016), " Strength Matters: Self-Presentation to the Strongest Audience Rather Than Lowest Common Denominator When Faced with Multiple Audiences in Social Network Sites ," Computers in Human Behavior , 61 , 56 – 62.
Martin Kelly D. , Murphy Patrick E.. (2017), " The Role of Data Privacy in Marketing ," Journal of the Academy of Marketing Science , 45 (2), 135 – 55.
Maslow Abraham Harold. (1943), " A Theory of Human Motivation ," Psychological Review , 50 (4) , 370 – 96.
Nambisan Satish , Baron Robert A.. (2009), " Virtual Customer Environments: Testing a Model of Voluntary Participation in Value Co-Creation Activities ," Journal of Product Innovation Management , 26 (4), 388 – 406.
Novak Thomas P. , Hoffman Donna L.. (2019), " Relationship Journeys in the Internet of Things: A New Framework for Understanding Interactions Between Consumers and Smart Objects ," Journal of the Academy of Marketing Science , 47 (2), 216 – 37.
Parker Geoffrey , Van Alstyne Marshall. (2018), " Innovation, Openness, and Platform Control ," Management Science , 64 (7), 3015 – 32.
Parker Geoffrey , Van Alstyne Marshall , Choudary Sangeet Paul. (2016), Platform Revolution: How Networked Markets are Transforming the Economy and How to Make Them Work for You. New York : W.W. Norton & Company.
Parmentier Marie-Agnès , Fischer Eileen. (2015), " Things Fall Apart: The Dynamics of Brand Audience Dissipation ," Journal of Consumer Research , 41 (5), 1228 – 51.
Payne Adrian F. , Storbacka Kaj , Frow Pennie. (2008), " Managing the Co-Creation of Value ," Journal of the Academy of Marketing Science , 36 (1), 83 – 96.
Perren Rebeca , Kozinets Robert V.. (2018), " Lateral Exchange Markets: How Social Platforms Operate in a Networked Economy ," Journal of Marketing , 82 (1), 20 – 36.
Pieters Rik , Baumgartner Hans , Allen Doug. (1995), " A Means-End Chain Approach to Consumer Goal Structures ," International Journal of Research in Marketing , 12 (3), 227 – 44.
Prahalad C.K. , Ramaswamy Venkat. (2004), " Co-Creation Experiences: The Next Practice in Value Creation ," Journal of Interactive Marketing , 18 (3), 5 – 14.
Price Linda , Epp Amber M.. (2015), " The Heterogeneous and Open-Ended Project of Assembling Family ," in Assembling Consumption: Researching Actors, Networks and Markets , Canniford R. , Bajde D. , eds. New York : Routledge , 119 – 34.
Ramaswamy Venkat , Ozcan Kerimcan. (2016), " Brand Value Co-Creation in a Digitalized World: An Integrative Framework and Research Implications ," International Journal of Research in Marketing , 33 (1), 93 – 106.
Ramaswamy Venkat , Ozcan Kerimcan. (2018), " Offerings as Digitalized Interactive Platforms: A Conceptual Framework and Implications ," Journal of Marketing , 82 (4), 19 – 31.
Rangaswamy Arvind , Moch Nicole , Felten Claudio , van Bruggen Gerrit , Wieringa Jaap E. , Wirtz Jochen. (2020), " The Role of Marketing in Digital Business Platforms ," Journal of Interactive Marketing , 51 , 72 – 90.
Reinartz Werner , Wiegand Nico , Imschloss Monika. (2019), " The Impact of Digital Transformation on the Retailing Value Chain ," International Journal of Research in Marketing , 36 (3), 350 – 66.
Rochet Jean-Charles , Tirole Jean. (2003), " Platform Competition in Two-Sided Markets ," Journal of the European Economic Association , 1 (4), 990 – 1029.
Ruckenstein Minna , Pantzar Mika. (2017), " Beyond the Quantified Self: Thematic Exploration of a Dataistic Paradigm ," New Media & Society , 19 (3), 401 – 18.
Schau Hope Jensen , Muñiz Albert M. , Arnould Eric J.. (2009), " How Brand Community Practices Create Value ," Journal of Marketing , 73 (5), 30 – 51.
Schlosser Ann E. (2005), " Posting Versus Lurking: Communicating in a Multiple Audience Context ," Journal of Consumer Research , 32 (2), 260 – 65.
Sherman Alex. (2019), " Netflix CEO Reed Hastings Says Subscriber Numbers Aren't the Right Metric to Track Competition ," CNBC (November 6) , https://www.cnbc.com/2019/11/06/netflix-ceo-reed-hastings-subscriber-numbers-are-not-that-important.html.
Sheth Jagdish N. , Newman Bruce I. , Gross Barbara L.. (1991), " Why We Buy What We Buy: A Theory of Consumption Values ," Journal of Business Research , 22 (2), 159 – 70.
Siebert Anton , Gopaldas Ahir , Lindridge Andrew , Simões Cláudia. (2020), " Customer Experience Journeys: Loyalty Loops Versus Involvement Spirals ," Journal of Marketing , 84 (4), 45 – 66.
Smith J. Brock , Colgate Mark. (2007), " Customer Value Creation: A Practical Framework ," Journal of Marketing Theory and Practice , 15 (1), 7 – 23.
Spaeth Sebastian , von Krogh Georg , He Fang. (2015), " Research Note—Perceived Firm Attributes and Intrinsic Motivation in Sponsored Open Source Software Projects ," Information Systems Research , 26 (1), 224 – 37.
Statista (2020), " Distribution of Branded and Unbranded Search Volume of Global Brands and SMBs Between August 2018 and August 2019 ," (April 21), https://www.statista.com/statistics/1143368/distribution-branded-search-volume-global-brands-smbs/.
Surowiecki James. (2004), The Wisdom of Crowds: Why the Many Are Smarter Than the Few and How Collective Wisdom Shapes Business, Economies, Societies, and Nations. New York : Doubleday & Co.
Swaminathan Vanitha , Sorescu Alina , Steenkamp Jan-Benedict E.M. , Gibson O'Guinn Thomas Clayton , Schmitt Bernd. (2020), " Branding in a Hyperconnected World: Refocusing Theories and Rethinking Boundaries , " Journal of Marketing , 84 (2), 24 – 46.
Sweeney Jillian C. , Soutar Geoffrey N.. (2001), " Consumer Perceived Value: The Development of a Multiple Item Scale ," Journal of Retailing , 77 (2), 203 – 20.
Täuscher Karl , Laudien Sven M.. (2018), " Understanding Platform Business Models: A Mixed Methods Study of Marketplaces ," European Management Journal , 36 (3), 319 – 29.
Thompson Scott A. , Kim Molan , Smith Keith Marion. (2016), " Community Participation and Consumer-to-Consumer Helping: Does Participation in Third Party–Hosted Communities Reduce One's Likelihood of Helping? " Journal of Marketing Research , 53 (2), 280 – 95.
Toma Catalina L. , Hancock Jeffrey T.. (2013), " Self-Affirmation Underlies Facebook Use ," Personality and Social Psychology Bulletin , 39 (3), 321 – 31.
Tuli Kapil R. , Kohli Ajay K. , Bharadwaj Sundar G.. (2007), " Rethinking Customer Solutions: From Product Bundles to Relational Processes ," Journal of Marketing , 71 (3), 1 – 17.
Under Armour (2020), " Under Armour Enters Into Definitive Agreement to Sell the MyFitnessPal Platform to Francisco Partners ," Cision PR Newswire (October 30), https://www.prnewswire.com/news-releases/under-armour-enters-into-definitive-agreement-to-sell-the-myfitnesspal-platform-to-francisco-partners-301163499.html.
Van Alstyne Marshall W. , Parker Geoffrey G. , Choudary Sangeet P.. (2016), " Pipelines, Platforms, and the New Rules of Strategy ," Harvard Business Review , 94 (4), 54 – 62.
Vargo Stephen L. , Lusch Robert F.. (2004), " Evolving to a New Dominant Logic for Marketing ," Journal of Marketing , 68 (1), 1 – 17.
Von Hippel Eric. (1994), " 'Sticky Information' and the Locus of Problem Solving: Implications for Innovation ," Management Science , 40 (4), 429 – 39.
Wolf Gary. (2010), " The Data-Driven Life ," The New York Times Magazine (April 28), https://www.nytimes.com/2010/05/02/magazine/02self-measurement-t.html.
Wu Yue , Zhang Kaifu , Padmanabhan V.. (2018), " Matchmaker Competition and Technology Provision ," Journal of Marketing Research , 55 (3), 396 – 413.
Xie Chunyan , Bagozzi Richard P. , Troye Sigurd V.. (2008), " Trying to Prosume: Toward a Theory of Consumers as Co-Creators of Value ," Journal of the Academy of Marketing Science , 36 (1), 109 – 22.
Zhao Yuxiang , Zhu Qinghua. (2014), " Evaluation on Crowdsourcing Research: Current Status and Future Direction ," Information Systems Frontiers , 16 (3), 417 – 34.
Zhou Qiang (Kris) , Allen B.J. , Gretz Richard T. , Houston Mark B.. (2021), " Platform Exploitation: When Service Agents Defect with Customers From Online Service Platforms ," Journal of Marketing (published online September 21), https://doi.org/10.1177/00222429211001311.
Zhu Feng , Furr Nathan. (2016), " Products to Platforms: Making the Leap ," Harvard Business Review , 94 (4), 72 – 8.
Zhu Feng , Iansiti Marco. (2019), " Why Some Platforms Thrive and Others Don't ," Harvard Business Review , 97 (1), 118 – 25.
~~~~~~~~
By Julian R.K. Wichmann; Nico Wiegand and Werner J. Reinartz
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 126- The Rise of New Technologies in Marketing: A Framework and Outlook. By: Hoffman, Donna L.; Moreau, C. Page; Stremersch, Stefan; Wedel, Michel. Journal of Marketing. Jan2022, Vol. 86 Issue 1, p1-6. 6p. 2 Diagrams. DOI: 10.1177/00222429211061636.
- Database:
- Business Source Complete
The Rise of New Technologies in Marketing: A Framework and Outlook
As a scholarly field, marketing has a long tradition of studying the adoption of new technologies. This attention is certainly warranted, as studies consistently demonstrate that, compared with firms that do not invest heavily in new technology, those that do are more agile and enjoy a strong competitive advantage ([10]). However, what has received less attention in the literature is how new technologies give rise to innovations in marketing techniques, tools, and strategies themselves. In particular, there is a need for marketing scholars to develop theoretical paradigms of how marketers use technologies to develop a competitive advantage.
This special issue on "New Technologies in Marketing" presents cutting-edge scholarly research that recognizes the foundational role of new technologies in driving marketing theory and practice. The articles in the special issue study a broad range of new technologies, and we hope they will stimulate further research concerning new technologies in marketing and their application in practice. In this editorial, we provide several frameworks for thinking about how new technology affects the marketing discipline. These frameworks serve to organize the portfolio of articles in the special issue, identify potential gaps worthy of further study, and propose an agenda for future research.
Prior research has defined "technology" as scientific knowledge and its applications to useful purposes (see, e.g., [11]). This definition recognizes that technology can relate both to the product or the service that follows from the scientific knowledge and to the knowledge itself. In doing so, it avoids the necessity of distinguishing between the product or service (e.g., chatbots) and the technology (e.g., artificial intelligence [AI]) it encompasses, which is, at times, impossible to do ([ 8]).
Because technology matures over time, we also define the term "new" as referring to recent applications of scientific knowledge that have not been replaced by others. In other words, technology is "new" when it is early in the adoption cycle for firms and/or consumers (i.e., in the innovator or early adopter phase).
The articles in this special issue examine a range of new marketing technologies that fall in one or both points of the adoption cycle. As such, these articles necessarily employ diverse research methodologies. Specifically, technologies that are further along are more likely to have produced hard data because a sufficient number of firms or consumers have adopted the technology to enable the antecedents or consequences of adoption to be empirically observed and quantitatively analyzed (e.g., augmented reality as in Tan, Chandukala, and Reddy [2022], livestream sales calls in Bharadwaj et al. [2022]). Other technologies fall earlier in the adoption cycle, and their adoption antecedents or consequences can only be established experimentally (e.g., chatbots as in [ 4], AI-based recommendations as in [12]) or conceptually (e.g., genetics as in [ 5], platforms as in [19], avatars as in [14]).
Integrating all these considerations, we define new technologies in marketing[ 1] as "scientific knowledge and/or its application in the early adoption cycle for firms and/or consumers with the potential to influence the activity, institutions, and processes for creating, communicating, delivering, and exchanging offerings that have value for customers, clients, partners, and society at large."
At a high level of abstraction, we observe that new technologies impact marketing in four broad, interconnected ways, as diagrammed in Figure 1. Specifically, new technology ( 1) supports new forms of interaction among consumers and firms, ( 2) provides new types of data that enable new analytic methods, ( 3) creates marketing innovations, and ( 4) requires new strategic marketing frameworks. It is important to keep in mind that different technologies can serve these multiple functions at the same time and to varying degrees. For this reason, some articles in the special issue appear in more than one cell in Figure 1.
Graph: Figure 1. Four ways new technologies impact marketing.
We start in the center of Figure 1 with how new technology may enable new forms of consumer-to-consumer, consumer-to-firm, firm-to-consumer, and firm-to-firm interactions. Many firms now enable direct consumer-to-consumer interactions by engaging consumers around brands. For example, brands such as Nike and Adidas have developed digital platforms to promote interactions among communities of runners and coaches as well as third parties ([19]).
New technologies have often been effectively deployed to improve firm–consumer interactions by providing new marketing tools. For example, AI is a powerful engine in replacing human representatives of the firm with machine agents, facilitating firm–consumer interactions via "word of machine" ([12]). Anthropomorphized chatbots can influence consumer response in consumer-initiated service interactions ([ 4]). In addition, avatars are increasingly used in firm–consumer interactions, where the extent of an avatar's form and behavioral realism is a major determinant of its effectiveness ([14]). Augmented reality (AR) is used in retailing to facilitate firm–consumer interactions, which, as a "try before you buy" technology, is especially effective when consumers are uncertain about products ([17]). Computer vision and facial recognition methods present new tools for marketers that can be used to enhance the effectiveness of livestream personal selling ([ 1]).
New technologies also give rise to new data and spawn new analytic methods, as shown on the left side of Figure 1. For example, [ 1] propose an analytic framework that utilizes computer vision methods to analyze the effectiveness of salespeople's facial expressions in livestream selling. [ 3] offer an approach for firms to assess the potential of new technologies to make informed product launch and product retirement decisions. Further, [ 5] portray a future in which consumers may consent to the use of their genetic data to improve customer targeting and new product development. These studies show that by altering consumer-to-consumer and consumer-to-firm interactions, new technologies produce new forms of data. In turn, these new forms often require the development of new methods or the adaptation of existing ones to process or analyze these data.
The top of Figure 1 highlights the potential for new technologies to provide new marketing tools and techniques that lead to innovations in the marketing of products and services. For example, [ 1] show how personal selling via livestreaming can be optimized with computer vision AI. [12] analyze the effectiveness of AI-based "word of machine," while [ 4] examine chatbot effectiveness. [17] analyze the effectiveness of AR in retailing. Studies like these reveal that new technologies enable marketers to develop and deploy new tools that render the marketing of products and services more effective.
Finally, new technologies enable new marketing strategies and strategic frameworks, as shown on the right side of Figure 1. [19] conceptualize digital platforms as places for consumer crowdsourcing and crowdsending of products and services. [14] propose a typology of avatars that guides marketers in their decisions about how to design and deploy avatars. [ 5] propose a framework that integrates the impact of genetics into consumer behavior theory and use that framework to provide an overview of marketing uses of genetic data. These articles show the value of new strategic frameworks in understanding the impact of new technologies on the marketing domain. They also provide guidance for how to formulate the most relevant research questions.
Having discussed the four fundamental ways that new technologies are influencing marketing practice, we now offer a framework for understanding how they are improving marketing decision making and associated firm and marketplace dynamics. Figure 2 presents a flywheel demonstrating the energy stored and deployed when firms invest in new marketing technologies to yield increasing returns. Starting on the left, new technology may enhance the richness, quality, and volume of market and consumer data (see Figure 1). For instance, the explosive growth of digital devices and software applications has created data streams that capture how consumers think, feel, behave, and interact with other consumers and firms at various points along the customer journey ([18]). Some data that may have previously been out of reach for many firms, such as eye movement, speech, facial recognition, and genetic data, may become ubiquitous as the cost to collect and analyze them rapidly decreases (see, e.g., [ 1]; [ 5]). Digital data-capture technology, which produces data on consumer–firm interactions via images, video, speech, and text, among others, has also enabled large-scale field experiments and A/B tests that allow companies to assess the causal effects of their marketing actions. With these experiments, marketers can optimize website designs, effectively retarget advertising, evaluate the effects of new marketing tools, and attribute effects to marketing actions throughout the customer journey.
Graph: Figure 2. New technology improves marketing decision making.
Moving clockwise around the flywheel, we argue that the increased availability of rich data leads to new and better methods for consumer and firm decision making (top of Figure 2). To a certain degree, data quality can be a substitute for model complexity. For instance, in A/B testing, the data-generating mechanism is controlled by the researcher or firm so that simpler models may be used (see [17]). However, the idea that simpler models are always adequate is misguided. Increased data richness (e.g., reviews, search, blogs, location, images, video, speech, eye, hand, head and body movements, genetic data) often requires more complex models or machine learning approaches. Further, greater data volume reduces sampling and measurement errors. Both the increased richness and volume of data allow for superior predictive performance of machine learning methods. As a case in point, [ 1] demonstrate how a large-scale application of computer vision methods for image analysis, coupled with sophisticated statistical techniques, leads to superior predictive performance of sales outcomes.
The right side of the flywheel highlights that better methods enable marketers to derive new and valuable insights. For example, [ 3] develop a method that gives managers insight into whether to adopt a new technology, continue capitalizing on the old technology, or invest in both the new and old technologies. Further, [17] research demonstrates that AR can reduce consumers' product uncertainty and thus improve marketing and sales outcomes.
At the bottom of the flywheel in Figure 2, we highlight how better insights derived from new technology can enable better and faster decision making by both consumers and firms. AR in retail can better inform and educate consumers and therefore improve their decision making on products and services ([17]). Chatbots enable real-time interactions with the firm that can provide consumers with insights and information to improve their satisfaction, firm evaluations, and purchase intentions ([ 4]). These interactions, in turn, generate more and richer data to continue the cycle.
This section pulls the different elements of our perspective together, emphasizing key learnings for marketers and marketing scholars. We also identify several high-potential future research directions. These recommendations are grounded in our own work in this area and fueled by editor efforts that preceded the special issue (e.g., the 2019 Theory+Practice in Marketing [TPM] conference focused on New Technologies in Marketing and a special session on the same topic at the 2019 European Marketing Academy Conference).
Technology-enabled interactions, methods, innovations, and frameworks (Figure 1), along with the market and marketing knowledge they generate (Figure 2), offer marketers the opportunity for real-time or automated decision making. For example, AI is beginning to play an important role in automatically generating tailored offerings based on individual consumers' search behaviors and navigational histories. Similarly, AI is starting to figure into firms' new product development decisions (e.g., go/no-go launch decisions). The current automation versus augmentation debate in AI ([15]) suggests that there is enormous opportunity for research on the types of human/AI collaborative teams that will be most effective in different marketing contexts.
Technology-enabled interactions also offer marketers opportunities to observe consumers engaged in new ways with products, brands, stores, firms, and other consumers. These observations provide a deeper understanding of consumers' relationships and preferences and give companies the opportunity to create new sources of value for both the consumer and the firm. In particular, personalization and recommendation systems will remain a key area for future research.
As the trend toward automated marketing decision making accelerates, it seems particularly important to establish the boundaries of increasingly popular machine learning and AI methods. For example, how sensitive are machine learning applications in marketing to small perturbations (adversarial attacks) of the input? How well do these systems perform on data/problems that fall outside the domain of training data? How can the interpretation of machine learning approaches applied to marketing problems be enhanced? What is the role of machine learning in making causal inferences from quasi- and nonexperimental data?
Technology is enabling consumers to interact with products, firms, and each other in virtual reality (VR) and AR. While many of the well-established theories in consumer behavior may extend quite naturally into virtual spaces, many may need to be updated significantly to accommodate consumers' search, choice, and consumption practices. For example, how is context-relevant information (e.g., product information, recommendations) presented in virtual or augmented environments processed? How do embodiment and the experience of presence facilitate information acquisition and consumer decision making in virtual contexts? What are the consequences of VR/AR engagement, including for self-image, anxiety, and interpersonal interactions? What roles do sharing platforms play in the development of consumers' self-presentation strategies, political views, body image, and values (e.g., materialism)?
Another important trend in consumer behavior involves the adoption of autonomous products. Traditionally, consumers have purchased products that assist them in accomplishing particular tasks (e.g., mowing the grass, cooking a meal). Now, autonomous products that remove the consumer from task accomplishment altogether are increasingly available. Consumer researchers need to extend existing theories to understand how consumers perceive, feel about, and interact with these autonomous devices ([ 6]). How do consumer experiences emerge from repeated interactions with AI-powered smart devices, as initiated in the work by [ 9] that situates consumer behavior in a broader, non-human-centric context? This "object-oriented" approach is not only novel but also important because consumers can increasingly actively interact with new technologies that now have their own capacity for autonomous action. To fully understand consumer experience with new technology, researchers must consider that consumers' perceptions of new technology go beyond internal subjective responses and are influenced by (and can influence) the agency, autonomy, and authority of technology itself.
While the prior two subsections leverage existing trends to identify important research questions, we also wanted to think further into the future to outline broader research directions for the field. To accomplish this goal, we first compared the flow of papers for both the TPM Conference at Columbia University that launched the special issue and the special issue itself with what we described in our original call for papers. Doing so allowed us to identify gaps between the topics we envisioned originally and the actual coverage these topics have received in this special issue. Second, we ideated along the dimensions of consumer behavior, market research, and marketing decision making to identify high-potential future research areas. We discuss each in turn.
The role of new technology in the marketing organization. How do new marketing technologies change marketing's role within the firm? What is the effect of the adoption of new technology by marketing on firm performance more generally? How does new technology change the way marketing collaborates with other functions, including the interfaces with operations (e.g., collaboration between the two functions on push notifications or product returns), research and development (e.g., new product/service development, product upgrades, quality control), and information technology (e.g., marketing technology budgets and decision making)?
The social and policy effects of new technology in marketing. How should firms react to policy initiatives that intend to protect consumer privacy and limit data access? How should firms that develop, commercialize, or use new marketing technologies act ethically in society and engage in the public debate about the consequences of their actions on firms, consumers, and society? How can we identify the public risks that new marketing technologies carry, and how should society manage those risks? What are the relevant consequences of new policy initiatives, such as those on climate change, for new technologies in marketing?
The "dark side" of new technology. Eliminating face-to-face interactions can make transactions more efficient from the perspective of the firm. What are the implications for firms' relationships with customers? Can this shift contribute to feelings of isolation and loneliness? The dramatic loss of privacy as the result of new technology usage has become a major concern for consumers. What are the behavioral consequences of the loss of privacy as a result of consumers' expanding digital footprint? What methods can be used to guard consumers' privacy and data security while allowing for real-time personalization?
As new technologies are integrated into marketing practice, questions emerge about the potential for bias in firms' decision making. Research involving AI should explicitly address ethical aspects of the AI technology on the constituent populations. The potential for algorithmic bias across all digital applications, particularly social media, requires a clear understanding of the ways in which these systems may operate suboptimally and negatively impact consumer welfare. In general, as we develop new marketing methodologies and populate them with richer and bigger data flows, we need to be aware that these "improvements" in marketing decision making may inadvertently harm stakeholders (see [13]).
In the face of the changing role of the marketer, the future research areas identified previously present opportunities for marketing scholars to remain relevant and influential. The pace at which new technologies are developed and implemented stands in sharp contrast to the speed with which scholars can access data derived from such technologies, develop rigorous frameworks to analyze the phenomena, and move the research through the review process (the present special issue included). Three guidelines for future scholarship may help address this dilemma for research on new technologies in marketing and, perhaps, for marketing scholarship more generally.
First, we can (more) effectively leverage our connection to practice (see [ 7]; [16]). Much of the new technology deployment in marketing happens at increasing speed inside companies. Within these companies, the insights function is learning quickly what does and does not work. Marketing scholars who develop close connections with practice can study the impact of what firms and organizations are doing in real time to maintain their role as knowledge developers, rather than becoming only knowledge distributors. As scholars, we may need to step up our efforts to work with firms to remain at the forefront of knowledge development.
Second, we can be more oriented toward the future by removing some of the limits created by empirical requirements. Specifically, we need to embrace conceptual research more enthusiastically (consistent with [20] call) to stay ahead of real-world implementation. We need to embrace mixed-methods approaches that can foster important insights rather than dogmatically viewing some approaches as easier to work with, more scientific, or more fashionable than others.
Finally, we need to encourage and reward scholars who address important forward-looking, even difficult, research questions that have the potential to influence practice. When we as marketing scholars rigorously tackle big questions with an open, future-oriented mind, in closer collaboration with business practice, we can exploit the vast research opportunities that the new technology revolution in marketing presents. Scholars who do so successfully will take center stage in our field and will inspire marketers to strengthen their firms' competitive advantage and continue to improve and enrich people's lives with new technology.
The articles in this special issue reveal the highly specialized knowledge that marketers need to possess to operate effectively in new technology environments. Given the overwhelming impact of new technology on marketing practice, firms are in increasing need of marketers who understand the full scope of the new technologies used by consumers, how new technologies can translate consumer data into insights, and the new technologies that the firm can employ to achieve more favorable marketing and consumer outcomes.
Those marketers should become cornerstones of the firm's digital transformation journey. Some have called these specialists "marketing technologists" (e.g., [ 2]). They understand that new technologies may gain traction at any time and disrupt the way in which companies most effectively serve their customers. They should anchor on the firm's purpose and goals and pursue the deployment of technologies in an agile manner. Moving forward, nondigital technologies (e.g., genetic data, decision science) will become part of marketers' portfolio as well. While the marketing technologist might become a specialized function, marketers in the future should incorporate some aspects of and be able to communicate with the (marketing) technologist.
This special issue on "New Technologies in Marketing" presents a broad spectrum of research that investigates how new technologies drive marketing practice and can stimulate further research. By elucidating how new technology enables new forms of interaction among consumers and firms, this research shows that new technology is spawning new types of data and analytic methods, creates marketing innovations, and gives rise to new strategic marketing frameworks. Collectively, the articles in this issue demonstrate the virtuous cycle whereby firms deploy new marketing technologies, which enhance the richness and volume of market data, which spawn new analytic methods, which enable novel insights, which support more effective marketing decisions, which improve the collection of additional market data, and so on. Against this backdrop, and inspired by the research presented in this special issue, we provide recommendations for future research and offer thoughts on how the marketing scholar and marketing practitioner can stay relevant in the context of rapid developments of new technology. We hope that the articles in this special issue will inspire marketing scholars to take on those future research challenges.
Footnotes 1 We adopt the American Marketing Association's definition of marketing (https://www.ama.org/the-definition-of-marketing-what-is-marketing/).
References Bharadwaj Neeraj , Ballings Michel , Naik Prasad A. , Moore Miller , Arat Mustafa. (2022), " A New Livestream Retail Analytics Framework to Assess the Sales Impact of Emotional Displays ," Journal of Marketing , 86 (1), 27–47.
2 Brinker Scott , Heller Jason. (2015), " Marketing Technology: What It Is and How It Should Work ," McKinsey & Co. (March 1) , https://www.mckinsey.com/business-functions/marketing-and-sales/our-insights/marketing-technology-what-it-is-and-how-it-should-work.
3 Chandrasekaran Deepa , Tellis Gerard J. , James Gareth M.. (2022), " Leapfrogging, Cannibalization, and Survival During Disruptive Technological Change: The Critical Role of Rate of Disengagement ," Journal of Marketing , 86 (1), 149–166.
4 Crolic Cammy , Thomaz Felipe , Hadi Rhonda , Stephen Andrew T.. (2022), " Blame the Bot: Anthropomorphism and Anger in Customer–Chatbot Interactions ," Journal of Marketing , 86 (1), 132–148.
5 Daviet Remi , Nave Gideon , Wind Jerry. (2022), " Genetic Data: Potential Uses and Misuses in Marketing ," Journal of Marketing , 86 (1), 7–26.
6 De Bellis Emanuel , Johar Gita Venkataramani , Poletti Nicola. (2021), " Meaning of Manual Labor Drives Consumer Adoption of Autonomous Products ," working paper.
7 Deighton John A. , Mela Carl F. , Moorman Christine. (2021), " Marketing Thinking and Doing ," Journal of Marketing , 85 (1), 1 – 6.
8 Glazer Rashi. (1991), " Marketing in an Information-Intensive Environment: Strategic Implications of Knowledge as an Asset ," Journal of Marketing , 55 (5), 1 – 19.
9 Hoffman Donna L. , Novak Thomas P.. (2018), " Consumer and Object Experience in the Internet of Things: An Assemblage Theory Approach ," Journal of Consumer Research , 44 (6), 1178 – 204.
IBM Institute for Business Value (2021), " 2021 CEO Study: Find Your Essential ," report (accessed November 4, 2021) , https://www.ibm.com/thought-leadership/institute-business-value/c-suite-study/ceo.
John George , Weiss Allen M. , Dutta Shantanu. (1999), " Marketing in Technology-Intensive Markets: Toward a Conceptual Framework ," Journal of Marketing , 63 (Special Issue), 78 – 91.
Longoni Chiara , Cian Luca. (2022), " Artificial Intelligence in Utilitarian Versus Hedonic Contexts: The 'Word-of-Machine' Effect ," Journal of Marketing , 86 (1), 91–108.
Mehrabi Ninareh , Morstatter Fred , Nripsuta Saxena , Lerman Kristina , Galstyan Aram. (2022), " A Survey on Bias and Fairness in Machine Learning ," ACM Computing Surveys , 54 (6), 115.
Miao Fred , Kozlenkova Irina V. , Wang Haizhong , Xie Tao , Palmatier Robert W.. (2022), " An Emerging Theory of Avatar Marketing ," Journal of Marketing , 86 (1), 67–90.
Raisch Sebastian , Krakowski Sebastian. (2021), " Artificial Intelligence and Management: The Automation-Augmentation Paradox ," Academy of Management Review , 46 (1), 192 – 210.
Stremersch Stefan , Winer Russell S. , Camacho Nuno. (2021), " Faculty Research Incentives and Business School Health: A New Perspective from and for Marketing ," Journal of Marketing , 85 (5), 1 – 21.
Tan Yong-Chin , Chandukala Sandeep R. , Reddy Srinivas K.. (2022), " Augmented Reality in Retail and Its Impact on Sales ," Journal of Marketing , 86 (1), 48–66.
Wedel Michel , Kannan P.K.. (2016), " Marketing Analytics for Data-Rich Environments ," Journal of Marketing , 80 (6), 97 – 121
Wichmann Julian R.K. , Wiegand Nico , Reinartz Werner J.. (2022), " The Platformization of Brands ," Journal of Marketing , 86 (1), 109–131.
Yadav Manjit S. (2010), " The Decline of Conceptual Articles and Implications for Knowledge Development ," Journal of Marketing , 74 (1), 1 – 19.nn
~~~~~~~~
By Donna L. Hoffman; C. Page Moreau; Stefan Stremersch and Michel Wedel
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 127- The Roar of the Crowd: How Interaction Ritual Chains Create Social Atmospheres. By: Hill, Tim; Canniford, Robin; Eckhardt, Giana M. Journal of Marketing. May2022, Vol. 86 Issue 3, p121-139. 19p. 3 Color Photographs, 1 Diagram, 3 Charts. DOI: 10.1177/00222429211023355.
- Database:
- Business Source Complete
The Roar of the Crowd: How Interaction Ritual Chains Create Social Atmospheres
Atmospheres are experiences of place involving transformations of consumers' behaviors and emotions. Existing marketing research reveals how atmospheric stimuli, service performances, and ritual place-making enhance place experiences and create value for firms. Yet it remains unclear how shared experiences of atmosphere emerge and intensify among groups of people during collective live events. Accordingly, this article uses sociological interaction ritual theory to conceptualize "social atmospheres": rapidly changing qualities of place created when a shared focus aligns consumers' emotions and behavior, resulting in lively expressions of collective effervescence. With data from an ethnography of an English Premier League football stadium, the authors identify a four-stage process of creating atmospheres in interaction ritual chains. This framework goes beyond conventional retail and servicescape design by demonstrating that social atmospheres are mobile and cocreated between firms and consumers before, during, and after a main event. The study also reveals how interaction rituals can be disrupted and offers insight as to how firms can balance key tensions in creating social atmospheres as a means to enhance customer experiences, customer loyalty, and communal place attachments.
Keywords: atmospheres; emotions; events; place; rituals; sport; emotional energy; entrainment
Atmospheres are experiences of place involving transformations of consumers' behaviors and emotions. When marketers get atmospheres right, firms benefit from enhanced customer experiences, loyalty, and place attachment ([ 9]). Given the value of atmospheres, it is fitting that research explains how firms can create atmospheric effects using collages of material objects ([ 4]; [44]; [73]). These stimulate consumers' senses, emotions, and behaviors in retail sites ([ 7]; [49]; [52]; [72]). Likewise, in themed stores, material stimuli create spectacular places that prime play and pleasure ([ 8]; [69]).
Marketing research also explains that experiences of place are socially constructed during interactions between staff and customers ([ 3]). Staff can enhance customer experiences of place by educating consumers about meanings of locations and by performing cultural scripts to elicit emotional responses ([ 3]; [22]; [62]; [69]). When service performances are enacted alongside atmospheric stimuli provided by material design, firms can transform places into "a separate and self-contained world" ([47], p. 662). For example, [23] show how opulent design and staff performances create the intimidating atmosphere of luxury stores.
Prior research has explained how firms can generate atmospheric effects through material and social constructions of place. Yet in contexts as diverse as nightclubs, festivals, and themed stores, consumers share atmospheric experiences in groups ([ 3]; [32]; [45]; [69]). Despite acknowledging the value of "boundary open" feelings of togetherness among consumers ([ 3]), marketing research has yet to explain specifically how group experiences of atmospheres are created. Thus, our primary research question asks, what are the social processes that create atmospheres?
Clues as to how these atmospheric interactions are created can be gathered from studies that show consumers cocreating place experiences ([ 3]; [65]). At historical reenactments, festivals such as Burning Man, and brandfests, cohorts of consumers act out consumption rituals ([ 5]; [45]; [53]). Over time, repetitions of these ritual processes can imprint spaces with a convivial "spirit of place," an identity that guides consumers' interpretations, emotions, and behaviors ([10]; [37]). Once instilled, these socially constructed qualities can endure, creating lasting place attachments ([ 9]; [67]).
In contrast to these enduring qualities constructed over time, however, social experiences of atmospheres can emerge and dissipate more rapidly ([14]; [38]). In places such as stadiums, nightclubs, religious sites, and street parades, for example, atmospheres often intensify and climax in pleasurable outbursts of shared emotion and behavior ([32]; [39]; [45]). As a valued aspect of collective live events, it is important to understand more about how ritual interactions create these shared intensifications of emotion and behavior that consumers value during atmospheric experiences of place.
This is especially the case given that consumer groups are often heterogeneous in nature ([53]; [77]), and misaligned values and behaviors can inhibit rituals, potentially spoiling atmospheres ([37]; [68]). Indeed, prior studies have recognized the necessity for managers to balance divergent expectations and behaviors in places cocreated by multiple actors including firms and consumers ([10]; [47]). To enable firms to benefit from creating social atmospheres and avoid failures of atmosphere, therefore, we ask two further research questions: How do atmospheres fail, and how can the social dimensions of atmosphere be managed?
To investigate how social atmospheres are created, how they fail, and how they can be managed, we next explain our theoretical lens and research procedures. We then present our findings, which reveal a four-stage process of creating "social atmospheres" in interaction ritual chains, and how social atmospheres can fail. Our discussion explains how these findings extend theories of material atmospherics, servicescapes, and ritual place-making by showing that social atmospheres are cocreated by firms and consumers in multiple sites before, during, and after a main event. We extend interaction ritual theory by explaining how atmospheres can be mobile and how learning enables social atmospheres. Finally, we offer advice for managers on how to create social atmospheres that enhance customer experiences, customer loyalty, and place attachment.
To understand the social processes by which atmospheres are created and how atmospheres can be managed, [15] theory of interaction ritual chains enables us to explain how shared emotions and behavior are created during collective live events. Interaction ritual chains are a "mechanism of change" ([15], p. 43) that aligns attention, behaviors, and emotions among group members. Interaction rituals begin when a symbolic object or action gathers a shared focus of attention, transforming individuals into ritual participants and transforming the atmosphere as a result. For example, when a conductor taps their baton on the music stand, the sounds of conversations and musicians warming up are replaced with an anticipatory silence.
If a group shares a common focus, then behaviors and emotions tend to align among ritual participants due to physiological predispositions to imitate others ([38]). To describe these alignments of behaviors and emotions, [15] provides the concept of "entrainment": the stimulation of common emotional responses is labeled "emotional entrainment"; the stimulation of common movements and vocalizations are labeled "behavioral entrainment." Consider the coordinated actions among a crowd at a tennis match as they follow the ball back and forth, for example, and how shared excitement intensifies the atmosphere.
[15] also enables us to explain how atmospheres climax. When crowds of entrained people generate expressions of their shared emotions, such as common vocalizations, gestures, and movements, Collins explains that ritual participants become aware that they are "doing the same thing" and "thinking the same thing" (2004, p. 33). During these instances, interaction rituals reach fever pitch. Here, Collins follows sociologist Émile [25], p. 424), who coined the term "collective effervescence"' to describe moments during rituals when participants "become hyper excited, the passions more intense, the sensations more powerful."
Social atmospheres encompass shared focus, entrainment, and collective effervescence, but it is particularly the latter that produces "emotional energy," a pleasurable feeling of group membership that motivates participants to repeat rituals ([15]). Importantly, the symbols that groups use during expressions of collective effervescence become representative of these pleasures. Collins suggests symbols used in interaction rituals can act like "batteries," storing emotional energy. As such, these charged symbols remind people of prior experiences and can be used as focal points to start future interaction rituals, linking rituals together in "chains."
Notwithstanding the enabling aspects of Collins's theory, it is not clear how group dynamics impact focus, entrainment, and collective effervescence ([28]). Although entrainment is stimulated by physiological mirroring, for instance, further investigation of how groups moderate alignments of behavior and emotion is warranted in light of our interest in community heterogeneity. This is particularly so in instances where consumers seek different emotional experiences at the same event ([19]; [37]).
Furthermore, [15] analyzes discrete interaction rituals that are enacted in particular places. Yet consumer research shows that place experiences can be anticipated by preparatory activities ([10]; [32]), and that consumers move from place to place during consumption experiences ([41]; [75]). Illustrating both possibilities, [11], p. 213) explain how preparations enacted by groups outside a stadium leave people "primed" and "psyched" for the game to follow. Despite these observations, it is not clear how group activities in one place might impact atmospheres in another. It is with these foregoing questions in mind that we describe our context for this investigation.
To understand how social interactions influence atmospheres and how these can be managed, we focused on the context of English Premier League (EPL) football stadiums. EPL stadiums are an ideal context to investigate atmospheres because these sites are associated with active and passionate supporters who cocreate the visual, sonic, and emotional qualities of these places. The EPL acknowledges that consumers "create an atmosphere that sets us apart from other leagues and competitions" ([63]). Moreover, industry leaders agree that these atmospheres generate commercial value for the EPL ([26]; [56]).
In keeping with interpretive marketing research studies of place ([22]; [47]; [69]), we focused on an exemplar case: Anfield, home of Liverpool Football Club (LFC), one of the wealthiest football clubs globally, and part of the Boston-based Fenway Sports Group. Anfield is considered one of the EPL's "most atmospheric stadiums" ([36]). Inside Anfield is the "The Kop" stand; this is where the ritual singing common in English football originated in the early 1960s, when fans began to sing the popular hit, "You'll Never Walk Alone" (from Rodgers and Hammerstein's Carousel). Ever since, the Kop has remained a bastion of impassioned vocal support ([42]), and this song has become the official club anthem.
Notwithstanding Anfield's reputation as an atmospheric place, EPL atmospheres are not always lively. Particularly in recent years, some consider that atmospheres have become flat and lifeless ([50]). Moreover, EPL stadiums represent contexts that require a variety of interventions to manage atmospheres. In the 1980s, two stadium disasters involving Liverpool supporters highlighted the risks of highly emotional crowd contexts: Hillsborough, where overcrowding led to 96 deaths, and Heysel, where crowd disorder led to 39 fatalities. These events motivated safety alterations at all EPL stadiums, ensuring that supporters remain seated and subject to security supervision ([43]). These aspects of our context provide opportunities to understand more about how atmospheres fail and how firms attempt to manage atmospheres.
We investigated how Anfield's atmosphere is created during fieldwork spanning seven years. Our battery of ethnographic techniques encompassed participant observation, interviews, as well as archival and online data collection ([ 3]; [46]; [69]). The data we collected capture supporters' experiences and interactions within Anfield, as well as in the spaces surrounding Anfield before and after matches ([76]). Our data set is summarized in Web Appendix 1 and comprises field notes, 60 in-depth interviews, and 4 audiencing interviews. Together these activities generated 410 pages of single-spaced text, 610 newspaper articles, 53 photographs, and 17.5 hours of video recordings.
To build a "thick description" of atmospheric experiences at Anfield, participant observation ([30]; [41]) occurred before and after matches during two consecutive EPL seasons, from November 2012 to August 2013 and from November 2013 to August 2014. In November 2012, the first author rented an apartment close to Anfield, and access to participants was facilitated by a local supporter who acted as a gatekeeper. This access enabled observations in supporters' homes, during social gatherings, on public transport to games, in pubs, and on the streets outside Anfield. In 2013 and 2014, the first author continued these procedures and attended 24 LFC matches: 18 at Anfield, and 6 at other EPL stadiums. Access to Anfield enabled the description of microinteractions between individuals as well as crowd-level expressions ([31]). To record these firsthand experiences of atmospheres, digitized field notes were compiled from observations, photographs of supporters' visual displays, and video footage recorded by the first author.
Access gained during the 2012–2013 season allowed for 60 in-depth interviews between November 2012 and March 2021. Interviews sought insight into supporters' life histories, the meanings they associate with Anfield, and their experiences of atmospheres. EPL stadiums provide an ideal instance of collective events characterized by consumer heterogeneity. To reflect this, we purposively sampled devoted supporters with local ties to LFC as well as casual supporters and tourists. Table 1 shows our sample, which represents a typical Anfield crowd by age, gender, and commitment level. To enhance insight into the creation and management of atmosphere we also sampled police, stadium architects, crowd safety experts, journalists, and sports-marketing consultants. Interviews began with grand-tour questions ([54]) about experiences of atmospheres at Anfield, as well as supporters' pre- and postmatch activities. Interviews lasted between 45 minutes and 3 hours; all were recorded and transcribed.
Graph
Table 1. Participants.
| Participant Type | Pseudonym | Approximate Age (Years) | Commitment | Interview Type |
|---|
| Supporters | Amanda (F) | 40s | Casual | Depth |
| June (F) | 50s | Devoted | Depth |
| Mark (M) | 50s | Devoted | Depth |
| Martin (M) | 30s | Casual | Depth |
| Tim (M) | 20s | Casual | Depth |
| Tommy (M) | 50s | Devoted | Depth |
| Adam (M) | 20s | Devoted | Depth |
| David (M) | 50s | Casual | Depth |
| Brian (M) | 60s | Devoted | Depth |
| Anne (F) | 60s | Devoted | Depth |
| Graham (M) | 50s | Devoted | Depth |
| Chris (M) | 20s | Devoted | Depth |
| Adam (M) | 20s | Devoted | Depth |
| Seb (M) | 30s | Casual | Depth |
| Bill (M) | 30s | Tourist | Depth |
| Paul (M) | 30s | Casual | Depth |
| Gary (M) | 60s | Devoted | Depth |
| Thomas (M) | 40s | Casual | Depth |
| Anthony (M) | 30s | Devoted | Depth |
| Michaela (F) | 20s | Devoted | Depth |
| James (M) | 30s | Devoted | Depth |
| Barry (M) | Teens | Devoted | Depth |
| Keith (M) | 40s | Devoted | Depth |
| Anne (F) | 60s | Devoted | Depth |
| Joel (M) | 20s | Devoted | Depth |
| Jon (M) | 50s | Casual | Depth |
| Peter (M) | 30s | Casual | Depth |
| Tony (M) | 70s | Devoted | Depth |
| Mick (M) | 40s | Devoted | Depth |
| Peter (M) | 50s | Casual | Depth |
| Steve (M) | 30s | Casual | Depth |
| Clive (M) | 50s | Casual | Depth |
| Sam (M) | 40s | Devoted | Depth |
| James (M) | 20s | Devoted | Depth |
| Lee (M) | Teens | Devoted | Depth |
| Craig (M) | 30s | Devoted | Depth |
| Duncan (M) | 40s | Casual | Depth |
| Brian (M) | 50s | Devoted | Depth |
| Steve (M) | 20s | Devoted | Depth |
| Andrew (M) | 20s | Devoted | Depth |
| Andy (M) | 30s | Devoted | Depth |
| David (M) | 40s | Casual | Depth |
| Joe (M) | 20s | Devoted | Depth |
| Mike (M) | 30s | Casual | Depth |
| Jenny (F) | 20s | Tourist | Depth |
| Sebastian (M) | 30s | Tourist | Depth |
| Ben (M) | 20s | Tourist | Depth |
| Jay (M) | 30s | Devoted | Audiencing |
| Gareth (M) | 30s | Devoted | Audiencing |
| Nev (M) | 50s | Devoted | Audiencing |
| Dan (M) | 30s | Devoted | Audiencing |
| Police | Jim (M) | 40s | — | Depth |
| Reg (M) | 40s | — | Depth |
| Stadium architects | Brian (M) | 50s | — | Depth |
| John (M) | 50s | — | Depth |
| Daniel (M) | 50s | — | Depth |
| Joe (M) | 30s | — | Depth |
| Crowd safety supervisors | Steve (M) | 60s | — | Depth |
| Keith (M) | 50s | — | Depth |
| Sports marketing consultants | Tim (M) | 50s | — | Depth |
| George (M) | 40s | — | Depth |
| Journalists | Stuart (M) | 40s | — | Depth |
| Rory (M) | 30s | — | Depth |
1 Notes: M = male; F = female.
Embodied practices and multisensory experiences of place can be difficult for respondents to recall or describe ([38]). "Audiencing" is a video elicitation technique that uses recorded footage of participants interacting with the environment to evoke discussion of significant and specific sensations, emotions, and actions ([55]). Accordingly, we used this technique to explore how atmosphere is created in Anfield as well as supporters' multisensory experiences of atmosphere and to interpret the role of material aspects of place during these experiences. Four such interviews took place. Participants were shown video footage recorded by the first author of a recent match at which they were present. This procedure lasted between one and two hours. These interviews were recorded and transcribed.
Data collection included newspaper coverage as a means to understand wider media representations of atmosphere, supporter behavior, and commercial/social/political issues related to EPL football stadiums. Using the LexisNexis archive ([24]), we collated documents referring to "atmosphere" in EPL contexts between September 2012 and August 2019 in national newspapers as well as a popular local newspaper, The Liverpool Echo. Finally, we monitored official LFC web platforms as well as social media where LFC supporters post content related to the club ([75]).
Open coding began with authors independently identifying common patterns and sequential moments among interview cases, field notes, and photographs ([74]). These procedures guided our analysis toward a focus on understanding the social processes that create effervescent atmospheres as well as what causes atmospheres to fail. As research progressed, our analysis identified place-based and processual occurrences; key actors, places, and material objects; effects on consumers' emotions and behaviors; and aggregate social effects within crowds. Our data set enabled us to identify interconnected ritual stages before, during, and after supporters' visits to Anfield. Moreover, given that consumers return to Anfield and surrounding places week after week, these repetitions enabled us to identify a cyclical process of creating social atmospheres. Interpretations were triangulated across methods and authors, and member-checks were used to solve coding disputes ([78]). The final stages of analysis involved tacking between emergent themes and existing knowledge of atmospheres and experiential place-making to refine and extend theory ([12]). Web Appendix 3 presents the final coding.
Our findings are organized to explain the process of creating social atmospheres through an interaction ritual chain that comprises four interlinked stages. Figure 1 illustrates how this process begins with preparations in private places, moves to activation in a service ecosystem, climaxes in an event space, and concludes with recovery in the service ecosystem after an event. Linking these stages are symbolic resources that enable consumers to transfer social atmospheres from one place to another, gathering focus, entrainment, and collective effervescence among increasingly large crowds on the way. Having unpacked this process, we explain how atmospheres can fail when shared focus is distracted or entrained groups are disaggregated.
Graph: Figure 1. Creating social atmosphere in the interaction ritual chain.
Atmospheric preparations are consumer-led activities prior to events in which symbolic resources and behavioral expectations are created to enable atmospheres. These preparations begin in relatively private places such as homes and involve families and friendship groups as well as entrepreneurial consumers. We reveal three modes of preparation: learning about cultural expectations to participate, making symbolic resources, and rehearsing entrained behaviors.
Learning to participate refers to experienced consumers teaching newer consumers to contribute to atmospheres. A sense of seriousness helps motivate participation in interaction rituals ([31]); Dan's first visit to Anfield as a boy was prefaced by "a serious chat with my Dad, where he explained how special Anfield is." Team affiliations among sports consumers are often inherited through families, where emotional and behavioral expectations are modeled. For example, Anthony describes how he taught his son about what going to Anfield means and the practices that supporters share there:
I think the most important thing is teaching the younger ones what it means to be a Red. You have to let them know that football is a shared experience, and you go to the match for the craic [fun and enjoyment], the camaraderie, and emotion. When I took our youngest, both me and my Dad told him about the Kop and what it's like to be in the Kop, what it means.... Sam had grown up singing the songs Michaela and I would sing around the house when he was really little, so when we thought he was old enough to come with us, he could join in.
Preparations teach consumers about the spirit of place ([69]), and the spirit of place at Anfield identifies expectation to participate in fun, friendship, and shared emotions, all of which encompass a moral responsibility to enact the rituals and traditions that identify and unite the LFC community ([59]; [65]).
Making symbolic resources involves the preparation of visual and aural symbols that represent consumer identity during interaction rituals. Dedicated supporter Chris and his friends create flags with motifs and slogans that represent the club: "We meet up on Wednesday nights and plan for the next match, or start thinking about bigger, more complicated designs for a particular occasion." Chris explains that these objects are meaningful for supporters: "Some of them are dead historical and were made for a particular moment in time, something like Hillsborough." Figure 2 shows these flags with visual motifs recalling Hillsborough as well as major cup victories: so valued are these resources that when they finally wear out after years of use and repairs, they are carefully replaced with a new version.
Graph: Figure 2. Commemorative flags symbolizing Hillsborough and league victories.
In addition to these visual symbols, supporters prepare aural symbols in the form of songs and chants. Dan explains that "music and football culture are closely entwined," and in 2012 he organized "Boss," an open mic night "for everyone to come together, to sing ... and build camaraderie." These events are opportunities for supporters to generate new songs. Dan explains that "songs that are created in the moment in events such as these spill into the ground, one moment you're playing around with your mates and then next weekend there's thousands of people singing it.... You feel dead proud, knowing that that's something you've created." LFC advertises these events, showing that firms can coproduce symbolic resources by partnering with "passionate entrepreneurs" like Dan ([34]).
Rehearsing entrained behaviors occurs before an event and involves individual consumers getting ready to participate in social atmospheres. Much like a sports team warms up before a game, supporters rehearse before matches. James explains that "you've got to get in the mood beforehand." Nev describes how he prepares himself for games by "whistling and humming songs as I'm getting ready. My son likes to put on some YouTube videos, or some of those emotional montages that the television companies like to produce." Online resources enable consumers to rehearse aural symbols at home, long before events begin. Both official LFC web media and supporter-run websites provide videos of excited crowds that inspire rehearsal. Given that not all supporters have family ties through which behavioral expectations are prepared, Andy, another passionate entrepreneur, explains that he founded the website "The Anfield Wrap" to "help fans across the world to learn how not only to experience and take in the atmosphere, but to contribute." To facilitate this outcome, Andy uses footage from inside the stadium that "captures the spectacle we supporters put on, the sights and sounds that we produce." Through these preparation activities, individual supporters familiarize themselves with songs and behaviors, such that the next stage of the ritual chain is made ready.
We define atmospheric activation as the creation of social atmospheres among smaller groups prior to a main event. This section first explains how groups of consumers activate shared focus and entrainment in a service ecosystem outside a main event space. Then, we explain that the social atmosphere intensifies when entrained groups gather into a larger crowd as they move toward an event. Finally, we highlight how entry into an event space can activate memories of prior atmospheric experiences through firm-generated objects.
Activating focus and entrainment transforms individual consumers into small groups that begin to create a social atmosphere. This occurs in the many hotels and pubs around Anfield stadium, a service ecosystem where supporters gather several hours before games. LFC devotee Gary's favorite pub is The Sandon, known as the birthplace of LFC in 1892. Gary returns here week after week to enjoy the same experience: "It's always a good time in the Sandon.... You know the same faces will be there." Studies of servicescapes note how consumers enjoy a collage of visual and sonic stimuli, which encourage sociality in small groups, much like the separate areas of ESPN Zone, a now-defunct sports-themed restaurant ([69]). In contrast, The Sandon is a bland, austere space. Yet as supporters arrive, the Sandon's managers allow them to pin their homemade flags to the walls. This act of "place-marking" transforms space into place and initiates dialogue among consumers ([79]).
The staff at the Sandon also help supporters move tables and chairs to clear a stage for the next moment in the interaction ritual. [15], p. 119) explains that certain individuals embody "symbolic resources" that enable them to entrain the behaviors of those around them. Gary fulfils this role: as the Sandon reaches capacity, he moves to the center of the room. Smaller groups of families and friends halt their conversations and share a focus. Field notes from January 18th, 2014, recall how Gary sings the first line of a song to the tune of Boney M.'s "Brown Girl in the Ring": "Poetry in motion," to which everyone responds, "Tra-la-la-la-la!"
Poetry in motion/Tra-la-la-la-la!
[repeat three times]
We're the best football team in the land/Yes we are!
We are Liverpool/Tra-la-la-la-la!
[repeat three times]
We're the best football team in the land!/Yes we are!
The call-and-response style enables supporters to begin entraining their behavior, breathing, and gestures, and physiological research suggests that even heartbeats can align among people who sing together ([58]). By sharing a focus and initiating behavioral entrainment, supporters start to build a social atmosphere hours before the game. Like other pregame rituals ([10]), these activities leave supporters like Gary pumped up with "emotional energy" ([15]): "It's the best pub to get you up for the game, it's noisy and no other pub can prepare you better to stand and sing your heart out on the Kop."
At some stage, consumers leave the service ecosystem and enter public spaces. Here, social atmospheres intensify as crowds gather and move toward an event space together. An hour before the game begins, supporters leave the pubs and join what some call "the pilgrimage," a procession through the streets toward Anfield. Stuart describes this as a "sea of people who descend at the same time on the most cherished symbol of the club." Sports and music venues are often likened to religious sites ([10]; [69]). Mark explains "there is nothing like being part of a crowd heading to the ground. Even though you don't know those around you, you feel like you're part of a gang, a religious group." Much like the shells carried by pilgrims on the Camino de Santiago ([41]), supporters' flags act as a mobile focus, helping gather people together along the way. Outbursts of laughter and singing light up the street like brushfires as groups of LFC supporters are united in the streets that lead toward the stadium.
Social atmospheres intensify when smaller groups are united into larger crowds. Like other appropriations of public space ([10]), thousands of supporters gather outside Anfield stadium to take part in what they call the "Welcoming Committee" that greets the LFC team bus as it arrives. Chris explains that this greeting was organized by "Spion Kop 1906," a group of devoted supporters that wanted "to do something outside the stadium to get people up for the match." Because of the presence of the team on board, and its proximity to the crowd, the bus is treated as a sacred object ([ 6]). As such, the bus provides another shared focus that gathers groups into an increasingly dense, entrained crowd. As the bus moves through this crowd, the streets are filled with pyrotechnic flares, smoke bombs, and flags.
The crowd grows.... Police remind people to not block the road. People climb on walls, encroach in people's gardens, mount scaffolding, and get on each other's shoulders. They sing, chant, fly banners and set off red smoke bombs. When the Liverpool bus arrives, it's impossible to see where the crowd starts or ends. As the bus creeps up Anfield Road at a snail's pace, more smoke bombs are set off, beer is thrown, and the crowd surrounds the vehicle, hitting it in support. (Field notes, February 10, 2014)
Figure 3 shows the chaotic scene of revelry that supporters create. The visual display of red and white flags and smoke is accompanied by the smell of beer and diesel fumes, the feeling of bodies pressed together, and the sounds of chanting, singing, and fists striking the bus. Members of this crowd become entrained to the excitement: the release of one flare or smoke bomb hastens the next; as one person chants, hundreds of others join in, producing outbursts of collective effervescence ([25]). This social atmosphere transforms the streets outside the stadium. As one tourist recalled, "The atmosphere before the game was crazy. I loved it."
Graph: Figure 3. Mobile symbols at the "Welcoming Committee."
Entering event spaces can activate strong feelings among consumers due to objects that stimulate memories and emotions ([37]; [69]). At Anfield, thousands of supporters file through the wrought iron "Paisley Gates," the iconic threshold of the stadium; they encounter statues of legendary players and the memorial shrine that commemorates the supporters killed at Hillsborough. Here many supporters stop to light a candle or stand in silent prayer. All of these objects symbolize memories of prior social atmospheres, and for Michaela these memories make Anfield special:
It could have been any one of us, it could have been our sons and daughters [and] the reason why so many people were adamant that we shouldn't build a new stadium is because of what happened at Anfield after Hillsborough: that mass outpouring of grief and the mourning that occurred here.
As much as place attachments can be associated with excitement, as with the Welcoming Committee, so too can they be associated with collective experiences of suffering and pain ([37]). In either case, [25], p. 220) explains that experiences of collective effervescence can instill a place with "exceptionally intense forces." For supporters like Michaela, memories of prior social atmospheres at Anfield enhance the attachment she feels toward this place ([ 9]; [20]). In other words, the material objects within event spaces can activate memories of prior social atmospheres, and this motivates consumers' loyalty to return to the sites where these experiences occurred.
Ritual climaxes mark the culmination of collective live events when lively social atmospheres are shared among large groups. First, we show how firms deliberately orchestrate a "formal climax" by triggering focus, entrainment, and collective effervescence to unite crowds. Following a formal climax, we reveal that "natural climaxes" can follow, where social atmospheres are created by consumers responding spontaneously to events, without leadership or control by firms.
A formal climax is a deliberately orchestrated moment of focus, entrainment, and collective effervescence that unites large groups. While other supporters are enjoying the Welcoming Committee or paying their respects at the Hillsborough Memorial, groups of devoted supporters expertly knot their homemade flags to poles and stanchions. As with similar activities in The Sandon, this place-marking ([79]) transforms the blank canvas of Anfield's interior into a place that boosts Steve's feelings of identity as an LFC supporter: "When you walk in and see them flying, you feel together.... You're there with your own people." At this stage, 50,000 people chat among themselves, waiting for the club anthem, "You'll Never Walk Alone." Having marked the beginning of games for more than 50 years, this song is Anfield's most iconic "sound-mark" ([64]), and its singing is a "formal ritual," a "recognized apparatus of ceremonial procedures" ([16], p. 50). This ceremony gathers focus and initiates entrainment among supporters who stand and begin to sing in unison:
Many close their eyes. Some have their arms wrapped around one another. Others extend their arms, as if they are at church.... Meanwhile, a gargantuan banner is passed over our heads, from left to right across the Kop, covering thousands of supporters. (Field notes, December 7, 2013)
Even when the music stops, the crowd continues singing, entrained to one rhythm. At the same time as consumers are cocreating Anfield's soundscape, an enormous banner dubbed "the surfer" is passed above supporters' heads from one side of the Kop to the other. The movement seems to symbolize the transformation of individuals into a unified crowd (see Figure 4). In one audiencing interview, devoted supporter Nev reflected on this multisensory experience as one of visceral connection to other supporters:
Graph: Figure 4. The formal ritual of "The Surfer" passing over supporters.
It sends a tingle down your neck, and makes your hair stand on end.... It feels like you have some deep connection with those around you and they might be strangers you know. We're doing this together, like, I'm expressing myself to you, and you are open to me. Where else can you get that connection?
Nev witnesses how this formal ritual breaks down the self–other divide, allowing for empathic experiences of shared emotions even among strangers ([32]; [37]). [15] explains that collective effervescence is facilitated when groups recognize that they are sharing emotions and behaviors. Firms play an important role in enabling these moments by providing event spaces that facilitate sensory experiences of others. Joe, a stadium architect, explains that the physical qualities of Anfield enable atmospheres because supporters can see and hear one another, "It's tight, compact, you feel hemmed in together, and the sight lines allow everyone to see what is happening on the pitch and to respond together. It doesn't take too long for sound to travel." As well as being pleasurable, this "heightened intersubjectivity" ([15], p. 35) among supporters becomes vital as the game begins.
Natural climaxes are pleasurable outpourings of shared emotion in which crowds of entrained consumers generate social atmospheres spontaneously, without leadership or orchestration from firms. As much as sport is a form of play ([10]), English football has also been described as an "audience-oriented conflict," where opposing supporters mimic intergroup violence ([17], p. 202). As such, most supporters do not wish to passively observe this conflict. Rather, they become a chorus that comments on the drama taking place on the field. During an audiencing interview, Jay described how this connection between the game and supporters produces an "incredible atmosphere":
Some games you're on your feet for the whole 90 minutes, and you can come out of the stadium and your head is banging, your mouth dry. It's really draining afterwards, you know.... That Chelsea game, everyone was involved, an incredible atmosphere, whistling nonstop, roaring and celebrating every challenge instantly.
Dan explains how he feels compelled to participate: "If there's someone giving it their all, you can't just let them do it by themselves. You want to get stuck in." It is noteworthy that the chorus chimes in without leadership. Thousands whistle and boo when the opposition are in possession of the ball; they chant rhythmically to propel their team forward or to humiliate their opponents. [15] uses the term "natural rituals" to describe events where entrained participants improvise according to emerging conditions. Fittingly, Joel describes the experience of being entrained with others as a "natural" reaction: "You're on the same wave-length. It's natural, you know, like I know what the others around me are feeling and thinking."
Photojournalist Stuart notes how entrained crowds appear to share emotions: "Everyone has the same look on their faces, that sort of clear connection of everyone being on the same page, it's like magic." For [15], it is when crowds become aware of their common experiences that exaggerated changes in a social atmosphere are likely to occur, such as when collective tension and anxiety transform into collective effervescence following a goal:
A Liverpool midfielder misplaces a pass, giving the ball to the opposition. The ground is starting to feel tense. Supporters sense that their team is vulnerable, they are rubbing their faces and biting their nails as they watch players fumble easy passes.... The team appear equally nervous, as each of their mistakes are met with an exasperated moan from the crowd. [Then] Liverpool score and the crowd erupts. The noise deafens. Arms and fists flail around as supporters rush to celebrate with each other. Bodies compress: supporters fall on top of one another. I am hauled into the row in front. Supporters scramble down the stairs that lead to the pitch. It is temporary chaos. (Field notes, January 28, 2014)
Collective effervescence is a pleasurable experience for supporters like James, who considers that crowds where "everyone is on the same page are a different animal. At the end of the 90 minutes, your head is banging, and your throat is raw. You know you've been in that crowd situation, an electric atmosphere." James' account of atmosphere as "electric" recalls Durkheim's ([25], p. 217) description of collective effervescence in crowds, where "a sort of electricity is generated from their closeness and quickly launches them to an extraordinary height of exaltation."
In the stage of atmospheric recovery, shared emotions and memories of atmosphere are stored in resources that make them available for future use. We first show how objects used within the stadium are charged up with emotions and memories, and how these objects will be recirculated by consumers in future interaction rituals. We then show how firms circulate representations of social atmospheres in social media, which helps kindle desires to repeat interaction rituals.
[15] explains that objects associated with experiences of collective effervescence become charged up like batteries, storing the emotional energy of an event. The large flags produced by Chris and his friends embody these feelings, for example, and are lovingly kept at the club for this reason. However, if LFC wins, supporters emerge from Anfield twirling red and white scarves above their heads and singing joyously. For tourist Ben, club merchandise helps him participate in a social atmosphere: "Every club I've been to watch, I've bought something ... to demonstrate that you want to fit in, and to contribute, to participate." Beyond fitting in, however, the symbolic resources that fans use in the stadium get charged up and become objects that inspire them to return. For example, Peter explains that through repeated use, his red and white scarf has become a "lucky charm. I've had it for decades, it's dead old, it's tatty. I don't bring it with me to the match every game, only the special, important ones, where the team really needs to win. It's become a bit of a joke, when I turn up at Anfield and those around me are like 'Oh, we're going to win this afternoon, are we?'"
Remembering atmosphere involves the use of digital merchandise to recall experiences and store emotional energy in ways that inspire future interaction rituals. Once the game is over, supporters filter away from the stadium, and the crowd dissipates into smaller groups. Although some groups move on to Liverpool's bustling bar and nightclub scene, Dan explains that many supporters return to the same service-ecosystem venues that they had frequented with family and friends before the game: "We used to go straight to into town, but now we wait for the traffic to die down, meet my Dad and his mates in the [pub], you know, we have a quiet drink and reflect on the match."
As part of these reflections, many seek out match highlights on their mobile phones. LFC's marketing team and websites like the Anfield Wrap share video footage of the crowd on social media. Jenny likes how this content "focuses on us, the fans, fan culture, the passion and the singing. They posted one before the Sunderland match.... Footage of fans celebrating wildly at Cardiff the previous week." Video media reminds supporters of the emotional energy felt during an experience of social atmosphere ([17]). In this way, atmospheric recovery links to the preparation stage of the next event, inspiring customer loyalty by encouraging repetitions of the ritual chain. Indeed, digital merchandise motivates Jenny to repeat the ritual chain next week: "I sort of build it up in my head, sing things in my head, watch videos of the crowd, and just get into the mood." James too explains that the songs sung in the stadium and recirculated in social media "are stuck in your head, rattling around in there."
Social atmospheres are created through interaction ritual chains in which shared focus and entrainment make possible expressions of collective effervescence. Social atmospheres can fail, however, when shared focus is distracted, or when entrained groups of consumers are disaggregated. We find that these instances can be caused by consumers as well as by firms, particularly during the climax stage of interaction ritual chains.
A shared focus prefigures entrainment and collective effervescence ([15]); it follows that social atmospheres are inhibited when shared focus is distracted. We reveal three means by which shared focus is distracted: first, when consumers have not prepared to focus; second, when firms remove symbolic objects from place; and third, when firms unwittingly distract from consumer-led atmospheres with spectacular atmospheric stimuli.
Unprepared consumers enter an event space without having learned about the behavioral expectations to participate or the symbols and objects that require shared focus. In such cases, consumers are unable to participate in creating social atmospheres. Anthony calls attention to the heterogeneous nature of EPL consumers, noting "how different supporters are, their mentalities, their mannerisms." Group heterogeneity can lead to misalignments of meanings and practices ([37]; [76]), and although devoted supporters have often been prepared to contribute to social atmospheres from a young age, Brian complains that some casual supporters and tourists enter the stadium
completely alien to the culture that has been nurtured in the Kop and in Anfield over decades.... It leaves a bitter taste for everyone when that culture and feeling is dying in front of your eyes, especially if we take that social experience of togetherness as one of those things that keeps you returning to the match. You get more people now playing on their phones, not paying any attention to what's happening. Some even are on their phones during "You'll Never Walk Alone".... They're just totally disconnected.
If consumers do not encounter the preparation stage of an interaction ritual chain, then they may not have learned about expected behaviors or what symbols require focus. By expressing "too casual an attitude toward the focus of attention" ([ 6], p. 11), consumers can appear irreverent during the formal rituals intended to unite crowds. Brian explains that if tourists fail to join a collective focus, this detracts from the atmosphere and the cultural experience of shared emotions that motivates supporters to repeat interaction rituals and maintain loyalty to Anfield.
Social atmospheres can be disrupted when firms physically alter event spaces ([51]), particularly by removing the symbolic and emotionally charged objects that serve as a shared focus during interaction rituals. When the EPL's popularity surged in the 1990s, many clubs renovated old stadiums or built new ones ([75]). Although improving safety and revenue, these alterations often removed material features of place associated with prior experiences of atmospheres. Amanda complains that these "cultural aspects of the stadium need to be respected. Clubs and architects need to understand and respect the community and the club. Arsenal's move to the 'Emirates Stadium'—a corporate name already—was undermined even more because they delayed moving Highbury's Clock into the stadium. It made it feel soulless." Recalling Michaela's place attachments detailed previously, we interpret the removal of these objects of shared focus as inhibiting those memories that help motivate place attachments and associated desires to repeat social atmospheres.
The removal of symbolic objects also occurs with respect to the mobile symbols that supporters use to gather shared focus among larger groups when activating social atmospheres. For reasons of cost, EPL clubs often outsource security staff, yet Chris explained that these staff can disrupt the visual display supporters create during formal rituals before the game: "They just don't get it, the flags and banners on the Kop. Most people understand that it's all part and parcel of what we do, our culture, but some stewards don't." Staff who remove these mobile symbols can undermine consumers' place-marking activities that enable feelings of shared identity.
Spectacular stimuli are material features of an event space that distract from consumer-led social atmospheres, thereby inhibiting shared focus. As with many experiential places, Anfield's managers provide entertainment and "wow factor" ([69]) with atmospheric stimuli including lights, music, and pyrotechnics. Yet Mark complains that this attempt to "pump in the atmosphere just feels wrong." [11] warn that spectacular environments can reduce consumer participation. In line with this view, Andy explains that spectacular stimuli distract from the shared focus created by supporters themselves: "You're a consumer. You passively just take it in, keep yourself to yourself. There's no opportunity for having a sing, cracking a joke, actually creating part of the atmosphere, there's none of that. It feels artificial." In contrast to literature that explains how firms can use spectacular stimuli to enhance place experiences ([ 9]; [69]), Andy warns that these well-intentioned atmospheric interventions change supporters' roles and divert attention from the place-marking and sound-marking activities consumers cocreate in the stadium. Flags and songs, for example, are symbols of group identity that enable shared focus among supporters and are central to the sensory experience of Anfield. Yet if firm-produced atmospheric stimuli distract from these resources, then place atmospheres can feel artificial.
Social atmospheres are intensified through emotional and behavioral entrainment that gathers among larger groups. Disaggregation refers to the opposite of this process, where entrained crowds are separated into smaller groups or individual consumers: this can inhibit atmosphere and cause disappointment for some consumers. In this subsection, we describe how disaggregation is caused by individualized consumers, unactivated consumers, and crowd pacification.
Individualized consumers refers to the outcome of ticket allocation practices that disaggregate entrained crowds, resulting in passive, lifeless social atmospheres. We have explained that groups of consumers activate entrained behaviors and emotions in the service ecosystem around the stadium before moving to Anfield, transferring a social atmosphere with them. Yet when supporters reach the stadium, these entrained groups can be split up ([75]). Particularly because of the way that tickets are allocated, Tony complains that although many supporters have activated an atmosphere before a game, on entering the stadium they are broken up such that "you no longer get consistent pockets of support, about 150 or 200 people who you could rely on to get the atmosphere going in different bits of the ground." When groups that have activated an atmosphere before an event are fragmented as they enter the event space, the social atmosphere is inhibited. Rather than enjoying the experience of being within an entrained crowd that has previously activated a social atmosphere, Gareth describes how this situation results in an individualized experience of shame:
Me and my mates may be together before the match thinking we'll get the atmosphere going, but because of how tickets are allocated and sold, we may be spread all over Anfield.... You've got to be brave to start the chanting when nobody you know is with you. There's nothing more embarrassing than raising your voice and you're a lone voice in a crowd of 100 people, it's dead shameful, head in your hands sort of stuff.
Unactivated consumers describes an aspect of consumer heterogeneity in which not all consumers are ready to align their emotions and behavior in a crowd. For example, many tourists enter Anfield without activating entrained behaviors before a game such that they are not ready to sing and chant with others. Indeed, Gareth's response indicates that in addition to being separated from those he has activated a social atmosphere with before the game, once in the stadium he may be seated with people who are content to watch the game passively. In such cases, the social atmosphere can feel flat. In response, devoted consumers within the stadium such as members of Spion Kop 1906 may try to rectify the situation by activating passive consumers "on the job," inviting them to join in with easy chants:
[Spion Kop] start a chant so simple, so rhythmic that even those not versed in Liverpool's hymn sheet could join in. The row behind me put more passion into it this time. It's an open invitation for the rest of the ground to join in. People sat close by cooperate, but take a while to adjust to the welcoming slower pace of the chant. The passion peters out. One supporter has had enough. "Are youse gonna give us a hand or what? Why are youse here?!" he shouts at the crowd. Some people in front turn around and roll their eyes, but the majority pretend not to hear him. (Field notes, January 18, 2014).
As these field notes show, attempts to activate entrained behavior can fail. Reflecting on instances where supporters are not able to get "on the same page," Michaela explains that "arguments have broken out in the crowd before." This can sour an atmosphere for Barry, who complains that sitting among unentrained groups of supporters "puts me in a foul mood, and I spend the match complaining. I'd rather watch it with my mates in the pub." Groups can harbor tensions when expectations and emotions are not aligned ([37]; [75]; [76]), leading audiences to dissipate ([61]). This problem motivates Barry's "place detachment" ([ 9]); where he used to loyally attend games at Anfield, he now prefers to watch LFC games on television in the pub with friends.
Crowd pacification refers to deliberate attempts to inhibit entrained behaviors and emotions. Following the accident at Hillsborough, EPL stadiums were modified to improve crowd safety. Where crowds once stood packed together in terraces, new regulations demanded that they be seated and supervised by stadium staff and police ([43]). Despite boasting some of the lowest incidents of crowd disorder in Europe, however, supporters such as Jon complain that these security interventions inhibit the creation of social atmospheres:
Let me show you. We're sitting down now.... I've got my hands in my lap, my shoulders and chest are all tight, and I'm stuck facing in one direction. How am I meant to express myself to others here? What can I do? I can clap, yes. Perhaps shout too, but it won't be loud as I've not got space to have a good old bellow. Without being aware of it, seats change your relationship to the match and others around you. You feel unconnected. Now, standing up [Jon gets up out of his seat]. Look at the difference now. I can move my arms. I can turn around. I can chat to the people behind me. I can take deep breaths and sing! You just feel completely different. You feel connected, able to respond to events on the pitch.
Jon explains that being seated can limit entrained behaviors such as singing and gesturing. Moreover, because seats direct attention toward the field rather than to other members of the crowd, the possibility for crowds to recognize their shared emotions and behaviors is also limited. This hinders the natural rituals that lead to collective effervescence ([15]). Moreover, when supporters ignore these strictures, such as in our description of celebrations following a goal, they risk being ejected by security guards. This problem has motivated some supporters to campaign to restore standing areas within EPL stadiums. Others, however, admit that they have stopped attending EPL games, preferring lower-league matches where standing is still permitted and where social atmospheres meet their expectations.
Social atmospheres are shared experiences of place involving transformations in consumers' emotions and behavior. Our study explains the processes by which these transformations are cocreated during a four-stage interaction ritual chain that spans several interconnected places. We next discuss three theoretical contributions of these findings before recommending how firms can manage social atmospheres in ways that enhance place experiences, customer loyalty, and communal place attachments. Finally, we conclude by considering boundary conditions, the transferability of our findings to other contexts, and opportunities for future research.
Our study makes three theoretical contributions. By explaining the process of creating social atmospheres, we contribute to marketing research in retail atmospherics and servicescapes. By discussing how consumer heterogeneity affects interaction ritual chains, we contribute to theory of interaction ritual chains. Finally, by uncovering how social atmospheres are created in a variety of places, we contribute to research on ritual place-making and interaction ritual chain theory.
Studies of retail environments show how firms create atmospheric experiences of place using multisensory stimuli ([ 4]; [73]) and that collages of these cues enhance place experiences by providing spectacles that prime playful behaviors ([47]; [69]). When service staff and consumers enact mythic and ideological meanings, experiences of place are further enhanced ([ 3]; [ 8]; [22]; [23]). In particular, ritual events imprint places with lasting identities ([10]; [37]). These social constructions of place are described as occurring in "long and slow processes" ([ 9], p. 892) and engender enduring qualities of place ([67]).
Our work extends literature on ritual place-making by theorizing more rapidly changing qualities of place created during collective live events in which emotions and behaviors intensify and climax within groups ([ 3]; [32]; [39]; [45]). We conceptualize "social atmospheres" as a means to explain the process by which individuals are aggregated into crowds by shared focus, and how shared emotions and behaviors can intensify in pleasurable outbursts of collective effervescence. This pleasurable aspect of interaction rituals enhances place experiences and motivates loyalty as desire to repeat social atmosphere. In addition to how communal meanings and practices can drive loyalty ([53]; [65]), we show that social atmospheres also drive loyalty because emotional experiences can help strengthen place attachments ([ 9]).
Moreover, although social atmospheres are conceptually distinct from a spirit of place ([10]), we find that locations with an established spirit of place offer ideal sites for interaction rituals. Our findings indicate that learning about the spirit of place facilitates participation in the cocreation of a social atmosphere. Furthermore, social atmospheres themselves help reconstruct a spirit of place: [25], p. 220) explains that repeated experiences of collective effervescence instill a place with "exceptionally intense forces." In short, in addition to the domesticity of "hestial" places ([10]; [69]), our study contributes an understanding of how an energetic spirit of place can be constructed through the creation of exciting social atmospheres.
Consumer heterogeneity is a common feature of communal consumption events ([77]). Because heterogeneous expectations, motivations, and resource use can lead audiences to dissipate ([37]; [61]), this common dimension of group dynamics is an important consideration in understanding how interaction ritual chains create atmospheres. Despite this, [15] tends to gloss over heterogeneity among interaction ritual participants ([28]). For this reason, our study extends Collins's ([15], [16]) interaction ritual theory by considering interaction ritual chains as social events involving heterogeneous groups of participants that include both consumers and managerial stakeholders.
More specifically, we challenge the notion that entrainment is an automatic result of people being together in a place. [15], p. xix) considers that under these conditions a natural physiological mirroring enables "emotions in one individual's body to become stimulated in the other person's body." Sociology has long discussed crowds as social configurations where individuals appear to think and act as one ([13]; [48]), and prior consumer research has described these effects as "affective contagion," where emotions spread via affective cues such as facial expressions, gestures, and touch ([37]; [38]). Yet our findings indicate that when devoted long-time consumers mix with first-time tourists who are unwilling or unable to participate in entrained behavior, social atmospheres can suffer.
As a corrective to [15] view of entrainment occurring automatically through physiological mirroring, we explain that this aspect of interaction rituals also depends on participants becoming "prepared" and "activated." In other words, shared emotions and behaviors are not entirely "natural"; rather, these effects depend on consumers' learning emotional scripts, procedures, and rules ([37]; [65]). Beyond the kinds of procedures and rules elaborated in prior work, our study shows that learning to participate in entrained behaviors can necessitate opportunities to learn coordinated behaviors such as singing. Consequently, our work extends Collins by explaining that, rather like a rudimentary knowledge of dance steps enables people to move among others on a dance floor, acquired abilities to synchronize group behavior can facilitate social atmospheres.
Third, our study clarifies how consumer interactions in one place can impact atmospheres in another. Atmospheric effects are commonly associated with bounded retail and servicescape locations ([ 4]; [44]; [69]; [73]). [15] too analyzes interaction rituals occurring in discrete contexts. In contrast to a view of places as bounded "containers," however, studies show that consumers prepare for place experiences in different locations and move from place to place as events progress ([10]; [41]; [75]). By tracing interaction rituals across multiple sites, we explain that the intense social atmospheres consumers enjoy at events do not begin spontaneously. Rather, at concerts, sports fixtures, carnivals, and festivals, atmospheres are often activated prior to, and outside of, the places where these events occur.
This contribution aligns with theorizations of place as connected to other places ([14]). Specifically, we view social atmospheres as implicating paths that lead toward and away from places ([18]). In terms of pathways that lead toward a place, social atmospheres are activated before the climax of a "main event" among smaller gatherings of consumers, much like the tailgating events described by [10]. In terms of the paths that lead away from a place, we find that a stage of recovery sees consumers calming and breaking back into smaller groups who frequent a wider service ecosystem ([ 1]) close to an event place, where they plan for repetitions of the social atmosphere.
Finally, with respect to the possibility for atmospheres in one place to move into another place, an important contribution of our study is to explain how different stages and contexts of the interaction ritual chain are linked. Our findings show that symbolic resources such as flags and songs are a constant presence throughout the interaction ritual chain. A unique implication of our study is that these symbolic resources can become mobile, and the transfer of these mobile resources from place to place connects the stages of an interaction ritual chain. Prepared in private places before being transferred through the service ecosystem and into an event space, visual and aural symbols provide a mobile focus that helps gather larger crowds, thereby facilitating the intensification of a social atmosphere as an interaction ritual chain progresses.
Managers play important roles in facilitating ritual interactions ([53]; [60]). To help firms benefit from social atmospheres, this subsection identifies and illustrates how a variety of managerial stakeholders—from small businesses to managers of large event spaces—can facilitate each stage of the interaction ritual chain (for a summary, see Table 2). Our insights go beyond existing managerial advice by illustrating best practice at different stages of the interaction ritual chain before, during, and after a main event. Finally, we explain how tensions in the creation of social atmospheres can be managed.
Graph
Table 2. Managerial Recommendations.
| Stage in the Interaction Ritual Chain | Group Size | Space | Firm's Managerial Roles | Roles of Other Stakeholders | Instances in Prior Research on Collective Live Events |
|---|
| Preparations: Consumer-led activities before events that provide resources to later enable atmospheres. | Intimate | Private | Educate consumers through marketing communications that stress behavioral expectations. | Family and friendship groups set behavioral expectations. | Festivals promote unique ethos through marketing communications (Kozinets 2002; Flinn and Frew 2014) |
| Promote the cocreation of focal symbolic resources. | Passionate entrepreneurs and brand devotees help make symbolic resources. | Nightclubbers prepare costumers for the event (Goulding et al. 2009) |
| Cocreate events with passionate entrepreneurs. | | TEDx events training (Fidelman 2012) |
| Activation: Creation of social atmosphere through focus and entrainment among small groups of consumers before a main event. | Small-scale | Ecosystem | Provide consumers with a visitors' guide that mythologizes local landmarks and places. | Businesses in service ecosystem help welcome and encourage consumers to activate. | Tailgating activities prior to American football games (Bradford and Sherry 2018) |
| Promote venues in the ecosystem in which activations occur. | Brand devotees lead activations. | Preparty events (Goulding et al. 2009) |
| Security services provide correct safety provisions for activations to occur. | Pre–Mardi Gras parades (De Jong 2015) |
| Climaxes: The culmination of collective live events when collective effervescence is shared among crowds. | Crowd | Event space | Stage formal climaxes by triggering focus, entrainment, and collective effervescence. | Brand devotees lead natural climax. | End of festival rituals (Anderton 2019; Kozinets 2002) |
| Enable natural atmospheres through spaces design to enable entrainment without distraction. | Casual consumers follow the lead offered by brand devotees. | Music concerts and events (Goulding et al. 2009) |
| Security services facilitate climax rituals through developing specialized knowledge. | Worship in megachurches (Wagner 2019) |
| | Live sports (Holt 1995) |
| Recovery: Shared emotions and memories of atmosphere are stored in resources that inspire repetitions of rituals. | Intimate | Ecosystem | Provide merchandise postevent. | Businesses in the ecosystem welcome and encourage consumers to recover together. | Digital merchandise produced out of music events (Anderton 2019) |
| Circulate video footage of consumers enjoying collective effervescence. | | Distributing recordings of climaxes in megachurches (Wagner 2019) |
Preparations facilitate social atmospheres by teaching consumers about behavioral and emotional expectations before an event. Our findings illuminate opportunities for firms to facilitate preparations. Firms can play important roles in educating consumers about cultural meanings and behavioral expectations using media communications and social networks ([65]) and by partnering with "passionate entrepreneurs," consumers who translate their knowledge of a community into offerings that support shared passions ([34]). For example, TED events are characterized by a highly charged elitist atmosphere. To create this experience, the TED organization provides guidelines to enthusiasts who arrange local TEDx events ([27]). These guidelines prepare speakers on how to "TED," covering topics such as creating material and rehearsing speakers' body language to elicit focus from the audience. These preparations also encourage consumers to become enthused and entrained when they attend, all of which benefits the TED brand by enhancing consumer experiences.
Activations facilitate atmosphere by providing opportunities for smaller groups to warm up before a main event. To enable activation, we recommend that firms create partnerships with businesses that operate in the broader ecosystem around a main event site, because these are places where consumers often gather before an event, as is the case with the bikers on their way to Sydney's Gay and Lesbian Mardi Gras ([21]). Riding in formation creates a shared focus and entrained behavior in small groups ([66]). Thereafter, bikers gather with others into larger unified crowds as they travel toward the carnival. Oftentimes these periods of mobility are as important as the event itself, and for this reason we recommend managers facilitate consumer "pilgrimages" to events by providing information about potential activities, venues, and points of interest along the way. For example, at both Rio de Janeiro's Carnival and New Orleans's Mardi Gras, consumers are guided toward the many events and activities on offer before the main parade. However, given that emotions and behaviors intensify among crowds in public spaces, event managers will benefit from dialogue with public authorities as well as police and security services.
Ritual climaxes provide pleasurable moments of open-boundary communication with others ([ 3]; [15]). This can enhance customer experiences and drive loyalty, as consumers seek to repeat rituals to reexperience collective effervescence. To facilitate climaxes, managers can first orchestrate formal rituals that focus and entrain smaller groups of consumers into larger audiences and crowds. Consider megachurches such as the enormously successful Hillsong. Much like musicians at concerts and festivals can unite huge crowds by having them clap and sing together, charismatic preachers begin events with formalized activities intended to focus and entrain a congregations' behaviors and emotions ([15]; [80]). Firms have important roles to play in providing spaces that foster these aspects of interaction rituals. For instance, megachurches benefit from architectural designs that facilitate consumers' awareness of their aligned emotions and behaviors. By enabling recognition of entrainment among a congregation, megachurches also facilitate natural rituals in which groups begin to express themselves more spontaneously. Preachers encourage worshippers to vocalize, or stand with hands raised, for example. When congregations can see, touch, and hear one another during these moments ([80]), their shared actions are more likely to climax in experiences of collective effervescence.
The recovery stage is when shared emotions and memories of atmosphere are stored in the symbolic resources that consumers have carried through interaction ritual chains. We have shown how merchandise associated with experiences of collective effervescence can store feelings of emotional energy. Such objects are used by consumers in the preparation and activation stages of future events and can therefore be useful drivers of loyalty. In addition, we recommend that firms create digital merchandise. By circulating footage of crowds enjoying a ritual climax, firms facilitate conversations between consumers about their shared emotional experiences. For example, music events such as Glastonbury and Coachella distribute media that focus on consumers as much as musicians to advertise climactic moments where an atmosphere becomes "electric" ([29]). By showcasing crowds enjoying shared focus, entrainment, and collective effervescence, digital merchandise can enhance loyalty by inspiring consumers' desires to repeat interaction rituals. Moreover, digital merchandise connects the recovery stage of one event to the preparation stages of future events by educating consumers about expectations such as dress and emotional conduct ([ 2]; [35]).
Despite opportunities for firms to facilitate interaction ritual chains, we recognize that social atmospheres can fail. Next, therefore, we provide recommendations to avoid failures by navigating three common tensions identified in our analysis.
Creating social atmospheres requires different management techniques compared with those used to manage environments intended to offer spectacular experiences, such as ESPN Zone ([47]; [69]), American Girl Place ([22]), or Disney World ([40]). Studies of these places note how managers assemble dazzling arrays of objects and "zoned" activities that provide myriad points of focus for smaller groups of consumers ([62]). In contrast, managers who want to enable social atmospheres should minimize distractions from a common focus of attention necessary to initiate entrainment and collective effervescence. For this reason, pumping in spectacular stimuli such as upbeat music, lights, and fireworks ([11]) can harm social atmospheres by distracting the crowd's attention. To facilitate shared focus in crowds, we recommend that firms use symbols that have previously been "charged" with emotions and group significance during prior interaction rituals ([15]). For instance, the Boston Red Sox have played Neil Diamond's "Sweet Caroline" during games for 20 years, providing a focus to prime a giant sing-along and a pleasurable social atmosphere ([71]). Managers should be aware that meaningless symbols are less likely to create shared focus. For example, when the Sydney Swans football club tried to activate similar conviviality using "Sweet Caroline," consumers complained the song "has nothing to do with the club" ([33]).
Previous research notes how heterogeneity among consumers can cause group tensions, particularly when "newcomers" mix face-to-face with established "insiders" ([66]; [77]). When music festivals grow in popularity, for example, a common complaint is that newcomers detract from the atmosphere ([ 2]). We have witnessed this tension between brand devotees and casual consumers, some of whom fail to join in with a shared focus and entrainment, leading to failures in social atmosphere. Managers can play a key role in solving this problem. A common strategy is to provide access to differentiated areas of the event space so as to separate consumer segments according to their expectations and behaviors, such as when season ticket holders are seated together. On the one hand, this possibility can help preserve groups of prepared and activated consumers who want to create social atmospheres. On the other hand, by separating devoted insiders from outsiders, these interventions may exacerbate differences between groups. Reinforcing differences in expectations and willingness to contribute to social atmospheres may prevent more experienced consumers from helping neophytes to join in ([65]). Brandfests offer a useful precedent here: when newer consumers mix with seasoned consumers, both groups benefit, as outsiders are able to learn and feel welcomed, while insiders often "relish the recognition and status" afforded them ([53], p. 42).
At collective live events, managers must inevitably assess myriad risks and implement safety precautions ([57]). How can the possibility to create collective effervescence be balanced against the necessity to manage risks and disorder? First, educate security and service staff to understand consumers' interaction ritual chains. Firms can avoid disrupting the entrainment and collective effervescence that contribute to social atmosphere by making staff aware of the places where consumers may have been before they arrive at an event, or the kinds of group behavior that they may have developed on the way there. Equally, service and security staff should be aware that collective effervescence may involve unusual behavior and unpredictable outbursts of emotion. For example, security at music festivals can misinterpret collective effervescence in a "mosh pit" as violent disorder, yet for consumers, this behavior is expected and enjoyed and, in most cases, is not harmful ([70]). These outcomes can benefit managers of security companies willing to build specialized knowledge of interaction rituals. Having witnessed ritual events time and again, security and service staff can assume roles of trusted facilitators, whose experience enables them to allow consumers to produce social atmospheres and to intervene only when necessary.
Our study explains the processes by which social atmospheres are created during interaction ritual chains intended to create the kinds of lively social atmosphere that typify many collective live events. Our analysis is transferable to other contexts that share characteristics encompassed in the following boundary conditions. First, our study is applicable to events at which the collective effervescence emerges as moments of positive affect. Future research could examine how interaction rituals produce atmospheres of grief, serenity, or outrage, which may deviate from the process we have described.
Second, our study is applicable to social atmospheres in both small contexts such as pubs and bars and larger, more organized live events such as sports and music events, religious ceremonies, festivals, and carnivals. A benefit of our theorization is to recognize how smaller businesses (e.g., pubs, cafes) are interconnected to larger organizations (e.g., football clubs) or places (e.g., stadiums). We note how smaller businesses that become aware of their place in an interaction ritual chain can facilitate activation and recovery activities; future research could examine how other kinds of small businesses in the ecosystem can benefit from their roles in these chains. Future research could also examine mobile social atmospheres in larger ecosystems, such as extended city-based events like the Edinburgh Fringe Festival or the Olympics.
Third, our theory is especially relevant to recurring events, as the cyclical nature of interaction ritual chains means that the recovery stage feeds into the preparation stage. Related to this, our analysis suggests that social atmospheres depend on the presence of consumer groups who have become used to contributing to atmosphere through repeated participation in interaction ritual chains. Thus, our insights are less relevant for one-off live events that are never repeated. Future research could investigate how social atmospheres can be developed at first-time or one-off events, which may differ from the process we have outlined.
Finally, our insights are most relevant in face-to-face contexts in which communal interactions occur. Our theory has less relevance in more individualized retail experiences for which the findings of conventional atmospherics research remain relevant. However, as a result of face-to-face interactions being limited during 2020 and beyond, the role of technology in facilitating social interactions is changing. For [15], face-to-face contact is critical for interaction rituals, and for this reason, he claims that interaction rituals cannot occur in digital environments. Nevertheless, rapid developments in virtual reality technologies may enable access to digitally mediated social atmospheres as a way of enhancing experiential places such as museums and historical and educational sites or allowing consumers to experience collective live events remotely. These possibilities present exciting opportunities for future research and practice, not least because people need connections with others and suffer without them. Marketing has important roles to play in facilitating connections and social systems, and we hope that this study inspires managerial practice that enables social atmospheres to flourish.
Footnotes 1 Ashlee Humphreys
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The authors would like to thank the University of Melbourne's International Research Scholarship for funding this research.
4 Tim Hill https://orcid.org/0000-0002-5758-7127 Giana M. Eckhardt https://orcid.org/0000-0001-7856-0719
5 Online supplement:https://doi.org/10.1177/00222429211023355
References Akaka Melissa A. , Vargo Stephen L.. (2015), " Extending the Context of Service: From Encounters to Ecosystems ," Journal of Services Marketing 29 (6/7), 453 – 62.
Anderton Chris. (2019), Music Festivals in the UK: Beyond the Carnivalesque. New York : Routledge.
Arnould Eric J. , Price Linda L.. (1993), " River Magic: Extraordinary Experience and the Extended Service Encounter ," Journal of Consumer Research , 20 (1), 24 – 45.
Baker Julie , Parasuraman Albert , Grewal Dhruv , Voss Glenn B.. (2002), " The Influence of Multiple Store Environment Cues on Perceived Merchandise Value and Patronage Intentions ," Journal of Marketing , 66 (2), 120 – 41.
Belk Russell W. , Costa Janeen Arnold. (1998), " The Mountain Man Myth: A Contemporary Consuming Fantasy ," Journal of Consumer Research , 25 (3), 218 – 40.
6 Belk Russell W. , Wallendorf Melanie , Sherry John F. Jr. (1989), " The Sacred and the Profane in Consumer Behavior: Theodicy on the Odyssey ," Journal of Consumer Research , 16 (1), 1 – 38.
7 Biswas Dipayan , Lund Kaisa , Szocs Courtney. (2019), " Sounds Like a Healthy Retail Atmospheric Strategy: Effects of Ambient Music and Background Noise on Food Sales ," Journal of the Academy of Marketing Science , 47 (1), 37 – 55.
8 Borghini Stefania , Diamond Nina , Kozinets Robert V. , McGrath Mary Ann , Muñiz Albert M. , Sherry John F. Jr. (2009), " Why Are Themed Brandstores So Powerful? Retail Brand Ideology at American Girl Place ," Journal of Retailing , 85 (3), 363 – 75.
9 Borghini Stefania , Sherry John F. Jr. , Joy Annamma. (2020), " Attachment to and Detachment from Favorite Stores: An Affordance Theory Perspective ," Journal of Consumer Research , 47 (6), 890 – 913.
Bradford Tonya Williams , Sherry John F. Jr. (2015), " Domesticating Public Space Through Ritual: Tailgating as Vestaval ," Journal of Consumer Research , 42 (1), 130 – 51.
Bradford Tonya Williams , Sherry John F. Jr. (2018), " Dwelling Dynamics in Consumption Encampments: Tailgating as Emplaced Brand Community ," Marketing Theory , 18 (2), 203 – 17.
Burawoy Michael. (1998), " The Extended Case Method ," Sociological Theory , 16 (1), 4 – 33.
Canetti Elias. (1981), Crowds and Power. New York : Continuum.
Coffin Jack , Chatzidakis Andreas. (2021) " The Möbius Strip of Market Spatiality: Mobilizing Transdisciplinary Dialogues Between CCT and the Marketing Mainstream ," AMS Review , 11 , 40 – 59.
Collins Randall. (2004), Interaction Ritual Chains. Princeton, NJ : Princeton University Press.
Collins Randall. (2014), " Interaction Ritual Chains and Collective Effervescence, " in Collective Emotions , von Scheve C. , Salmela M. , eds. Oxford, UK : Oxford University Press , 299 – 311.
Collins Randall. (2016), " Micro-Sociology of Sport: Interaction Rituals of Solidarity, Emotional Energy, and Emotional Domination ," European Journal for Sport and Society , 13 (3), 197 – 207.
Cresswell Tim. (2004), Place: An Introduction. London : John Wiley & Sons.
Cronin James , Cocker Hayley L.. (2019), " Managing Collective Effervescence: 'Zomsumption' and Postemotional Fandom ," Marketing Theory , 19 (3), 281 – 99.
Debenedetti Alain , Oppewal Harmen , Arsel Zeynep. (2014), " Place Attachment in Commercial Settings: A Gift Economy Perspective ," Journal of Consumer Research , 40 (5), 904 – 23.
De Jong Anna. (2015), " Dykes on Bikes: Mobility, Belonging and the Visceral ," Australian Geographer , 46 (1), 1 – 13.
Diamond Nina , Sherry John F. Jr. , Muñiz Albert M. , McGrath Mary Ann , Kozinets Robert V. , Borghini Stefania. (2009), " American Girl and the Brand Gestalt: Closing the Loop on Sociocultural Branding Research ," Journal of Marketing , 73 (3), 118 – 34.
Dion Delphine , Borraz Stéphane. (2017), " Managing Status: How Luxury Brands Shape Class Subjectivities in the Service Encounter ," Journal of Marketing , 81 (5), 67 – 85.
Dolbec Pierre-Yann , Fischer Eileen. (2015), " Refashioning a Field? Connected Consumers and Institutional Dynamics in Markets ," Journal of Consumer Research , 41 (6), 1147 – 68.
Durkheim Émile. (1995), The Elementary Forms of Religious Life. New York : The Free Press.
Ebner Sarah. (2013), " History and Time are Key to Power of Football, Says Premier League Chief," The Times (July 2), https://www.thetimes.co.uk/article/history-and-time-are-key-to-power-of-football-says-premier-league-chief-3d3zf5kb35m.
Fidelman Mark. (2012), "Here's Why TED and TEDx Are So Incredibly Appealing," Forbes (June 19), https://www.forbes.com/sites/markfidelman/2012/06/19/heres-why-ted-and-tedx-are-so-incredibly-appealing-infographic.
Fine Gary A. (2005), " Interaction Ritual Chains ," Social Forces , 83 (3), 1287 – 88.
Flinn Jenny , Frew Matt. (2014), " Glastonbury: Managing the Mystification of Festivity ," Leisure Studies , 33 (4), 418 – 33.
Geertz Clifford. (1973), The Interpretation of Cultures. New York : Basic Books.
Goffman Erving. (1967), Interaction Ritual: Essays on Face-to-Face Behavior. New York : Pantheon.
Goulding Christina , Shankar Avi , Elliott Richard , Canniford Robin. (2009), " The Marketplace Management of Illicit Pleasure ," Journal of Consumer Research , 35 (5), 759 – 71.
Griffiths Neil. (2017), " Sydney Swans Smack Down Petition to Stop 'Sweet Caroline' Being Played at Games," The Music (May 8), https://themusic.com.au/news/new-petition-calls-for-sydney-swans-to-stop-playing-sweet-caroline-at-home-games/kwKEh4aJiIs/09-05-17/.
Guercini Simone , Cova Bernard. (2018), " Unconventional Entrepreneurship ," Journal of Business Research , 92 , 385 – 91.
Haider Arwa. (2019), " From Woodstock to Coachella: The Ultimate Music Festivals," (April 5), https://www.bbc.com/culture/article/20190405-from-woodstock-to-coachella-the-ultimate-lost-weekends.
Handler Paul. (2014), " Ear-Splitting Anfield Harks Back to Football Days Gone By," Manchester Evening News (April 15), https://www.manchestereveningnews.co.uk/sport/football/football-news/anfield-atmosphere-liverpool-vs-manchester-6987405.
Higgins Leighanne , Hamilton Kathy. (2019), " Therapeutic Servicescapes and Market- Mediated Performances of Emotional Suffering ," Journal of Consumer Research , 45 (6), 1230 – 53.
Hill Tim , Canniford Robin , Mol Joeri. (2014), " Non-Representational Marketing Theory ," Marketing Theory , 14 (4), 377 – 94.
Holt Douglas B. (1995), " How Consumers Consume: A Typology of Consumption Practices ," Journal of Consumer Research , 22 (1), 1 – 16.
Houston Rika , Meamber Laurie. (2011), " Consuming the 'World': Reflexivity, Aesthetics and Authenticity at Disney World's EPCOT Center ," Consumption Markets & Culture , 14 (2), 177 – 91.
Husemann Katharina C. , Eckhardt Giana M.. (2019), " Consumer Deceleration ," Journal of Consumer Research , 45 (6), 1142 – 63.
Kelly Stephen. (2007), The Kop: Liverpool's Twelfth Man. London : Virgin.
King Anthony. (2002), End of the Terraces: The Transformation of English Football. London : Bloomsbury.
Kotler Phillip. (1974), " Atmospherics as Marketing Tool ," Journal of Retailing , 49 (4), 48 – 84.
Kozinets Robert V. (2002), " Can Consumers Escape the Market? Emancipatory Illuminations from Burning Man ," Journal of Consumer Research , 29 (1), 20 – 38.
Kozinets Robert V. (2020), Netnography: The Essential Guide to Qualitative Social Media Research. London : Sage.
Kozinets Robert V. , Sherry John F. Jr. , Storm Diana , Duhachek Adam , Nuttavuthisit Krittinee , DeBerry-Spence Benét. (2004), " Ludic Agency and Retail Spectacle ," Journal of Consumer Research , 31 (3), 658 – 72.
Le Bon Gustave. (1960), The Crowd: A Study of the Popular Mind. New York : Viking Press.
Lee Na Young , Noble Stephanie M. , Biswas Dipayan. (2018), " Hey Big Spender! A Golden (Color) Atmospheric Effect on Tipping Behavior." Journal of the Academy of Marketing Science , 46 (2), 317 – 37.
MacInnes Paul. (2016), " Premier League Must Fight the Tide of Increasingly Quiet Crowds," The Guardian (November 16), https://www.theguardian.com/football/blog/2016/nov/16/premier-league-quiet-crowds-crystal-palace.
Maclaran Pauline , Brown Stephen. (2005), " The Center Cannot Hold: Consuming the Utopian Marketplace ," Journal of Consumer Research , 32 (2), 311 – 23.
Madzharov Adriana V. , Block Lauren G. , Morrin Maureen. (2015), " The Cool Scent of Power: Effects of Ambient Scent on Consumer Preferences and Choice Behavior ," Journal of Marketing , 79 (1), 83 – 96.
McAlexander James H. , Schouten John W. , Koenig Harold F.. (2002), " Building Brand Community ," Journal of Marketing , 66 (1), 38 – 54.
McCracken Grant. (1988), The Long Interview. New York : SAGE Publications.
Merchant Stephanie. (2011), " The Body and the Senses: Visual Methods, Videography and the Submarine Sensorium ," Body & Society , 17 (1), 53 – 72.
Millward Peter. (2011), The Global Football League. London : Palgrave.
Mowen Andrew J. , Vogelsong Hans G. , Graefe Alan R.. (2002), " Perceived Crowding and Its Relationship to Crowd Management Practices at Park and Recreation Events ," Event Management , 8 (2), 63 – 72.
Müller Victor , Lindenberger Ulman. (2011), " Cardiac and Respiratory Patterns Synchronize Between Persons During Choir Singing ," PloS One , 6 (9), e24893.
Muñiz Albert M. and Thomas C. O'Guinn (2001), " Brand Community ," Journal of Consumer Research , 27 (4), 412 – 32.
Otnes Cornelia C. , Ilhan Behice Ece , Kulkarni Atul. (2012), " The Language of Marketplace Rituals: Implications for Customer Experience Management ," Journal of Retailing , 88 (3), 367 – 83.
Parmentier Marie-Agnes , Fischer Eileen. (2015), " Things Fall Apart: The Dynamics of Brand Audience Dissipation ," Journal of Consumer Research , 41 (5), 1228 – 51.
Peñaloza Lisa. (1998), "Just Doing It: A Visual Ethnographic Study of Spectacular Consumption Behavior at Nike Town," Consumption , Markets and Culture , 2 (4), 337 – 400.
Premier League (2020), " Full and Vibrant Stadiums," (accessed July 28, 2020), https://www.premierleague.com/this-is-pl/the-football/408985.
Schafer R. Murray. (1993), The Soundscape: Our Sonic Environment and the Tuning of the World. New York : Simon and Schuster.
Schau Hope Jensen , Muñiz Albert M. , Arnould Eric J.. (2009), " How Brand Community Practices Create Value ," Journal of Marketing , 73 (5), 30 – 51.
Schouten John W. , McAlexander James H.. (1995), " Subcultures of Consumption: An Ethnography of the New Bikers ," Journal of Consumer Research , 22 (1), 43 – 61.
Seamon David. (2018), Life Takes Place: Phenomenology, Lifeworlds, and Place Making. New York : Routledge.
Seregina Anastasia , Weijo Henri. (2017), " Play at Any Cost: How Cosplayers Produce and Sustain their Ludic Communal Consumption Experiences ," Journal of Consumer Research 44 (1), 139 – 59.
Sherry John F. Jr. , Kozinets Robert V. , Storm Diana , Duhachek Adam , Nuttavuthisit Krittinee , DeBerry-Spence Benét. (2001), " Being in the Zone: Staging Retail Theater at ESPN Zone Chicago ," Journal of Contemporary Ethnography , 30 (4), 465 – 510.
Sinclair Gary , Dolan Paddy. (2015), " Heavy Metal Figurations: Music Consumption, Subcultural Control and Civilizing Processes ," Marketing Theory , 15 (3), 423 – 41.
Slane Kevin. (2017), " How 'Sweet Caroline' Became Fenway's Beloved (and Detested) Ballpark Anthem," Boston.com (October 8), https://www.boston.com/sports/boston-red-sox/2017/10/08/how-sweet-caroline-became-fenways-beloved-and-detested-ballpark-anthem/.
Spangenberg Eric R. , Crowley Ayn E. , Henderson Pamela W.. (1996), " Improving the Store Environment: Do Olfactory Cues Affect Evaluations and Behaviors? " Journal of Marketing , 60 (2), 67 – 80.
Spence Charles , Puccinelli Nancy M. , Grewal Dhruv , Roggeveen Anne L.. (2014), " Store Atmospherics: A Multisensory Perspective ," Psychology & Marketing , 31 (7), 472 – 88.
Spiggle Susan. (1994), " Analysis and Interpretation of Qualitative Data in Consumer Research ," Journal of Consumer Research , 21 (3), 491 – 503.
Steadman Chloe , Roberts Gareth , Medway Dominic , Millington Steve , Platt Louise. (2021), " (Re)thinking Place Atmospheres in Marketing Theory ," Marketing Theory , 21 (1), 135 – 54.
Thomas Tandy Chalmers , Epp Amber M. , Price Linda L.. (2020), " Journeying Together: Aligning Retailer and Service Provider Roles with Collective Consumer Practices ," Journal of Retailing , 96 (1), 9 – 24.
Thomas Tandy Chalmers , Price Linda L. , Schau Hope J.. (2013), " When Differences Unite: Resource Dependence in Heterogeneous Consumption Communities ," Journal of Consumer Research , 39 (5), 1010 – 33.
Thompson Craig J. , Locander William B. , Pollio Howard R.. (1989), " Putting Consumer Experience Back into Consumer Response: The Philosophy and Method of Existential-Phenomenology ," Journal of Consumer Research , 16 (2), 133 – 46.
Visconti Luca M. , Sherry John F. Jr. , Borghini Stefania , Anderson Laurel. (2010), " Street Art, Sweet Art? Reclaiming the 'Public' in Public Place ," Journal of Consumer Research , 37 (3), 511 – 29.
Wagner Tom. (2019), Music, Branding and Consumer Culture in Church: Hillsong in Focus. London : Routledge.
~~~~~~~~
By Tim Hill; Robin Canniford and Giana M. Eckhardt
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 128- The Upside of Negative: Social Distance in Online Reviews of Identity-Relevant Brands. By: Ordabayeva, Nailya; Cavanaugh, Lisa A.; Dahl, Darren W. Journal of Marketing. Mar2022, p1. DOI: 10.1177/00222429221074704.
Ahead of Print- Database:
- Business Source Complete
Record: 129- They're Just Not That into You: How to Leverage Existing Consumer–Brand Relationships Through Social Psychological Distance. By: Connors, Scott; Khamitov, Mansur; Thomson, Matthew; Perkins, Andrew. Journal of Marketing. Sep2021, Vol. 85 Issue 5, p92-108. 17p. 2 Charts, 2 Graphs. DOI: 10.1177/0022242920984492.
- Database:
- Business Source Complete
They're Just Not That into You: How to Leverage Existing Consumer–Brand Relationships Through Social Psychological Distance
While prevailing marketing practice is to encourage ever-stronger relationships between consumers and brands, such relationships are rare, and many consumers are relationship-averse or content with the status quo. The authors examine how marketers can more effectively manage existing brand relationships by focusing on the psychological distance between consumers and brands in order to match close (distant) brands with concrete (abstract) language in marketing communications. Through such matching, marketers can create a beneficial mindset-congruency effect leading to more favorable evaluations and behavior, even for brands that are relatively distant to consumers. Study 1 demonstrates the basic mindset-congruency effect, and Study 2 shows that it is capable of affecting donation behaviors. Study 3 documents two brand-level factors (search vs. experience goods, brand stereotypes) that moderate this effect in managerially relevant ways. Study 4 shows that activation of the mindset-congruency effect influences consumers to spend more and that these behaviors are moderated by consumer category involvement. The authors conclude with marketing and theoretical implications.
Keywords: brand relationship type; brand relationship management; consumer–brand relationships; construal level; mindset congruency; psychological distance
Just as parents find it difficult to be objective about their children, so it is with marketing managers and their brands. It is hard to see the brand from the consumer's perspective. It is difficult to appreciate the minor role the brand plays in the life of the consumer....No one else loves your brand as much as you do.
—[40], pp. 6–7)
Marketing managers want consumers to form strong connections with their brands. Building on [29] consumer–brand relationship (CBR) framework, research has mapped more than 50 relationship types ([85]), many of them meaningful and closely tied to the consumer's sense of self. In the academic literature, there has been a notable propensity to focus on these kinds of relationships, such as committed partnerships and best friendships ([30]). This focus is also reflected in prevailing brand-management approaches that aim to move consumers from "weak or indifferent" relationships ([30], p. 253) to stronger ones in which the consumer is more attached to, connected to, or in love with a brand ([58]). After all, if stronger brand relationships are commercial assets that "offer the greatest economic profit potential" ([67], p. 379), then pursuing stronger relationships seems to be a sound strategy.
However, as indicated by the sheer volume and variation of CBR types, consumers often do not experience or seek strong brand relationships. Strong relationships are, in fact, "rare in a brand context" ([77], p. 89), and as many as 77% of consumers report that they do not forge strong relationships with brands ([32]). Similarly, large-scale practitioner research by [38] suggests that a majority of brand content is not meaningful, and consumers would not care if most brands "disappeared." To underscore this point, we conducted a simple test: we recruited 323 consumers from a private research panel (Mage = 37 years; 47% male) and asked them to list all the brands that were important to them. On average, respondents listed only 2.15 brands (SD = 1.50), and fewer than 1% of respondents listed 10 or more brands. Thus, while most consumers have at least one brand that they feel strongly about, there appears to be a low ceiling to this phenomenon. All of this speaks to a situation lamented decades ago that persists today ([31], p. 44):
Every company wants the rewards of long term, committed partnerships. But people maintain literally hundreds of one-on-one relationships in their personal lives...and clearly, only a handful of them are of a close and committed nature. How can we expect people to do any more in their lives as consumers?
The implication is that marketers are fixated on building the types of relationships that countless consumers simply may not want, in essence choosing a potentially wasteful relationship-upgrading strategy as a result of disregarding consumer preferences. In response, we highlight the value of marketers' embracing the relationship status quo and argue in favor of a simplified strategy based on a phenomena-to-construct ([57]) assessment of CBRs. We leverage the fact that all the major CBR constructs (e.g., love, attachment, identification) implicitly or explicitly reflect the idea of self–brand distance, defined as the psychological proximity between a brand and the consumer's self-concept. We show that different types of brand relationships are associated with varying levels of psychological distance ([78]) and expand on this theoretical mapping to demonstrate how to better leverage existing CBRs. Drawing on construal level theory ([78]), we establish a congruency effect, showing that matching the psychological distance associated with a CBR to an appropriate level of construal or message concreteness improves brand evaluations and spending for both close and distant consumers. In five studies, we offer the first empirical demonstration that social psychological distance is common to many major CBR constructs, that matching distance and construal level in marketing communications results in superior consumer evaluations and behaviors, and that these effects are moderated by variables with strong implications for how marketers can respond more effectively to consumers' existing brand relationships.
Consumers engage in many types of brand relationships, and most implicate their self-concept. This basic idea is captured with different terms such as self–brand connection ([24]), self–brand overlap ([ 9]), self–brand distance ([66]), self-connection ([29]), and self-concept connection (e.g., [76]). While each possesses its own nuance, they converge on reflecting "the extent to which the brand overlaps with or is included in the self; that is, the extent to which the brand is me and I am the brand" ([56], p. 364). Extensive evidence of this self–brand distance is embedded in core CBR concepts such as brand identification, commitment, attachment, and love. For example, previous work proposed that self-connection—the extent to which a brand reflects and expresses important aspects of the self—is a vital component of how brands can become meaningful relationship partners ([29]). Other work draws from self-expansion theory ([71]) to posit that brand relationships are formed as part of an unconscious motivation to expand the self and include close others in the self-concept. What these differing accounts make clear is that many brand relationships largely implicate closeness of the brand to the self-concept (see Table 1).
Graph
Table 1. Self–Brand Distance Across a Selection of CBR Constructs.
| CBR Construct (Focal Component) | Sources | Sample Items |
|---|
| Attachment–aversion relationship (Brand self-distance) | Park, Eisingerich, and Park (2013) | I am personally connected to [personally disconnected from] the brand; The brand is very close to me [very far away] and [not] who I am. |
| Brand attachment (Brand self-connection) | Park et al. (2010) | To what extent is [brand] part of you and who you are? To what extent do you feel personally connected to [brand]? |
| Brand commitment (Affective commitment) | Fullerton (2005); Lee, Huang, and Hsu (2007) | [Brand] has a great deal of personal meaning to me; I feel a strong sense of identification with [brand]; I feel emotionally attached to [brand]. |
| Brand identification | Einwiller et al. (2006); Homburg, Wieseke, and Hoyer (2009); Stokburger-Sauer, Ratneshwar, and Sen (2012) | I strongly identify with this [brand]; I feel attached to this [brand]; Being a customer of [brand] is part of my sense of who I am. |
| Brand love (Self–brand integration) | Bagozzi, Batra, and Ahuvia (2017); Batra, Ahuvia, and Bagozzi (2012) | My personal identity and this brand's identity match; Using this brand says something "true" and "deep" about who I am as a person. |
| Brand relationship quality (Self-connection or self-concept connection) | Lee and Aaker (2004); Swaminathan, Page, and Gurhan-Canli (2007) | This brand says a lot about the kind of person I would like to be; This brand makes a statement about what is important to me in life. |
| Brand self-expression | Carroll and Ahuvia (2006) | This brand symbolizes the kind of person I really am inside; The brand is an extension of my inner self. |
| Brand self-relevance | Eisingerich and Rubera (2010) | [Brand] means a great deal to me; I cannot imagine life without [brand]. |
| Ego involvement | Beatty, Homer, and Kahle (1988); Beatty and Kahle (1988) | I can make many connections or associations between my use of [brand] and experiences in my life; The brands I use say a lot about who I am. |
| Inclusion of brand in self (IOS) | Aron, Aron, and Smollan (1992); Reimann et al. (2012) | Zipper scale |
| Psychological distance | Choi and Winterich (2013) | Zipper scale |
| Self–brand congruence (Actual, ideal) | Malar et al. (2011) | The personality of [brand] is consistent with how I see myself (my actual self); The personality of [brand] is a mirror image of the person I would like to be (my ideal self). |
| Self–brand connection | Escalas (2004); Escalas and Bettman (2003) | I consider this brand to be "me"; This brand reflects who I am. |
We employ psychological distance as a useful complement to the CBR literature as a means of conceptualizing self–brand distance. Psychological distance refers to the "subjective experience that something is close or far away from the self, here, and now" ([78], p. 440). At its core, psychological distance reflects the subjective feeling of how far, in abstract psychological space, a target (e.g., object, event) is perceived to be from the self ([ 7]). We suggest that psychological distance can be construed as the foundation underlying the numerous conceptualizations of self–brand distance in the CBR literature. In support of this contention, we report empirical evidence from a large-scale pilot study implying that the common self–brand constructs in the marketing literature (e.g., self–brand connection [[25]], self-connection [[ 1]]) load on a single factor interpretable as psychological distance (see Pilot Study A, Web Appendix W1).
While psychological distance can vary based on geographic, temporal, or probabilistic proximity (e.g., [63]), numerous aspects of social cognition have also been shown to alter perceptions of psychological distance. For example, the psychological distance of a target is smaller for an in-group member (psychologically close; e.g., sister) and larger for an out-group member (psychologically distant; e.g., waiter) ([53]). Further, similar others are perceived to be more psychologically close than dissimilar others ([54]), and objects are perceived to be psychologically closer when imagined from the first-person versus third-person perspective ([69]).
Findings in the literature are consistent with the idea that the social component of psychological distance may explain how consumers interact with brands. For example, examining moral identity in the context of out-group brands, [20] do not explicitly address the social dimension of psychological distance but suggest that "although psychological distance tends to be examined as the distance between two people rather than between a consumer and a brand, it is possible that the perception of distance from others applies to brands given the relationships and group associations with brands" (p. 100). Even so, there is nearly no explicit consideration of the social component of psychological distance in the branding literature.
Because CBRs represent socially construed dyads that are in many ways akin to an interpersonal relationship ([29]), the array of consumer–brand relationships identified in previous research should vary predictably along the social dimension of psychological distance, based on the relational norms and behaviors that constitute each relationship. Consider two examples. With "committed" brand relationships, consumers are faithful to the brand in some lasting way and think about these brands relatively similarly to their more intimate interpersonal connections ([61]). In this case, much like personal relationships (e.g., [53]), it is clear that the brand will be perceived as psychologically close and incorporated into the self-concept ([29]). Conversely, "secret affair" brand relationships, also characterized by high levels of affect, imply that brands are kept hidden to avoid a public association. Indeed, their nearest relational neighbor is the "complete stranger" type ([88]), underlining that secret affair brands lie more in the domain of "not me." In this case, despite positive feelings toward the brand, secret affair brands will be perceived as more psychologically distant because the brand is incorporated into the self-concept to a lesser extent and the consumer actively seeks distance from it ([10]). More generally, we use "close brand-relationship types" to refer to those associated with a low level of perceived psychological distance between the self and the brand, and "distant brand-relationship types" to refer to those associated with a high level of perceived psychological distance between the self and the brand. Pilot Study B provides empirical support for the level of psychological distance as a common foundation to a number of CBR types (Web Appendix W2).
If brand relationships indeed involve consumers' perceptions of psychological distance, it should benefit the brand to align this social distance with the construal level of brand information offered by marketers. Construal level theory ([78]) suggests that, on the one hand, the greater an object's psychological distance from a person, the greater the likelihood that it will be conceptualized at a higher level of abstraction. On the other hand, objects that are psychologically close are represented by more concrete, low-level construals. Abstract, high-level construals are "schematic, decontextualized representations that extract the gist from the available information," whereas concrete, low-level construals are "relatively unstructured, contextualized representations that include subordinate and incidental features of events" ([79], p. 83).
Matching the psychological distance of an object with an appropriate level of construal or concreteness of brand information results in mindset-congruency effects that have been shown to lead to information being perceived as more persuasive (e.g., [49]; [79]) and more likely to be accurately stored and retained in memory ([47]). Such mindset-congruency effects have been observed in marketing. For example, research on message framing and construal suggests that high-level, abstract language versus low-level, concrete language improves conservation behavior ([84]), increases effectiveness of charitable appeals and health messaging ([37]; [55]), and explains consumer evaluations of brand extensions ([60]).
We harness the concept of psychological distance that underlies the various brand relationship types to establish actionable strategies focused on leveraging the relationships that form organically between consumers and brands, rather than attempting to lead often-unwilling consumers into stronger relationships. Specifically, we expect that evaluations and behavior directed toward psychologically close brand relationships will be more favorable when consumers are presented with low-level, concrete brand information, whereas evaluations and behavior directed toward psychologically distant brand relationships will be more favorable when consumers are presented with high-level, abstract brand information.
The concept of processing fluency has been defined in a number of ways, generally referring to the ease with which a person is able to process information and assess meaning ([ 7]; [50]). Fluency has been shown to increase as a result of construal-based mindset congruencies (e.g., matching loss- [gain]-framed messages with concrete [abstract] mindsets), leading to more favorable behaviors ([83]; [84]). Research has further shown that heightened processing fluency resulting from a fit-based mindset congruency can lead to greater message persuasion ([49]) and positively influences a variety of judgments such as liking ([ 6]). Thus, further corroborating our psychological distance-based account, we expect that perceptions of fluency will increase when the degree of psychological distance implied by a particular brand relationship is matched with an appropriate construal level or concreteness of brand information.
We report four studies. Study 1 demonstrates our mindset-congruency effect by embedding a construal manipulation within brand information and documents processing fluency as a mediator. Study 2 embeds a manipulation of construal level in an advertisement and shows that the mindset-congruency effect can increase donations to a brand-supported cause. Study 3 elicits psychologically close and distant brands and examines both brand stereotypes (i.e., warmth and competence) and the search versus experience nature of the brand as moderators to shed light on how to more effectively manage existing CBRs. Finally, Study 4 employs a concreteness manipulation in a field study to demonstrate that the mindset-congruency effect is sufficiently strong to influence consumer spending and establishes an actionable segmentation moderator: category usage rate.
Study 1 examines the brand relationship mindset-congruency effect using a construal manipulation embedded within a brand communication. If different brand relationship types are associated with varying levels of closeness to the self, this should result in a mindset-congruency effect when processing that brand information at an appropriate construal level. To achieve this, we elicit two types of brand relationships—"committed" and "secret affair"—based on the results of Pilot Study B (see Web Appendix W2) that examined 12 brand relationships along the psychological distance dimension in addition to other dimensions currently used to conceptualize CBRs. As committed relationships are closer to the self ([29]; [61]), we expect to see improved brand evaluations following low-construal brand information processing. Conversely, because "secret affairs" are more distant from the self ([29]; [61]), brand evaluations should be more positive following exposure to high-construal brand information. Furthermore, by showing mediation by processing fluency, we provide further evidence to support our claim that we are documenting a construal–mindset congruency effect driven by psychological distance.
We recruited 266 undergraduate students (33% female; Mage = 18.2 years) from a large public university in exchange for partial course credit. Twelve were removed for failing an attention check or for not following instructions, and five were removed due to incomplete responses (N = 249). Participants were randomly assigned to conditions in a 2 (brand relationship type: committed, secret affair) × 2 (construal level: low, high) between-subjects design.
Participants were informed that they would be evaluating the quality of various brands using an online system, International Standard, which (ostensibly) compiles information from a variety of online sources (e.g., consumer reports, online product reviews) and produces brand-quality scores (see Web Appendix W3). After learning about the system, participants completed an established brand relationship elicitation task in which they were asked to nominate a brand that fit the given brand relationship type: committed or secret affair ([61]; see Web Appendix W1). In all conditions, participants were asked to reflect on brands that they use regularly in their daily life, so the only difference between the conditions was the nature of the brand relationship elicited. On the following screen, participants were asked to wait ten seconds while the system calculated the International Standard quality score for their nominated brand. The system then informed all participants that their brand had scored 9.2 (out of 10). To manipulate construal level of the brand information, the next screen provided a list of the "top five factors that contributed to the International Standard score that the brand received," which varied by condition based on a "how versus why" manipulation of construal level. Repeatedly focusing on how something is done elicits a low-level, concrete mindset, whereas focusing on why it is done elicits a high-level, abstract mindset ([33]). In the low-level condition, participants were shown five claims about the brand pertaining to how (concrete) the brand earned the score that it did (e.g., "by creating products that continually meet or exceed the expectations of its customers—according to consumer reports"). In the high-level condition, participants were shown five claims about the brand pertaining to why (abstract) it received the score that it did (e.g., "because it creates products that continually meet or exceed the expectations of its customers—according to consumer reports"). The content of the five claims did not differ between the two conditions—the only difference was the manipulation of construal level. We pretested the manipulation to assess the construal level of the brand information and attitudes toward the information (for pretest results for construal manipulations used in all studies, see Web Appendices W4 and W5).
Next, participants were asked to complete a three-item measure of processing fluency (α =.91; [49]; e.g., "How easy was this information to comprehend?," "How easy was this information to process?") on a 1–7 scale ("very difficult–very easy"). This was followed by evaluations of the brand using indices of attitudes (α =.95; nine items from [14]; e.g., "unfavorable–favorable," "dislike–like"), trust (α =.81; three items from [19]]; e.g., "I trust this brand," "I rely on this brand"), and satisfaction (α =.96; three items from [27]; e.g., "How satisfied are you with this brand?," "How content are you with this brand?"), all measured on 1–7 scales. Finally, participants completed basic demographics and manipulation checks for psychological distance using the Inclusion of Other in the Self (IOS) zipper scale ([20]; i.e., "Please indicate which case best describes the level of overlap between your own self-definition or identity and this brand") and self–brand connection (α =.92; seven items from [24]]; e.g., "This brand reflects who I am," "I can identify with this brand") scales. The latter measure was captured on a 1–7 scale ("not at all–extremely well"; for measures used in all studies, see Web Appendix W6).
Manipulation checks revealed that the brand closeness manipulation was successful. A brand relationship type × construal level condition analysis of variance (ANOVA) on psychological distance revealed a significant main effect of brand relationship type (F( 1, 243) = 16.73, p <.001). Participants in the committed brand relationship condition perceived the brand to be closer to the self (M = 4.80) than those in the secret affair condition (M = 3.96). As an additional check of psychological distance, we examined self–brand connection; as we expected, committed brand relationships were perceived to be closer (M = 4.27) than secret affair brand relationships (M = 2.97; F( 1, 245) = 55.86, p <.001).
A principal components analysis indicated that all three dependent variables (attitudes, trust, and satisfaction) were unidimensional (all loadings >.89), so we indexed them to form a brand evaluations composite (α =.88). We use a composite in all remaining analyses to economize reporting, though the choice is bolstered by research showing strong correlations and theoretical links across these variables (e.g., [36]). A 2 (brand relationship type: committed, secret affair) × 2 (construal level: high, low) ANOVA yielded a significant main effect of brand relationship on brand evaluations (F( 1, 245) = 48.79, p <.001; η2 =.17). In general, brands were evaluated more positively in the committed condition than in the secret affair condition. Importantly, the results show a significant interaction (F( 1, 245) = 11.81, p <.01; η2 =.05) that remained significant after controlling for both age and gender (F( 1, 243) = 11.73, p <.01; η2 =.05), which we include as covariates in all further studies.
Follow-up simple effects revealed that for committed relationships, brand evaluations were significantly more favorable in the low-level condition (M = 6.08) than the high-level condition (M = 5.66; F( 1, 243) = 5.08, p <.05; η2 =.02). On the other hand, for secret-affair relationships, brand evaluations were significantly more favorable in the high-level condition (M = 5.20) than in the low-level condition (M = 4.72; F( 1, 243) = 6.74, p <.05; η2 =.03).
We conducted a conditional process analysis with the PROCESS macro (Model 8, [39]) using a bootstrap procedure ( 5,000 draws) to construct bias-corrected confidence intervals. Results suggest that processing fluency mediates the focal relationship because the indirect effect of the highest-order interaction (brand relationship type × construal level) through fluency was significant (B =.25, SE =.10, 95% confidence interval [CI] = [.09,.48]). That is, the effect of construal level on brand evaluations through processing fluency is conditional on relationship type. For committed relationships, results show a significant indirect effect (B =.09, SE =.06, 95% CI = [.0004,.2224]), whereas for secret-affair relationships, the direction of the indirect effect reverses (B = −.16, SE =.07, 95% CI = [−.31, −.05]).
Study 1 provides evidence that consumers' relationships with brands can be effectively managed by attending to the associated degree of psychological distance. For consumers who are in psychologically close brand relationships (e.g., committed), claims made using low-level, concrete language result in increased processing fluency, leading to the brand being perceived more favorably than when high-level abstract claims are made. Conversely, for consumers who are in a more distant relationship with a brand (e.g., secret affair), high-level abstract claims result in increased processing fluency, leading to the brand being perceived more favorably than when low-level concrete claims are made. However, it should be noted that prior research associating construal level with positive affect (e.g., [48]) has suggested a potential alternative explanation for our congruency effect. To address this, we conducted a replication of our congruency effect to experimentally and statistically rule out the role of affect and to enhance robustness by using a stronger, direct manipulation of construal level (see Replication Study in Web Appendix W7). Finally, although the evidence of mediation by processing fluency supports our contention that our results are attributable to a construal–mindset congruency effect ([ 6]; [49]; [84]; 2019), in Study 2 we aim to provide further evidence that this congruency effect is driven by the brand's psychological distance and address selection concerns.
Study 1 examined the impact of the mindset-congruency effect on brand evaluations. It remains to be seen whether the effect is strong enough to affect consumer spending. To this end, Study 2 employs a more realistic application of construal level by embedding a construal-based manipulation in an advertisement for a charitable cause. We examine differences in consumer donations as a joint function of the construal level of brand communications and the psychological distance of the target brand. Finally, Study 2 uses a single target brand assigned to all respondents, an approach that avoids selection effects and more closely represents the type of decisions typically made by a marketer managing a single brand.
We recruited 156 student and nonstudent community volunteers (75% female; Mage = 22.2 years) through a large public university in exchange for $8.00. Participants were informed that they would be taking part in a study to assess their thoughts, feelings, and attitudes toward a brand. Following general demographic questions, participants rated their brand closeness using two measures ([20]; [24]; α =.94). We chose the brand Molson Canadian because a pretest (N = 48) suggested that it elicited considerable variance in self–brand distance in the study population. Next, participants viewed a fundraising advertisement titled "Lend a Hand to Man's Best Friend" (Web Appendix W8) that was cobranded by Molson Canadian and the local Society for the Prevention of Cruelty to Animals. We created a between-subjects construal-level manipulation by embedding "how" versus "why" language similar to that used in Study 1. In the low-level construal condition, the advertisement featured the question "How does your donation make a difference?" along with four statements answering the question (e.g., "by providing medical care"). In the high-level construal condition, the advertisement featured the question "Why is your donation important?" along with four statements (e.g., "because it ensures healthy animals"). Finally, after viewing the advertisement, participants were told that Molson Canadian was raising money for the Society for the Prevention of Cruelty to Animals and asked how much of their $8.00 payment they would like to donate to the cause, with their donation to be deducted from their payment at the end of the study (donation amount). Although all participants actually received payment of the $8.00 at the end of the study, they did not know this at the time they were asked to make a donation.
We regressed donation amount on construal level, brand closeness, the two-way interaction term, and the age and gender covariates. The expected construal level × brand closeness interaction was significant (B =.70, t(150) = 2.85, p <.01; f2 =.05). A floodlight analysis revealed that the effect of construal level was significant and negative for brand closeness scores below 1.77 (B = −1.00, t(150) = −1.98, p =.05) and significant and positive for brand closeness scores above 5.12 (B = 1.35, t(150) = 1.98, p =.05). That is, for more distant brands, donations were at least 67% higher when the advertisement featured high-level (vs. low-level) language, whereas for closer brands, donations were at least 88% higher when the ad featured low-level (vs. high-level) language.
Study 2 measured consumers' perceived closeness to a brand to demonstrate that the construal–mindset congruency effect can impact the amount of money donated to a charity affiliated with a target brand. These findings reveal that the mindset-congruency effect is strong enough to shift consumers' brand-endorsed donation behavior, providing behavioral support for our primary finding. Furthermore, it shows that embedding a construal-level manipulation in an advertisement is a practical, effective means of establishing a construal–mindset congruency effect. Importantly, the effect is replicated using a different brand-selection procedure (experimenter-provided vs. self-selected brand) that guards against idiosyncratic brand effects.
Study 3 was designed to address key brand-level moderators to provide greater insight into the practical application of our mindset congruency effect. Because our congruency effect is predicated on the construal level of brand information, we focus on two brand-level moderators that pose specific implications for how consumers process this brand information. First, we examine search versus experience brands to examine how our congruency effect is impacted by differences in the availability and diagnosticity of brand information. Second, we examine how strongly held brand stereotypes (e.g., warmth, competence) can inhibit the processing of new brand information.
Search attributes ([64]) are those "qualities of a brand that the consumer can determine by inspection prior to purchase" ([28], p. 434) and can be effectively discovered without the consumer interacting with the brand or product ([43]). In contrast, experience attributes refer to those product attributes that cannot be determined prior to inspection, as they typically require purchase to understand ([ 5]). Adapting these definitions to the brand level, we define search (experience) brands as those for which the brand attributes most important to consumers can be effectively evaluated using the information available before (only after) purchase—that is, the brand is primarily characterized by search (experience) attributes.
Recall that our theorizing suggests that consumers will prefer low-level concrete information for close brands and high-level abstract information for distant brands. However, in the context of search brands, we expect this pattern to be reversed. For search brands, the information typically sought by consumers is readily available prior to purchase ([64]) as consumers will typically have extensive knowledge of ([62]) and be less skeptical of claims made by such brands ([28]). Therefore, search brands are likely to be characterized by information saturation: close (distant) search brands provide all concrete (abstract) information consumers need in advance. Because most CBRs are based on consumers' regularly interacting with brands, additional construal-congruent information becomes highly redundant and is unlikely to gain attention or be processed extensively. Rather, with search brands, we expect that the novelty of being exposed to information that is incongruent with the associated construal mindset—that is, abstract information for close brands and concrete information for distant brands—will better capture consumer attention and influence their subsequent evaluations. This view is supported by research suggesting that information is novel when it breaks from preexisting schemas and can lead to heightened attention, arousal, and more favorable responses ([ 8]).
However, because the qualities of experience brands are difficult to evaluate in advance of purchase, consumers tend to expend more effort gathering information about them ([62]) and undertake more processing in relation to them ([43]), yet they still often end up in a state of greater ambiguity and uncertainty compared with search brands ([41]). Thus, as a result of this subjectivity, it is less likely that consumers will reach a point of information saturation when considering experience brands, meaning that information that is congruent with their construal-mindset will continue to be evaluated more favorably. Therefore, we expect to obtain our construal–mindset congruency effect for such brands.
Building on the Stereotype Content Model of interpersonal interaction ([26]), extant literature suggests that consumers typically maintain two fundamental perceptions or beliefs about brands ([46]). The first is brand warmth, which captures the extent to which a brand is perceived as having positive intentions. The second is brand competence, indicating whether the brand is perceived to have the ability to carry out these intentions. Both warmth and competence share characteristics with other constructs such as a brand's personality (i.e., sincerity and competence; [56]) and power (i.e., communion and agency; [86]). Warmth and competence are pivotal in the management of product, service, human, and destination brands ([17]; [59]; [65]; [80]) due to their importance in shaping consumer evaluations and behaviors ([ 3]; e.g., [ 2]).
Brands that are consistently positioned over time (e.g., Samuel Adams) may become stereotyped by virtue of being perceived as being very warm and/or competent ([34]; [46]), a situation that we expect will present a boundary condition for our mindset-congruency effect. Specifically, we anticipate that highly stereotyped brands will lead consumers not to attend to information provided to them in marketing communications but to rely on their existing brand beliefs. These strongly stereotyped beliefs are highly accessible, stable, enduring ([ 2]; [34]; [70]) and "resistant to change, regardless of the nature of the new information" ([44], p. 468). As a result, people with very strongly held stereotypes tend to expend less cognitive effort on stereotype-consistent information ([74]). In the current context, any mindset-congruency effect that results from a matching of self–brand distance with the concreteness of brand communications would be eliminated for those with strong beliefs about the warmth or competence of a brand, because this new information would not change existing stereotyped beliefs about the brand.
We recruited 201 participants (59% female; Mage = 38.0 years) from Amazon Mechanical Turk in exchange for a nominal fee. Of those participants, 7 were removed for failing an attention check (N = 194). Participants were randomly assigned to conditions in a 2 (brand closeness: close, distant) × 2 (construal level: low, high) between-subjects design. We used the same International Standard approach as Study 1 with three key differences. First, rather than using specific brand relationship types (i.e., committed vs. secret affair) to manipulate psychological distance, we took a more direct approach by eliciting psychologically close versus distant brands. Pilot Study B and extant research (e.g., [29]) show that brand relationships vary on dimensions other than psychological distance, such as their valence, hierarchy, and whether the products or services tend to be publicly consumed. Because some of these dimensions are uncorrelated with psychological distance (e.g., hierarchy, public/private; see Web Appendix W2), they are unlikely to confound our results. Still, for the remaining studies, we thought it prudent to take different approaches, which is why going forward we either explicitly manipulate or measure psychological distance. Participants were shown an image of [ 9] IOS scale with large overlap (close) or separate (distant) pairing circled (see Web Appendix W9) and asked to think of a brand they use in their daily lives that they felt best characterized this high level (low level) of self–brand overlap.
Second, we strengthened our "how versus why" construal-level manipulation by altering the concreteness of the information returned by the International Standard procedure (see Web Appendix W10). Prior research has shown that construal level can be manipulated by varying the level of concreteness or abstraction of written language (e.g., [78]; [84]) such that concrete language engages low-level construals and abstract language engages high-level construals. Third, we measured brand stereotypes of warmth (α =.95; e.g., "warm," "friendly") and competence (α =.93; e.g., "competent," "effective"; both four-item scales from [ 3]) on 1–7 scales ("not at all–very much"). We further captured the extent to which the brand is primarily a search versus experience good (α =.85; lower scores = experience good, higher scores = search good; [73]) and single-item measures[ 7] of public/private, symbolic/utilitarian, and political orientation (1 = very liberal, and 7 = "very conservative"). Two sample items for search versus experience are "I can get all the information about this brand before buying it" and "I can evaluate the quality of this brand before buying it." In all other respects, this study mirrored Study 1.
First, a manipulation check revealed that the brand closeness manipulation was successful. A brand closeness × construal-level condition ANOVA on self-brand connection revealed only a significant main effect of brand closeness (F( 1, 190) = 134.05, p <.001). Participants in the close condition perceived the brand to be closer to the self (M = 5.87) than those in the distant condition (M = 3.28). Next, we created a brand evaluations composite (α =.98) based on brand attitude (α =.97), trust (α =.89), and satisfaction (α =.97). Second, results from detailed analyses indicate that the focal brand closeness, search versus experience, and brand stereotypes variables represent independent constructs (see Web Appendix W12). Finally, each of the three moderators were examined in separate regression analyses in which we used PROCESS Model 3 ([39]) to regress brand evaluations on construal level (dummy coded), brand closeness (dummy coded), the continuous moderator, all two-way interactions, the three-way interaction term, and age and gender covariates.
Results indicated a significant three-way interaction (B = −1.41, t(184) = −5.71, p <.001; f2 =.11; see Figure 1). Floodlight analysis indicated two significant Johnson–Neyman inflection points. Specifically, the simple interaction effect of construal level and brand closeness was significant and positive for any search score below 5.31 (B =.64, t(184) = 1.97, p =.05) and significant and negative for any search score above 6.34 (B = −.83, t(184) = −1.97, p =.05). That is, while the congruency effect holds for brands containing some level of experience attributes, the effect reverses for brands characterized predominantly by search attributes. Next, we used spotlight analyses to probe the simple interaction of construal level × brand closeness at two levels: experience (−1 SD) and search (+1 SD). Supporting our mindset-congruency hypothesis, for experience brands the effect of concreteness was significant and positive for close brands (B = 1.41, t(184) = 4.58, p <.001) and significant and negative for distant brands (B = −1.02, t(184) = −3.20, p <.01). Thus, for experience brands, evaluations of close (distant) brands were higher when concrete (abstract) information was used. In contrast, for search brands the effect of concreteness was significant and negative for close brands (B = −.69, t(184) = −2.02, p <.05) and marginally significant and positive for distant brands (B =.61, t(184) = 1.85, p <.07). Thus, for search brands, evaluations of close (distant) brands were higher when abstract (concrete) information was used.
Graph: Figure 1. Brand relationship type × construal level effect for search versus experience brands (Study 3).*p <.07.**p <.05.***p <.01.****p <.001.Notes: Error bars = ±1 SE.
For competence, there was a significant three-way interaction (B = −.69, t(184) = −2.25, p <.05; f2 =.01). Floodlight analysis indicated that the simple interaction effect of construal level and brand closeness was significant and positive for any competence score below 6.22 (B =.58, t(184) = 1.97, p =.05). That is, the mindset-congruency effect was attenuated at high levels of competence. Next, a spotlight analysis showed that for low-competence brands (−1 SD), the effect of concreteness was significant and negative for distant brands (B = −.57, t(184) = −2.31, p <.05) and significant and positive for close brands (B =.82, t(184) = 2.16, p <.05). Thus, evaluations of close (distant) brands were higher when concrete (abstract) information was used. In contrast, no significant effects were found for high-competence brands (+1 SD).
For warmth, the results yielded a similar three-way interaction (B = −.73, t(183) = −4.00, p <.001; f2 =.04). Floodlight analysis indicated that the simple interaction effect of construal level and brand closeness was significant and positive for any warmth score below 5.23 (B =.59, t(183) = 1.97, p =.05). A spotlight analysis showed that for low-warmth brands (−1 SD), the effect of concreteness was significant and negative for distant brands (B = −.45, t(183) = −1.94, p =.05) and significant and positive for close brands (B = 1.60, t(183) = 4.70, p <.001). In contrast, no significant effects were found for high-warmth brands (+1 SD). Thus, the results indicate that our mindset-congruency effect is effectively attenuated at high levels of either the warmth or competent brand stereotype (for figures, see Web Appendix W13).
Study 3 provides further replication of our mindset-congruency effect, identifies boundary conditions to inform its practical application, and highlights the unique implications of search brands, for which our effect is reversed. Specifically, our findings suggest that marketers can enhance the success of strategies that match brand closeness with construal level by focusing on brands without strongly developed brand stereotypes. Furthermore, we argue that managing brands predominantly characterized by search attributes necessitates that marketers follow a reversed strategy by matching close (distant) search brands with abstract (concrete) language.
Study 2 demonstrated that the impact of the mindset-congruency effect was sufficient to positively influence consumer spending. However, this effect was observed indirectly in the context of a charity cobrand, using dollars donated to the charity partner as the dependent variable. In contrast, Study 4 was conducted to increase the robustness and ecological validity of our findings through a field study examining direct purchase behavior. Furthermore, while our previous studies primarily employed more subtle "how versus why" construal-based manipulations, Study 4 adopts a purely concreteness-based manipulation of construal level in order to improve the practical application of our effect. Finally, Study 3 examined how brand-based differences in the availability of brand information (i.e., search vs. experience brands) and the strength with which this brand knowledge is held in memory (i.e., brand stereotypes) moderate our mindset congruency effect. In Study 4, we build on our exploration of boundary conditions by examining a brand-level moderator, category usage rate, that affects consumers' motivation to process brand information.
We use category usage rate, capturing the amount of a product consumed by an individual in an average week, to tap the notion of category involvement and to reflect a useful segmentation variable ([21]). Extant research has found that low involvement levels are associated with a lack of active information seeking, little motivation to compare across product attributes, and limited personal relevance of what the product has to offer ([87]). We suggest that consumers who are less involved (i.e., low-volume, light users) are unmotivated and unlikely to be sufficiently attentive to the content of a brand communication effort for the mindset-congruency effect to emerge. In contrast, more regular and involved users should be more able and willing to attend to (and be impacted by) subtle differences in how brand information is presented. Thus, we expect that because light users are comparably less involved in the category, they will be less likely to attend to and process brand communications, mitigating any potential effect that the concreteness or abstractness of this message may have.
It may be useful to compare this prediction with Study 3, in which we suggested that search brands are characterized by information saturation, meaning that consumers should pay attention to and process construal-incongruent (vs. congruent) information. Here, we propose that when consumers are uninvolved with a product category, they will be less motivated to process any brand information. Rather than flipping the mindset-congruency effect as search brands did in Study 3, we anticipate the lack of attention associated with low-involvement consumers will eliminate the mindset-congruency effect.
One hundred fifty-eight student and community volunteers were recruited through a large public university. Eighteen were removed, as they had not heard of the target brand prior to the study (N = 140, 51% female; Mage = 22 years). In a central location on the university campus, we set up a trade table for a well-known, high-end tea brand, TWG. Prior to completing our study, participants had received $5.00 for taking part in an ostensibly unrelated study. Upon receiving payment, participants were informed that TWG was promoting a new line of teas for the upcoming season, and they were asked to stop by the trade booth as they went on their way. TWG was selected because a pretest (N = 32) showed that it elicited considerable variance on the focal brand closeness variable (M = 3.52, SD = 1.89) and that this variability was not correlated with how positively the brand was viewed by the study population (r = −.25, p =.19).
Upon approaching the TWG booth, participants were greeted by a confederate acting as a TWG employee who provided information about the company and its products. Participants were shown an advertisement (Web Appendix W14) that invited them to "accept our invitation to experience our new tea collection." They were randomly assigned to conditions that varied the concreteness of the messaging in the ad (concrete vs. abstract). In the concrete condition, the advertisement used more concrete language (e.g., "Allow the teabag to steep for four minutes—no more, no less. During this time the tea leaves open, hydrate, and infuse the cup with the essence and aroma of tea fruit and flowers"). In the abstract condition, the advertisement used more abstract language (e.g., "While the tea steeps, the ethereal essence envelops and soothes because of each tea's playful yet calming aromas").
After viewing the advertisement, participants were given an opportunity to purchase a sample pack of three of the featured teas using a "pay what you will" structure. Those who opted to purchase the product (any amount higher than $0) paid as much as they were willing for a TWG-branded package. Importantly, all brand payment decisions were binding such that the branded package was always provided in exchange for the indicated amount. All participants were then asked to complete a short feedback card that captured demographic information, category usage (i.e., "how many cups of tea do you drink per week"; continuous), and the IOS measure of psychological distance ([20]) as well as perceived similarity to the spokesperson (1 = "not at all similar," and 7 = "very similar") and perceived usefulness of information (1 = "not at all useful," and 7 = "very useful").[ 8] Finally, participants were asked if they would be interested in providing their email address to receive future communication for the TWG brand as an additional behavior-dependent variable (0 = no, 1 = yes).
We first wanted to ensure that the concreteness manipulation did not inadvertently affect perceptions of psychological distance. A one-way ANOVA verified that there was no significant effect of concreteness on self–brand distance (F( 1, 138) < 1).
To examine the mindset-congruency effect on consumer spending, we first regressed purchase price on concreteness (dummy coded), brand closeness, the two-way interaction term, and age and gender included as covariates. Results indicate significant main effects of concreteness (B = −1.91, t(134) = −2.74, p <.01), brand closeness (B = −.77, t(134) = −2.61, p <.05), and the two-way interaction (B =.56, t(134) = 3.10, p <.01; f2 =.07). A floodlight analysis revealed that the effect of concreteness was significant and negative for brand closeness scores below 2.04 (B = −.77, t(134) = −1.98, p =.05) and significant and positive for brand closeness scores above 4.64 (B =.68, t(134) = 1.98, p =.05). That is, for more distant brands, the amount paid was at least 35% higher when the ad featured abstract (vs. concrete) language, whereas for closer brands, the amount paid was at least 28% higher when the ad featured concrete (vs. abstract) language.[ 9]
First, PROCESS Model 3 ([39]) was used to regress purchase price on concreteness (dummy coded), brand closeness, category usage rate, all two-way interactions, the three-way interaction term, and age and gender covariates. Results reveal a significant three-way interaction (B =.10, t(130) = 2.02, p <.05; f2 =.03). Floodlight analysis indicates that the simple interaction effect of concreteness and brand closeness was significant and positive for any category usage rate above 3.28 (B =.39, t(130) = 1.98, p =.05). That is, the mindset-congruency effect was attenuated for individuals with an average category usage rate of 3.28 cups/week and below. Next, we used spotlight analyses to test the effect of concreteness across closeness scores at two levels of category usage: light user (−1 SD) and heavy user (+1 SD). In support of our mindset-congruency hypothesis, for heavy users the effect of concreteness was significant and negative for those who perceived the brand to be psychologically distant (B = −1.86, t(130) = −2.40, p <.05) and significant and positive for those who perceived the brand to be psychologically close (B = 1.47, t(130) = 2.77, p <.01). In contrast, no significant effects were found for light users (see Figure 2).
Graph: Figure 2. Brand relationship type × construal level effect for light versus heavy users (Study 4).*p <.05.**p <.01.Notes: Error bars = ±1 SE.
Study 4 was a field study designed to boost managerial relevance by showing behavioral outcomes in an ecologically valid setting. Our results suggest that our core mindset-congruency effect persists for consumer spending in a realistic and branded trade booth setting. For close (distant) consumers, the amount paid for the brand was greater, and consumers were more likely to engage with the brand when the message used concrete (abstract) language. The results shed light on a boundary condition to the observed mindset-congruency effect: category involvement. Consumers with lower category involvement (i.e., those that drink tea rarely) attend to category-related information less, regardless of their distance to the focal brand, mitigating any benefit of matching message concreteness to brand distance.
It is unlikely that consumers will ever care about as many brands as marketers would want, so it is imperative that marketers learn to thrive within the constraints of existing brand relationships, many of which are rather distant. Our findings suggest that marketing communications promoting such brands are more successful if they employ high-level, abstract language. Across all studies (see Table 2), we underscore that self–brand distance is a thread weaving through many major CBR measures, including brand attachment, brand love, self–brand connection, brand commitment, and brand identification. This in turn establishes the groundwork for specific consumer-based strategies to extract value from preexisting brand relationships using the theoretical lens of social psychological distance. We show that level of psychological distance associated with a brand relationship can trigger a favorable congruency effect when matched with the appropriate level of construal or concreteness of a marketing message, resulting in enhanced processing fluency, more favorable brand evaluations, higher donations, and more spending. We also identify several theoretical and practical moderators of our effect.
Graph
Table 2. Overview of Studies.
| Sample | Construal Variable | Psychological Distance Variable | DVs | Covariates | Moderators/Mediators |
|---|
| Study 1: Embedded construal | Undergraduate students (N = 249) | International Standard: Construal manipulation embedded within brand information | Brand relationship: committed (close) vs. secret affair (distant) | Brand evaluations: attitudes, trust, satisfaction | Age, gender | Mediator: processing fluency |
| Replication Study: Web Appendix W7 | Amazon's Mechanical Turk (N = 126) | "How vs. why": construal mindset manipulation | Brand relationship: committed (close) vs. secret affair (distant) | Brand evaluations: attitudes, trust, satisfaction | Affect, age, gender | |
| Study 2: Donation study | Undergraduate students and community volunteers (N = 156) | Charity cobrand: construal manipulation embedded within brand communications | Measured | Donation behavior | Age, gender | |
| Study 3: Eliciting close vs. distant brands | Amazon's Mechanical Turk (N = 201) | International Standard: concreteness manipulation embedded within brand information | Elicited brands via IOS manipulation | Brand evaluations: attitudes, trust, satisfaction | Age, gender | Moderators: search vs. experience, brand stereotypesa |
| Study 4: Field study | Undergraduate students and community volunteers (N = 140) | Trade show materials: concreteness manipulation embedded within brand communications | Measured | Purchase behavior, email engagement | Age, gender | Moderators: category involvementb |
1 a Study 3 also included functional/symbolic, private/public brands, and political orientation (moderators included on an exploratory basis).
2 b Study 4 also included usefulness/relevance of information (alternate mediator) and salesperson similarity (exploratory moderator).
We contribute to the understanding of how marketers can better manage the full spectrum of consumer–brand relationships. It is well understood that a high level of closeness between the consumer's self and a brand is a desirable marketing outcome and an effective input to brand loyalty ([25]). However, as the current research shows, it is not only the distance between the consumer and the brand that matters—it is the manner in which this distance interacts with how marketers speak to consumers about brands. Our mindset-congruency effect sheds light on a significantly overlooked aspect of brand relationships by demonstrating how managers can better realize value from relationships in which the brand is not close to the self.
Importantly, the flip side of this congruency effect demonstrates how marketers can better leverage close brands. We find that the use of concrete language within marketing communications results in more positive brand attitudes and increased trust, satisfaction, and spending. That is, while a high level of closeness between the consumer and the brand is beneficial, this outcome is made even more positive by tailoring the concreteness of brand language to match the psychological distance implied by that relationship.
Psychological distance should thus be given due consideration as a segmentation variable. For example, consider Walmart, which was listed by different respondents in several of our samples as a psychologically close or psychologically distant brand. People who exhibit a close relationship likely comprise a segment of working families who rely on Walmart's low prices to accommodate budgets. Conversely, those who relate to the brand along the lines of a distant relationship are likely younger and more brand conscious, relying on Walmart's prices but preferring to avoid being seen using Walmart-branded products. Our results imply that Walmart can profit from both groups by leveraging its accumulated customer relationship management databases pertaining to psychographics (e.g., spending, shopping habits) and demographics (e.g., age) to customize the concreteness of their marketing messaging.
In fact, identifying segments of consumers based on their relative self–brand distance should be fairly straightforward. For example, using what consumers write on social media, market research firms or in-house research teams could easily develop real-time monitoring tools based on dictionaries that reflect relative distance and then target consumer segments accordingly. Another approach could be based on surveying consumers directly. Firms already do this prolifically with the Net Promoter Score (NPS), which is essentially a future-looking word-of-mouth metric. Like the IOS scale used in our studies, NPS is a single-item metric, but the former has advantages. For example, unlike NPS, which lacks a "strong theoretical development" ([52], p. 81), IOS boasts a rich theoretical tradition, does not require transformation (in our samples it tended to be normally distributed), does not disregard the middle of the scale, and can be treated as a continuous measure. We were curious about whether NPS could be used in place of IOS and conducted a high-powered online experiment (cell sizes ∼130) that failed to find anything resembling the results we demonstrate in our studies. We ran an additional study (cell sizes ∼100) using trust instead of psychological distance and, again, the results were not promising.
Importantly, the approach implied by our research involves minimal investment. For example, in Studies 2 and 4 we found that simple changes in how information was presented in brand communications (Molson and TWG ads) caused distant consumers to donate and spend more than they otherwise would have and to even spend as much as close consumers. To illustrate, the Study 4 spotlight analyses showed that distant participants spent an average of $2.98 after viewing an ad with abstract language, whereas close participants spent an average of $2.79 across both information types. This result may be surprising when framed in light of marketers' enthusiasm for relationship building, but other scholars working in the construal-level domain have found similar tangible advantages emerging from simple changes to message framing (e.g., [84]). Compared with the resource-intensive process of solidifying relationships, we document a comparatively low threshold for making changes that should have positive and immediate financial impact. Of course, we are not suggesting that marketers abandon relationship-building efforts, but different tactics may be more beneficial with distant consumers. Future research could assess the comparative value of these two strategies.
Our studies also identify both brand-level and segmentation moderators with relevant implications for the application of our congruency effect. First, the recommendation implied by our mindset-congruency effect is contingent on the level of search versus experience attributes that characterize the brand. While our standard congruency effect holds for brands that possess even reasonably small levels of experience attributes (experience/search < 5.30), the effect is reversed for brands characterized by predominantly search attributes (experience/search > 6.34). For high-search brands—those that consumers can reliably evaluate the brand before purchase (e.g., clothing, jewelry, furniture)—managers should focus on aligning close (distant) search brands with abstract (concrete) brand communications. Second, when the brand possesses a well-developed brand stereotype (e.g., very high levels of warmth or competence), our mindset-congruency effect is mitigated. Such stereotypes are already prominent brand-management considerations ([17]; [59]; [65]), and extant research (e.g., [46]) suggests that only a few exemplary brands such as Coca-Cola and Campbell's ever reach high levels—according to our data, 6.22/7 on competence and 5.22/7 on warmth—where our mindset-congruency effect is unlikely to work. It also might be noted that a few select brands reach superior levels on both dimensions ([ 2]), but we would anticipate the same basic mitigation result: such brands are so resolutely positioned in this "golden quadrant" ([ 2], p. 191) that they would resist updating through the types of marketing communications examined in this article. We would advise brand managers that using a mindset-congruency strategy in such a situation would be wasteful.
In addition, consumers with lower category involvement are unlikely to demonstrate a mindset-congruency effect. For example, when consumers report a low category usage rate (i.e., drink tea rarely), they attend to category-related information less, regardless of their distance to the focal brand. Here, it is important to note that while consumers are less likely to form connections in low-involvement product categories (e.g., [71]), they still often do (e.g., [82]), meaning that psychological distance with a brand is not confounded with category involvement. For example, in our Study 4 we found only a small correlation between the measures of category usage and self–brand distance (r =.26, p <. 01), suggesting that consumers have separate sentiments about brands and categories.
We contribute to the marketing literature on consumer–brand relationships ([29]). By laying out how the social psychological distance associated with a brand relationship is a core dimension of numerous brand relationship types, we perform a useful phenomena-to-construct mapping ([57]), which enables us to develop a simplifying strategy. In contrast to research in which brand dimensions proliferate without an attendant level of clarity concerning how to put those dimensions to optimal use, we posit that psychological distance is inherent to brand relationships, is functionally synonymous with many concepts that scholars use to explore self–brand linkages, and is statistically unidimensional. Furthermore, in demonstrating mindset-congruency effects, our findings suggest that brand relationships parallel interpersonal relationships in terms of level of psychological distance associated with the relationship partner.
Second, we contribute back to construal level theory ([78]). Previous literature examining the social dimension of psychological distance has predominantly focused on the effects of in-groups versus out-groups (e.g., [53]; [54]), such that in-groups are perceived as close, whereas out-groups are perceived as distant. In contrast, we demonstrate that the norms that constitute a given relationship can offer subtle variations to these effects. That is, even positive, in-group relationships can be psychologically distant if the norms that govern the relationship imply distance (e.g., secret affair). Furthermore, we identify the search versus experience nature of the brand as a boundary condition of construal level theory that is unique to the marketing context. We show that the mindset-congruency effect is overridden and reversed for high-search brands for which, over the course of a brand relationship, consumers have reached a point of information saturation. Thus, for high-search brands, construal-incongruent information better captures consumer attention, leading to more favorable brand evaluations.
Third, our article provides the first empirical application of the social dimension of psychological distance to nonhuman targets. While the psychological distance of inanimate objects can be altered along temporal, spatial, and hypothetical dimensions (e.g., [78]), the social dimension has been applied only to human targets due to an underlying assumption that nonhuman entities are not truly social. However, through our examination of brand relationships, our insights suggest that the psychological distance of such objects can be influenced by social concepts such as ascribed relational norms. This in turn advances the possibility for psychological distance–based construal effects that would otherwise not be predicted by extant literature. These effects are made possible by consumers' tendency to anthropomorphize brands and see them akin to a relationship partner (e.g., [29]).
First, we propose a simplifying strategy that rests on a single idea: many of the somewhat disparate, even fragmented constructs appearing in the CBR literature share a latent feature tied to social psychological distance. Given the proliferation of brand constructs in the marketing literature, future research could productively adopt a similar approach. That is, researchers would benefit from taking a step back from or looking across the many measures and constructs in the CBR literature to identify those that have unique meaning versus those that have shared meaning, and to understand whether there are other latent lenses through which the field may continue to consolidate and clarify. Doing so would likely improve relationship marketing practice, make marketing spending more efficient, and reduce some of the redundancies that seem apparent in the CBR literature.
Second, we focus primarily on brand-specific consequences of our mindset-congruency effect (i.e., evaluations, spending) across our studies, but it is possible that the effect may similarly influence aspects of consumer judgment and decision making. For example, research could build on existing self-control ([81]) and gift-giving studies ([12]), which report that construal level is associated with a preference for feasibility or desirability attributes. Thus, the closeness of a consumer's relationship with a brand may represent a way to identify and cater to consumer attribute preferences during the decision-making process, such that as close (distant) relationships evoke a low-level (high-level) mindset they should lead to a greater emphasis on feasibility (desirability) attributes.
Third, our findings suggest a nuance in social-based construal-level effects in that they may depend on the real versus fictional nature of the relationship. Although extant research has shown that the priming of specific relationship norms can influence construal for fictional relationships (e.g., [ 4]), we show that when the relationship is lived and experienced, effects due to perceived psychological distance appear to supersede the priming effect. Thus, future research should examine the juxtaposition of these two competing effects in order to disentangle relationship norm predictions based on real versus fictional brands.
Fourth, we find evidence to suggest that our mindset-congruency effect is effectively mitigated for brands that are strongly positioned. In Study 3 we examine this boundary condition using two fundamental brand stereotypes (i.e., warmth and competence; [46]) for which the stereotype literature predicts that consumers will rely on existing beliefs rather than new information when forming object evaluations ([74]). To further ground this finding in the branding literature, it would be worthwhile to examine whether the same pattern of results extends to other brands that are superlatively positioned but on less fundamental, nonstereotype dimensions, such as brands that are viewed as very exciting or powerful ([56]; [86]).
Finally, although we examine several factors, the complexity of consumer interactions with brands necessitates that further research explore moderating variables or boundary conditions to our brand closeness-based congruency effect. For example, there may be situations in which a consumer's construal mindset does not align with the traditional predictions of construal level theory. Consider a frequent user of the Tide brand who feels that the brand is distant from their self-concept.[10] As a result of their use of the brand (to do laundry), they tend to think most often about more concrete aspects of using the brand (e.g., measuring out detergent, adding to a wash). Thus, it is possible that this consumer would tend to view this distant brand in a more concrete as opposed to abstract manner. While we address this empirically by ruling out any moderating role of functional versus symbolic and public versus private products, additional research should explore this potential occurrence for distant brands.
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242920984492 - They're Just Not That into You: How to Leverage Existing Consumer–Brand Relationships Through Social Psychological Distance
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242920984492 for They're Just Not That into You: How to Leverage Existing Consumer–Brand Relationships Through Social Psychological Distance by Scott Connors, Mansur Khamitov, Matthew Thomson and Andrew Perkins in Journal of Marketing
Footnotes 1 The first two authors contributed equally to this article.
2 C. Page Moreau
3 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
4 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research has benefited from financial support from the Social Science and Humanities Research Council of Canada.
5 Scott Connors https://orcid.org/0000-0002-9076-9221 Mansur Khamitov https://orcid.org/0000-0002-2444-0030
6 Online supplement: https://doi.org/10.1177/0022242920984492
7 Symbolic/functional (p =.40) and public/private (p =.12) were included on an exploratory basis and were found to not moderate the construal level × brand closeness interaction. We also explored whether political orientation ([45]) moderated our mindset-congruency effect. The results show a significant three-way interaction with the congruency interaction significant for more liberal respondents and not significant for more conservative respondents, but the key spotlight analyses were nonsignificant (for liberal participants [−1 SD], the effect of concreteness was negative for distant brands [p >.10] and positive for close brands [p >.13]). We further investigated this effect using a separate replication of Study 3 (see Web Appendix W11), where all the spotlight analyses are significant (p <.05).
8 We examined usefulness of information as an alternative mechanism and found that it did not mediate the effect of the construal level × brand closeness interaction on brand evaluations (β =.02, 95% CI = [−.03,.08]). Salesperson similarity was also included as an exploratory moderator. Results are reported in Web Appendix W15.
9 A binary logistic regression yielded a similar interaction for participants' likelihood to engage with the brand via email (B = −3.93, χ2(1) = 9.27, p <.01; see Web Appendix W16). Floodlight analyses indicate that the pattern of the interaction matches that of the dollars spent dependent variable.
We thank the Associate Editor for this suggestion.
References Aaker Jennifer L. , Fournier Susan , Adam Brasel S.. (2004), " When Good Brands Do Bad ," Journal of Consumer Research , 31 (1), 1 – 16.
Aaker Jennifer L. , Garbinsky Emily N. , Vohs Kathleen D.. (2012), " Cultivating Admiration in Brands: Warmth, Competence, and Landing in the 'Golden Quadrant' ," Journal of Consumer Psychology , 22 (2), 191 – 94.
Aaker Jennifer L. , Vohs Kathleen D. , Mogilner Cassie. (2010), " Nonprofits Are Seen as Warm and For-Profits as Competent: Firm Stereotypes Matter ," Journal of Consumer Research , 37 (2), 224 – 37.
Aggarwal Pankaj , Law Sharmistha. (2005), " Role of Relationship Norms in Processing Brand Information ," Journal of Consumer Research , 32 (3), 453 – 64.
Alba Joseph W. , Lynch John , Weitz Barton , Janiszewski Chris , Lutz Richard , Sawyer Alan , et al. (1997), " Interactive Home Shopping: Consumer, Retailer, and Manufacturer Incentives to Participate in Electronic Marketplaces ," Journal of Marketing , 61 (3), 38 – 53.
Allard Thomas , Griffin Dale. (2017), " Comparative Price and the Design of Effective Product Communications ," Journal of Marketing , 81 (5), 16 – 29.
Alter Adam L. , Oppenheimer Daniel M.. (2008), " Effects of Fluency on Psychological Distance and Mental Construal (or Why New York Is a Large City, but New York Is a Civilized Jungle) ," Psychological Science , 19 (2), 161 – 67.
Ang Swee Hoon , Lee Yih Hwai , Leong Siew Meng. (2007), " The Ad Creativity Cube: Conceptualization and Initial Validation ," Journal of the Academy of Marketing Science , 35 (2), 220 – 32.
Aron Arthur , Aron Elaine N. , Smollan Danny. (1992), " Inclusion of Other in the Self Scale and the Structure of Interpersonal Closeness ," Journal of Personality and Social Psychology , 63 (4), 596 –612.
Arsel Zeynep , Stewart Scott. (2015), " Identity Degrading Brands ," in Strong Brands, Strong Relationships , Fournier Susan , Breazeale Michael , Avery Jill , eds. New York : Routledge , 106 – 16.
Bagozzi Richard P. , Batra Rajeev , Ahuvia Aaron. (2017), " Brand Love: Development and Validation of a Practical Scale ," Marketing Letters , 28 (1), 1 – 14.
Baskin Ernest , Wakslak Cheryl J. , Trope Yaacov , Novemsky Nathan. (2014), " Why Feasibility Matters More to Gift Receivers than to Givers: A Construal-Level Approach to Gift Giving ," Journal of Consumer Research , 41 (1), 169 – 82.
Batra Rajeev , Ahuvia Aaron , Bagozzi Richard P.. (2012), " Brand Love ," Journal of Marketing , 76 (2), 1 – 16.
Batra Rajeev , Stayman Douglas M.. (1990), " The Role of Mood in Advertising Effectiveness ," Journal of Consumer Research , 17 (2), 203 – 14.
Beatty Sharon E. , Homer Pamela , Kahle Lynn R.. (1988), " The Involvement–Commitment Model: Theory and Implications ," Journal of Business Research , 16 (2), 149 – 67.
Beatty Sharon E. , Kahle Lynn R.. (1988), " Alternative Hierarchies of the Attitude-Behavior Relationship: The Impact of Brand Commitment and Habit ," Journal of the Academy of Marketing Science , 16 (2), 1 – 10.
Bennett Aronté Marie , Malone Chris , Cheatham Kenyn , Saligram Naina. (2019), " The Impact of Perceptions of Politician Brand Warmth and Competence on Voting Intentions ," Journal of Product and Brand Management , 28 (2), 256–73.
Carroll Barbara A. , Ahuvia Aaron C.. (2006), " Some Antecedents and Outcomes of Brand Love ," Marketing Letters , 17 (2), 79 – 89.
Chaudhuri Arjun , Holbrook Morris B.. (2001), " The Chain of Effects from Brand Trust and Brand Affect to Brand Performance: The Role of Brand Loyalty ," Journal of Marketing , 65 (2), 81 – 93.
Choi Woo Jin , Winterich Karen Page. (2013), " Can Brands Move in from the Outside? How Moral Identity Enhances Out-Group Brand Attitudes ," Journal of Marketing , 77 (2), 96 – 111.
Dillon William R. , Gupta Sunil. (1996), " A Segment-Level Model of Category Volume and Brand Choice ," Marketing Science , 15 (1), 38 – 59.
Einwiller Sabine A. , Fedorikhin Alexander , Johnson Allison R. , Kamins Michael A.. (2006), " Enough Is Enough! When Identification No Longer Prevents Negative Corporate Associations ," Journal of the Academy of Marketing Science , 34 (2), 185 – 94.
Eisingerich Andreas B. , Rubera Gaia. (2010), " Drivers of Brand Commitment: A Cross-National Investigation ," Journal of International Marketing , 18 (2), 64 – 79.
Escalas Jennifer Edson. (2004), " Narrative Processing: Building Consumer Connections to Brands ," Journal of Consumer Psychology , 14 (1–2), 168 – 80.
Escalas Jennifer Edson , Bettman James R.. (2003), " You Are What They Eat: The Influence of Reference Groups on Consumers' Connections to Brands ," Journal of Consumer Psychology , 13 (3), 339 – 48.
Fiske Susan T. , Cuddy Amy J.C. , Glick Peter , Xu Jun. (2002), " A Model of (Often Mixed) Stereotype Content: Competence and Warmth Respectively Follow from Perceived Status and Competition ," Journal of Personality and Social Psychology , 82 (6), 878 – 902.
Fletcher Garth J.O. , Simpson Jeffry A. , Thomas Geoff. (2000), " The Measurement of Perceived Relationship Quality Components: A Confirmatory Factor Analytic Approach ," Personality and Social Psychology Bulletin , 26 (3), 340 – 54.
Ford Gary T. , Smith Darlene B. , Swasy John L.. (1990), " Consumer Skepticism of Advertising Claims: Testing Hypotheses from Economics of Information ," Journal of Consumer Research , 16 (4), 433 – 41.
Fournier Susan. (1998), " Consumers and Their Brands: Developing Relationship Theory in Consumer Research ," Journal of Consumer Research , 24 (4), 343 – 73.
Fournier Susan , Alvarez Claudio. (2013), " Relating Badly to Brands ," Journal of Consumer Psychology , 23 (2), 253 – 64.
Fournier Susan , Dobscha Susan , Mick David Glen. (1998), " The Premature Death of Relationship Marketing ," Harvard Business Review , 76 (1), 42 – 51.
Freeman Karen , Spenner Patrick , Bird Anna. (2012), " Three Myths About What Customers Want ," Harvard Business Review (May 23), https://hbr.org/2012/05/three-myths-about-customer-eng?goback=.gde_1952271_member_119509501.
Freitas Antonio L. , Gollwitzer Peter , Trope Yaacov. (2004), " The Influence of Abstract and Concrete Mindsets on Anticipating and Guiding Others' Self-Regulatory Efforts ," Journal of Experimental Social Psychology , 40 (6), 739 – 52.
Freling Traci H. , Forbes Lukas P.. (2005), " An Empirical Analysis of the Brand Personality Effect ," Journal of Product and Brand Management , 14 (7), 404 – 13.
Fullerton Gordon. (2005), " The Impact of Brand Commitment on Loyalty to Retail Service Brands ," Canadian Journal of Administrative Sciences/Revue Canadienne des Sciences de l'Administration , 22 (2), 97 – 110.
Garbarino Ellen , Johnson Mark S.. (1999), " The Different Roles of Satisfaction, Trust, and Commitment in Customer Relationships ," Journal of Marketing , 63 (2), 70 – 87.
Han DaHee , Duhachek Adam , Agrawal Nidhi. (2016), " Coping and Construal Level Matching Drives Health Message Effectiveness via Response Efficacy or Self-Efficacy Enhancement ," Journal of Consumer Research , 43 (3), 429 – 47.
Havas (2020), " Meaningful Brands ," (accessed December 9, 2019), https://www.meaningful-brands.com/en.
Hayes Andrew F.. (2018), Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach , 2nd ed. New York : Guilford Press.
Heckler Donna S. , Till Brian D.. (2009), Creating Brands People Love. Upper Saddle River, NJ : Pearson Education.
Hoch Stephen J. , Deighton John. (1989), " Managing What Consumers Learn from Experience ," Journal of Marketing , 53 (2), 1 – 20.
Homburg Christian , Wieseke Jan , Hoyer Wayne D.. (2009), " Social Identity and the Service-Profit Chain ," Journal of Marketing , 73 (2), 38 – 54.
Huang Peng , Lurie Nicholas H. , Mitra Sabyasachi. (2009), " Searching for Experience on the Web: An Empirical Examination of Consumer Behavior for Search and Experience Goods ," Journal of Marketing , 73 (2), 55 – 69.
Johar Gita Venkataramani , Sengupta Jaideep , Aaker Jennifer L.. (2005), " Two Roads to Updating Brand Personality Impressions: Trait Versus Evaluative Inferencing ," Journal of Marketing Research , 42 (4), 458 – 69.
Jost John T. , Glaser Jack , Kruglanski Arie W. , Sulloway Frank J.. (2003), " Political Conservatism as Motivated Social Cognition ," Psychological Bulletin , 129 (3), 339 – 75.
Kervyn Nicolas , Fiske Susan T. , Malone Chris. (2012), " Brands as Intentional Agents Framework: How Perceived Intentions and Ability Can Map Brand Perception ," Journal of Consumer Psychology , 22 (2), 166 – 76.
Kisielius Jolita , Sternthal Brian. (1986), " Examining the Vividness Controversy: An Availability-Valence Interpretation ," Journal of Consumer Research , 12 (4), 418 – 31.
Labroo Aparna A. , Patrick Vanessa M.. (2009), " Providing a Moment of Respite: Why a Positive Mood Helps Seeing the Big Picture ," Journal of Consumer Research , 35 (5), 800 – 809.
Lee Angela Y. , Aaker Jennifer L.. (2004), " Bringing the Frame into Focus: The Influence of Regulatory Fit on Processing Fluency and Persuasion ," Journal of Personality and Social Psychology , 86 (2), 205 – 18.
Lee Angela Y. , Labroo Aparna A.. (2004), " The Effect of Conceptual and Perceptual Fluency on Brand Evaluation ," Journal of Marketing Research , 41 (2), 151 – 65.
Lee Kuan-Yin , Huang Hui-Ling , Hsu Yin-Chiech. (2007), " Trust, Satisfaction and Commitment: On Loyalty to International Retail Service Brands ," Asia Pacific Management Review , 12 (3), 161 – 69.
Lemon Katherine N. , Verhoef Peter C.. (2016), " Understanding Customer Experience Throughout the Customer Journey ," Journal of Marketing , 80 (6), 69 – 96.
Linville Patricia W. , Fischer Gregory W. , Yoon Carolyn. (1996), " Perceived Covariation among the Features of Ingroup and Outgroup Members: The Outgroup Covariation Effect ," Journal of Personality and Social Psychology , 70 (3), 421 – 36.
Liviatan Ido , Trope Yaacov , Liberman Nira. (2008), " Interpersonal Similarity as a Social Distance Dimension: Implications for Perception of Others' Actions ," Journal of Experimental Social Psychology , 44 (5), 1256 – 69.
MacDonnell Rhiannon , White Katherine. (2015), " How Construals of Money Versus Time Impact Consumer Charitable Giving ," Journal of Consumer Research , 42 (4), 551 – 63.
MacInnis Deborah J. , Folkes Valerie S.. (2017), " Humanizing Brands: When Brands Seem to be Like Me, Part of Me, and in a Relationship with Me ," Journal of Consumer Psychology , 27 (3), 355 – 74.
MacInnis Deborah J. , Morwitz Vicki G. , Botti Simona , Hoffman Donna L. , Kozinets Robert V. , Lehmann Donald R. , et al. (2020), " Creating Boundary-Breaking, Marketing-Relevant Consumer Research ," Journal of Marketing , 84 (2), 1 – 23.
Malar Lucia , Krohmer Harley , Hoyer Wayne D. , Nyffenegger Bettina. (2011), " Emotional Brand Attachment and Brand Personality: The Relative Importance of the Actual and the Ideal Self ," Journal of Marketing , 75 (4), 35 – 52.
Malone Chris , Fiske Susan T.. (2013), The Human Brand: How We Relate to People, Products, and Companies. San Francisco : John Wiley & Sons.
Meyvis Tom , Goldsmith Kelly , Dhar Ravi. (2012), " The Importance of the Context in Brand Extension: How Pictures and Comparisons Shift Consumers' Focus from Fit to Quality ," Journal of Marketing Research , 49 (2), 206 – 17.
Miller Felicia , Fournier Susan , Allen Chris. (2012), " Exploring Relationship Analogues in the Brand Space ," in Consumer-Brand Relationships: Theory and Practice , Fournier Susan , Breazeale Michael , Fetscherin Marc , eds. London : Routledge , 30 – 56.
Mitra Kaushik , Reiss Michelle C. , Capella Louis M.. (1999), " An Examination of Perceived Risk, Information Search and Behavioral Intentions in Search, Experience and Credence Services ," Journal of Services Marketing , 13 (3), 208 – 28.
Murdock Mitchel R. , Rajagopal Priyali. (2017), " The Sting of Social: How Emphasizing Social Consequences in Warning Messages Influences Perceptions of Risk ," Journal of Marketing , 81 (2), 83 – 98.
Nelson Phillip. (1970), " Information and Consumer Behavior ," Journal of Political Economy , 78 (2), 311 – 29.
Packard Grant , Moore Sarah G. , McFerran Brent. (2020), " Speaking to Customers in Uncertain Times ," MIT Sloan Management Review , 60 (2), 68 – 71.
Park C. Whan , Eisingerich Andreas B. , Park Jason Whan. (2013), " Attachment–Aversion (AA) Model of Customer–Brand Relationships ," Journal of Consumer Psychology , 23 (2), 229 – 48.
Park C. Whan , MacInnis Deborah J. , Priester Joseph R.. (2009), " Research Directions on Strong Brand Relationships ," in Handbook of Brand Relationships , MacInnis Deborah J. , Whan Park C. , Priester Joseph R. , eds. Armonk, NY : M.E. Sharpe , 379 – 91.
Park C. Whan , MacInnis Deborah J. , Priester Joseph , Eisingerich Andreas B. , Iacobucci Dawn. (2010), " Brand Attachment and Brand Attitude Strength: Conceptual and Empirical Differentiation of Two Critical Brand Equity Drivers ," Journal of Marketing , 74 (6), 1 – 17.
Pronin Emily , Ross Lee. (2006), " Temporal Differences in Trait Self-Ascription: When the Self Is Seen as an Other ," Journal of Personality and Social Psychology , 90 (2), 197 – 209.
Puzakova Marina , Kwak Hyokjin , Taylor Charles R.. (2013), " The Role of Geography of Self in 'Filling In' Brand Personality Traits: Consumer Inference of Unobservable Attributes ," Journal of Advertising , 42 (1), 16 – 29.
Reimann Martin , Aron Arthur. (2009), " Self-Expansion Motivation and Inclusion of Brands in Self ," in Handbook of Brand Relationships , MacInnis Deborah J. , Whan Park C. , Priester Joseph R. , eds. Armonk, NY : M.E. Sharpe , 65 – 81.
Reimann Martin , Castaño Raquel , Zaichkowsky Judith , Bechara Antoine. (2012), " How We Relate to Brands: Psychological and Neurophysiological Insights Into Consumer–Brand Relationships ," Journal of Consumer Psychology , 22 (1), 128 – 42.
Sharma Piyush , Sivakumaran Bharadhwaj , Marshall Roger. (2014), " Exploring Impulse Buying in Services: Toward an Integrative Framework ," Journal of the Academy of Marketing Science , 42 (2), 154 – 70.
Sherman Jeffrey W. , Stroessner Steven J. , Conrey Frederica R. , Azam Omar A.. (2005), " Prejudice and Stereotype Maintenance Processes: Attention, Attribution, and Individuation ," Journal of Personality and Social Psychology , 89 (4), 607 – 22.
Stokburger-Sauer Nicola , Ratneshwar Srinivasan , Sen Sankar. (2012), " Drivers of Consumer–Brand Identification ," International Journal of Research in Marketing , 29 (4), 406 – 18.
Swaminathan Vanitha , Page Karen L. , Gurhan-Canli Zeynep. (2007), " 'My' Brand or 'Our' Brand: The Effects of Brand Relationship Dimensions and Self-Construal on Brand Evaluations ," Journal of Consumer Research , 34 (2), 248 – 59.
Thomson Matthew , MacInnis Deborah J. , Park C. Whan. (2005), " The Ties That Bind: Measuring the Strength of Consumers' Emotional Attachments to Brands ," Journal of Consumer Psychology , 15 (1), 77 – 91.
Trope Yaacov , Liberman Nira. (2010), " Construal-Level Theory of Psychological Distance ," Psychological Review , 117 (2), 440 – 63.
Trope Yaacov , Liberman Nira , Wakslak Cheryl. (2007), " Construal Levels and Psychological Distance: Effects on Representation, Prediction, Evaluation, and Behavior ," Journal of Consumer Psychology , 17 (2), 83 – 95.
Volos Vadim. (2020), " London Ranks as the Top 'City Brand.' Sydney Emerges in Second Place, While Paris Declines from First in 2017 to Third in 2020 ," Ipsos (April 6), https://www.ipsos.com/en/2020-anholt-ipsos-city-brand-index.
Wan Echo Wen , Agrawal Nidhi. (2011), " Carryover Effects of Self-Control on Decision Making: A Construal-Level Perspective ," Journal of Consumer Research , 38 (1), 199 – 214.
Warrington Patti , Shim Soyeon. (2000), " An Empirical Investigation of the Relationship between Product Involvement and Brand Commitment ," Psychology and Marketing , 17 (9), 761 – 82.
White Katherine , Habib Rishad , Hardisty David J.. (2019), " How to SHIFT Consumer Behaviors to Be More Sustainable: A Literature Review and Guiding Framework ," Journal of Marketing , 83 (3), 22 – 49.
White Katherine , MacDonnell Rhiannon , Dahl Darren W.. (2011), " It's the Mind-Set That Matters: The Role of Construal Level and Message Framing in Influencing Consumer Efficacy and Conservation Behaviors ," Journal of Marketing Research , 48 (3), 472 – 85.
Wittenbraker John , Zeitoun Helen , Fournier Susan. (2015), " Using Relationship Metaphors to Understand and Track Brands ," in Strong Brands, Strong Relationships , Fournier Susan , Breazeale Michael , Avery Jill , eds. London : Routledge , 360 – 75.
Yang Linyun W. , Aggarwal Pankaj. (2019), " No Small Matter: How Company Size Affects Consumer Expectations and Evaluations ," Journal of Consumer Research , 45 (6), 1369 – 84.
Zaichkowsky Judith Lynne. (1985), " Measuring the Involvement Construct ," Journal of Consumer Research , 12 (3), 341 – 52.
Zayer Linda Tuncay , Neier Stacy. (2011), " An Exploration of Men's Brand Relationships ," Qualitative Market Research: An International Journal , 14 (1), 83 – 104.
~~~~~~~~
By Scott Connors; Mansur Khamitov; Matthew Thomson and Andrew Perkins
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 130- Traveling with Companions: The Social Customer Journey. By: Hamilton, Ryan; Ferraro, Rosellina; Haws, Kelly L.; Mukhopadhyay, Anirban. Journal of Marketing. Jan2021, Vol. 85 Issue 1, p68-92. 25p. 1 Black and White Photograph, 6 Charts, 1 Graph. DOI: 10.1177/0022242920908227.
- Database:
- Business Source Complete
Traveling with Companions: The Social Customer Journey
When customers journey from a need to a purchase decision and beyond, they rarely do so alone. This article introduces the social customer journey, which extends prior perspectives on the path to purchase by explicitly integrating the important role that social others play throughout the journey. The authors highlight the importance of "traveling companions," who interact with the decision maker through one or more phases of the journey, and they argue that the social distance between the companion(s) and the decision maker is an important factor in how social influence affects that journey. They also consider customer journeys made by decision-making units consisting of multiple individuals and increasingly including artificial intelligence agents that can serve as surrogates for social others. The social customer journey concept integrates prior findings on social influences and customer journeys and highlights opportunities for new research within and across the various stages. Finally, the authors discuss several actionable marketing implications relevant to organizations' engagement in the social customer journey, including managing influencers, shaping social interactions, and deploying technologies.
Keywords: customer experience; customer journey; joint consumption; social distance; social influence
Many consumer decisions do not occur in isolation but rather within an interactive context of social relationships and societal concerns, often facilitated by technology. In a paradox of modern society, then, the more that technology allows us to connect with others, the more fragmented society becomes, with increasing alienation and disconnection ([76]). Modern marketing came of age in postwar America, in a familiar milieu featuring nuclear families, physical retail locations, mass-printed catalogs, and broad consumer segments. Consumers decided what to buy on the basis of one-way communications from advertisers and two-way interactions with friends and neighbors. Capturing the zeitgeist, [46] based his theory of social influence on the assumption of people living "in a community marked by geographic immobility and lifelong friendships" (p. 1375), in which social ties are created by distinct and often predictable social contexts.
Today, much has changed. Nuclear families have given way to single-parent and blended family households, physical stores and printed catalogs have been replaced by online retailers that use algorithm-powered interactive tools, and personalization has supplanted segment-based targeting. Digital technology has vastly expanded the contexts within which people socialize, allowing them to form and maintain relationships beyond limits circumscribed by geography. Artificial intelligence (AI) agents, such as Amazon's Alexa and Apple's Siri, play increasingly relational roles in consumers' daily lives, complementing and even substituting for other social interactions ([94]). Thus, it seems inevitable that consumer decision making should evolve along with these technological and social changes. Consequently, it is vital for marketers and researchers alike to acknowledge, investigate, and delineate these fundamental shifts. This article takes on this challenge and examines the changing and pervasive role of social influence throughout the consumer decision-making process.
We approach the challenge of understanding social influence on consumer decision making from the perspective of customer journeys, which break decisions into a series of steps that constitute a path to purchase and beyond. These "journeys" (or, when a drop-off is expected in the number of people across successive steps, "funnels") were recognized at least as far back as the late 1800s, when marketing experts decomposed the effectiveness of advertising into a series of staged effects. The most influential of these early stepwise models evolved into the AIDA framework—Awareness, Interest, Desire, and Action—and is still popular in both academic settings and marketing practice ([116]).
Over time, depictions of the customer journey, commonly referred to as customer journey maps, have become more complex and specialized, with additional stages and extensions added both before and after the purchase. Recent contributions include conceptualizing a nonlinear customer journey ([26]), emphasizing the various stakeholders who "own" different touchpoints along the journey ([74]), and identifying a set of journey archetypes that acknowledge the diverse cognitive and behavioral states that motivate purchases ([73]). Others have argued that more than constituting simply a path to purchase, customer journeys are depictions of the entire customer experience ([101]) or even paths to achieving life goals ([49]). In depicting the customer journey, these maps delineate the factors that may influence consumers along their decision-making process. Table 1 contains a cross-section of customer journey frameworks proposed by experts from the worlds of academic research, marketing practice, and marketing education. Although by no means exhaustive, the compiled list of journey frameworks emphasizes the different approaches adopted in the literature.
Graph
Table 1. Customer Journey Models.
| Customer Journey Stages |
|---|
| Source and Type | −5 | −4 | −3 | −2 | −1 | 0 | 1 | 2 | 3 |
|---|
| Belch and Belch (2015) (T) | | | Problem recognition | Information search | Alternative evaluation | Purchase decision | Postpurchase evaluation | | |
| Chernev (2014)b (T) | | Awareness | Understanding | Attractiveness | Purchase intent | Purchase | Satisfaction | Repurchase intent | |
| Court et al. (2017), Traditional (P) | | | Awareness | Familiarity | Consideration | Purchase | Loyalty | | |
| Court et al. (2017), Updateda (P) | | | | Initial consideration set | Active evaluation | Moment of purchase | Postpurchase experience | Loyalty loop | |
| Donavan, Minor, and Mowen (2016)a (T) | Involvement | Memory | Exposure | Attention | Comprehension | | | | |
| Edelman and Singer (2015)a (P) | | | | Consider | Evaluate | Buy | Enjoy | Advocate | Bond |
| Edwards (2011) (A) | | | | Awareness | Consideration | Buy | Use | Form opinion | Talk |
| Forrester (2010)a (P) | | | | Discover | Explore | Buy | Engage | | |
| Howard and Sheth (1969) (O) | | Attention | Brand comprehension | Attitude | Intention | Purchase | | | |
| Kotler and Armstrong (2012) (T) | | | Need recognition | Information search | Evaluation of alternatives | Purchase decision | Postpurchase behavior | | |
| Lemon and Verhoef (2016)a (A) | | | | | Prepurchase stage | Purchase stage | Postpurchase stage | | |
| Lewis (1908) (O) | | Awareness | Opinion | Consideration | Preference | Purchase | | | |
| Mothersbaugh and Hawkins (2015) (T) | | | Problem recognition | Information search | Alternative evaluation | Purchase | Postpurchase | | |
| Puccinelli et al. (2009) (A) | | | Need recognition | Information search | Evaluation | Purchase | Postpurchase | | |
| Richardson (2010) (P) | | | | | Engage | Buy | Use | Share | Complete |
| Solomon (2015) (T) | | | Problem recognition | Information search | Evaluation of alternatives | Product choice | Outcomes | | |
| Strong (1925) (T) | | | Attention | Interest | Desire | Action | | | |
| Wiesel, Pauwels, and Arts (2011) (A) | | | | Information | Evaluation | Purchase | | | |
1 a Nonlinear journey model.
- 2 b Combination of two customer journey frameworks.
- 3 Notes: The customer journey frameworks have been aligned by the purchase stage (0), with the stages leading up to purchase identified with negative numbers and those occurring after purchase having positive numbers. Publication types are as follows: A = peer-reviewed academic research, P = practitioner publication (e.g., Harvard Business Review), T = textbook, O = other.
One common feature of these journey frameworks is that they are individual customer journeys, focusing on isolated consumers as the decision-making unit (DMU). When prior research has accommodated social influences in the customer journey, it has done so in a general way, either by emphasizing the importance of social factors without articulating those influences within the specific stages of the journey, or by viewing them as part of the set of broader environmental conditions throughout the journey. For example, [122] consider the "social environment" as one of seven primary factors influencing customer experience. [101] incorporate social influences by acknowledging social cues as part of the atmospherics of a retail experience along with other elements such as design and ambiance. [74] identify "social/external" as one category of touchpoints in the prepurchase, purchase, and postpurchase stages of their model. In all these cases, the authors endorse the importance of social influence but treat it as an independent factor or as one of several broad influences that can affect the journey.
A more recent depiction, the "needs-adaptive" shopper journey model ([73]), de-emphasizes the multistage journey model in favor of a more fluid set of "states" consumers may experience. By using differences in the movement among these states, the authors identify journey archetypes of common purchase situations. Consistent with previous work, this model acknowledges "peer-to-peer/social" influences as one of the "groups of factors that influence a consumer's shopping process" (p. 279). However, it also allows for a more specific approach for incorporating social influences by identifying several archetypes that are inherently social: the "joint journey," the "social network journey," and the "outsourced journey." [73] also acknowledge that social influences can play a role even in archetypes not explicitly centered on social influence (e.g., "retail therapy journey" and "learning journey" both implicitly incorporate social influences). This customer-journey-as-archetype model provides a novel way of characterizing customer experiences and produces new insights. While these archetypal journeys clearly recognize the importance of social influences on the customer journey, the approach is deliberately divorced from previous work defining a customer journey as a set of defined stages. In this article, we suggest that returning to the classic customer journey model to investigate social influences can help generate complementary insights for researchers and practitioners.
With the exceptions noted previously, we believe that the core premise of existing frameworks is fundamentally decontextualized from social influences. This lack of focus on social influence in the customer journey stands in stark contrast to research documenting the many ways in which social contexts influence customer decisions. For example, consumers trade off their own preferences for those of the group ([ 9]) and consider how their purchases will be perceived by others ([13]). The influence of social others is also relevant in business-to-business (B2B) contexts ([110]), including in sales interactions ([ 1]) and in relationship marketing ([89]). Thus, a somewhat surprising gap exists in the literature, as researchers have independently studied individual customer journeys and various aspects of decision making in social contexts but have left customer journeys in a social context largely unexamined.
In this article, we present the "social customer journey" and introduce the notion of "traveling companions" (or social others on the journey), who interact, directly or indirectly, with the decision maker through one or more phases of the journey. The introduction of these companions allows us to highlight not merely how they might influence the decision maker, but also how the companions' own journeys may be influenced in turn. The goal of this article is not to create a more "accurate" journey map by accounting for a wider set of customer decisions, nor to provide a complete review of research on social influences in decision making. Instead, we highlight key insights about social influences throughout the customer journey and identify promising areas for new research and marketing insight by reflecting on a customer journey made increasingly complex by the ways in which it is affected by others. We also emphasize the interactive nature of customer journeys. Just as individuals are influenced by social others, those others, as individuals, groups, and even society as a whole, are influenced by the journeys of the individuals around them.
While consumers have long sought input from others in their decision making, particularly from those who are socially close, new technologies and societal changes have significantly altered the manner in which and extent to which purchases are influenced in some way by others. Moreover, the advent of online platforms and social media has changed the very definition of social closeness. The opinions of anonymous others and the aggregated ratings of groups of others are readily available in ways previously unimaginable. Consequently, even people who are seemingly socially distant may exert a powerful influence on one's decisions. For example, consumers easily and eagerly ask their Facebook "friends," whom they may never have met in person, for input into what restaurant to dine at while on vacation, and they actively seek out recommendations from quasi-celebrity "mommy bloggers" about what diaper bag to buy. After a purchase, consumers frequently share information about product performance, even for mundane purchases, on retailer websites and social media, potentially influencing the decision making of others with their ratings and reviews. The rise in the use of email lists and online industry forums, and greater participation in professional social networks, facilitated by technologies like LinkedIn, mean that social influences are increasingly as likely to affect business decision making as consumer decision making.
We employ a linear six-step journey (depicted in Figure 1) that builds on the greatest points of similarity from previous models: motivation, information search, evaluation, decision, satisfaction, and postdecision sharing. We acknowledge that the linear depiction does not fully capture the dynamics of decision making, as consumers may iterate between stages or drop out at any stage to restart the journey later. Moreover, nonlinear journey frameworks are useful in emphasizing the ongoing relationships consumers have with brands and retailers rooted in repeat purchases ([26]; [74]). We represent these nonlinear paths with the circular loops in the figure. However, the linear model we use to frame this discussion provides a parsimonious and generalizable foundation for our analysis.
Graph: Figure 1. The social customer journey.Notes: Social others (i.e., traveling companions), individually or in aggregate, can influence an individual customer's decision journey at the various stages while also themselves being influenced by that customer. The social others are depicted by the individuals and groups below the solid line, with the bidirectional arrows representing the bidirectional influence between the focal customer and social others. The gray figure on the journey path represents joint journeys (plural DMUs). The circular arrows qualify the otherwise linear journey depiction, acknowledging that social customer journeys may often deviate from the simple path.
Our journey model extends the customer journey framework in two critical ways. First, it introduces the notion that social others (i.e., traveling companions), individually or in aggregate, can influence an individual customer's decision journey at various stages, while also themselves being influenced by that customer. This is depicted in the figure by the individuals and groups below the solid line. The bidirectional arrows emphasize that social influences flow in both directions as these social others are on their own journeys. Second, it recognizes that some customer journeys occur for a DMU of more than one individual, depicted by the black and gray figures moving together in a joint journey.
Social others can play many different roles in a social customer journey (e.g., a face-to-face interaction with a close friend who raves about a new restaurant vs. the likes and glowing reviews of anonymous customers on social media and review sites). To help understand and characterize these various roles and influences, we propose a continuum of social others based on the closeness of others to the DMU. Social distance, as we are conceptualizing it here, can be affected by many social relationship dimensions, but we highlight five that are especially relevant to marketing researchers: number of social others, extent to which the other is known, temporal and physical presence, group membership, and strength of ties (see Figure 2). We suggest that these dimensions converge to form a global sense of social distance, but that not all dimensions need to be on the extreme ends of the continuum for the social other to be interpreted as overall more proximal or distal. Rather, we suggest that a preponderance of the factors will determine how close the social other is perceived to be.
Graph: Figure 2. Social distance considerations for the social customer journey.Note: Social distance is defined on a continuum by a preponderance of factors that make a social other proximal or distal.
"Proximal social others" are typically specific, individuated others that provide distinct, discrete, articulated inputs to the focal customer's journey. They tend to be close—in terms of temporal and physical proximity—members of the customer's in-group and have strong ties to the focal consumer. For example, a consumer's evaluation of a potential vacation destination may be influenced by inputs from a proximal social other, such as a single, close friend representing one well-known, physically present in-group member with strong social ties.
"Distal social others" can be larger groups or the whole of society, whose members may not be individuated, present, temporally proximal, or even known to the consumer. When a distal other is a single individual, this individual will tend to be someone the consumer does not know personally, such as a YouTube tutorialist or an anonymous review writer. The same vacation-planning consumer may also be influenced by distal social others, including the reviews of hundreds of others on a travel website representing many, relatively unknown, not physically present social others with only weak social ties and unlikely membership in a readily identifiable in-group.
This social distance continuum and the dimensions highlighted in Figure 2 are consistent with existing theories of social influence. [15] model the self-concept with three levels of representation: personal, relational, and collective. The personal self reflects the self as an individual differentiated from others; the relational self reflects one's self view vis-à-vis one's close relationships, often in dyadic or small-group contexts; and the collective self suggests an even broader social perspective, viewed through the lens of group membership. Our view of social others parallels Brewer and Gardner's view of the self, in which others exist on a continuum that moves from very proximal (personal) through increasing separation (relational) to very distal others (collective). Our framework also shares elements with construal level theory ([120]), which suggests that psychologically proximal versus distal objects, events, or people are viewed in more concrete versus abstract terms, respectively. Other people may be perceived as more proximal to or distal from the self; for example, in-groups are viewed as more proximal than out-groups, and those with close ties are viewed as more proximal than those with weak ties ([43]). Finally, our framework incorporates elements from social impact theory ([72]), which suggests that the degree of impact from the social environment depends on the size (i.e., number of people), immediacy (i.e., physical or temporal proximity), and strength (i.e., importance to the individual) of the group. The size and immediacy dimensions are incorporated directly as dimensions of social distance, with small (large) size and greater (lesser) immediacy being consistent with the proximal (distal) end of the continuum. We draw on these models, as they all highlight how the perceived social distance between the self and others affects decision making.
Importantly, this social distance distinction leads to meaningful differences in how social influence affects a customer's journey. One straightforward proposition is that more proximal social others will have a stronger influence on customer journeys than more distal social others. Prior research supports this perspective, as social others with whom one shares strong ties have been shown to be influential ([16]). While we acknowledge that this may generally be the case, we also seek to highlight some sources of possible exceptions to this rule: ( 1) the importance of more distal social influences in certain contexts; ( 2) changes in perceived social distance, along one or more of the dimensions, often brought about by technological changes; and ( 3) the nature of, rather than simply the magnitude of, the roles that proximal and distal social others may play.
Some research suggests that under some circumstances, distal influences can be more powerful than proximal ones. For example, [33] found that consumers with high subjective knowledge, or those who encounter instrumental cues, tend to seek recommendations from more distal social sources. Adding a further nuance, [64] found that preference for socially proximal sources was moderated by temporal distance, while [129] showed that others' recommendations are most persuasive when their social distance aligns with the temporal distance of the decision itself. More generally, powerful normative influences often reflect the preferences of diffuse, unknown others.
Changes in the actual or perceived nature of one or several of the social distance dimensions can move the social other from the more distal toward the more proximal end of the continuum, and vice versa. Relatedly, the perceived distance from social others is malleable. For example, reviewers may be perceived as similar to the consumer ([93]), which would narrow the perceived social distance between the consumer and a typically distal social other, potentially increasing their influence ([42]). Similarly, influence as a function of changes in perceived social distance is an idea that must increasingly be applied to nonhuman social companions, such as Amazon's Alexa and Apple's Siri ([94]). As technology evolves, these entities will increasingly take on the person-roles of information gatekeepers, experts, and possibly even decision makers (a type of outsourced journey, per [73]). As they become more familiar and proximal—in our homes, in our cars, and on our persons—these AI agents may be seen as more proximal social others by some consumers.
Finally, we suggest that the precise roles of social others may change along with differences in perceived social distance. A friend may provide valuable information about a product, but the friend's social closeness to the consumer may exert influence in other ways, such as by serving as a basis for social comparison or because the consumer desires to maintain a good relationship with that friend. Reviewers are more distal in social connection, and their impact is likely to be more informational. Yet as some or all of the social distance factors begin moving toward the more proximal or the more distal end of the continuum, corresponding changes in the respective roles may also change.
While technology and other societal forces have considerably broadened where and how traveling companions can influence the customer journey, perhaps the most fundamentally social journey is one wherein two or more consumers journey together. With respect to our social distance continuum, when a certain threshold is surpassed, social others may become incorporated into the DMU itself, creating a joint journey characterized by interdependence in most or all stages of the customer journey. This results in a pluralized DMU (see Figure 1, black and gray figures), where two or more people travel on a "joint decision, joint consumption" journey together ([45]). Decision making in such situations is qualitatively different because the members of the DMU have interdependent utilities ([51]), and the individual members of the DMU may, at each stage of the journey, base their own responses on the responses of the other. Consequently, joint journeys are complex and distinct from individual journeys because of the relationship dynamics that must be managed ([113]).
Joint journeys were a key focus in early consumer research, with an emphasis on the family as the DMU ([29]; [111]). For example, [18] investigated how husband–wife dyads arrived at decisions involving the purchase of a family car. They found that the members of this dyad can vary in expertise, experience, and preferences, and their levels of involvement and empathy dictate their interactions across the stages of the decision process to arrive at a joint decision that is mutually acceptable. Since this early research, joint journeys have received sporadic interest (e.g., [24]); however, they have recently begun to reemerge as an important area of study both in dyads (e.g., [34]) and in family units (e.g., [37]; [118]). In a B2B context, decision making can be viewed through the lens of the joint customer journey as there are often several individuals playing a role in the decision-making process ([110]). We present the joint journey as a special case of the social customer journey and highlight the specific considerations these journeys entail.
We next discuss the effect of traveling companions at each stage of the social customer journey. As mentioned, a large literature has examined the effects of social influence on consumer decision making (for reviews, see [65] and [67]), but most of that research has not been framed within the customer journey. In what follows, we provide selected research insights that are relevant to the specific stage of the social customer journey, placing emphasis on insights highlighting proximal versus distal social influences. We also discuss research relevant to the special case of the joint journey. For each stage, we offer ideas for how viewing the customer journey through a social lens can generate new research, and we provide more detailed research questions in Table 2, including those related to both bidirectional and cross-stage effects.
Graph
Table 2. Emerging Research Questions from the Social Customer Journey.
| Motivation |
| Proximal social drivers | How does the influence of others, distal or proximal, change as motivations change over time? How are motivational conflicts stemming from inputs from different social others resolved? How does the impact of social comparison on one's purchase motivations change according to social distance?
|
| Distal social drivers | How do consumers navigate conflicts between existing brand preferences and alignment with their broader social networks' values (e.g., political ideologies or social causes)? How do consumers manage situations in which proximal motivational inputs are in opposition to distal societal norms?
|
| Joint journeys | How are conflicting motivations of joint journey members reconciled? How does being the catalyst for starting a joint journey affect the journey path?
|
| Information Search |
| Nature of social information | Might consumers turn to different sources of information for various inputs (e.g., social media influencers for normative information vs. proximal others for technical information)? Which product attributes or features are more likely to be influenced by social information versus other information sources?
|
| Sources of social information | When will consumers seek out information from AI agents rather than (human) social others? How do consumers sift through and integrate social and other information, and how does technology assist in this process? Might various social others become the go-to source for certain product categories or types of information, thereby actually shortening the information search process? How do echo chambers affect confidence? Do they change the type of information that is sought or that influences choice?
|
| Joint journeys | How do relative expertise and power affect the social sources of information gathered in a joint journey? How are the nature and diagnosticity of social sources of information viewed differently in the case of joint journeys?
|
| Evaluation |
| Evaluation of the source | How and when does greater social distance enhance perceptions of objectivity or affect the activation and use of persuasion knowledge? How is expertise or credibility established for various social others, including social media influencers? How does perceived intent and content affect assessments of the validity and importance of various information sources (including traditional endorsers vs. influencers)? How do consumers assess and incorporate information from AI agents, and what biases do they perceive from these sources?
|
| Evaluation of the information | When do consumers disregard information coming from a social source in favor of other sources of information, such as that provided directly by the brand itself or AI sources? How does the social distance of the source of information affect how that information is weighted in evaluation?
|
| Joint journeys | How might power dynamics influence dyadic interactions in joint journey evaluations? How is the expertise of a DMU member weighed against that of outside social others?
|
| Decision |
| Physical presence of social others | How and when are a consumer's choices influenced by the anticipation of others' responses? How do self-service technologies differentially affect consumers in the presence of known and unknown social others? How do the observable characteristics of others (demographic and behavioral) influence responses in various environments? How do physically present others influence the consumer who is shopping online? How does the impact of social others change the decision stage in higher-involvement contexts, such as financial decision making and health care?
|
| Decision |
| Felt presence of social others | How and when do AI agents serve as substitutes or complements to the influences of human others? Under what conditions will robots and other technology serve as surrogates for humans in the retail space? How and when can the virtual presence of social others affect online purchase decisions? What specific decision-making biases are based on assumptions about the motivations of social others? How do these change according to social distance?
|
| Joint journeys | How does the collective identity of a DMU impact the ultimate decision? How do the dynamics and relative power structures of joint DMUs within firms influence the point of decision in B2B contexts? How might the timing of the purchase (vs. the purchase decision) be of particular importance in joint journeys?
|
| Satisfaction |
| Satisfaction when consuming with others | What factors might cause consumers to experience "virtual" joint consumption, where independent experiences feel shared because consumers know others are experiencing something similar? When do consumers seek shared or social consumption versus individual or isolated experiences? When does shared consumption enhance versus detract from satisfaction? Will consumers be as satisfied with products that are used by others in access-based service contexts as they would be if they owned the products outright?
|
| Satisfaction based on inputs from others | When does consumption influenced by proximal versus distal others result in higher levels of satisfaction? When is satisfaction influenced by the consumer's anticipated ability to later share the experience with social others? How do the inputs from various social others positively and negatively influence consumption?
|
| Joint journeys | How is satisfaction in a joint journey independently determined by each member? What leads to the greatest or least divergence in satisfaction among members of the DMU?
|
| Postdecision Sharing |
| Reasons for sharing | What are the consumer well-being implications of sharing product experiences with others? When (and with whom) might consumers be more likely to share about experiences for which their own evaluations are less certain? Does the propensity to lie, offend, or humblebrag differ in online versus face-to-face contexts? Does this depend on the social distance of the social others?
|
| Audiences for sharing | When are consumers most likely to share product experiences with proximal versus distal others, and what type of information do they share? What leads consumers to complain to social others instead of the firm after product or service failures? How can virtual communities influence the social customer journey and potentially initiate new, unrelated customer journeys?
|
| Joint journeys | How is sharing implemented by individual members of a joint journey? When does the likelihood of sharing increase or decrease for a joint journey? How does the nature of what is shared change when more than one customer is involved?
|
| Additional Social Customer Journey Considerations |
| Nature of social influence | How do social inputs that are normative versus informational differentially affect the social journey? Do consumers respond differently to intentional versus unintentional social inputs along the journey? Under what circumstances are implicit social influences more powerful than explicit ones? What differences emerge when social others are physically present versus virtually present? How do cultural factors, such as an individualistic versus collectivistic group orientation, affect the social customer journey?
|
| Bidirectional social influence effects | What are the benefits to a focal consumer of seeing social others' journeys affected by their own social influence? What leads consumers to want to become more proximal influencers on others' customer journeys? At what stages of their own journey is this desire most likely to arise? How is sharing motivated by the consumer's desire to trigger a customer journey for others?
|
| Additional Social Customer Journey Considerations |
| How do the aggregated journeys of individuals and groups affect the journey of societies in the establishment of new social norms? Will a customer motivated by general social trends (i.e., distal others) be more likely to broadly advocate for the product after adopting it than if the motivation had come from a single, proximal other? How do social phenomena such as fear of missing out (FOMO) affect one's interpretations of others' journeys? How do they influence other consumers at each stage of the journey? How do power dynamics affect the bidirectional influences throughout the social customer journey?
|
| Cross-stage influences | Which social others are most likely to move consumers from the motivation to the search stage, and from the search stage to the evaluation stage, and so on, or backward in their journeys? How do the sources of information used in making purchase decisions affect the likelihood that consumers will share their experiences and the format in which they do so? How does additional social information encountered after a purchase decision influence satisfaction and sharing? How will the number or source of likes or comments on social media increase or decrease the time to next purchase? When will these social media cues lead to additional information seeking? When satisfaction is low, how do consumers change the way in which they attend to social others in the next journey? Are cross-stage effects in a joint journey distinct from those of the more general social journey? How might postdecision reactions from social others retroactively change a consumer's stated/remembered motivations? How do social customer journeys with fewer/more proximal traveling companions differ from those with many/more distal ones? How does this difference influence the overall timing of the cycle, which stages to skip or include, and the time until the next journey?
|
Social others often shape the motivations that initiate a customer journey. Consumers are frequently motivated by interactions with and observation of proximal and distal others. From feeling the need to "keep up with the Joneses" to seeing a friend's gushing review on Facebook to reading a celebrity's social media post about a new exercise regimen, decision journeys are often inspired by in-person or virtual contact with others. While small groups of well-known others are likely to exert direct influence over what motivates a consumer to begin a journey, distal others have an increasingly powerful impact as technology allows one's circle of influential others to expand beyond geographic proximity. In this section, we discuss social drivers of behavior that are primarily based on the proximal versus distal nature of the social other as key considerations for the motivation stage.
Consumers are motivated to affiliate with others and may do so by matching the consumption behavior of others (e.g., [117]). For example, what or how much one chooses to eat is often motivated by wanting to associate with social others more than one's own physiological need ([85]). That desire to affiliate depends on social distance; for example, consumers form stronger connections with brands that are associated with in-groups as opposed to out-groups ([38]). Yet consumers also make purchases as a way to differentiate from others ([19]), particularly in product categories that convey identity ([13]). Consumers are less likely to choose products that are associated with out-groups. For example, [126] showed that men had more negative evaluations of, and were less inclined to choose, a product associated with a female reference group. In short, the desire to affiliate or dissociate may be as dominant a motivation as addressing the underlying functional need (e.g., food, clothing, shelter), highlighting the importance of social motivation in understanding customer behavior.
Consumers are also motivated to engage in identity signaling, and they do so through the consumption of brands, products, and experiences that contain cultural meanings or associations with social status, traits, and aspirations ([65]; [98]). For example, choice of feature-rich products can signal wealth, technological skills, and openness to new experiences ([119]). Even if consumers are motivated to pursue a purchase for functional or other benefits, they may leverage the consumption opportunity for identity signaling. For example, consumers who are motivated to consume in an environmentally responsible manner may also use that consumption to signal how virtuous they are ([48]). The desire to signal status transcends socioeconomic boundaries. [95] found that social competition can lead to increased conspicuous consumption among consumers of low socioeconomic status. While most signaling via consumption is likely done to influence proximal others, we expect that such signaling may be especially effective among distal others, as proximal others will tend to have more existing knowledge of the consumer.
Distal social others may also have a significant impact on consumer decision making, especially on the altruistic or prosocial motivations of consumers ([20]; [98]). For example, consumers may be motivated to consume (or not consume) for purposes of the greater good, and their decisions may include various forms of environmentally responsible consumption ([52]), as illustrated by recent interest in discontinuing the usage of single-use plastics ([83]). In such cases, motivational influences can come from abstractly defined groups of social others and yet have a strong impact on one's own motivation for both adoption and disadoption of certain products or practices. Businesses are not immune from this kind of distal social pressure. Firms' decisions to source sustainably, to support (or not support) LGBTQ communities, or to force environmental, safety, or ethical standards on their suppliers are all examples of distal social influence on managers' motivations.
An intriguing example of how distal social groups can influence motivation relates to the role of political orientation in consumer decision making ([27]; [61]). [96] showed that political ideology affects how consumers differentiate from others, such that conservatives are more likely to choose products signaling hierarchical status whereas liberals are more likely to choose products signaling uniqueness. [88] found that higher CEO-to-worker pay ratios were associated with lower consumer purchase intentions, but only among Democrats and independents. [63] found that political ideology affects the relative motivation of maintaining status versus advancing status in the preference for luxury goods. In particular, political conservatives are more motivated by status maintenance in their consumption of luxury goods than are political liberals.
Social media platforms have increased the general revelation of one's political beliefs as well as the intermixing of consumer decision making and politics. Bringing political views and related social causes to the forefront and aligning such views with influencers or brands may change the perceived social distance of the traveling companion vis-à-vis the alignment with one's own political viewpoint. Interesting questions arise as to how consumers navigate conflicts between, for example, existing brand preferences and alignment with their social networks' political ideologies.
When the DMU has more than one member, there are unique opportunities for understanding the social nature of the customer journey because joint decision making introduces the need to negotiate different, often conflicting, motivations. Each member of the DMU must be on board with the motivation to engage in a specific journey, which means that individual members may need to persuade one another. Often this occurs in family settings in which DMUs share a collective identity and goals ([118]). This situation introduces roles for pro-relationship behaviors (e.g., [34]), such as empathy ([18]), as well as pitfalls involving power ([24]), which may even influence future decision journeys in unrelated domains. In a B2B context, decision making often formalizes and codifies the split motivations of a DMU. When a major purchase is under consideration, the motivations of employees representing purchasing, engineering, legal, and operations may be at odds (e.g., minimizing cost vs. maximizing reliability). A particularly interesting question relates to how conflicting motivations among individual members of a joint journey DMU are reconciled, and what marketers can do to facilitate and influence that reconciliation.
Viewing the motivation phase of the social customer journey with an emphasis on both proximal and distal others can lead to exciting research avenues with clear implications for researchers and practitioners. At the more distal level, cultural values affect the form that needs take (e.g., functional vs. symbolic) as well as the means consumers use to fulfill those needs (e.g., luxury vs. nonluxury car). Social media brings to light new and conflicting views about what is desirable, raising interesting research questions. When are desires more influenced by distal groups rather than by proximal influences? How does the consumer reconcile situations in which those closest to the consumer provide motivational inputs that are in opposition to larger but more distal societal norms? Some motivational conflicts may lead to journey abandonment instead of continuation, whereas others may involve a choice between two different journey paths. As [80] found, perceived progress toward one goal can motivate people to pursue alternative goals if the focal goal is close. The factors determining perceived progress are contextually dependent, making this a fruitful area for future research. The converse direction of influence is also worth considering. The motivation of the decision maker may exert its own influence on the social other. People infer the motivation of others from observable cues ([22]), and an actor who signals a high level of motivation may well motivate observers to set off on their own journeys.
In addition to the influence of social others on a consumer's motivations, these traveling companions can also influence when and how motivation leads to action. It is possible, even likely, that traveling companions can spur decision makers to the next stage of the journey. If so, which companions are most likely to move a consumer from the motivation stage to the information search stage and beyond? Motivations stemming from proximal social others might lead to quicker movement than motivations arising from distal others; understanding when this is and is not the case is important. The precise genesis of one's socially inspired motivations is likely to influence how one goes about continuing a customer journey.
Information search involves accessing memory and the external environment for relevant product information. A large literature from economics and marketing suggests that consumers search for product information until the costs of acquiring additional information exceed the benefit of that additional information. New technology has made information search less costly and allowed consumers to access a wealth of new information sources. Information acquired from social others through word-of-mouth has long been an important source of product information and is often viewed as less biased than information coming from a firm ([40]). For many purchases, other customers have now become the go-to source for product information. Technology has enabled the collation of word-of-mouth communications beyond those of physically close others. Of course, information search may still entail talking to proximal others, such as family, friends, and neighbors, but increasingly, consumers first turn to information proffered by distal others via effectively anonymous product reviews or recommendations by social media influencers ([21]). The level, type, sequence, and amount of search varies dramatically, and understanding this variation has been emphasized in prior research (e.g., [31]; [56]). We suggest that a deeper understanding of how various proximal and distal social inputs shape the search process is crucial in expanding overall understanding of this stage. In highlighting relevant issues at the intersection of social influence and information search, we emphasize the importance of the nature and sources of social information as key areas of consideration.
Not only has technology dramatically changed consumers' ability to search for information by enabling access to practically infinite sources of information, it has also changed the nature of the information on offer. Consumers may still access traditional sources of independent information (e.g., Consumer Reports), but they also may access personalized recommendations and evaluative information presented by countless distal consumers through product reviews. Recent research reflects the importance of product reviews within the customer journey. Key findings suggest that online reviews do not necessarily converge with independent expert reviews, such as those provided by Consumer Reports, and that consumers heavily weight average reviewer ratings without appropriately accounting for the sample size ([30]). Consequently, the inferences that consumers draw as they encounter and process this information directly affect the relative weight given to it in the evaluation stage. Consumers also draw inferences even in the absence of explicitly offered information, such as when social others display their preferences on platforms such as Pinterest. Even silence, particularly from a proximal social other, may be construed as assent or dissent with the decision maker's own opinion, despite the social other not having intended to play an active role in the decision-making process ([125]).
The idea that the silence of one's traveling companions can influence a customer's journey highlights the profound shifts in information search brought about by technology. Technology has helped blur the barriers between unknown sources, representing weak ties, and known sources, representing strong ties. This shift has led to the rise of interest groups focused on specific topics, to which people with similar tastes turn to find information, but it has also led to the rise of echo chambers, where people gain comfort from others holding similar opinions ([112]) and where negative opinions lead to more negative opinions ([54]). However, technology has also resulted in information searches that are broader and more open-ended, with consumers crowdsourcing inputs from social networks and thereby incorporating social others who previously would not have had any input on that journey. Finally, the opinion leaders and market mavens of times past are increasingly being replaced by social media influencers who cultivate fame by providing product information through blogs and other online platforms ([59]).
As [111] and [18] demonstrate, DMU members vary in expertise. They have a priori access to different information, and such differences will naturally manifest in the information search stage. Joint information search raises the possibility of an expanded knowledge base or one that is acquired more efficiently. However, it also raises the possibility that decision-relevant information will not be equally well understood by all members, creating asymmetries. Furthermore, joint decision making may change the nature of the information acquired. Members of the DMU may seek different types of information regarding choice options when they anticipate having to defend their evaluations or persuade partners than when they are gathering information for an independent decision. They may also change their reliance on various traveling companions to increase persuasiveness within the DMU; for example, they may have personally been persuaded by a social media influencer, whereas they believe that a decision partner may be more influenced by a perceived expert, and thus they may adjust their own search accordingly. Similarly, information gathering in a B2B decision setting may be motivated by the need to defend one's preferred option against the anticipated arguments of others in the DMU with a different set of preferences.
Although a large literature has addressed information search, we suggest that the focus on social influence opens opportunities to expand that research further. One drawback of easy access to information from social others is a sense of information overload, and with more information comes more variation in the nature and valence of the information. How do consumers sift and integrate this information, and what role does social distance play in determining the type of information that is most impactful? It is likely that consumers will turn to different sources of information for various inputs, such that certain social others become the go-to source for certain product categories or types of information. We further expect that the ease of access to multiple influencers, ironically, may serve as an impediment to information gathering, as the proliferation of experts makes information search seem never truly complete. Just as "Sale!" signs become less effective when too many items in a category have them ([ 5]), the simplifying heuristic of seeking out an expert's opinion can lose its simplifying power when dozens of experts' opinions must be weighed against each other.
Because technology has reduced search costs, it is likely to generate feelings of an incomplete information search. Thus, understanding how and when the information obtained from various proximal and distal social others propels a consumer on to the evaluation stage deserves further attention. [55] find that exposure to information on Facebook leads to more conventional feature choices, as looking at Facebook makes social others more salient, thereby increasing fear of negative evaluation from those others. Given that consumers acquire product information via Facebook, this may affect how they evaluate the possible options. Do social information sources further blur the line between information search and evaluation as others tend to simultaneously provide information and recommendations? Is the form of social influence different for specific social sources (e.g., informational from proximal others and normative from distal others), and how do consumers organize and categorize this information?
Perhaps the most influential aspect of the role of traveling companions in evaluation relates to the actual inputs provided by those companions. But the manner in which information is interpreted is often influenced by whom the information comes from. Persuasion models, such as the elaboration likelihood model ([99]), identify influencer characteristics that drive persuasion, including credibility and likability. We suggest, however, that distal or proximal others could be more or less persuasive, depending on contextual factors. In highlighting relevant issues at the intersection of social influence and evaluation, we first emphasize important evaluations of the source itself (e.g., credibility, liking) and then the information obtained as key areas of interest.
Given the emergence of new social sources of information, including social media influencers, it is important to reconsider classic notions of what makes a source persuasive. Credibility, trustworthiness, likability, and attractiveness have been examined in many consumer contexts for their role in shaping evaluations of a source. Extensive research on the customers' perceptions of salespeople has yielded relevant insights. For example, attractive or likable salespeople are more persuasive because they are seen as less likely to have an ulterior motive, compared with their unattractive or disliked counterparts ([104]). Furthermore, an attractive celebrity endorser positively affects brand attitudes under low involvement ([99]), or under high involvement when attractiveness is relevant to the product category ([62]).
In the online environment, mechanisms for establishing credibility for distal influencers include inputs such as the number of likes, followers, or reviews. As consumers regularly interact with bloggers and social media influencers, those relationships become more familiar and thereby more proximal, potentially enhancing credibility. However, this credibility can also be altered when firms become involved, for example, through sponsored posts. Recent findings suggest that greater blogger expertise has stronger effects for raising awareness than for spurring product trial, and expertise has differential effects based on the specific online platform (e.g., Facebook vs. a blog; [59]).
Although they are closely intertwined, the evaluation of the source of social information and of the information itself can be different, such that the nature of the social information customers receive can influence the evaluation of that specific product information. Learning about an attribute from multiple sources might increase the perceived importance of that attribute; for example, hearing from several friends that a movie has surprising twists might increase the importance of plot complexity in evaluating information about the movies currently showing. Structured customer evaluations (e.g., Audible.com encourages customer ratings and feedback along several specific dimensions) are likely to increase the importance of comparable attributes relative to anecdotal word-of-mouth, which might privilege noncomparable attribute information. For informationally complex products, customers can simplify the choice by evaluating the overall customer rating to the exclusion of nearly all other information. Increasingly, information in the form of aggregate customer reviews serves as a substitute for specific attribute information. Interestingly, moderately positive reviews have been found to be more persuasive than extremely positive ones because they are perceived to be more thoughtful, thereby enhancing credibility ([69]). Social influence operating through social norms can also affect how attribute information is weighted. Attributes that might have received little attention in the past, including whether something has a small carbon footprint, is cruelty-free, or is made from recycled materials, can drive choice when social norms change what is important to customers. Firms have proven to be especially sensitive to changes in these norms, both downstream, as a way of enticing customers, and upstream, as a criterion for selecting among vendors.
We expect social distance to further moderate the way social influence affects the evaluation of information. While social norms are a distal social influence, previous research suggests that when these norms are filtered through socially proximal exemplars, the effects on how information is evaluated can be even more influential. For example, in the context of green consumption, [ 4] found that energy usage was affected by information about the energy consumption of one's neighbors, and [44] showed that hotel guests' reuse of towels was affected by the towel usage behavior of other hotel guests. These findings suggest that combinations of distal social influences (e.g., norms) and proximal influences (e.g., neighbors, previous occupants) might be especially powerful in influencing information evaluation.
Multiple members of a DMU exerting influence on the outcome opens the door to new factors influencing the evaluation process, including the perceived fairness of the decision-making process and the weighting of outcomes of past journeys. Members of a DMU may track who "won" in past decision journeys and use this history to "equalize gains" this time around ([24]). As [47] put it, the relationship between the parties is not a "constraint on utility maximization," which causes a "deviation from the base state of independent or autonomous decision making" (p. 170). Rather, joint utility maximization is the "central explanatory concept" for understanding the decision-making process.
These interesting dynamics have led to research on how evaluation is affected by the relative standing of individuals within the DMU. When members within a DMU differ in their preferences, the weight that each member's preferences receive is based on influence within the DMU ([82]). [41] found that couples who pooled finances in a joint bank account were more likely to choose utilitarian products when spending from that joint account than when spending from a separate account, due to the need to justify the decision. Members of DMUs also have individual, relational, and collective identities that ebb and flow over time ([37]). Hence, individual preferences may change when members consider their collective identity as a part of the DMU. For example, looking at joint decisions relating to self-control (e.g., spending, healthy eating), [34] found that dyads with a member low in self-control tended to make decisions reflecting reduced self-control because the member high in self-control relinquished that self-control to maintain harmony within the dyad.
Social others have a significant impact on how consumers formulate and evaluate consideration sets based on both who they are and what information they provide. Much is known about this process, but key questions remain. For example, how does social distance affect the evaluation and weighting of information on product attributes? As is the case with the entirety of the customer journey, conventional wisdom may suggest that those closest to the consumer have the greatest influence, but the familiarity of close others may make perceived biases, particularly those relevant to evaluation, more transparent to the consumer. This is especially relevant in the presence of conflicting opinions about product alternatives. When would a more credible, but distal, social other fare better against a less credible but proximal one? Furthermore, one's close social group may hold a contrary evaluation of a product compared with the wisdom of thousands of reviewers on Amazon. In such instances, people must grapple with clear trade-offs between going with the views of their smaller and more proximal in-group or those of a larger, but less known and more distal group of social others. Consumers may well develop heuristics to deal with such situations, and understanding precisely what these heuristics are deserves further attention.
As highlighted by [59], new technology has moved the power of endorsements from traditional celebrities to social media influencers. We see significant opportunities to examine how credibility, trustworthiness, likability, and attractiveness are established for these influencers, and how the intent, content, and platform all combine to drive the effectiveness of these sources. [ 6] view characteristics of influencers as key variables in how brands will choose to use them for conveying their messages. Relatedly, how do consumers apply persuasion knowledge ([40]) when evaluating the potential usefulness or truthfulness of other consumers' articulated experiences or opinions, and does social distance affect the level of scrutiny?
We also see opportunities for research related to the role of AI-based recommendation agents in this stage of the social customer journey ([68]). These AI agents are designed to learn from the customer's expressed preferences in concert with aggregated data on choice patterns. Such algorithms should improve with usage and, ideally, be able to accurately predict preferences. What might be the optimal consideration set size provided by an AI agent? Could the set consist of just one option? The better an algorithm performs, the more likely it is to serve as a substitute rather than a complement to human social influence. Thus, the reverse may also be possible: the more a consumer sees an AI agent as a traveling companion, the more likely the consumer is to rely exclusively on its evaluations. The effect of social closeness of AI agents on advice acceptance is a fascinating area for future investigation as it will likely affect perceptions of credibility and trustworthiness and differ across contexts; for example, within health care, patients are concerned with how AI agents account for their unique circumstances ([79]), but in other product categories, such as entertainment or music, advice from the AI agent may be more readily accepted.
A decision is the culmination of the predecision stages, suggesting that the decision reflects all the social influences encountered thus far. [73] proposed dividing what was typically considered a single stage into two separate states: "decide: to make up one's mind about whether to buy, and if so, which particular brand or product to purchase" and "purchase: to buy a brand or product item that one has decided upon" (p. 280). We consider this a useful distinction. From the perspective of the current framework, we argue that traveling companions influence both the decide state, as they can shape whether and what to buy, and the purchase state, as they can play a key role at the point of purchase. In highlighting relevant issues at the intersection of social influence and decision, we emphasize the role of physical and felt social presence at the time of purchase.
Much research has examined how the physical presence of social others at the point of purchase can affect the outcome. Such presence makes their influence more proximal and may affect consumers because of self-presentational or other concerns. For instance, [70] found that male consumers tend to spend more when they shop with a friend than when they shop alone, and [10] showed that the likelihood of using coupons decreased in the presence of others. In both cases, the physical presence of others activated a desire to avoid appearing cheap. Impression management at the point of purchase is not limited to situations involving known others. Even the mere presence of others can affect the experience of various emotions, including embarrassment, at the point of purchase ([ 8]; [28]). More generally, from a marketer's perspective, the crowding or social density of one's surroundings can have both positive (e.g., increased brand attachment; [58]) and negative (e.g., decreased purchase intentions; [128]) effects on consumers. The presence of others may also be relevant at the time of purchase as social others may provide valuable product information or express their own preferences. Such "moment of truth" social inputs may well override prior inputs gathered in the predecision stages because of their proximity to the purchase. For example, the presence of others during a grocery shopping trip is associated with increased in-store decision making ([60]) and a more positive shopping experience ([77]).
Just as technology has changed the role of social influence in the predecision stages, it has also enhanced the felt presence of social others at the point of purchase. As noted by [106] in discussing social factors in retail contexts, consumers may connect virtually through FaceTime or other technologies to get live insights from others who are not physically present with them. In the online context, firms use tactics that explicitly highlight felt social presence by noting other consumers' interest in a product (e.g., "30 other customers have booked this today," "5 people are viewing this right now!"). Such tactics may trigger scarcity concerns or make salient that a product is highly desirable ([53]). Technology has also enabled a subtler use of social influence as firms leverage information about the decisions of social others (e.g., electricity usage of neighbors; [ 4]) to influence decision making and short-circuit the "last mile problem" in a customer's journey, whereby the final transition from evaluation to action fails ([115]). Furthermore, we suggest that AI agents can be viewed as a form of felt social presence at the point of purchase, particularly in online shopping environments.
In a joint journey, the decision is influenced by both the individual and joint utilities of the DMU members. The more closely members of the DMU have traveled together along the predecision phases of the journey, the more straightforward the decision phase will likely be. However, because the decision is the stage of ultimate commitment, it might also be the stage where disagreements or misaligned motives and preferences come to a head. Indeed, different members of the DMU may use different evaluation models (e.g., one spouse uses a lexicographic model, while the other spouse uses a weighted-additive model), and these differences would only become evident at this stage. If so, negotiations might cause the DMU to return to any of the previous stages of the journey.
Furthermore, the point of the actual purchase has special significance in many joint journeys given that such decisions often have significant financial or other implications, so decisions about the timing and mechanism of purchase also become important. Joint decision journeys may be especially likely to evoke risk aversion at the decision stage, relative to solo customer journeys. The old B2B sales aphorism, "Nobody ever got fired for buying IBM," speaks to the perceived importance of sticking with "safe" decision options—the status quo option or the dominant brand—when employees know they will eventually have to justify their decisions to more powerful members of the DMU.
An excellent review by [ 7] synthesizes findings of social influences in retail contexts. They draw important distinctions between active and passive social influences and then further distinguish (within passive influences) those aimed at a focal recipient versus those merely witnessed by other customers. The authors provide keen insights about future research directions that fit primarily within the decision stage of our social customer journey. We also note opportunities to focus on other specific decision contexts wherein social influences are likely to be powerful, including decisions about experiences, financial decision making, and decisions in health care settings. Furthermore, given that much of the research about the impact of present social others is in physical contexts, we see opportunities for understanding the impact of present and virtual traveling companions in online consumption contexts.
The decision may also be influenced by anticipation of others' responses to the choice. More proximal traveling companions might naturally be expected to have a greater influence on the current decision than more distal others—even those who may have played an important role at earlier stages in the journey, as closer others are more likely to see the consumer using the product. Yet clearly the breadth of the audience that can be made aware of one's decision has expanded through the same technologies discussed throughout. Thus, once the decision has been made and executed, traveling companions will play a key role in how the focal consumer uses and evaluates the product.
Nonhuman entities are increasingly entering the retail space, changing the nature of what social presence means. [124] found that the use of social cues on online retail sites led consumers to rate the websites as more helpful and intelligent, and [87] found increased food consumption in response to the discomfort felt from being served by a robot. This raises questions about whether and how robots and other technological devices will serve as surrogates for humans in the retail space. We expect that these devices will tend to be perceived as distal "others"; however, this will likely change as their physical proximity and ubiquity—in our homes, on our wrists—are likely to shift perceptions, making them seem more proximal. Important questions emerge about how and when these technologies will serve as substitutes or complements to the influences of human others. It seems likely that consumers will begin to substitute artificial social others for the flesh-and-blood variety for purchases that are especially embarrassing, that are ephemeral or low-stakes, or that are beyond the technical expertise of the customer to evaluate.
Traveling companions are likely to play a crucial role in how consumers experience a product in terms of both usage and evaluation. One's own evaluations and potential postpurchase regret may be influenced by the evaluations of others; for example, receiving compliments on one's new haircut or watching one's family enjoy an escape room experience can directly enhance one's own satisfaction with those purchases. In addition, technology has significantly expanded the types of social others that consumers might draw on as they consume and evaluate acquired products. In highlighting relevant issues at the intersection of social influence and satisfaction, we emphasize the effects of consuming with others and learning from others as key areas of consideration.
Prior research has demonstrated how joint consumption, even when the decisions are made independently, affects evaluation of experiences ([81]). For example, experiencing something in the presence of others can lead to emotional contagion, whereby individuals share more consonant emotions than when experiencing the same thing independently, leading to enhanced enjoyment or dissatisfaction ([102]). On the other hand, recent research suggests that consumers overestimate how much they will enjoy an activity engaged in with another person versus alone. Specifically, [103] demonstrated that consumers often feel inhibited from engaging in hedonic activities alone, especially when these activities are observable by others, as they anticipate that others will make negative inferences about them. As it turns out, consumers enjoy the activity just as much whether it is undertaken alone or with another person.
Even when consumers do not consume a product jointly, others may have a significant influence on usage and satisfaction. Consumers care about why others choose the same option that they did and even experience reduced confidence in their own decisions when others arrive at the same decision for different reasons ([71]). Furthermore, social others help consumers to more fully experience the goods and services they have chosen. For products that require learning, the role of traveling companions in this learning process is crucial in influencing usage and satisfaction. A consumer who purchases an Instant Pot cooker on the basis of a friend's recommendation may ask that friend for usage tips and recipes, or the consumer may look to social media (i.e., distal others) for that information. How-to videos garnered more than 42 billion views in 2018 alone ([84].) Technology has likely made acquiring usage information from distal others less costly in terms of time and effort than acquiring the information from proximal others. Furthermore, information from distal others may be more useful as they may offer more varied and possibly less biased inputs on usage.
In the case of joint journeys, the decision and often the consumption are experienced as a joint DMU. Yet the members' individual assessments of satisfaction may vary. These satisfaction assessments could pertain to both the outcome and the process: dissatisfaction with the process or even with the other members of the DMU may well manifest itself in the individual's level of satisfaction with the product or service and/or even in unrelated future journeys embarked on by the DMU. Furthermore, each member may obtain usage information and tips from different social sources or may invest varying amounts of time and effort in using a product, thereby leading to different experiences.
The many ways traveling companions can affect postdecision satisfaction remain largely open for investigation. When does consumption influenced by distal (vs. proximal) others result in greater satisfaction? This question would depend on whether expectations are set differently depending on who is observing the consumption and also on whether feedback about the experience would be shared or not. In some contexts, the act of consumption, not just the feedback, is shared. How does "virtual" joint consumption, where independent experiences feel shared because the individuals know that others are concurrently having similar experiences, affect satisfaction? And when does shared consumption enhance versus detract from satisfaction? Much of the extant research on joint consumption has focused on food, but opportunities are available for research in new settings such as online gaming or online exercise classes, where joint consumption occurs virtually. Likewise, technology has changed what were previously solitary experiences (e.g., watching a sporting event at home) into things that can be jointly experienced and evaluated via real-time online discussions, leading to the question of the circumstances in which consumers seek shared consumption versus individual experiences. The rise of access-based products, for which a consumer pays per usage, in an increasing number of industries (e.g., clothing, boats) raises new questions about satisfaction with products that have been used by many others. Some research points to psychological benefits of consuming previously owned and used products (e.g., [107]), but there are likely cases where the opposite occurs, and further research could shed light on both the costs and benefits.
Finally, as consumers seek out how to use new products, how do others affect that learning and satisfaction? The degree of collaboration in expectation setting, perceived togetherness in consumption, and the extent to which the feedback is shared publicly versus experienced privately should all influence satisfaction with joint consumption experiences. As consumers evaluate their satisfaction with products in the context of the social influences discussed previously, they are likely also to be developing a sense of how willing they might be to share their experiences with others, raising interesting questions such as when consumption, influenced by others, is more likely to lead to postdecision sharing.
Sharing experiences through postpurchase word-of-mouth is an inherently social process. Consumers have always been able to tell friends, neighbors, and coworkers about their purchases and consumption experiences. However, a qualitative shift has occurred in the relative ease with which a consumer can share opinions with large audiences, both known and unknown, through social media and product review platforms. Some of the motivations that compel the customer to share experiences with others, including social affiliation and identity signaling, may have motivated the entire customer journey. But the motivations to share may be different from the motivations that lead to the purchase decision, and must be considered independently. We also highlight the importance of the expanding audiences with which consumers share their product experiences.
Prior research suggests that the reasons for sharing product experiences and the consequences thereof are wide-ranging. We suggest that social distance may influence the likelihood that product experience will be conveyed and the form that this sharing will take. In general, consumers are more likely to share experiences for which they have stronger attitudes, positive or negative ([ 2]); however, this likelihood may depend on social distance (e.g., [21]). One reason that consumers share consumption experiences is that sharing serves as a mechanism for identity signaling. Before social media, consumers conveyed their identity through readily observable material purchases, such as clothing or a car. Now, people convey identity through what they post, including their tastes in music, books, and food, as well as via the brands and influencers that they follow. Unlike the offline context, where product choices are limited by personal (e.g., income) or contextual (e.g., audience size) factors, online sharing allows consumers more freedom to carefully curate the identities they convey through discussion or display of products and experiences, whether or not they actually own or use them. In this way, technology has brought distal others closer and given them the opportunity to be included in a focal customer's journey. Ironically, this has happened even though many people use social media to express themselves at a distance, not communicating directly with targeted recipients ([17]).
Consistent with the increased ease of sharing of experiences and opinions is an increased potential for some of this shared information to be less objective or sincere. In research conducted prior to mass use of social media, [108] found that consumers may conceal that they purchased a product at a regular price when it is important to be perceived as a smart shopper. Concomitantly, technology has facilitated the phenomenon of humblebragging ([109]), a form of insincere sharing that is common on social media. Clearly, one's motives for sharing affect what is shared and to whom the sharing is targeted.
Consumers also share their product experiences to affiliate with others, and who these others are affects the nature of the sharing. For example, consumers are more likely to share negative product experiences with friends to connect emotionally, whereas they are more likely to share positive product experiences with strangers to give a better impression of themselves ([21]). Furthermore, the nature of what is shared changes with audience size: people are more other-focused and share more useful content with one person (i.e., narrowcasting) but are more concerned with not looking bad when sharing with larger audiences (i.e., broadcasting; [11]). Sharing may have additional unintended consequences as well. For example, [91] demonstrate that consumers providing reviews feel an emotional boost from the enhanced social connectedness they experience that then leads them to buy impulsively.
Another form of postdecision sharing is done within brand communities, in which people bond over their common affiliation with a brand ([92]). It is likely that the more socially close one feels to a community of brand loyalists, the more likely one is to become an advocate for the brand, as strong relationships with community members enhance identification with the community, leading to further engagement. Yet negative effects of brand community membership have been identified, as communities exert pressure to conform to rules and practices ([ 3]). [130] found that participation in a brand community can lead to riskier financial decision making because members perceive that others in the community will help and support them if the decision turns out poorly. Furthermore, reputations are built within the community and affect how community members engage with each other ([50]). In sum, technology has significantly increased the number of potential audiences and communities with which consumers can share their consumption experiences.
As with information search, it is likely that members of a joint DMU will engage in postdecision sharing as individuals rather than as a pluralized DMU. For example, presumably most reviews are composed by single individuals, even if the decision was the result of a joint journey. It is also likely that sharing behaviors in joint journeys are less tightly coupled within the DMU than predecision cognitions and behaviors are. Although a joint DMU needs to reach some form of consensus before a decision is made, members of the DMU may go their separate ways in evaluating the experience and sharing those opinions. Alternatively, some joint journeys might culminate in social pressure to align in terms of evaluations and sharing, potentially reducing the likelihood of sharing widely. Consider a family vacation wherein multiple members of the family participated in the motivation, information gathering, decision, and consumption phases of the trip. They may feel subsequent pressure to come to a consensus evaluation of the vacation, and to share stories that are consistent and complementary across the family DMU.
Many interesting questions emerge regarding postdecision sharing, including how consumers curate their consumption experiences for others. Identity signaling depends on the audience ([14]), and thus many opportunities are available to examine the distance between the focal consumer and the audience, and how that distance affects sharing. How does a consumer's likelihood of sharing and the type of information shared vary with the stage of the listener's customer journey? What level of detail might consumers provide about experiences to different types of traveling companion? When do consumers refrain from sharing even the best or worst experiences because they fear a negative response from others? When and with whom might consumers be more likely to share less certain evaluations? Related questions concern the venue or medium through which consumers share. Different platforms facilitate different types of communication. Consumers writing a review on Amazon, calling out a brand on Twitter, or telling a customer experience story on Reddit will be operating in environments with different norms of communication and different capabilities facilitated by the platform. Recent research has found that the interface through which consumers write reviews (smartphone vs. computer) affects the emotionality of the content ([86]) such that the more constrained nature of smartphones leads to shorter but more emotional reviews. Parallel research investigating how customers' postconsumption sharing is influenced by where they choose to share is likely to uncover similarly interesting findings.
Questions arise as to whether postdecision sharing enhances or hurts consumers' well-being. For example, postdecision sharing can bolster affiliation and deepening of relationships—[ 6] identify social media as a vehicle for combating loneliness—but social media usage can also trigger feelings of envy and isolation ([76]). Additional questions relate to how consumers curate their consumption experiences for impression management. Interestingly, consumers may, at times, desire to elicit negative responses from others. For example, consumers may share their consumption experiences to offend others, allowing them to enhance a normatively negative self-identity ([78]). The topics of lying and humblebragging about purchases also lead to interesting research questions, such as whether the propensity to engage in these actions varies with social distance. If consumers suspect misrepresentation by social others, this suspicion could reduce the persuasiveness that one's postdecision sharing has on others' information search and evaluation. Today's virtual communities are places where consumers discuss not just brands but all aspects of life (e.g., healthy eating, raising kids). This raises interesting questions about how community members bring brands into community discussions. Finally, bringing the discussion full circle, how is a consumer's sharing motivated by a personal motivation to trigger a customer journey for a traveling companion?
Customer journey frameworks are among the few concepts that lie squarely at the intersection of marketing theory and practice. Their utility to both practitioners and researchers has led to the creation of a wide range of alternative formulations, of increasing complexity. The debate over the shape and length of customer journeys has obscured the fact that these frameworks have tended to take the perspective of individual consumers, operating independently. Without denying the usefulness of that approach, we suggest that such a focus fails to capture a rich variety of consumer decisions that are inherently social in nature. In fact, we suggest that most consumer decisions involve some form of traveling companion. We highlighted two social lenses to layer on the classic decision-making journey: ( 1) we introduced the notion that various social others, individually or in aggregate, influence a given individual customer's decision journey at the various stages while themselves being influenced by that customer, and ( 2) we recognized that some customer journeys occur within a DMU of more than one individual, and that these joint journeys are also influenced at various stages by traveling companions. We pointed to representative examples of social influence effects from prior research and identified a variety of research opportunities within each of the key stages in the decision journey.
Transcending the specific research questions that are pertinent where discussed, several broader emergent themes also provide further opportunities for meaningful research. This section highlights key additional considerations about the nature of social influence, bidirectional social influence effects, and cross-stage social influences as key categories of future research, and we provide further specifics in Table 2.
Our focus has been on the role that the social distance between the focal consumer and traveling companions plays in the way that those traveling companions affect the customer journey. We recognize that an important additional layer for examining the role of social influence lies in considering the nature of the social influence. Social influence will vary in ways beyond social distance, some of which we have touched on already. For example, influence exerted by a social other may be normative or informational. Moreover, the influence attempt may be intended or unintended, whereby the consumer either is specifically targeted or is an incidental recipient of persuasive information from social others. Influence may be direct, with the social other being in physical or virtual proximity to the customer, or indirect, operating through third parties. Indeed, the influence may even be implicit, in that it is perceived by the focal customer without any intention to influence on the part of the social other (see [ 7] for similar distinctions). These important differences in the very nature of the social influence may change the path of a journey as the consumer moves forward or turns back to revisit prior stages. These distinctions also raise questions about whether consumers in collectivistic cultures are more interested in exerting, or are more susceptible to, social influence, than consumers in individualistic cultures. Each of these considerations adds nuance to the research questions highlighted throughout.
The bidirectional arrows between the focal consumer and the traveling companions in Figure 1 are intended to emphasize that as consumers experience social influence, they simultaneously influence the customer journeys of their companions. This bidirectional influence may take many forms, including a simple back-and-forth conversation about a product between friends that initiates customer journeys for both or the posting of a review by one customer at the end of a journey intended to help distal others during the early stages of their own journeys.
The bidirectionality of social influence throughout the customer journey can also be seen in the way individuals react to distal, macro-level social forces, which in turn affect the contributions individual consumers make to their social networks and society as a whole. For example, [123] found that income inequality was related to the sharing stage of the customer journey, with luxury brand mentions on Twitter more likely in geographic regions with high inequality, potentially motivating customer journeys in those regions and reinforcing the cycle. Arguably, the burgeoning sharing economy ([100]) is further evidence of the bidirectionality of social influence in the customer journey. The move from ownership to consumer-to-consumer rental transactions breaks down the traditional roles of buyer and seller, facilitating two-way social influence throughout the customer journey. We view these as interesting examples, but many questions regarding how macro-level social factors, such as inequality, corruption, and aging populations, interact with individual customer journeys remain.
The bidirectional nature of social influence might be further extended and used to understand the creation of consumption norms for groups of people and even entire societies. Consider how norms regarding single-use straws have changed over time, with consumers moving away from usage of nonbiodegradable, single-use straws to usage of reusable or biodegradable straws or to abandoning the category entirely ([83]). These norms have changed as the result of the decision journeys of many individuals influencing other individuals and groups until, ultimately, a shift in society's path is detectable. We propose that one may treat shifts in norms as the "decision" a cultural group has reached after a societal-level journey through motivation, information search, and evaluation to a collective decision. From this perspective, just as social others can affect the journey of individuals, so too do the aggregated journeys of individuals affect the journey of societies in the establishment of new social norms, practices, and laws.
As mentioned, traveling companions may influence one or multiple phases of the customer journey. For example, the same person(s) or group(s) may enter one's journey at just one stage, play a role in the entire journey, or stroll in and out of the journey at various stages, all while separate companions do likewise. Numerous emerging questions are based on how either the same or different sources of social influence at the various stages of the customer journey merge to influence the decision process. For instance, will a customer motivated by general social trends (i.e., distal others) be more likely to advocate for a product after purchase than if the motivation had come from a single, proximal other? How does sharing with traveling companions affect the time until one starts another, similar journey, motivated in part by the desire to share again? Furthermore, when does the motivation to eventually share the outcome initiate a journey? When consumption satisfaction is low, do consumers change the way in which they attend to their traveling companions in the next journey?
Another cross-stage insight afforded by the social lens is that traveling companions may catalyze a customer's progress from one stage of a journey to the next. We highlighted several of these possibilities in the discussion of the individual stages, and we present additional questions probing these cross-stage influences in Table 2. A traveling companion may provide a tipping point in moving the focal customer from information search to evaluation of the options or by pressuring the focal customer to instead slow down and reconsider the motivations for the journey and the information gathered thus far. A business leader's initial motivation to investigate some particular technology or strategy could be based on the social influence of friends or competitors, but these social influences could also be the impetus to move from information gathering to more serious evaluation of the options, or from evaluation to purchase. Interesting questions relate to the form of these tipping points and how they are affected by the social distance of others as well as companions' desire to play either a transient or more pervasive role in the customer journey. Other interesting questions relate to how expectations of others' reactions at the sharing stage affect how the focal customer goes about gathering information and evaluating the alternatives.
Although the main objective of this article was to cast a new light on customer journeys in order to spur fresh research, viewing the customer journey in its social setting clearly has implications for marketers throughout the journey stages, particularly in the areas of new product development, communication and sales strategies, online and in-store technologies, and B2B sales.
Table 3 highlights some practical questions and insights for each of the social customer journey stages. Key managerial issues across the entire social customer journey involve how and when to become involved in what might otherwise be only consumer-to-consumer interactions. Considerations might include when and how firms should respond to negative customer reviews or social media callouts, when to highlight a social media influencer who is implicitly or explicitly endorsing one's product, how to manage sponsored blog posts, and when to provide corrective information when consumers are exposed to unfavorable product information by their peers. Of particular interest to marketers, social influence may be used to nudge consumers from evaluation to decision. As discussed, technology has increased the number of opinions that bear on a customer's journey and has even begun to provide a decision-support system wherein the customer and AI agent together reach a final decision. Firms must carefully consider their usage of AI technologies, attending specifically to the social implications.
Graph
Table 3. Emerging Implications for Marketing Practice.
| Questions | Implications |
|---|
| Motivation | |
| • Which proximal social motivations (e.g., affiliation, signaling) motivate customers, and what actions might a brand take in response? | Consumption decisions are often driven by the need to affiliate with or distinguish oneself from other people. Meeting these needs requires knowing not just about the customers' desires but also something about the groups that the customers wish to affiliate with or distinguish themselves from—and the dimensions on which they would want to affiliate or distinguish themselves. Brands should carefully consider how they approach social influencers, with a keen understanding of underlying motivations of both customers and the groups to which they belong. |
| • How should a brand be positioned relative to distal social drivers? | Neutrality on social and political issues (e.g., LGBTQ rights, environmental policies, nationalism) is an increasingly untenable position for brands. Brands must weigh the implications of picking a side on hot-button issues—with the understanding that they could be forced into taking a position by customers or activists. |
| Information Search |
| • What is the nature of the social information influencing customers? | Target customers may be more persuaded by aggregated ratings, detailed individual reviews, or in-person testimonials from close friends depending on the purchase context. Positive word-of-mouth communications may not affect customer behavior if the information does not come in the form that is most compatible with how those customers make their choices. |
| • How do customers cope with too much social information? | Customers may utilize a variety of heuristics to manage the often overwhelming amount of information available from other customers, from simplified information-processing strategies (e.g., a "satisficing" rule) to turning to aggregate statistics to limiting information acquisition to a small number of influential sources. Brands need to find ways to positively influence these various sources. |
| Evaluation |
| • Could AI agents serve as traveling companions for customers? | AI agents can become complements to social influences (e.g., by facilitating access to or navigating through customer reviews), but they might also serve as substitutes for traditional social influences by replacing some of the evaluative roles historically filled by social others. Brands must consider whether the replacement of social influences with AI agents represents an opportunity or a threat. |
| • How might firms present product assortments differently in store or online depending on readily available social information at the time of evaluation? | Brands make decisions about both product assortments and how to organize or structure these assortments, both online and offline. When curating options and developing consideration sets, they should take into account social inputs. For example, popularity cues and customer reviews can be provided in more detail at the time of first browsing or after the customer has already narrowed the choice sets. Brands decide both how readily to facilitate product comparisons and whether and how to include social information in these comparisons. |
| Decision |
| • Are customers affected more by physically present others or by virtually present others? | Social influences on customers' decisions can come from the concurrent presence of others during a decision or from the felt presence of others virtually present in the setting, including the imagined presence of friends or family who will later see the purchased item and even the nebulous and diffused influence that comes from social norms. For example, the decision to use or not use disposable plastic straws could be influenced by either the arguments of specific persuaders or by the sense of a broad, societal consensus for or against. Brands will need to consider how they might provide access to the most relevant forms of social information. |
| • What are the unique challenges of social influences at the point of decision in B2B contexts? | Given that in many B2B contexts, brands have undergone a thorough decision-making process prior to the point of decision, it is important to guard against irrelevant social influence at the final point of decision. Plural DMUs, including the various stakeholders involved in a B2B purchase, develop strategies for handling decision conflict. Failing to understand and anticipate how those conflicts are negotiated can scuttle an otherwise potentially successful purchase. |
| Satisfaction |
| • How can a brand offering's value be enhanced via input from social others? | Whether through brand communities, wiki-type collective instruction databases, or "modding" customization groups, social others can provide network-effect-based additional value to purchases. Arguably, part of the success of Instant Pot was attributable to the amateur chefs who posted recipes online to share with their fellow pressure cooker owners. |
| • When and how should brands encourage joint consumption with other people? | For both material goods and experiences, consuming with others often leads to richer experiences and satisfaction. Firms should creatively look for ways to enhance satisfaction with their products by incorporating opportunities for joint usage with others. |
| Postdecision Sharing |
| • How can understanding customers' motivations for sharing improve marketing strategy? | The motivation for sharing an opinion about a purchase can be social (e.g., identity signaling, social closeness) or self-serving (e.g., self-expression, symbolic self-completion). Brands that rely on word-of-mouth communications would be well advised to provide both reasons to purchase and reasons to share after the decision. The types of information that customers are most likely to share may or may not be the information that leads other customers to purchase. |
| • What can brands learn for improving their product designs or marketing strategies from their monitoring of social information sharing? | Brands need to systematically monitor information that their customers are sharing with others, but this needs to be converted into meaningful improvements in product design or other marketing mix elements, or even broader marketing strategies. The customer voices that speak the loudest in social media (e.g., those with large followings) may not accurately represent the experiences and preferences of a brand's target customers. Information gathered by monitoring social information must also be weighted according to its value in representing actual or preferred customers. |
| Cross-Stage Influences |
| • How are B2B contexts specifically affected by influences across the social customer journey? | Many B2B marketers have already developed sophisticated journey maps for their customers. Such maps should include the various sources of social influence that could facilitate or inhibit a customer's path to purchase. A useful first step would be identifying the social others influencing B2B customers (e.g., other customers, trade groups and trade conference participants, consultants). |
| • How should brands allocate resources to influencing social information across the decision journey? | Brands should bring an understanding of social influence into marketing strategy. Decisions involving facilitating access to social information and allocating resources to encourage versus discourage social interaction are critical and must be considered in terms of the allocation of resources. In some contexts, brands will benefit from limiting customers' access to social information within their customer environments, and in other contexts a brand's attempt at policing the sharing of information between customers will result in backlash. |
| • How should brands most effectively utilize social influencers throughout the social customer journey? | More and more brands are actively managing social influencers, for example, through sponsored blog posts and content on social media platforms. Managers must grapple with which sources to focus on and what portion of their marketing budget to deploy to shape various forms of online word-of-mouth. Brands must consider how these efforts should be integrated across the stages of the journey. The role of social media influencers in B2B contexts is underresearched but is likely to be substantial in some industries. |
4 Notes: This table summarizes and expands on implications of the social customer journey framework. The questions are intended as a starting point for managers looking to apply the insights discussed in the article.
Finally, firms must continue to find new ways to gather, analyze and use information collected from online and peer-to-peer communications to develop useful metrics to understand the changing motivations, decision heuristics, and satisfaction assessments of customers. Promising research in this area continues apace; [97] demonstrate a text analysis tool to examine the sentiment of online reviews, and [121] propose examining the effectiveness of various story lines captured with reviews. New methods for effectively creating usable knowledge from the abundance of socially developed information is critical for marketers.
With the hyperconnectivity brought about by technology, and the concomitant rise of social influences in consumer decision making, the conceptualization of customer journeys as paths populated by individual consumers traveling alone can be limiting from the perspective of generating ideas for substantive research. We opened this article with a discussion of recent research that has identified negative effects of technology on the social milieu. Our analysis of social influences on the customer journey presents a picture that we hope is more positive. The increasingly social nature of the customer journey has the potential to influence and improve people's lives in many ways. Accordingly, we hope that this article will inspire new avenues of research into topics that matter to both managers and researchers.
Footnotes 1 Author Contributions All authors contributed equally.
2 Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 Funding The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The fourth author gratefully acknowledges financial support from the HKUST endowed professorship.
4 ORCID iD Rosellina Ferraro https://orcid.org/0000-0002-2960-0646
References Agnihotri Raj, Dingus Rebecca, Hu Michael Y., Krush Michael T. (2016), "Social Media: Influencing Customer Satisfaction in B2B Sales," Industrial Marketing Management, 53, 172–80.
Akhtar Omair, Christian Wheeler S. (2016), "Belief in the Immutability of Attitudes Both Increases and Decreases Advocacy," Journal of Personality and Social Psychology, 111 (4), 475–92.
Algesheimer Rene, Dholakia Utpal M., Herrmann Andreas. (2005), "The Social Influence of Brand Community: Evidence from European Car Clubs," Journal of Marketing, 69 (3), 19–34.
Allcott Hunt, Rogers Todd. (2014), "The Short-Run and Long-Run Effects of Behavioral Interventions: Experimental Evidence from Energy Conservation," American Economic Review, 104 (10), 3003–37.
5 Anderson Eric T., Simester Duncan I. (2001), "Are Sale Signs Less Effective When More Products Have Them?" Marketing Science, 20 (2), 121–42.
6 Appel Gil, Grewal Lauran, Hadi Rhonda, Stephen Andrew T. (2020), "The Future of Social Media in Marketing," Journal of the Academy of Marketing Science, 48, 79–95.
7 Argo Jennifer J., Dahl Darren W. (2020), "Social Influence in the Retail Context: A Contemporary Review of the Literature," Journal of Retailing, forthcoming, DOI:10.1016/j.jretai.2019.12.005.
8 Argo Jennifer J., Dahl Darren W., Manchanda Rajesh V. (2005), "The Influence of a Mere Social Presence in a Retail Context," Journal of Consumer Research, 32 (2), 207–12.
9 Ariely Dan, Levav Jonathan. (2000), "Sequential Choice in Group Settings: Taking the Road Less Traveled and Less Enjoyed," Journal of Consumer Research, 27 (3), 279–90.
Ashworth Laurence, Darke Peter R., Schaller Mark. (2005), "No One Wants to Look Cheap: Trade-offs Between Social Disincentives and the Economic and Psychological Incentives to Redeem Coupons," Journal of Consumer Psychology, 15 (4), 295–306.
Barasch Alixandra, Berger Jonah. (2014), "Broadcasting and Narrowcasting: How Audience Size Affects What People Share," Journal of Marketing Research, 51 (3), 286–99.
Belch George E., Belch Michael A. (2015). Advertising and Promotion: An Integrated Marketing Communications Perspective, 10 th ed. New York : McGraw-Hill Education.
Berger Jonah, Heath Chip. (2007), "Where Consumers Diverge from Others: Identity Signaling and Product Domains," Journal of Consumer Research, 34 (2), 121–34.
Berger Jonah, Ward Morgan, (2010), "Subtle Signals of Inconspicuous Consumption," Journal of Consumer Research, 37 (4), 555–69.
Brewer Marilynn B., Gardner Wendi. (1996), "Who Is This 'We'? Levels of Collective Identity and Self Representations," Journal of Personality and Social Psychology, 71 (1), 83–93.
Brown Jacqueline J., Reingen Peter H. (1987), "Social Ties and Word-of-Mouth Referral Behavior," Journal of Consumer Research, 14 (3), 350–62.
Buechel Eva, Berger Jonah. (2018), "Microblogging and the Value of Undirected Communication," Journal of Consumer Psychology, 28 (1), 40–55.
Burns Alvin C., Granbois Donald H. (1977), "Factors Moderating the Resolution of Preference Conflict in Family Automobile Purchasing," Journal of Marketing Research, 14 (1), 77–86.
Chan Cindy, Berger Jonah, Boven Leaf Van. (2012), "Identifiable but Not Identical: Combining Social Identity and Uniqueness Motives in Choice," Journal of Consumer Research, 39 (3), 561–73.
Chaney Kimberly E., Sanchez Diana T., Maimon Melanie R. (2019), "Stigmatized-Identity Cues in Consumer Spaces," Journal of Consumer Psychology, 29 (1), 130–41.
Chen Zoey. (2017), "Social Acceptance and Word of Mouth: How the Motive to Belong Leads to Divergent WOM with Strangers and Friends," Journal of Consumer Research, 44 (3), 613–32.
Cheng Yimin, Mukhopadhyay Anirban, Williams Patti. (2020) "Smiling Signals Intrinsic Motivation," Journal of Consumer Research, 46 (5), 915–35.
Chernev Alexander. (2014), Strategic Marketing Management, 8 th ed. Chicago : Cerebellum Press.
Corfman Kim P., Lehmann Donald R. (1987), "Models of Cooperative Group Decision-Making and Relative Influence: An Experimental Investigation of Family Purchase Decisions," Journal of Consumer Research, 14 (1), 1–13.
Court David, Elzinga Dave, Finneman Bo, Perrey Jesko. (2017), "The New Battleground for Marketing-Led Growth," McKinsey Quarterly (February), https://www.mckinsey.com/business-functions/marketing-and-sales/our-insights/the-new-battleground-for-marketing-led-growth.
Court David, Elzinga Dave, Mulder Susan, Vetvik Ole Jorgen. (2009), "The Consumer Decision Journey," McKinsey Quarterly (June), https://www.mckinsey.com/business-functions/marketing-and-sales/our-insights/the-consumer-decision-journey.
Crockett David, Wallendorf Melanie. (2004), "The Role of Normative Political Ideology in Consumer Behavior," Journal of Consumer Research, 31 (3), 511–28.
Dahl Darren W., Manchanda Rajesh V., Argo Jennifer J. (2001), "Embarrassment in Consumer Purchase: The Roles of Social Presence and Purchase Familiarity," Journal of Consumer Research, 28 (3), 473–81.
Davis Harry L. (1970), "Dimensions of Marital Roles in Consumer Decision-Making," Journal of Marketing Research, 7 (2), 168–77.
De Langhe Bart, Fernbach Philip M., Lichtenstein Donald R. (2015), "Navigating by the Stars: Investigating the Actual and Perceived Validity of Online User Ratings," Journal of Consumer Research, 42 (6), 817–33.
Diehl Kristin. (2005), "When Two Rights Make a Wrong: Searching Too Much in Ordered Environments," Journal of Marketing Research, 42 (3), 313–22.
Donavan Todd, Minor Michael S., Mowen John. (2016), Consumer Behavior. Chicago : Chicago Business Press.
Duhan Dale F., Johnson Scott D., Wilcox James B., Harrell Gilbert D. (1997), "Influences on Consumer Use of Word-of-Mouth Recommendation Sources," Journal of the Academy of Marketing Science, 25 (4), 283 –95.
Dzhogleva Hristina, Lamberton Cait. (2014), "Should Birds of a Feather Flock Together? Understanding Self-Control Decisions in Dyads," Journal of Consumer Research, 41 (2), 361–80.
Edelman David C., Singer Marc. (2015), "Competing on Customer Journeys," Harvard Business Review (November), 88–100.
Edwards Steven M. (2011), "A Social Media Mindset," Journal of Interactive Advertising, 12 (1), 1–3.
Epp Amber M., Price Linda L. (2008), "Family Identity: A Framework of Identity Interplay in Consumption Practices," Journal of Consumer Research, 35 (1), 50–70.
Escalas Jennifer E., Bettman James R. (2003), "You Are What They Eat: The Influence of Reference Groups on Consumers' Connections to Brands," Journal of Consumer Psychology, 13 (3), 339–48.
Forrester (2010), " Time to Bury the Marketing Funnel: Marketers Must Embrace the Customer Life Cycle," https://www.forrester.com/report/Its+Time+To+Bury+The+Marketing+Funnel/-/E-RES57495#.
Friestad Marian, Wright Peter. (1994), "The Persuasion Knowledge Model: How People Cope with Persuasion Attempts," Journal of Consumer Research, 21 (1), 1–31.
Garbinsky Emily N., Gladstone Joe J. (2019), "The Consumption Consequences of Couples Pooling Finances," Journal of Consumer Psychology, 29 (3), 353–69.
Gershoff Andrew D., Mukherjee Ashesh, Mukhopadhyay Anirban. (2003), "Consumer Acceptance of Online Agent Advice: Extremity and Positivity Effects," Journal of Consumer Psychology, 13 (1/2), 161–70.
Gilbert Daniel T. (1998), " Ordinary Personology," in The Handbook of Social Psychology, Vol. 2, Gilbert Daniel T., Fiske Susan T., Gardner Lindzey, eds. New York : McGraw-Hill, 89–150.
Goldstein Noah J., Cialdini Robert B., Griskevicius Vladas. (2008), "A Room with a Viewpoint: Using Social Norms to Motivate Environmental Conservation in Hotels," Journal of Consumer Research, 35 (3), 472–82.
Gorlin Margarita, Dhar Ravi. (2012), "Bridging the Gap Between Joint and Individual Decisions: Deconstructing Preferences in Relationships," Journal of Consumer Psychology, 22 (3), 320–33.
Granovetter Mark S. (1973), "The Strength of Weak Ties," American Journal of Sociology, 78 (6), 1360–80.
Greenhalgh Leonard, Chapman Deborah I. (1995), " Joint Decision Making," in Negotiation as a Social Process, Kramer Roderick M., Messick David M., eds. Thousand Oaks, CA : Sage Publications, 166–85.
Griskevicius Vladas, Tybur Joshua M., Bergh Bram Van den. (2010), "Going Green to Be Seen: Status, Reputation, and Conspicuous Conservation," Journal of Personality and Social Psychology, 98 (3), 392–404.
Hamilton Rebecca, Price Linda L. (2019), "Consumer Journeys: Developing Consumer-Based Strategy," Journal of the Academy of Marketing Science, 47 (2), 187–91.
Hanson Sara, Jiang Lan, Dahl Darren W. (2019), "Enhancing Consumer Engagement in an Online Brand Community via User Reputation Signals: A Multi-Method Analysis," Journal of the Academy of Marketing Science, 47 (2), 349–67.
Hartmann Wesley R., Manchanda Puneet, Nair Harikesh, Bothner Matthew, Dodds Peter, Godes David, et al. (2008), "Modeling Social Interactions: Identification, Empirical Methods and Policy Implications," Marketing Letters, 19 (3/4), 287–304.
Haws Kelly L., Winterich Karen Page, Naylor Rebecca Walker. (2014), "Seeing the World Through GREEN-Tinted Glasses: Green Consumption Values and Responses to Environmentally Friendly Products," Journal of Consumer Psychology, 24 (3), 336–54.
He Youngfu, Oppewal Harmen. (2018), "See How Much We've Sold Already! Effects of Displaying Sales and Stock Level Information on Consumers' Online Product Choices," Journal of Retailing, 94 (1), 45–57.
Hewett Kelly, Rand William, Rust Roland T., van Heerde Harald J. (2016), "Brand Buzz in the Echoverse," Journal of Marketing, 80 (3), 1–24.
Hildebrand Christian, Schlager Tobias. (2019), "Focusing on Others Before You Shop: Exposure to Facebook Promotes Conventional Product Configurations," Journal of the Academy of Marketing Science, 47 (2), 291–307.
Honka Elisabeth, Chintagunta Pradeep. (2017), "Simultaneous or Sequential? Search Strategies in the U.S. Auto Insurance Industry," Marketing Science, 36 (1), 21–44.
Howard John A., Sheth Jagdish N. (1969), The Theory of Buyer Behavior. New York : John Wiley & Sons.
Huang Xun (Irene), (Tak) Huang Zhongqiang, Wyer Robert S. Jr. (2018), "The Influence of Social Crowding on Brand Attachment," Journal of Consumer Research, 44 (5), 1068–84.
Hughes Christian, Swaminathan Vanitha, Brooks Gillian. (2019), "Driving Brand Engagement Through Online Social Influencers: An Empirical Investigation of Sponsored Blogging Campaigns," Journal of Marketing, 83 (5), 78–96.
Inman J. Jeffrey, Winer Russell S., Ferraro Rosellina. (2009), "The Interplay Among Category Characteristics, Customer Characteristics, and Customer Activities on In-Store Decision Making," Journal of Marketing, 73 (5), 19–29.
Jost John T. (2017), "The Marketplace of Ideology: 'Elective Affinities' in Political Psychology and Their Implications for Consumer Behavior," Journal of Consumer Psychology, 27 (4), 502–20.
Kahle Lynn R., Homer Pamela M. (1985), "Physical Attractiveness of the Celebrity Endorser: A Social Adaptation Perspective," Journal of Consumer Research, 11 (4), 954–61.
Kim J. Christine, Park Brian, Dubois David. (2018), "How Consumers' Political Ideology and Status-Maintenance Goals Interact to Shape Their Desire for Luxury Goods," Journal of Marketing, 82 (6), 123–49.
Kim Kyeongheui, Zhang Meng, Li Xiuping. (2008), "Effects of Temporal and Social Distance on Consumer Evaluations," Journal of Consumer Research, 35 (4), 706–13.
Kirmani Amna, Ferraro Rosellina. (2017), " Social Influence in Marketing: How Other People Influence Consumer Information Processing and Decision Making," in The Oxford Handbook of Social Influence, Harkins Stephen G., Williams Kipling D., Burger Jerry, eds. New York : Oxford University Press, 415–30.
Kotler Philip, Armstrong Gary. (2012), Principles of Marketing, 1 4t h ed. Boston : Pearson Prentice Hall.
Kristofferson Kirk, White Katherine. (2015), " Interpersonal Influences in Consumer Psychology: When Does Implicit Social Influence Arise? " in Handbook of Consumer Psychology, Norton Michael I., Rucker Derek D., Lamberton Cait, eds. Cambridge, UK : Cambridge University Press, 419–45.
Kumar V., Dixit Ashutosh, Javalgi Rajshekar G., Dass Mayukh. (2016), "Research Framework, Strategies, and Applications of Intelligent Agent Technologies (IATs) in Marketing," Journal of the Academy of Marketing Science, 44 (1), 24–45.
Kupor Daniella, Tormala Zakary. (2018), "When Moderation Fosters Persuasion: The Persuasive Power of Deviatory Reviews," Journal of Consumer Research, 45 (3), 490–510.
Kurt Didem, Inman J.Jeffrey, Argo Jennifer J. (2011), "The Influence of Friends on Consumer Spending: The Role of Agency–Communion Orientation and Self-Monitoring," Journal of Marketing Research, 48 (4), 741–54.
Lamberton Cait Poynor, Naylor Rebecca W., Haws Kelly L. (2013), "Same Destination, Different Paths: The Effect of Observing Others' Divergent Reasoning on Choice Confidence," Journal of Consumer Psychology, 23 (1), 74–89.
Latane Bibb. (1981), "The Psychology of Social Impact," American Psychologist, 36 (4), 343–56.
Lee Leonard, Inman J.Jeffrey, Argo Jennifer J., Böttger Tim, Dholakia Utpal, Gilbride Tim, et al. (2018), "From Browsing to Buying and Beyond: The Needs-Based Shopper Journey Model," Journal of the Association for Consumer Research, 3 (3), 277–93.
Lemon Katherine N., Verhoef Peter C. (2016), "Understanding Customer Experience Throughout the Customer Journey," Journal of Marketing, 80 (6), 69–96.
Lewis Elias St. Elmo. (1908), Financial Advertising. Indianapolis : Levey Brothers.
Lin Liu Yi, Sidani Jaime E., Shensa Ariel, Radovic Ana, Miller Elizabeth, Colditz Jason B., et al. (2016), "Association Between Social Media Use and Depression Among US Young Adults," Depression and Anxiety, 33 (4), 323–31.
Lindsey-Mullikin Joan, Munger Jeanne. (2011), "Companion Shoppers and the Consumer Shopping Experience," Journal of Relationship Marketing, 10 (1), 7–27.
Liu Peggy J., Lamberton Cait, Bettman James R., Fitzsimons Gavan J. (2019), "Delicate Snowflakes and Broken Bonds: A Conceptualization of Consumption-Based Offense," Journal of Consumer Research, 45 (6), 1164–93.
Longoni Chiara, Bonezzi Andrea, Morewedge Carey K. (2019), "Resistance to Medical Artificial Intelligence," Journal of Consumer Research, 46 (4), 629–50.
Louro Maria J.S., Pieters Rik, Zeelenberg Marcel. (2007), "Dynamics of Multiple Goal Pursuit," Journal of Personality and Social Psychology, 93 (1), 174–93.
Lowe Michael L., Haws Kelly L. (2014), "(Im)Moral Support: The Social Outcomes of Parallel Self-Control Decisions," Journal of Consumer Research, 41 (2), 489–505.
Lowe Michael, Nikilova Hristina, Miller Chadwick J., Dommer Sara L. (2019), "Ceding and Succeeding: How the Altruistic Can Benefit from the Selfish in Joint Decisions, " Journal of Consumer Psychology, 29 (4), 652–61.
Madrigal Alexis C. (2018), "Disposable America: A History of Modern Capitalism from the Perspective of the Straw. Seriously," The Atlantic (June 21), https://www.theatlantic.com/ technology/archive/2018/06/disposable-america/563204/.
Marshall Carla. (2019), "Online Tutorials: Viewers Flock to YouTube for How-To Videos," Tubular Insights (January 23), https://tubularinsights.com/how-to-videos/.
McFerran Brent, Dahl Darren W., Fitzsimons Gavan J., Morales Andrea C. (2010), "I'll Have What She's Having: Effects of Social Influence and Body Type on the Food Choices of Others," Journal of Consumer Research, 36 (6), 915–29.
Melumad Shiri, Inman J.Jeffrey, Pham Michel Tuan. (2019), "Selectively Emotional: How Smartphone Use Changes User-Generated Content," Journal of Marketing Research, 56 (2), 259–75.
Mende Martin, Scott Maura L., van Doorn Jenny, Grewal Dhruv, Shanks Ilana. (2019), "Robots Rising: How Humanoid Robots Influence Consumers' Service Experiences," Journal of Marketing Research, 56 (4), 535–56.
Mohan Bhavya, Schlager Tobias, Deshpande Rohit, Norton Michael I. (2018), "Consumers Avoid Buying from Firms with Higher CEO-to-Worker Pay Ratios," Journal of Consumer Psychology, 28 (2), 344–52.
Morgan Robert M., Hunt Shelby D. (1994), "The Commitment-Trust Theory of Relationship Marketing," Journal of Marketing, 58 (3), 20–38.
Mothersbaugh David L., Hawkins Delbert I. (2015), Consumer Behavior: Building Marketing Strategy, 1 3th ed. New York : McGraw-Hill Education.
Motyka Scott, Grewal Dhruv, Aguirre Elizabeth, Mahr Dominik, de Ruyter Ko, Wetzels Martin. (2018), "The Emotional Review-Reward Effect: How Do Reviews Increase Impulsivity?" Journal of the Academy of Marketing Science, 46 (6), 1032–51.
Muniz Albert M. Jr., O'Guinn Thomas C. (2001), "Brand Community," Journal of Consumer Research, 27 (4), 412–32.
Naylor Rebecca Walker, Lamberton Cait Poynor, West Patricia M. (2012), "Beyond the 'Like' Button: The Impact of Mere Virtual Presence on Brand Evaluations and Purchase Intentions in Social Media Settings," Journal of Marketing, 76 (6), 105–20.
Novak Thomas P., Hoffman Donna L. (2019), "Relationship Journeys in the Internet of Things: A New Framework for Understanding Interactions Between Consumers and Smart Objects," Journal of the Academy of Marketing Science, 47 (2), 216–37.
Ordabayeva Nailya, Chandon Pierre. (2011), "Getting Ahead of the Joneses: When Equality Increases Conspicuous Consumption Among Bottom-Tier Consumers," Journal of Consumer Research, 38 (1), 27–41.
Ordabayeva Nailya, Fernandes Daniel. (2018), "Better or Different? How Political Ideology Shapes Preferences for Differentiation in the Social Hierarchy," Journal of Consumer Research, 45 (2), 227–50.
Ordenes Francisco V., Ludwig Stephan, de Ruyter Ko, Grewal Dhruv, Wetzels Martin. (2017), "Unveiling What Is Written in the Stars: Analyzing Explicit, Implicit, and Discourse Patterns of Sentiment in Social Media," Journal of Consumer Research, 43 (6), 875–94.
Oyserman Daphna. (2009), "Identity-Based Motivation and Consumer Behavior," Journal of Consumer Psychology, 19 (3), 276–79.
Petty Richard E., Cacioppo John T., Schumann David. (1983), "Central and Peripheral Routes to Advertising Effectiveness: The Moderating Role of Involvement," Journal of Consumer Research, 10 (2), 135–46.
Price Linda L., Belk Russell W. (2016), "Consumer Ownership and Sharing: Introduction to the Issue," Journal of the Association for Consumer Research, 1 (22), 193–97.
Puccinelli Nancy M., Goodstein Ronald C., Grewal Dhruv, Price Robert, Raghubir Priya, Stewart David. (2009), "Customer Experience Management in Retailing: Understanding the Buying Process," Journal of Retailing, 85 (1), 15–30.
Ramanathan Suresh, McGill Ann L. (2007), "Consuming with Others: Social Influences on Moment-to-Moment and Retrospective Evaluations of an Experience," Journal of Consumer Research, 34 (4), 506–24.
Ratner Rebecca K., Hamilton Rebecca W. (2015), "Inhibited from Bowling Alone," Journal of Consumer Research, 42 (2), 266–83.
Reinhard Marc-Andre, Messner Matthias, Sporer Siegfried L. (2006), "Explicit Persuasive Intent and Its Impact on Success at Persuasion: The Determining Roles of Attractiveness and Likeableness," Journal of Consumer Psychology, 16 (3), 249–59.
Richardson Adam. (2010), "Using Customer Journey Maps to Improve Customer Experience," Harvard Business Review (November 15), https://hbr.org/2010/11/using-customer-journey-maps-to.
Roggeveen Anne L., Grewal Dhruv, Schweiger Elisa. (2020), "The DAST Framework for Retail Atmospherics: The Impact of In- and Out-of-Store Retail Journey Touchpoints on the Customer Experience," Journal of Retailing, forthcoming, DOI:10.1016/j.jretai.2019.11.002.
Sarial-Abi Gulen, Vohs Kathleen D., Hamilton Ryan, Ulqinaku Aulona. (2017), "Stitching Time: Vintage Consumption Connects the Past, Present, and Future," Journal of Consumer Psychology, 27 (2), 182–94.
Sengupta Jaideep, Dahl Darren W., Gorn Gerald J. (2002), "Misrepresentation in the Consumer Context," Journal of Consumer Psychology, 12 (2), 69–79.
Sezer Ovul, Gino Francesca, Norton Michael I. (2018), "Humblebragging: A Distinct—and Ineffective—Self-Presentation Strategy," Journal of Personality and Social Psychology, 114 (1), 52–74.
Sheth Jagdish N. (1973), "A Model of Industrial Buyer Behavior," Journal of Marketing, 37 (4), 50–56.
Sheth Jagdish N. (1974), " A Theory of Family Buying Decisions," in Models of Buyer Behavior, Sheth Jagdish N., ed. New York : Harper and Row, 17–33.
Shore Jesse, Baek Jiye, Dellarocas Chrysanthos. (2018), "Network Structure and Patterns of Information Diversity on Twitter," MIS Quarterly, 42 (3), 849–72.
Simpson Jeffry A., Griskevicius Vladas, Rothman Alexander J. (2012), "Consumer Decisions in Relationships," Journal of Consumer Psychology, 22 (3), 304–14.
Solomon Michael R. (2015), Consumer Behavior: Buying, Having, and Being, 11th e d. Boston : Pearson.
Soman Dilip. (2015), The Last Mile: Creating Social and Economic Value from Behavioral Insights. Toronto : University of Toronto Press.
Strong Edward K. Jr., (1925), The Psychology of Selling and Advertising. New York : McGraw-Hill.
Tanner Robin J., Ferraro Rosellina, Chartrand Tanya L., Bettman James R., van Baaren Rick B. (2008), "Of Chameleons and Consumption: The Impact of Mimicry on Choices and Preferences," Journal of Consumer Research, 34 (6), 754–66.
Thomas Tandy Chalmers, Epp Amber, Price Linda. (2020), "Journeying Together: Aligning Retailer and Service Provider Roles with Collective Consumer Practices," Journal of Retailing, forthcoming, DOI:10.1016/j.jretai.2019.11.008.
Thompson Debora V., Norton Michael I. (2011), "The Social Utility of Feature Creep," Journal of Marketing Research, 48 (3), 555–65.
Trope Yaacov, Liberman Nira. (2010), "Construal-Level Theory of Psychological Distance," Psychological Review, 117 (2), 440–63.
Van Laer Tom, Escalas Jennifer Edson, Ludwig Stephan, van den Hende Ellis A. (2019), "What Happens in Vegas Stays on Trip Advisor? Understanding the Role of Narrativity in Consumer Reviews," Journal of Consumer Research, 46 (2), 267–85.
Verhoef Peter C., Lemon Katherine N., Parasuraman A., Roggeveen Anne, Tsiros Michael, Schlesinger Leonard A. (2009), "Customer Experience Creation: Determinants, Dynamics, and Management Strategies," Journal of Retailing, 85 (1), 31–41.
Walasek Lukasz, Bhatia Sudeep, Brown Gordon D.A. (2018), "Positional Goods and the Social Rank Hypothesis: Income Inequality Affects Online Chatter About High- and Low-Status Brands on Twitter," Journal of Consumer Psychology, 28 (1), 138–48.
Wang Liz C., Baker Julie, Wagner Judy A., Wakefield Kirk. (2007), "Can a Retail Web Site Be Social?" Journal of Marketing, 71 (3), 143–57.
Weaver Kimberlee, Hamby Anne. (2019), "The Sounds of Silence: Inferences from the Absence of Word-of-Mouth," Journal of Consumer Psychology, 29 (1), 3–21.
White Katherine, Dahl Darren W. (2006), "To Be or Not Be? The Influence of Dissociative Reference Groups on Consumer Preferences," Journal of Consumer Psychology, 16 (4), 404 –14.
Wiesel Thorsten, Pauwels Koen, Arts Joep. (2011), "Marketing's Profit Impact: Quantifying Online and Off-Line Funnel Progression," Marketing Science, 30 (4), 604–11.
Zhang Xiaoling, Li Shibo, Burke Raymond R., Leykin Alex. (2014), "An Examination of Social Influence on Shopper Behavior Using Video Tracking Data," Journal of Marketing, 78 (5), 24–41.
Zhao Min, Xie Jinhong. (2011), "Effects of Social and Temporal Distance on Consumers' Responses to Peer Recommendations," Journal of Marketing Research, 48 (3), 486–96.
Zhu Rui, Dholakia Utpal M., Chen Xinlei, Algesheimer Rene. (2012), "Does Online Community Participation Foster Risky Financial Behavior?" Journal of Marketing Research, 49 (3), 394–407.
~~~~~~~~
By Ryan Hamilton; Rosellina Ferraro; Kelly L. Haws and Anirban Mukhopadhyay
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 131- Values Created from Far and Near: Influence of Spatial Distance on Brand Evaluation. By: Chu, Xing-Yu; Chang, Chun-Tuan; Lee, Angela Y. Journal of Marketing. Nov2021, Vol. 85 Issue 6, p162-175. 14p. 2 Graphs. DOI: 10.1177/00222429211000706.
- Database:
- Business Source Complete
Record: 132- Variable Compensation and Salesperson Health. By: Habel, Johannes; Alavi, Sascha; Linsenmayer, Kim. Journal of Marketing. May2021, Vol. 85 Issue 3, p130-149. 20p. 1 Diagram, 7 Charts, 1 Graph. DOI: 10.1177/0022242921993195.
- Database:
- Business Source Complete
Variable Compensation and Salesperson Health
Positive effects of incentives on salespeople's motivation, effort, and performance are well-established in literature. This article takes a novel look at their influence on salespeople's health. The results of four empirical studies, including more than 1,400 salespeople, suggest that an increasing variable compensation share (i.e., greater pay-for-performance component in salespeople's compensation plans) increases salespeople's stress, resulting in emotional exhaustion and more sick days. These outcomes are more likely for salespeople with lower personal ability and fewer social resources. The harmful effects of the variable compensation share on salespeople's health reduce the positive effects of this incentive on sales performance. To practice better marketing for a better world, if variable compensation is used, the authors recommend that managers screen new hires for job-related resources and help their existing staff build such resources. In addition, companies may personalize incentive schemes and sensitize managers to the stress-inducing effects of variable compensation shares.
Keywords: absenteeism; emotional exhaustion; health; pay-for-performance; sales force incentives; stress; variable compensation
Variable compensation plans are used widely to motivate salespeople and account for approximately 40% of total sales compensation in the United States—equivalent to over $320 billion ([66]). Variable compensation is issued on top of a base salary, and the amount is contingent on performance. For example, a salesperson with a target salary of $100,000 and a variable compensation share of 20% would receive $80,000 as a fixed amount, then the remaining $20,000 contingent on the achievement of sales targets.
Substantial research has established that variable compensation plans increase salespeople's motivation and efforts and thus their performance ([14]; e.g., [ 9]; [45]). Another research stream challenges this positive view, however, by establishing some early evidence that variable pay might impose health-threating pressure and stress on employees (Table 1). This emerging literature helps motivate the current research in four main ways.
Graph
Table 1. Literature Review: Effects of Monetary Incentives on Health-Related Outcomes.
| Authors | Methodology | Dependent Variable | Context | Nonlinear Effects? | Contingencies? | Mediation Mechanism? | Effects on Performance? | Results |
|---|
| Dahl and Pierce (2020) | Survey data, regression analyses | Use of antidepressant medication
| Unspecified firm worker/employees | No | Performance Age Gender
| No | No | Pay for performance increases mental health problems, and the extent depends on contingency factors. |
| Parker et al. (2019) | Online experiment, online survey, structural equation model | Emotional exhaustion Prosocial behavior
| Unspecified employees | No | No | Intrinsic motivation Identified regulation External regulation
| No | The more employees appraise performance-based pay as a hindrance (challenge), the more (less) they will feel strain and the less (more) they will behave prosocially. |
| DeVaro and Heywood (2017) | Manager survey panel data, fixed effects regressions | Perceived ailment by management Perceived absence by management
| Unspecified firm worker/employees | No | No | No | Yes | Performance pay increases ailments and absences. The elevated rates of ailments reduce financial performance. |
| Davis (2016) | Worker survey, random effects model | Perceived emotional health Perceived physical health
| Garment workers | No | No | No | No | Wage incentives reduce perceived emotional and physical health. |
| Yeh et al. (2009) | Survey data, regression analyses | Burnout Work stress
| Mix of manual and nonmanual workers | No | No | No | No | Workers with a performance-based pay system show higher levels of burnout. |
| Eisenberger and Aselage (2009) | Lab experiment, survey, regression analyses | Performance pressure Creativity
| Students / alumni | No | No | No | No | High performance rewards increase performance pressure, which in turn increases creativity. |
| Shirom et al. (1999) | Survey data, regression analyses | Depression Somatic complaints
| Blue-collar workers | No | No | Work monotony | No | Performance-contingent pay increases depression and somatic complaints. The effect is mediated by work monotony. |
| Timio, Gentili, and Pede (1979, 1976) | Longitudinal field experiment | Occupational stress (adrenaline and noradrenaline measures)
| Metal workers | No | No | No | No | Payment by results strongly increases levels of adrenaline and noradrenaline, in effects that persist even after six months. |
| Our research | Natural experiment, DiD analysis; survey data, path models | Sick days Emotional exhaustion Sales performance
| Salespeople | Yes | Personal ability Social resources
| Stress Effort
| Yes | The variable compensation share progressively increases health problems for salespeople with volatile performance and few resources. |
First, the stream is nascent, requiring further research to reach a point of empirical generalizability. Such a goal is important given one in four employees feel burned out at work "very often or always," and almost half of them feeling this way "sometimes" ([41]). Second, the nature of the relationship between variable pay and health remains unclear in several respects. To begin, only one field study provides causal evidence from the field ([69]), while the remaining studies offer either correlational evidence ([17]; [18]) or evidence from the lab ([56]). Further, most studies ignore important contingencies that might influence this relationship (the study by [17]] is an exception). Finally, no studies examine the presence of nonlinear effects.
Third, the effects of an opportunity to earn variable pay on health mainly have been identified for blue-collar workers (e.g., [18]; [69]), not in sales contexts ([28]). This gap is striking given that variable compensation plans are so ubiquitous in sales force management. Furthermore, managers are uncertain about the effect of variable pay on salesperson health. A survey of 182 U.S. sales managers via an online panel (53.3% female, average age 34.5 years) asked them to rate their agreement with the statement "Variable compensation increases health issues among salespeople" and found an average score of 4.63 (SD = 1.59) on a 7-point scale. Fourth, no study has examined the effect of variable pay on performance through health-related effects. Given that health problems can reduce salespeople's performance ([24]), it is important to assess the performance cost of variable pay due to these negative health costs.
With these research motivations, we examine the contingent effects of variable sales compensation shares on salesperson health and performance outcomes. Leveraging the conservation of resources (COR) theory ([34]), we predict that variable compensation share, defined as the contractually agreed-on share of variable relative to total pay that salespeople receive if they achieve their targets ([39]), can lead to health problems because it renders salespeople's compensation uncertain and thereby induces stress. We examine whether the effect is nonlinear and propose five contingencies reflecting job-related resources that influence this relationship. We also examine how health problems due to variable compensation reduce sales performance.
We test our framework with four empirical studies. The first is a natural experiment, involving a company that changed the variable compensation share in one of its business units. Results show that a higher variable compensation share is associated with enhanced sales performance but also with more sick days, which, in turn, reduce the gains to sales performance. We replicate our findings in a cross-industry survey in Study 2, which uses emotional exhaustion as an alternative measure of health problems. Furthermore, the study reveals a nonlinear, J-shaped effect of the variable compensation share on health problems. Study 3, a single-company study that tests for moderating and mediating mechanisms, reveals that the effect of the variable compensation share on emotional exhaustion is mediated by stress and moderated by salespeople's experience, team identification, and relationship quality with their leader. Study 4 is a follow-up experiment that reveals that participants chose lower variable compensation shares if they were made aware of the stress induced by their decisions, particularly if they scored high on the empathy dimension of perspective taking.
These findings have important implications for both marketers and the academic discipline that are important to generating better-world outcomes. Notably, managers should strive to build sales forces that possess the job-related resources required to cope with the potential stress arising from variable compensation shares. In addition, managers should modulate the use of variable compensation overall or personalize to the salespeople to better manage its negative effect on health. For theory, given that we link sales incentives to health outcomes, our research should raise awareness of the human cost of sales performance. We also extend knowledge about health-related outcomes beyond sales performance outcomes of variable compensation share by identifying nonlinear, moderated, and mediated effects.
As Figure 1 shows, we predict that as the variable compensation share increases, salespeople's sales performance increases (direct effect), but the increases will be lower for salespeople who experience stress-induced health problems (indirect effect). We include two indicators of health problems: ( 1) physician-attested sick days and ( 2) emotional exhaustion, defined as "feelings of being overextended and depleted of one's emotional and physical resources" ([49], p. 399). The former measure aims to capture both physical and mental health problems severe enough to prompt a visit to a physician, and the latter is an indicator of mental health concerns.
Graph: Figure 1. Conceptual framework and study overview.
The extent to which an increase in the variable compensation share causes stress-induced sick days and emotional exhaustion also may depend on moderators associated with salespeople's job-related resources, or aspects that "reduce job demands and the associated physiological and psychological costs" ([ 6], p. 312). Furthermore, we propose the effect of the variable compensation share on health problems to be nonlinear and mediated by salespeople's stress and effort.
Two findings from prior literature form the basis of our framework, so we offer two corresponding replication hypotheses as a starting point for our theorizing. First, incentives, including variable compensation shares, motivate salespeople to work harder and increase their sales performance ([14]; e.g., [ 9]). We predict the following:
- H1: An increase in a salesperson's variable compensation share leads to an increase in sales performance.
Second, health problems reduce salesperson productivity, so sales performance decreases ([24]). Using both sick days and emotional exhaustion as indicators of health problems, we posit:
- H2: An increase in a salesperson's (a) sick days and (b) emotional exhaustion leads to a decrease in sales performance.
To extend knowledge on the effect of a salesperson's variable compensation share, we further propose that the variable compensation share increases stress and health problems. As the basis for these predictions, we turn to COR theory, according to which people strive to develop, preserve, and protect their own resources ([34]). In this view, any expected or actual loss of resources leads people to suffer stress (e.g., [44]; [71]).
Compensation constitutes a valued resource for employees ([ 6]; [20]). An increase in the variable compensation share creates more uncertainty and according to COR theory, this uncertainty induces stress. Stress, in turn, causes health problems such as emotional exhaustion ([37]) and heightened susceptibility to illnesses (e.g., [54]). In addition, people suffering extreme stress may feel the need for a break, so they take time off ([24]; [71]).
We test these propositions in Study 1 in a natural experiment that enables us to compare a variable compensation share of 20% with a variable compensation share of 80%. In this study, we operationalize health problems as a salesperson's physician-attested sick days taken (see Figure 1). Therefore,
- H3a: An increase in variable compensation share increases sick days taken.
While this natural experiment enables us to compare a relatively low with a relatively high variable compensation share, the question arises which functional form the effect on health problems takes for variable compensation shares in between. Building on the notion that salespeople may experience increasing variable compensation shares with heightened sensitivity, we predict a progressive (J-shaped), nonlinear effect of the rising variable compensation share on health problems. Specifically, when variable compensation share increases at low levels, the salesperson should perceive low additional compensation uncertainty. Conversely, when the variable compensation share is high and thus fixed compensation is relatively low, a salesperson's total compensation might vary considerably depending on the salesperson's performance inducing stress. We test this relationship in our second study, which uses survey data, operationalizes health problems as emotional exhaustion, and includes fine-grained, continuous measurements of variable compensation share (see Figure 1). Therefore,
- H3b: An increase in variable compensation share leads to progressively greater increases in emotional exhaustion.
We argue that the extent to which salespeople experience stress and health problems due to variable compensation share depends on the availability of job-related resources to cope with these pressures. Coping refers to "cognitive and behavioral efforts to manage the internal and external demands of transactions that tax or exceed a person's resources" ([42], p. 483). We propose five such job-related resources, three related to salespeople's personal ability (i.e., prior sales performance level, prior sales performance volatility, experience) and two that reflect resources in the social work environment (i.e., team identification, leader relationship quality).
A salesperson's personal ability, defined as knowledge, skills, and experiences they possess, should give them confidence in their future performance and compensation ([39]). We consider three likely personal ability moderators: past sales performance level, performance volatility, and sales experience.
A salesperson's past performance level should serve as a reference point for predicting future sales performance and as the basis for assessing sales ability. If prior performance is relatively poor, perceived ability is likely to be low. As such, an increase in variable compensation share should lead a salesperson to perceive a greater increase in their compensation uncertainty. This in turn leads to a greater increase in health problems.
Similarly, higher performance volatility—variability in past sales revenues—should lead a salesperson to have lower confidence in their ability to perform well in the future. Again, an increase in variable compensation share should lead to a greater increase in the salesperson's compensation uncertainty, leading to stress and health problems.
Therefore, we hypothesize the following:
- H4: As a salesperson's prior sales performance level decreases, an increase in variable compensation share leads to progressively greater increases in emotional exhaustion.
- H5: As a salesperson's prior sales performance volatility increases, an increase in variable compensation share leads to progressively greater increases in emotional exhaustion.
The remaining hypotheses refer to Study 3, which like Study 2 operationalizes health problems as emotional exhaustion and employs a continuous measurement of variable compensation share. Salesperson's sales experience is another personal ability resource that should reduce the harmful effect of the variable compensation share on emotional exhaustion. The reason is that salespeople with more experience have accumulated deeper knowledge about their profession, so they should be confident in their own ability ([57]). In contrast, novice employees have fewer resources for handling the type of stress induced by a variable compensation share. Together with our previous arguments for nonlinear effects of the variable compensation share on emotional exhaustion, we propose:
- H6: As a salesperson's experience decreases, an increase in variable compensation share leads to progressively greater increases in emotional exhaustion.
Our framework examines the moderating effect of two social resources—the salesperson's team identification (i.e., degree to which a salesperson perceives oneness with a team; [72]) and leader relationship quality (i.e., degree to which a salesperson is on good terms with a superior; [50]). Such social resources help salespeople cope with stressors and should thus increase their confidence in their ability to perform well ([ 7]; [44]). As such, as a salesperson's social resources increase, an increase in variable compensation share should lead to a smaller increase in salesperson's compensation uncertainty. This, in turn, should lead to a smaller increase in stress, resulting in a smaller increase in emotional exhaustion. Thus,
- H7: As a salesperson's team identification decreases, an increase in variable compensation share leads to progressively greater increases in emotional exhaustion.
- H8: As a salesperson's relationship quality with a leader decreases, an increase in variable compensation share leads to progressively greater increases in emotional exhaustion.
Our previous hypotheses suggest that increasing variable compensation share leads salespeople with fewer job-related resources to experience greater health problems, such as emotional exhaustion. We have attributed these effects to the experience of greater uncertainty and thus more stress, in line with COR theory. To test the validity of this attribution, we posit the following:
- H9: A salesperson's stress mediates the effect of variable compensation share on emotional exhaustion.
Finally, an increasing variable compensation share is designed to incentivize salespeople to expend higher effort to increase their sales performance. However, as per the logic of COR theory, greater effort (i.e., working more hours) also depletes salespeople's resources and may increase perceived work overload and work–family conflicts, which are two well-known antecedents of emotional exhaustion ([24]). Therefore,
- H10: A salesperson's effort mediates the effect of variable compensation share on emotional exhaustion.
To test H1, H2a, and H3a, we analyzed a natural experiment with a company in Germany that reduced the variable compensation share in one of its business units (treatment BU) and kept it constant in another business unit (control BU). Both business units are located in Germany and market consumables, tools, accessories, and services for professional users in the construction and automotive industries. Salespeople, who sell products directly to customers, have traditionally had a variable compensation share of 80%, which comprised a sales commission and a bonus if they achieved sales targets. Following a shift in its philosophy regarding salesperson compensation, the treatment BU dropped the variable compensation share from 80% to 20% for all salespeople, then paid bonuses upon the achievement of individual sales targets. The compensation plans before and after the change were equivalent, promising the same total compensation in case of the achievement of sales targets (for details on compensation plan equivalency, see [40]] and [ 9]]). Critically, the treatment BU presented this change as permanent, and it has remained in place.
With this natural experiment, we can examine the effect of a reduction in the variable compensation share. We extracted data for all salespeople employed by the company on the day the treatment took effect (N = 1,028) but dropped 175 salespeople who joined the firm within six months prior to the treatment or who left the firm within six months after the treatment. We also dropped 50 salespeople for whom the company could not provide records of historical sales performance data, due to technology maintenance issues. The resulting balanced panel of 803 salespeople (401 in the treatment BU and 402 in the control BU) involved 12 months of data and 9,636 observations.
Sales performance equals the monthly sales revenue generated by a salesperson, extracted from the company's enterprise resource planning system. We divided this variable by 10,000 to improve the readability of our estimated coefficients. Furthermore, we log-transformed the variable to mitigate skewness.
We focus on the health problem indicator of sick days, because, unlike emotional exhaustion, this measure is available over time both before and after the change in the variable compensation share in the treatment BU. Thus, we can analyze the data using a difference-in-differences (DiD) method. We obtained the count of sick days from the company's human resource records. In the country of study, sick days are strictly regulated by law and require certification by a physician (at the latest on the third day of the leave). Thus, salespeople with more than three sick days in a given month are likely to have substantial health problems. Therefore, the sick days variable starts counting after the third sick day in a month, and it ranges between 0 and 24 days, with a mean value of.74 days (SD = 2.80). We log-transformed the variable to mitigate skewness. To verify that we can analyze the data using a DiD method, we ensured that average monthly sales performance and sick days exhibited parallel trends in both BUs before the change in the variable compensation share.
As is common practice in DiD analyses (e.g., [63]), we specified two dummy variables and their interaction. The first variable (treatment BU dummy) indicates the treatment BU (= 1) or control BU (= 0). The second (postperiod dummy) indicates the months in which the treatment was in place. The change to the variable compensation share in the treatment BU took place in the seventh month, so this dummy variable equals 0 for months 1–6 and 1 for months 7–12. Their multiplication (treatment BU dummy × postperiod dummy) indicates the treatment effect.
To account for heterogeneity across salespeople, we controlled for the mean values of sick days (log-transformed) and sales (log-transformed) in the six months before the treatment took effect. Furthermore, we controlled for salespeople's tenure with the business unit (measured in years). Table 2 contains the descriptive statistics and correlations.
Graph
Table 2. Studies 1–3: Descriptive Statistics, Psychometric Properties, and Correlations.
| Study 1 | V1 | V2 | V3 | V4 | V5 | V6 | V7 | | | | | | | | |
|---|
| V1: Sales performancea | — | .03 | .00 | .00 | .01 | .77 | .18 | | | | | | | | |
| V2: Sick daysa | −.16** | — | .00 | .00 | .15 | .00 | .00 | | | | | | | | |
| V3: Treatment BU dummy | −.06** | −.06** | — | .00 | .00 | .00 | .00 | | | | | | | | |
| V4: Postperiod dummy | −.04** | .05** | −.00 | — | .00 | .00 | .00 | | | | | | | | |
| V5: Pretreatment sick daysa | −.09** | .39** | −.04** | −.00 | — | .01 | .00 | | | | | | | | |
| V6: Pretreatment salesa | .88** | −.06** | −.04** | .01 | .11** | — | .26 | | | | | | | | |
| V7: Tenure | .43** | .01 | .03** | −.00 | .02* | .51** | — | | | | | | | | |
| Mean | 1.01 | .21 | — | — | .29 | 1.04 | 9.41 | | | | | | | | |
| Standard deviation | .44 | .60 | — | — | .50 | .42 | 8.35 | | | | | | | | |
| Study 2 | V1 | V2 | V3 | V4 | V5 | V6 | V7 | V8 | V9 | V10 | | | | | |
| V1: Variable compensation sharea | — | .00 | .01 | .03 | .01 | .04 | .04 | .01 | .00 | .62 | | | | | |
| V2: Emotional exhaustion | .05 | — | .01 | .00 | .06 | .00 | .00 | .02 | .05 | .00 | | | | | |
| V3: Sales performance | .12* | −.09 | — | .07 | .01 | .00 | .05 | .04 | .00 | .00 | | | | | |
| V4: Prior sales performance level | .18** | −.03 | .26** | — | .00 | .00 | .04 | .01 | .00 | .02 | | | | | |
| V5: Prior sales performance volatility | .10 | .25** | −.11* | −.07 | — | .00 | .04 | .00 | .01 | .01 | | | | | |
| V6: Genderb | −.19** | .05 | .01 | .00 | .04 | — | .00 | .05 | .00 | .04 | | | | | |
| V7: Competitive intensity | .20** | .04 | .22** | .19** | .20** | −.04 | — | .00 | .01 | .02 | | | | | |
| V8: Task variety | −.08 | −.13* | .20** | .08 | −.07 | .22** | .07 | — | .01 | .00 | | | | | |
| V9: Job tenure | −.01 | −.23** | .04 | .04 | −.08 | −.03 | .10 | .09 | — | .00 | | | | | |
| V10: Past var. compensation share | .79** | .06 | .07 | .13* | .09 | −.19** | .13* | −.05 | .00 | — | | | | | |
| Mean | 1.22 | 3.81 | 5.34 | 67.56 | 3.28 | — | 4.53 | 5.10 | 6.61 | 1.34 | | | | | |
| SD | 1.49 | 1.53 | 1.07 | 33.48 | 1.53 | — | 1.25 | 1.37 | 6.46 | 18.92 | | | | | |
| Cronbach's α | — | .93 | .91 | — | .90 | — | .79 | .96 | — | — | | | | | |
| AVE | — | .72 | .77 | — | .76 | — | .57 | .85 | — | — | | | | | |
| Study 3 | V1 | V2 | V3 | V4 | V5 | V6 | V7 | V8 | V9 | V10 | V11 | V12 | V13 | V14 | V15 |
| V1: Variable compensation sharea | — | .01 | .02 | .01 | .01 | .03 | .00 | .01 | .01 | .05 | .00 | .00 | .04 | .00 | .00 |
| V2: Stress | .09 | — | .01 | .37 | .00 | .04 | .10 | .03 | .07 | .00 | .00 | .00 | .00 | .00 | .14 |
| V3: Effort | .15* | .12 | — | .03 | .01 | .00 | .00 | .01 | .00 | .02 | .00 | .01 | .00 | .04 | .00 |
| V4: Emotional exhaustion | .11 | .61** | .17* | — | .00 | .04 | .10 | .06 | .04 | .01 | .00 | .01 | .00 | .00 | .13 |
| V5: Experience | .11 | −.00 | −.09 | −.04 | — | .02 | .02 | .00 | .01 | .00 | .02 | .00 | .00 | .00 | .00 |
| V6: Team identification | .18** | −.21** | .02 | −.19** | .13 | — | .14 | .10 | .35 | .02 | .03 | .06 | .03 | .07 | .04 |
| V7: Leader relationship quality | −.03 | −.31** | −.04 | −.32** | −.14 | .38** | — | .03 | .11 | .00 | .07 | .03 | .00 | .02 | .06 |
| V8: Job significance | .09 | −.18* | .09 | −.25** | .03 | .31** | .18* | — | .17 | .00 | .02 | .10 | .00 | .27 | .05 |
| V9: Organizational identification | .08 | −.26** | −.01 | −.19** | .12 | .59** | .33** | .41** | — | .02 | .00 | .06 | .02 | .06 | .01 |
| V10: Genderb | −.22** | −.01 | −.13 | −.08 | −.02 | −.13* | .06 | −.05 | −.14* | — | .00 | .00 | .00 | .00 | .01 |
| V11: Competitive intensity | .04 | −.04 | .03 | −.02 | −.13 | .16* | .26** | .14 | .05 | .06 | — | .01 | .00 | .04 | .01 |
| V12: Task variety | −.07 | −.02 | .08 | −.08 | .04 | .24** | .16* | .32** | .25** | −.05 | .09 | — | .00 | .24 | .01 |
| V13: Number of meetingsa | .19* | −.00 | −.04 | −.01 | −.01 | .18* | .04 | .06 | .15* | −.04 | −.05 | .07 | — | .00 | .02 |
| V14: Time pressure | .01 | .05 | .20** | .06 | .05 | .26** | .14 | .52** | .25** | .03 | .20** | .49** | −.02 | — | .00 |
| V15: Neuroticism | .01 | .38** | −.01 | .36** | .07 | −.19** | −.24** | −.23** | −.12 | .12 | −.12 | −.11 | .14 | −.01 | — |
| Mean | 3.19 | 3.43 | 47.82 | 2.56 | 9.25 | 5.65 | 6.03 | 6.14 | 6.12 | — | 5.71 | 5.79 | .96 | 5.94 | 2.81 |
| SD | 1.05 | 1.64 | 8.47 | 1.43 | 7.76 | 1.01 | 1.21 | .96 | .79 | — | 1.11 | 1.07 | .58 | .95 | 1.17 |
| Cronbach's α | — | .95 | — | .90 | — | .79 | .97 | .85 | .69 | — | .81 | .84 | — | .81 | .76 |
| AVE | — | .82 | — | .70 | — | .50 | .93 | .69 | .39 | — | .59 | .87 | — | .68 | .52 |
- 270022242921993180 *p <.05.
- 280022242921993180 **p <.01.
- 290022242921993200 a Log-transformed.
- 300022242921993200 b1 (0) for female (male).
- 310022242921993200 Notes: Numbers below the diagonal are bivariate correlations (r), numbers above indicate average variance shared (r2).
H1 and H2a propose that variable compensation share and sick days are antecedents of a salesperson's sales performance. To test these hypotheses, we initially specify a model that regresses sales performance on the treatment BU dummy, the postperiod dummy, and the interaction of the two; sick days; and controls:
Sales_performanceit=α+b1×treatment_BUi+b2×postperiodt+b3×treatment_BUi×postperiodt+ b4×sick_daysit+b5−7×controls+ϵit,1
where regression coefficients are given as b and ϵ is the error term. Variables with the subscripts i and t vary across salespeople and months, respectively. We estimate the model using STATA 16 and cluster the standard errors to account for repeated salesperson observations. The results appear in Table 3, Model 1.
Graph
Table 3. Study 1: Results of Difference-in-Differences Estimation.
| DV: Sales Performance | DV: Sick Days |
|---|
| Model 1 | Model 2 | Model 3 | Model 4 | Model 5 | Model 6 |
|---|
| Intercept | .037*** | 1.550*** | −.022 | .077*** | .046*** | .065* |
| Treatment BU dummy | −.006** | −.592*** | −.020* | −.026** | −.060*** | .012 |
| Postperiod dummy | −.020*** | −.020*** | −.016* | .092*** | .092*** | .111*** |
| Treatment BU dummy × postperiod dummy | −.030*** | −.031*** | −.034*** | −.063* | −.063* | −.082** |
| (i.e., effect of a reduction of VC share) | (H1: ✓) | (H1: ✓) | (H1: ✓) | (H3a: ✓) | (H3a: ✓) | (H3a: ✓) |
| Sick days | −.098*** | −.099*** | −.094*** | | | |
| (H2a: ✓) | (H2a: ✓) | (H2a: ✓) | | | |
| Controls | | | | | | |
| Pretreatment sick days | .053*** | | .053*** | .470*** | | .349*** |
| Pretreatment sales | .961*** | | .847*** | −.035† | | .001 |
| Tenure | −.001** | | −.007** | .001 | | .002 |
| Salesperson fixed effects | | Yes | | | Yes | |
| Inverse Mills ratio | | | .109* | | | −.018 |
| Propensity score matched data set | | | Yes | | | Yes |
| Number of observations | 9,549 | 9,549 | 7,961 | 9,636 | 9,636 | 8,040 |
| R2 | .800 | .819 | .804 | .160 | .266 | .084 |
| Adjusted R2 | .800 | .802 | .803 | .160 | .200 | .083 |
- 80022242921993200 †p <.10.
- 90022242921993200 *p <.05.
- 100022242921993200 **p <.01.
- 110022242921993200 ***p <.001.
- 120022242921993200 Notes: Two-tailed tests of significance. Unstandardized coefficients. DV = dependent variable; BU = business unit; VC = variable compensation; ✓ = supported.
In line with our prediction in H1 that a salesperson's variable compensation share fosters sales performance, the treatment effect of the reduction in variable compensation share is negative and significant (Model 2: b3 = −.030, p <.01). Furthermore, in line with H2a a salesperson's sick days decrease sales performance (b4 = −.098, p <.001).
In H3a, we argued that an increase in the variable compensation share can lead to an increase in sick days; correspondingly, the decrease in the variable compensation share examined in this natural experiment should decrease sick days. To test this prediction, we specify a second model to regress the number of monthly sick days for each salesperson on the treatment BU dummy, the postperiod dummy, the interaction of the two, and controls:
Sick_daysit=α+b1×treatment_BUi+b2×postperiodt+b3×treatment_BUi×postperiodt+b4−6×controls+ϵit.2
We estimate the model using STATA 16 and cluster the standard errors by salespeople. The results are in Table 3, Model 4. The treatment effect on sick days is negative and significant (b3 = −.063, p <.05), providing support for H3a.
The indirect effect of the treatment on sales performance via sick days is positive and significant (bindirect =.006, p <.01, 95% confidence interval [CI95%] = [.001,.010]; 5,000 bootstrap iterations). Because the direct effect of the treatment on sales performance is significant and negative (see H1), this suggests we have a case of competitive mediation ([75], p. 199). Thus, the positive indirect effect reduces 20% (= 100% ×.006/.030) of the negative direct effect ([70]). That is, the variable compensation share directly increases sales performance but indirectly decreases sales performance through more sick days. The total effect of the treatment on sales performance is significantly negative (btotal = −.024, p <.01, CI95% = [−.040, −.007]).
We conducted three steps to test the robustness of our results against endogeneity issues. First, to alleviate concerns that our estimates are biased by omitted variables, we replaced our control variables with salesperson fixed effects. Results in Table 3, Models 2 and 5, are fully in line with the results of Models 1 and 4.
Second, salespeople were not randomly assigned to the treatment and control groups. We employed genetic propensity score matching ([33]) to create comparable groups by selecting subsamples with similar values on sick days and years of tenure with the business unit. The matched data set comprises 670 salespeople (401 in the treatment BU and 269 in the control BU) and significantly improved their comparability on a Q-Q plot.
Third, the causality of the estimated effects might be questionable if the treatment led to sorting effects, prompting salespeople with more sick days to resign. Yet in the treatment BU, leavers' sick days (M =.597, SD =.981) and nonleavers' sick days (M =.480, SD = 1.048, t = −.750, p =.454) do not differ significantly, nor do leavers' sick days in the treatment BU (M =.597, SD =.981) and the control BU (M =.733, SD = 1.404, t =.565, p =.573). To further verify the robustness of the model to sorting effects, we performed Heckman's selection correction ([32]) by estimating a probit model to predict whether a salesperson left after the treatment as a function of the salesperson's business unit, mean pretreatment sick days and sales, tenure, and, as an instrument, pretreatment sales trend.[ 5] Pretreatment sales trend satisfies the exclusion restriction because it predicts whether a salesperson left the company, but it should be independent from salespeople's performance and sick days and does not have significant effects on sales performance and sick days ([ 3]; [60]). Web Appendix 1.1 contains the results of the probit model.
With the propensity score–matched data set and the inverse Mills ratio as additional predictors, we reestimated Models 1 and 4. Because the inverse Mills ratio is salesperson-specific and time-invariant, we exclude salesperson fixed effects. The results, presented as Models 3 and 6, are in line with our previous findings.[ 6] In summary, our analyses suggest that our results are not unduly influenced by endogeneity.
If the effect of the variable compensation share on sick days is due to stress, the effect should be particularly pronounced for salespeople who are more sensitive to stress. To test this proposition, we surveyed 533 of the 803 salespeople in Study 1 who still worked in the two business units (259 in the treatment BU and 274 in the control BU). We obtained 238 responses, for a response rate of 45% (91 in the treatment BU and 147 in the control BU; self-reported mean age 48 years, 98% male). Our measure reflects the notion of "choking under pressure" ([ 8]), with items adapted from research on performance anxiety under pressure ([13]): "When I feel pressure at work, I easily become nervous," "At work I react very sensitively to pressure," and "When pressure is exerted on me, I cannot work properly" (Cronbach's α =.882, average variance extracted [AVE] =.723). We reestimated Equation 2 with the main effect of sensitivity to stress, relevant two-way interaction effects, and the three-way interaction effect of sensitivity to stress, the treatment BU dummy, and the postperiod dummy. Results in Web Appendix W1.2 show that sensitivity to stress moderates the treatment effect negatively (bDiD×sens. = −.042, p <.05), suggesting that the reduction of the variable compensation share improved the health of particularly those salespeople who are sensitive to stress.
This natural experiment provides evidence of a health–performance trade-off. Companies can increase salespeople's performance by increasing variable compensation share, but this increase comes at the expense of deteriorated health (more sick days). Furthermore, sick days reduce sales performance, to such an extent that they offset the direct, positive performance effects of increasing the variable compensation share. These findings link previously disconnected research contributions regarding the performance effects and health effects of incentives and suggest their close interaction.
Study 2 aims to remedy four limitations of Study 1. First, because Study 1 analyzed a natural experiment with a variable compensation share of 80% versus 20%, we could not examine our hypothesis that variable compensation share can exhibit nonlinear effects on health problems. In Study 2, we collected a wide range of variable compensation shares to examine such nonlinear effects. Second, Study 1 used data from a single company and industry. To test the external validity of our findings, Study 2 assesses the robustness of our Study 1 findings by including a more diverse sample from different industries. Third, Study 1 did not test contingencies of the effect of variable compensation share on health problems. Study 2 tests contingencies associated with sales performance level (H4) and sales performance volatility (H5). Fourth, to further assess the robustness of our findings, Study 2 considers a wider set of control variables and validates emotional exhaustion as an indicator of health problems by linking it with specific diseases (e.g., depression, high blood pressure, migraine) and alternative illness indicators (e.g., sick days, sleep, medication, exercise).
We test the nonlinear, progressively positive effect of variable compensation share on salespeople's emotional exhaustion, as influenced by salespeople's prior sales performance level and volatility (see Figure 1) with data collected from a cross-industry sample of salespeople. An online panel provider collected initial data from 400 salespeople, but we dropped 27 observations due to failed attention checks and 13 due to missing values, resulting in a sample size of 360. The represented industries are diverse (e.g., 21% retail, 11% health care systems, 9% financial services), the gender distribution is approximately equal, and the average age of the respondents is 36.6 years; they have 6.2 years' tenure on average. There was significant heterogeneity in the compensation system design and variable compensation share across salespeople, such that their variable compensation includes bonuses and commissions tied to target achievement based on sales volume, sales revenue, or profit.
The Appendix lists all measurement items. To obtain a quasiobjective measure of the variable compensation share, we asked salespeople to provide a percentage that represents their variable compensation share ([ 1]; [35]). The mean variable compensation share in the sample is 9.64% (SD = 17.44%) with a minimum of 0% and a maximum of 90%. To measure their emotional exhaustion, we used five items from [48], as reported in the Appendix. On a 7-point scale, the salespeople report an average emotional exhaustion level of 3.81 (SD = 1.55). We developed one item to assess prior sales performance level in the past 12 months ("To what extent did you achieve your target in the past 12 months?," answered as a percentage) and three items to assess prior sales performance volatility (sample item: "In the past 12 months, I think my selling performance was very volatile"; α =.90). Our ultimate dependent variable, sales performance, was measured with three items from [36].
We include a set of control variables related to salespeople's general characteristics and task environment to reduce omitted variable bias and account for potential alternative influences. Existing research indicates that several traits can affect how employees react to organizational stressors and how they perform, so we control for salespeople job role ([15]), gender, job tenure, and the squared term of job tenure ([ 4]; [23]). Moreover, the task environment governs job demands and available job resources, so it influences the level of stress and emotional exhaustion salespeople experience ([16]). Accordingly, we include task variety (degree of job responsibility breadth) and competitive intensity (degree to which a salesperson perceives other providers as competitive) as well as past share of variable compensation as control factors ([11]; [20]). Task variety might reduce emotional exhaustion by increasing salespeople's engagement ([29]), but competitive intensity should increase emotional exhaustion, in light of the enhanced work challenges it creates ([26]). We include salespeople's past share of variable compensation because effects of stressors on salespeople can vary depending on whether they occur permanently or rather situationally ([61]).
To examine the convergent and discriminant validity of the multi-item scales, we followed standard procedures to estimate confirmatory factor analyses ([ 5]). All scales conform to prescribed values for the item reliabilities, composite reliability, and AVE. Table 3 contains the descriptive statistics and correlations.
We specified a path model as in Figure 1. First, we specify the quadratic and linear effects of variable compensation share (in the equation: VC_share) on salespeople's emotional exhaustion and the effect of salespeople's emotional exhaustion on sales performance. Second, we include the linear and quadratic interaction effects of variable compensation share and salespeople's prior sales performance level on emotional exhaustion, as well as the corresponding interactive effects of variable compensation share and prior sales performance volatility. Third, we examine the health–performance trade-off induced by variable compensation share by adding its direct effect on salespeople's sales performance. Formally,
Sales_performancei=α+b1×VC_sharei+b2×emotional_exhaustioni+b3−11×controls+ϵi,3
Emotional_exhaustioni=γ+b12×VC_sharei+b13×VC_sharei2+b14×prior_sales_performance_leveli+b15×prior_sales_performance_volatilityi+b16×VC_sharei×prior_sales_performance_leveli+b17×VC_sharei2×prior_sales_performance_leveli+b18×VC_sharei×prior_sales_performance_volatilityi+b19×VC_sharei2×prior_sales_performance_volatilityi+b20−25×controls+ϵi.4
We estimate the model in a stepwise approach using STATA 16 (see Table 4). Model 3 is our full model and the basis for our hypothesis tests. With regard to H1 and in line with previous research, we find that the variable compensation share increases salespeople's sales performance (b1 =.084, p <.05). In H2b, we proposed that salespeople's emotional exhaustion reduces sales performance, for which we find support (b2 = −.073, p <.05). In H3b, we proposed that an increase in variable compensation share leads to progressively greater increases in emotional exhaustion. In support of this hypothesis, the quadratic effect of variable compensation share on emotional exhaustion is significantly positive (b13 =.125, p <.05). These results align with and extend findings from Study 1 and support the health–performance trade-off induced by the variable compensation share.
Graph
Table 4. Study 2 Results.
| Estimated Effects | Hypotheses | Model 1 | Model 2 | Model 3 |
|---|
| Main Links in the Model | | | | |
| Variable compensation share | → Emotional exhaustion | | −.054 | −.036 | −.042 |
| Variable compensation share2 | → Emotional exhaustion | H3b: + | .114* | .153* | .125* |
| Variable compensation share | → Sales performance | H1: + | .086* | .084* | .084* |
| Emotional exhaustion | → Sales performance | H2b: − | −.066† | −.073* | −.073* |
| Main Effects of Moderators | | | | |
| Prior sales performance level | → Emotional exhaustion | | — | −.000 | −.000 |
| Prior sales performance volatility | → Emotional exhaustion | | — | .203*** | .245*** |
| Interaction Effects | | | | |
| Variable compensation share × Prior sales performance level | → Emotional exhaustion | H4: − | — | — | .002 |
| Variable compensation share2 × Prior sales performance level | → Emotional exhaustion | — | — | .001 |
| Variable compensation share × Prior sales performance volatility | → Emotional exhaustion | H5: + | — | — | .082* |
| Variable compensation share2 × Prior sales performance volatility | → Emotional exhaustion | — | — | −.013 |
| Controls | | | | |
| Gender | → Emotional exhaustion | | — | .219† | .230† |
| Job tenure | → Emotional exhaustion | | — | −.046** | −.043** |
| Job tenure2 | → Emotional exhaustion | | — | −.000 | −.000 |
| Competitive intensity | → Emotional exhaustion | | — | .039 | .045 |
| Task variety | → Emotional exhaustion | | — | −.136** | −.143** |
| Past variable compensation share | → Emotional exhaustion | | — | −.007 | −.007 |
| Gender | → Sales performance | | — | .051 | .051 |
| Job tenure | → Sales performance | | — | .003 | .003 |
| Job tenure2 | → Sales performance | | — | .000 | .000 |
| Job role dummies | → Sales performance | | — | Included | Included |
| Clustered standard errors by industries | | — | Yes | Yes |
| Intercepts | | | | |
| Sales performance | | 5.590*** | 5.294*** | 5.294*** |
| Emotional exhaustion | | 3.543*** | 3.234*** | 3.234*** |
| Model Fit | | | | |
| R2 Sales performance | | 2.2% | 6.4% | 6.4% |
| R2 Emotional exhaustion | | 1.4% | 13.2% | 14.5% |
- 320022242921993200 †p <.10.
- 330022242921993200 *p <.05.
- 340022242921993200 **p <.01.
- 350022242921993200 *** p <.001.
- 360022242921993200 Notes: Two-tailed tests of significance. Unstandardized coefficients are displayed. Gender is coded 1 (0) for female (male).
Turning to the moderating effects, we predict in H4 that increases in the variable compensation share lead to progressively greater increases in emotional exhaustion for salespeople with lower prior sales performance level. The interaction effect between the variable compensation share and prior sales performance level is insignificant, leading us to reject H4. However, in H5 we predict that increases in the variable compensation share lead to progressively greater increases in emotional exhaustion for salespeople with higher prior sales performance volatility. Results reveal that the effect of the variable compensation share on emotional exhaustion is positively moderated by salespeople's prior sales performance volatility (b16 =.082, p <.05). According to the interaction diagram and simple slopes, when prior sales performance volatility is higher, higher variable compensation share increases emotional exhaustion (b12, high share =.395, p <.05), whereas low and medium levels of this share do not (b12, low share = −.230, p >.10; b12, med. share =.083, p >.10). Figure 2 illustrates the interaction effect of variable compensation share and prior sales performance volatility.
Graph: Figure 2. Studies 2 and 3: interaction plots.Notes: For moderators, a low (high) value refers to a value of one standard deviation below (above) the mean value of the variable. The variable compensation share is plotted along the entire range of values for Study 2 and for the mean ±1 SD in Study 3. RQ = relationship quality.
We calculate conditional indirect effects at different levels of prior sales performance volatility using a bootstrapping approach with 5,000 iterations ([59]; [75]). Following the procedure suggested by [30], we initially calculated an index of moderated mediation, which serves as a quantification of whether a moderator significantly influences the strength of an indirect effect. If the index of moderated mediation is significantly different from zero, it implies that the indirect effect is significantly moderated such that "any two conditional indirect effects estimated at different values of the moderator are significantly different from each other" ([30], p. 2). In our model, the index of moderated mediation (IMM) is negative and significant at the 90% confidence interval, indicating that higher levels of performance volatility may exacerbate the harmful indirect effect of variable compensation share on sales performance through elevated emotional exhaustion (IMM = −.006, CI90% = [−.012, −.0003]). This enables us to inspect the pattern of effects in more detail. First, the direct effect of variable compensation share on sales performance is positive and significant (bdirect =.084, CI95% = [.018,.158]). However, with increasing variable compensation share and prior performance volatility, the total effect of variable compensation share on sales performance diminishes. That is, at a variable compensation share as well as prior sales performance volatility of one standard deviation above the mean, variable compensation share exhibits a marginally significant negative indirect effect on sales performance through emotional exhaustion (bindirect = −.029, CI90% = [−.090, −.002]). This negative indirect effect reduces 35% (= 100% ×.029/.084) of the positive direct effect.
To examine associations of salespeople's emotional exhaustion with additional indicators of health problems, we included a broad set of alternative, established indicators of health and disease. In line with existing health-related research, salespeople's emotional exhaustion significantly correlates with a diverse set of physical and mental health issues. For example, emotional exhaustion is correlated with depression (r =.187, p <.001) and migraines (r =.168, p <.001) (see Web Appendix W2.1 for details).
Beyond the initial set of controls, we check the stability of our findings and reduce omitted variable bias by considering a broad range of incentive systems and contextual controls that capture different facets of salespeople's specific incentive (e.g., incentive time horizons) and family (e.g., number of children) conditions. The model estimation with this expanded set of controls remains stable and the interpretation of the hypothesized effects does not change (Web Appendix W2.2).
While our focal health-related outcome in this study is emotional exhaustion, a variable compensation share might trigger further mechanisms that link to health problems and sick days taken. Web Appendix W2.3 tests four of such mechanisms associated with reciprocity, perceived entitlement, and share of nonincentivized tasks (vs. incentivized tasks) as alternative mechanisms. We find that the hypothesized effects from variable compensation share on emotional exhaustion and from emotional exhaustion on sick days remain fully supported when we include these alternative explanations.
In line with our theorizing, Study 2 indicates a progressive effect of the variable compensation share on salespeople's emotional exhaustion, contingent on their prior sales performance volatility. Salespeople's prior sales performance level did not moderate this relationship. A possible explanation is that high-performing salespeople may feel more confident in their selling abilities and thus less stressed by a variable compensation share, but they also may realize that more is at stake for them, which could increase their stress (e.g., [ 2]). Furthermore, Study 2 confirms that the variable compensation share induces a trade-off between salespeople's health and sales performance. Finally, Study 2 shows that emotional exhaustion relates to different mental and physical diseases, as well as sick days, and tests the robustness of our findings against a broad set of controls and alternative mediating mechanisms.
Study 3 aims to test the moderating influence of salespeople's experience and social resources on the progressive effect of the variable compensation share, as described in H6–H8, and the corresponding mediation effects through stress (H9) and effort (H10). For these purposes, we surveyed salespeople of an international business-to-business supplier whose core competence lies in engineering services and the sale of construction accessories and solutions. The 185 experienced salespeople in the Study 3 sample are assigned to 89 sales managers with an average professional experience of 9.25 years, and 95% are men. The average variable compensation share was 33.55% (SD = 22.18%), and their mean emotional exhaustion score was 2.57 (SD = 1.44) on a 7-point scale. The salespeople worked in 43 countries, including the United States (22 observations); Brazil, Chile, and Germany (each with 13 observations); and a range of other countries (with fewer than 10 observations each). The compensation systems included bonuses and commissions tied to individual sales and profit targets.
As in Study 2, salespeople's variable compensation share represents the key independent variable, which we measured quasiobjectively ([ 1]; [35]). Our key dependent variable is emotional exhaustion ([48]). We capture the mediating variables related to salespeople's stress and effort with established scales (e.g., [12]).
Salespeople's experience (H6) is measured as the number of years in their current position. For team identification (H7), we adopted a scale from [47], which we slightly adjusted to the firm's context (see Appendix; [72]). The measure of leader relationship quality (H8) is based on [55].
As in Study 2, we included control variables related to the task environment and salespeople's general characteristics. For the task environment, we consider time pressure, task variety, number of meetings, and competitive intensity as additional control factors ([20]) that likely shape the job demands employees face and the resources available to them. Specifically, time pressure and number of meetings likely threaten available time resources, which might exacerbate stress and emotional exhaustion ([34]). Regarding salespeople's general characteristics, we control for gender, work division, experience, and neuroticism. The latter measure reflects emotional stability and is a predictor of susceptibility to stress ([67]). We also include the squared term of experience, because stress and emotional exhaustion appear to vary with salespeople's different experience levels ([10]; [23]). Furthermore, we control for salespeople's organizational identification (degree to which a salesperson perceives oneness with their employer; [19]) and job significance (degree to which a salesperson perceives their job as important for their company), spanning both the linear and quadratic interaction effect of variable compensation share with job significance ([73]). Job significance and organizational identification may be psychologically empowering and thereby establish key psychological resources that employees can leverage to reduce their stress and emotional exhaustion ([29]). The Appendix contains all the measures, and Table 2 displays the descriptive statistics, correlations, and reliability diagnostics. All constructs were discriminant according to the [27] criterion. Given all the variables were gathered from the same survey, we took several steps to reduce the risk of common method bias (see Web Appendix W3).
The structural equation model, depicted in Figure 1, includes the direct effects of the variable compensation share (VC_share) on salespeople's stress and effort and then the effect of these two variables on emotional exhaustion, controlling for the direct effect of variable compensation share on emotional exhaustion. To assess the nonlinear, progressive effect of variable compensation share, we included its squared term. We also specified interactive effects of the moderators through their multiplicative terms with the linear and quadratic values of variable compensation share. We thus estimate the following system of equations:
Emotional_exhaustioni=α+b1×stressi+b2×efforti+b3×VC_sharei+b4×VC_sharei2+b5×experiencei+b6×team_identificationi+b7×leader_relationship_qualityi+b8−22×controls+ϵi,5
Stressi=γ+b23×VC_sharei+b24×VC_sharei2+b25×efforti+b26×experiencei+b27×team_identificationi+b28×leader_relationship_qualityi+b29×VC_sharei×experiencei+b30×VC_sharei×team_identificationi+b31×VC_sharei×leader_relationship_qualityi+b32×VC_sharei2×team_identificationi+b33×VC_sharei2×leader_relationship_qualityi+b34×VC_sharei2×experiencei+b35−51×controls+ϵi,6
Efforti=δ+b52×VC_sharei+b53×experiencei+b54×team_identificationi+b55×leader_relationship_qualityi+b56−71×controls+b74+ϵi.7
To estimate the specified model, we used Mplus 8.0 and a maximum likelihood estimator with robust standard errors (Table 5, Model 1). It exhibits a good fit with the data (comparative fit index =.94; standardized root mean residual =.01; root mean square error of approximation =.06; χ2/d.f. = 1.74). To account for the nesting of salespeople within sales managers and countries we employed the "sandwich estimator" approach that computes adjusted standard errors and test statistics while accounting for the dependence of individual salesperson responses within clusters ([53]). Specifically, Model 2 in Table 5 accounts for the nesting in sales managers, and Model 3 accounts for the nesting in both sales managers and countries. The latter model constitutes the basis for the following hypothesis testing and bootstrapping approach.
Graph
Table 5. Study 3 Results.
| Estimated Effects | Model 1 | Model 2 | Model 3 |
|---|
| Main Relationships in Model | | | |
| Variable compensation share | → Stress | .194† | .194 | .194† |
| Variable compensation share2 | → Stress | .273* | .273* | .273* |
| Variable compensation share | → Effort | .147* | .147† | .147† |
| Stress | → Emotional exhaustion | .489*** | .489*** | .489*** |
| Effort | → Emotional exhaustion | .122* | .122* | .122** |
| Effort | → Stress | .099 | .099 | .099 |
| Main Effects of Moderators | | | |
| Experience | → Stress | .109 | .109 | .109 |
| Team identification | → Stress | .072 | .072 | .072 |
| Leader relationship quality | → Stress | −.090 | −.090 | −.090 |
| Experience | → Effort | −.243* | −.243* | −.243* |
| Team identification | → Effort | −.014 | −.014 | −.014 |
| Leader relationship quality | → Effort | −.081 | −.081 | −.081 |
| Interaction Effects | | | |
| Variable compensation share × experience | → Stress | −.198* | −.198* | −.198* |
| Variable compensation share2 × experience | → Stress | −.157 | −.157 | −.157 |
| Variable compensation share × team identification | → Stress | −.233* | −.233* | −.233* |
| Variable compensation share2 × team identification | → Stress | −.216* | −.216* | −.216* |
| Variable compensation share × leader relationship quality | → Stress | −.194† | −.194† | −.194† |
| Variable compensation share2 × leader relationship quality | → Stress | −.281† | −.281* | −.281* |
| Controlled Effects | | | |
| Gender | → Stress | −.110 | −.110 | −.110 |
| Competitive intensity | → Stress | .055 | .055 | .055 |
| Task variety | → Stress | −.003 | −.003 | −.003 |
| Division dummies | → Stress | Included | Included | Included |
| Time pressure | → Stress | .122 | .122 | .122 |
| Number of meetings | → Stress | −.093 | −.093 | −.093 |
| Neuroticism | → Stress | .243** | .243** | .243** |
| Experience2 | → Stress | −.134 | −.134 | −.134 |
| Organizational identification | → Stress | −.113 | −.113 | −.113 |
| Job significance | → Stress | −.331** | −.331** | −.331** |
| Variable compensation share × job significance | → Stress | .427** | .427** | .427** |
| Variable compensation share2 × job significance | → Stress | .521* | .521* | .521* |
| Gender | → Effort | −.076 | −.076 | −.076 |
| Competitive intensity | → Effort | .050 | .050 | .050 |
| Task variety | → Effort | .047 | .047 | .047 |
| Division dummies | → Effort | Included | Included | Included |
| Time pressure | → Effort | .175 | .175 | .175 |
| Number of meetings | → Effort | .026 | .026 | .026 |
| Experience2 | → Effort | .135 | .135 | .135 |
| Neuroticism | → Effort | −.004 | −.004 | −.004 |
| Organizational identification | → Effort | .022 | .022 | .022 |
| Job significance | → Effort | −.018 | −.018 | −.018 |
| Gender | → Emotional exhaustion | −.080* | −.080† | −.080* |
| Competitive intensity | → Emotional exhaustion | −.016 | −.016 | −.016 |
| Task variety | → Emotional exhaustion | −.052 | −.052 | −.052 |
| Division dummies | → Emotional exhaustion | Included | Included | Included |
| Time pressure | → Emotional exhaustion | .090 | .090 | .090 |
| Number of meetings | → Emotional exhaustion | −.085 | −.085 | −.085 |
| Neuroticism | → Emotional exhaustion | .092 | .092 | .092 |
| Organizational identification | → Emotional exhaustion | .066 | .066 | .066 |
| Job significance | → Emotional exhaustion | −.174* | −.174* | −.174** |
| Variable compensation share | → Emotional exhaustion | .193* | .193* | .193* |
| Variable compensation share2 | → Emotional exhaustion | .206* | .206* | .206* |
| Experience | → Emotional exhaustion | −.053 | −.053 | −.053 |
| Experience2 | → Emotional exhaustion | .046 | .046 | .046 |
| Team identification | → Emotional exhaustion | −.074 | −.074 | −.074 |
| Leader relationship quality | → Emotional exhaustion | −.110† | −.110† | −.110 |
| Intercepts | | | |
| Emotional exhaustion | .107 | .107 | .107 |
| Effort | 5.766*** | 5.766*** | 5.766*** |
| Stress | 1.565** | 1.565** | 1.565** |
| Model Fit | | | |
| R2 Emotional exhaustion | 49.5% | 49.5% | 49.5% |
| R2 Effort | 17.6% | 17.6% | 17.6% |
| R2 Stress | 35.0% | 35.0% | 35.0% |
- 460022242921993200 †p <.10.
- 470022242921993200 *p <.05.
- 480022242921993200 **p <.01.
- 490022242921993200 ***p <.001.
- 500022242921993200 Notes: Two-tailed tests of significance. Standardized coefficients are displayed. Gender is coded 1 (0) for female (male).
We propose indirect effects of variable compensation share on emotional exhaustion through salespeople's stress and effort (H6–H10). The indirect effect through stress is further contingent on the levels of the proposed moderators. To verify H6–H8, we follow established recommendations by [65] and [59] and calculate conditional indirect effects using a bootstrapping approach based on 5,000 iterations. That is, we bootstrap the standard errors of parameter estimates and estimate confidence intervals for the indirect and direct effects.
Table 5 contains the results of the model estimation and Table 6 reports results of the conditional indirect effects of variable compensation share on emotional exhaustion through stress (H6–H8). Figure 2 illustrates the interaction effects.
Graph
Table 6. Study 3: Bootstrapped Conditional Indirect Effects of Variable Compensation Share on Emotional Exhaustion via Stress.
| | Variable Compensation Share |
|---|
| Hypothesis | Moderator Valuesa | Low | Medium | High |
|---|
| H6 | Low experience | .001 [−.106,.108] | .192* [.022,.362] | .383* [.062,.704] |
| High experience | −.067 [−.248,.114] | −.003 [−.089,.084] | .062 [−.132,.256] |
| H7 | Low team identification | −.027 [−.179,.126] | .214** [.068,.360] | .455** [.165,.744] |
| High team identification | −.039 [−.163,.085] | −.024 [−.168,.119] | −.010 [−.259,.240] |
| H8 | Low leader relationship quality | −.064 [−.173,.046] | .182** [.053,.310] | .427** [.123,.730] |
| High leader relationship quality | −.002 [−.134,.130] | .008 [−.144,.160] | .018 [−.256,.292] |
- 510022242921993200 * Significant at the 95% CI.
- 520022242921993200 ** Significant at the 99% CI.
- 530022242921993200 a High/low values at mean ±1 SD.
- 540022242921993200 Notes: Standardized coefficients and 95% confidence intervals are reported, based on 5,000 iterations.
As in Study 2, before inspecting the detailed pattern of indirect effects, we estimated indexes of moderated mediation to verify whether the proposed moderators experience, team identification, and leader relationship quality significantly affect the indirect effect of variable compensation share on emotional exhaustion ([30]). The indices of moderated mediation, based on 5,000 bootstrapped iterations, are negative and significant indicating, in line with our expectations, that salespeople's mental and social resources may buffer harmful effects of variable compensation share on emotional exhaustion (IMMexperience = −.022, CI95% = [−.046, −.002]; IMMteam identification = −.227, CI99% = [−.51, −.035]; IMMrelationship quality = −.157, CI90% = [−.298, −.021]).
Drawing on these IMMs, we examined conditional indirect effects at different moderator values. With regard to H6, the bootstrapped indirect effects confirm that at low levels of salesperson experience and variable compensation share, the indirect effect on emotional exhaustion is insignificant (βindirect =.001, CI95% = [−.106,.108]). At low levels of salesperson experience and medium or high levels of variable compensation share, the indirect effects are positive and significant (medium: βindirect =.192, CI95% = [.022,.362]; high: βindirect =.383, CI95% = [.062,.704]). When salesperson experience is high, the indirect effects on emotional exhaustion are insignificant at all levels of variable compensation share (Table 6).
We also predicted that increases in variable compensation share lead to greater increases in emotional exhaustion, especially for salespeople with low team identification and poor relationships with their leaders. In support of H7, at low levels of team identification and low variable compensation share, the indirect effect is insignificant, but at medium and high variable compensation share levels, they are positive and significant (medium: βindirect =.214, CI95% = [.068,.360]; high: βindirect =.455, CI95% = [.165,.744]). When team identification is high, the indirect effects on emotional exhaustion are insignificant at all levels of variable compensation share (Table 6). In support of H8, poor leader relationship quality and low variable compensation share reveals an insignificant indirect effect, but at medium and high levels, the indirect effects are positive and significant (medium: βindirect =.182, CI95% = [.053,.310]; high: βindirect =.427, CI95% = [.123,.730]). If relationship quality is high, the indirect effects are insignificant (Table 6).
Moreover, H9 predicted that salespeople's stress mediates the increasingly positive effect of the variable compensation share on emotional exhaustion. To verify H9, we had estimated conditional indirect effects of variable compensation share on emotional exhaustion via stress, contingent on the level of salespeople's experience, team identification, and leader relationship quality. In support of H9, we find that variable compensation share significantly increases emotional exhaustion through elevated stress, at high variable compensation shares, and if experience is low (bindirect =.383, CI95% = [.062,.704]), team identification is low (bindirect =.455, CI95% = [.165,.744]), or leader relationship quality is low (bindirect =.427, CI95% = [.123,.730]). Finally, with regard to salespeople's effort, variable compensation share marginally increases effort (β =.147, p =.057), and effort significantly increases salespeople's emotional exhaustion (β =.122, p =.009), but the corresponding indirect effect is only marginally significant (βindirect =.018, CI90% = [.001,.035]), so we only partially confirm H10. The harmful effects of variable compensation share on emotional exhaustion appear to manifest mainly through stress, and to a smaller extent through increased levels of effort.
With Study 3, we replicate the nonlinear, progressive effect of variable compensation share on salespeople's emotional exhaustion and assess salespeople's experience and social resources as contingencies of this effect. As a complement to Study 2, we verify resource-related contingencies and thereby validate the conceptual framework based on COR theory. The results indicate that resource-related moderators can buffer the harmful effects of variable compensation share, in accordance with the basic tenets of COR theory. Furthermore, we confirm the nonlinear, progressive relationship between the variable compensation share and salespeople's emotional exhaustion, governed by salespeople's resource endowment. Finally, given Study 3 uses data from more than 40 countries and uses a more complex solution context, it further establishes the external validity of our findings.
Managers often rely on variable compensation shares to increase salespeople's performance, and this strategy seems to work ([66]). However, our results show that this incentive also can create health problems that reduce the hoped-for performance benefits, particularly for salespeople with weak job-related resources. The average effect of the variable compensation share on sales performance is positive, so managers would be ill-advised to eliminate it. Rather, to practice better marketing for a better world, they should incentivize sales teams through variable compensation shares while mitigating any health problems those incentives create. We propose strategies for newly hired staff (i.e., ex ante strategies) and for existing staff (i.e., ex post strategies).
If a company's variable compensation share is high, managers should carefully screen salespeople and sales supervisors before hiring them. For example, while interviewing salespeople and reviewing references, managers might probe the stability of their past performance, their experience, and their tendency to build relationships with leaders and peers. If these resources are lacking or unobservable (e.g., for first-time employees), managers might screen for other stress-related resources, such as strong personal resilience or social networks. When hiring sales supervisors, managers also should screen for these applicants' willingness to help salespeople cope with stress and the ability to build strong relationships with and among team members.
Among existing staff, first, managers should help salespeople build their job-related resources. For example, by encouraging salespeople to manage their sales pipeline for a steady stream of sales, managers can help them reduce performance volatility. With regard to social resources, companies should train supervisors to adopt leadership techniques related to relationship and community building and encourage supportive networks among salespeople, such as through team-building events.
Second, if legally and culturally possible, managers could personalize incentive schemes (e.g., [ 9]; [66]). They could assign a high variable compensation share to salespeople with high job-related resources who prefer extrinsic rewards but limit this share for more stress-vulnerable salespeople.
Third, companies concerned with their salespeople's stress and health should sensitize their managers to the health-harming effects of their compensation decisions. In an experimental follow-up Study 4 (see Web Appendix W4), we examined decisions about a sales team's variable compensation share when the participating managers were aware or unaware of the link to stress. These participants chose lower variable compensation shares if they were made aware of the stress induced by their decisions, particularly if they scored high on the empathy dimension of perspective taking ([51]). Although this follow-up study involves an online simulation, it provides tentative evidence that companies can implement better marketing for a better world by making the unintended consequences of managers' decisions transparent.
Our study contributes to several literature streams. First, literature on sales force incentives traditionally has focused on economic performance outcomes (e.g., [ 9]; [14]). To the best of our knowledge, this study is the first to address how variable compensation shares can lead to stress, emotional exhaustion, and sick days among salespeople varying in job-related resources. Our findings highlight that variable compensation plans include hidden costs that threaten to offset the very objective for which they are employed ([21]). This finding lays the groundwork for future research. For example, how do variable compensation plans affect stress, health, and productivity in the long term? Do different sales force incentives moderate these effects differently (e.g., bonuses vs. commissions, all-or-nothing vs. proportional payment, monetary vs. nonmonetary incentives)? How do team incentives affect salespeople's stress? Research could also consider how framing of compensation plans influences stress responses (e.g., $100,000 with 20% variable pay vs. $80,000 with an opportunity for a $20,000 bonus).
Second, we confirm studies of the health-related outcomes of pay-for-performance plans (e.g., [17]; [18]; [56]) in the specific context of sales and salespeople ([28]). This confirmation is important, because researchers have questioned whether the influence of pay-for-performance plans on health are sufficiently important to be studied at all ([58]). Our results provide evidence that whether and how a variable compensation share induces health problems is worth studying, particularly if the marketing discipline aims to contribute to a better world. Furthermore, we extend knowledge about these health-related outcomes by identifying nonlinear, moderated, and mediated effects. Continued research can draw on these findings to build more complete models of health-related outcomes of pay-for-performance plans.
Finally, our study contributes to literature on employee stress and burnout. The emergence of stress can be explained by several resource-related, such as cognitive appraisal ([43]) and job demands–resources ([ 6]) theories. These theories assert that perceptions of stress arise from environmental stressors and that people cope with stressors by leveraging resources. Future research could more systematically integrate variable compensation shares as a potential stressor into these theories.
Supplemental Material, sj-docx-1-jmx-10.1177_0022242921993195 - Variable Compensation and Salesperson Health
Supplemental Material, sj-docx-1-jmx-10.1177_0022242921993195 for Variable Compensation and Salesperson Health by Johannes Habel, Sascha Alavi and Kim Linsenmayer in Journal of Marketing
Graph
Appendix Survey Items for Studies 2 and 3.
| Study 2 | Study 3 |
|---|
Variable Compensation Sharea (based on Alavi et al. [2018])According to your current employment contract, what percentage of your total remuneration is variable (e.g., paid out as commissions, bonuses, profit sharing), given 100% target achievement?Emotional Exhaustion (based on Maslach and Jackson [1981])bI feel emotionally drained from my work. I feel fatigued when I get up in the morning and have to face another day on the job. I feel burned out from my work. I feel frustrated by my job. I feel used up at the end of the workday. Sales Performance (based on Homburg, Müller, and Klarmann [2011])bHow do you evaluate your current level of sales performance in comparison with your colleagues?The achieved sales? The achieved orders? The achieved contribution margin? Prior Sales Performance Level (self-developed)aTo what extent did you achieve your target in the past 12 months?Prior Sales Performance Volatility (self-developed)bIn the past 12 months, I think that my selling performance was very volatile. In the past 12 months, my selling performance fluctuated a lot. In the past 12 months, my level of selling performance was varying strongly. GendercWhat is your sex?Competitive Intensity (based on Jaworski and Kohli1993)bWith respect to my sales district, competition in our industry is cutthroat. With respect to my sales district, price competition is a hallmark of our industry. With respect to my sales district, our competitors are quite strong. Task Variety (based on Morgeson and Humphrey [2006])bThe job involves a great deal of task variety. The job involves doing a number of different things. The job requires the performance of a wide range of tasks. The job involves performing a variety of tasks. Job TenureaHow many years of professional experience do you have in your current position?Job RoledWhich description best matches your role in your job?Past Variable Compensation Sharea (based on Alavi et al. [2018])Think of the past FIVE YEARS: What was the average percentage of your total remuneration that was variable (e.g., payed out as commissions, bonuses, profit sharing) given 100% target achievement? | Variable Compensation Sharea (based on Alavi et al. [2018])According to your current employment contract, what percentage of your total compensation is variable?Stress (based on Loerbroks et al. [2010])bMy working environment puts a strain on me. My working environment stresses me. My working environment is a source of stress for me. My working environment saps my energy. EffortaHow many hours do you work in a typical working week in total?Emotional Exhaustion (based on Maslach and Jackson [1981])bI feel emotionally drained from my work. I feel fatigued when I get up in the morning and have to face another day on the job. I feel burned out from my work. I feel frustrated by my job. ExperienceaHow many years have you been working in your current job (at [company] or other firms)?Team Identification (based on Mael and Ashforth [1992])bWhen someone criticizes the colleagues of my department, it feels like a personal insult. I am very interested in what others think about the colleagues in my department. When I talk about the colleagues in my department, I usually say "we" rather than "they." I identify with my department. Leader Relationship Quality (based on Palmatier et al. [2007])bMy relationship with my direct superior is good. I consider my relationship with my direct superior as positive. I get along well with my direct superior.
| Job Significance (based on Hackman and Oldham [1975])bMy job tasks are important for [company]'s company success. My job tasks are meaningful for [company]'s company success. My job tasks are significant for [company]'s company success. Organizational Identification (based on Mael and Ashforth [1992])bWhen someone criticizes [company], it feels like a personal insult. I am very interested in what others think about [company]. When I talk about [company], I usually say "we" rather than "they". I identify with [company]. Competitive Intensity (based on Jaworski and Kohli [1993])bWith respect to my sales district, competition in our industry is cutthroat. With respect to my sales district, price competition is a hallmark of our industry. With respect to my sales district, our competitors are quite strong. Task Variety (based on Hackman and Oldham [1975])bMy job tasks offer a lot of variety. My job tasks are multifaceted. My job tasks are far from monotonous. Number of MeetingsaHow many meetings with colleagues or superiors in person do you have each day, approximately?Time Pressure (self-developed)bIn my job, time is short. Time is a valuable aspect in my job. Time is an important factor in my job. Neuroticism (based on Donnellan et al. [2006])bI do not get upset easily.r I am relaxed most of the time.r My mood is usually stable over time.r
|
1 aOpen-ended text field.
- 2 bSeven-point scale from "strongly disagree" to "strongly agree."
- 3 c Single choice ("male," "female").
- 4 d Single choice ("salesperson," "sales manager," "sales clerk", "sales consultant," "service employee," "consultant," and "other").
- 5 r Reverse-coded.
Footnotes 1 Ajay Kohli
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Online supplement: https://doi.org/10.1177/0022242921993195
5 To calculate pretreatment sales trend, we regressed each salesperson's monthly sales before the treatment on a time count variable and then extracted each salesperson's regression coefficient for the time count variable. We divided this regression coefficient by a salesperson's mean pretreatment sales, thus obtaining a relative trend measure.
6 As before, the indirect effect of the treatment on sales performance via sick days is positive and significant (bindirect =.008, p <.01, CI95% = [.003,.012]; 5,000 bootstrap iterations) and reduces the direct negative effect of the treatment on sales performance by 24% (= 100% ×.008/.034). The total effect of the treatment on sales performance is significantly negative (btotal = −.026, p <.01, CI95% = [−.045, −.008]).
References Alavi Sascha, Habel Johannes, Guenzi Paolo, Wieseke Jan. (2018), "The Role of Leadership in Salespeople's Price Negotiation Behavior," Journal of the Academy of Marketing Science, 46 (4), 703–24.
Arrondel Luc, Duhautois Richard, Laslier Jean-François. (2019), "Decision Under Psychological Pressure: The Shooter's Anxiety at the Penalty Kick," Journal of Economic Psychology, 70, 22–35.
Atefi Yashar, Ahearne Michael, Hohenberg Sebastian, Hall Zachary, Zettelmeyer Florian. (2020), "Open Negotiation: The Back-End Benefits of Salespeople's Transparency in the Front End," Journal of Marketing Research, 57 (6), 1076–94.
Avey James B., Luthans Fred, Smith Ronda M., Palmer Noel F. (2010), "Impact of Positive Psychological Capital on Employee Well-Being over Time," Journal of Occupational Health Psychology, 15 (1), 17–28.
Bagozzi Richard P., Yi Youjae. (1989), "On the Use of Structural Equation Models in Experimental Designs," Journal of Marketing Research, 26 (3), 271–84.
Bakker Arnold B., Demerouti Evangelia. (2007), "The Job Demands–Resources Model: State of the Art," Journal of Managerial Psychology, 22 (3), 309–28.
7 Bakker Arnold B., Demerouti Evangelia, Euwema Martin C. (2005), "Job Resources Buffer the Impact of Job Demands on Burnout," Journal of Occupational Health Psychology, 10 (2), 170–80.
8 Baumeister Roy F. (1984), "Choking Under Pressure: Self-Consciousness and Paradoxical Effects of Incentives on Skillful Performance," Journal of Personality and Social Psychology, 46 (3), 610–20.
9 Bommaraju Raghu, Hohenberg Sebastian. (2018), "Self-Selected Sales Incentives: Evidence of Their Effectiveness, Persistence, Durability, and Underlying Mechanisms," Journal of Marketing, 82 (5), 106–24.
Brewer Ernest W., Shapard Leslie. (2004), "Employee Burnout: A Meta-Analysis of the Relationship Between Age or Years of Experience," Human Resource Development Review, 3 (2), 102–23.
Brown Steven P., Peterson Robert A. (1993), "Antecedents and Consequences of Salesperson Job Satisfaction: Meta-Analysis and Assessment of Causal Effects," Journal of Marketing Research, 30 (1), 63–77.
Brown Steven P., Peterson Robert A. (1994), "The Effect of Effort on Sales Performance and Job Satisfaction," Journal of Marketing, 58 (2), 70–80.
Cheng Wen-Nuan Kara, Hardy Lew, Markland David. (2009), "Toward a Three-Dimensional Conceptualization of Performance Anxiety: Rationale and Initial Measurement Development," Psychology of Sport and Exercise, 10 (2), 271–78.
Chung Doug J., Steenburgh Thomas, Sudhir K. (2014), "Do Bonuses Enhance Sales Productivity? A Dynamic Structural Analysis of Bonus-Based Compensation Plans," Marketing Science, 33 (2), 159–314.
Churchill Gilbert A.Jr, Ford Neil M., Hartley Steven W., Walker Orville C.Jr. (1985), "The Determinants of Salesperson Performance: A Meta-Analysis," Journal of Marketing Research, 22 (2), 103–18.
Crawford Eean R., Pine Jeffery A. Le, Rich Bruce Louis. (2010), "Linking Job Demands and Resources to Employee Engagement and Burnout: A Theoretical Extension and Meta-Analytic Test," Journal of Applied Psychology, 95 (5), 834–48.
Dahl Michael S., Pierce Lamar. (2020), "Pay-for-Performance and Employee Mental Health: Large Sample Evidence Using Employee Prescription Drug Usage," Academy of Management Discoveries, 6 (1), 12–38.
Davis Mary E. (2016), "Pay Matters: The Piece Rate and Health in the Developing World," Annals of Global Health, 82 (5), 858–65.
Deckop John R., Mangel Robert, Cirka Carol C. (1999), "Getting More than You Pay for: Organizational Citizenship Behavior and Pay-for-Performance Plans," Academy of Management Journal, 42 (4), 420–28.
Demerouti Evangelia, Bakker Arnold B., Nachreiner Friedhelm, Schaufeli Wilmar B. (2001), "The Job Demands–Resources Model of Burnout," Journal of Applied Psychology, 86 (3), 499–512.
DeVaro Jed, Heywood John S. (2017), "Performance Pay and Work-Related Health Problems: A Longitudinal Study of Establishments," ILR Review, 70 (3), 670–703.
Donnellan M. Brent, Oswald Frederick L., Baird Brendan M., Lucas Richard E. (2006), "The Mini-IPIP Scales: Tiny-yet-Effective Measures of the Big Five Factors of Personality," Psychological Assessment, 18 (2), 192–203.
Dunford Benjamin B., Shipp Abbie J., Boss Wayne R., Angermeier Ingo, Boss Alan D. (2012), "Is Burnout Static or Dynamic? A Career Transition Perspective of Employee Burnout Trajectories," Journal of Applied Psychology, 97 (3), 637–50.
Edmondson Diane R., Matthews Lucy M., Ambrose Scott C. (2019), "A Meta-Analytic Review of Emotional Exhaustion in a Sales Context," Journal of Personal Selling & Sales Management, 39 (3), 275–86.
Eisenberger Robert, Aselage Justin. (2009), "Incremental Effects of Reward on Experienced Performance Pressure: Positive Outcomes for Intrinsic Interest and Creativity," Journal of Organizational Behavior, 30 (1), 95–117.
Fletcher Thomas D., Major Debra A., Davis Donald D. (2008), "The Interactive Relationship of Competitive Climate and Trait Competitiveness with Workplace Attitudes, Stress, and Performance," Journal of Organizational Behavior, 29 (7), 899–922.
Fornell Claes, Larcker David F. (1981), "Evaluating Structural Equation Models with Unobservable Variables and Measurement Error," Journal of Marketing Research, 18 (1), 39–50.
Ganster Daniel C., Kiersch Christa E., Marsh Rachel E., Bowen Angela. (2011), "Performance-Based Rewards and Work Stress," Journal of Organizational Behavior Management, 31 (4), 221–35.
Hackman J. Richard, Oldham Greg R. (1975), "Development of the Job Diagnostic Survey," Journal of Applied Psychology, 60 (2), 159–70.
Hayes Andrew F. (2015), "An Index and Test of Linear Moderated Mediation," Multivariate Behavioral Research, 50 (1), 1–22.
Hays Ron D., Sherbourne Cathy Donald, Mazel Rebecca M. (1993), "The RAND 36-Item Health Survey 1.0," Health Economics, 2 (3), 217–27.
Heckman James J. (1979), "Sample Selection Bias as a Specification Error," Econometrica, 47 (1), 153–62.
Ho Daniel E., Imai Kosuke, King Gary, Stuart Elizabeth A. (2011), "MatchIt: Nonparametric Preprocessing for Parametric Causal Inference," Journal of Statistical Software, 42 (8), 21603.
Hobfoll Stevan E. (2001), "The Influence of Culture, Community, and the Nested-Self in the Stress Process: Advancing Conservation of Resources Theory," Applied Psychology, 50 (3), 337–421.
Homburg Christian, Klarmann Martin, Reimann Martin, Schilke Oliver. (2012), "What Drives Key Informant Accuracy?" Journal of Marketing Research, 49 (4), 594–608.
Homburg Christian, Müller Michael, Klarmann Martin. (2011), "When Should the Customer Really Be King? On the Optimum Level of Salesperson Customer Orientation in Sales Encounters," Journal of Marketing, 75 (2), 55–74.
Jaramillo Fernando, Mulki Jay Prakash, Boles James S. (2013), "Bringing Meaning to the Sales Job: The Effect of Ethical Climate and Customer Demandingness," Journal of Business Research, 66 (11), 2301–07.
Jaworski Bernard J., Kohli Ajay K. (1993), "Market Orientation: Antecedents and Consequences," Journal of Marketing, 57 (3), 53–70.
John George, Weitz Barton. (1989), "Salesforce Compensation: An Empirical Investigation of Factors Related to Use of Salary Versus Incentive Compensation," Journal of Marketing Research, 26 (1), 1–14.
Kishore Sunil, Rao Raghunath Singh, Narasimhan Om, John George. (2013), "Bonuses Versus Commissions: A Field Study," Journal of Marketing Research, 50 (3), 317–33.
Kraft Sheryl. (2018), "Companies Are Facing an Employee Burnout Crisis," CNBC (August 14), https://www.cnbc.com/2018/08/14/5-ways-workers-can-avoid-employee-burnout.html.
Latack Janina C., Havlovic Stephen J. (1992), "Coping with Job Stress: A Conceptual Evaluation Framework for Coping Measures," Journal of Organizational Behavior, 13 (5), 479–508.
Lazarus R.S., Folkman Susan. (1984), Stress, Appraisal, and Coping. New York: Springer.
LePine Marcie, Zhang Yiwen, Crawford Eean R., Rich Bruce Louis. (2016), "Turning Their Pain to Gain: Charismatic Leader Influence on Follower Stress Appraisal and Job Performance," Academy of Management Journal, 59 (3), 1036–59.
Lim Noah, Ahearne Michael J., Ham Sung H. (2009), "Designing Sales Contests: Does the Prize Structure Matter?" Journal of Marketing Research, 46 (3), 356–71.
Loerbroks A., Gadinger Michael C., Bosch Jos A., Stürmer Til, Amelang Mafred. (2010), "Work-Related Stress, Inability to Relax after Work and Risk of Adult Asthma: A Population-Based Cohort Study," Allergy, 65 (10), 1298–305.
Mael Fred, Ashforth Blake E. (1992), "Alumni and Their Alma Mater: A Partial Test of the Reformulated Model of Organizational Identification," Journal of Organizational Behavior, 13 (2), 103–23.
Maslach Christina, Jackson Susan E. (1981), "The Measurement of Experienced Burnout," Journal of Organizational Behavior, 2 (2), 99–113.
Maslach Christina, Schaufeli Wilmar B., Leiter Michael P. (2001), "Job Burnout," Annual Review of Psychology, 52 (1), 397–422.
Maslyn John M., Uhl-Bien Mary. (2001), "Leader–Member Exchange and Its Dimensions: Effects of Self-Effort and Other's Effort on Relationship Quality," Journal of Applied Psychology, 86 (4), 697–708.
McBane Donald A. (1995), "Empathy and the Salesperson: A Multidimensional Perspective," Psychology & Marketing, 12 (4), 349–70.
Morgeson Frederick P., Humphrey Stephen E. (2006), "The Work Design Questionnaire (WDQ): Developing and Validating a Comprehensive Measure for Assessing Job Design and the Nature of Work," Journal of Applied Psychology, 91 (6), 1321–39.
Muthén L.K., Muthén B.O. (2017), Mplus User's Guide (8th ed.). Los Angeles: Muthén and Muthén.
Nixon Ashley E., Mazzola Joseph J., Bauer Jeremy, Krueger Jeremy R., Spector Paul E. (2011), "Can Work Make You Sick? A Meta-Analysis of the Relationships Between Job Stressors and Physical Symptoms," Work and Stress, 25 (1), 1–22.
Palmatier Robert W., Scheer Lisa K., Houston Mark B., Evans Kenneth R., Gopalakrishna Srinath. (2007), "Use of Relationship Marketing Programs in Building Customer–Salesperson and Customer–Firm Relationships: Differential Influences on Financial Outcomes," International Journal of Research in Marketing, 24 (3), 210–23.
Parker Stacey L., Bell Katrina, Gagné Marylene, Carey Kristine, Hilpert Thomas. (2019), "Collateral Damage Associated with Performance-Based Pay: The Role of Stress Appraisals," European Journal of Work and Organizational Psychology, 28 (5), 691–707.
Peterson Robert A. (2020), "Self-Efficacy and Personal Selling: Review and Examination with an Emphasis on Sales Performance," Journal of Personal Selling & Sales Management, 40 (1), 57–71.
Pfeffer Jeffrey. (2019), "'Crying Wolf': A Comment on Dahl and Pierce and a Suggestion on Using (Danish) Prescription Data," Academy of Management Discoveries, 6 (1), 137–39.
Preacher Kristopher J., Rucker Derek D., Hayes Andrew F. (2007), "Addressing Moderated Mediation Hypotheses: Theory, Methods, and Prescriptions," Multivariate Behavioral Research, 42 (1), 185–227.
Puhani Patrick. (2000), "The Heckman Correction for Sample Selection and Its Critique," Journal of Economic Surveys, 14 (1), 53–68.
Ryari Hanaa, Alavi Sascha, Wieseke Jan. (2020), "Drown or Blossom? The Impact of Perceived Chronic Time Pressure on Retail Salespeople's Performance and Customer–Salesperson Relationships," Journal of Retailing(published online June 24), https://doi.org/10.1016/j.jretai.2020.05.005.
Schaubroeck John, Merritt Deryl E. (1997), "Divergent Effects of Job Control on Coping with Work Stressors: The Key Role of Self-Efficacy," Academy of Management Journal, 40 (3), 738–54.
Schmitz Christian, Friess Maximilian, Alavi Sascha, Habel Johannes. (2020), "Understanding the Impact of Relationship disruptions," Journal of Marketing, 84 (1), 66–87.
Shirom Arie, Westman Mina, Melamed Samuel. (1999), "The Effects of Pay Systems on Blue-Collar Employees' Emotional Distress: The Mediating Effects of Objective and Subjective Work Monotony," Human Relations, 52 (8), 1077–97.
Shrout Patrick E., Bolger Niall. (2002), "Mediation in Experimental and Nonexperimental Studies: New Procedures and Recommendations," Psychological Methods, 7 (4), 422–45.
Steenburgh Thomas, Ahearne Michael. (2012), "Motivating Salespeople: What Really Works," Harvard Business Review, 90 (7/8), 70–75.
Swider Brian W., Zimmerman Ryan D. (2010), "Born to Burnout: A Meta-Analytic Path Model of Personality, Job Burnout, and Work Outcomes," Journal of Vocational Behavior, 76 (3), 487–506.
Timio Mario, Gentili Simonetta. (1976), "Adrenosympathetic Overactivity Under Conditions of Work Stress," Journal of Epidemiology & Community Health, 30 (4), 262–65.
Timio Mario, Gentili Simonetta, Pede Sergio. (1979), "Free Adrenaline and Noradrenaline Excretion Related to Occupational Stress," British Heart Journal, 42 (4), 471–74.
Van Heerde Harald J., Moorman Christine, Moreau C. Page, Palmatier Robert W. (2021), "Reality Check: Infusing Ecological Value into Academic Marketing Research," Journal of Marketing, 85 (2), 1–13.
Van Woerkom Marianne, Bakker Arnold B., Nishii Lisa H. (2016), "Accumulative Job Demands and Support for Strength Use: Fine-Tuning the Job Demands–Resources Model Using Conservation of Resources Theory," Journal of Applied Psychology, 101 (1), 141–50.
Wieseke Jan, Kraus Florian, Ahearne Michael, Mikolon Sven. (2012), "Multiple Identification Foci and Their Countervailing Effects on Salespeople's Negative Headquarters Stereotypes," Journal of Marketing, 76 (3), 1–20.
Xie Jia L., Johns Gary. (1995), "Job Scope and Stress: Can Job Scope Be Too High?" Academy of Management Journal, 38 (5), 1288–309.
Yeh Wan-Yu, Cheng Yawen, Chen Chiou-Jung. (2009), "Social Patterns of Pay Systems and Their Associations with Psychosocial Job Characteristics and Burnout Among Paid Employees in Taiwan," Social Science & Medicine, 68 (8), 1407–15.
Zhao Xinshu, Lynch John G.Jr, Chen Qimei. (2010), "Reconsidering Baron and Kenny: Myths and Truths About Mediation Analysis," Journal of Consumer Research, 37 (2), 197–206.
~~~~~~~~
By Johannes Habel; Sascha Alavi and Kim Linsenmayer
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 133- Virtual Reality in New Product Development: Insights from Prelaunch Sales Forecasting for Durables. By: Harz, Nathalie; Hohenberg, Sebastian; Homburg, Christian. Journal of Marketing. May2022, Vol. 86 Issue 3, p157-179. 23p. 1 Black and White Photograph, 1 Diagram, 7 Charts, 2 Graphs. DOI: 10.1177/00222429211014902.
- Database:
- Business Source Complete
Virtual Reality in New Product Development: Insights from Prelaunch Sales Forecasting for Durables
This investigation examines how consumer durable goods producers can leverage virtual reality for new product development. First, the authors develop a prelaunch sales forecasting approach with two key features: virtual reality and an extended macro-flow model. To assess its effectiveness, the authors collect data from 631 potential buyers of two real-world innovations. The results reveal that the new approach yields highly accurate prelaunch forecasts across the two field studies: compared with the actual sales data tracked after the product launches, the prediction errors for the aggregated first-year sales are only 1.9% (Study 1a, original prelaunch sales forecast),.0% (Study 1b, forecast with actual advertisement spending), and 20.0% (Study 1b, original prelaunch forecast). Moreover, the average mean absolute percentage error for the monthly sales is only 23% across both studies. Second, to understand the mechanisms of virtual reality, the authors conduct a controlled laboratory experiment. The findings reveal that virtual reality fosters behavioral consistency between participants' information search, preferences, and buying behavior. Moreover, virtual reality enhances participants' perceptions related to presence and vividness, but not their perceptions related to alternative theoretical perspectives. Finally, the authors provide recommendations for when and how managers can use virtual reality in new product development.
Keywords: innovative durables; new product development; sales forecasting; virtual reality
Introducing a new consumer durable product to the market is an important but risky endeavor ([31]). Research thus far provides rich insights regarding various aspects of new product development (NPD) that help firms manage these risks. For instance, investigators have examined customer integration during NPD ([14]) and drivers of new product success ([35]). However, knowledge on how to leverage new technologies such as virtual reality to improve NPD is scarce, despite their high potential ([31]) and several recent calls for such research ([50]; [70]).
The paucity of research on virtual reality use in NPD is especially surprising in a consumer durables context, as virtual reality technologies have improved significantly in terms of their visualization and automated-tracking capabilities, and they can now elicit and track extensive consumer experiences ([16]). Leveraging these technological improvements could help durables producers perform their NPD tasks. For instance, because the market success of durables may strongly depend on consumers' experiences ([48]), better simulation and capture of these experiences could improve prelaunch product assessments ([37]) and help firms decrease the high failure rates of durables, which range between 40% and 90% ([26]; [42]). In addition, evidence indicates that extensive experiences can be created and assessed in virtual reality, even for products that do not yet exist as firms only need a virtual blueprint or 3D model of the new product ([36]; [62]). Thus, durables producers could get consumer insights earlier in the NPD process by relying on virtual reality, resulting in advantages such as better aligned production and commercialization plans or cost reductions ([11]; [51]).
For these reasons, we examine how consumer durable goods producers can utilize virtual reality—a simulated environment that allows the consumer to interact with it ([49])—to improve their NPD. We focus on virtual reality for prelaunch sales forecasting and then discuss implications for other NPD tasks. In particular, we examine the following:
- RQ1: (How) does virtual reality improve prelaunch sales forecasting?
- RQ2: Why do virtual reality simulations (vs. traditional studio tests with real products) lead to advantages in forecasting?
To examine RQ1, we developed a new prelaunch sales forecasting approach that employs virtual reality, together with an extended macro-flow model. We tested the new approach in two field studies (Studies 1a and 1b) with two real-world innovations that our collaborating companies actually introduced. In total, a representative sample of 631 potential buyers participated in the field studies. Using one year of actual sales data provided by GfK, a leading European market research agency, we validated our prelaunch sales forecasts after the launch of the two innovative durables. The results show that the new approach achieves high prelaunch forecasting accuracy with errors for the aggregated first-year sales forecasts of 1.9% (Study 1a),.0% (Study 1b, forecast with actual advertisement spending), and 20.0% (Study 1b, original prelaunch forecast) as well as a mean absolute percentage error (MAPE) of 18.6% (Study 1a), 26.5% (Study 1b, forecast with actual advertisement spending), and 44.4% (Study 1b, original prelaunch forecast) for the monthly sales forecasts across the two studies. In addition, a comparison of our forecasts with the strongest benchmark model demonstrates that this new forecasting approach improves monthly forecasting accuracy by over 30%. In a supplemental analysis, we furthermore approximated that a substantial part of this improvement can be attributed to the virtual reality key feature of the approach.
To address RQ2, because we could not determine from the field studies why virtual reality contributes to sales forecasting accuracy, we conducted a laboratory experiment (Study 2, n = 210). Participants were randomly assigned to one of three conditions: ( 1) lab virtual reality, ( 2) online virtual reality, or ( 3) a studio test with real products. The findings reveal that the forecasting advantages of virtual reality emerged because virtual reality participants on average behaved more consistently in terms of their information search, preferences, and purchase behavior within the simulation than the studio test participants. The results also indicate that these differences can be explained by increased presence (i.e., the extent to which participants feel they are actually in the simulation) and vividness (i.e., the extent to which participants feel that the simulation is detailed and easy to imagine), but not through alternative theoretical lenses, such as decision uncertainty and convenience.
Overall, our approach and findings advance the understanding of whether and how virtual reality can improve sales forecasting (RQ1: Studies 1a and 1b) and of the reasons why these advantages arise (RQ2: Study 2). In concluding, we discuss the studies' implications for future research—including a road map for future marketing studies on better prognoses, diagnostics, and customer linking through virtual reality ([19])—and nonacademic audiences—including an action plan for the use of virtual reality in NPD.
Researchers have focused on forecasting models for new durables that are applied close to launch ([ 7]; [25]). However, firms need a precise sales forecast early in NPD to improve their investment decisions ([51]) and to align their production, marketing, and distribution plans ([11]). Although knowledge is lacking on the prelaunch sales forecasting of durables, sophisticated models for consumables exist that achieve high forecasting accuracy by using a trial-and-repeat logic ([22]; [63]). However, these models are not applicable to durables, which are purchased infrequently and are therefore not subject to trial-and-repeat purchases ([29]). We address this research gap and develop a new prelaunch sales forecasting approach for consumer durables with two key features: virtual reality and an extended macro-flow model.
The use of a virtual reality simulation for the new approach was motivated by virtual reality's visualization and automated-tracking capabilities. We explain these two capabilities and then discuss the virtual reality types that leverage these capabilities to a varying extent (Table 1).
Graph
Table 1. Differences and Similarities of Physical Product Experience, Virtual Reality, and Multimedia Approaches.
| Capability | | Physical Product Experience | Virtual Reality | Multimedia Approaches |
|---|
| Studio Test with Real Products in a Shelf(e.g.,Castro, Morales, and Nowlis 2013; Basis for Benchmarkin Study 2) | Lab Virtual Reality(This Investigation:Studies 1a, 1b, 2) | Online Virtual Reality(This Investigation:Studies 1a, 1b, 2) | Information Acceleration(Urban et al. 1997; Basis for Benchmark in Studies 1a and 1b) | Virtual Shelf Simulation(Burke 1996) | Web-Animated Prototypes(Dahan and Srinivasan 2000) |
|---|
| Visualization | Simulation scope | MediumFocused on a shelf due to limited studio space with many different products and detailed information on products | Very HighComprehensive environment (e.g., entire local retail store) with many different products and detailed information on products | Very HighComprehensive environment (e.g., entire local retail store) with many different products and detailed information on products | MediumSections of environment with new product and one competitor product with detailed information on the two products | LowFocused on a virtual shelf with different products and some information on products | LowWebpage showing different products |
| Similarity to reality | MediumStanding in front of a shelf with real products | Very High3D view of the environment and products from 360 degrees and first-person perspective based on 360-degree pictures of actual setting | High2D view of the environment and products from 360 degrees and first-person perspective based on 360-degree pictures of actual setting | Medium2D view of sections of environment and two products from different angles based on pictures and/or videos of actual setting | Low2D view of programmed virtual shelf incl. products from different angles | Low2D view of animated products viewed |
| Immersion | MediumSenses with regard to product interactions:Use of own eyes for seeing the shelf with the products and hands for interactions | Very HighSenses with regard to product and environment interactions:Use of a head-mounted display to see the whole store and products, use of and controllers for immersing user's motions and interactions | HighSenses with regard to product and environment interactions:Use of a computer to see the whole store and products, use of keyboard and mouse for immersing user's interactions | MediumSenses with regard to product and sections of environment interactions:Use of a computer to see sections of environment and two products, use of keyboard and mouse for interactions | MediumFocused on senses with regard to product interactions:Use of a computer to see the shelf and different products, use of keyboard and mouse for interactions | LowFocused on seeing the product:Use of a computer to see different animated products |
| Tracking of behavior | Interactivity | High Interactivity, Tracking Through ObservationWalk through store excerpt (i.e., shelf) Explore different products placed on shelf Pick products up from shelf and turn them to see them from all angles Put products in basket
| High Interactivity, Automated TrackingWalk through store Explore different products placed on shelf Pick products up from shelf and turn them to see them from all angles Put products in basket Go to checkout and purchase any product
| High Interactivity, Automated TrackingWalk through store Explore different products within the store Pick products up from shelf and turn them to see them from all angles Put products in basket Go to checkout and purchase any product
| Medium Interactivity, Tracking Was Not ImplementedChange to predefined positions to view the product from different angles Ask predefined questions regarding the product
| Medium Interactivity, Automated TrackingExplore different products placed on shelf Pick products up from shelf and turn them to see them from all angles Put products in basket for purchase
| Low Interactivity, No TrackingReplay the animation/video of the product
|
By "visualization capability," we refer to the ability to simulate new products, customer touchpoints, and environments in a comprehensive, realistic, and engaging manner. Drawing on previous work ([16]; [49]; [52]; [68]), we theorized that the visualization capability would be valuable for forecasting: if consumers are immersed into the simulation and the simulation is comprehensive and similar to reality, they might be more likely to act throughout the simulation as they would in a real durable purchasing situation. As Table 1 shows, virtual reality has a higher visualization capability than previous multimedia approaches or studio tests used for prelaunch forecasting. This advantage can be attributed to virtual reality's simulation scope, similarity to reality, and immersion.
First, virtual reality can span a very high simulation scope. This attribute is due to its ability to simulate all facets of purchase journeys with many different products and detailed depictions of various information sources and shopping environments, such as online shops or local retail stores ([28]; [45]). In building on these insights, we expected that, compared with previous multimedia approaches and studio tests, virtual reality would motivate more realistic consumer behavior due to the increased liveliness of the simulation ([52]; [61]). This anticipation is because, to reduce costs and complexity, prior approaches relied on more focused and less detailed simulations, such as displaying the new product on a shelf alongside only a few competitors (Table 1).
Second, virtual reality can achieve a (very) high similarity to reality. This attribute is due to its ability to showcase the simulated shopping environment and products from 360 degrees and first-person perspective based on 360-degree pictures of the actual setting. Thus, in line with previous research in related fields ([28]), we expected that virtual reality would increase similarity to reality compared with previous multimedia approaches and physical product experiences, such as studio tests (Table 1). This anticipation is because previous approaches relied on either third-person perspectives via pictures, image galleries, and infomercials or on simulating parts of reality (e.g., a shelf with physical products or only the products) in a less genuine manner.
Third, virtual reality simulations can achieve a (very) high immersion of consumers. This attribute exists because, using additional virtual reality equipment (e.g., head-mounted displays, motion controllers), it can deeply transport consumers into a simulation by stimulating their senses intensively ([66]). In contrast, previous approaches would either immerse fewer senses or focus on consumers' immersion related to the product (but not the environment) (Table 1). Thus, we expected that virtual reality (vs. previous approaches) would motivate more realistic behavior by further increasing consumers' perceived transportation into the simulation.
Automated-tracking capability is the ability to directly collect data of consumers' interactions with the simulation. The recorded data can encompass the type of interaction performed by the consumer (e.g., entering a simulated environment, operating a product in the use environment) as well as the duration of each interaction. We theorized that an automated-tracking capability would be valuable for sales forecasting: prelaunch sales forecasting models could integrate such behavioral data as inputs, which might increase forecasting accuracy ([52]; [67]).
Virtual reality has an automated-tracking capability advantage over previous approaches, which relied on observations to capture consumers' actions and focused on collecting data on interactions with the products (Table 1). In contrast, due to sophisticated computer-based simulation technology and motion-tracking sensors, virtual reality offers additional options to interact with both the simulated environment and the products ([16]). Moreover, because these interactions occur digitally, they can be directly tracked ([16]; [28]).
Virtual reality simulations essentially fall into two main categories: lab virtual reality and online virtual reality. Lab virtual reality requires data gathering in a central location, as it uses additional equipment for viewing and interacting with the simulated environment (e.g., head-mounted displays, sensory input devices, power walls). This equipment allows lab virtual realities to offer many highly intuitive interaction possibilities ([16]). For instance, head-mounted displays automatically react to the user's head movements and change the perspective and picture. Moreover, hand movements can be detected by motion-tracking sensors or virtual reality controllers and mimic and perform these actions within the simulation. Thus, using gesture control, consumers within a lab virtual reality can move around freely in the simulation, pick up items from shelves, and operate or buy products (e.g., through hand movements).
In contrast, an online virtual reality can be accessed from anywhere via the internet as it is displayed on a computer and does not need additional equipment. Owing to sophisticated graphics and 360-degree views from the first-person perspective, online virtual realities can realistically depict products and environments at comparatively low costs ([ 8]). However, compared with lab virtual realities, online virtual realities stimulate fewer senses and provide less intuitive interaction possibilities (e.g., clicking on a mouse vs. motion-tracking) while still offering high degrees of interactivity ([16]). For instance, in online virtual reality, consumers can freely navigate in any direction within the simulated environment by clicking in that direction, and the position and perspective changes automatically. Participants can select different products with the cursor, pick a product up by clicking on it, and then view it from 360 degrees, put it back on the shelf, or buy it as they would in a real store.
Thus, as Table 1 shows, lab virtual realities and online virtual realities are very similar in terms of their automated-tracking capability but differ with respect to their visualization capability. Both lab and online virtual realities can achieve similarly high levels of simulation scope and interactivity, but lab virtual realities are likely to create more immersive and realistic environments, due to the usage of additional virtual reality equipment. Figure 1 provides visuals and screenshots that illustrate both virtual reality types. See the subsection "Studies 1a and 1b: Research Setting, Data, and Analyses" for a depiction on how we use both virtual reality types in field studies.
Graph: Figure 1. Visuals and screenshots of virtual reality environments.Notes: *Due to confidentiality agreements, we are not allowed to show the real products and their actual functionalities. Therefore, we show placeholders instead.
Macro-flow models forecast the sales of new products over time before their market launch ([69]). In contrast, alternative forecasting models either do not allow for a prediction over time (e.g., conjoint analysis) or cannot be applied prior to the launch (e.g., diffusion models). To derive a sales forecast over time, macro-flow models predict the flows of potential buyers from one state of the purchase journey to the next. To make these predictions prelaunch, they require a purchase journey simulation ([68]). Macro-flow models consist of three elements: behavioral states, the flows between these states, and the determinants of these flows (Figure 2).
Graph: Figure 2. Behavioral states, flows, and determinants in the extended macro-flow model.Notes: The behavioral states shown on the left are simplified. For a detailed depiction, see Figure W1.1 in Web Appendix.
Behavioral states are the attitudes and actions through which consumers flow in their purchase journey ([69]). As Figure 2 shows, we extended previous macro-flow models in the awareness, information search, and purchase decision state. For creating awareness, previous macro-flow models considered word of mouth (WOM) and advertisements ([69]; [68]). To adjust for the presence of the internet (e.g., social media, blogs) and its impact on consumer behavior ([71]), we added awareness via third-party information. For information search, previous macro-flow models included a showroom visit, a test drive ([68]), or a consumer magazine and recommendations from others ([67]). To capture consumers' information search more broadly ([45]), our extended macro-flow model includes five frequently used information sources: a local retail store, an online shop, the producer's website, an online magazine, and a use environment. Finally, previous macro-flow models captured the purchase decision state in terms of buying either the new product or a competitor product, thus forcing participants to make a decision. In contrast, to account for the possibility that consumers delay their purchase or refrain from buying ([29]; [65]), we added a no-purchase option.
Flows between the states are defined as the proportion of consumers transitioning from one behavioral state to another ([58]; [69]). All flows build on logit and negative exponential models ([33]). We focus the description below on the new and updated flows but summarize all equations in Table 2.
Graph
Table 2. Overview of Flows in the Extended Macro-Flow Model.
| Flow | Rationale | Sourcea |
|---|
| (1) | The probability of becoming aware of the new product via WOM of previous buyers depends on the number of previous buyers and the susceptibility to WOM. | Urban, Hauser, and Roberts (1990) |
| (2) | The probability of becoming aware of the new product via advertisements depends on the advertisement spending and the susceptibility to advertisements. | Urban, Hauser, and Roberts (1990) |
| (3) , where | The probability of becoming aware of the new product via third-party information depends on the time since announcement of the new product and the number of similar products. | New |
| (4) where | The overall probability of becoming aware depends on the awareness via WOM, ads, and third-party information and has an upper boundary of 100%.b | New |
| (5) | The intensity of information search depends on the actions, time, and visits spent on each product at each information source. | New |
| (6) | The preference for the new product depends on the points given to the new product compared with the points given to the other products in the relevant set during the preference measurement. | New, based on Luce (1959) |
| (7). | The probability of buying the new product depends on the preference for it and the buying behavior during the virtual reality simulation. | New |
| (8) | The overall purchase probability depends on the awareness, information search, and the probability of buying the new product. | New, based on Urban, Hauser, and Roberts (1990) |
| (9) | The sales of the new product depend on the overall purchase probability for it and the market size of the product category. | Urban, Hauser, and Roberts (1990) |
1 a For differences and similarities across macro-flow models, see Web Appendix W1.
2 b From a theoretical point of view, the equation of awareness needs to subtract consumers, who become aware through two or more ways. To simplify the equation, we omitted this subtraction because even if the consumer becomes aware through multiple ways, the maximum awareness will still be 100%, which is why we introduced the upper boundary of 100%.
The first new flow is the probability of becoming aware via third-party information (Equation 3, Table 2). We model this flow, , dependent of the amount of buzz for the product. Specifically, buzz factor Bt computes as two years[ 5] minus the time since the announcement divided by the number of similar products, which accounts for decreasing relevance of buzz over time ([71]) and a decrease in buzz as the number of similar products grow ([37]). Integrating this extension with the two components of previous macro-flow models (Equation 4, Table 2) allows for calculation of the overall probability of awareness, .[ 6]
The next updated flow is from awareness to information search (Equation 5, Table 2). As the time and effort in the information search allocated to particular products indicates purchase interest in those products ([ 9]), we model the flow from awareness to information search, P(Information Search), for each consumer for each product depending on the number of actions, visits, and time spent at each information source.
The next new flow is from information search to the purchase decision. This flow, P(Behavior), is calculated as the mean of two components (Equation 7, Table 2): preferences and the observed virtual purchases. We model the preference of each participant, P(Preference), by asking participants to divide a fixed number of points according to their preferences between the products they would consider buying ([46]; [47]).[ 7] We observe each participant's virtual purchase decision, P(Virtual Reality), in the simulation and assign different purchasing probability values to it (i.e., 0 if no product or a competitor is bought,.2 if the product is on the wish list, and.7 if the product is bought), guided by previous findings on conversion schemes for purchase intentions ([13]; [41]). We calculated several sensitivity checks for these choices (Web Appendix W2) and found that forecasting accuracy only slightly decreases if alternative choices based on competing rationales are made.
We calculate the overall purchase probability for the new product by combining all states through multiplication following the logic of macro-flow modeling (Equation 8, Table 2). However, we also adjust the overall purchase probability of each participant depending on their intensity of information search for the new product relative to the strongest competitors as in the end of the purchase decision consumers commonly select between the top two products ([43]). Finally, to calculate the sales of the new durable at time (t), we multiply the mean of the purchase probability of all participants for the new durable with the market size of the product category (Equation 9, Table 2).
The determinants of flows are the independent variables that drive the flows between the behavioral states of the macro-flow model ([69]). We used the same determinants as [69] for the adopted flows and new determinants for all newly introduced or updated flows (see Figure 2).
We implemented our new forecasting approach in two large-scale field studies (Studies 1a and 1b) collaborating with two different durables' producers planning to launch new products. For data collection, we also collaborated with GfK to enlist participants from GfK's representative retail panels. Following standard market research procedures, we employed a screener to identify a representative sample of consumers in the market willing to buy a product in this product category (Web Appendix W3).
For both field studies, the purchase journey simulation started by creating awareness, as awareness is the necessary precondition for consumers' subsequent purchase journey behaviors, and because the new durable had not been announced to the market ([ 3]). We created awareness by showing all participants actual marketing stimuli provided by the company launching the new product. Next, participants searched for information on the products in the purchase journey simulation. The simulation ended when participants bought any of the displayed products or they ended the simulation without making a purchase ([20]). For both field studies, we captured a virtual purchase decision because an actual preordering of the new products was not yet possible. After the simulation, participants allocated a fixed number of points among the products in a preference measurement ([47]). Finally, we conducted a survey with questions on additional input parameters (e.g., susceptibility to WOM) for the extended macro-flow model.
The purchase journey was simulated in virtual reality. Because our collaboration partners required accurate and cost-effective results, we used both types of virtual reality: a lab virtual reality (via a head-mounted display with virtual reality controllers and motion-tracking sensors for interactions) and an online virtual reality (via a computer screen with mouse and keyboard for interactions). Both types showed the same products in exactly the same environments (i.e., local retail store, online shop, producers' webpages, online magazine, and use environment) and had exactly the same study flow (i.e., awareness creation, free information search, and the purchase decision). In addition, both types tracked identical individual-level behavioral data (Table 3): the type and number of actions per product at each information source, the time per action and product at each information source, and the visits per information source and product.
Graph
Table 3. Data Automatically Tracked in the Lab and Online Virtual Reality Simulations.
| Information Source | Variables Automatically Tracked in Virtual Reality Environments |
|---|
| Actions(Number per Product) | Time(Per Action and Product) | Visits(Per Information Source and Product) |
|---|
| Local retail store | Product picked from shelf Product viewed (in 360 degrees) Product returned to shelf Product put in the basket Product removed from basket Product purchased
| Total time in local retail store Total time per product Time of viewing each product (in 360 degrees)
| Number of visits at local retail store Number of visits per product
|
| Online shop | Product selected for detail view Product viewed (in 360 degrees) Returned to overview page Search bar used Filters used Sorting used Product put on wish list Product removed from wish list Product put in the basket Product removed from basket Product purchased
| Total time in online shop Total time per product Time of viewing each product (in 360 degrees)
| Number of visits at online shop Number of visits per product
|
| Producers' webpages | Product selected for detail view Product viewed (in 360 degrees) Product put in the basket Product removed from basket Product purchased
| Total time on producers' webpages Total time on each producer's webpage Total time per product
| Number of visits at each producer's webpage Number of visits per product
|
| Online magazine | Viewed product review
| Total time on online magazine Total time per product
| Number of visits at online magazine Number of visits per product
|
| Use environment | Product picked up/selected Product operated Product deselected
| Total time in use environment Total time per operated product
| Number of visits at use environment Number of visits per product
|
Study 1a forecasts the sales of a new kitchen appliance. The market for this appliance is extremely competitive even though the product is highly priced. Thus, we agreed with the collaborating producer to include 18 products in the simulation to ensure appropriate market coverage. In total, 305 potential buyers participated in Study 1a, with 80 in the lab and 225 in the online virtual reality. Participants' average age was 45 years, with 45% women. Study 1b forecasts the sales of a highly innovative gardening tool. Because of the tool's novelty, we decided with the collaborating producer that the simulation would include nine competitor products with varying prices, from the product category most similar to the new tool's area of application. Study 1b had 326 participants with an average age of 48 years and 50% women. In total, 101 participants completed the lab and 225 participants the online virtual reality simulation. We used the formulas summarized in Table 2 to calculate the forecasts in both studies. We report details on estimated flows in Web Appendix W4.
Figure 3 presents the sales forecasts calculated prior to launch for Studies 1a and 1b. We organize the aggregated results according to their focus on external validity (i.e., forecasts vs. actual sales and additional external validations) and internal validity (i.e., benchmark macro-flow model, other benchmarks, model variations, and comparison of virtual reality types) assessments of our new virtual reality forecasting approach.
Graph: Figure 3. Studies 1a and 1b: sales forecasts compared with actual sales after launch.
To assess the accuracy of our new forecasting approach, we compared the prelaunch sales forecasts with the actual monthly sales of the respective new durable (i.e., aggregated number of units sold in one calendar month), as recorded by GfK through its point-of-sales scanning. To assess first-year model performance, we summed the monthly sales over the first 12 months after launch.
The results show that the prelaunch sales forecast (vs. actual sales data) in Study 1a achieves high forecasting accuracy with a MAPE for the monthly sales forecast of 18.6% and a forecasting error of only 1.9% for aggregated first-year sales. For Study 1b, the MAPE for the monthly sales forecast is 44.4%, and the forecast deviates from aggregated first-year sales by 20.0%, which is less accurate, mainly owing to deviations in months 9–12 (Figure 3). As our collaboration partner pointed out, these deviations were caused by the company's decision to increase ad spending after initial sales were below expectations. To reflect the company's decision in a recalculated sales forecast, we used its actual ad spending for months 9–12 as input (Equation 2 in Table 2). The forecast with the actual ad spending for months 9–12 is also shown in Figure 3 and achieves a high forecasting accuracy (i.e., 26.5% MAPE over 12 months and.0% forecasting error for aggregated first-year sales).
Because, in both studies, the sales forecasting accuracy varies in different periods, we analyzed when and why deviations occur. We identified three factors that caused such deviations: external influences (e.g., unusual weather conditions that affected sales of gardening tools, as in Study 1b), problems during the product launch (e.g., supply chain issues, as in Study 1a, where sales started two weeks later than planned), and changes in marketing plans (e.g., an increase in ad spending, as in Study 1b). As outlined in Web Appendix W5, aggregating sales for longer time spans (e.g., quarterly), relying on confidence intervals, and calculating an updated forecast can remedy these deviations.
In addition, we assessed longer-term predictions, the speed of adoption, and early uses of our new sales forecasting approach (Web Appendix W6). The results show that the approach can be used for longer-term predictions, as we demonstrate by validating a forecast for 22 months with actual sales data from Study 1a. The results reveal that this longer-term forecast is also highly accurate, with a MAPE between 14% and 23% for the monthly sales. Moreover, the cumulative sales forecast can be used to assess the speed of adoption and identify the takeoff ([24]). In addition, the results show how the approach can be used early in NPD if a virtual blueprint of the new product exists: either by calculating confidence intervals, as in the real-world application, or by making scenario-based predictions.
To validate our new sales forecasting approach's extended macro-flow model, we benchmarked it with its strongest precursor "information acceleration" ([67]). To ensure comparability, the benchmark macro-flow model uses the same behavioral states, flows, and determinants from management judgment, survey data, and virtual reality simulations as our new sales forecasting approach. However, we changed one component: the data used for estimating the purchase probability. Instead of using preferences and virtual purchases, we use the percentage of people who intended to buy the new product, as measured in the survey after the purchase journey simulation. This choice permits a prediction more in line with information acceleration and previous macro-flow models ([69]; [67]).
The sales forecast derived from the benchmark macro-flow model is also depicted in Figure 3 and has an average MAPE over the first 12 months after introduction of 49.6% in Study 1a and 74.4% in Study 1b (forecast with actual ad spending: 67.2%). It is, on average, more than 30% less accurate than the sales forecasts derived from the extended macro-flow model (Study 1a: 18.6%; Study 1b: 44.4% [forecast with actual ad spending: 26.5%]). This comparison of the extended macro-flow model with the benchmark macro-flow model, which still uses virtual reality and all other amendments of the macro-flow model, provides strong evidence of the new approach's incremental contribution to the literature. However, we cannot assess from these results the incremental value of either the other amendments to the macro-flow model (for the value of different extensions, see the "Model Variations" subsection) or the use of virtual reality (for the value of virtual reality for prelaunch sales forecasting, see the "Supplemental Analysis" section).
For further benchmarking the sales forecast calculated using the extended macro-flow model, we also compare it with conjoint analysis, other approaches to sales forecasting that require different data such as diffusion and utility models, and purchase intention conversion schemes (Web Appendix W7). The results reveal that the new approach overall achieves the highest forecasting accuracy. However, other forecasting approaches, such as conjoint analysis or surveys on purchase intentions, are generally associated with fewer data-gathering efforts and therefore provide alternatives for firms that have time or money constraints or need different or more high-level insights (e.g., estimating willingness to pay or making go/no-go decisions).
We validated the extensions of the macro-flow model using model variations, which take out one new component at a time. Importantly, we could take out only the amendments we made to the macro-flow model as they are not shown as a stimulus in the simulation. Assessing the relative impact of amendments related to the simulation (e.g., use of virtual reality, the no-buy option, or the search for information) is not possible because participants' experiences in the simulation cannot be undone. To assess the impact of using virtual reality we ran a supplemental analysis, which we discuss after Study 2.
Comparing the model variations (Table 4), we conclude that the greatest impact on forecasting accuracy due to macro-flow model extensions comes from the combination of virtual purchases with preferences in the purchase decision stage (Equation 7, Table 2; average accuracy increases across Studies 1a and 1b by 54.0%, with 15.8% coming from virtual reality data on the purchase decision and 38.2% from the preference measurement; see Variants 5 and 6 in Table 4). Adding awareness via third-party information (Equation 3, Table 2) leads to an average accuracy increase across the two studies by 29.8% (see Variant 1 in Table 4), whereas the contribution to forecasting accuracy is lower for the extensions of previous macro-flow models in the information search states (Equation 5, Table 2; average accuracy increase across the two studies by 3.2%; see Variants 2–4 in Table 4).
Graph
Table 4. Impact of Macro-Flow Model Extensions.
| Model | Awareness | Information Search | Purchase Decision | Accuracy | Accuracy Increase |
|---|
| Ads | WOM | Third | Actions | Time | Visits | Virtual Purchases | Preferences | Monthly Forecast Study 1a | Monthly Forecast Study 1b | Mean Studies 1a and 1b | Mean Studies 1a and 1b |
|---|
| Extended macro-flow model | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | 18.6% | 44.4% | 31.5% | Basis for comparison |
| Variant 1 | ✓ | ✓ | | ✓ | ✓ | ✓ | ✓ | ✓ | 40.4% | 82.1% | 61.3% | +29.8% |
| Accuracy Increase Due to Extensions in the Awareness Stage | | +29.8% |
| Variant 2 | ✓ | ✓ | ✓ | | ✓ | ✓ | ✓ | ✓ | 20.2% | 44.6% | 32.4% | +.9% |
| Variant 3 | ✓ | ✓ | ✓ | ✓ | | ✓ | ✓ | ✓ | 23.7% | 43.9% | 33.8% | +2.3% |
| Variant 4 | ✓ | ✓ | ✓ | ✓ | ✓ | | ✓ | ✓ | 18.6% | 44.5% | 31.5% | +/−.0% |
| Accuracy Increase Due to Extensions in the Information Search Stage | | +3.2% |
| Variant 5 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | | ✓ | 22.7% | 72.0% | 47.3% | +15.8% |
| Variant 6 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | | 88.4% | 51.0% | 69.7% | +38.2% |
| Accuracy Increase Due to Extensions in the Purchase Decision Stage | | +54.0% |
3 Notes: The checkmark (✓) indicates that the model component is included. With this comparison, we investigate the incremental value of the extensions we made to the macro-flow model. However, it is impossible to extract the incremental value of virtual reality for the new forecasting approach from this table, because all depicted variants are computed based on an entire virtual reality consumer purchase journey and market environment simulation. For the value of virtual reality, see Table 6.
Notably, findings showed that the preferences and the virtual purchases complement one another (Variants 5 and 6 in Table 4). Specifically, although observing a participant's virtual purchase (i.e., yes/no decision between purchasing vs. not purchasing) may be more representative of their choice, observing a nuanced preference may help incorporate some of the participant's uncertainty when making this choice. These rationales are also reflected in the results. First, the results reveal the highest forecasting accuracy across Studies 1a and 1b when virtual purchases and preferences are used together, confirming the intuition that combining the approaches allows for leveraging the advantages of both and increases forecasting accuracy ([27]). Second, the results show that while preferences have a higher incremental value in Study 1a (expensive product from a high-involvement product category), the incremental value of including the virtual purchases is higher in Study 1b (lower-cost product from a low-involvement product category). This is in line with studies showing that uncertainty can lead to increased selection of the no-choice option when products are associated with either high involvement or high financial risk ([ 6]; [30]).
Finally, we compared the lab and online virtual realities. To assess whether the differences in the similarity to reality and immersion between the two (Table 1) lead to differences in sales forecasting accuracy, we compared aggregate sales forecasts derived solely on the basis of the lab participants and the online participants. In line with our expectations, the results reveal that in both studies the forecasting accuracy is higher for the lab (vs. online) virtual reality and increases on average by 6%.
To offer nuanced insight, we predicted individual purchase behavior within the virtual reality simulation from each participant's information search behavior and preferences. For this purpose, we used Equations 5 and 6 and an adapted version of Equation 8, without awareness and using only preferences. The results reveal that across both studies, the lab (vs. online) virtual reality condition, on average, predicts participants' purchase behavior within the simulation 6.5% more accurately. For Study 1b (Study 1a), we correctly predicted the purchase decision of 72.6% (54.1%) of the lab and only 62.4% (51.2%) of the online participants. The percentage of correct predictions is higher in Study 1b because Study 1a covers more products (Study 1b: 10 products; Study 1a: 18 products). In line with these aggregated findings, individual behavior prediction results show that, on average, lab (vs. online) virtual reality participants' behavior in the simulation was more consistent with actual buyer behavior.
Behavioral consistency refers to the extent to which consumers move through their purchase journey (i.e., from searching information to building preferences and making the purchase decision) in a coherent manner ([64]). We focus on behavioral consistency, because actual purchase behavior for durables tends to be a result of—and therefore, consistent with—consumers' information search behavior and preferences ([32]; [38]; [40]; [45]). Understanding differences in behavioral consistency between the virtual reality simulation types could therefore provide first insights regarding the drivers of the prediction accuracy differences. More precisely, in integrating insights from research on durable purchasing behavior ([32]; [45]) and virtual reality in other fields ([16]), we expect that lab virtual reality would motivate higher levels of behavioral consistency than online virtual reality—for instance, because it creates high immersion and comprehensive, realistic experiences (Table 1).
To operationalize behavioral consistency, we developed a score reflecting these rationales: we counted the occurrence of consistent behavior throughout the simulation and the preference measurement. The score increases by a count of 1 when consistent behavior occurs (e.g., if a participant assigns a preference to a product that they have previously viewed, or if a participant purchases any product for which they previously assigned a preference). Therefore, the minimum score is 0, indicating no consistent behavior at all (i.e., the participant only assigned preferences to products that they did not view and purchased a product to which they assigned a preference of 0). The maximum consistency score is the number of products plus 1. This maximum score can be reached if a participant assigns a preference to each product, views each product, and purchases one.[ 8] Confirming our expectations, the results reveal significant differences between lab and online virtual reality for Studies 1a and 1b, with lab participants behaving on average more consistently than online participants (Study 1a: F( 1, 303) = 60.45, p =.000; MlabVR = 2.60 vs. MonlineVR =.93; Study 1b: F( 1, 323) = 27.82, p =.000, MlabVR = 2.08 vs. MonlineVR = 1.33).
The results of Studies 1a and 1b show highly beneficial consequences of using virtual reality technologies for prelaunch sales forecasting. Moreover, the comparison between lab and online virtual reality participants hints at increased behavioral consistency between information search, preferences, and virtual reality buying behavior as a reason behind their superior forecasting performance. However, the two field studies can neither disentangle the underlying mechanisms of these benefits nor conclusively quantify the specific contribution of using virtual reality for forecasting accuracy. To address this void, we conducted Study 2 to examine why virtual reality simulations lead to forecasting advantages by comparing a lab versus an online virtual reality purchase journey simulation versus a traditional studio test with real products (RQ2). Finally, we use Study 2 data and results to approximate the impact of virtual reality within the new prelaunch sales forecasting approach.
Prior work in other contexts provides several clues why virtual reality might lead to advantages compared with previous visualization approaches. We organize these clues into two theoretical perspectives: presence and vividness. In addition, we consider alternative lenses (i.e., convenience and decision uncertainty), which we explain in the "Results" section. The focus on presence and vividness is in line with our conceptualization of virtual reality for prelaunch sales forecasting (Table 1). We argue that both lab and online virtual realities differ from traditional studio tests with real products in terms of simulation scope, similarity to reality, and immersion, whereas lab and online virtual realities differ only in their similarity to reality and immersion.
"Presence" refers to the phenomenon of the participant feeling as if they are actually in the simulation ([ 8]). Consequently, presence has an internal component (i.e., immersion—the degree to which senses are stimulated via interactions with the simulation) and an external component (i.e., realism of environment—the resemblance of the simulated environment of a real environment) ([ 2]; [16]; [49]; [59]).
Studies indicate that virtual reality may enhance consumers' presence. For instance, virtual reality simulations provide visual connections and realistic experiences ([28]) or dynamic presentation formats that create feelings that the user is actually experiencing the product ([59]). We therefore expect that lab and online virtual reality (vs. a traditional studio test with real products) leads to advantages in prelaunch sales forecasting. In addition, because lab virtual reality uses additional equipment that may facilitate both participant immersion and an environment that further resembles an actual environment (Table 1), we expect that individuals in a lab (vs. online) virtual reality would experience higher presence. Owing to higher presence, consumers would better imagine the simulated situation and thus behave in virtual reality much as they would in the real world ([16]; [52]).
Vividness denotes the extent to which consumers feel that a simulation is lively and detailed as well as easy to imagine and remember ([44]). According to vividness theory ([55]), lively simulations enhance consumers' imagination, which reduces the perceived effort to process the displayed information ([49]) and increases their involvement ([28]).
In addition, studies show that interacting with a virtual object increases vividness and produces more images in the participant's mind than static pictures or text ([61]). Building on these insights, we expect that virtual reality simulations (vs. traditional studio tests with real products) will elicit higher vividness, which manifests in more consistent and realistic behavior. Because both virtual reality types can generate a very high simulation scope, we do not expect differences in vividness between them (Table 1).
Study 2 is a between-subjects experiment (n = 210) in which we compare a purchase journey simulation between lab virtual reality, online virtual reality, and a studio test with real products. We randomly assigned participants to one of the three conditions, resulting in 70 participants per condition. All three groups were asked to assess the same ten gardening tools in the simulated local retail store as presented in Study 1b, which was a do-it-yourself (DIY) store selling hardware and tools for home improvement (for the lab and online virtual reality visuals, see Figure 1). The choice to use an excerpt from Study 1b is in line with our goal for Study 2 to be comparable to Studies 1a and 1b but to offer more control.
All information on the products was identical between the three conditions. However, in contrast to the virtual reality conditions with an entire DIY store with the ten gardening tools (very high scope of the simulation), participants in the studio test saw only a shelf with all ten gardening tools (i.e., medium simulation scope). This more focused visualization follows state-of-the art market research practice emphasizing the need for a focused, but realistic, depiction of the new product and its key competitors on a shelf as they would be in-store, but within a neutral and quiet environment to avoid any distraction. Compared with virtual reality simulations with 360-degree depictions of the environments and products, standing in front of a shelf with real products may create only partial similarity to reality. Moreover, the studio test is likely to create medium immersion, as it only stimulates senses with regard to product interactions (vs. stimulating senses with regard to product and environment interactions). Yet, in all three groups, participants could interact with the products and the shelf in a similar manner: they walked to the shelf, could pick up any of the ten products from the shelf as often as they liked, hold each one for as long as they liked, turn it to assess it from 360 degrees, return it to the shelf, or put it in their shopping basket for purchase. As in Studies 1a and 1b, participants could decide whether to assess all of the products or only some of them.
Before the simulation, participants were asked questions regarding demographics and general characteristics. Afterward, we conducted a preference measurement and a survey asking for participants' assessment of the products and the simulation. Using confirmatory factor analysis, we assessed reliability and validity for each measure. Overall, the scales exhibit sufficient psychometric properties: for all constructs, the values for item reliabilities, composite reliability, and average variance extracted surpass the recommended thresholds ([ 1]; [39]). Moreover, the analysis of [23] criterion indicates that discriminant validity exists for all constructs. Table 5 provides an overview of the variables and their measurement.
Graph
Table 5. Study 2: Overview of Variables.
| Variable | Items (Seven-Point Likert Scale) | IL | IR | CR | AVE |
|---|
| Presence Perspective |
| Presence | I felt like I'd just been to a real store. | .93 | .88 | .93 | .83 |
| I forgot my actual surroundings during the simulation. | .80 | .64 |
| It was like I was really visiting a store. | .95 | .96 |
| Immersion | The simulation immersed my senses. | .80 | .67 | .86 | .68 |
| I was able to do everything in the simulation that would have been possible in a real store. | .73 | .54 |
| The simulation activates the same senses as a real store visit. | .85 | .83 |
| Realism of environment (Cho, Shen, and Wilson 2014) | The visual depiction of the store seemed real. | .96 | .92 | .98 | .95 |
| The visual depiction of the store was realistic. | .98 | .97 |
| The store shown during the simulation looked real. | .97 | .94 |
| Vividness Perspective |
| Vividness(Keller and Block 1997) | How would you rate the depiction in the simulation? | | | | |
| Not vivid vs. vivid | .69 | .48 | .83 | .63 |
| Not concrete vs. concrete | .82 | .79 |
| Not easy to picture vs. easy to picture | .77 | .62 |
| Convenience Perspective |
| Convenience(Donthu and Gilliland 1996) | The simulation was convenient. | .82 | .69 | .88 | .72 |
| The simulation made it easy to gather information for making a purchase decision. | .87 | .85 |
| The simulation helped me to quickly gather information. | .77 | .60 |
| Usefulness(Davis 1989) | I found the simulation useful. | .86 | .79 | .90 | .75 |
| The simulation made the purchase decision easier. | .81 | .68 |
| The simulation enhanced my effectiveness in making a purchase decision. | .86 | .77 |
| Ease of use(Davis 1989) | I found the simulation easy to use. | .88 | .82 | .91 | .77 |
| The use of the simulation was clear and understandable. | .85 | .75 |
| Learning how to operate the simulation was easy. | .85 | .75 |
| Decision Uncertainty Perspective |
| Decision uncertainty (Rosbergen, Pieters, and Wedel 1997) | I am uncertain of my choice. | .86 | .75 | .93 | .82 |
| I did not know if I made the right purchase. | .90 | .84 |
| I felt a bit at a loss in choosing one of the products. | .91 | .88 |
| Additional Variables of Interest |
| Realism of product (Cho, Shen, and Wilson 2014) | The visual depiction of the products seemed real. | .95 | .94 | .91 | .77 |
| The visual depiction of the products was realistic. | .94 | .89 |
| I could experience the products as in real life. | .69 | .47 |
| Task involvement (Pham 1996) | I was very motivated to reach an accurate evaluation of the products. | .77 | .61 | .86 | .68 |
| I did put much effort into the evaluation of the products. | .84 | .81 |
| It was important for me to examine the products carefully. | .77 | .61 |
| Manipulation Checks |
| Risk aversion(Donthu and Gilliland 1996) | I prefer not to take risks when purchasing anything. | .82 | .73 | .84 | .65 |
| In general, I avoid risky things. | .83 | .78 |
| I would rather be safe than sorry. | .66 | .43 |
| Technology experience (Beatty and Smith 1987) | Compared to the average person,... | | | | |
| ...I am highly knowledgeable regarding new technologies. | .90 | .81 | .95 | .87 |
| ...I regard myself as very experienced in new technologies. | .95 | .94 |
| ...I know a lot about new technologies. | .92 | .85 |
| Need for touch(Peck and Childers 2003) | I rather trust products that can be touched before purchase. | .89 | .85 | .91 | .77 |
| I feel more comfortable purchasing a product after physically examining it. | .89 | .83 |
| I feel more confident making a purchase after touching a product. | .79 | .63 |
4 Notes: IL = item loading; IR = item reliability; CR = composite reliability; AVE = average variance extracted.
We used analysis of variance (ANOVA) to analyze the data obtained from the controlled laboratory experiment and summarize the results in Figure 4, Panels A–I. The findings reveal that the randomization of participants worked, as the groups did not differ with respect to gender (F( 2, 207) =.11, p =.894) and other traits, such as risk aversion (F( 2, 207) =.04, p =.961), technology experience (F( 2, 207) = 1.05, p =.350), and need for touch (F( 2, 207) = 1.40, p =.248).
Graph: Figure 4. Study 2: ANOVA results for virtual reality mechanisms.*p <.1.**p <.05.***p <.01.aDirect measure of theoretical perspectives.Notes: Error bars = ±1 SEs.
Building on Studies 1a and 1b, using the individual-level information search, preference, and purchase behavior data from virtual reality, we assessed behavioral consistency. The results reveal significant differences in behavioral consistency between virtual reality and non–virtual reality participants (Figure 4, Panel A). Specifically, confirming the results of Studies 1a and 1b, we find that behavioral consistency is highest for the lab virtual reality, followed by online virtual reality and the studio test. Therefore, in line with our expectations, virtual reality participants show behaviors in the simulation that are closer to actual consumer behavior than studio test participants.
The ANOVA results provide support for the presence and vividness perspectives (Figure 4, Panels B and E). The findings show great differences between the groups for the direct measures of presence and vividness. The findings also show significant group differences for the constructs associated with the presence perspective (Figure 4, Panels C and D): immersion and realism of the environment.[ 9] Overall, the results confirm the rationales for why the virtual reality advantage occurs: lab virtual realities, due to the additional equipment use, generate higher presence than online virtual realities, whereas both virtual realities have advantages over a traditional studio test due to their higher presence and vividness.
As Table 5 shows, we measured further constructs to rule out alternative explanations. For example, research on virtual reality suggests that it may facilitate consumers' convenience (i.e., the extent to which consumers perceive the simulation as easy to use and useful) by expediting their access to relevant decision parameters or enhancing the possibilities to explore the products relative to visualization modes not relying on new technologies ([28]; [34]; [49]). Moreover, virtual reality simulations could reduce participants' decision uncertainty (i.e., the extent of difficulty in making a purchase decision) by collecting more information and better integrating information than visualization modes that do not use virtual reality ([ 5]; [49]). However, the complexity of a virtual reality simulation, as well as usage troubles (e.g., dizziness, fatigue), could diminish potential convenience and decision uncertainty reduction advantages ([53]). In line with these rationales, the results do not reveal differences in the direct measures of decision uncertainty (Figure 4, Panel I) and convenience (Figure 4, Panel F), or for constructs associated with convenience (Figure 4, Panels G and H).
To ensure that the virtual product representations in the lab and online virtual realities were displayed adequately, we assessed product realism (i.e., the extent to which the depicted product resembles the real product). We expected no differences, as the virtual product representations were created from 360-degree pictures taken of the actual products. The results confirm these expectations and show that the physical and virtual products are adequate benchmarks (F( 2, 207) =.60, p =.550; MlabVR = 5.03 vs. MonlineVR = 4.90 vs. Mstudio test = 5.14). In addition, the results on task involvement (i.e., the extent to which participants were focused on the experimental task at hand) show no significant differences between the groups (F( 2, 207) = 1.09, p =.338; MlabVR = 5.84 vs. MonlineVR = 5.61 vs. Mstudio test = 5.62).
Studies 1a and 1b reveal beneficial consequences of using virtual reality technologies for prelaunch sales forecasting. Study 2 offers suggestive evidence why these benefits occurred. However, due to the absence of a non–virtual reality control condition in Studies 1a and 1b and the more focused design of Study 2 (i.e., virtual reality mechanisms), we still cannot conclusively disentangle how much of the observed forecasting advantage can be attributed to virtual reality. To approximate this, we jointly used the results from Studies 1b and 2: Although Study 2 has a smaller scope and does not completely replicate the simulation of Study 1b, both studies are comparable because they include the identical DIY store simulation. We conducted the approximation in two steps.
First, we approximated the total effect of using (vs. not using) virtual reality for prelaunch sales forecasting. To do so, we built an artificial, non–virtual reality control condition for Study 1b in leveraging the non–virtual reality control condition of Study 2 (i.e., studio test using real products): we screened Study 2 data for differences between the virtual reality and non–virtual reality conditions in metrics that serve as inputs for forecasting. This screening revealed that while the preferences for the focal new durable product were highly similar across studies for the lab and online virtual reality participants, these preferences were 1.2 times higher for non–virtual reality participants. To account for this difference, we multiplied the preferences of Study 1b participants by 1.2 to create the artificial, non–virtual reality control condition. Using this control condition, we then calculated another sales forecast for Study 1b (Modification 1 in Table 6), which uses neither virtual reality data (i.e., no actions, time, or visits in the information search stage and no virtual purchases in the purchase decision stage) nor virtual reality for the visualization of the purchase journey (i.e., because participants' preferences were adjusted; see previous step). Comparing the Modification 1 forecast with the Study 1b forecast indicates that the usage of virtual reality accounts for an accuracy increase of roughly 50%–70% (Table 6).
Graph
Table 6. Approximation of Virtual Reality's Impact on Forecasting Accuracy.
| Extended Macro-Flow Model | Visualizationof Purchase Journey | MAPEMonthly ForecastStudy 1b(Original Prelaunch Forecast) | MAPEMonthly ForecastStudy 1b (Forecast with Actual Ad Spending) |
|---|
| Awareness | Information Search | Purchase Decision |
|---|
| Ads | WOM | Third | Actions | Time | Visits | Virtual Purchases | Preferences |
|---|
| | | Data Provided byAutomated Tracking | |
|---|
| Step 1: Approximatingthe total effect of using (vs. not using) virtual reality for prelaunch sales forecasting | Virtual reality prelaunch sales forecasting approach | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | Lab and online virtual reality(Study 1b) | 44.4% | 26.5% |
| Modification 1:No virtual reality data and visualization | ✓ | ✓ | ✓ | — | — | — | — | ✓ | Studio test (Study 2) | ∼100% | ∼95% |
| Total Effect: Approximated Accuracy Increase Due to Using Virtual Reality: | ▵ ≈ 55% | ▵ ≈ 70% |
| Step 2: Splitting thetotal effect betweenvirtual reality's automated trackingand visualization capability | Virtual reality prelaunch sales forecasting approach | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | Lab and online virtual reality(Study 1b) | 44.4% | 26.5% |
| Modification 2:No virtual reality data | ✓ | ✓ | ✓ | — | — | — | — | ✓ | Lab and online virtual reality(Study 1b) | ∼70% | ∼50% |
| Effect from Virtual Reality's Automated-Tracking Capability: | ▵ ≈ 25% | ▵ ≈ 25% |
| Effect from Virtual Reality's Visualization Capability:(i.e., difference between total effect and effect from virtual reality's automated-tracking capability) | ▵ ≈ 30% | ▵ ≈ 45% |
Second, we split the total effect between virtual reality's automated-tracking and visualization capability. To do so, we calculated one more sales forecast (Modification 2 in Table 6) that contains virtual reality for the visualization of the purchase journey (i.e., by using original preferences of Study 1b participants) but does not include any virtual reality data (i.e., no data on actions, time, or visits during information search and no virtual purchases). Thus, comparing this modification with the original sales forecast of Study 1b provides insights on the contribution of virtual reality's automated-tracking capability, which is roughly 25%. Next, to isolate the contribution of virtual reality's visualization capability, we jointly interpreted the results of Modifications 1 and 2. The joint interpretation is necessary because it is impossible to calculate a sales forecast using data tracked in virtual reality without having a visualization in virtual reality. Findings indicate that virtual reality's visualization capability accounts for a forecasting accuracy of roughly 30%–45% in Study 1b (Table 6). Thus, the impact of virtual reality's visualization capability seems to be higher than its automated-tracking capability. However, due to the artificial nature of the non–virtual reality condition, we still cannot conclusively disentangle the exact impact of virtual reality.
The results of Study 1a (kitchen appliance) and Study 1b (gardening tool) show that our new virtual reality sales forecasting approach yields highly accurate predictions. Moreover, these results indicate that these benefits occurred because lab virtual reality motivated, on average, consistent and realistic consumer behavior, whereas these metrics were at a lower level for online virtual reality. To provide further insight, in Study 2 we compared the two virtual reality types with a studio test with real products. The findings confirm that lab virtual reality simulations motivated more consistent consumer behavior than online virtual reality, and that virtual reality creates superior behavioral consistency compared with the studio test. Moreover, the findings indicate that these effects are due to advantages that virtual reality creates in terms of presence and vividness, not alternative theoretical explanations, such as convenience and decision uncertainty. Using Study 2 and Study 1b findings jointly allowed us to approximate that the forecasting advantage attributed to virtual reality is likely to range between 50% and 70% (Table 6).
This investigation yields results that advance prior work in several ways. First, researchers have developed forecasting models for new durables that are applied close to launch ([25]). We advance the literature by conceptualizing and testing a novel virtual reality sales forecasting approach that can also be applied in earlier NPD stages because firms do not need a fully developed product but only a virtual blueprint, which serves as input for the programming of the virtual reality simulation. This flexible application of the new approach is an important advancement, because in earlier NPD stages, investments in a new product are relatively small compared with investments in later stages ([51]). Future work could build on these insights and investigate important extensions. For example, in both field studies, our collaborating partners were able to categorize the products, anticipate the time until launch, as well as the market environment at launch. However, because for some innovations and at very early stages of NPD the categorization of a new durable may be difficult (Web Appendix W6), future investigators could develop more precise solutions that help anticipate the future market environment or simulate the exact time until launch.
Second, in developing the new virtual reality forecasting approach, we extended previous macro-flow models in three ways. We added new behavioral states (e.g., aware by third-party information, extensive and flexible information search); we introduced new equations for calculating the flows (e.g., Equation 5 in Table 2) and new determinants of the flows (e.g., behavioral data from virtual reality). Through these amendments, the extended macro-flow model covers the purchase journey comprehensively and realistically. It is therefore likely to be more generally applicable, as in our two field studies with two very different durables, than previous macro-flow models that focused on one particular industry (e.g., automotive in [69]]). Future work could build on these insights and examine how macro-flow models as well as the virtual reality simulations need to be adapted to adequately predict buyer behavior for other types of products or markets and therefore extend the external validity of the new forecasting approach. For example, if the product's value is low (and buyers are not very systematic in their behavior), how should behavioral data and preference measurement data be combined for the purchase decision stage (Equation 7, Table 2)? If the innovation is a service, are there additional determinants that need to be included in the information search stage (Equation 5, Table 2) to account for the service's individuality? How should competing services be depicted in virtual reality? Finally, if the new product is sold in a business-to-business market with different members involved in the decision making, should they jointly use the virtual reality simulation? How should their data be aggregated to predict the buying center purchase decision (Equation 8, Table 2)?
Third, researchers have focused on developing and testing prelaunch sales forecasting approaches ([69]; [67]). We advance the literature by providing additional insights into why forecasting advantages occur. In doing so, the findings also expand the knowledge on the mechanisms of virtual reality. However, while we show across two large-scale field studies that the two key features we add—virtual reality and macro-flow model extensions—substantially improve the forecasting accuracy, we cannot conclusively assess how much of the accuracy increase is due to virtual reality. Thus, we encourage future studies on virtual reality sales forecasting to include a non–virtual reality control condition and to more directly isolate the impact of virtual reality and the contributions by each of its mechanisms.
Finally, we offer the first systematic marketing study of virtual reality in NPD. Future work should build on these insights to further examine how firms can use virtual reality to advance their capabilities. For example, as [19] has shown, firms' marketing capabilities fall into two main categories: customer linking and market sensing. While we leverage virtual reality for an improved market sensing in terms of highly accurate prognoses, future work could focus on how to use virtual reality to improve the market sensing related to diagnostics (e.g., by studying consumer heterogeneity or idiosyncratic purchase journeys, testing marketing materials, or optimizing the marketing mix) and customer-linking purposes (e.g., by integrating this knowledge into new products, offering remote services or trainings).
Our study contains actionable implications that fall into three categories: ( 1) application fields of virtual reality for NPD, ( 2) visualization mode selection, and ( 3) guidance for implementing virtual reality for NPD. These implications are derived from our field studies (Studies 1a and 1b), the controlled laboratory experiment (Study 2), and several in-depth interviews that we conducted with senior managers of leading durable producers.
Despite the increasing interest in virtual reality, the in-depth interviews revealed that many companies remain reluctant to use virtual reality as a matter of routine, given, for instance, the complexity and costs of programming virtual reality environments. Moreover, the interviews surfaced that companies experience challenges when implementing virtual reality in NPD due to a lack of knowledge within the company and that effective use cases for this technology remain scarce. We identified prelaunch sales forecasting as a promising application field for virtual reality. Furthermore, the interviews hinted at additional applications for virtual reality in NPD associated with its superior visualization capability, as a senior manager of a global technology company pointed out:
Virtual reality is a great tool to showcase products, objects and entire worlds. We can depict those products, objects and worlds in a realistic manner, show them to other people and especially make our products more easily understandable. This is something very valuable during new product development.
Drawing on this insight and blending it with the existing literature, we identified other use cases for virtual reality across all NPD stages. For example, research has shown that virtual reality can be beneficial in the early stages of NPD and improve idea generation outcomes by creating immersive collaboration platforms ([ 5]). In addition, we demonstrate how virtual reality can improve forecasting outcomes using realistic and comprehensive purchase journeys (Studies 1a and 1b). Moreover, using the interview insights, we also derive use cases for NPD's medial stages, such as concept and product development. For instance, one of our interview partners, a senior manager of a leading household appliance producer, reported very positive experiences with virtual prototype testing:
To make our innovation process more agile, we recently experimented with virtual prototypes and compared the results to actual prototype testing. Stunningly, we did not find any differences in diagnostic information between the two![10]
The results show that virtual reality can reap large benefits, such as more flexible use throughout NPD or more accurate prognoses. However, the results also reveal some limitations of virtual reality: it can be expensive and have some use constraints when testing haptics ([16]; [53]). We therefore present specific guidelines for managers on how to choose between lab virtual reality and online virtual reality, as well as when to eschew virtual reality (Figure 5).
Graph: Figure 5. Guidance for managers on when to utilize virtual reality in NPD.
First, in weighing the costs and benefits of virtual reality, we recommend that managers use virtual reality simulations to improve NPD decisions that require early, detailed, and precise consumer information (i.e., user experience testing, purchase journey analyses, prelaunch sales forecasting, and point-of-sale optimization). Moreover, we suggest that the business case for using virtual reality is especially strong if a new durable is expensive, is highly innovative, or requires considerable explanation. We make this recommendation because findings reveal that using virtual reality results in realistic consumer behavior even for these products, whereas prior work has shown that accurately predicting sales for such products is hardly possible with traditional methods ([37]).
Second, we suggest that managers carefully choose between lab and online virtual reality as the visualization mode. Managers should choose lab virtual reality if highly accurate consumer insights are required, because our findings reveal that lab virtual reality generates significantly higher presence, which manifests in more consistent behavior. In contrast, managers should use online virtual realities when cost constraints are high, data gathering needs to be conducted promptly, and large samples are required. If companies require scalability and very detailed and realistic insights, they can combine online and lab virtual realities (Figure 5). Finally, managers should refrain from using virtual reality when they only need high-level insights. For example, if they need to make general go/no-go decisions or long-term market share predictions, virtual reality simulations may be too expensive and traditional market research methods, such as purchase intention surveys or conjoint analysis, are likely to have a superior cost–benefit ratio. In addition, when testing haptics, traditional studio tests with physical prototypes may lead to better diagnostic information as simulating haptics in virtual reality is difficult ([16]).
We have two recommendations for managers interested in implementing virtual reality in their firms' NPD. We recommend clearly defining the aims of using virtual reality and specifying a clear and comprehensive implementation plan to ensure focus throughout the project (Web Appendix W8). Our experience from working with and interviewing durable producers on new virtual reality use cases reveals that many companies get very excited about this technology, sometimes resulting in unrealistic expectations regarding simulation scope and length. To avoid overburdening of the virtual reality simulations, we recommend refraining from "all-in-one-solutions" and aim for one tool for one purpose (e.g., forecasting, prototype testing). Using virtual reality several times during NPD creates synergies because programming of virtual prototypes or environments can be slightly adjusted to different tests and contexts.
Supplemental Material, sj-pdf-1-jmx-10.1177_00222429211014902 - Virtual Reality in New Product Development: Insights from Pre-launch Sales Forecasting for Durables
Supplemental Material, sj-pdf-1-jmx-10.1177_00222429211014902 for Virtual Reality in New Product Development: Insights from Pre-launch Sales Forecasting for Durables by Nathalie Harz, Sebastian Hohenberg and Christian Homburg in Journal of Marketing
Footnotes 1 Jacob Goldenberg
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Online supplement: https://doi.org/10.1177/00222429211014902
5 We select two years because for many new durables, new versions are introduced within one to two years after launch (e.g., smartphones, kitchen appliances, TVs).
6 Awareness drives the dissemination of the new product in the market. Therefore, correct calibration of awareness flow rates is critical for accurate forecasts. However, estimating flow rates, especially without early sales data, is difficult ([58]).
7 Alternative approaches include paired comparisons (e.g., [63]) or purchase intentions. Paired comparisons are not feasible, given the number of products and attributes ([54]), and purchase intentions are assessed in Web Appendix W7.
8 Consistency tends to be higher when viewing and assigning preferences for many products compared with either when viewing only one product or when viewing many products but assigning a preference for and purchasing only one product.
9 Realism of environment does not reflect realism of the simulation (comprising environment, products, interactions, and the purchase decision), as the items in Table 5 illustrate.
To follow up, we conducted a lab experiment (n = 100) and randomly assigned participants to two conditions: virtual reality prototype testing or physical prototype testing. ANOVA results showed no significant differences in the perceived product advantage between groups (Mvirtual = 4.61, SD = 1.52 vs. Mphysical = 4.81, SD = 1.30; F(1, 98) =.47, p =.496), purchase intentions (Mvirtual = 3.74, SD = 1.79 vs. Mphysical = 4.24, SD = 1.75; F(1, 98) = 1.99, p =.161), or user experience (Mvirtual = 5.66, SD = 1.17 vs. Mphysical = 5.81, SD =.94; F(1, 98) =.50, p =.481).
References Bagozzi Richard P. , Yi Youjae. (2012), " Specification, Evaluation, and Interpretation of Structural Equation Models, " Journal of the Academy of Marketing Science , 40 (1), 8 – 34.
Baños Rosa María , Botella Cristina , Garcia-Palacios Azucena , Martin Helena Villa , Perpiñá Conxa , Raya Mariano Alcañiz. (2000), " Presence and Reality Judgment in Virtual Environments: A Unitary Construct? " CyberPsychology & Behavior , 3 (3), 327 – 35.
Barroso Alicia , Llobet Gerard. (2012), " Advertising and Consumer Awareness of New, Differentiated Products, " Journal of Marketing Research , 49 (6), 773 – 92.
Beatty Sharon E. , Smith Scott M.. (1987), " External Search Effort: An Investigation Across Several Product Categories, " Journal of Consumer Research , 14 (1), 83 – 95.
Bhagwatwar Akshay , Massey Anne , Dennis Alan. (2018), " Contextual Priming and the Design of 3D Virtual Environments to Improve Group Ideation, " Information Systems Research , 29 (1), 169 – 85.
Bloch Peter H. , Richins Marsha L.. (1983), " A Theoretical Model for the Study of Product Importance Perceptions, " Journal of Marketing , 47 (3), 69 – 81.
Bolton Lisa E.. (2003), " Stickier Priors: The Effects of Nonanalytic Versus Analytic Thinking in New Product Forecasting, " Journal of Marketing Research , 40 (1), 65 – 79.
Bowman Doug A. , McMahan Ryan P.. (2007), " Virtual Reality: How Much Immersion Is Enough? " Computer , 40 (7), 36 – 43.
Bronnenberg Bart J. , Kim Jun B. , Mela Carl F.. (2016), " Zooming in on Choice: How Do Consumers Search for Cameras Online? " Marketing Science , 35 (5), 693 – 712.
Burke Raymond R.. (1996), " Virtual Shopping: Breakthrough in Marketing Research, " Harvard Business Review , 74 (March–April), 120 – 31.
Cao Xinyu , Zhang Juanjuan. (2021), " Preference Learning and Demand Forecast, " Marketing Science , 40 (1), 62 – 79.
Castro Iana A. , Morales Andrea C. , Nowlis Stephen M.. (2013), " The Influence of Disorganized Shelf Displays and Limited Product Quantity on Consumer Purchase, " Journal of Marketing , 77 (4), 118 – 33.
Chandon Pierre , Morwitz Vicki G. , Reinartz Werner J.. (2005), " Do Intentions Really Predict Behavior? Self-Generated Validity Effects in Survey Research, " Journal of Marketing , 69 (2), 1 – 14.
Chang Woojung , Taylor Steven A.. (2016), " The Effectiveness of Customer Participation in New Product Development: A Meta-Analysis, " Journal of Marketing , 80 (1), 47 – 64.
Cho Hyunyi , Shen Lijiang , Wilson Kari. (2014), " Perceived Realism: Dimensions and Roles in Narrative Persuasion, " Communication Research , 41 (6), 828 – 51.
Cipresso Pietro , Giglioli Irene Alice Chicchi , Raya Mariano Alcañiz , Riva Giuseppe. (2018), " The Past, Present, and Future of Virtual and Augmented Reality Research: A Network and Cluster Analysis of the Literature, " Frontiers in Psychology , 9 (November) , 1 – 20.
Dahan Ely , Srinivasan V.. (2000), " The Predictive Power of Internet-Based Product Concept Testing Using Visual Depiction and Animation, " Journal of Product Innovation Management , 17 (2), 99 – 109.
Davis Fred D.. (1989), " Perceived Usefulness, Perceived Ease of Use, and User Acceptance of Information Technology, " MIS Quarterly , 13 (3), 319 – 40.
Day George S.. (1994), " The Capabilities of Market-Driven Organizations, " Journal of Marketing , 58 (4), 37 – 52.
Dhar Ravi , Simonson Itamar. (2003), " The Effect of Forced Choice on Choice, " Journal of Marketing Research , 40 (2), 146 – 60.
Donthu Naveen , Gilliland David I.. (1996), " Observations: The Infomercial Shopper, " Journal of Advertising Research , 36 (2), 69 – 76.
Fader Peter , Hardie Bruce G.S. , Huang Chun-Yao. (2004), " A Dynamic Changepoint Model for New Product Sales Forecasting, " Marketing Science , 23 (1), 50 – 65.
Fornell Claes , Larcker David F.. (1981), " Evaluating Structural Equation Models with Unobservable Variables and Measurement Error, " Journal of Marketing Research , 18 (1), 39 – 50.
Golder Peter N. , Tellis Gerard J.. (1997), " Will It Ever Fly? Modeling the Takeoff of Really New Consumer Durables, " Marketing Science , 16 (3), 256 – 70.
Goodwin Paul , Meeran Sheik , Dyussekeneva Karima. (2014), " The Challenges of Pre-Launch Forecasting of Adoption Time Series for New Durable Products, " International Journal of Forecasting , 30 (4), 1082 – 97.
Gourville John T.. (2006), " Eager Sellers and Stony Buyers: Understanding the Psychology of New-Product Adoption, " Harvard Business Review , 84 (6), 98 – 106.
Granger Clive W.J.. (1989), " Invited Review Combining Forecasts—Twenty Years Later, " Journal of Forecasting , 8 (3), 167 – 73.
Grewal Dhruv , Noble Stephanie M. , Roggeveen Anne L. , Nordfalt Jens. (2020), " The Future of In-Store Technology, " Journal of the Academy of Marketing Science , 48 (1), 96 – 113.
Grewal Rajdeep , Mehta Raj , Kardes Frank R.. (2004), " The Timing of Repeat Purchases of Consumer Durable Goods: The Role of Functional Bases of Consumer Attitudes, " Journal of Marketing Research , 41 (1), 101 – 15.
Gunasti Kunter , Ross William T.. (2009), " How Inferences About Missing Attributes Decrease the Tendency to Defer Choice and Increase Purchase Probability, " Journal of Consumer Research , 35 (5), 823 – 37.
Hauser John R. , Tellis Gerard J. , Griffin Abbie. (2006), " Research on Innovation: A Review and Agenda for Marketing Science, " Marketing Science , 25 (6), 687 – 717.
Hauser John R. , Urban Glen L. , Weinberg Bruce D.. (1993), " How Consumers Allocate Their Time When Searching for Information, " Journal of Marketing Research , 30 (4), 452 – 66.
Hauser John R. , Wisniewski Kenneth J.. (1982), " Dynamic Analysis of Consumer Response to Marketing Strategies, " Management Science , 28 (5), 455 – 86.
Heller Jonas , Chylinski Mathew , de Ruyter Ko , Mahr Dominik , Keeling Debbie I.. (2019), " Let Me Imagine That for You: Transforming the Retail Frontline Through Augmenting Customer Mental Imagery Ability, " Journal of Retailing , 95 (2), 94 – 114.
Henard David H. , Szymanski David M.. (2001), " Why Some New Products Are More Successful than Others, " Journal of Marketing Research , 38 (3), 362 – 75.
Hershfield Hal E. , Goldstein Daniel G. , Sharpe William F. , Fox Jesse , Yeykelis Leo , Carstensen Laura L. , et al. (2011), " Increasing Saving Behavior Through Age-Progressed Renderings of the Future Self, " Journal of Marketing Research , 48 (Special Issue), 23 – 37.
Hoeffler Steve. (2003), " Measuring Preferences for Really New Products, " Journal of Marketing Research , 40 (4), 406 – 20.
Howard John A. , Sheth Jagdish. (1969), The Theory of Buyer Behavior. New York : John Wiley & Sons.
Hu Li-tse , Bentler Peter M.. (1995), " Evaluating Model Fit, " in Structural Equation Modeling: Concepts, Issues, and Application , Hoyle R.H. , ed. Thousand Oaks, CA : SAGE Publications , 76 – 99.
Hu Ye , Du Rex Yuxing , Damangir Sina. (2014), " Decomposing the Impact of Advertising: Augmenting Sales with Online Search Data, " Journal of Marketing Research , 51 (3), 300 – 319.
Jamieson Linda F. , Bass Frank M.. (1989), " Adjusting Stated Intention Measures to Predict Trial Purchase of New Products: A Comparison of Models and Methods, " Journal of Marketing Research , 26 (3), 336 – 45.
Jhang Ji Hoon , Grant Susan Jung , Campbell Margaret C.. (2012), " Get It? Got It. Good! Enhancing New Product Acceptance by Facilitating Resolution of Extreme Incongruity, " Journal of Marketing Research , 49 (2), 247 – 59.
Ke T. Tony , Shen Zuo-Jun Max , Villas-Boas J. Miguel. (2016), " Search for Information on Multiple Products, " Management Science , 62 (12), 3576 – 3603.
Keller Punam Anand , Block Lauren G.. (1997), " Vividness Effects: A Resource-Matching Perspective, " Journal of Consumer Research , 24 (3), 295 – 304.
Lemon Katherine N. , Verhoef Peter C.. (2016), " Understanding Customer Experience Throughout the Customer Journey, " Journal of Marketing , 80 (6), 69 – 96.
Lim Noah , Ahearne Michael J. , Ham Sung H.. (2009), " Designing Sales Contests: Does the Prize Structure Matter? " Journal of Marketing Research , 46 (3), 356 – 71.
Luce R. Duncan. (1959), Individual Choice Behavior. New York : John Wiley & Sons.
Luo Lan , Kannan P.K. , Ratchford Brian T.. (2008), " Incorporating Subjective Characteristics in Product Design and Evaluations, " Journal of Marketing Research , 45 (2), 182 – 94.
Lurie Nicholas H. , Mason Charlotte H.. (2007), " Visual Representation: Implications for Decision Making, " Journal of Marketing , 71 (1), 160 – 77.
Marketing Science Institute (2018), " 2018–2020 Research Priorities: Marketers' Strategic Imperatives, " (May 13) , https://www.msi.org/articles/marketers-top-challenges-2018-2020-research-priorities/.
Markovitch Dmitri G. , Steckel Joel H. , Michaut Anne , Philip Deepu , Tracy William M.. (2015), " Behavioral Reasons for New Product Failure: Does Overconfidence Induce Overforecasts? " Journal of Product Innovation Management , 32 (5), 825 – 41.
Morales Andrea C. , Amir On , Lee Leonard. (2017), " Keeping It Real in Experimental Research—Understanding When, Where, and How to Enhance Realism and Measure Consumer Behavior, " Journal of Consumer Research , 44 (2), 465 – 76.
Munafo Justin , Diedrick Meg , Stoffregen Thomas. (2017), " The Virtual Reality Head-Mounted Display Oculus Rift Induces Motion Sickness and Is Sexist in Its Effects, " Experimental Brain Research , 235 (3), 889 – 901.
Netzer Oded , Srinivasan V.. (2011), " Adaptive Self-Explication of Multiattribute Preferences, " Journal of Marketing Research , 48 (1), 140 – 56.
Nisbett Richard E. , Ross Lee. (1980), Human Inference: Strategies and Shortcomings of Social Judgment. Englewood Cliffs, NJ : Prentice Hall.
Peck Joann , Childers Terry L.. (2003), " Individual Differences in Haptic Information Processing: The 'Need for Touch' Scale, " Journal of Consumer Research , 30 (3), 430 – 42.
Pham Michel Tuan. (1996), " Cue Representation and Selection Effects of Arousal on Persuasion, " Journal of Consumer Research , 22 (4), 373 – 87.
Roberts John H. , Nelson Charles J. , Morrison Pamela D.. (2005), " A Prelaunch Diffusion Model for Evaluating Market Defense Strategies, " Marketing Science , 24 (1), 150 – 64.
Roggeveen Anne L. , Grewal Dhruv , Townsend Claudia , Krishnan R.. (2015), " The Impact of Dynamic Presentation Format on Consumer Preferences for Hedonic Products and Services, " Journal of Marketing , 79 (6), 34 – 49.
Rosbergen Edward , Pieters Rik , Wedel Michel. (1997), " Visual Attention to Advertising: A Segment-Level Analysis, " Journal of Consumer Research , 24 (3), 305 – 14.
Schlosser Ann E.. (2006), " Learning Through Virtual Product Experience: The Role of Imagery on True Versus False Memories, " Journal of Consumer Research , 33 (3), 377 – 83.
Serranoa Berenice , Botellaa Cristina , Baños Rosa M. , Raya Mariano Luis Alcañiz. (2013), " Using Virtual Reality and Mood-Induction Procedures to Test Products with Consumers of Ceramic Tiles, " Computers in Human Behavior , 29 (3), 648 – 53.
Silk Alvin J. , Urban Glen L.. (1978), " Pre-Test-Market Evaluation of New Packaged Goods: A Model and Measurement Methodology, " Journal of Marketing Research , 15 (2), 171 – 91.
Smith Robert E. , Swinyard William R.. (1983), " Attitude-Behavior Consistency: The Impact of Product Trial Versus Advertising, " Journal of Marketing Research , 20 (3), 257 – 67.
Srinivasan Shuba , Rutz Oliver J. , Pauwels Koen. (2016), " Paths to and off Purchase: Quantifying the Impact of Traditional Marketing and Online Customer Activity, " Journal of the Academy of Marketing Science , 44 (4), 440 – 53.
Swaminathan Vanitha , Sorescu Alina , Steenkamp Jan-Benedict E.M. , O'Guinn Thomas Clayton Gibson , Schmitt Bernd. (2020), " Branding in a Hyperconnected World: Refocusing Theories and Rethinking Boundaries, " Journal of Marketing , 84 (2), 24 – 46.
Urban Glen L. , Hauser John R. , Qualls William J. , Weinberg Bruce D. , Bohlmann Jonathan D. , Chicos Roberta A.. (1997), " Information Acceleration: Validation and Lessons from the Field, " Journal of Marketing Research , 34 (1), 143 – 53.
Urban Glen L. , Weinberg Bruce D. , Hauser John R.. (1996), " Premarket Forecasting of Really-New Products, " Journal of Marketing , 60 (1), 47 – 60.
Urban Glen L. , Hauser John R. , Roberts John H.. (1990), " Prelaunch Forecasting of New Automobiles, " Management Science , 36 (4), 401 – 21.
Weingarden Matt. (2018), " Call for Papers | Journal of Marketing Special Issue: New Technologies and Marketing, " American Marketing Association (November 1) , https://www.ama.org/2018/11/01/call-for-papers-journal-of-marketing-special-issue-new-technologies-and-marketing/.
Xiong Guiyang , Bharadwaj Sundar. (2014), " Prerelease Buzz Evolution Patterns and New Product Performance, " Marketing Science , 33 (3), 401 – 21.
~~~~~~~~
By Nathalie Harz; Sebastian Hohenberg and Christian Homburg
Reported by Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 134- Visual Elicitation of Brand Perception. By: Dzyabura, Daria; Peres, Renana. Journal of Marketing. Jul2021, Vol. 85 Issue 4, p44-66. 23p. 3 Color Photographs, 2 Black and White Photographs, 1 Illustration, 5 Charts, 3 Graphs. DOI: 10.1177/0022242921996661.
- Database:
- Business Source Complete
Visual Elicitation of Brand Perception
Understanding consumers' associations with brands is at the core of brand management. However, measuring associations is challenging because consumers can associate a brand with many objects, emotions, activities, sceneries, and concepts. This article presents an elicitation platform, analysis methodology, and results on consumer associations of U.S. national brands. The elicitation is direct, unaided, scalable, and quantitative and uses the power of visuals to depict a detailed representation of respondents' relationships with a brand. The proposed brand visual elicitation platform allows firms to collect online brand collages created by respondents and analyze them quantitatively to elicit brand associations. The authors use the platform to collect 4,743 collages from 1,851 respondents for 303 large U.S. brands. Using unsupervised machine-learning and image-processing approaches, they analyze the collages and obtain a detailed set of associations for each brand, including objects (e.g., animals, food, people), constructs (e.g., abstract art, horror, delicious, famous, fantasy), occupations (e.g., musician, bodybuilder, baker), nature (e.g., beach, misty, snowscape, wildlife), and institutions (e.g., corporate, army, school). The authors demonstrate the following applications for brand management: obtaining prototypical brand visuals, relating associations to brand personality and equity, identifying favorable associations per category, exploring brand uniqueness through differentiating associations, and identifying commonalities between brands across categories for potential collaborations.
Keywords: brand associations; brand collages; branding; image processing; latent Dirichlet allocation; machine learning
Understanding how consumers perceive brands is at the core of brand management. It helps managers develop and position new products, understand the competitive landscape, and create effective marketing communications. Brand perception is often conceptualized as an associative network, where concepts related to the brand attributes, benefits, and attitudes are represented as memory nodes. [15] argues that these associations are diverse: they can relate to the brand's functional benefits, to its symbolic value, to the marketing-mix elements, to consumer experiences and attitudes, and to usage situations. The favorability, strength, and uniqueness of these associations determine the brand's position relative to other brands, its competitive advantage, and its brand equity. In this framework, a brand manager's task is to manage the associations—that is, strengthen desired associations and weaken undesired ones. Because consumers can associate a brand with any number of objects, emotions, activities, sceneries, and concepts, it is challenging to elicit and measure them in an interpretable way across brands and individuals.
Ideally, a comprehensive elicitation of brand associations should have several properties. First, it should not require predefining the set of associations of interest but rather should elicit them in an unaided way. Second, it should be scalable and quantitative to allow for monitoring a large number of respondents and brands. Third, to minimize the effect of intervening variables, the elicitation task should directly ask respondents for their associations rather than tease them out from a secondary source such as social media. The existing methods for obtaining brand associations are broadly categorized into quantitative surveys, qualitative surveys, and social media mining. Quantitative surveys (e.g., brand personality [[ 1]], brand equity [[26]]) are perhaps the most widely used. They typically define several theoretically driven brand attributes, on which participants are asked to rate brands. While these methods are scalable and quantitative, they are not ideal for free, unaided mining of associations. Qualitative surveys, such as collage methods ([43]) or association maps ([13]), are known to elicit a broad, diverse, and detailed range of associations; however, they are costly to apply on a large number of brands and respondents and cannot generate quantitative assessment.
The proliferation of online social media platforms has enabled scalable, quantitative, and unaided brand tracking by mining user-generated content (UGC) ([ 5]; [16]; [20]; [21]; [27]; [28]; [29]; [39]). However, for understanding consumer associations, UGC suffers from some shortcomings. First, it is available for only certain categories; whereas Nike generates a lot of social media chatter, social media posts about Colgate, for instance, are less abundant ([22]). Second, it is difficult to control for the characteristics of the content contributors. For example, users with a stronger relationship with the brand ([19]) or those who hold a particularly strong positive or negative opinion may contribute more ([22]). Finally, even a given consumer who contributes brand content may not offer their true opinion of the brand: consumers may post strategically to signal about themselves to the public ([ 9]; [22]) and serve their self-presentation needs ([37]).
In this article, we propose an elicitation that is direct, unaided, scalable, and quantitative and use it to retrieve the associations of a large number of national U.S. brands. Our elicitation consists of a platform and an analysis methodology. Inspired by qualitative elicitation approaches in psychology and marketing, we developed an online brand visual elicitation platform (B-VEP) that asks respondents to create an online collage of images representing their relationship with the brand. Participants can choose photos for their collages from a broad repository of tens of thousands of photos, using free browsing as well as keyword search. We analyze the collages using a machine-learning back end to derive brand associations at the individual-respondent level. The content extraction combines several machine-learning algorithms: image tagging, word embedding, and topic modeling. The combination of word embedding and topic modeling is a unique contribution of this research. By using unaided elicitation and unsupervised learning algorithms, we do not limit the dimensions on which the brand perceptions are measured. Unlike most existing unaided surveys, our approach allows for scaling to a large consumer population.
We use the proposed approach to elicit the brand associations of 303 major national U.S. brands using 4,743 collages from 1,851 respondents. We retrieve 150 brand associations relating to objects, actions, adjectives, characters, places, sceneries, concepts, and metaphors, on which all of these brands are mapped, to form the equivalent of a very high-dimensional perceptual map. Figure 1 presents three sample brands from our data—Axe, Degree, and Secret—and their most frequently occurring associations. Note that these associations relate to attributes, benefits, and attitudes ([15]) that go beyond the standard dimensions of brand personality and brand equity. Although the three brands describe functionally similar deodorants, each brand has a distinctive set of associations: Axe is associated with fashion, urban youth, astronomy, and bodybuilding. Degree has athletic associations, such as running, water, sports, and fitness. Secret's associations are more romantic and delicate, including lingerie, rain, and beauty salon.
Graph: Figure 1. An illustration of the strongest associations,in decreasing order, for Axe, Degree, and Secret.
We demonstrate the power of these findings through several potential applications for the creative and strategic functions of the brand management team. First, we show how to create prototypical collages by indexing photo repositories according to their fit to a given brand's associations. These collages can serve as mood boards to help graphic designers generate visual brand content and to visually convey the brand's associations. Second, we relate the associations to the well-established brand personality ([ 1]) and brand equity ([23]; [26]) metrics, such that each metric has a clear, specific set of related associations (e.g., the "wholesome" metric is associated with herbs, baby, winter, happy nature, and insects; the "masculine" metric is associated with bicycle, military, heavy vehicle, auto racing, and photography; see the Web Appendix). Third, we relate the associations to brand favorability to identify desirable and undesirable associations in each of the nine product categories in our data. Fourth, we show how to measure brand uniqueness relative to its category—namely, what consumers associate with the brand significantly more or less than with other brands in its category. Finally, we show how to use the distance in the association space to detect potentially valuable commonalities between brands (e.g., for potential collaborations).
Our methodology shares some elements with Zaltman's metaphor elicitation technique (ZMET), a collage-based interviewing technique ([43]; [44]). In ZMET, participants are asked to create a collage of pictures to represent how they view a brand. The method, which has been widely used by practitioners ([ 3]), argues that consumers store a rich visual representation of their relationship with the brand, and these relationships can be efficiently elicited by creating collage metaphors ([43]). Like other qualitative direct-elicitation approaches (for a review, see [38]]), ZMET results in data that are less directed by consumers' strategic goals when posting on social media, can be applied for any brand, and can be used to gather responses from a controlled sample of consumers. ZMET also has the advantage of being fully unaided and free-form, allowing consumers to express their views in terms of a wide range of concepts. However, because it requires the presence of an interviewer, it is costly to conduct at scale.
The basic premise behind using visuals is that although the exact representation of brand associations in the human brain is not known, thoughts occur, in many cases, as images and visual metaphors. Therefore, visual research methods are considered to better reflect the emotions, cultural experiences, and attitudes that constitute the associations, in contrast to verbal methods, which focus more on the discourse of these experiences ([36]). In addition, use of images has been demonstrated to successfully disrupt well-rehearsed narratives ([36]) and thus might be effective in revealing hidden, often unarticulated associations and ideas.
The practice of using images to reveal brand associations is supported by the extensive use of visual stimuli by firms ([42]). The human ability to process pictures and images ([18]; [30]) and to associate them with feelings and emotions ([ 4]) makes visual elements a key factor in brand communications ([24]; [42]). Research has shown that visual elements such as product packaging ([ 8]), store design ([25]), graphic design of ads ([33]; [35]; [41]), and the visual context in which the brand is displayed ([ 4]) have considerable impact on consumers' responses to brands.
Our contribution is methodological, substantive, and managerial. Our methodology, consisting of an elicitation platform and an analysis procedure, is unaided, scalable, quantitative, and direct. We validate the associations and show that our method is superior to free verbal elicitation. Substantively, we obtain, for the first time, a detailed set of associations for 303 major U.S. brands from nine product categories. Our associations contain objects, actions, constructs, occupations, sceneries, and institutions. Managerially, we show how B-VEP can aid and enhance the creative and strategic functions of the brand management team. The creative teams can use B-VEP to index photo repositories and generate visual brand content to convey the brand's associations. They can also connect each brand metric (e.g., "young," "confident") to sets of visuals. Strategically, insights from B-VEP can be used to manage brand health by relating the associations to brand favorability and identifying desirable and undesirable associations in each category. They can also help monitor the brand's unique positioning in its category. Finally, by identifying brands in different categories with similar associations, B-VEP can be used as an aid to suggest strategic alliances.
Measuring how consumers perceive brands has received much attention in the academic literature, as well as in practitioners' best practices. We review some of these methods next and summarize them in Table 1.
Graph
Table 1. Mapping the Literature on Brand Perception Measurement Methods: Description and Benefits.
| Research | Elicitation Type | Data | Method and Type of Associations | Text/ Visual | Scalable | Used on Any Brand | Sample Selection | Robust to Strategic Posting | Discover Dimensions |
|---|
| Aaker (1997) | Direct elicitation: survey | 631 questionnaires on 37 brands | Rating brands on five dimensions of brand personality: sincerity, excitement, competence, sophistication, and ruggedness | Text | ✓ | ✓ | ✓ | ✓ | |
| Mizik and Jacobson (2008) | Direct elicitation: survey | BAV database | Rating brands on four dimensions of brand equity: differentiation, relevance, esteem, and knowledge | Text | ✓ | ✓ | ✓ | ✓ | |
| John et al. (2006) | Direct elicitation: qualitative | 194 concept maps for Mayo Clinic | Arranging predefined sets of 25 associations into maps and depth interviews | Text | | ✓ | ✓ | ✓ | ✓ |
| Zaltman and Coulter (1995) | Direct elicitation: qualitative | 20 projects | Collages and depth interviews | Visual | | ✓ | ✓ | ✓ | ✓ |
| Steenkamp and Van Trijp (1997) | Direct elicitation: qualitative | Interviews with 99 female participants | Comparing three attribute generation methods: free elicitation, hierarchical dichotomy, and repertory grid | Text | | ✓ | ✓ | ✓ | ✓ |
| Nam, Joshi, and Kannan (2017) | UGC | Social textual tags for Apple from the Delicious app | Co-occurrence of words to extract nine associations, including functional and intangible attributes (e.g., hardware, computer, design, fun, cool); individual and aggregate respondent levels | Text | ✓ | | | | ✓ |
| Culotta and Cutler (2016) | UGC | 239 brands on Twitter | Similarity in the follower set with "exemplar" brands; intangible brand attributes: ecofriendliness, luxury, and nutrition | Text | ✓ | | | | ✓ |
| Klostermann et al. (2018) | UGC | Images, text, and tags for McDonald's from Instagram | Clustering image content; 20 associations: attributes, animals, objects, and sceneries | Visual | ✓ | | | | ✓ |
| Netzer et al. (2012) | UGC | Consumer forums on car and drug brands | Co-occurrence between brands, other brands, and topics; functional and usage attributes | Text | ✓ | | | | ✓ |
| Liu, Dzyabura, and Mizik 2020 | UGC | 56 apparel and beverage brands on Instagram | Creating visual representations for "glamorous," "healthy," "fun," and "rugged" | Visual | ✓ | | | | |
| Gelper, Peres, and Eliashberg (2018) | UGC | Prerelease social media posts from 157 movies. | Using MTurk to classify posts on movie dimensions: topic, tone, sentiment, and event drivers | Text | ✓ | | | | |
| Lee and Bradlow (2011) | UGC | Digital camera reviews over 3.5 years | Word vectors created from pro/con statements; functional attributes: battery life, optical zoom, and photo quality | Text | ✓ | | | | ✓ |
| Tirunillai and Tellis (2014) | UGC | Online reviews from 16 brands in 5 categories | Topic extraction to derive positive and negative attributes; functional dimensions | Text | ✓ | | | | ✓ |
| This article | Direct elicitation: survey | Online collages created by users on 303 U.S. national brands | Online collage creation, image tagging, topic modeling (guided LDA). Topic distributions obtained at individual respondent level. 150 associations: objects, actions, adjectives, characters, places, sceneries, concepts, and metaphors. | Visual | ✓ | ✓ | ✓ | ✓ | ✓ |
40022242921996664 Notes: UGC = user-generated content.
Traditional brand perception methods approach respondents directly, asking them for their perceptions of the brand. Some of these methods are surveys in which respondents rate the brands on sets of theoretically derived predefined attributes. The brand personality scale ([ 1]) rates brands on sets of five personality traits: sincerity, excitement, competence, sophistication, and ruggedness. The BrandAsset Valuator (BAV) scale, developed by Young & Rubicam, rates brands on four dimensions (differentiation, relevance, esteem, and knowledge) that have been shown to relate to brand financial performance ([26]) and to the volume of the brand's online and offline word of mouth ([22]). Survey methods have become very popular due to their clear advantages: they are scalable for a large number of brands and respondents, can be applied to any brand, and enable the researcher to choose the sample according to the research needs (e.g., brand loyalists, potential users, a specific target market). The main drawback of surveys is that they require the researcher to predefine a set of attributes and thus cannot be used to discover new dimensions and associations.
To reveal new brand associations from consumer responses, researchers have developed qualitative methods. These methods usually involve a one-on-one interview, a detailed protocol for how the interview should be conducted, and post hoc guidelines for interpreting the data. Similar to surveys, some qualitative methods use a predefined set of attributes. For example, [13] presented respondents with a set of 25 associations derived from conversations with consumers and marketing professionals and asked them to arrange these associations into a map. Other methods, such as free elicitation (in which respondents are asked to describe the relevant dimension of the brands within a product category), hierarchical dichotomization (in which respondents classify brands into groups based on perceived similarity), and the repertory grid (in which respondents are asked to indicate similarity between triads of brands) enable researchers to elicit relevant attributes from the respondents (see [38]).
A notable qualitative technique, which served as a motivation for this article, is ZMET ([43]; [44]). In ZMET, participants are given seven to ten days to either take their own photographs or cut out pictures from books and magazines and arrange them into a collage describing how they view a brand. Then, respondents sit for a guided one-on-one conversation with an interviewer to describe their collage. The method, which has been widely used by practitioners ([ 3]), argues that consumers store a rich visual representation of the brand's associations and metaphors, and creating collages is an efficient method for eliciting them ([43]). [43], p. 40) state that because consumers create their own collage, rather than being presented with stimuli by the researcher, it is the consumers themselves (rather than the researchers) who are "in control of the stimuli used in the guided conversation." The salient advantage of the qualitative techniques is that, being less restrictive and unaided (or nearly so), they do not constrain the respondents to dimensions predefined by the researchers and therefore generate new sets of associations and concepts. However, being qualitative, they are costly to scale and cannot be used to generate quantitative measures. In B-VEP, we aim to combine the advantages of the unaided, less restrictive elicitation methods with the scaling and quantification of the survey approaches.
Recently, consumers have begun contributing a large quantity of brand-related content on social media outlets, such as Twitter and Instagram. These data have the advantage of scalability due to the abundance of data contributed by consumers. These data are also unaided, as consumers are free to discuss any topic. A stream of research has developed ways to use this UGC for deriving valuable insights on product and brand perceptions. Researchers have used text data such as reviews ([20]), blogs ([ 7]), microblogs ([ 5]), social tags ([27]; [28]), and discussion forums ([29]), as well as visual social media content ([12]; [21]; [31]), for this purpose.
Some of the aforementioned studies exploited the richness of the data to apply unsupervised algorithms to derive relevant associations. Such associations include functional attributes (e.g., for cameras: battery life, optical zoom, photo quality [[20]]; for cars: full warranty, roomy, highway mileage, cargo capacity [[29]]; for Motorola mobile phones: instability, portability, receptivity, compatibility [[39]]) and some intangible brand attributes (e.g., for Apple: design, fun, cool [[27]]).
Other research has used these data to query for predefined attributes of interest. Such attributes include brand functional attributes (e.g., in UGC on movies: opinion, call for action, actor, storyline [[ 7]]) and intangible attributes (e.g., ecofriendliness, luxury, nutrition [[ 5]] and glamorous, healthy, fun, rugged [[21]]). [12] extract color features such as hue, color, and brightness from brand Instagram photos and relate them to click-through rates. [16] combined Instagram images, post text, and tags for McDonald's. Rather than querying for predefined attributes, they conduct unsupervised clustering on the image labels to extract brand associations. Their article demonstrates the power of unsupervised analysis on visual data: while the UGC text contains associations related to brand functional and intangible attributes, unsupervised analysis of images generates a broad spectrum of associations for the brand, ranging from burger and McCafé to cartoon and urban.
As explained previously, UGC data have several shortcomings: they are available for some brands but not others; they do not enable the researcher to easily select the sample according to the research goals (e.g., those who purchased the brand, those who are in the market for a car); and, finally, they may lack validity, as consumers do not simply respond about their brand perceptions but also strategically signal social and personal cues to their own target audiences.
Our goal in this article is to build on the aforementioned literature to create a platform and conduct a large-scale brand-mapping process that combines the following five benefits: ( 1) it is fairly robust to strategic posting, ( 2) it allows for flexibility and control over the sample as needed, ( 3) it can be applied to any brand, ( 4) it is able to discover new dimensions of associations from the data, and ( 5) it should be scalable to any number of brands and respondents. Table 1 describes selected research along these benefits and positions B-VEP's contribution. Our elicitation task is direct: it asks respondents explicitly for their associations, and thus it is less subject to strategic signaling by consumers and can be applied to any brand and sample. The data collection using online collages is unaided, and the analysis uses unsupervised machine learning to extract a data-driven set of diverse associations that go way beyond functional or intangible attributes. The online data collection and the quantitative analysis make the method scalable to a large number of brands and respondents.
Following the large body of literature demonstrating the power of visuals in extracting deep metaphors and depicting consumers' attitudes, moods, and associations ([12]; [21]; [31]; [36]; [43]; [45]), our method uses data of visual images. However, extant visual methods either were qualitative ([43]) or used predesigned sets of attributes ([12]; [21]; [31]; [45]). Our elicitation is also among the first (see also [16]) to provide unsupervised extraction of associations from visual data.
Our main data-collection tool is a software platform that we developed, on which consumers can create collages for brands. Collage creation is an expressive technique that has been used in psychology ([17]) and marketing ([43]; [44]). Collage making is an unaided visual elicitation technique that helps uncover hidden associations and emotions that could have remained undetected by other techniques ([ 6]; [17]) and therefore is appropriate for eliciting visual brand representation. Although collage making is traditionally a qualitative research method, we develop an online collage-creating platform that can be used for a large number of brands and respondents and analyzed quantitatively.
The collage-making procedure was conducted as follows: a respondent was shown several instruction screens explaining how to create the collage. Then, they were assigned a brand and asked to think, "What are your emotions, associations, and expectations with respect to the brand? What does the brand mean to you? Recall your experiences with the brand. What are the colors associated with the brand? What shapes? What objects? What images?" Next, the respondent was taken to the collage-making screen. Figure 2 provides a screenshot of the screen on which the collages were created.[ 5] The screen is divided into two sections: The left-hand side is the "canvas" on which the respondent creates the collage (in Figure 2, the brand is Starbucks), and the right-hand side contains a large repository of photos for the respondent to choose from. Respondents could drag photos from the right- to the left-hand side to create the collage onto the canvas. They could move, resize, and rotate the images once they had dropped them on the canvas. Respondents were able to either scroll through photos randomly or search for keywords and retrieve photos relevant to that keyword. For example, in the screenshot in Figure 2, the user searched for the keyword "laptop."
Graph: Figure 2. The collage canvas.Notes: The photo repository is on the right side of the screen, and the canvas is on the left.
The photo repository is a key element in the platform. First, it should be large and diverse enough for respondents not to feel constrained by the images and to be able to accurately convey their perception of the brand with the available images. Second, the images should prompt the respondents to think about the entire spectrum of associations, beyond the product-related attributes or the obvious brand elements, so that a collage for Levi's, for example, will not simply be a collection of photographs of jeans or the Levi's logo.
With these goals in mind, we created the photo repository and designed the right-hand side of the screen. We began by downloading a large set of photographs from Flickr, a photo-sharing website that allows users to label the photographs that they upload or view. To make the repository as rich as possible, and to ensure that participants could find photographs that represented what they were trying to communicate, we queried Flickr's application programming interface for the top 4,000 nouns, verbs, and adjectives in English and downloaded the first 50 photo results for each. The resulting image database consists of ∼100,000 unique photos (many photos have multiple labels).
We also embedded a search feature on the platform that returns photos, in randomized order, that have the labels of the queried term on Flickr. For example, in Figure 2, the user searched for "laptop." The ability to retrieve photos by search terms helped respondents tailor the collage to better represent their brand perception. As each search term retrieved many photos (e.g., 46 photos labeled with "laptop," 318 labeled with "family," 3,621 labeled with "nature"), the search option did not limit the users but rather was used as an initial aid in browsing through the repository. We wanted to ensure that the collage represented the respondent's perception of the brand beyond simply the product category and the company's own marketing efforts. We also wanted to encourage respondents to retrieve personal and meaningful associations. To that end, we constrained the terms for which respondents could search. The system does not allow them to search for the brand itself, the category, or the product type. If they did, they saw an error message, saying the term was not allowed as a search term for this brand. For example, when creating a collage for Levi's, the user would not have been able to search for "Levi's," "clothing," "apparel," or "jeans." Research assistants manually generated the list of these "banned" keywords for each brand.
Each respondent was assigned elicitation tasks for three brands sequentially. To ensure that respondents created collages only for brands with which they were familiar, respondents first had to rate their familiarity with ten brands on a five-point scale (1 = "not at all familiar," and 5 = "very familiar"). Three focal brands were selected randomly from those that the respondent rated as a 4 or 5. If a respondent was not familiar with any of the brands, another set of ten brands was presented, and if, after three sets of ten brands, no brand was scored a 4 or 5 on familiarity, the survey terminated for that respondent.
Respondents were encouraged to spend as much time as needed to create a thoughtful collage. If a respondent submitted a collage after less than two minutes, or if the collage contained fewer than six photographs, a pop-up screen appeared asking them if they were sure they wanted to submit. After submitting the collage, respondents were asked to score the task's level of difficulty on a five-point scale, with 5 being very difficult. To ensure that respondents understood the task, respondents were also asked to briefly describe the collage and explain their choice of images. Finally, research assistants checked each collage manually[ 6] and removed the data if the participant did not appear to have invested sufficient effort in the collage. The criteria for deletion were to delete collages that took less than one minute to make, that used only one or two photos, and for which the responses for the brand characteristics were identical for all items (e.g., respondent rated the brand only 1 or only 5 on all 49 items). In total, 17% of the collages were removed.
Designing the software platform was a major undertaking. Its user-friendliness and clarity were essential to engaging respondents and obtaining high-quality collages. The user interface was designed following design best practices ([14]), using professional web designers. All screen, instruction, and error messages were extensively tested for comprehensibility by an internal team of 10 users and an external beta test team of 50 Amazon Mechanical Turk (MTurk) users.
Respondents for the task were recruited on MTurk and received $2.50 for completing the entire task. Although our sample was not created to be demographically representative, it is quite balanced, skewed toward younger adults. In total, all of our respondents were U.S. residents, 43.5% were male (56.5% female), 26% were 18–29 years old, 41% were 30–39 years old (the age group 18–39 forms 36.5% of the U.S. population), and 33% were 40–69 years old (this age group forms 54% of the U.S. population). Note that we used MTurk as a proof of concept and a means to recruit a large number of respondents from the general population. If needed, a firm could use a more representative sample of respondents. Each respondent completed the task for up to three brands, or 30 minutes, whichever came first. That is, if the respondent was only finished with their first or second brand after 30 minutes, they were taken to the final screen, which thanked them for participating and terminated the study. The time limit helped us avoid fatigued respondents.
We collected 4,743 collages from 1,851 respondents (3,937 were approved by the research assistants). The data include an average of 15.6 collages per brand, for 303 national U.S. brands from nine categories: beauty (40 brands), beverages (65 brands), cars (29 brands), clothing (23 brands), department stores (17 brands), food and dining (84 brands), health products and services (10 brands), home design and decoration (16 brands), and household cleaning products (19 brands). The brand list is an updated version of [22], excluding TV shows, video games, movies, and since-discontinued brands. Web Appendix A presents the full list of brands. The average collage took eight minutes to create and included 11.45 photos. The average reported level of difficulty of creating the collages was 2.5 (on a 1–5 scale).
Mostly, respondents used mixed methods of browsing through the photo repository and searching for specific terms. The search feature was not always used: 690 collages (17.5% of the approved collages) did not use it at all. The median number of search terms used in a collage was 5, and the average was 6.4. In addition, respondents did not make many attempts to use the "banned" words: of the 25,262 search terms used, only 1,111 (<5%) attempts were made to use the "banned" words. To further verify that the search function did not restrict or bias the collages, we compared, for each brand, the associations derived from the brand's collages that used an above-median number of search words with collages in which the number of search words used was below median. We found that this specific split was not significantly different from any random split of collages (for details, see the Appendix).
Figure 3 presents a sample of four collages for the brand Starbucks, from four different respondents, along with the verbal description. The collages contain rich, meaningful information about the brands' associations: they do not simply show people drinking coffee or images related to Starbucks' brand elements. At first glance, these collages appear to be very different from each other, without obvious commonality. However, as we show next, they share specific visual elements that create a distinctive set of associations that is unique to Starbucks and differentiates it from the other brands in the sample as well as from brands in its category. Next, we discuss how associations are extracted.
Graph: Figure 3. Examples and verbal descriptions of four collages for Starbucks, made by four different respondents.
In addition to the collage, we collected data on respondents' perceptions of the brands on well-established brand metrics. After completing each collage, respondents rated the brand on each of 49 items on a five-point scale. The set of items is the combination of [ 1] personality dimensions and BAV brand equity items ([23]). The items were presented in a randomized order. We consulted with a BAV team to operationalize the survey as closely as possible to how they operationalize theirs. A major difference is that BAV's survey is conducted on a representative sample, whereas our sample, as explained previously, is not truly representative. The correlation between the average brand score on each BAV item in our survey and the scores that we received from BAV 2016–2017 data for these brands is.58 (p <.05). This correlation is high, given that the survey was conducted in a different format, included additional items, surveyed a different population (e.g., those who indicated high familiarity with the brand), and was done a year later.
The collage-creation task generated a set of collages for each brand. Our goal was to extract and summarize interpretable associations from the collages and organize them into a single, unified space on which all brands can be mapped and analyzed. To do so, we use image-tagging to extract the visual elements of the collages and identify patterns among tags in brand collages.
In many image-processing applications (e.g., [21]), the goal is to solve an image-classification problem. The visual features extracted from images do not have to be interpretable and typically include low-level features such as edges, corners, color histograms, shapes, line directions, and texture, or even more abstract deep-learned features. Our goal in this article is different from previous applications: we do not use the visual elements as an intermediate stage in solving a prediction problem. Rather, we look for what associations set one brand's collages apart from others, thus creating mapping from visuals to brands and brand characteristics. Therefore, we are interested in extracting and summarizing interpretable features. For this reason, we turn to image tagging.
We used a commercially available image tagging tool called Clarifai ([34]), which is pretrained on a corpus of millions of photos and uses deep convolutional neural networks to classify the content of photos into over 11,000 semantic tags (labels) relating to the objects, scenery, actions, emotions, adjectives, and other visual elements ([10]). Clarifai offers several options for pretrained models, of which we used "general 1.3." Each photo is assigned the 20 tags with the highest confidence scores. For example, the photo at the bottom left of the bottom-right collage for Starbucks in Figure 3, showing men in a running competition, is tagged with athlete, competition, race, runner, marathon, track and field, jogger, running, athletics, fitness, action, energy, exercise, footrace, hurry, endurance, motion, effort, jog, man, and sport. The 4,743 collages in our data set contain 91,856 photos, yielding 5,426 unique tags (the approved 3,937 collages had 4,601 unique tags). We next extract the associations in each collage and compare the associations across brands and individuals.
We analyzed the tags using a topic modeling approach called guided latent Dirichlet allocation (LDA), a semisupervised variation of the popular unsupervised topic-modeling algorithm, LDA ([11]). LDA is a widely used text-mining approach that discovers topics in documents (e.g., research articles, books, news articles). Each topic is a sparse Dirichlet probability distribution over all of the words in the vocabulary, and each document has a probability distribution over all topics. The model is estimated using an iterative Bayesian approach, with one parameter being updated in each iteration. Typically, this procedure is initialized with a uniform distribution, with all words being equally likely to occur in all topics.
We treat each collage as an individual document and a tag as a word. On average, each collage contains 11–12 photos with 20 tags each, resulting in a very short document for training the LDA. We avoided aggregating collages of a single brand into one document, as we wanted to extract associations at a collage level to be able to relate the associations to individual brand perceptions. To overcome the challenge of the short documents, we used guided LDA. The guided LDA method changes the priors of certain words to increase the probability that they have a high weight in one topic. For example, one might seed the words "girl," "boy," and "child" toward being included in topic 1. As we show subsequently, we chose the priors using a method that groups words according to their linguistic similarity. This enabled us to incorporate knowledge of word meanings, which LDA alone does not take into account.
We obtained the sets of seed words for the guided LDA by computing a word embedding for each tag and clustering the tags in the embedded space. We used Stanford's Global Vectors tool (GloVe), an unsupervised algorithm pretrained on over 6 billion text tokens from Wikipedia and the linguistic data from English GigaWord, 5th edition. During the training phase, the algorithm uses global matrix factorization methods, in combination with local context window methods, to create a 300-vector dimensional space ([32]). The algorithm takes into consideration factors such as word-to-word co-occurrence, context similarities, and word analogies. We used this 300-dimensional space provided by GloVe as input to our analysis and represented each tag in our data set as a point in this space. As is common in text mining, we removed the most and least frequently occurring tags. Specifically, we removed tags occurring fewer than 10 and more than 2,000 times in the corpus, resulting in a total vocabulary of 2,596 unique tags (out of the original 4,601).
We then clustered the resulting vectors using a k-means clustering algorithm (Scikit-learn machine learning Python package). The role of the clustering step is to generate seeds for the guided LDA. We did not require a full clustering of the vocabulary; rather, we wanted to identify groups of only very similar words. Words that do not have linguistically similar neighbors do not need to be included. Therefore, we began with a large number of 465 clusters (10% of the vocabulary). Naturally, this resulted in many spurious clusters. We removed clusters that occurred in fewer than 50 collages and fewer than 6 tags, leaving us with 120 word clusters to use as the priors (presented in Web Appendix B). In this procedure, we balanced the need to obtain meaningful seeds on the one hand without allowing the seeds to dominate the collage data.[ 7]
The guided LDA process can be described as the following generative model: in regular LDA ([ 2]), for a vocabulary of tags of size , containing documents (the set of tags for a single collage in our case), each tag is drawn from a topic-tag distribution that is multinomial with parameter , corresponding to a given topic , . is a vector of size , drawn from a Dirichlet distribution, . can be interpreted as the probability of generating each of the tags in the vocabulary, given topic k. The topic k is drawn from a document-topic distribution for collage , which is also multinomial with parameter . is a vector of size , drawn from a Dirichlet distribution , and can be interpreted as the probability of generating each of the topics, given collage . The model cycles through the documents (the first tag of a collage, the second tag, etc.), and generates a topic and a tag from that topic. Thus, it generates an entire set of tags for the collage. The parameters α, β, ϕ, and θ are optimized to maximize the likelihood of the observed data. The estimation is performed via collapsed Gibbs sampling.
In guided LDA, a topic is a mixture of two multinomial distributions: a "seed-topic" distribution and a "regular-topic" distribution . The seed-topic distribution is constrained to generate words from a corresponding predefined set . In our case, a seed set is a cluster, resulting from the previous stage, and there is a total of such clusters, while the regular-topic distribution can generate any tag from the vocabulary . The probability of drawing a tag from the seed-topic distribution versus the regular-topic distribution is determined through the seed-confidence parameter , which we took to be.3.[ 8] Following [11], the process can be described as follows:
- Input: , , .
- For each topic .
- Draw regular-topic .
- Draw seed-topic .
- For each collage , draw .
For each in the collage: \\ is the number of tags in collage .
- Draw a topic .
- Draw an indicator \\ select the regular or the seed topic distribution.
- If is 0: Draw a tag .
- If is 1: Draw a tag .
For our analysis, we used the implementation of this process of the Python library guidedlda2.0.0.dev22.
The guided LDA provides two outputs: ( 1) a set of topics, each topic being a distribution over tags, and ( 2) the distribution over topics for each collage. The main parameter to be set by the researcher is the total number of topics . We experimented with various values of this parameter ranging from 10 to 200. We looked for parameter values that maximize the model's likelihood (similar to [27]]). We also qualitatively checked the resulting topics to ensure that they achieve the balance between insufficiency and redundancy—on the one hand, no topic/cluster is a mixture of multiple semantic topics; on the other hand, no two topics/clusters are too similar and could be merged into a single group. We ended up with 150 topics.
We named each of the 150 topics manually, using three research assistants, all English literature majors, based on the tags with the highest probability of the topic, to ensure that topic names were meaningful. These 150 topics form a rich set of brand associations, including objects (e.g., animals, food, people), constructs (e.g., abstract art, horror, contemporary, delicious, famous, fantasy, illness), occupations (e.g., musician, bodybuilding, baking), nature (e.g., beach, misty, snowscape, wildlife), and institutions (e.g., corporate, army, investment, school). In [15] terminology, these associations represent product-related attributes (e.g., alcoholic drinks); non-product-related attributes (e.g., baby, holiday party); functional, experiential, and symbolic benefits (e.g., fitness, cityscape, pop star); and attitudes (e.g., American flag).
Web Appendix C contains the distribution of tags in the association topics, as well as the topic names. For example, the aeronautics association topic has the tags air, flight, airplane, aircraft, flying, military, and jet, with probabilities 7.7%, 7.4%, 6.3%, 6.3%, 5.1%, 4.8%, and 3.9%, respectively. The cityscape topic has downtown, cityscape, skyline, skyscraper, modern, office, tower, and bridge tags, with probabilities 9.4%, 9.4%, 7.9%, 7.4%, 7.4%, 5.3%, 5.0%, and 4.4%, respectively. The running topic is composed of athlete, runner, race, action energy exercise, fitness, marathon, and jogger tags, with probabilities 7.8%, 6.4%, 5.9%, 5.8%, 5.7%, 5.7%, 5.0%, and 4.5%, respectively. These 150 association topics ("associations" hereinafter) constitute the set of dimensions on which we will map the brands. Note that the dimensions may change for a different set of brands.
To validate the results of the association extraction, we ran an additional study as follows. Participants were given a set of associations extracted from a collage, and two different collages to choose from, where only one is the correct collage (from which the presented associations were extracted). They were asked to indicate which of the two collages best matches the presented set of associations. Participants were recruited on MTurk and paid $1. A total of 46 participants completed the study, each completing 20 tasks, giving us a total of 920 choices. Of these, 784, or 85.2% were correct, which validates the association extraction algorithm.[ 9]
Note that our association extraction methodology combines two state-of-the-art text-mining methods in a novel way and is a unique contribution of this research. Recall that extracting associations from collages is a challenging problem, especially when allowing for a large number of topics, and off-the-shelf tools are not able to extract good associations. Although topic models such as LDA work well on long documents, such as books or articles, we only have about 200 image tags per collage. Moreover, the LDA model does not take into account the meanings of words but rather the co-occurrence of words in documents. Clustering word embeddings, in contrast, does not take into account the co-occurrence of tags in one collage. Using the word-embedding clusters to seed the guided LDA enables us to extract the most appropriate associations that take into account both types of information.
Indeed, the clustering alone (without the LDA) does not result in good associations. First, as with any clustering algorithm, it allows each tag to belong to only one association. However, a word can be a part of multiple associations. For example, "flowers" can belong to a romantic or wedding association, as well as a nature association. Clearly, a brand associated with nature differs from one associated with wedding. In LDA, each topic is a probability distribution over all the words in the vocabulary; therefore, each word is included in all the topics, with different weights. Second, embeddings are based on linguistic similarity, which does not necessarily mean that they are a part of the same association. For example, one of our resulting clusters (see Web Appendix B) included humor, creepy, amusing, affectionate, and erotic, which are all emotions occurring in similar contexts in ordinary texts but have different implications for branding. Third, clustered word embeddings do not use the fact that tags appear in the same collages (for example, "humor" and "erotic" are not together in the same collages). The LDA step provides that.
For each brand, we averaged the association distribution extracted from the brand's collages. Guided LDA outputs the probability of each of the 150 topics occurring in each collage. Let represent the probability with which association occurs in collage . We compute the average of the association distributions across the collages for a given brand, namely, , where is the set of collages for brand b. Table 2 presents the top 5 highest weighted associations for all of the brands in the beauty category. The results for all brands are presented in Web Appendix D.
Graph
Table 2. The Top Five Most Frequently Occurring Associations for Beauty Brands (in Decreasing Order of Probabilities).
| Brand Name | Most Frequent Associations (Top Five) |
|---|
| Always | Glamour | Therapy | Flowers (botanical) | Beach | Water |
| Aveeno | Fashion | Streams | Flowers (botanical) | Baby | Water |
| Avon | Glamour | Hand | Flowers (botanical) | Produce | Frosty |
| Axe | Fashion | Urban youth | Flowers (romantic) | Astronomy | Bodybuilding |
| Bath & Body Works | Flowers (romantic) | Water | Flowers (botanical) | Therapy | Fruits |
| Caress | Flowers (romantic) | Fruits | Water | Streams | Glamour |
| Chanel beauty | Flowers (romantic) | Lingerie | Geometric | Jewelry | Alcoholic drinks |
| Charmin | Water | Flowers (romantic) | Bedroom | Cat | Family |
| Clean & Clear | Beauty salon | Bathroom | Flowers (botanical) | Flowers (tropical) | Water |
| Clinique | Glamour | Hairstyling | Flowers (romantic) | Eye | Painting |
| Colgate | Water | Family | Herbs | Glamour | Child |
| CoverGirl | Flowers (romantic) | Glamour | Holiday party | Water | Child |
| Crest | Water | Power energy | Bathroom | Rain | Frosty |
| Degree | Running | Water | Sports | Fitness | Ball sports |
| Dial Soap | Water | Rainstorm | Ocean | Bedroom | Bathroom |
| Dove | Flowers (romantic) | Streams | Water | Warm fabrics | Erotic |
| Garnier Fructis | Fruits | Streams | Fashion | Flowers (romantic) | Flowers (botanical) |
| Gillette | Wedding | Suit | Sailing | Water sports | Modern building |
| Head & Shoulders | Hairstyling | Flowers (tropical) | Water | Juice | Beach |
| Herbal Essences | Flowers (botanical) | Rain | Hairstyling | Flowers (romantic) | Juice |
| Irish Spring | Streams | Water | Erotic | Mountain | Bathroom |
| Jergens | Flowers (romantic) | Baby | American flag | Birds of prey | Fruits |
| Kleenex | Rainstorm | Furniture | Child | Ocean | Baby |
| Kotex | Flowers (romantic) | Fashion | Child | Water | Glamour |
| L'Oréal | Birds of prey | Hairstyling | Glamour | Fashion | Church |
| Mary Kay | Glamour | Hairstyling | Fashion | Flowers (romantic) | Lingerie |
| Maybelline | Glamour | Eye | Hairstyling | Fruits | Lingerie |
| Neutrogena | Water | Flowers (romantic) | Flowers (botanical) | Hairstyling | Bathroom |
| Nivea | Glamour | Flowers (botanical) | Flowers (romantic) | Water | Lingerie |
| Olay | Flowers (romantic) | Flowers (botanical) | Glamour | Hairstyling | Water |
| Old Spice | Bodybuilding | Bathroom | Heavy vehicle | Cat | Running |
| Pantene | Hairstyling | Bathroom | Flowers (romantic) | Rain | Beach |
| ProActiv | Water | Hairstyling | Flowers (romantic) | Beauty salon | Produce |
| Revlon | Glamour | Fashion | Flowers (botanical) | Rainstorm | Modern building |
| Scott Tissue | Cat | Frosty | Birds of prey | Delicate fabric | Child |
| Secret | Flowers (romantic) | Lingerie | Rain | Running | Beauty salon |
| Sephora | Hairstyling | Flowers (romantic) | Fruits | Glamour | Water |
| Suave | Water | Flowers (botanical) | Rainstorm | Streams | Flowers (romantic) |
| Tampax | Beauty salon | Running | Lingerie | Flowers (romantic) | Fashion |
| TRESemmé | Hairstyling | Flowers (romantic) | Glamour | Fashion | Flowers (botanical) |
Consider, for example, the shampoo brands Garnier Fructis and Pantene. Both are associated with romantic flowers; Garnier Fructis has stronger associations for fruits and fashion, while Pantene is more strongly associated with hairstyling and bathroom. Of the beer brands (Web Appendix D), Budweiser is associated with ball sports, fire, water, auto racing, and youth and Corona with beach, ocean, breakfast, lingerie, and pool. Recall that each of these associations represents a large number (see Web Appendix C) of objects, concepts, emotions, and activities. The associations relate to the brand's product attributes; usage; users; functional, symbolic, and experiential benefits; and attitudes toward the brand. The association weights enable us to measure the strength of each association for a given brand. As we show in the "Applications for Brand Management" section, the association weights also allow for the calculation of brand uniqueness, relative positioning, favorability, and relationships with other brand metrics.
To validate the brand–association relationship, we ran an additional validation study that follows a similar format to the study used for collage validation. Participants were given a set of associations for a brand and two different brands to choose from, one of which is the correct brand (for which the presented associations were extracted). They were asked to indicate which of the two brands best matches the presented set of associations. Participants were recruited on MTurk and paid $1. A total of 91 participants completed the study, giving us 1,707 choices. Of these, 1,280, or 75% were correct, which validates the brand association relationship.[10]
We assessed split-half reliability (similar to [13]]) to determine the consistency of the elicited brand associations across respondents. We randomly divided the collages of each brand in half and calculated the average topic weights for all 150 topics. That is, each brand had two 150-dimensional topic distribution vectors, one for each split. We then computed the correlation across these two vectors. For 222 out of the 303 brands, the correlations are positive and significant at p <.05. The remaining correlations are not statistically significant. This is a high number given that the brands have only 7–10 collages in each half-split (the original sample size of 15 per brand was optimized to balance costs and stability of associations).
Table 2 indicates that associations such as flowers, water, and hairstyling are particularly prevalent in the beauty category. Table 3 presents category averages (i.e., the five most frequently occurring associations in each category) and their average probability of occurring in a collage. The results have face validity in that most of the high-probability associations are closely related to the category (e.g., "traffic" for cars, "furniture" for home design and decoration).
Graph
Table 3. The Top Five Most Frequently Occurring Associations for Each Category with Their Category Averages (in Decreasing Order of Probabilities).
| Category | Most Frequent Associations (Top Five) |
|---|
| Beauty | Flowers (romantic) | Water | Hairstyling | Flowers (botanical) | Glamour |
| .046 | .045 | .034 | .033 | .032 |
| Beverages | Water | Streams | Ball sports | Fruits | Ocean |
| .031 | .025 | .023 | .022 | .021 |
| Cars | Traffic | Car | Cityscape | Finance | Steel |
| .049 | .044 | .037 | .028 | .024 |
| Clothing | Fashion | Sports | Clothing | Band | Street art |
| .028 | .025 | .021 | .02 | .018 |
| Department stores | Retail | Finance | Clothing | Business school | School |
| .05 | .032 | .025 | .023 | .022 |
| Food and dining | Dining | Family | Youth | Baking | Child |
| .064 | .03 | .027 | .021 | .02 |
| Health products and services | Family | Hospital | Flowers (botanical) | Business school | Child |
| .032 | .03 | .027 | .026 | .024 |
| Home design and decoration | Furniture | Steel | House | Modern building | Water |
| .047 | .034 | .029 | .026 | .022 |
| Household products | Water | Flowers (romantic) | Furniture | Flowers (botanical) | Frosty |
| .05 | .03 | .029 | .029 | .024 |
To demonstrate the richness of associations elicited via images relative to those elicited via text, we created an elicitation tool[11] that is identical to B-VEP, except that instead of creating a collage, participants write a free-text paragraph describing their associations with the focal brand. The task is an online large-scale version of free elicitation described by [38].
Similar to the B-VEP task, participants were paid $2.50 for completing three brand descriptions. To ensure the quality of these text responses and to make our comparison as conservative as possible, we asked participants to write their own original text (not copy from the internet) and disabled the paste option on the page. Research assistants manually evaluated all responses and rejected those that were copied (by searching for the exact text). We collected these descriptions for the 40 brands in the beauty category, 5–7 descriptions per brand, for a total of 235 descriptions by 85 participants.
The key takeaway is that it is hard to obtain brand associations in this format. First, many participants found it difficult to write detailed descriptions. A typical description was 50–60 words (3–5 sentences). Second, the associations derived from this task are very similar to product reviews. Despite our clear instructions to "recall experiences" and describe the "objects, feelings, actions and images" associated with the brand, which successfully generated rich and diverse visual collages, in the textual setting, respondents tended to review the brand. Therefore, the descriptions are not distinctive and do not yield terms that are unique to the brand.
Specifically, participants focused on product/brand evaluation (e.g., "I like the smell," "brand that I can trust and rely upon"), product usage (e.g., "Nivea is a company that I use quite often," "I am regularly using this brand [Pantene]"), target audience (e.g., "Revlon makes beauty products for older people"), and functional attributes (e.g., "Garnier products are environmentally friendly").
Table 4, Panel A, presents the ten most frequently occurring terms for the three brands in Figure 1 (Axe, Degree, and Secret). The most frequent terms for all 40 brands appear in Web Appendix E. For Axe, the words "use" and "used" pertain to usage; "good," "love" (as in "I love its smell"), and "favorite" pertain to evaluation; and "body" (as in "body spray") is a functional attribute. For Degree and Secret, the words "frequently" and "using" pertain to product usage; "enjoys," "good," and "famous" (as in "famous brand") pertain to evaluation; and "women" is the target audience. A TFIDF analysis reveals that the most salient differences in verbal descriptions between brands are the brand name and product type.
Graph
Table 4. Top Ten Most Frequently Occurring Words in Verbal Descriptions.
| A: Verbal Free Elicitation | B: Verbal Description of the Collage |
|---|
| Axe | Degree | Secret | Axe | Degree | Secret |
|---|
| product | Degree | brand | Axe | Degree | think |
| Axe | brand | frequently | hard | fresh | Secret |
| brand | product | using | like | clean | like |
| use | work | really | day | makes | feel |
| good | believe | deodorant | men | think | makes |
| used | enjoys | usually | things | strong | happy |
| one | products | unique | brand | feel | life |
| favorite | higher | famous | young | use | deodorant |
| love | good | sold | good | cool | confident |
| body | women | women | looking | brand | fresh |
We also computed the most frequently occurring bigrams, or two-word sequences, in all of the text descriptions combined. The top five most frequent bigrams are "skin care," "good product," "good quality," "toilet paper," and "product used." This bigram analysis further demonstrates the focus of these descriptions on product quality and usage, much like a product review. While this is certainly valuable information, it is less relevant for brand association research.
These findings are consistent with the literature on brand associations: Table 1 indicates that most text-based elicitation methods (e.g., [27]; [29]) extract information on the brands' functional attributes and brand evaluation. In contrast, the associations generated by unaided visual methods ([16]; [43]) contain a broad range of objects, emotions, activities, sceneries, and concepts.
We also analyzed the verbal descriptions of the collages provided by B-VEP participants after completing their collage task. In B-VEP, after completing the collage, participants were asked to "describe how your collage relates to brand X." The sequence of completing a collage and then explaining it verbally was similar to the ZMET process. Table 4, Panel B, presents the top ten most frequently occurring terms in the collage verbal descriptions for the same three deodorant brands. The terms for all the brands are presented in Web Appendix E. Although some words relate to evaluation (e.g., "good," "like") and usage (e.g., "use"), there are also words such as "confident," "happy," "strong," and "cool," which relate more to brand associations and intangible attributes. While this is far from substituting for a direct analysis of the visual content of the collages, the mere fact that respondents wrote the verbal description after creating a collage results in more relevant associations than does the free-elicitation verbal task.
Recall that in B-VEP, participants select the images for their collages from a large photo repository (right-hand side of Figure 2). To aid in browsing, we implemented the option to search for keywords. Participants were not requested to use the search option; it was implemented as an aid to help users navigate through this very large repository. One might wonder if the search terms are the associations themselves: when respondents create their collage, they may search for their association, such as "flowers," to find corresponding pictures. However, as we show next and in the Appendix, the association elicitation process is more intricate, and constraining the analysis to the search terms generates only a small fraction of the association space.
First, search terms are not used in every collage: 17.5% of collages did not use the keyword search at all; their creators simply scrolled through the photo repository (in random order). The median number of search terms (e.g., "pine tree"), used in a collage is 5, and the average number of key terms used per collage is 6.4, which is low compared with the 11.45 photos and 229 extracted tags per collage on average. Second, the keywords used in the search are limited and repetitive. The top 30 words (.5% of total unique words) account for 17% of total searches. This limits the search words' ability to provide unique associations. Third, each search retrieves a large number of photos. For example, "family" retrieves 318 photos, "nature" retrieves 3,621 photos, "child" retrieves 629 photos, and "happy" retrieves 210 photos. The chosen photo contains additional information about the respondent's relationship with the brand, over and above the search term. Using only the search terms ignores this additional information, which, as our collage analysis shows, contains meaningful, distinctive associations. For more details on the search words analysis, see the Appendix.
We also tested whether the search terms provide additional information that is not contained in the collage itself. In line with the pooling approach suggested by [16], we appended the search words to the Clarifai image tags as an input to the LDA algorithm. The resulting distribution of topics does not change relative to having the image tags only.
B-VEP can be used to support both the creative function and the strategic function of the brand management team. For the creative function, we demonstrate how to create a prototypical collage, or a mood board, for each brand—that is, a collection of photos that together capture the average distribution of associations and provide a visual representation of the brand. This is done by indexing a photo repository and computing, for each set of photos, how closely it resembles the association distribution of the brand. We also use B-VEP to match brand commonly used brand metrics—brand personality ([ 1]) and brand equity ([26])—with brand association (e.g., the personality trait "charming" is positively associated with the visuals of hand, wedding, painting, eye, and beauty salon). The creative personnel could use this matching for choosing associations that execute the brand's desired positioning.
For the strategic function of brand management, insights from B-VEP are useful in assessing brand health, positioning relative to other brands, and collaboration opportunities. We measured the relationship between the associations and brand favorability and found that the corresponding associations differ by category (e.g., favorable associations for cars differ from those for beverages). We also measured brand uniqueness: for each brand, we tested how its set of associations differs from those of other brands in its category. Finally, we use similarity and distance in the association space to detect potentially valuable commonalities between brands, which could prove useful for potential collaborations.
Our method can be used to index repositories of images as per their fit with the association set of a brand. Such indexing can have various applications. For example, it can help brand managers and graphic designers search for images that reflect the current set of a brand's associations and create mood boards, or prototypical collages for brands, by visually displaying their associations.
First, we used guided LDA to calculate the distribution of associations of all of the images in our photo repository. Then, we chose for each brand the set of ten photos (not contained in any of the original brand collages) that together generate the highest similarity to the brand associations vector. To reduce computational complexity, we used a greedy algorithm that adds photos to the set one at a time to move the collective topic distribution maximally toward the desired distribution. We measured cosine similarity between the normalized 150-dimensional topic vectors of the photos and the brand. The average similarity between the brand association vector and the representative collage is.899, indicating that the collages are prototypical.[12]Figure 4 presents the prototypical collages for the three deodorant brands Axe, Degree, and Secret. Note that we could have chosen photos from other photo repositories or created collages containing more or fewer than ten photos.
Graph: Figure 4. Prototypical collages for Axe, Degree, and Secret, based on cosine similarity between the brand association distribution and the photos in the photo repository.
We explore the relationship of the brand associations extracted from the collages with the frequently used brand characteristics: Aaker's brand personality characteristics ([ 1]) and Young & Rubicam's BAV equity characteristics ([23]). Understanding the relationships between specific brand associations with brand dimensions of personality and equity can assist brand managers in cultivating and using the visual representation that will support the brand's desired personality and equity characteristics. For example, what brand associations should a manager develop to make the brand more down-to-earth?
Recall that each respondent, after completing the collage for a brand, was asked to score the brand on the items of the brand personality and brand equity characteristics. Altogether, the respondents rated the brand on 49 characteristics, a combined set of the Aaker brand personality traits and the BAV brand equity pillars. To measure relationships between these characteristics and our identified brand associations, we regressed these ratings on the corresponding collage's distribution of topics (associations).
Specifically, let be the set of collages, be the set of associations, and be the set of brand characteristics, rated by each respondent on a 1–5 scale. A total of 49 characteristics were in the survey (i.e., ). Let be the rating on characteristic s corresponding to collage i. We ran the following regression for each characteristic:
Graph
The resulting coefficients . represent the extent to which topic occurs more/less in collages in which the brand is rated higher on characteristic s. We obtained 3,937 observations and 150 regressors for each regression. Note that for this analysis, it is essential that the associations were elicited at the individual collage level. Thus, we can link associations to individual brand perceptions.
Figures 5 and 6 present the significant associations with the five most positive coefficients and the five with the most negative coefficients, for each of the items used to construct the "sophistication" and "ruggedness" factors of brand personality characteristics (Figure 5) and the "differentiation" equity pillar (Figure 6). Web Appendix F presents the full results. For example, the personality trait "glamorous" (part of the "upper-class" facet in the "sophistication" personality factor of [ 1]] scheme) is associated with wedding, eye, fashion, and glamour. It is not associated with heavy vehicles, construction, and patriotism (for the complete associations, see Web Appendix D), meaning that brands that are rated high on "glamorous" contain fewer of these visuals in their collages. The personality trait "rugged" (which is part of the "ruggedness" factor in Aaker's scheme) is positively associated with the associations of heavy vehicle, military, bicycle, industry, and desert and not associated with therapy, church, candy, arts and crafts, and sparkling (for the complete tags related to each association, see Web Appendix C). The equity characteristic "innovative," which is part of the "differentiation" BAV equity pillar, is correlated with high frequency of associations such as hand, religion, painting, cityscape, and light and negatively correlated with patriotism, chest, ruin, symbol, and cowboy, meaning that brands that score high on "innovative" will contain fewer visuals of these associations in their collages.
Graph: Figure 5. The associations with strongest positive and negative coefficients relating to the personality characteristics of "sophistication" and "ruggedness."Notes: The positive associations are arranged in decreasing order (from left to right), and the negative are arranged in increasing order (left to right, from the most negative to the least negative). N = 3,937.
Graph: Figure 6. The associations with the strongest positive and negative coefficients associated with the "differentiation" brand equity pillar.Notes: The positive associations are arranged in decreasing order, and the negative are arranged in increasing order (from the most negative to the least negative). N = 3,937.
The results presented in Figures 5 and 6 and Web Appendix F demonstrate that brand personality and equity traits systematically relate to particular associations. Mapping brands in this very rich, unstructured space of visual content reveals that the meaning of certain visual content is systematically related to established brand measures.
Next, we identify desirable and undesirable associations in each category. Recall that after submitting each brand collage, the respondent was asked a series of questions about the brand. One of the survey items was to rate the brand on being "high quality," with 1 being the lowest quality and 5 being the highest. We regressed this rating on the associations extracted from the collage. One collage is a data point, and we ran the regressions on collages separately for each category. Web Appendix G presents the results.
For example, for cars, the associations alcoholic drink, cityscapes, house, fashion, and suit have positive and significant coefficients—that is, they occur more frequently in collages for which the respondent rates the brand as higher quality. As presented in Web Appendix D, car brands whose collages include alcoholic drinks as one of their top associations are the luxury brands—Audi, Lamborghini, Porsche, and Mercedes Benz—and their collages contain images of mansions, Riviera vacations, and expensive alcohol. The associations music festival, healthy cooking, breakfast, rain, dance, and ruin have negative coefficients. Interestingly, while certain associations, such as ruin, have either a negative or nonsignificant coefficient for all categories, some associations have opposite signs in some categories. For example, while breakfast and healthy cooking are negative for cars, both are positive in food and dining. Negative associations for food and dining include pollution, traffic, industry, vehicle, finance, computer, and ruin. The beach association is positive for food and dining but negative for beverages. The house association is positive for cars but negative for beverages.
Because the "high-quality" characteristic is vertical (i.e., one on which all brands would want to be rated highly), we conducted this analysis at the category level. Indeed, one would expect positive and negative associations to be specific to a product category. Next, we examine more horizontal brand characteristics (i.e., those that some brands want to have, and others do not). For example, while some brands want to be perceived as "sincere" and "down-to-earth," others may aim to be perceived as "glamorous" or "sophisticated."
To determine how a brand stands out from others in its category, we tested whether an association occurs with a significantly higher/lower probability in collages for the focal brand than for other brands in the same category. We chose to compare the brand with its category, rather than simply with all other brands in the set, to remove category-level averages. For example, beauty brands have on average more flowers, water, hairstyling, and glamour associations than car brands. Specifically, we performed a Mann–Whitney test to compare to , for each association . Recall that is the set of collages for brand b, and is the weight of association k in collage i. We report for each brand the associations for which these two samples are statistically significantly different. We present the results for the brands in the beauty category in Table 5. Web Appendix H presents results for the full set of brands.
Graph
Table 5. Beauty Brands: Associations with Significantly Higher/Lower Probability Relative to Other Brands in Its Category, Ordered from Most to Least Unique.
| Brand Name | Most Associated With (Top Five), Relative to the Category | Least Associated With (Top Five), Relative to the Category |
|---|
| Always | Therapy | Dance | Running | Glamour | Flowers (tropical) | Hairstyling | Computer | Youth | Beach | Pool |
| Aveeno | Streams | Frosty | Foggy landscape | Diving | | Abstract art | Misty | | | |
| Avon | Seats | Cutlery | Healthy cooking | Dining | | Water | American flag | Car | Aeronautics | Erotic |
| Axe | Urban youth | Bodybuilding | Band | Ball sports | Suit | Water | Military | Family | American flag | Birds of prey |
| Bath & Body Works | Therapy | Fruits | Aeronautics | Juice | Water | Cat | Running | Music festival | Sports | Bodybuilding |
| Caress | Streams | Abstract art | Water | | | Steel | Carnival | Desert | Flowers (romantic) | Juice |
| Chanel beauty | Light | Mountaineering | Beauty salon | Science | | Countryside | Baby | Autumn | Bedroom | Dogs |
| Charmin | Bedroom | Water | Cat | Delicate fabric | Snowscape | Horror | Mountain | Baby | Child | Therapy |
| Clean & Clear | Happy nature | American flag | | | | Theater | Misty | Cottage | Dining | Autumn |
| Clinique | Glamour | Hairstyling | Eye | Painting | Arts and crafts | Horror | House | | | |
| Colgate | Water | Family | Herbs | Child | Dental | Flowers (romantic) | Fruits | Water sports | Running | Music festival |
| CoverGirl | Holiday party | Music festival | Alcoholic drinks | Wedding | Religion | Streams | Fashion | Glamour | Hairstyling | Military |
| Crest | Bathroom | Steel | Healthy cooking | Dental | Abstract art | Birds of prey | Candy | Construction | Sparkling | Industry |
| Degree | Running | Sports | Fitness | Ball sports | Water | Hairstyling | Glamour | Birds of prey | Wedding | Abstract art |
| Dial Soap | Water | Ocean | Streams | Popstar | Clothing | Glamour | Fashion | Wedding | Jewelry | Suit |
| Dove | Flowers romantic | Streams | Erotic | Water | Bathroom | Fruits | Produce | Sports | Old town | Business school |
| Garnier Fructis | Streams | Diving | Religion | Aquarium | | Cat | Family | Birds of prey | Furniture | Youth |
| Gillette | Suit | Sailing | Water sports | Modern building | Bodybuilding | Flowers (romantic) | Fruits | Holiday party | Rally | Ball sports |
| Head & Shoulders | Juice | Herbs | Aeronautics | Snowscape | City twilight | Rainstorm | Fruits | Wheat | Produce | Popstar |
| Herbal Essences | Water birds | Autumn | Birds of prey | Fowl | Wildlife | Bathroom | Youth | Bedroom | Family | Child |
| Irish Spring | Streams | Water | Mountain | Countryside | Rain | Hairstyling | Glamour | Lingerie | Cat | Bedroom |
| Jergens | | | | | | Streams | Youth | Pool | Misty | Happy nature |
| Kleenex | Rainstorm | Photography | Candy | | | Glamour | Flowers (romantic) | Fashion | Lingerie | Flowers (botanical) |
| Kotex | Fashion | Business school | Baby | Warm fabrics | Wheat | Streams | Alcoholic drinks | Pool | Industry | Erotic |
| L'Oréal | Hairstyling | Glamour | Horror | | | Water | Countryside | Flowers tropical | Rain | Flowers romantic |
| Mary Kay | Glamour | Curved lines | Fashion | Symbol | Sparkling | Fruits | Countryside | Mountain | Sailing | Misty |
| Maybelline | Glamour | Eye | Hairstyling | Curved lines | Candy | Frosty | Countryside | Mountaineering | | |
| Neutrogena | Baby | Ocean | Foggy landscape | Hairstyling | Urban youth | Cat | Steel | Herbs | Military | Religion |
| Nivea | Aeronautics | Street art | Candle | Clock | | Rainstorm | Geometric | Sailing | Abstract art | Warm fabrics |
| Olay | Flowers (romantic) | Flowers (botanical) | Flowers (tropical) | Insects | Adventure quest | Running | Erotic | Bodybuilding | Fitness | Construction |
| Old Spice | Bodybuilding | Heavy vehicle | Running | Sports | Military | Flowers (romantic) | Water | Fashion | Birds of prey | Delicate fabric |
| Pantene | Beach | Autumn | Ocean | Lingerie | Kitchen | Therapy | Horror | Baby | Geometric | Train |
| ProActiv | Eye | Metalwork | Bathroom | Computer | Cartoon | Car | Theater | Pollution | Symbol | |
| Revlon | Fashion | Glamour | Flowers (botanical) | Modern building | Street art | Streams | Rain | Mountain | Mountaineering | Aeronautics |
| Scott Tissue | Cat | Frosty | Birds of prey | Child | Animals | Water | Clothing | Glamour | Flowers (botanical) | Jewelry |
| Secret | Flowers (romantic) | Cottage | Baking | Golf | Healthy cooking | Ocean | Finance | Rally | Photography | Dining |
| Sephora | Hairstyling | Fruits | Flowers (romantic) | Eye | Lingerie | Power energy | Juice | Herbs | Industry | American flag |
| Suave | Band | | | | | Fashion | Bedroom | Furniture | Delicate fabric | Geometric |
| Tampax | Beauty salon | Modern building | | | | Streams | Holiday party | Light | American flag | Alcoholic drinks |
| TRESemmé | Hairstyling | Fashion | Glamour | Seats | Diving | Fruits | Rain | Streams | Foggy landscape | Autumn |
The left-hand columns of Table 5 ("Most Associated With") contain the five associations that occur significantly more frequently in the collages for the brand, whereas the right-hand set of columns ("Least Associated With") contain the five associations that occur significantly less frequently for this brand relative to other brands in the category. We see that some brands are prototypical of their categories, while others stand out. For example, Dasani and Diet Pepsi are not much differentiated from the beverages category (at p <.05), while Jamba Juice has more distinctive associations.
Relative to the average beauty brand, the deodorant brand Axe is more associated with urban youth, bodybuilding, band, ball sports, and suit, meaning that these associations appear in its collages significantly more frequently than they do in the average beauty brand. The romantic flowers association has a strong presence in the category; although it exists in Axe's associations (Table 2), it does not differentiate Axe from the category. In the cars category, while most cars are associated with traffic, cityscape, and steel (see Table 3), Ferrari has, relative to other car brands, strong associations with aeronautics, delicate fabrics, and lingerie and less strong associations with industry, school, and church than the average car brand (see Web Appendix H). Jeep, positioned as an outdoorsy brand, has significantly lower weights, in the association distribution vector of its collages, of cityscape and modern building than the average car brand.
Note that the usage and interpretation of these results should be done with caution. The large number of tests is subject to multiple comparisons concerns. A brand manager who runs 150 tests at the p <.05 level would expect to find 7–8 spurious effects when running such an analysis.
Our association elicitation method enables measuring the similarity of associations between brands. We calculated the cosine similarity between the normalized (sum of squares is equal to 1) association distribution vectors of all of the brand pairs in our sample. Cosine similarity is a way to compare two vectors by calculating the angle that they create. The number ranges from 0 to 1, where 1 indicates identical vectors. Web Appendix I describes the similarity matrix, a symmetric 303 × 303 matrix, whose elements take values between 0 and 1, with higher values corresponding to more similar brands and diagonal elements equal to 1. We see, for example, that the associations of Cheesecake Factory are highly similar to the baking appliances brand KitchenAid (cosine similarity of.84): they share common associations (baking, dining, candle, family). The family dining chain Golden Corral is very similar in its associations to the supermarket convenience food brand Hormel (cosine similarity of.91). Barnes and Noble has similar associations to the pain drug brand Aleve (cosine similarity of.7, sharing the associations of school and bedroom). Febreze has a.7 cosine similarity to Ashley Furniture. These similarities can be an indication for potential brand alliance, cross-category perceptual maps, and positioning inquiries.
In this article, we propose and implement a novel brand-association-elicitation tool (which we term B-VEP). The elicitation task enables participants to portray their relationships with brands through a collage of photographs. Visual images have the advantage of better reflecting the emotions, cultural experiences, and attitudes that constitute consumer associations, as opposed to verbal methods that focus more on the discourse of these experiences ([36]). Use of images has been demonstrated to successfully disrupt well-rehearsed narratives, revealing hidden, unarticulated ideas. The analysis uses unsupervised machine learning methods to avoid "strangling" the data: rather than looking for specific predefined associations, we let the data speak and identify associations using topic modeling. The resulting set of associations is rich and spans a variety of objects, occupations, natural elements, constructs, and institutions, to name just a few.
Using this tool, we gathered a large set of consumer brand perceptions on 303 brands. We applied it to explore several important challenges for brand management: creating mood boards for each brand, consisting of a collection of photographs that capture the distribution of consumers' associations with the brand; testing which associations are related to commonly used brand metrics such as brand personality and brand equity; identifying favorable and unfavorable associations for each category; finding unique associations, in which the brand differs from others in its product category; and, finally, measuring association-based similarities between brands from different categories, which may identify potential for brand alliances or strategic partnerships.
We see these applications as just scratching the surface of the potential of using visual elicitation. We hope that future research will build on this work in other directions. One future direction might be identifying brand extension strategies. Starbucks's top associations include baking and dining (see Web Appendix D). While Starbucks does offer baked goods and food, this association might imply a need for more dining choices. Interestingly, Dunkin Donuts, which by definition offers baked goods, has much weaker association with baking and dining. In the beauty category (Table 2), Clinique has a strong association with hairstyling; however, its product line contains few hair products. These insights can be a starting point for exploring brand extensions.
Another potential avenue for future work is to identify systematic relationships between perceptual dimensions and elements of visual design, such as shapes, colors, texture, and so on. While modern visual design provides many guidelines on how these elements can be used in a composition to create a certain perception, few of these are empirically tested on brand-related imagery. In addition, elements such as a photo's location on the canvas relative to other photos, its size, and its rotation angle may carry additional meaning of which we are currently unaware.
An interesting theoretical question is the evolution of brand associations and their relationships with brand characteristics ([40]). On the one hand, one could argue that consumers think about brands in terms of characteristics such as personality and equity and then create in their minds images to represent these characteristics (e.g., they perceive the brand as innovative, and the concept of innovativeness evokes metaphors such as transistors, and therefore, they associate the brand with visuals containing transistors). On the other hand, one could think of the brand as evoking sets of metaphors, and the characteristics of these metaphors reflect, in turn, how consumers perceive the brand (e.g., the brand evokes the association of a transistor, transistors are perceived as innovative, which forms, inter alia, the innovative perception of the brand). B-VEP can help address this question through tasks such as collage building of synthetic brands with predefined controlled characteristics, or creating collages describing characteristics (e.g., innovative) and testing their similarity to associations of brands.
Our tool can aid in exploring heterogeneity among consumers' brand perceptions. By collecting a large number of collages per brand, we can learn how individual differences in personality, values, lifestyle, and other variables of interest influence brand perception. Insights from such studies can be useful for performing segmentation, optimizing marketing communications, and creating a better fit between brands and their consumers. This could be done by methods such as hierarchical topic models, unsupervised machine-learning-based clustering, and latent class models.
In summary, modern software and image-processing tools open many new opportunities for marketing researchers. B-VEP enables researchers and firms to gather and harvest visual brand-related data directly from consumers, which complements existing brand metrics as well as the rapidly growing field of visual social media monitoring.
Supplemental Material, sj-xlsx-1-jmx-10.1177_0022242921996661 - Visual Elicitation of Brand Perception
Supplemental Material, sj-xlsx-1-jmx-10.1177_0022242921996661 for Visual Elicitation of Brand Perception by Daria Dzyabura and Renana Peres in Journal of Marketing
Participants select the images for their collages from a large photo repository (the right-hand side of Figure 2). To help them browse through the repository, we implemented the option to search for keywords. Because the search uses words, one may wonder whether this undermined B-VEP's main focus as a visual elicitation tool. To verify that this is not the case, we have conducted the following tests and measurements:
- Usage of the search keywords is infrequent: Figure A1 displays the distribution of search words per collage. Of the 3,937 approved collages, 690 (17.52%) did not use any search word. The median number of search terms used in a collage is 5, indicating that half the collages used 5 search terms or fewer. The average number of search terms per collage is 6.41.
- Search words are repetitive: The search words used are limited and are often repetitive. Out of the 25,262 search terms (consisting of 28,505 separate words) used by respondents, only 6,475 were unique (21.3%). The top 30 words (.5% of the number of unique words) are responsible for 4,465 searches (17% out of the total number of searches). Figure A2 shows a histogram of the common search terms. These search terms are not brand specific and might have a limited power in providing a unique brand association.
- Each search retrieves a large number of photos: The photo repository has ∼100,000 photos, each given dozens of labels by Flickr users. Therefore, each search word retrieves multiple photos, from which the user needs to keep scrolling to choose the most appropriate one. For example, the search word "family" retrieves 318 photos, "nature" retrieves 3,621 photos, "child" retrieves 629 photos, and "happy" retrieves 210 photos. In Figure 2, the participant searched for the word "laptop" and retrieved 46 Flickr photos labeled "laptop." The participant could have chosen a laptop with people sitting next to it, children playing on a laptop in a restaurant, a laptop in an office or a school, and so on. The chosen picture contains many additional visual items that, we believe, reflect additional feelings, attitudes, and associations the user had for the brand that might not be even related to the original search word "laptop." Therefore, the search can be viewed as an aid in the browsing, but not one that limits or constrains it.
- Users rarely use the "banned" words: Users were directed to "not choose pictures that show the brand logo (or a logo of any other brand), type of product, or product category." If they did so, they received an error message. Out of the 25,262 search terms used, only 1,111 (<5%) attempts were made to use the "banned" words. Despite this restriction, the collages still capture brand functionality. Table 3 demonstrates that the category information is present and significant in the collages. Thus, our restriction helps respondents create rich and meaningful collages without disrupting the flow of the collage making.
- Search word usage does not impact collage: To determine whether collages that used more search words generate different associations than collages that used fewer search words, we carried out the following procedure:
- Split: Split the collages of each brand into two equal groups: collages that used a below-median number of search terms (1,968 collages, average of 2.9 search terms per collage) and collages that used an above-median number of search terms (1,968 collages, average of 11.6 search words per collage).
- Elicit associations: We applied our association elicitation method (the guided LDA topic extraction) on each of these two groups and extracted associations.
- Test similarity: For each brand, we calculated the cosine similarity between the normalized (sum of squares is equal to 1) association distribution vectors of the above-median and below-median groups. Recall that cosine similarity is a way to compare two vectors, by calculating the size of angle they create. The number ranges from 0 to 1, where 1 indicates identical vectors.
- Estimate similarity relative to random partition: We compared the cosine similarity of the association vectors of the above-/below-median split with the cosine similarity values obtained by 100 other random equal partitions of the brand collages. That is, if a brand has n collages, they form partitions of size n/2. We sampled 100 of these partitions for each brand, calculated the cosine similarity of their association vectors, and checked what percentile the above/below similarity falls into. If, indeed, it is equivalent to any other partition, the percentile should fall in the range 0–1 in a uniform distribution. Figure A3 presents the percentiles for the 303 brands. Indeed, the distribution is not significantly different from uniform (χ2p-value =.12).
Graph: Figure A1. The distribution of search words used in a collage across collages.
Graph: Figure A2. The top 30 search words over all the collages.
Graph: Figure A3. The ordered percentiles, for all brands, of the cosine similarity LDA topic distribution vector of the above-below median partition within 100 other equal size random partitions.
Footnotes 1 David Schweidel
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This article was supported by the Marketing Science Institute, the Israeli Science Foundation, and the KMart foundation.
4 Online Supplement: https://doi.org/10.1177/0022242921996661
5 For the collage task as well as the other parts of the questionnaire, see http://bvep.ResearchSoftwareHosting.org.
6 In principle, this data-cleaning process could be automated.
7 We tested the initial number of clusters with various numbers of collages and concluded that for the collage data, 10% is a starting point that clusters similar words but does not strangle the data. Note, however, that for data sets of a different nature (e.g., articles, movie scripts, user reviews) the starting point might be different.
8 We chose this value for our data because it is the minimal value that has impact on the LDA output yet still allows the LDA the freedom to move the words around. The results showed low sensitivity to this parameter.
9 For the user interface of the validation experiment, see http://collages.researchsoftwarehosting.org.
For the user interface of the validation experiment, see http://positiveness.researchsoftwarehosting.org.
For the user interface of the free-text elicitation task, see http://bvep-text.researchsoftwarehosting.org.
For the representative collages, see https://www.dropbox.com/sh/t1gc61mkx2k5lyz/AACL6rXp0le-SisLK8jXuhX2a?dl=0.
References Aaker Jennifer L. (1997), "Dimensions of Brand Personality," Journal of Marketing Research, 34 (3), 347–56.
Blei David M., Ng Andrew Y., Jordan Michael I. (2003), "Latent Dirichlet Allocation," Journal of Machine Learning Research, 3 (1), 993–1022.
Catchings-Castello Gwendolyn. (2000), "The ZMET Alternative," Marketing Research, 12 (2),6–12.
Cho Hyejeung, Schwarz Norbert, Song Hyunjin. (2008), "Images and Preferences: A Feelings-as-Information Analysis," in Visual Marketing: From Attention to Action, Wedel Michel, Pieters Rik, eds. New York: Lawrence Erlbaum Associates, 259–76.
Culotta Aron, Cutler Jennifer. (2016), "Mining Brand Perceptions from Twitter Social Networks," Marketing Science, 35 (3), 343–62.
Davis Donna, Butler-Kisber Lynn. (1999), "Arts-Based Representation in Qualitative Research: Collage as a Contextualizing Analytic Strategy," paper presented at the Annual Meeting of the American Educational Research Association, Montreal, QC (April 19–23).
Gelper Sarah, Peres Renana, Eliashberg Jehoshua. (2018), "Talk Bursts: The Role of Spikes in Prerelease Word-of-Mouth Dynamics," Journal of Marketing Research, 55 (6), 801–17.
Greenleaf Eric, Raghubir Priya. (2008), "Geometry in the Marketplace," in Visual Marketing: From Attention to Action, Wedel Michel, Pieters Rik, eds. New York: Lawrence Erlbaum Associates, 113–42.
Han Young Jee, Nunes Joseph C., Drèze Xavier. (2010), "Signaling Status with Luxury Goods: The Role of Brand Prominence," Journal of Marketing, 74 (4), 15–30.
Howard Andrew G. (2013), "Some Improvements on Deep Convolutional Neural Network Based Image Classification," https://arxiv.org/ftp/arxiv/papers/1312/1312.5402.pdf.
Jagarlamudi Jagadeesh, Daumé HalIII, Udupa Raghavendra. (2012), "Incorporating lexical Priors into Topic Models," in Proceedings of the 13th Conference of the European Chapter of the Association for Computational Linguistics. Stroudsburg, PA: Association for Computational Linguistics, 204–13.
Jalali Nima Y., Papatla Purushottam. (2016), "The Palette That Stands Out: Color Compositions of Online Curated Visual UGC That Attracts Higher Consumer Interaction," Quantitative Marketing and Economics, 14 (4), 353–84.
John Deborah Roedder, Loken Barbara, Kim Kyeongheui, Monga Alokparna Basu. (2006), "Brand Concept Maps: A Methodology for Identifying Brand Association Networks," Journal of Marketing Research, 43 (4), 549–63.
Johnson Jeff. (2013), Designing with the Mind in Mind: Simple Guide to Understanding User Interface Design Guidelines. New York: Elsevier.
Keller Kevin Lane. (1993), "Conceptualizing, Measuring, and Managing Customer-Based Brand Equity," Journal of Marketing, 57 (1), 1–22.
Klostermann Jan, Plumeyer Anja, Böger Daniel, Decker Reinhold. (2018), "Extracting Brand Information from Social Networks: Integrating Image, Text, and Social Tagging Data," International Journal of Research in Marketing, 35 (4), 538–56.
Koll Oliver, Von Wallpach Sylvia, Kreuzer Maria. (2010), "Multi-Method Research on Consumer–Brand Associations: Comparing Free Associations, Storytelling, and Collages," Psychology and Marketing, 27 (6), 584–602.
Kress Gunther R., Van Leeuwen Theo. (1996), Reading Images: The Grammar of Visual Design. New York: Routledge.
Labrecque Lauren I. (2014), "Fostering Consumer–Brand Relationships in Social Media Environments: The Role of Parasocial Interaction," Journal of Interactive Marketing, 28 (2), 134–48.
Lee Thomas Y., Bradlow Eric T. (2011), "Automated Marketing Research Using Online Customer Reviews," Journal of Marketing Research, 48 (5), 881–94.
Liu Liu, Dzyabura Daria, Mizik Natalie V. (2020), "Visual Listening in: Extracting Brand Image Portrayed on Social Media," Marketing Science, 39 (4), 669–86.
Lovett Mitchell, Peres Renana, Shachar Ron. (2013), "On Brands and Word of Mouth," Journal of Marketing Research, 50 (4), 427–44.
Lovett Mitchell, Peres Renana, Shachar Ron. (2014), "A Dataset of Brands and Their Characteristics," Marketing Science, 33 (4), 609–17.
McQuarrie Edward F. (2008), "Differentiating the Pictorial Element in Advertising: A Rhetorical Perspective," in Visual Marketing: From Attention to Action, Wedel Michel, Pieters Rik, eds. New York: Lawrence Erlbaum Associates, 91–112.
Meyers-Levy Joan, Zhu Rui. (2008), "Perhaps the Store Made You Purchase It: Toward an Understanding of Structural Aspects of Indoor Shopping Environments," in Visual Marketing: From Attention to Action, Wedel Michel, Pieters Rik, eds. New York: Lawrence Erlbaum Associates, 193–224.
Mizik Natalie, Jacobson Robert. (2008), "The Financial Value Impact of Perceptual Brand Attributes," Journal of Marketing Research, 45 (1), 15–32.
Nam Hyoryung, Joshi Yogesh V., Kannan P.K. (2017), "Harvesting Brand Information from Social Tags," Journal of Marketing, 81 (4), 88–108.
Nam Hyoryung, Kannan Pallassana Krishnan. (2014), "The Informational Value of Social Tagging Networks," Journal of Marketing, 78 (4), 21–40.
Netzer Oded, Feldman Ronen, Goldenberg Jacob, Fresko Moshe. (2012), "Mine Your Own Business: Market-Structure Surveillance Through Text Mining," Marketing Science, 31 (3), 521–43.
Palmer Stephen E. (1999), Vision Science: Photons to Phenomenology. Cambridge, MA: MIT Press.
Pavlov Eugene, Mizik Natalie. (2019), "Increasing Consumer Engagement with Firm-Generated Social Media Content: The Role of Images and Words," working paper, Foster School of Business, University of Washington.
Pennington Jeffrey, Socher Richard, Manning Christopher D. (2014), "GloVe: Global Vectors for Word Representation," in Proceedings of the Empirical Methods in Natural Language Processing (EMNLP 2014). Stroudsburg, PA: Association for Computational Linguistics, 1532–43.
Pieters Rik, Rosbergen Edward, Wedel Michel. (1999), "Visual Attention to Repeated Print Advertising: A Test of Scanpath Theory," Journal of Marketing Research, 36 (4), 424–38.
Rangel José Carlos, Cazorla Miguel, García-Varea Ismael, Martínez-Gómez Jesus, Fromont Élisa, Sebban Marc. (2016), "Scene Classification Based on Semantic Labeling," Advanced Robotics, 30 (11/12), 758–69.
Rayner Keith, Miller Brett, Rotello Caren M. (2008), "Eye Movements When Looking at Print Advertisements: The Goal of the Viewer Matters," Applied Cognitive Psychology, 22 (5), 697–707.
Reavey Paula. (2011), "The Return to Experience: Psychology and the Visual," in Visual Methods in Psychology: Using and Interpreting Images in Qualitative Research, Chap. 1. London: Routledge.
Seidman Gwendolyn. (2013), "Self-Presentation and Belonging on Facebook: How Personality Influences Social Media Use and Motivations," Personality and Individual Differences, 54 (3), 402–07.
Steenkamp Jan-Benedict, Van Trijp Hans. (1997), "Attribute Elicitation in Marketing Research: a Comparison of Three Procedures," Marketing Letters, 8 (2), 153–65.
Tirunillai Seshadri, Tellis Gerard J. (2014), "Mining Marketing Meaning from Online Chatter: Strategic Brand Analysis of Big Data Using Latent Dirichlet Allocation," Journal of Marketing Research, 51 (4), 463–79.
Torres Anna, Bijmolt Tammo H. (2009), "Assessing Brand Image Through Communalities and Asymmetries in Brand-to-Attribute and Attribute-to-Brand Associations," European Journal of Operational Research, 195 (2), 628–40.
Wedel Michel, Pieters Rik. (2000), "Eye Fixations on Advertisements and Memory for Brands: A Model and Findings," Marketing Science, 19 (4), 297–312.
Wedel Michel, Pieters Rik. (2008), "Introduction to Visual Marketing," in Visual Marketing: From Attention to Action, Wedel Michel, Pieters Rik, eds. New York: Lawrence Erlbaum Associates, 1–8.
Zaltman Gerald, Coulter Robin Higie. (1995), "Seeing the Voice of the Customer: Metaphor-Based Advertising Research," Journal of Advertising Research, 35 (4), 35–51.
Zaltman Gerald, Zaltman Lindsay H. (2008), Marketing Metaphoria: What Deep Metaphors Reveal About the Minds. Boston: Harvard Business Press.
Zhang Shunyuan, Lee Dokyun, Singh Param Vir, Srinivasan Kannan. (2017), "How Much Is an Image Worth? Airbnb Property Demand Estimation Leveraging Large Scale Image Analytics," working paper, Tepper School of Business, Carnegie Mellon University.
~~~~~~~~
By Daria Dzyabura and Renana Peres
Reported by Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 135- Wage Inequality: Its Impact on Customer Satisfaction and Firm Performance. By: Bamberger, Boas; Homburg, Christian; Wielgos, Dominik M. Journal of Marketing. Nov2021, Vol. 85 Issue 6, p24-43. 20p. 2 Diagrams, 6 Charts, 1 Graph. DOI: 10.1177/00222429211026655.
- Database:
- Business Source Complete
Record: 136- When Algorithms Fail: Consumers' Responses to Brand Harm Crises Caused by Algorithm Errors. By: Srinivasan, Raji; Sarial-Abi, Gülen. Journal of Marketing. Sep2021, Vol. 85 Issue 5, p74-91. 18p. 1 Chart. DOI: 10.1177/0022242921997082.
- Database:
- Business Source Complete
When Algorithms Fail: Consumers' Responses to Brand Harm Crises Caused by Algorithm Errors
Algorithms, increasingly used by brands, sometimes fail to perform as expected or, even worse, cause harm, leading to brand harm crises. Algorithm failures are unfortunately increasing in frequency, yet little is known about consumers' responses to brands following such crises. Extending developments in the theory of mind perception, the authors hypothesize that, following a brand harm crisis, consumers respond less negatively if the error was caused by an algorithm (vs. a human). The authors further hypothesize that consumers' lower mind perception of agency of the algorithm (vs. a human), which lowers their perceptions of the algorithm's responsibility for the harm, mediates this relationship. Four moderators of this relationship are hypothesized: two algorithm characteristics (whether the algorithm is anthropomorphized and whether it involves machine learning) and two characteristics of the task for which the algorithm is deployed (whether the task is subjective [vs. objective] and whether it is interactive [vs. noninteractive]). The authors find support for the hypotheses in eight experimental studies. The effects of two managerial interventions to manage brand harm crises caused by algorithm errors are examined. This research advances the literature on brand harm crises, algorithm usage, and algorithmic marketing and generates managerial guidelines to address such crises.
Keywords: algorithm errors; algorithmic marketing; brand harm crises; theory of mind perception
Given the explosive growth in the volume of data, dramatic developments in software programs, and the decreasing cost of cloud computing, the use of algorithms—software programs that organize data, predictions, and decisions—has grown exponentially. Although this has occurred across many contexts, algorithm usage in the marketing context—algorithmic marketing—has increased dramatically. Algorithmic marketing has many advantages, including lower costs, high efficiency, and high effectiveness ([18]). Despite these advantages, though, there is growing evidence of algorithm failures across multiple contexts ([24]). In marketing, algorithm errors harm consumers and/or violate consumers' expectations of the brand's values, creating brand harm crises. In a survey of chief marketing officers (CMOs), fielded by the CMO Council and Dow Jones Inc., most CMOs (78%) expressed concern about the threats to their brands' reputations from algorithm errors ([55]).
Although algorithms operate in the digital domain, algorithm errors have many real-world consequences, including causing substantive harm to brands. We discuss two examples to provide additional context. First, there is evidence ([ 9]; [52]) of algorithmic defamation in online searches. Algorithm-based Google search auto-completion routines make incorrect defamatory associations about groups of people ([ 5]). For example, searching for certain ethnic names on Google provides results including advertisements for bail bonds or criminal record checking. Second, Apple Credit Card, launched in partnership by Apple Inc. and Goldman Sachs Inc. in August 2019, faced reputational harm when users noticed that it offered lower lines of credit to women than to men of equal or even lower financial standing ([53]). In response, the New York Department of Financial Services announced an investigation of Apple Inc. to assess a breach of federal financial rules regarding equal financial access. Cognizant of the potential harm that could be caused by algorithm errors, for the first time, Google's parent company, Alphabet Inc. (February 2019), and Microsoft Inc. (August 2018) acknowledged in their annual reports that "flawed" algorithms could result in "brand or reputational harm" and have an "adverse effect" on financial performance ([54]). In summary, algorithm errors are a key and growing source of brand harm crises.
Brand harm crises are adverse negative events inconsistent with a brand's values. In a brand harm crisis, the brand's ability to deliver promised benefits to consumers is compromised or, even worse, the brand causes physical harm to consumers ([11]; [42]), causing consumers to respond negatively to the brand ([ 3]; [34]; [51]). Consumers' attributions about what caused the harm influence their subsequent responses to the brand ([15]; [16]). Consumers feel angry and seek revenge if they believe that the firm was responsible for the harm and could have prevented it ([17]). See [ 8] for a comprehensive review of the brand harm crises literature. Because the growth in algorithmic marketing is fairly recent, research has not yet addressed harm crises caused by algorithm errors.
There is a large body of research in multiple literatures, including in marketing, on people's responses to nonhuman agents (e.g., algorithms, computers, robots). People treat computers as social actors although they know that computers do not possess feelings, intentions, motivations, or "selves" ([38]; [40]). Other work ([ 7]) suggests that humanoid (vs. nonhumanoid) service robots are more strongly associated with warmth (but not competence).
Past work on algorithm usage has examined people's responses to using algorithms ([ 6]; [10]). Individuals prefer doing a task themselves or having it done by their peers with whom they have more in common ([41]), rather than using an imperfect algorithm (i.e., people display algorithm aversion [[10]]). This preference for humans over algorithms persists even when it worsens outcomes. In contrast, in the advice-giving context (absent of algorithm errors), [35] report algorithm appreciation (i.e., people incorporate advice from algorithms more than from humans). Related recent work on automated vehicles operated by algorithms ([ 4]; [19]) suggests that individuals consider harm to pedestrians by an automated vehicle (vs. themselves as the driver in a regular car) to be more permissible. See [ 6] for a good overview of the research on algorithm usage (specifically Table 1).
In summary, past research on algorithms has overlooked how consumers respond to a brand following a brand harm crisis caused by an algorithm error (vs. human error), which is the focus of this research. Distinct from past research on algorithm usage that considers the individual's decision to use the algorithm, in this research context, the decision to use the algorithm is taken by the brand manager rather than by the consumer who experiences the harm caused by the algorithm error. Furthermore, the dependent variable here is the consumer's response to the brand and not to the algorithm that commits the error, which has been the focus of past research on algorithm usage. Furthermore, we examine the moderation effects of two algorithm characteristics and two characteristics of the task in which the error occurs on this relationship. As consumers' responses to a brand harm crisis are always negative ([34]), we examine consumers' negative responses to a brand following a brand harm crisis caused by an algorithm error.
We apply the theory of mind perception ([13]; [20]; [22]), which proposes that individuals ascribe minds to other entities (e.g., individuals, animals, robots) and consider the contents of these entities' minds. Specifically, we examine consumers' mind perception of the algorithm's agency (i.e., the error-committing entity's perceived capacity to intend and to act).
Features of an entity can change people's mind perception of the entity's agency ([56]). Accordingly, we hypothesize that, following a brand harm crisis caused by an algorithm error (vs. human error), consumers will, ceteris paribus, have lower mind perception of agency of the algorithm (the entity) and assign it lower responsibility for the harm caused, weakening their negative responses to the brand. Furthermore, individuals' responses to an algorithm vary depending on task characteristics ([ 6]). Accordingly, we consider four moderators of consumers' responses to a brand following a harm crisis caused by an algorithm error: two algorithm characteristics (whether the algorithm is anthropomorphized and whether it involves machine learning) and two characteristics of the task in which the algorithm error occurs (whether the task is subjective [vs. objective] and whether it is interactive [vs. noninteractive]). We test and find support for the hypotheses in eight experimental studies, including an incentive-compatible study with a consequential outcome (donation to a charity) and two studies with behavioral measures.
This research's insights extend the literature on harm crises by studying an inanimate source of errors, algorithms, which have hitherto been overlooked as a source of brand harm crises in the marketing literature. Second, in a novel extension to the algorithm usage literature, which has thus far focused on consumers' responses to an algorithm, we find that consumers' responses to the brand are more forgiving of algorithm errors when they do not have the authority to choose whether to use the algorithm. Third, we identify consumers' mind perception of algorithms' agency as a potential key building block, relevant in the development of a theory of algorithmic marketing. Fourth, by identifying the moderating role of algorithm and task characteristics, this research's insights make a novel contribution to the harm crises literature, which has not examined characteristics of the sources of the error and the task as factors affecting outcomes in harm crises. Using insights from the findings of the studies involving the four moderators and a managerial interventions study, we provide guidance to managers on the deployment of algorithms given their effects on consumers' responses when they commit errors, as well as how to manage the aftermath of such brand harm crises.
Early work on people's responses to nonhuman agents (e.g., computers) suggests that consumers mindlessly apply social norms ([38]) in their interactions with such agents, including displaying a self-serving bias in attributions of responsibility to positive versus negative service encounters ([39]). Building on these ideas, we apply the theory of mind perception in the psychology literature ([20]; [22]) about people's perceptions of the minds of other entities to algorithms to develop the hypotheses. We first provide a brief overview of the theory of mind perception and then develop the hypotheses.
Mind perception, also known as humanizing or mentalizing, involves individuals making inferences about their own and others' (entities) mental states by positing unobservable properties such as intentions, desires, goals, beliefs, and secondary emotions to serve as mediators between their sensory inputs and subsequent actions ([20]; [22]). According to the theory of mind perception, a perceiver needs to implicitly determine the extent to which an entity has a mind and then determine that entity's state of mind. In addition to perceiving the minds of other humans, people are capable of perceiving minds in nonhuman entities such as animals, gadgets, or software.
People perceive other entities' minds in terms of two psychological capacities: agency and experience ([20]). Mind perception of the entity's agency is its perceived capacity to intend and to act (e.g., self-control, judgment, communication, thought, memory), and mind perception of experience is the entity's perceived capacity for sensation and feeling (e.g., hunger, fear, pain, pleasure, consciousness). When discussing mind perception of agency, [20] posit that agency qualifies entities as moral agents, capable of reasoned actions and having the capacity to do right or wrong ([21]; [23]), whereas experience qualifies entities as moral patients, capable of benefiting from good or suffering from evil acted upon them.[ 5]
In this research, we consider consumers' mind perception of agency of the algorithm that has committed the error and do not consider consumers' mind perception of the algorithm's experience as a moral patient being acted upon by others, which is not relevant when the algorithm commits errors. We note that individuals' mind perception of agency of an entity are positively related to judgments of the entity's responsibility for harm caused ([57]), which is consistent with common law practice that holds individuals with diminished mental capacity as being less responsible for their transgressions.
We propose that, following a brand harm crisis caused by an algorithm error (vs. human error), consumers will have lower mind perception of agency of the algorithm (than they would for humans) for the error and assign lower responsibility to the algorithm for the harm caused, resulting in a less negative response to the brand.
Features of the entity can change people's mind perception of its agency ([56]). Furthermore, individuals' responses to algorithms vary depending on the characteristics of the task for which the algorithm is deployed ([35]). Extending these two ideas, we propose four factors that will moderate consumers' responses to a brand following a brand harm crisis caused by an algorithm error: two algorithm characteristics (whether the algorithm is anthropomorphized and whether it involves machine learning) and two characteristics of the task in which the error occurs (whether the task is subjective [vs. objective] and whether it is interactive [vs. noninteractive]).
An entity's mind perception of agency to intend and to act affects individuals' perception of the entity's responsibility for its actions. For example, people have lower mind perception of agency of an inanimate robot than of a man or of a young girl ([20]), suggesting that they would perceive lower responsibility for a robot's harmful actions. Extending this idea to algorithm errors, we propose that people will have lower mind perception of the agency of the algorithm (vs. a human) that commits the error responsible for the brand harm crisis and assign lower responsibility[ 6] to the algorithm for the harm caused.
The notion that people consider algorithms to have lower agency than the humans who developed them is consistent with early research on individuals' interactions with computers and robots ([38]) and the recent research on algorithm aversion ([10]) and algorithm appreciation ([35]; [41]). This argument is consistent with other evidence about algorithms ([36]) showing that because algorithms do not have "human-like" qualities, people may not hold them fully responsible for actions that cause harm.
Accordingly, we propose that consumers' responses to a brand following a brand harm crisis caused by an algorithm error (vs. human error) will be less negative. We further propose that consumers' responses to the brand will be serially mediated by their lower mind perception of the algorithm's agency, which, in turn, will lower their perceptions of the algorithm's responsibility for the harm caused by the error. Thus, we propose:
- H1: Consumers' responses to a brand following a brand harm crisis caused by an algorithm error (vs. human error) are less negative.
- H2: Consumers' lower mind perception of the algorithm's agency, which lowers their perceptions of the algorithm's responsibility for the harm caused by the error, mediates the relationship in H1.
Anthropomorphism is the process of inductive inference in which people attribute distinctively human characteristics to inanimate objects, including brands, machines, technologies, and software ([29]). Anthropomorphizing an entity involves the use of human traits (e.g., humanlike face and name) so that individuals attribute essential human characteristics (e.g., humanlike mind capable of thinking and feeling) to the entity. A common marketing practice is to name products with human names with the intent of anthropomorphization (e.g., IBM's artificial intelligence software "Watson," Bank of America's virtual financial assistant "Erica," Amazon's virtual assistant "Alexa").
The effects of anthropomorphization on consumer behaviors have received attention from marketing scholars ([ 1]; [ 2]; [28]). The overall evidence suggests that the higher a product's anthropomorphization, the higher consumers' evaluations of it ([ 1]) and the higher its sales ([32]). With regard to harm crises, anthropomorphization of a product lowers consumers' evaluations of it ([43]), which is consistent with the main effect in H1.
In the technology context, relevant to this research, firms anthropomorphize products to make them user-friendly and less intimidating ([31]). Anthropomorphizing technology-driven products increases consumers' positive feelings toward the products, reduces people's fear of technology, and suggests that the products can perform their intended functions well ([57]). This results in consumers assigning higher responsibility to anthropomorphized products, indeed, at a level comparable to what they assign to humans ([12]).
Accordingly, we suggest that when an anthropomorphized (vs. not) algorithm is the source of the error that causes a brand harm crisis, consumers will consider the anthropomorphized algorithm to have higher mind perception of agency and assign higher responsibility to it for the harm caused by the algorithm error. We hypothesize that consumers' responses to a brand following a brand harm crisis caused by an algorithm error will be more negative when the algorithm is anthropomorphized (vs. not). Thus, we propose:
- H3: Consumers' responses to a brand following a brand harm crisis are more negative when the error is caused by an anthropomorphized (vs. not) algorithm.
Machine learning algorithms learn "by themselves" (i.e., independently using historical data, models, and analyses). In other words, machine learning algorithms are programmed such that they can modify themselves (i.e., without human intervention) to improve their performance. The availability of "big data," growing computational power, and developments in software technology enable such machine learning algorithms to learn independently from their experiences working repeatedly on large data sets ([26]). Machine learning algorithms know users' behaviors and leverage that knowledge to recommend products that match users' preferences. Such machine learning algorithms power Amazon, Netflix, and Spotify recommendations; Google Maps; and much of the content on Facebook, Instagram, and Twitter.
Developments in bioethics consider an entity's capacity for learning, including the ability to think, reason, and remember, to be indicative of superior mental abilities and to define the degree of its humanness ([14]). Reiterating this view, [21] compared people's perceptions of mentally competent (vs. mentally challenged) adults and found them to be higher on mental abilities associated with learning and mind perception of agency.
Applying these ideas, we propose that consumers will ascribe more humanness to a machine learning (vs. not) algorithm. Following a brand harm crisis resulting from an error caused by a machine learning (vs. not) algorithm, people may perceive the machine learning algorithm to have higher agency and, therefore, higher responsibility for the harm caused. Thus, we hypothesize that, following a brand harm crisis caused by an error of a machine learning (vs. not) algorithm, consumers' responses to the brand will be more negative. Thus, we propose:
- H4: Consumers' responses to a brand following a brand harm crisis are more negative when the error is caused by a machine learning (vs. not) algorithm.
Following [ 6], a subjective task is open to interpretation on the basis of an individual's personal opinion, whereas an objective task is one that involves factors that are quantifiable and measurable. People perceive subjective tasks as requiring intuition and objective tasks as requiring human traits such as logical, rule-based analysis ([27]). Although algorithms are proficient at objective tasks, the growth of big data and lower costs of computing have resulted in a dramatic increase in the use of algorithms for subjective tasks ([30]). Companies routinely use algorithms for subjective tasks, such as selecting applicants (e.g., Indeed.com, university admissions) and recommending clothing for consumers (e.g., J. Jills, Stitchfix.com). Ceteris paribus, consumers perceive that algorithms lack the ability to perform subjective tasks ([ 6]), although making an algorithm more humanlike is effective at increasing its usage for subjective tasks. In other words, individuals ascribe higher humanness to an algorithm deployed for subjective tasks.
Applying this logic, we propose that when an algorithm is used for a subjective (vs. objective) task that requires intuition and an algorithm error causes the brand harm crisis, consumers will perceive the algorithm as having higher mind perception of agency and hold it more responsible for the harm caused. Therefore, we posit that in a brand harm crisis caused by an algorithm error, when the algorithm error occurs during a subjective (vs. objective) task, consumers' responses to the brand will be more negative. Thus, we propose:
- H5: Consumers' responses to a brand following a brand harm crisis are more negative when the algorithm error occurs during a subjective (vs. objective) task.
A key characteristic of interactive communications between entities (say, a human and a computer) is contingency in responses ([48]). In an interactive communication ([49]), each entity acknowledges and incorporates the other entity's prior communications. Higher interactivity between two individuals in an online context heightens perceptions of each other's humanness ([50]). Interactivity between an individual and a nonhuman entity (e.g., an algorithm) makes the entity more humanlike because it mimics the contingency in real-time interactive exchanges between humans ([44]). Therefore, people may perceive the algorithm in an interactive task as being capable of communication, an integral aspect of people's mind perception of agency of an entity ([23]). Indeed, algorithms are now widely used by marketers in interactive communications including in customer service chatbots (e.g., Spotify) and product recommendations (e.g., Stitchfix).
Applying these ideas, we anticipate that consumers will have higher mind perception of agency when the task in which the algorithm error occurs is interactive (vs. noninteractive) between consumers and the algorithm. We propose that, following a brand harm crisis caused by an algorithm error in an interactive (vs. noninteractive) task, consumers will hold the algorithm more responsible for the harm, causing their responses to the brand to be more negative. Thus, we propose:
- H6: Consumers' responses to a brand following a brand harm crisis caused by an algorithm error are more negative when the error occurs in an interactive (vs. noninteractive) task.
We conducted a prestudy that examined consumers' responses to a brand when there are no errors to ensure that the effects we theorize relate only to the error caused by the algorithm (vs. human) and not to algorithms in general.[ 7] We preregistered this prestudy on AsPredicted.org (#43436). We provide stimuli for the prestudy and all other studies in the Web Appendix and summary of the studies and findings in Table 1.
Graph
Table 1. Overview of Studies.
| Study | Participants | Context; Dependent Variable (DV) | Conditions and Results | Conclusion |
|---|
| Prestudy | N = 403, online | Financial investments; DV: brand attitude | Error | No error | When there is no error, there is no difference in consumers' responses to a brand that uses an algorithm versus a human.As hypothesized in H1, consumers' responses to a brand following a brand harm crisis are less negative when the error is caused by an algorithm (vs. human). |
| Algorithm 4.55 (1.56)5 | Human 3.63 (1.79) | Algorithm 5.55 (1.03) | Human 5.31 (1.07) |
| 1a | N = 157, online | Mistake in headline of a fundraising advertisement;DV: amount of donation | Algorithm error 20.71 (41.91) | Human error 7.84 (22.34) | Support for H1. |
| 1b | N = 233, online | Online platform made a mistake and provided bad financial advice;DV: advice provided | Algorithm error 3.19 (2.18) | Human error 2.49 (1.91) | Support for H1. |
| 1c | N = 177, U.S. undergraduate students | Glitch in the Qualtrics online computer system;DV: % intention to re-engage with the brand | Algorithm error 64.3% | Human error 42.1% | Support for H1. |
| 2 | N = 251, online | Recall of 4.8 million vehicles;DV: brand attitudemediators: (1) mind perception of source of the error's agency in committing the error and (2) perceptions of source of the error's responsibility for the harm caused to the brand. | Algorithm error 4.59 (1.61) | Human error 4.17 (1.69) | Support for H1, H2, and for serial mediation by lower agency of the algorithm and responsibility for the harm caused by the algorithm error. |
| 3 | N = 372, online | Mistake in investment decisions for customers of a financial investment company;DVs: brand attitude, donation amount | Human error 3.63 (1.81) 156.88 (165.19) | Algorithm error 4.25 (1.84) 204.98 (180.05) | Anthropomorphized algorithm Error 3.74 (1.76) 160.40 (157.47) | As hypothesized in H3, consumers' responses to a brand following a brand harm crisis caused by an algorithm error are more negative when the error is caused by an anthropomorphized (vs. not) algorithm. |
| 4 | N = 310, online | Mistake in users' Twitter timelines;DV: brand attitude | Human error 4.20 (1.47) | Algorithm error 4.76 (1.62) | Machine learning algorithm error 4.21 (1.45) | As hypothesized in H4, consumers' responses to a brand following a brand harm crisis caused by an algorithm error are more negative when the error is caused by a machine learning (vs. not) algorithm. |
| 5 | N = 400, online | A leading university in the United States faced a crisis because of an error in the subjective (vs. objective) assessment of Asian American students' applicationsDV: brand attitude | Algorithm error | Human error | As hypothesized in H5, consumers' responses to a brand following a brand harm crisis caused by an algorithm error are more negative when the algorithm error occurs during a subjective (vs. objective) task. |
| Subjective task 3.76 (1.64) | Objective task 4.46(1.49) | Subjective task 4.28 (1.73) | Objective task 3.99 (1.64) |
| 6 | N = 328, U.S. undergraduate students | Mistake in product selection by a personal stylist of a fashion retailer companyDV: brand attitude | Algorithm error | Human error | As hypothesized in H6, consumers' responses to a brand following a brand harm crisis caused by an algorithm error are more negative when the error occurs during an interactive (vs. noninteractive) task. |
| Interactive task 3.41 (1.05) | Noninteractive task 3.82(1.22) | Interactive task 3.48 (.98) | Noninteractive task 3.37 (.95) |
| M1 | N = 368, online | Recall of 4.8 million vehiclesDV: brand attitude | Algorithm error | Human error | Responses to a brand following a brand harm crisis caused by an algorithm error are more negative when there is more human supervision (vs. technological supervision). |
| Technological supervision 4.71 (1.69) | Human supervision 4.13 (1.56) | Technological supervision 4.15 (1.65) | Human supervision 4.63 (1.69) |
1 Note: Figures in parentheses are standard deviations.
A total of 403 adults participated in the experiment on Amazon Mechanical Turk (MTurk) in exchange for $.50 (219 male; Mage = 37.73 years, SD = 12.40). The study used an error (vs. no error) and algorithm (vs. human) between-subjects design.
We randomly assigned participants to error and "no error" conditions. Participants in the error condition read that HMS Investments, a leading financial investment company, was facing a crisis. In the "no error" condition, participants read that HMS Investments reduces risks for its clients. We also randomly assigned participants to algorithm and human conditions. Participants in the algorithm (human) error condition read that HMS Investments, a leading financial investment company, was facing a crisis because a financial algorithm program (financial manager) had committed an error, resulting in financial losses for its customers. Participants in the algorithm (human) "no error" condition read that HMS Investments reduces risks for its clients with its strong computer algorithms (employees). We measured participants' attitude toward HMS Investments using a five-item scale ([ 3]; [51]): "bad/good," "low quality/high quality," "undesirable/desirable," "harmful/beneficial," and "unfavorable/favorable" (α =.96). Participants then provided their basic demographic information.
An analysis of variance (ANOVA) on brand attitude reveals the predicted interaction effect of error (vs. no error) and algorithm (vs. human) conditions, F( 1, 399) = 5.86, p =.016. There is no main effect of error (vs. no error), p =.157, and algorithm (vs. human) conditions, p =.335. Participants' responses to a brand following a brand harm crisis caused by an algorithm (vs. human) error are less negative: MAE = 4.55, SD = 1.56 versus MHE = 3.63, SD = 1.79, F( 1, 399) = 21.63, p <.001. However, when there is no error, there is no difference in participants' responses when the brand uses algorithms versus when they use humans: Malgorithm = 5.55, SD = 1.03 versus Mhuman = 5.31, SD = 1.07, F( 1, 399) = 1.53, p =.217. There is no effect of age, p =.085, or gender, p =.612.
Thus, when there is no error, consumers' responses to a brand are similar regardless of whether the brand uses an algorithm or a human. However, as hypothesized in H1, consumers' responses to a brand following a brand harm crisis caused by an algorithm (vs. human) error are less negative. These findings provide preliminary support for H1 and indicate that algorithm error, and not the mere presence of the algorithm, drives the results in the subsequent studies.
In Study 1a, we examine consumers' responses to a brand following a brand harm crisis caused by an algorithm error (vs. human error) with an incentive-compatible experiment using a consequential outcome (donation to a charity suggested by the brand) as the dependent variable. A lower donation to the charity denotes a more negative response to the brand.
Participants read about a consumer electronics retailer for which an algorithm error or a human error had caused a brand harm crisis. Participants then indicated the amount that they were willing to donate to the World Health Organization through the electronics retailer from the compensation that they would receive in the study.
A total of 157 U.S. adults participated in the experiment on MTurk in exchange for $1.50 (84 male; Mage = 40.69 years, SD = 10.37). All participants read that a consumer electronics company, Qualtronics, was facing a harm crisis. This was because their fundraising campaign, BanishCovid19, aimed at combating the COVID-19 pandemic, implied that a Chinese virus caused COVID-19. The fundraising campaign was for the World Health Organization.
Participants in the algorithm error condition read, "because the disease was first detected in Wuhan Province of China, Qualtronics used computer algorithms to design the advertisement and released the campaign with the headline 'Contribute to BanishCovid19 and Destroy the Chinese Virus.'" Participants in the human error condition read, "because the disease was first detected in Wuhan Province of China, Qualtronics' managers designed the advertisement and released the campaign with the headline 'Contribute to BanishCovid19 and Destroy the Chinese Virus.'" In both conditions, participants read that following negative feedback from their customers, Qualtronics apologized to its customers and changed the advertisement headline to "Contribute to BanishCovid19 and Destroy the Corona Virus."
We then provided participants in both conditions the opportunity to donate to the World Health Organization through Qualtronics. The maximum amount they could donate was the $1.50 they would earn in the study. Participants indicated the amount they would donate on a sliding scale (M = 14.40 cents, SD = 34.47 cents).
As a manipulation check, participants indicated the extent to which they believed the error was caused by a human in Qualtronics (1 = "not at all," and 7 = "very much"). Participants also indicated the extent to which they were concerned about COVID-19 and the extent to which COVID-19 had impacted their community on seven-point scales (1 = "not at all," and 7 = "very much"). Participants then provided their basic demographic information, including race.
As intended, participants in the human error (vs. algorithm error) condition indicated that the source of the error in Qualtronics is more human: MHE = 6.29, SD = 1.34 versus MAE = 5.81, SD = 1.57, t(155) = 2.03, p =.044.
The results indicate a significant effect of algorithm error (vs. human error) condition on the donation amount: MAE = 20.71 cents, SD = 41.91 cents versus MHE = 7.84 cents, SD = 22.34 cents, F( 1, 155) = 5.70, p =.018. When we included the three control variables of participants' race, concerns about COVID-19, and COVID-19's impact on their community in the model, the effect of algorithm error (vs. human error) on the amount of the donation is still significant, F( 1, 152) = 5.11, p =.024. There is no main effect of the three control variables or of age, p =.289, or gender, p =.202.
In Study 1a, consumers' donations following a brand harm crisis caused by an algorithm error (vs. human error) are higher. This finding supports the prediction (H1) that consumers' responses to the brand following a brand harm crisis caused by an algorithm error (vs. human error) are less negative.
In Study 1b, we examine consumers' behavioral responses to a brand harm crisis caused by an algorithm error (vs. human error) at a fictitious global platform company, Life Skills Without Borders, an online advice crowdsourcing website for young adults. We randomly assigned participants to either the algorithm error or human error condition and measured the number of items of advice provided by participants to Life Skills Without Borders following a brand harm crisis.
The experiment used the algorithm error (vs. human error) condition as a between-subjects design. A total of 233 participants participated in the experiment on the Prolific online platform in exchange for one British pound (101 male; Mage = 35.89 years, SD = 12.36).
All participants read about Life Skills Without Borders, a global crowdsourcing platform for providing life skills advice to young adults. We randomly assigned participants to either the algorithm error or human error conditions. Participants in the algorithm error (human error) condition read that a computer algorithm (an employee) at Life Skills Without Borders had made a mistake and provided bad financial advice to poor young couples, resulting in financial losses.
We then informed participants that Life Skills Without Borders was presently crowdsourcing ideas for providing postgraduation career advice to young adults. We asked participants to provide advice to Life Skills Without Borders, which was the study's dependent variable. We used the number of unique items of advice provided by each participant (e.g., "follow your heart and work at a job that you think you might like," "if you get an offer for a higher paying job at a different company be sure you want the job before you take it") as the dependent variable. Because the only difference between the two (between-subjects) conditions was the source of the error (algorithm vs. human), we considered higher numbers of pieces of advice provided by participants as indicative of participants' less negative response to Life Skills Without Borders. Participants then provided their basic demographic information.
One of the authors coded the number of unique items of advice provided by the participants (Madvice# = 2.52, SD = 2.13). A t-test shows that participants in the algorithm error condition provided more advice than participants in the human error condition: MAE = 3.19, SD = 2.18 versus MHE = 2.49, SD = 1.91, t(229) = 72.56, p =.011. The results of Study 1b support our prediction (H1) that consumers' behavioral responses to a brand following a brand harm crisis caused by an algorithm error (vs. human error) are less negative.
In Study 1c (details in the Web Appendix), we examine consumers' reengagement behaviors with the brand following a brand harm crisis caused by a failure in an online computer system. An algorithm error (vs. human error) disrupted an online task in Qualtrics, a software program used for lab experiments. We randomly assigned participants either to the algorithm error or human error condition and noted their decision to repeat or not repeat the online task (i.e., reengage with Qualtrics). Participants' willingness to repeat the task would indicate a less negative response to the Qualtrics brand. The results support our prediction (H1) that consumers' reengagement behaviors with the brand (i.e., whether they would repeat the online task) following a brand harm crisis are less negative when caused by an algorithm error (vs. human error).
In Study 2, we examine the role of consumers' mind perception of the source of the error's agency in committing the error and responsibility for the harm caused in serially mediating consumers' responses to the brand following a brand harm crisis caused by an algorithm error (vs. human error) (H2).[ 8] For a test of the mediation (H2), we measured participants' mind perception of agency of the source of the error that caused the brand harm crisis and perceptions of the source of the error's responsibility for the harm caused by the error.
A total of 251 adults participated in the between-subjects experiment on the MTurk online platform in exchange for $.50 (137 male; Mage = 34.98 years, SD = 11.19). All participants saw a tweet on the official Twitter account of The New York Times announcing the recall of 4.8 million Fiat Chrysler vehicles because of a cruise control problem. We randomly assigned participants to either the algorithm error or human error condition. Participants in the algorithm error (human error) condition read that a Fiat Chrysler computer algorithm (Fiat Chrysler employees) had made a mistake resulting in a defect in the cruise control system, causing a safety hazard.
We measured participants' attitude toward the Fiat Chrysler brand using a five-item scale ([ 3]; [51]): "bad/good," "low quality/high quality," "undesirable/desirable," "harmful/beneficial," and "unfavorable/favorable" (α =.96). We measured participants' mind perception of agency of the source of the error using [20] seven-item scale, which consisted of the following items: ( 1) telling right from wrong, ( 2) remembering things, ( 3) understanding how others feel, ( 4) conveying thoughts to others, ( 5) making plans, ( 6) exercising self-restraint over impulses, and ( 7) thinking (1 = "not at all," and 7 = "very much"; α =.95). We then measured participants' perceptions of the source of the error's responsibility for the harm caused by the error using [57] four-item scale. The items are the extent to which the source of the error at Fiat Chrysler ( 1) was responsible, ( 2) must be held to account, ( 3) deserves blame, and ( 4) was blameworthy for the harm caused by the error (1 = "not at all," and 7 = "very much"; α =.93).
As a manipulation check, we asked participants to indicate the extent to which they thought the source of the error was a human and the extent to which they thought the source of the error was a computer algorithm (1 = "not at all," and 7 = "very much"). Finally, participants provided their basic demographic information.
As intended, participants in the human error (vs. algorithm error) condition indicated that the source of the error is more human: MHE = 5.32, SD = 1.41 versus MAE = 4.22, SD = 1.73, F( 1, 249) = 30.18, p <.001. Participants in the algorithm error (vs. human error) condition indicated that the source of the error is more algorithm-like: MAE = 5.07, SD = 1.60 versus MHE = 4.03, SD = 1.61, F( 1, 249) = 26.33, p <.001.
A one-way ANOVA on participants' attitude toward the brand, Fiat Chrysler, is significant: F( 1, 249) = 4.09, p =.044. Supporting H1, participants' responses to the brand following a brand harm crisis are less negative when the error is an algorithm error (vs. human error): MAE = 4.59, SD = 1.61 versus MHE = 4.17, SD = 1.69.
Next, we tested the mediating roles of mind perception of agency of the source of the error and the source of the error's responsibility for the harm caused by the error in mediating participants' responses to the brand following a brand harm crisis (H2). We note that the means of mind perception of agency and responsibility for the harm caused in the algorithm error and human error conditions are, respectively, as follows: MAE = 3.65, SD = 1.65 versus MHE = 4.87, SD = 1.37, F( 1, 249) = 40.66, p <.001; MAE = 4.53, SD = 1.64 versus MHE = 5.11, SD = 1.35, F( 1, 249) = 9.56, p =.002.
We first regressed participants' perceptions of the source of the error's responsibility for the harm caused on algorithm error (vs. human error) condition and found a significant effect: β =.19, p =.002. We then regressed participants' mind perception of the source of the error's agency in committing the error on algorithm error (vs. human error) condition and found a significant effect: β =.37, p <.001. We then regressed participants' perception of the source of the error's responsibility for the harm caused on both algorithm error (vs. human error) condition and mind perception of the source of the error's agency in committing the error. While there is no effect of algorithm error (vs. human error) condition, β =.04, p =.54, there is a significant effect of mind perception of the source of the error's agency in committing the error: β =.41, p <.001.
Next, we formally tested the proposed serial mediation model (H2). We used PROCESS Macro Model 6 ([25]), where algorithm error (vs. human error) was the independent variable, participants' mind perception of the source of the error's agency in committing the error and perception of the source of the error's responsibility for the harm were the serial mediators, and brand attitude was the dependent variable. The model first tested the effect of the algorithm error (vs. human error) and mind perception of the source of the error's agency in committing the error on perception of the source of the error's responsibility for the harm caused. The results show no effect of algorithm error (vs. human error), β =.1167, 95% CI = [−.2535,.4869], but a significant effect of mind perception of the source of the error's agency in committing the error on the source of the error's responsibility for the harm caused: β =.3911, 95% CI = [.2762,.5059].
The model then tested for the effects of algorithm error (vs. human error), mind perception of the source of the error's agency in committing the error, and perception of the source of the error's responsibility for the harm caused on brand attitude. The results show a significant effect of algorithm error (vs. human error) (β = −.6306, 95% CI = [−1.0555, −.2058]), participants' mind perception of the source of the error's agency in committing the error (β =.3105, 95% CI = [.1673,.4536]), and perception of the source of the error's responsibility for the harm caused (β = −.2794, 95% CI = [−.4228, −.1360]) on brand attitude. The 95% bias-corrected bootstrap CI for the indirect effect of algorithm error (vs. human error) on brand attitude is significant (β = −.1309; 95% CI = [−.2413, −.0498]), indicating serial mediation by mind perception of the source of the error's agency in committing the error and perception of the source of the error's responsibility for the harm caused.
The results of Study 2 offer two findings. First, supporting H1, consumers' responses to the brand following a brand harm crisis caused by an algorithm (vs. human) error are less negative. Second, in support of H2, following a brand harm crisis caused by an algorithm error, participants' mind perception of the source of the error's agency in committing the error and perception of the source of the error's responsibility for the harm serially mediate consumers' less negative responses to the brand. As the serial mediation is only partial, there may be other theoretical mechanisms that can be explored in future research.
In Study 3, we examine H3, which proposes that consumers' responses to a brand following a brand harm crisis will be more negative when the error is caused by an anthropomorphized (vs. not) algorithm. We use an incentive-compatible experimental design with a consequential outcome, a donation to a Feeding America network suggested by the brand, as the dependent variable. We also measure participants' brand attitude. In this study, a fictitious financial investment company, HMS Investments, is facing a crisis because it made a mistake in its investment decisions for its customers, resulting in financial losses for them. We preregistered this study on AsPredicted.org (#44090).
A total of 372 adults (180 female, Mage = 36.34 years, SD = 14.03) participated in the experiment on the MTurk online platform in exchange for one U.S. dollar. As it is not meaningful to consider anthropomorphized humans, we used a three-factor experimental design consisting of algorithm error, human error, and anthropomorphized algorithm error conditions, to which we randomly assigned participants. We informed participants that HMS Investments, a leading financial investment company, was facing a crisis because a financial algorithm program (financial manager or financial algorithm program Charles) had committed an error, resulting in financial losses for its customers.
We measured participants' attitude toward the brand, HMS Investments, using the same five-item scale used in Study 2 (α =.97). We informed participants that the study's researchers decided to randomly give 20 participants a five-dollar bonus, from which participants could donate to Feeding America, the largest domestic hunger-relief organization in the United States through HMS Investments. We informed them that each dollar donated provides about ten meals to families in need through the Feeding America's network of food banks. Participants indicated the amount they were willing to donate to Feeding America, ranging from nothing to five dollars.
As a manipulation check, we asked participants to indicate the extent to which they thought the source of the error at HMS Investments was a human (and algorithm) on a two-item scale (1 = "not at all," and 7 = "very much"). Participants also provided perceptions of the extent to which the news was from a credible source and the extent to which the news was believable on a two-item scale (1 = "not at all," and 7 = "very much"). Results show no effect of algorithm error versus anthropomorphized algorithm error versus human error conditions on the news' credibility, F( 2, 369) =.23, p =.794, or its believability, F( 2, 369) =.653, p =.521. Finally, participants provided their basic demographic information.
As intended, participants in the human error (vs. algorithm error and vs. anthropomorphized algorithm error) condition indicated that the source of the error at HMS Investments was more human: MHE = 5.84, SD = 1.24 versus MAE = 4.15, SD = 1.95 versus MAAE = 4.26, SD = 1.81, F( 2, 369) = 38.48, p <.001. As intended, participants in the algorithm error and anthropomorphized algorithm error (vs. human error) conditions indicated that the source of the error at HMS Investments was more algorithm-like: MAE = 5.40, SD = 1.63 versus MAAE = 5.35, SD = 1.64 versus MHE = 3.49, SD = 1.96, F( 2, 369) = 47.93, p <.001.
Consistent with H3, a one-way ANOVA on participants' brand attitude toward HMS Investments is significant: F( 2, 369) = 4.19, p =.016. Supporting H3, participants' responses to the brand following a brand harm crisis caused by an algorithm error are more negative when the algorithm is anthropomorphized (vs. not): MAAE = 3.74, SD = 1.76 versus MAE = 4.25, SD = 1.84, p =.027. Furthermore, in support of H1, following a brand harm crisis caused by an algorithm error (vs. human error), participants' brand attitude is less negative, MAE = 4.25, SD = 1.84 versus MHE = 3.63, SD = 1.81, p =.008. In addition, there is no difference in participants' brand attitude when a brand harm crisis is caused by an anthropomorphized algorithm error versus a human error: MAAE = 3.74, SD = 1.76 versus MHE = 3.63, SD = 1.81, p =.617.
Consistent with H3, a one-way ANOVA on participants' donation is significant: F( 2, 369) = 3.18, p =.043. Participants' donations to Feeding America are higher when the brand harm crisis at HMS Investments is caused by an algorithm error (vs. anthropomorphized algorithm error): MAAE = 160.40 (cents), SD = 157.47 versus MAE = 204.98 (cents), SD = 180.05, p =.039. Following a brand harm crisis, participants' donations to Feeding America are higher when the error is caused by an algorithm error (vs. human error): MAE = 204.98 (cents), SD = 180.05 versus MHE = 156.88 (cents), SD = 165.19, p =.029. Furthermore, there is no difference in participants' donations to Feeding America when the brand harm crisis is caused by an anthropomorphized algorithm error versus a human error: MAAE = 160.40 (cents), SD = 157.47 versus MHE = 156.88 (cents), SD = 165.19, p =.864.
The results of Study 3 offer two key findings. First, supporting H3, consumers' responses to the brand following a brand harm crisis caused by an algorithm error are more negative when the algorithm is anthropomorphized (vs. not). Second, in support of H1, consumers' responses to the brand following a brand harm crisis are less negative when the error is caused by an algorithm (vs. human).
In Study 4, we examine H4, which proposes that consumers' responses to a brand following a brand harm crisis will be more negative when the error is caused by a machine learning (vs. not) algorithm. In this study, Twitter was facing a crisis because it made a mistake in the timelines of its users so that some of the displayed tweets had inappropriate and offensive content. We measured participants' attitude toward Twitter following the brand harm crisis.
A total of 310 adults participated in the experiment on the MTurk online platform in exchange for one U.S. dollar (155 male; Mage = 34.95 years, SD = 11.04). We informed participants that when users log in to Twitter, their home timelines display a stream of tweets from accounts that they have chosen to follow on Twitter.
As in Study 3, it was not useful to consider machine learning humans. Therefore, we again used a three-factor experimental design consisting of algorithm error, human error, and machine learning algorithm error conditions to which we randomly assigned participants. In the algorithm error (human error) condition, participants read that Twitter uses algorithms (employees) to evaluate scores and determine which tweets to display. In the machine learning algorithm error condition, participants read that Twitter uses machine learning algorithms to determine which tweets to display. In addition, participants in the machine learning algorithm error condition read that machine learning algorithms are algorithms that learn from past data and analyses to make their decisions. We then informed participants that there had been some problems in Twitter timelines that resulted in the incorrect display of tweets for users. Some of these incorrectly displayed tweets had inappropriate content that had offended some Twitter users.
We measured participants' attitude toward Twitter, using the same five-item scale used in Study 2 (α =.94). As a manipulation check, we asked participants for their perceptions of whether the source of the error at Twitter was a human (1 = "not at all," and 7 = "very much"). Participants also provided perceptions of the extent to which the news was from a credible source and the extent to which the news was believable on a two-item scale (1 = "not at all," and 7 = "very much"). Results show no effect of algorithm error versus machine learning algorithm error versus human error conditions on the news' credibility, F( 2, 307) = 1.51, p =.22, or its believability, F( 2, 307) = 1.26, p =.28. Finally, participants indicated whether they had a Twitter account and provided their basic demographic information.
As intended, participants in the human error (vs. algorithm error) condition indicated that the source of the error at Twitter was more human: MHE = 5.06, SD = 1.75 versus MAE = 4.49, SD = 1.60, p =.014. There is no significant difference between the participants in the human error (vs. machine learning algorithm error) condition regarding the extent to which they thought the source of the error at Twitter was more human: MHE = 5.06, SD = 1.75 versus MMLAE = 4.79, SD = 1.54, p =.243. There is also no significant difference between participants in the algorithm error (vs. machine learning algorithm error) condition regarding the extent to which they thought the source of the error at Twitter was more human: MAE = 4.49, SD = 1.60 versus MMLAE = 4.79, SD = 1.54, p =.176.
Consistent with H4, a one-way ANOVA on participants' attitude toward Twitter is significant, F( 2, 307) = 4.72, p =.010. Supporting H4, participants' responses to the brand following a brand harm crisis are more negative when the error was caused by a machine learning (vs. not) algorithm: MMLAE = 4.21, SD = 1.45 versus MAE = 4.76, SD = 1.62, p =.011. Furthermore, in support of H1, following a brand harm crisis caused by an algorithm error (vs. human error), participants' responses to the brand are less negative: MAE = 4.76, SD = 1.62 versus MHE = 4.20, SD = 1.47, p =.009. There is no difference in participants' responses to the brand following a brand harm crisis caused by a machine learning algorithm versus a human: MMLAE = 4.21, SD = 1.45 versus MHE = 4.20, SD = 1.47, p =.95. Whether participants have a Twitter account did not change the effect of the algorithm error versus machine learning algorithm error versus human error condition on their attitude toward Twitter, F( 2, 306) = 4.58, p =.011.
The results of Study 4 offer two key findings. First, supporting H4, consumers' responses to the brand following a brand harm crisis are more negative when the error is caused by a machine learning (vs. not) algorithm. Second, in support of H1, following a brand harm crisis caused by an algorithm error (vs. human error), consumers' responses to the brand are less negative.
In Study 5, we test H5, which proposes that consumers' responses to a brand following a brand harm crisis will be more negative when the algorithm error occurs during a subjective (vs. objective) task. In this study, we informed participants that a leading university in the United States was facing a crisis because of an error in the subjective (vs. objective) assessment of Asian American students' applications. We measured participants' attitude toward the university. We preregistered the study at AsPredicted.org (#43396).
A total of 400 adults (199 female, Mage = 35.70 years, SD = 11.97) participated in the study on the MTurk online platform in exchange for monetary compensation. We used a 2 (algorithm error, human error) × 2 (subjective task, objective task) between-subjects design.
We informed participants that a leading university in the United States was experiencing a crisis because of a mistake in assessments of the applications of prospective Asian American students. We randomly assigned participants to either the algorithm error or human error condition. In the algorithm error (human error) condition, we informed participants that the computer algorithm (employees) had made the mistake.
We informed participants that the university used both subjective and objective methods in their admissions process. The subjective methods included analyzing the applicant's personality and social skills, including "positive personality," likability, courage, kindness, and being "widely respected." The objective methods included reviewing the applicant's test scores and grades. Participants in the subjective (objective) task condition read that the error was in the subjective (objective) assessment of the application. Participants in the subjective task condition read that Asian American applicants were rated lower than other applicants on traits like positive personality, likability, courage, kindness, and being widely respected. Participants in the objective task condition read that the error was based on the incorrect use of lower test scores for the Asian American applicants. In both conditions, participants read that the error resulted in the university incorrectly declining applications of hundreds of qualified and acceptance-worthy Asian American applicants.
We measured participants' attitude toward the university using the same five-item scale used in Study 2 (α =.95). As a manipulation check, we asked participants to indicate the extent to which they thought the source of the error was human, the extent to which they thought the source of the error was an algorithm, and the extent to which they thought the error was related to an objective task (1 = "not at all," and 7 = "very much"). Participants also provided perceptions of the extent to which the news was from a credible source and the extent to which the news was believable (1 = "not at all," and 7 = "very much"). Results showed no effect of algorithm error versus human error conditions and subjective versus objective task conditions on the news' credibility, F( 1, 396) =.00, p =.994, or its believability, F( 1, 396) =.454, p =.501. Finally, participants provided their basic demographic information.
As intended, participants in the human error (vs. algorithm error) condition indicated that the source of the error was more human: MHE = 5.33, SD = 1.41 versus MAE = 4.45, SD = 1.51, t(398) = 6.003, p <.001. Participants in the algorithm error (vs. human error) condition indicated that the source of the error was more algorithm-like: MAE = 4.68, SD = 1.64 versus MHE = 3.84, SD = 1.68, t(398) = −5.047, p <.001. Participants in the objective task (vs. subjective task) condition indicated that the error was likely to have occurred during an objective task: Mobjective = 4.60, SD = 1.41 versus Msubjective = 3.94, SD = 1.74, t(398) = −4.184, p <.001.
Consistent with H5, an ANOVA on participants' brand attitude reveals the predicted interaction effect of algorithm error (vs. human error) and subjective (vs. objective) task conditions, F( 1, 396) = 9.15, p =.003. Supporting H5, participants' responses to a brand following a brand harm crisis caused by an algorithm error are more negative when the error occurs during a subjective (vs. objective) task: Msubjective = 3.76, SD = 1.64 versus Mobjective = 4.46, SD = 1.49, F( 1, 396) = 9.06, p =.003. There is no difference in participants' responses to a brand following a brand harm crisis caused by a human error when the error occurs during a subjective versus objective task: Msubjective = 4.28, SD = 1.73 versus Mobjective = 3.99, SD = 1.64, F( 1, 396) = 1.58, p =.21.
In support of H1, participants' responses to a brand following a brand harm crisis caused by an algorithm error (vs. human error) are less negative when the error occurs during an objective task: MAE = 4.46, SD = 1.49 versus MHE = 3.99, SD = 1.64, F( 1, 396) = 4.13, p =.043. Participants' responses to a brand following a brand harm crisis caused by an algorithm error (vs. human error) are more negative when the error occurs during a subjective task: MAE = 3.76, SD = 1.64 versus MHE = 4.28, SD = 1.73, F( 1, 396) = 5.04, p =.025.
The results of Study 5 offer two key findings. First, supporting H5, consumers' responses to a brand following a brand harm crisis are more negative when the algorithm error occurs during a subjective (vs. objective) task. Second, supporting H1, consumers' responses to the brand following a brand harm crisis caused by an algorithm error (vs. human error) are less negative when the error occurs during an objective task.
In Study 6, we test H6, which proposes that consumers' responses to a brand following a brand harm crisis caused by an algorithm error will be more negative when there is interactivity (vs. not) with the algorithm during the task in which the error occurs. We informed participants that a fictitious leading fashion retailer brand, D&J, has been facing growing customer complaints because of its personal stylists. Participants were randomly assigned to algorithm error (vs. human error) and interactive (vs. noninteractive) task conditions. We measured participants' brand attitude.
A total of 328 students (206 female, Mage = 20.12 years, SD = 1.64) from a university in the southern United States participated in the laboratory experiment in exchange for course credit. We used a 2 (algorithm error, human error) × 2 (interactive task: yes, no) between-subjects design.
We randomly assigned participants to either the algorithm error or human error conditions. Participants in the algorithm error (human error) condition read that in recent weeks, D&J, a leading fashion retailer brand, had been facing growing customer complaints because of problems caused by its algorithm (human) personal stylists, which were recently introduced by D&J and were intended to personalize products for customers to reflect and accentuate their personalities. Participants were assigned to the interactive and noninteractive task conditions. To ensure realism, we did not use the word "human" in the human error condition.
Participants in the interactive task condition read that customers who wanted to use the interactive algorithm (personal) stylists first completed an online form, which collected a personal photograph and details of customers' height, weight, and personal likes and dislikes of different colors and styles. Then, the D&J algorithm (personal) stylists interacted with customers such that customers could see how the products would look on them and could work with the D&J algorithm (personal) stylists to choose the right products. The customer was thus actively involved in the selection of products by algorithm (personal) stylists. Using the information provided by the customer, the algorithm (personal) stylists chose and shipped products to customers.
Participants in the noninteractive task condition read that customers who used the algorithm (personal) stylists first completed an online form that collected a personal photograph and details of their height, weight, and personal likes and dislikes of different colors and styles. Then, the D&J algorithm (personal) stylists chose the right products for the customer. The customer was not involved in the selection of products done by algorithm (personal) stylists. Using the information provided by the customer, the algorithm (personal) stylists chose and shipped products to customers. Participants read that customers stated stylists misled them, causing them to buy very expensive products that did not reflect their personalities and were, in fact, mismatched with their personalities. Customers were now demanding refunds for these products and threatening to sue D&J.
We measured participants' brand attitude, using the same five-item scale used in Study 2 (α =.88). As a manipulation check, we asked participants to indicate the extent to which they thought the source of the error at D&J was a human and the extent to which they thought the error occurred during a task in which there was communication between the personal stylist and the customer, which indicates interactivity on a two-item scale (1 = "not at all," and 7 = "very much"). Participants also provided perceptions of the extent to which the news was from a credible source and the extent to which the news was believable on a two-item scale (1 = "not at all," and 7 = "very much"). Results showed no effect of algorithm error versus human error condition and interactive versus noninteractive task conditions on the news' credibility, F( 1, 324) = 1.49, p =.22, or its believability, F( 1, 324) =.018, p =.89. Finally, participants provided their basic demographic information.
As intended, participants in the human error (vs. algorithm error) condition indicated that the source of the error was more human: MHE = 5.19, SD = 1.47 versus MAE = 4.40, SD = 1.51, t(326) = 4.83, p <.001. Participants in the interactive (vs. noninteractive) task indicated that the error was more likely to have occurred on a task in which there was more communication between the personal stylist and the customer: Minteractive = 3.51, SD = 1.57 versus Mnoninteractive = 3.04, SD = 1.41, t(326) = −2.90, p =.004.
Consistent with H6, an ANOVA on brand attitude reveals the predicted interaction effect of algorithm error (vs. human error) and interactive (vs. noninteractive) task conditions, F( 1, 324) = 5.05, p =.025. Supporting H6, participants' responses to a brand following a brand harm crisis caused by an algorithm error are more negative when there is interactivity (vs. not) with the algorithm during the task in which the error occurs: Minteractive = 3.41, SD = 1.05 versus Mnot = 3.82, SD = 1.22, F( 1, 324) = 6.33, p =.012. There is no difference in participants' responses to a brand following a brand harm crisis caused by a human error when there was interactivity versus no interactivity with the employee during the task: Minteractive = 3.48, SD =.98 versus Mnoninteractive = 3.37, SD =.95, F( 1, 324) =.44, p =.51.
In support of H1, participants' responses to a brand following a brand harm crisis caused by an algorithm error (vs. human error) are less negative when the task in which the error occurs is noninteractive: MAE = 3.82, SD = 1.22 versus MHE = 3.37, SD =.95, F( 1, 324) = 7.59, p =.006. There is no difference in participants' responses to a brand following a brand harm crisis caused by an algorithm error versus a human error when the task was interactive: MAE = 3.41, SD = 1.05 versus MHE = 3.48, SD =.98, F( 1, 324) =.179, p =.672.
The results of Study 6 offer two findings. First, supporting H6, consumers' responses to a brand following a brand harm crisis caused by an algorithm error are more negative when the task in which the error occurs is interactive (vs. noninteractive). Second, supporting H1, consumers' responses to the brand following a brand harm crisis caused by an algorithm (vs. human) error are less negative when the task in which the error occurs is noninteractive.
As algorithm errors are, unfortunately, common in business practice, firms undertake interventions to manage the aftermath of such brand harm crises. The baseline intervention in algorithm errors is technological supervision of the algorithm (e.g., facial recognition algorithm failures at Microsoft; [45]). Another common intervention following brand harm crises caused by an algorithm error is to increase human supervision of the algorithm ([33]). As Sheryl Sandberg, Chief Operating Officer, Facebook, noted (in [46]) after an algorithm error caused the display of anti-Semitic ads, "we're adding more human review and oversight to our automated processes....From now on we will have more manual review of new ad targeting options to help prevent offensive terms from appearing." To generate managerial guidance, we conducted a study (M1) in which we examined consumers' responses to human supervision and technological supervision following brand harm crises caused by an algorithm (vs. human) error.
A total of 368 adults (171 female, Mage = 35.08 years, SD = 11.06) participated in the study on the MTurk online platform in exchange for monetary compensation. We used a 2 (algorithm error, human error) × 2 (human supervision, technological supervision) between-subjects design. We preregistered the study at AsPredicted.org (#53178).
All participants saw a tweet on the official Twitter account of The New York Times announcing the recall of 4.8 million Fiat Chrysler vehicles due to a cruise control problem. We assigned participants to the algorithm error and human error conditions. Participants in the algorithm error condition read that the computer algorithm at Fiat Chrysler had made a mistake resulting in a defect in the cruise control system, causing a safety hazard. Participants in the human error condition read that the employees of Fiat Chrysler had made a mistake resulting in a defect in the cruise control system, causing a safety hazard. We then randomly assigned participants to human supervision and technological supervision conditions. We informed participants in the human supervision condition that Fiat Chrysler would increase managerial supervision in its manufacturing processes to prevent such errors. We informed participants in the technological supervision condition that Fiat Chrysler would increase technological supervision in the manufacturing processes to prevent such errors.
We measured participants' attitude toward the Fiat Chrysler brand using the same five-item scale used in Study 2 (α =.96). As a manipulation check, we asked participants to indicate the extent to which they thought the source of the error was human, the extent to which they thought the source of the error was an algorithm, the extent to which they thought there will be more human supervision at Fiat Chrysler in the future, and the extent to which there will be more technological supervision at Fiat Chrysler in the future on four seven-point scales (1 = "not at all," and 7 = "very much"). Participants also provided perceptions of the extent to which the news was believable (1 = "not at all," and 7 = "very much"). Results show no effect of algorithm error versus human error conditions and human supervision versus technological supervision conditions on the news' believability, F( 1, 364) =.152, p =.697. Finally, participants provided their basic demographic information.
As intended, participants in the human error (vs. algorithm error) condition indicated that the source of the error was more human: MHE = 5.12, SD = 1.54 versus MAE = 4.54, SD = 1.75, t(366) = −3.41, p =.001. Participants in the algorithm error (vs. human error) condition indicated that the source of the error was more algorithm-like: MHE = 3.93, SD = 1.64 versus MAE = 4.92, SD = 1.57, t(366) = 5.90, p <.001. Participants in the human supervision (vs. technological supervision) condition indicated, going forward, there will be more human supervision at Fiat Chrysler: MHS = 5.41, SD = 1.44 versus MTS = 5.06, SD = 1.47, t(366) = 2.28, p =.023. Participants in the technological supervision (vs. human supervision) condition indicated, going forward, there will be more technological supervision at Fiat Chrysler: MHS = 4.95, SD = 1.67 versus MTS = 5.29, SD = 1.40, t(366) = −2.09, p =.037.
An ANOVA on brand attitude reveals the predicted interaction effect of algorithm error (vs. human error) and human supervision (vs. technological supervision) conditions: F( 1, 364) = 9.25, p =.003. Participants' responses to a brand following a brand harm crisis caused by an algorithm error are more negative when there is more human supervision (vs. technological supervision): MHS = 4.13, SD = 1.56 versus MTS = 4.71, SD = 1.69, F( 1, 364) = 5.44, p =.020. Participants' responses to a brand following a brand harm crisis caused by a human error are marginally more negative when there is more technological supervision (vs. human supervision): MHS = 4.63, SD = 1.69 versus MTS = 4.15, SD = 1.65, F( 1, 364) = 3.87, p =.050.
Participants' responses to a brand following a brand harm crisis caused by an algorithm error (vs. human error) are less negative when there is more technological supervision: MAE = 4.71, SD = 1.69 versus MHE = 4.15, SD = 1.65, F( 1, 364) = 5.27, p =.022. Participants' responses to a brand following a brand harm crisis caused by an algorithm error (vs. human error) are more negative when there is more human supervision: MAE = 4.13, SD = 1.56 versus MHE = 4.63, SD = 1.69, F( 1, 364) = 4.02, p =.046.
Study M1's findings indicate that consumers' responses to a brand following a brand harm crisis caused by an algorithm error are more (less) negative when there is human (technological) supervision of the algorithm following the harm crisis. The practical implication of these findings is that marketers should not (should) publicize human (technological) supervision of algorithms in communications with customers to ensure the best possible responses from consumers when the brand introduces such supervision following a brand harm crisis caused by an algorithm error.
"AI algorithms may be flawed...These deficiencies could undermine the decisions, predictions, or analysis AI applications produce, subjecting us to competitive harm, legal liability, and brand or reputational harm."
[37].
The use of algorithmic marketing across many applications is growing dramatically in many sectors. Moreover, there is growing evidence of the occurrence of algorithm errors that cause brand harm crises. However, there are few insights in the marketing literature on consumers' responses to brands following brand harm crises caused by algorithm errors.
Addressing this research gap, we develop and find support for a theory of consumers' responses to a brand following a brand harm crisis caused by an algorithm error. The findings from eight experimental studies that support our hypotheses are robust across multiple contexts (e.g., products, financial services, and online services), different samples (e.g., students and adults), and different responses, including attitudinal, behavioral, and consequential actions (in two incentive-compatible experimental designs). We conclude with a discussion of the findings' theoretical contributions, managerial implications, and limitations and opportunities for further research.
Distinct from past research on consumers attributing product failures to managerial (i.e., human) errors, we consider brand harm crises caused by inanimate entities: algorithms that are software programs. Consumers perceive that inanimate algorithms have lower agency over the error and therefore lower responsibility for the harm caused by the error.
Applying the theory of mind perception ([20]) to algorithms committing errors that cause brand harm crises, we find that consumers have lower mind perception of agency of the algorithm for the error and assign lower responsibility to the algorithm for the harm caused (H2), resulting in less negative responses to the brand (H1). Furthermore, consumers' responses to the brand following a brand harm crisis caused by an algorithm error are more negative when ( 1) the algorithm is anthropomorphized (vs. not) (H3), ( 2) the algorithm is a machine learning algorithm (vs. not) (H4), ( 3) the algorithm error occurs during a subjective (vs. objective) task (H5), and ( 4) the algorithm error occurs during an interactive (vs. noninteractive) task (H6).
Taken together, the support for the four moderation effects (i.e., whether the algorithm is anthropomorphized, whether the algorithm involves machine learning, whether the task is subjective, and whether the task is interactive), each of which humanize the algorithm, indirectly support the serial mediation by lower mind perception of agency of the algorithm for the error and, in turn, consumers' lower perception of the algorithm's responsibility for the harm caused. Given the growing prevalence of inanimate entities (e.g., algorithms, robots, drones) in practice, this research's findings make a novel contribution to the literature on harm crises, which has not examined consumers' responses to errors caused by inanimate entities. Furthermore, extant literature has also not examined moderators of the sources of harm crises and characteristics of the task in which the error occurs. Finally, the support for partial serial mediation by agency of the algorithm and, in turn, algorithms' lower responsibility for the harm caused suggests there may be other theoretical processes that offer future research opportunities.
Extant research on algorithm usage (e.g., [10]; [35]; [41]) has focused on consumers' decisions to use (or continue to use) an algorithm. However, there may be situations in practice, such as in algorithmic marketing, in which others, not the algorithm users, decide whether to deploy the algorithm. Algorithm errors frequently occur even in these contexts, which is an issue overlooked in extant research. We address this gap and consider consumers' responses to the brand following a brand harm crisis caused by an algorithm error (vs. human error) in which brand managers (not consumers) decide to deploy the algorithm. In what we consider a novel finding, when an algorithm commits an error and causes a brand harm crisis, consumers' responses to the brand following the crisis are less negative than if the firm's managers committed the same error. That is, consumers are more forgiving of algorithm errors, which suggests individuals' receptivity to algorithms when they do not have the decision-making authority on whether to use the algorithm.
Further research on individuals' responses to algorithm errors will be useful, for example, in health care, in which there is increasing application of algorithms where users have no say in their usage. For example, in the diagnosis and treatment of health conditions for which big data are used, different types of errors may occur (e.g., omission or commission, Type I and Type II errors resulting in false positives and false negatives) that could affect consumers' responses to the brand using the algorithm and to the algorithm itself.
To the best of our knowledge, this is the first study to examine consumers' responses to algorithmic errors. We identify consumers' mind perception of agency of algorithms as a building block that is relevant in the study of algorithmic marketing. Moreover, drawing on the findings of the four moderation effects, we identify conditions related to error source and task characteristics that modify the main effect of the algorithm error on consumers' responses to the brand. In doing so, we identify building blocks for developing a comprehensive theory of algorithmic marketing. Relevant questions for further research include how consumers may respond to the brand across different algorithm errors in product development, advertising, and targeting settings. A research area with policy implications is the ethicality of algorithmic marketing (e.g., facial recognition algorithms inappropriately targeting/excluding minorities) ([47]).
Our findings offer actionable guidance to managers on the deployment of algorithms in marketing contexts. First, consumers' responses to a brand following a brand harm crisis caused by an algorithm error (vs. human error) are less negative. In addition, consumers' perceptions of the algorithm's lower agency for the error and resultant lower responsibility for the harm caused mediate their responses to a brand following a brand harm crisis caused by an algorithm error. In summary, consumers penalize brands less when an algorithm (vs. human) causes an error that leads to a brand harm crisis.
Second, the findings show that consumers' responses to a brand are more negative following a brand harm crisis caused by an algorithm error when the algorithm appears more human. Thus, a brand's risk exposure to the harm caused by an algorithm error is higher when the algorithm is anthropomorphized (vs. not), involves machine learning (vs. not), is used in a subjective (vs. objective) task, or is used in an interactive (vs. noninteractive) task. Marketers must be aware that in contexts in which an algorithm appears to be more human, it would be wise to have heightened vigilance in its deployment and monitoring and to allocate resources for managing the aftermath of any brand harm crises caused by algorithm errors.
Third, to manage the aftermath of brand harm crises caused by algorithm errors, managers can highlight the role of the algorithm and the lack of agency of the algorithm for the error, which may attenuate consumers' negative responses to the brand. However, we caution that highlighting the role of the algorithm will worsen the situation by strengthening consumers' negative responses if the error involves an anthropomorphized algorithm or a machine learning algorithm, or if the algorithm error occurs in a subjective or interactive task.
Fourth, the insights from Study M1 generate concrete guidance for effectively managing the aftermath of brand harm crises caused by algorithm errors. Marketers should not publicize human supervision of algorithms (which may actually be effective in fixing the algorithm) in communications with their customers following brand harm crises caused by algorithm errors. However, if a brand uses technological supervision of the algorithm to respond to an error, marketers should publicize this fact to leverage the benefit identified in Study M1 (i.e., that consumers are less negative when there is technological supervision of the algorithm following a brand harm crisis).
First, in this initial study on brand harm crises caused by algorithm errors, we focus on consumers' negative responses to brands following one algorithm error. We do not consider the effects of repeated algorithm errors. We also do not consider very serious harm crises with dozens of fatalities (e.g., the Lion Air plane crash in Indonesia in October 2018 and the Ethiopian Airlines plane crash in March 2019 caused by Boeing 737 Max 8 automated flight system algorithms). In such cases, we anticipate extremely negative responses in both algorithm and human error conditions, precluding lab experiments for theory testing. Furthermore, we do not consider marketing mix remedies (e.g., advertising, promotions) that may be effective in handling the aftermath of brand harm crises. Further research on brand harm crises caused by algorithm errors in which brands incorporate marketing mix remedies and the effects on brand performance using less intrusive, qualitative methods, including observational studies, would be useful. Second, we focus only on errors in algorithmic marketing. Additional research on harm crises caused by algorithmic errors in other contexts (e.g., health care, justice) in which algorithm usage is increasing and errors have substantive consequences with policy implications would be useful. Third, with respect to the various parties involved, we consider the brand as the focus of our research without consideration of whether there is a distinction between blaming the algorithm itself, the person who designed it, or the person/company that chose to implement it. Whether consumers differentiate between the brand, the designer, or the person implementing the algorithm (e.g., brand manager) is a future research opportunity.
In summary, we view this study as a useful first step in exploring algorithmic marketing by focusing on brand harm crises caused by algorithm errors that, unfortunately, are now rather common in marketing practice. We hope that this research stimulates further work on algorithmic marketing strategies and related consumer behaviors.
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921997082 - When Algorithms Fail: Consumers' Responses to Brand Harm Crises Caused by Algorithm Errors
Supplemental Material, sj-pdf-1-jmx-10.1177_0022242921997082 for When Algorithms Fail: Consumers' Responses to Brand Harm Crises Caused by Algorithm Errors by Raji Srinivasan and Gülen Sarial-Abi in Journal of Marketing
Footnotes 1 Don Lehmann
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Online supplement: https://doi.org/10.1177/0022242921997082
5 The common everyday meaning of "experience" as "practical contact with and observation of facts or events" (from Merriam-Webster) is distinct from the use of the term "experience" in the theory of mind perceptions, defined as the capacity for sensations (i.e., algorithms being capable of feeling).
6 We note that the meaning of the term "responsibility" has three commonplace meanings (from Merriam-Webster): (1) the state or fact of having a duty to deal with something or of having control over someone, (2) the state or fact of being accountable or to blame for something, and (3) the opportunity or ability to act independently and make decisions without authorization. Our use of the term "responsibility" is as per the second definition. Consistent with this interpretation, "blame" is a synonym for "responsibility" at https://www.merriam-webster.com/dictionary/responsibility. Thus, our view of responsibility for the harm is consistent with blame for the harm.
7 The Institutional Review Board of the authors' home institutions reviewed and approved the experimental design before commencing the research. Participants in all studies provided informed consent before participation. As an empirical practice, we had a rule of thumb of ensuring at least 30 participants per cell for lab studies and at least 75 participants per cell for online studies. For the lab studies using student participants, we did not know, a priori, the number of participants. We report all variables collected and all conditions in the studies and do not exclude data from the analyses unless otherwise noted for clearly identified reasons. We report the number of excluded participants and do not add data from additional participants in any study, following the analyses. We preregistered analyses (and exclusions) at https://AsPredicted.org for the prestudy as well as Studies 3, 5, and the managerial intervention study. We conducted Studies 1a−1c, 2, 4, and 6 before preregistration became our standard practice. Anonymized links to the preregistrations of studies are available on request. We conducted all analyses on SPSS Statistics 23 and 25 IBM software.
8 Following the suggestion of a reviewer, we conducted a pretest (N = 153) that rules out the alternative possibility that people attribute more agency to algorithms because when algorithms make mistakes, people may consider that "even a superior entity that has higher capacities made a mistake." In this pretest, people significantly attributed more agency to humans than they did to algorithms (Mhuman = 5.95, SD =.94, Malgorithm = 4.47, SD = 1.42, t(152) = 10.409, p <.001).
References Aggarwal Pankaj , McGill Ann L.. (2007), " Is That Car Smiling at Me? Schema Congruity as a Basis for Evaluating Anthropomorphized Products ," Journal of Consumer Research , 34 (4) , 468 – 79.
Aggarwal Pankaj , McGill Ann L.. (2012), " When Brands Seem Human, Do Humans Act Like Brands? Automatic Behavioral Priming Effects of Brand Anthropomorphism ," Journal of Consumer Research , 39 (2), 307 – 23.
Ahluwalia Rohini , Burnkrant Robert E. , Rao Unnava H.. (2000), " Consumer Response to Negative Publicity: The Moderating Role of Commitment ," Journal of Marketing Research , 37 (2), 203 – 14.
Awad Edmond , Levine Sydney , Kleiman-Weiner Max , Dsouza Sohan , Tenenbaum Joshua B. , Shariff Azim , et al. (2020), " Drivers are Blamed More Than Their Automated Cars When Both Make Mistakes ," Nature Human Behaviour , 4 (2), 34 – 143.
Badger Emily. (2019), " Who's to Blame When Algorithms Discriminate? " The New York Times (August 20) , https://www.nytimes.com/2019/08/20/upshot/housing-discrimination-algorithms-hud.html.
Castelo Noah , Bos Maarten W. , Lehmann Donald R.. (2019), " Task-Dependent Algorithm Aversion ," Journal of Marketing Research , 56 (5), 809 – 25.
Choi Sungwoo , Mattila Anna S. , Bolton Lisa E.. (2020), " To Err Is Human(-oid): How Do Consumers React to Robot Service Failure and Recovery? " Journal of Service Research (published online December 16), DOI:10.1177/1094670520978798.
Cleeren Kathleen , Dekimpe Marnik G. , Heerde Harald van. (2017), " Marketing Research on Product-Harm Crises: A Review, Managerial Implications, and an Agenda for Future Research ," Journal of the Academy of Marketing Science , 45 (5), 593 – 615.
9 Diakopoulos Nicholas. (2013), " Race Against the Algorithms ," The Atlantic (October 3) https://www.theatlantic.com/technology/archive/2013/10/rage-against-the-algorithms/280255/.
Dietvorst Berkeley , Simmons Joseph P. , Massey Cade. (2015), " Algorithm Aversion: People Erroneously Avoid Algorithms After Seeing Them Err ," Journal of Experimental Psychology: General , 144 (1), 114 – 26.
Dutta Sujay , Pullig Chris. (2011), " Effectiveness of Corporate Responses to Brand Crises: The Role of Crisis Type and Response Strategies ," Journal of Business Research , 64 (12), 1281 – 87.
Epley Nicholas , Caruso Eugene , Bazerman Max H.. (2006), " When Perspective Taking Increases Taking: Reactive Egoism in Social Interaction ," Journal of Personality and Social Psychology , 91 (5), 872 – 89.
Epley Nicholas , Waytz Adam. (2009), " Mind Perception ," in The Handbook of Social Psychology , 5th ed , Fiske Susan T. , Gilbert Daniel T. , Lindzey Gardener , eds. New York, NY : Wiley , 498 – 541.
Fletcher Joseph F.. (1979), Humanhood: Essays in Biomedical Ethics. Buffalo, NY : Prometheus Books.
Folkes Valerie S.. (1984), " Consumer Reactions to Product Failure: An Attributional Approach ," Journal of Consumer Research , 10 (4), 398 – 409.
Folkes Valerie S.. (1990). " Conflict in the Marketplace: Explaining Why Products Fail ," in Attribution Theory: Applications to Achievement, Mental Health, and Interpersonal Conflict , Graham Sandra , Folkes Valerie S. , eds. Hillsdale, NJ : Lawrence Erlbaum , 143 – 60.
Folkes Valerie S. , Koletsky Susan , Graham John L.. (1987), " A Field Study of Causal Inferences and Consumer Reaction: The View from the Airport ," Journal of Consumer Research , 13 (4), 534 – 39.
Gal Michal S. , Elkin-Koren Niva. (2017), " Algorithmic Consumers ," Harvard Journal of Law & Technology , 30 (2), 309 – 53.
Gill Tripat. (2020), " Blame It on the Self-Driving Car: How Autonomous Vehicles Can Alter Consumer Morality ," Journal of Consumer Research , 47 (2), 272 – 91.
Gray Heather M. , Gray Kurt , Wegner Daniel M.. (2007), " Dimensions of Mind Perception ," Science , 315 (5812), 619.
Gray Kurt , Wegner Daniel M.. (2009), " Moral Typecasting: Divergent Perceptions of Moral Agents and Moral Patients ," Journal of Personality and Social Psychology , 96 (3), 505 – 20.
Gray Kurt , Wegner Daniel M.. (2012), " Feeling Robots and Human Zombies: Mind Perception and the Uncanny Valley ," Cognition , 125 (1), 125 – 30.
Gray Kurt , Young Liane , Waytz Adam. (2012), " Mind Perception Is the Essence of Morality ," Psychological Inquiry , 23 (2), 101 – 24.
Griffith Eric. (2017), " 10 Embarrassing Algorithm Fails ," PCMag (September 23) , https://www.pcmag.com/feature/356387/10-embarrassing-algorithm-fails.
Hayes Andrew F. , Preacher Kristopher J.. (2014), " Statistical Mediation Analysis with a Multicategorical Independent Variable ," British Journal of Mathematical and Statistical Psychology , 67 (3), 451 – 70.
Heller Martin. (2019), " Machine Learning Algorithms Explained ," InfoWorld (May 9) , https://www.infoworld.com/article/3394399/machine-learning-algorithms-explained.html.
Inbar Yoel , Cone Jeremy , Gilovich Thomas. (2010), " People's Intuitions About Intuitive Insight and Intuitive Choice ," Journal of Personality and Social Psychology , 99 (2), 232 – 47.
Kim Hyeongmin Christian , Kramer Thomas. (2015), " Do Materialists Prefer the 'Brand-as-Servant?' The Interactive Effect of Anthropomorphized Brand Roles and Materialism on Consumer Responses ," Journal of Consumer Research , 42 (2), 284 – 99.
Kim Sara , McGill Ann. (2011), " Gaming with Mr. Slot or Gaming the Slot Machine? Power, Anthropomorphism, and Risk Perception ," Journal of Consumer Research , 38 (1), 94 – 107.
Kleinberg Jon , Lakkaraju Himabindu , Leskovec Jure , Ludwig Jens , Mullainathan Sendhil. (2018), " Human Decisions and Machine Predictions ," The Quarterly Journal of Economics , 133 (1), 237 – 93.
Lafrance Adrienne. (2014), " Why People Name Their Machines ," The Atlantic (June 23) , https://www.theatlantic.com/technology/archive/2014/06/why-people-give-human-names-to-machines/373219/.
Landwehr Jan R. , McGill Ann , Herrmann Andreas. (2011), " It's Got the Look: The Effect of Friendly and Aggressive 'Facial' Expressions on Product Liking and Sales ," Journal of Marketing , 75 (3), 132 – 46.
Lee Nicol Turner , Resnick Paul , Barton Genie. (2019), " Algorithmic Bias Detection and Mitigation: Best Practices and Policies to Reduce Consumer Harms ," Brookings (May 22) , https://www.brookings.edu/research/algorithmic-bias-detection-and-mitigation-best-practices-and-policies-to-reduce-consumer-harms/.
Lei Jing , Dawar Niraj , Gürhan-Canli Zeynep. (2012), " Base-Rate Information in Consumer Attributions of Product-Harm Crises ," Journal of Marketing Research , 49 (3), 336 – 48.
Logg Jennifer , Minson Julia , Moore Don A. , (2019), " Algorithm Appreciation: People Prefer Algorithmic to Human Judgment ," Organizational Behavior and Human Decision Processes , 151 , 90 – 103.
McCullom Rod. (2017), " Facial Recognition Technology Is Both Biased and Understudied," Undark (May 17) , https://undark.org/article/facial-recognition-technology-biased-understudied/.
Microsoft Annual Report (2018), United States Securities and Exchange Commission Form 10-K (accessed June 1, 2021) , https://www.sec.gov/Archives/edgar/data/789019/000156459018019062/msft-10k%5f20180630.htm.
Moon Youngme. (2000), " Intimate Exchanges: Using Computers to Elicit Self-Disclosure from Consumers ," Journal of Consumer Research , 26 (4), 323 – 39.
Moon Youngme. (2003), " Don't Blame the Computer: When Self-Disclosure Moderates the Self-Serving Bias ," Journal of Consumer Psychology , 13 (1/2), 125 – 37.
Nass Clifford , Moon Youngme. (2000), " Machines and Mindlessness: Social Responses to Computers ," Journal of Social Issues , 56 (1), 81 – 103.
Prahl Andrew , Swol Lyn van. (2017), " Understanding Algorithm Aversion: When Is Advice from Automation Discounted? " Journal of Forecasting , 36 (6), 691 – 702.
Pullig Chris , Netemeyer Richard G. , Biswas Abhijit. (2006), " Attitude Basis, Certainty, and Challenge Alignment: A Case of Negative Brand Publicity ," Journal of the Academy of Marketing Science , 34 (4), 528 – 42.
Puzakova Marina , Kwak Hyokjin , Rocereto Joseph F.. (2013), " When Humanizing Brands Goes Wrong: The Detrimental Effect of Brand Anthropomorphization Amid Product Wrongdoings ," Journal of Marketing , 77 (3), 81 – 100.
Rafaeli Sheizaf. (1988), " Interactivity: From New Media to Communication ," in SAGE Annual Review of Communication Research: Advancing Communication Science , Vol. 16 , Hawkins Robert P. , Wiemann John M. , Pingree Suzanne , eds. Beverly Hills, CA : SAGE Publications , 110 – 34.
Roach John. (2018), " Microsoft Improves Facial Recognition Technology to Perform Well Across All Skin Tones, Genders ," Microsoft: The AI Blog (June 26) , https://blogs.microsoft.com/ai/gender-skin-tone-facial-recognition-improvement/.
Sandberg Sheryl. (2017), Facebook Post (September 20) , https://www.facebook.com/sheryl/posts/10159255449515177.
Spirina Katrine. (2009), " Ethics of Facial Recognition: How to Make Business Uses Fair and Transparent ," Towards Data Science (April 2019) , https://towardsdatascience.com/ethics-of-facial-recognition-how-to-make-business-uses-fair-and-transparent-98e3878db08d.
Sundar S. Shyam. (2009), " Media Effects 2.0: Social and Psychological Effects of Communication Technologies ," In The SAGE Handbook of Media Processes and Effects , Nabi Robin L. , Oliver Mary Beth , eds. Thousand Oaks, CA : SAGE Publications , 545 – 60.
Sundar S. Shyam , Bellur Saraswathi , Oh Jeeyun , Jia Haiyan , Kim Hyang Sook. (2016), " Theoretical Importance of Contingency in Human–Computer Interaction: Effects of Message Interactivity on User Engagement ," Communication Research , 43 (5), 595 – 625.
Sundar S. Shyam , Go Eun , Kim Hyang-Sook , Zhang Bo. (2015), " Communicating Art, Virtually! Psychological Effects of Technological Affordances in a Virtual Museum ," International Journal of Human–Computer Interaction , 31 (6), 385 – 401.
Swaminathan Vanitha , Page Karen L. , Gürhan-Canli Zeynep. (2007), " 'My' Brand or 'Our' Brand: The Effects of Brand Relationship Dimensions and Self-Construal on Brand Evaluations ," Journal of Consumer Research , 34 (2), 248 – 59.
Sweeney Latanya. (2013), " Discrimination in Online Ad Delivery ," Queue , 56 (5).
Vigdor Neil. (2019), " Apple Card Investigated After Gender Discrimination Complaints ," The New York Times (November 10) , https://www.nytimes.com/2019/11/10/business/Apple-credit-card-investigation.html.
Vincent James. (2019), " Google and Microsoft Warn Investors That Bad AI Could Harm Their Brand ," The Verge (February 11) , https://www.theverge.com/2019/2/11/18220050/google-microsoft-ai-brand-damage-investors-10-k-filing.
Vizard Sarah. (2017), " The Brand Safety Fallout: Three in Four Marketers Say Brand Reputation Has Taken a Hit ," Marketing Week (September 26) , https://www.marketingweek.com/2017/09/26/brand-safety-fallout/.
Waytz Adam , Cacioppo John , Epley Nicholas. (2010), " Who Sees Human? The Stability and Importance of Individual Differences in Anthropomorphism ," Perspectives on Psychological Science , 5 (3), 219 – 32.
Waytz Adam , Heafner Joy , Epley Nicholas. (2014), " The Mind in the Machine: Anthropomorphism Increases Trust in an Autonomous Vehicle ," Journal of Experimental Social Psychology , 52 , 113 – 17.
~~~~~~~~
By Raji Srinivasan and Gülen Sarial-Abi
Reported by Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 137- When the Honeymoon Is Over: A Theory of Relationship Liabilities and Evolutionary Processes. By: Chmielewski-Raimondo, Danielle A.; Shamsollahi, Ali; Bell, Simon J.; Heide, Jan B. Journal of Marketing. Feb2022, p1. DOI: 10.1177/00222429211062247.
Ahead of Print- Database:
- Business Source Complete
Record: 138- When to Use Markets, Lines, and Lotteries: How Beliefs About Preferences Shape Beliefs About Allocation. By: Shaddy, Franklin; Shah, Anuj K. Journal of Marketing. May2022, Vol. 86 Issue 3, p140-156. 17p. 1 Color Photograph, 1 Diagram, 6 Graphs. DOI: 10.1177/00222429211012107.
- Database:
- Business Source Complete
When to Use Markets, Lines, and Lotteries: How Beliefs About Preferences Shape Beliefs About Allocation
When allocating scarce goods and services, firms often either prioritize those willing to spend the most resources (e.g., money, in the case of markets; time, in the case of lines) or simply ignore such differences and allocate randomly (e.g., through lotteries). When do these resource-based allocation rules seem most appropriate, and why? Here, the authors propose that people are more likely to endorse markets and lines when these systems increase the likelihood that scarce goods and services go to those who have the strongest preferences—that is, when they help sort preferences. This is most feasible when preferences are dissimilar (i.e., some consumers want something much more than others). Consequently, people are naturally attuned to preference variance: when preferences for something are similar, markets and lines seem less appropriate, because it is unlikely that the highest bidders or those who have waited the longest actually have the strongest preferences. However, when preferences are dissimilar, markets and lines seem more appropriate, because they can more easily sort preferences. Consumers thus react negatively when firms use resource-based allocation rules in situations where preferences cannot be easily sorted (e.g., when preferences are similar).
Keywords: allocation; customer segmentation; fairness; lines; lotteries; markets; queues; scarcity
When allocating scarce goods and services, there are many ways to determine who gets what ([63]). Often, however, firms either prioritize those willing to spend the most resources (e.g., money, in the case of markets; time, in the case of lines) or simply ignore such differences and allocate randomly (e.g., through lotteries). For example, Live Nation, a concert promoter, auctions tickets to the highest bidders (i.e., to those willing to spend the most money; "Ticketmaster Auctions"). In contrast, its chief rival, AEG, administers lotteries, selling face-value tickets to randomly selected fans ("Fair AXS"). Or consider the television network NBC, which allocates advance tickets to tapings of Saturday Night Live via lottery before the start of each season. After the start of the new season, however, it allocates standby tickets via lines (i.e., to those willing to spend the most time).
Given the considerable differences between these systems, when do markets, lines, and lotteries seem most appropriate, and why?
The economics of market design (e.g., auction theory [[55]], matching theory [[64]]) offers a rich toolkit for determining which allocation rules are optimized for different goals, but it does not provide guidance for which the public will most readily endorse or regard as most fair. Yet, this is a critical issue for marketing theory and practice: beliefs about fairness not only pose a fundamental psychological question for researchers but also place significant constraints on firms ([42]; [62]).
In this research, we suggest that people more strongly endorse markets and lines when they believe these resource-based allocation rules increase the likelihood that scarce goods and services will go to the consumers who have the strongest preferences—that is, when markets and lines can help sort preferences. Critically, people believe this is most feasible when preferences are dissimilar (i.e., some consumers want something much more than others). So, for example, it might seem more appropriate for concert promoters and television networks to use a market or line when tickets are broadly available to the general public (where preferences are dissimilar) but less appropriate when tickets are available only to a fan club (where preferences are similar).
There are many reasons why markets might seem appropriate for determining who gets what. Markets facilitate price discovery. They can help supply meet demand. They might also encourage innovation and entrepreneurship and are generally viewed as legitimate and just ([40]). In addition, the norms of exchange underlying markets in consumer contexts are a basic feature of social relationships more broadly ([26]). As a result, markets have sprung up in many unconventional settings. For example, some food banks bid on donations ([60]), some college students bid on classes ([13]), and even some prisoners of war invented currency to bid on rations ([61]).
Likewise, there are many reasons why lines, queues, or first-come, first-served policies might seem appropriate. Firms benefit when lines signal positive product or firm characteristics, particularly when demand exceeds supply ([ 2]; [ 8]), and consumers benefit from their inherent egalitarianism. They require people to spend time, a resource believed to be more equally distributed than money ([67]).
Yet there are also many compelling reasons why these allocation rules might seem inappropriate, particularly with respect to markets. For example, people believe it is taboo to exchange resources such as money for something sacred, such as human organs ([27]; [54]; [77]). Moreover, consumers are wary of the possibility that markets will generate unfair profits ([42]; [57]) or incentivize actions inconsistent with social good ([11]). Meanwhile, because waiting can be aversive, lines sometimes trigger negative reactions from customers, including frustration, anxiety, and boredom ([19]; [22]; [46]; [74]; [82]).
Prior research has therefore identified many specific instances in which people endorse or resist markets and lines, but there is not yet a systematic framework for understanding these beliefs more broadly. Indeed, even studies that directly compare these allocation rules with each other (e.g., [28]; [41]; [65]) primarily describe consumer attitudes without explaining why they hold them. Furthermore, prior work does not predict when one approach might seem more appropriate than another. Our theory aims to address this gap.
We assert that beliefs about when to use markets and lines depend on the extent to which these allocation rules can help sort preferences. This assertion is based on prior research, which shows that people care a great deal about distributive efficiency, or the allocation of goods and services to those with the strongest preferences ([47]; [48]; [49]; [81]). Moreover, recent work demonstrates that people view allocation rules as fairer when they make it possible for consumers to signal their preferences clearly ([67]). So, although there are many goals that markets and lines can potentially help achieve, people seem particularly focused on whether these allocation rules ensure that scarce goods and services go to those who want them the most. But when and how is this possible?
We suggest that the answer depends on beliefs about preference variance, which, in turn, shape attitudes about whether preference sorting is feasible. In particular, we propose that when people believe everyone has dissimilar preferences for something (e.g., some consumers want it much more than others), they anticipate that it will be easier for a market or line to sort those with stronger preferences from those with weaker preferences; conversely, when consumers believe everyone has similar preferences (e.g., all consumers want something to roughly the same degree), they anticipate that sorting them will be more difficult.
This reasoning suggests that consumers will view markets and lines as less appropriate (and less fair) when preferences are similar and more appropriate (and fairer) when they are not. This is because when preferences are similar, people will doubt that the highest bidders or those who wait the longest actually have the strongest preferences. In other words, preference sorting seems less feasible. So, it might seem fairer to simply ignore these trivial differences, which would be difficult to accurately detect anyway. Instead, it could seem more appropriate to allocate randomly (i.e., use a lottery). However, when preferences are dissimilar, it will seem more plausible that the highest bidders or those who wait the longest actually have the strongest preferences. Now, preference sorting seems more feasible, and ignoring those nontrivial differences in preferences (e.g., by using a lottery) would seem unfair, because someone with very weak preferences would have the same chance at acquiring something as someone with very strong preferences.
It is worth noting that it is mechanically the case that a market or line can more easily sort preferences when they are dissimilar. Yet it is unclear whether consumers acknowledge or appreciate this basic economic insight, much like they fail to acknowledge or appreciate others. For example, people often do not recognize the positive gains from trade (instead assuming exchanges are zero sum; [ 3]; [36]; [39]) or the incentive value of profit (instead viewing it harmful to society; [11]).
Beliefs about the appropriateness of markets and lines could be more strongly tied to any number of other factors aside from their ability to sort preferences. For example, they could depend on which allocation rule reflects the status quo ([43]), whether prices reflect quality ([16]; [51]; [75]), reference transactions ([ 1]; [30]; [42]), or religious and moral views ([27]; [54]; [77]). But if our assertion holds, then people's intuitions about preference sorting may represent a key way in which lay economic beliefs align with textbook economic principles.
First, we propose that the distribution of preferences will influence endorsement of markets, lines, and lotteries—as well as perceptions of fairness (because we assume that people endorse allocation rules they regard as fair).
- H1: Consumers are more likely to endorse and regard as fair resource-based allocation rules (e.g., markets and lines) when they believe preferences are dissimilar.
Second, intuitions about preference sorting will play an explanatory role.
- H2: The belief that resource-based allocation rules (e.g., markets and lines) help sort preferences mediates the effect of preference variance on endorsement of resource-based allocation rules.
Several theoretical and managerial implications follow from these predictions (Figure 1). First, implicit in H1 and H2 is the assumption that willingness to spend resources and preferences are correlated (if sometimes imperfectly; e.g., [69]; [72]; [80]; [82]). Therefore, factors that undermine this correlation should attenuate the effect. One such variable is inequality salience. For example, people find that it is easier to infer preferences from the amount of time someone is willing to spend to acquire something than from the amount of money they are willing to spend. This is, in part, because time is believed to be more equally distributed than money ([67]). So, if inequality in the distribution of a resource were salient, it might reduce the perceived ratio of signal (e.g., preferences) to noise (e.g., spending uncorrelated with preferences). This, in turn, would render preference sorting less feasible—even if preferences were dissimilar. As such, moderation by inequality salience would corroborate our proposed preference sorting mechanism.
- H3: Inequality in the distribution of a resource, when salient, moderates the effect, attenuating endorsement of resource-based allocation rules.
Graph: Figure 1. Conceptual framework.
Second, there may be certain goods or services that people simply think should never be allocated on the basis of willingness to spend resources. For example, people treat wants (learned desires) differently than needs (basic requirements; [10]; [21]; [44]; [52]), which are protected by sacred values ([ 4]; [76]) and governed by moral reasoning ([ 6]; [38]; [73]). As a result, people are often uncomfortable with using markets to allocate needs ([ 5]; [53]; [66]), especially when the neediest have the fewest resources. This suggests that even if preferences for something construed as a need were dissimilar—and furthermore even if those preferences could be sorted by a market or line—people would nevertheless prefer a different basis for allocation (likely one sensitive to differences in need, rather than want). So, for needs, preference sorting should no longer matter.
- H4: The type of good or service, when perceived as a need, moderates the effect, attenuating endorsement of resource-based allocation rules.
Finally, when firms misapply these allocation rules (i.e., choose the option regarded as less appropriate), the resulting perceptions of unfairness will yield negative downstream consequences. This is consistent with work showing that consumers are less likely to patronize businesses believed to have engaged in unfair practices ([14]; [33]). Perceptions of unfairness reduce willingness to pay (WTP; [12]), trigger complaints ([37]), decrease satisfaction ([34]; [58]), and can even arouse a desire for vengeance ([ 7]).
- H5: Misapplication of these allocation rules (e.g., use of a resource-based allocation system when preferences are similar) reduces purchase intentions.
We conducted a total of 13 studies (N = 5,159; Table 1) to explore this account, and we report all variables tested. For studies that included instructional manipulation checks ([59]), we excluded failures prior to analysis. Data, stimuli, and code are publicly available.[ 6]
Graph
Table 1. Overview of Studies.
| Study | N | Hyp. | Contribution | Main Finding | Endorsement of Markets and Lines | Sig. |
|---|
| High Pref. Variance | Low Pref. Variance |
|---|
| Pilot A | 200 | H1 | Establishes correlation | Beliefs about the distribution of preferences for 25 real-world items were correlated with endorsement of a market for allocating them. | — | — | *** |
| Pilot B | 199 | H1 | Establishes correlation | Beliefs about the distribution of preferences for 25 real-world items were correlated with endorsement of a line for allocating them. | — | — | *** |
| 1a | 525 | H1 | Establishes causal effect | When preferences were dissimilar, participants endorsed a market. | 47% | 31% | *** |
| 1b | 602 | H1 | Establishes causal effect | When preferences were dissimilar, participants endorsed a line. | 59% | 35% | *** |
| 2a | 405 | H1 | Reveals that consumers try to infer preference variance (without prompting) | Participants more strongly endorsed a market for allocating concert tickets to the general public (dissimilar preferences) than a fan club (similar preferences). | 3.66 | 2.43 | *** |
| 2b | 222 | H1 | Reveals that consumers try to infer preference variance (without prompting) | Participants more strongly endorsed a line for allocating concert tickets to the general public (dissimilar preferences) than a fan club (similar preferences). | 4.80 | 3.89 | *** |
| 3 | 366 | H2 | Tests mediation by preference sorting; examines fairness | Participants more strongly endorsed a market or line for allocating beer when preferences were dissimilar, due to preference sorting. | 3.51 | 2.07 | *** |
| 4 | 202 | H2 | Tests mediation by preference sorting; presents a consequential choice | Participants cast consequential votes for allocating a prize to the highest bidder (e.g., use a market) when preferences for it were dissimilar. | 53% | 37% | * |
| 5 | 566 | H3 | Shows that inequality salience breaks the link between preference variance and preference sorting | When preferences for an electric truck were dissimilar, participants endorsed a market for allocating it, but not when inequality was salient. | 3.43 | 2.49 | *** |
| 6 | 376 | H4 | Shows that product type breaks the link between preference sorting and endorsement of resource-based allocation rules | When families differed according to how much they wanted rental cabins, participants endorsed market for allocating them, but not when families differed according to how much they needed rental cabins. | 4.01 | 3.05 | ** |
| 7 | 508 | H5 | Documents implications for purchase intentions | Misapplication of these allocation rules (e.g., using a market when preferences are similar) reduced purchase intentions. | 3.64 | 3.05 | * |
| Supp. 1a | 493 | H1 | Extends causal effect | When WTP for basketball tickets varied, participants endorsed a market. | 37% | 22% | *** |
| Supp. 1b | 495 | H1 | Extends causal effect | When willingness to wait for basketball tickets varied, participants endorsed a line. | 88% | 78% | ** |
1 *p < .05.
- 2 **p < .01.
- 3 ***p < .001.
Specifically, in Pilots A and B, we examine the relationship between beliefs about preference variance and endorsement of markets, lines, and lotteries for 25 real-world products and services. We then manipulate preference variance directly (Studies 1a and 1b) and indirectly (i.e., leaving participants to infer it; Studies 2a and 2b). Next, to probe our proposed mechanism, we explore whether intuitions about preference sorting mediate the effect (Studies 3 and 4) and test two theoretically derived moderators: inequality salience (Study 5) and product type (Study 6). Finally, to highlight managerial implications, we examine whether misapplication of these allocation rules reduces purchase intentions (Study 7).[ 7]
We first tested whether beliefs about preference variance predict endorsement of markets (Pilot A) and lines (Pilot B) for allocating 25 real-world goods and services.
For Pilot A, we recruited 200 Amazon Mechanical Turk (MTurk) participants (Mage = 32.81 years; 84 women, 116 men); for Pilot B, we recruited 199 MTurk participants (Mage = 34.96 years; 82 women, 117 men). Both pilots employed a within-subject design, in which participants evaluated 25 items along two dimensions, in two counterbalanced blocks: preference variance and endorsement of markets (vs. lotteries; Pilot A) or endorsement of lines (vs. lotteries; Pilot B).
We measured preference variance by asking participants "whether people differ in how much they want or need 25 different products and services." Specifically, for each item (presented in random order), we asked, "For [item], what do you think is generally the case?" (1 = "Some people want/need to purchase [item], while some people do not want/need to purchase [item]," and 7 = "Everyone wants/needs to purchase [item].").
We measured endorsement of markets, lines, and lotteries by asking participants "how 25 different products and services should be allocated." Specifically, for each item (presented in random order), we asked, "Imagine that at the current price there are not enough available [item] for everyone who wants or needs them. How should the [item] be allocated?" One option was a lottery: "Use a lottery (i.e., select people randomly) to decide who gets to purchase the [item]. The people selected can get [item] at the current price. The people not selected will not be able to get [item]." In Pilot A, the alternative was a market: "Sell the [item] to the people who will pay the most. The people willing to pay the most will get [item] at the price they offer. The people willing to pay the least will not be able to get [item]." In Pilot B, the alternative was a line: "Use a first-come, first-served policy to decide who gets to purchase the [item]. The people who are the first to request (or have spent the most time waiting) will be able to get [item]. The people who are the last to request (or have spent the least time waiting) will not be able to get [item]."
We reverse-coded preference variance ratings for ease of explication (so higher numbers correspond to greater preference variance). We then calculated the correlation between preference variance and endorsement of a market (market = 1, lottery = 0) or a line (line = 1, lottery = 0) across all items (i.e., using 25 pairs of observations). We observed a positive relationship in both pretests (Pilot A: r = .86, p < .001; Pilot B: r = .77, p < .001; Figure 2).
Graph: Figure 2. Pilots A and B: Perceived preference variance for a product or service correlates with endorsement of markets and lines for allocating that product or service.
We further analyzed this relationship at the participant level. We fit a random-effects logistic regression (to account for repeated measurement) with preference variance as the independent variable and allocation decision as the dependent variable. We observed a positive relationship between preference variance and both endorsement of a market (z = 21.61, p < .001) and endorsement of a line (z = 11.06, p < .001).
These initial findings characterize a strong, positive relationship between beliefs about preference variance and endorsement of resource-based allocation rules. However, this could simply be a feature of the particular set of products and services that we tested. And, of course, these pilots are correlational. Do beliefs about preference variance actually have a causal effect?
Study 1 tests whether beliefs about preference variance increases endorsement of both markets (Study 1a) and lines (Study 1b) (H1).
For Study 1a, we recruited 525 MTurk participants (Mage = 35.77 years; 313 women, 212 men); for Study 1b, we recruited 602 MTurk participants (Mage = 37.16 years; 286 women, 316 men). Both studies employed a single-factor (condition: variance vs. no variance—high vs. no variance—low), between-subjects design. Participants were randomly assigned to a condition and one of two scenarios (product vs. ticket). In the product scenario, participants read that "a retailer has a limited supply of a very popular product, and there is just one item left." In the ticket scenario, participants read that "a venue has a limited supply of tickets for a very popular upcoming event, and there is just one ticket left."
In the variance condition, we told participants that "three people all want the [product/ticket] to varying degrees" and that Persons A, B, and C, respectively, were "extremely," "moderately," or "only a little [interested in the product/excited about the event]." In the no variance—high condition, we told participants "three people all want the [product/ticket] to the same extent" and that Persons A, B, and C were all "extremely [interested in the product/excited about the event]." The no variance—low condition was identical to the no variance—high condition, but Persons A, B, and C were all "only a little [interested in the product/excited about the event]." We included the no variance—low condition to account for the possibility that consumers are simply uncomfortable using markets for allocation when demand is uniformly high ([11]; [42]).
We then asked participants to choose between a resource-based allocation rule versus random allocation (counterbalanced). In Study 1a, participants selected either "choose someone randomly" (lottery) or "choose the person who is willing to pay the most money" (market). In Study 1b, participants selected either "choose someone randomly" (lottery) or "choose the person who is willing to wait the longest in line" (line).
For these analyses, we collapsed across the product and ticket scenarios (and note that the effects did not vary by scenario). In Study 1a, participants were more likely to endorse a market, relative to a lottery, in the variance condition (47%, 95% confidence interval [CI]: [40%, 55%]; Figure 3) than in both the no variance—low condition (32%, 95% CI: [26%, 39%]; z = 2.93, p = .003, Φc = .16) and the no variance—high condition (30%, 95% CI: [23%, 37%]; z = 3.38, p = .001, Φc = .18). In Study 1b, participants were more likely to endorse a line, relative to a lottery, in the variance condition (59%, 95% CI: [52%, 65%]) than in both the no variance—low condition (36%, 95% CI: [30%, 43%]; z = 4.41, p < .001, Φc = .22) and the no variance—high condition (33%, 95% CI: [27%, 40%]; z = 5.00, p < .001, Φc = .25).
Graph: Figure 3. Studies 1a and 1b: Preference variance increases endorsement of markets and lines.
These results illustrate that when some consumers have much stronger preferences than others, markets and lines seem more appropriate. We also find a similar effect when participants view only proxies for preferences (e.g., WTP, wait times) that vary a lot or a little (see Supplemental Studies 1a and 1b in the Web Appendix). However, in these initial studies, we explicitly gave participants this information. Do people naturally attend to preference variance in the absence of such prompting?
Because different consumers maintain different preferences ([32]; [70]), what "works" for one group might not for another. We predicted that endorsement of resource-based allocation rules would depend on whether participants thought about a group that they inferred had similar or dissimilar preferences (H1).
For Study 2a, we recruited 405 MTurk participants (Mage = 37.46 years; 204 women, 201 men); for Study 2b, we recruited 222 participants from the behavioral laboratory at a West Coast business school (Mage = 22.05 years; 169 women, 53 men). Both studies employed a single-factor (condition: general public vs. fan club), between-subjects design. We described a scenario in which the band Radiohead was playing "a one-night-only show in Los Angeles" and made tickets available either to "the general public" or to "members of its Los Angeles fan club." We expected that participants would infer lower preference variance within the fan club than among the general public.
In Study 2a, we asked whether the band should allocate tickets, at face value, via "a lottery" or "sell the tickets," at their stated price, to those "willing to pay the most" (1 = "Definitely use a lottery," and 7 = "Definitely sell the tickets to the people willing to pay the most"). In Study 2b, we asked whether the band should allocate the tickets, at face value, via "a lottery" or "sell the tickets," at face value, to those "willing to wait the longest in line" (1 = "Definitely use a lottery," and 7 = "Definitely sell the tickets to the people willing to wait the longest").
Finally, as a manipulation check, we measured inferences about preference variance: "Among members of the [general public (i.e., everyone in the city of Los Angeles)/members of the Los Angeles fan club (i.e., die-hard fans)], what do you think is generally the case?" (1 = "Everyone is interested in the tickets to a similar degree," and 7 = "Some are not interested in the tickets at all, some are moderately interested in the tickets, and some are extremely interested in the tickets").
Confirming the effect of the manipulation, in Study 2a, participants inferred greater preference variance in the general public condition (M = 5.97, 95% CI: [5.76, 6.18]) than in the fan club condition (M = 3.58, 95% CI: [3.28, 3.87]; t(403) = 13.00, p < .001, d = 1.29). In Study 2b, participants inferred greater preference variance in the general public condition (M = 6.06, 95% CI: [5.84, 6.29]) than in the fan club condition (M = 4.50, 95% CI: [4.13, 4.86]; t(220) = 7.24, p < .001, d = .97).
Moreover, in Study 2a, participants were more likely to endorse a market, relative to a lottery, in the general public condition (M = 3.66, 95% CI: [3.34, 3.97]) than in the fan club condition (M = 2.43, 95% CI: [2.16, 2.70]; t(403) = 5.88, p < .001, d = .58). In Study 2b, participants were more likely to endorse a line, relative to a lottery, in the general public condition (M = 4.80, 95% CI: [4.46, 5.14]) than in the fan club condition (M = 3.89, 95% CI: [3.51, 4.27]; t(220) = 3.55, p < .001, d = .48).
These additional analyses corroborate our claim that the appropriateness of markets, lines, and lotteries depends not only on what is being allocated, but to whom. What explains this pattern, though? We propose that consumers endorse markets and lines when they believe these resource-based allocation rules increase the likelihood that those who want or need something the most will get it—that is, when they help sort preferences. In addition, thus far we have asked participants what should be done, rather than what would be the fairest thing to do. While our account implies beliefs about the latter shape beliefs for the former, we have yet to test this assumption empirically.
Study 3 offers initial process evidence for our account by testing whether intuitions about preference sorting explain why preference variance increases both endorsement and the perceived fairness of markets and lines (H1). We predicted that these intuitions would play a mediating role (H2).
We recruited 366 MTurk participants (Mage = 43.80 years; 186 women, 180 men). Study 3 employed a 2 (condition: variance vs. no variance) × 2 (resource: money vs. time), between-subjects design. All participants read, "A local craft brewery has just released a new, limited-edition beer. This new beer is an India pale ale (IPA), and there are only 10 available cases. The brewery announced the release in a Facebook post to its 100 followers."
In the variance condition, participants read, "All 100 followers would be willing to purchase a case, but some of these followers are more excited than others (i.e., some love IPAs, while others only somewhat like IPAs)." In the no-variance condition, participants read, "Because the company is known for its IPAs, all 100 followers are extremely excited and would be willing to purchase a case (i.e., they all love IPAs)."
We then explained, "One option is to enter all 100 followers into a lottery. The 10 cases would be sold to 10 randomly selected people (at the standard price)." In the money condition, we said, "Another option is to offer the available cases to those who are willing to pay the most. The 10 cases would be sold to the 10 people willing to pay the most (at their stated price)." In the time condition, we said, "Another option is to offer the available cases on a first-come, first-served basis. The 10 cases would be sold to the 10 people willing to wait in line the longest."
We asked (counterbalanced), "What should the brewery do?" and "What would be the fairest thing for the brewery to do?" (money condition: 1 = "Definitely use a lottery," and 7 = "Definitely sell the cases to those who are willing to pay the most"; time condition: 1 = "Definitely use a lottery," and 7 = "Definitely sell the cases to those who are willing to wait in line the longest").
Finally, we measured intuitions about preference sorting: "If the brewery sold the available cases to [those who are willing to pay the most/wait in line the longest], how likely is it that the available cases would end up going to the people who want them the most?" (1 = "Not at all likely," and 7 = "Extremely likely").
Beliefs about what the brewery "should" do and what would be the "fairest thing for the brewery to do" did not meaningfully differ (α = .89), so we formed a composite by taking the average. An analysis of variance (ANOVA) of this composite on condition, resource, and their interaction revealed only a main effect of preference variance (F( 1, 362) = 49.43, p < .001, d = .68), such that participants were more likely to endorse a market or line, relative to a lottery, in the variance condition (M = 3.51, 95% CI: [3.18, 3.84]) than in the no-variance condition (M = 2.07, 95% CI: [1.82, 2.32]). The simple effect of condition was significant for each resource (money condition: F( 1, 362) = 15.83, p < .001; time condition: F( 1, 362) = 35.70, p < .001).
We next examined beliefs about preference sorting. An ANOVA of these beliefs on condition, resource, and their interaction revealed a main effect of preference variance (F( 1, 362) = 30.94, p < .001, d = .52), such that participants believed that a market or line would do a better job sorting preferences in the variance condition (M = 5.66, 95% CI: [5.45, 5.87]) than in the no-variance condition (M = 4.82, 95% CI: [4.57, 5.06]). The simple effect of condition was significant for each resource (money condition: F( 1, 362) = 16.80, p < .001; time condition: F( 1, 362) = 14.17, p < .001). We also observed a main effect of resource (F( 1, 362) = 9.65, p = .002, d = .25), such that participants believed that preference sorting was more likely in the time condition (M = 5.44, 95% CI: [5.22, 5.65]) than in the money condition (M = 5.04, 95% CI: [4.79, 5.29]).
Finally, we tested for mediation. Indeed, beliefs about preference sorting mediated the effect of condition on endorsement of a market or line, relative to a lottery (based on 10,000 bootstrapped resamples: indirect effect = .32, SE = .07, 95% CI: [.193,.486]).
This result supports the notion that consumers believe preference sorting is a basic function of markets and lines, and this is why they seem both more appropriate and fairer when preferences are dissimilar. All of the studies thus far have been hypothetical, however. Next, we test whether these findings hold when participants face real consequences for their allocation decisions. We note that although our framework applies to both markets and lines, in the remaining studies we focus specifically on attitudes toward markets and market pricing (predicting conceptually similar results for first-come, first-served policies).
Study 4 tests whether preference variance affects real decisions for how something should be allocated. We predicted that when participants believed that preferences for a prize varied, they would be more likely to cast votes for a market (vs. a lottery). We again predicted intuitions about preference sorting would play a mediating role (H2).
We recruited 202 MTurk participants (Mage = 35.82 years; 72 women, 130 men). Study 4 employed a single-factor (condition: high variance vs. low variance), between-subjects design. We first told all participants that they would be participating in a trivia game and that their goal would be to identify characters from a popular television show (The Office). We also told participants they would have the chance to win a prize, depending on their performance.
After reviewing these instructions and launching the trivia game, participants had 45 seconds to evaluate ten photos (Figure 4). They were asked to indicate which photo depicted each of ten characters that were listed below the table in random order.
Graph: Figure 4. Study 4: Trivia Game Stimuli.
After completing the trivia game participants read, "We have one ( 1) The Office-theme card game (see below) to offer as a gift to participants in this study." We displayed the prize and asked, "How much of your base pay ($1.00) would you be willing to exchange for this gift?" Participants responded on a sliding scale, ranging from 0 to 100 cents.
In the high-variance condition, we told participants, "All participants, regardless of their score, will be eligible to receive this gift. And we are asking all participants, regardless of their score, to vote on how this gift will be awarded." In the low-variance condition, we told participants, "Only those participants who earned a perfect score will be eligible to receive this gift. But we are asking all participants, regardless of their score, to vote on how this gift will be awarded." We then asked, "Should we choose [someone/one of these die-hard fans] randomly, or should we 'sell' it to the highest bidder (i.e., the participant who is willing to give up the most of his/her $1.00 base pay)? Note that we will actually tally these votes and use the outcome to decide how to award this gift" ("Choose randomly" or "'Sell' it to the highest bidder").
Finally, after casting a vote, participants responded to four follow-up questions. First, to test our proposed mechanism, we captured intuitions about preference sorting: "If we 'sold' it to the highest bidder (among [everyone who scored between 0%–100%/only those who scored 100%]), would that make it more likely or less likely that the person who wants this card game the most will be able to get it?" (1 = "Less likely," 4 = "Neither," and 7 = "More likely"). We also asked participants to guess how many characters they correctly identified (0–10) and to indicate whether they were familiar with (1 = "Not at all familiar," and 7 = "Very familiar") and a fan of (1 = "Definitely not," and 7 = "Definitely") the television show.
Participants were likelier to vote for a market (i.e., sell the prize to the highest bidder), relative to a lottery, in the high-variance condition (53%, 95% CI: [43%, 62%]) than in the low-variance condition (37%, 95% CI: [28%, 47%]; χ2( 1) = 5.18, p = .023, Φc = .16).[ 8] Participants also indicated that they believed a market would make it more likely that the person who wanted the card game the most would be able to get it (i.e., sort preferences) in the high-variance condition (M = 5.80, 95% CI: [5.52, 6.08]) than in the low-variance condition (M = 5.42, 95% CI: [5.12, 5.72]; t(200) = 1.86, p = .064, d = .26). Furthermore, these beliefs mediated the effect of condition on voting for a market (based on 10,000 bootstrapped resamples: indirect effect = .02, SE = .02, bias-corrected 95% CI: [.001,.066]).
It is also worth pointing out that unlike in the previous studies, participants here voted for an allocation rule to which they, themselves, would be subjected. Interestingly, objective performance and endorsement of a market were weakly but negatively correlated (r = −.13, p = .06). In other words, those with low scores—those less likely to be fans of the show and consequently those with lower WTP—nevertheless tended to believe the prize should be "sold" to the highest bidder, possibly recognizing the potential to improve distributive efficiency (even though allocation through market pricing would mean they, themselves, were unlikely to win).
Studies 3 and 4 reveal that people more strongly endorse resource-based allocation rules when preferences are dissimilar, because markets and lines are likelier to allocate scarce goods and services to those with the strongest preferences (i.e., sort preferences). Next, we turn to two theoretically derived moderators of our basic model.
Previous research has found that inequality in the distribution of a resource makes it difficult to clearly signal preferences ([67]). So, when inequality is salient, preference variance should no longer matter because there is no reliable way to sort those differences (H3).
We recruited 566 Prolific participants (Mage = 37.71 years; 279 women, 287 men). Study 5 employed a 2 (condition: variance vs. no variance) × 2 (inequality: salient vs. baseline), between-subjects design. All participants first reviewed a vignette describing the introduction of "a new, highly anticipated, all-electric pickup truck." We explained that because "the company can only produce a limited supply," potential buyers would need to submit a waitlist application that included contact information, a description of their interest, and a refundable deposit.
In the variance condition, participants read, "The people on the waitlist each have dramatically different levels of desire for the truck." In the no-variance condition, participants read, "The people on the waitlist all have exactly the same level of desire for the truck." Participants in the inequality-salient condition were told, "The people on the waitlist each earn dramatically different incomes." In the baseline condition, participants read nothing else. Finally, we asked, "How should the company allocate the available trucks?" (1 = "Choose people randomly [and sell at list price]," and 7 = "Choose the people willing to pay the most money [and sell at the offered price]").
An ANOVA of allocation rule on condition, inequality, and their interaction revealed a main effect of condition (F( 1, 562) = 7.10, p = .008), which was qualified by an interaction (F( 1, 562) = 7.93, p = .005). Decomposition revealed a simple effect of condition at baseline (F( 2, 562) = 15.46, p < .001, d = .45; Figure 5), replicating the basic effect: participants were more likely to endorse a market, relative to a lottery, in the variance condition (M = 3.43, 95% CI: [3.10, 3.75]) than in the no-variance condition (M = 2.49, 95% CI: [2.16, 2.83]). However, there was no such simple effect of condition when inequality was salient (F( 2, 562) = .01, p = .916, d = .01; Mvariance = 2.81, 95% CI: [2.47, 3.15]; Mno variance = 2.83, 95% CI: [2.49, 3.18]).
Graph: Figure 5. Study 5: Inequality salience attenuates the effect of preference variance on endorsement of markets.
Study 5 confirms that when inequality is salient, preference sorting seems less feasible—even when preferences are dissimilar—so resource-based allocation rules seem less appropriate. The next study tests whether there are certain goods or services that people simply think should never be allocated on the basis of willingness to spend resources.
People often disapprove of resource-based allocation rules for allocating needs, which can impose taboo trade-offs ([ 5]; [53]). We therefore expected that for something people need (as opposed to merely want), resource-based allocation rules seem less appropriate (H4).
We recruited 376 MTurk participants (Mage = 34.59 years; 167 women, 209 men). Study 6 employed a 2 (condition: variance vs. no variance) × 2 (type: want vs. need), between-subjects design. All participants first read: "Throughout the country, the U.S. Forest Service maintains a number of restricted-use cabins on protected land. These cabins are not typically open to the public, but are rather used for operational purposes."
In the want condition, we explained that "the agency has decided to make these cabins available for short-term rental to people who are interested in vacationing at these sites." In the need condition, we explained that because "forest fires near one residential neighborhood have significantly diminished air quality and now pose a serious safety hazard, ... the Forest Service is making some cabins available for short-term rental." Participants then read, "There is now only one cabin left and several families still [want/need] it." In the variance condition, we told participants, "These families, however, each have dramatically different levels of [need/desire] for the cabin." In the no-variance condition, we told participants, "These families, however, all have exactly the same level of [need/desire] for the cabin."
Finally, we asked (counterbalanced), "What should the Forest Service do?" and "What would be the fairest thing for the Forest Service to do?" (1 = "Choose a family randomly," and 7 = "Choose the family willing to pay the most money for it").
Beliefs about what the Forest Service "should" do and what would be the "fairest thing for the Forest Service to do" did not meaningfully differ (α = .88), so we formed a composite by taking the average. An ANOVA of this composite on condition, type, and their interaction revealed main effects of condition (F( 1, 372) = 4.09, p = .044) and type (F( 1, 372) = 4.53, p = .034), which were qualified by an interaction (F( 1, 372) = 5.40, p = .021). Decomposition revealed a simple effect of condition for wants (F( 1, 372) = 9.38, p = .002; d = .43), replicating the basic effect: participants were more likely to endorse a market, relative to a lottery, in the variance condition (M = 4.01, 95% CI: [3.56, 4.46]) than in the no-variance condition (M = 3.05, 95% CI: [2.64, 3.46]) (Figure 6). However, there was no such simple effect of condition for needs (F( 1, 372) = .05, p = .831; d = .03; Mvariance = 3.03, 95% CI: [2.61, 3.45]; Mno variance = 3.10, 95% CI: [2.62, 3.57]).
Graph: Figure 6. Study 6: Preference variance increases endorsement of resource-based allocation rules for wants, but not for needs.
Study 6 reveals that even when preferences for something construed as a need are dissimilar—and furthermore even when those preferences could be sorted by a market—people still resist resource-based allocation rules. This could be due to hesitance regarding taboo trade-offs, which possibly shift people from consequentialist moral reasoning (see the "General Discussion" section). Or it may be that in these situations people prefer a different basis for allocation (likely one sensitive to differences in need, rather than want). In our final study, we examine how consumers respond when they cannot choose the allocation rule themselves (as is typically the case), underscoring the managerial implications of our theory.
An expansive body of literature has documented the numerous negative consequences that result from perceptions of unfairness in the marketplace (e.g., [ 7]; [12]; [14]; [33]; [34]; [37]; [58]). This suggests that firms may be penalized for choosing allocation rules that our framework characterizes as inappropriate (H5).
We recruited 508 MTurk participants (Mage = 40.17 years; 272 women, 236 men). Study 7 employed a 2 (condition: variance vs. no variance; within-subjects) × 2 (system: market vs. lottery; between-subjects) mixed design. All participants first read, "The U.S. Centers for Disease Control and Prevention (CDC) recommends wearing face masks to help slow the spread of the coronavirus." While "surgical masks and cloth masks are widely available," N95 respirators "are still in short supply."[ 9] We then explained that the "largest domestic manufacturer of N95 respirators is 3M, which also makes a wide array of other products, including sticky notes, tape, bandages, air filters, water filters, sponges, and much more."
We then described two cities, one with greater preference variance than the other: "In the city of Springfield, each resident has dramatically different desire for an N95 respirator"; alternatively, "In the city of Greenville, all residents have identical desire for an N95 respirator." Preferences for N95s, therefore, varied in Springfield, but not in Greenville.
Those assigned to the market system indicated how fair it would be "if 3M used an auction to allocate its available N95s to the highest bidders (at their stated price)" in each of Springfield and Greenville (counterbalanced). Those assigned to the lottery system indicated how fair it would be "if 3M used a lottery to allocate its available N95s randomly (at list price)" in each of Springfield and Greenville (counterbalanced; for both questions, 1 = "Extremely unfair," and 7 = "Extremely fair").
On the next page, we measured purchase intentions (counterbalanced): "If 3M used [an auction/a lottery] to allocate N95s in Springfield, would that affect your willingness to purchase 3M products?" And: "If 3M used [an auction/a lottery] to allocate N95s in Greenville, would that affect your willingness to purchase 3M products?" (for both questions, 1 = "It would make me less likely to purchase other 3M products," and 7 = "It would make me more likely to purchase other 3M products").
A mixed ANOVA of fairness on system (between-subjects), variance (within-subjects), and their interaction revealed a main effect of system (F( 1, 467) = 100.30, p < .001) and a main effect of variance (F( 1, 467) = 37.68, p < .001), which were qualified by an interaction (F( 1, 467) = 141.02, p < .001). Decomposition revealed that participants believed it was fairer to use a market to allocate the available N95s in the city where preferences varied (i.e., Springfield; M = 3.62, 95% CI: [3.38, 3.86]) than in the city where preferences did not vary (i.e., Greenville; M = 3.04, 95% CI: [2.81, 3.28]; F( 1, 467) = 16.21, p < .001, d = .30) (Figure 7). By contrast, participants believed it was fairer to use a lottery to allocate the available N95s in the city where preferences did not vary (i.e., Greenville; M = 5.64, 95% CI: [5.43, 5.84]) than in the city where preferences varied (M = 3.82, 95% CI: [3.57, 4.06]; F( 1, 467) = 164.70, p < .001, d = .89).
Graph: Figure 7. Study 7: Misapplication of these allocation rules (e.g., use of a market when preferences are similar) reduces perceptions of fairness and purchase intentions.
A mixed ANOVA of purchase intentions on system (between-subjects), variance (within-subject), and their interaction revealed a main effect of system (F( 1, 467) = 86.12, p < .001) and a main effect of variance (F( 1, 467) = 3.78, p = .052), which were qualified by an interaction (F( 1, 467) = 45.82, p < .001). Decomposition revealed that participants were less likely to purchase other 3M products if the company used a market to allocate the available N95s in the city where preferences did not vary (i.e., Greenville; M = 2.87, 95% CI: [2.68, 3.05]) than in the city where preferences varied (i.e., Springfield; M = 3.16, 95% CI: [2.99, 3.34]; F( 1, 467) = 11.46, p < .001, d = .20) (Figure 7). By contrast, participants were less likely to purchase 3M products if the company used a lottery to allocate the available N95s in the city where preferences varied (i.e., Springfield; M = 3.72, 95% CI: [3.57, 3.87]) than in the city where preferences did not vary (i.e., Greenville; M = 4.23, 95% CI: [4.08, 4.39]); F( 1, 467) = 38.54, p < .001, d = .47).
This final study demonstrates that when a company fails to apply the more appropriate allocation rule (as characterized by our framework), purchase intentions suffer. However, we acknowledge the possibility that participants could have made different inferences about the two cities (given our within-subject design), though it is not clear in what direction this would have systematically affected judgments. For example, residents of a city might express uniformly high desire for N95 respirators because their public health infrastructure is poorly equipped (and thus lacks supplies) or well equipped (reflecting a citizenry that enthusiastically adopts new mitigation tactics). Nevertheless, the findings here highlight the importance of anticipating how the appropriateness of allocation rules for some products can potentially affect downstream purchase intensions for other products.
In this research, we offer a general account of when and why people favor the use of markets, lines, and lotteries. We believe that understanding these lay economic beliefs is of broad theoretical interest. Yet these intuitions also have practical consequences, as they shape perceptions of fairness in the marketplace. To that end, our account builds on prior work showing that consumers care deeply about distributive efficiency ([47]; [48]; [49]). Perhaps as a result of this, we find that people are naturally attuned to how preferences are distributed. And thus their views about when to use markets, lines, and lotteries depend on the extent to which they believe preferences vary.
Of course, preference variance is not the only factor that shapes views about how to allocate things. For example, Study 5 demonstrates that inequality reduces support for resource-based allocation rules. People are uncomfortable with inequality in general ([24]), but our results reveal that at least some of this discomfort stems from skepticism about whether resource-based allocation rules can improve distributive efficiency when spending is uncorrelated with preferences. In addition, inequality may furthermore affect perceptions of unfairness simply because people regard any form of inequality as unfair (e.g., [23]; [45]; [56]; [71]; [79]).
It is further plausible that the source of inequality could matter as well. For example, inequality arising from differences in work ethic probably attenuate the effect less than inequality arising from differences in inheritance ([17]). And resources themselves can often be exchanged for each other (e.g., paying money to jump a queue and save time), suggesting another potential moderator future research might explore.
More broadly, our findings enrich the literature exploring when people most readily adopt preference-based versus other allocation norms. For example, prior work has argued that people especially desire improvements in distributive efficiency when there is an insufficient supply of something and preferences vary ([20]; [68]; [81]). Our conceptual framework reveals that these are necessary, but not sufficient, conditions: people also need to believe that stated preferences (signaled by the resources consumers are willing to spend) are appropriate for determining who should get what and that resources spent reliably signal those preferences.
We believe that our work yields several additional theoretical insights. We identify a novel source of market aversion. For example, previous work has found that market aversion can occur when people attach moral value to things ([77]) or react negatively to profit-taking ([11]; [42]; [57]). Here, we propose that market aversion can also be traced to views about the very purpose of markets to begin with. That is, consumers seem to believe that a primary function is to help sort preferences—identifying those who most want something and allocating accordingly. And so they will exhibit market aversion when this goal is infeasible (because preferences are too similar).
This basic insight might apply to other allocation rules in nonconsumer contexts, as well. For example, a primary function of admissions committees at elite universities can be viewed as "merit sorting"—allocating limited seats in each freshman class to the most qualified applicants. However, merit sorting should be similarly infeasible when, in this case, qualifications are too similar. This has led some experts to call for lottery admissions for applicants who meet certain academic thresholds ([ 9]; [18]; [35]). To the extent that many other potential bases for allocation exist—for example, differences in need (e.g., Study 6), future potential ([78]), and emotional resonance ([31]; [50])—our framework might similarly apply.
It is also likely that there exist other moderators for the model described here. For example, one interpretation of Study 6 is that the prospect of allocating wants versus needs shifts people from consequentialist moral reasoning (wherein they think about the costs and benefits of using markets and lines) to deontological moral reasoning (wherein they adhere to simple ethical rules and heuristics; [ 6]; [38]; [73]). If true, then the numerous other factors that have been shown to shift reliance on consequentialist versus deontological moral reasoning (e.g., whether outcomes are framed as gains vs. losses or benefits vs. harms; [ 4]; [29]) might further moderate the effects we document.
More practically, highlighting preference variance might soften some resistance to market pricing. For example, during emergencies, demand for certain products or services can increase dramatically. When prices follow suit, firms are often accused of price gouging ([25]), a practice that people seem to oppose uniformly ([15]; [42]). However, our findings suggest some potential nuance: consumers might actually tolerate raising prices if they appreciate that doing so can help direct scarce goods and services to those who will make the best use of them—that is, improve distributive efficiency through preference sorting. Indeed, previous work has argued that while raising the price of hotel rooms in the path of a hurricane "does not literally increase the supply of hotel rooms, it increases the available supply" ([83], p. 363).
Implications for segmentation are worth highlighting, as well. As underscored by Studies 2a and 2b, preference variance between segments often differs, suggesting another consideration marketers should weigh with respect to their pricing tactics. To adapt a classic example: business travelers usually pay higher fares for flights than leisure travelers with the exact same itinerary. Airlines are able to price discriminate thusly because business travelers typically make purchases much later than leisure travelers (and fares tend to increase over time). However, flights are often oversold, requiring airlines to set prices for not completing a trip as planned (i.e., compensating a traveler for instead taking the next available flight). Here, preference variance among business travelers, who probably have tighter schedules, is likely lower than preference variance among leisure travelers, who are less likely to have appointments to keep. So, when deciding whom to leave behind, it could make more sense to use random allocation (e.g., a lottery) for the business segment and a resource-based allocation rule (e.g., a market based on willingness to accept) for the leisure segment.
Finally, although consumers generally view lines as a fairer alternative to markets ([28]; [41]; [65]), our theory cautions against their uncritical adoption. We find that the same conditions that give rise to market aversion also dampen support for first-come, first-served policies: if consumers believe lines cannot accurately sort preferences (because they are too similar), then they will resist using them all the same.
People often disagree about how to allocate things fairly, and it can sometimes seem like these disagreements stem from intractable differences in moral convictions or political philosophies (e.g., socialism vs. capitalism). However, our work suggests a more flexible view. People actually seem to earnestly try to discern the nature of preferences and choose an allocation rule that fits. It thus reveals an interesting way in which people apply their lay economic beliefs. Consumers desire distributive efficiency in that they believe things should go to those who want them the most, but psychology shapes views about when this goal is possible and how best to achieve it.
sj-pdf-1-jmx-10.1177_00222429211012107 - Supplemental material for When to Use Markets, Lines, and Lotteries: How Beliefs About Preferences Shape Beliefs About Allocation
Supplemental material, sj-pdf-1-jmx-10.1177_00222429211012107 for When to Use Markets, Lines, and Lotteries: How Beliefs About Preferences Shape Beliefs About Allocation by Franklin Shaddy and Anuj K. Shah in Journal of Marketing
Footnotes 1 Dilip Soman
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship, and/or publication of this article.
4 Franklin Shaddy https://orcid.org/0000-0002-1153-4839
5 Online supplement:https://doi.org/10.1177/00222429211012107
6 See https://osf.io/9eg8u/.
7 We note that in all studies, we compare endorsement of markets with lotteries and lines with lotteries, but we do not compare markets with lines. Much prior work (e.g., [28]; [41]; [65]) has already shown that that lines are generally believed to be fairer than markets (while not the focus of our hypotheses, our studies empirically confirm this). We designed our experiments in this way because we are primarily interested in understanding when people endorse resource-based allocation rules, which sort preferences. Consequently, this framework should potentially apply to any resource-based allocation rule (in addition to markets and lines; see the "General Discussion" section).
8 Because there were more votes overall for the lottery (111) than the market (91), we randomly chose a participant as the winner. We offered to either (a) purchase and send the prize to the winner or (b) issue an MTurk bonus in the amount of the retail price of the prize. The winner chose option (b).
9 Note that we chose to test N95 respirators specifically because they are extremely effective at preventing transmission (and therefore should be carefully allocated when supply is limited); however, they are not strictly a necessity (e.g., as in Study 6), due to the availability of other types of masks and preventative measures.
References Anderson Eric T. , Simester Duncan I.. (2008), " Research Note—Does Demand Fall When Customers Perceive That Prices Are Unfair? The Case of Premium Pricing for Large Sizes ," Marketing Science , 27 (3), 492 – 500.
Banerjee Abhijit V. (1992), " A Simple Model of Herd Behavior ," Quarterly Journal of Economics , 107 (3), 797 – 817.
Baron Jonathan , Kemp Simon. (2004), " Support for Trade Restrictions, Attitudes, and Understanding of Comparative Advantage ," Journal of Economic Psychology , 25 (5), 565 – 80.
Baron Jonathan , Ritov Ilana. (2009), " Protected Values and Omission Bias as Deontological Judgments ," in Psychology of Learning and Motivation , Vol 50 , Bartels D.M. , Bauman C.W. , Skitka L.J. , Medin D.L. , eds. Cambridge, MA : Elsevier Academic Press , 133 – 67.
Baron Jonathan , Spranca Mark. (1997), " Protected Values ," Organizational Behavior and Human Decision Processes , 70 (1), 1 – 16.
Bartels Daniel M. (2008), " Principled Moral Sentiment and the Flexibility of Moral Judgment and Decision Making ," Cognition , 108 (2), 381 – 417.
Bechwati Nada Nasr , Morrin Maureen. (2003), " Outraged Consumers: Getting Even at the Expense of Getting a Good Deal ," Journal of Consumer Psychology , 13 (4), 440 – 53.
Becker Gary S. (1991), " A Note on Restaurant Pricing and Other Examples of Social Influences on Price ," Journal of Political Economy , 99 (5), 1109 – 16.
Bellafante Ginia. (2020), "Should Ivy League Schools Randomly Select Students (at Least for a Little While)?" The New York Times (December 18), https://www.nytimes.com/2020/12/18/nyregion/ivy-league-admissions-lottery.html.
Berry Christopher J. (1994), The Idea of Luxury: A Conceptual and Historical Investigation. Cambridge, UK : Cambridge University Press.
Bhattacharjee Amit , Dana Jason , Baron Jonathan. (2017), " Anti-Profit Beliefs: How People Neglect the Societal Benefits of Profit ," Journal of Personality and Social Psychology , 113 (5), 671 – 96.
Bolton Lisa E. , Warlop Luk , Alba Joseph W.. (2003), " Consumer Perceptions of Price (Un)Fairness ," Journal of Consumer Research , 29 (4), 474 – 91.
Budish Eric , Cantillon Estelle. (2012), " The Multi-Unit Assignment Problem: Theory and Evidence from Course Allocation at Harvard ," American Economic Review , 102 (5), 2237 – 71.
Campbell Margaret C. (1999a), " Perceptions of Price Unfairness: Antecedents and Consequences ," Journal of Marketing Research , 36 (2), 187 – 99.
Campbell Margaret C. (1999b), " Pricing Strategy & Practice: 'Why Did You Do That?' The Important Role of Inferred Motive in Perceptions of Price Fairness ," Journal of Product & Brand Management , 8 (2), 145 – 53.
Chernev Alexander , Carpenter Gregory S.. (2001), " The Role of Market Efficiency Intuitions in Consumer Choice: A Case of Compensatory Inferences ," Journal of Marketing Research , 38 (3), 349 – 61.
Chow Rosalind M. , Galak Jeff. (2012), " The Effect of Inequality Frames on Support for Redistributive Tax Policies ," Psychological Science , 23 (12), 1467 – 69.
Conley Dalton. (2018), "Enough Fretting over College Admissions. It's Time for a Lottery," The Washington Post (August 13) , https://www.washingtonpost.com/opinions/enough-fretting-over-college-admissions-its-time-for-a-lottery/2018/08/13/f65a072c-9a74-11e8-8d5e-c6c594024954%5fstory.html.
Davis Mark M. , Vollmann Thomas E.. (1990), " A Framework for Relating Waiting Time and Customer Satisfaction in a Service Operation ," Journal of Services Marketing , 4 (1), 61 – 69.
Deutsch Morton. (1975), " Equity, Equality, and Need: What Determines Which Value Will Be Used as the Basis of Distributive Justice? " Journal of Social Issues , 31 (3), 137 – 49.
Dhar Ravi , Wertenbroch Klaus. (2000), " Consumer Choice Between Hedonic and Utilitarian Goods ," Journal of Marketing Research , 37 (1), 60 – 71.
Efrat-Treister Dorit , Daniels Michael A. , Robinson Sandra L.. (2020), " Putting Time in Perspective: How and Why Construal Level Buffers the Relationship Between Wait Time and Aggressive Tendencies ," Journal of Organizational Behavior , 41 (3), 294 – 309.
Farmer Adam , Kidwell Blair , Hardesty David M.. (2020), " Helping a Few a Lot or Many a Little: Political Ideology and Charitable Giving ," Journal of Consumer Psychology , 30 (4), 614 – 30.
Fehr Ernst , Schmidt Klaus M.. (1999), " A Theory of Fairness, Competition, and Cooperation ," Quarterly Journal of Economics , 114 (3), 817 – 68.
Ferguson Jodie L. , Ellen Pam Scholder , Piscopo Gabriela Herrera. (2011), " Suspicion and Perceptions of Price Fairness in Times of Crisis ," Journal of Business Ethics , 98 (2), 331 – 49.
Fiske Alan P. (1992), " The Four Elementary Forms of Sociality: Framework for a Unified Theory of Social Relations ," Psychological Review , 99 (4), 689 – 723.
Fiske Alan Page , Tetlock Philip E.. (1997), " Taboo Trade-Offs: Reactions to Transactions That Transgress the Spheres of Justice ," Political Psychology , 18 (2), 255 – 97.
Frey Bruno S. , Pommerehne Werner W.. (1993), " On the Fairness of Pricing—An Empirical Survey Among the General Population ," Journal of Economic Behavior & Organization , 20 (3), 295 – 307.
Gamez-Djokic Monica , Molden Daniel. (2016), " Beyond Affective Influences on Deontological Moral Judgment: The Role of Motivations for Prevention in the Moral Condemnation of Harm ," Personality and Social Psychology Bulletin , 42 (11), 1522 – 37.
Gershoff Andrew D. , Kivetz Ran , Keinan Anat. (2012), " Consumer Response to Versioning: How Brands' Production Methods Affect Perceptions of Unfairness ," Journal of Consumer Research , 39 (2), 382 – 98.
Goenka Shreyans , Van Osselaer Stijn M.J.. (2019), " Charities Can Increase the Effectiveness of Donation Appeals by Using a Morally Congruent Positive Emotion ," Journal of Consumer Research , 46 (4), 774 – 90.
Goodman Joseph K. , Broniarczyk Susan M. , Griffin Jill G. , McAlister Leigh. (2013), " Help or Hinder? When Recommendation Signage Expands Consideration Sets and Heightens Decision Difficulty ," Journal of Consumer Psychology , 23 (2), 165 – 74.
Guo Xiaomeng , Jiang Baojun. (2016), " Signaling Through Price and Quality to Consumers with Fairness Concerns ," Journal of Marketing Research , 53 (6), 988 – 1000.
Haws Kelly L. , Bearden William O.. (2006), " Dynamic Pricing and Consumer Fairness Perceptions ," Journal of Consumer Research , 33 (3), 304 – 11.
Hess Frederick. (2019), "A Modest Proposal Regarding College Admissions," Forbes (March 15) , https://www.forbes.com/sites/frederickhess/2019/03/15/a-modest-proposal-regarding-college-admissions/?sh=75534f71fe52.
Hiscox Michael J. (2006), " Through a Glass and Darkly: Attitudes Toward International Trade and the Curious Effects of Issue Framing ," International Organization , 60 (30), 755 – 80.
Huppertz John W. , Arenson Sidney J. , Evans Richard H.. (1978), " An Application of Equity Theory to Buyer-Seller Exchange Situations ," Journal of Marketing Research , 15 (2), 250 – 60.
Iliev Rumen , Sachdeva Sonya , Bartels Daniel M. , Joseph Craig , Suzuki Satoru , Medin Douglas L.. (2009), " Attending to Moral Values ," in The Psychology of Learning and Motivation , Bartels D.M. , Bauman C.W. , Skitka L.J. , Medin D.L. , eds. Burlington, VT : Academic Press , 169 – 92.
Johnson Samuel G.B. (2018), " Why Do People Believe in a Zero-Sum Economy? " Behavioral & Brain Sciences , 41 (172), 29 – 30.
Jost John T. , Blount Sally , Pfeffer Jeffrey , Hunyady György. (2003), " Fair Market Ideology: Its Cognitive-Motivational Underpinnings ," Research in Organizational Behavior , 25 , 53 – 91.
Kahneman Daniel , Knetsch Jack L. , Thaler Richard (1986a), " Fairness and the Assumptions of Economics ," Journal of Business , 59 (4), 285 – 300.
Kahneman Daniel , Knetsch Jack L. , Thaler Richard (1986b), " Fairness as a Constraint on Profit Seeking: Entitlements in the Market ," American Economic Review , 76 (4), 728 – 41.
Kimes Sheryl E. (1994), " Perceived Fairness of Yield Management: Applying Yield-Management Principles to Rate Structures Is Complicated by What Consumers Perceive as Unfair Practices ," Cornell Hotel and Restaurant Administration Quarterly , 35 (1), 22 – 29.
Kivetz Ran , Simonson Itamar. (2002), " Self-Control for the Righteous: Toward a Theory of Precommitment to Indulgence ," Journal of Consumer Research , 29 (2), 199 – 217.
Kuziemko Ilyana , Norton Michael I. , Saez Emmanuel , Stantcheva Stefanie. (2015), " How Elastic Are Preferences for Redistribution? Evidence from Randomized Survey Experiments ," American Economic Review , 105 (4), 1478 – 1508.
Larson Richard C. (1987), " OR Forum—Perspectives on Queues: Social Justice and the Psychology of Queueing ," Operations Research , 35 (6), 895 – 905.
Lerner Abba P. (1944), The Economics of Control: Principles of Welfare Economics. New York : Macmillan.
Leventhal Gerald S. (1980), " What Should Be Done with Equity Theory? " in Social Exchange: Advances in Theory and Research , Gergen K. , Greenberg M. , Willis R. , eds. New York : Springer , 27 – 55.
Leventhal Gerald S. , Karuza Jurgis , Fry William R.. (1980), " Beyond Fairness: A Theory of Allocation Preferences ," Justice and Social Interaction , 3 (1), 167 – 218.
Liang Jianping , Chen Zengxiang , Lei Jing. (2016), " Inspire Me to Donate: The Use of Strength Emotion in Donation Appeals ," Journal of Consumer Psychology , 26 (2), 283 – 88.
Lichtenstein Donald R. , Burton Scot. (1989), " The Relationship Between Perceived and Objective Price-Quality ," Journal of Marketing Research , 26 (4), 429 – 43.
Maslow Abraham H. (1970), Motivation and Personality. New York : Harper & Row.
McGraw A. Peter , Schwartz Janet A. , Tetlock Philip E.. (2012), " From the Commercial to the Communal: Reframing Taboo Trade-Offs in Religious and Pharmaceutical Marketing ," Journal of Consumer Research , 39 (1), 157 – 73.
McGraw A. Peter , Tetlock Philip E.. (2005), " Taboo Trade-Offs, Relational Framing, and the Acceptability of Exchanges ," Journal of Consumer Psychology , 15 (1), 2 – 15.
Milgrom Paul R. , Weber Robert J.. (1982), " A Theory of Auctions and Competitive Bidding ," Econometrica , 50 (5), 1089 – 1122.
Norton Michael I. , Ariely Dan. (2011), " Building a Better America—One Wealth Quintile at a Time ," Perspectives on Psychological Science , 6 (1), 9 – 12.
Okun A. (1981), Prices and Quantities: A Macroeconomic Analysis. Washington, DC : Brookings Institution.
Oliver Richard L. , Swan John E.. (1989), " Equity and Disconfirmation Perceptions as Influences on Merchant and Product Satisfaction ," Journal of Consumer Research , 16 (3), 372 – 83.
Oppenheimer Daniel M. , Meyvis Tom , Davidenko Nicolas. (2009), " Instructional Manipulation Checks: Detecting Satisficing to Increase Statistical Power ," Journal of Experimental Social Psychology , 45 (4), 867 – 72.
Prendergast Canice. (2017), " How Food Banks Use Markets to Feed the Poor ," Journal of Economic Perspectives , 31 (4), 145 – 62.
Radford Robert A. (1945), " The Economic Organisation of a POW Camp ," Economica , 12 (48), 189 – 201.
Roth Alvin E. (2007), " Repugnance as a Constraint on Markets ," Journal of Economic Perspectives , 21 (3), 37 – 58.
Roth Alvin E.. (2015), Who Gets What—and Why: The New Economics of Matchmaking and Market Design. New York : Houghton Mifflin Harcourt.
Roth Alvin E. , Sotomayor Marilda. (1992), " Two-Sided Matching ," in Handbook of Game Theory with Economic Applications , Vol. 1 , Aumann R. , Hart S. , Zamir S. , Young P. , eds. Amsterdam : Elsevier , 485 – 541.
Savage David A. , Torgler Benno. (2010), " Perceptions of Fairness and Allocation Systems ," Economic Analysis and Policy , 40 (2), 229 – 48.
Shaddy Franklin , Fishbach Ayelet , Simonson Itamar. (2021), " Trade-Offs in Choice ," Annual Review of Psychology , 72 , 181 – 206.
Shaddy Franklin , Shah Anuj K.. (2018), " Deciding Who Gets What, Fairly ," Journal of Consumer Research , 45 (4), 833 – 48.
Skitka Linda J. , Tetlock Philip E.. (1992), " Allocating Scarce Resources: A Contingency Model of Distributive Justice ," Journal of Experimental Social Psychology , 28 (6), 491 – 522.
Soman Dilip. (1999), " Effects of Payment Mechanism on Spending Behavior: The Illusion of Liquidity ," Journal of Consumer Research , 27 (4), 460 – 74.
Spiller Stephen A. , Belogolova Lena. (2017), " On Consumer Beliefs About Quality and Taste ," Journal of Consumer Research , 43 (6), 970 – 91.
Stiglitz Joseph E. (2012), The Price of Inequality: How Today's Divided Society Endangers Our Future. New York : WW Norton & Company.
Sunstein Cass R. (2007), " Willingness to Pay vs. Welfare ," Harvard Law & Policy Review , 1 (2), 303 – 30.
Tanner Carmen , Medin Douglas L. , Iliev Rumen. (2008), " Influence of Deontological Versus Consequentialist Orientations on Act Choices and Framing Effects: When Principles Are More Important Than Consequences ," European Journal of Social Psychology , 38 (5), 757 – 69.
Taylor Shirley. (1994), " Waiting for Service: The Relationship Between Delays and Evaluations of Service ," Journal of Marketing , 58 (2), 56 – 69.
Tellis Gerard J. , Wernerfelt Birger. (1987), " Competitive Price and Quality Under Asymmetric Information ," Marketing Science , 6 (3), 240 – 53.
Tetlock Philip E. (2003), " Thinking the Unthinkable: Sacred Values and Taboo Cognitions ," Trends in Cognitive Sciences , 7 (7), 320 – 324.
Tetlock Philip E. , Kristel Orie V. , Beth Elson S. , Green Melanie C. , Lerner Jennifer S.. (2000), " The Psychology of the Unthinkable: Taboo Trade-Offs, Forbidden Base Rates, and Heretical Counterfactuals ," Journal of Personality and Social Psychology , 78 (5), 853 – 70.
Tormala Zakary L. , Jia Jayson S. , Norton Michael I.. (2019), " The Preference for Potential ," Journal of Personality and Social Psychology , 103 (4), 567 – 83.
Walasek Lukasz , Bhatia Sudeep , Brown Gordon DA. (2018), " Positional Goods and the Social Rank Hypothesis: Income Inequality Affects Online Chatter About High-and Low-Status Brands on Twitter ," Journal of Consumer Psychology , 28 (1), 138 – 48.
Warren Caleb , Peter McGraw A. , Van Boven Leaf. (2011), " Values and Preferences: Defining Preference Construction ," Wiley Interdisciplinary Reviews: Cognitive Science , 2 (2), 193 – 205.
Yaari Menahem E. , Bar-Hillel Maya. (1984), " On Dividing Justly ," Social Choice and Welfare , 1 (1), 1 – 24.
Zhou Rongrong , Soman Dilip. (2003), " Looking Back: Exploring the Psychology of Queuing and the Effect of the Number of People Behind ," Journal of Consumer Research , 29 (4), 517 – 30.
Zwolinski Matt. (2008), " The Ethics of Price Gouging ," Business Ethics Quarterly , 18 (3), 347 – 78.
~~~~~~~~
By Franklin Shaddy and Anuj K. Shah
Reported by Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 139- Who We Are and How We Govern: The Effect of Identity Orientation on Governance Choice. By: Heide, Jan B.; Bell, Simon J.; Tracey, Paul. Journal of Marketing. May2022, p1. DOI: 10.1177/00222429221094027.
Ahead of Print- Database:
- Business Source Complete
Record: 140- Why Is It Wrong to Sell Your Body? Understanding Liberals’ Versus Conservatives’ Moral Objections to Bodily Markets. By: Goenka, Shreyans; van Osselaer, Stijn M.J. Journal of Marketing. Nov2021, p1. DOI: 10.1177/00222429211046936.
Ahead of Print- Database:
- Business Source Complete
Record: 141- Why Salespeople Avoid Big-Whale Sales Opportunities. By: Xu, Juan; van der Borgh, Michel; Nijssen, Edwin J.; Lam, Son K. Journal of Marketing. Sep2022, Vol. 86 Issue 5, p95-116. 22p. 2 Diagrams, 4 Charts, 2 Graphs. DOI: 10.1177/00222429211037336.
- Database:
- Business Source Complete
Why Salespeople Avoid Big-Whale Sales Opportunities
Contrary to the intuition that salespeople gravitate toward big-whale sales opportunities, in reality they often avoid them. To study this phenomenon, the authors integrate contingent decision-making and conservation-of-resources theories to develop and test a framework of salespeople's decision making when prospecting. Study 1 reveals that the performance impact of salesperson initial judgment of opportunity magnitude follows an inverted U-shape, indicating that salespeople's avoidance of large opportunities results from rational benefit–cost analyses due to their conservation of resources. Interestingly, salespeople use a calibration decision-making strategy (i.e., calculating expected benefits by accounting for conversion uncertainty) at the portfolio rather than prospect level, in solution- but not product-selling contexts. Ignoring this calibration effect can lead to under- or overestimation of conversion rates of up to 100%. Study 2 shows that salespeople's past performance success and experience bias this calibration. Simulations reveal that when high performers or inexperienced salespeople believe their portfolio magnitude is large and conversion uncertainty low, they are less concerned about resource conservation and improve their quota attainment by 50%. Study 3 confirms the theoretical mechanism. These findings shed new light on salespeople's decision making and suggest ways for sales professionals to improve effectiveness when prospecting.
Keywords: salesperson judgment; uncertainty; solution selling; prospecting; conservation-of-resources theory
Central to a firm's customer acquisition is salesperson prospecting, which involves identifying sales opportunities among potential customers. Prior research on salesperson prospecting has underscored its importance not only for firms' customer relationship management (CRM) but also for salesperson performance (e.g., [48]). However, more than 40% of salespeople report that prospecting is challenging and full of uncertainty, taking on average 25% of their time ([ 8]). Whereas some practitioners emphasize the pursuit of large prospects because these "big whales" help firms and salespeople achieve rapid sales growth, others warn against prioritizing such prospects because they can easily drain salesperson and company resources ([32]). Moreover, "in the time it takes to land one major deal, [the salesperson] could have closed five smaller deals" ([19]). Although practitioners appear to recognize salespeople's benefit–cost trade-offs when prospecting, academic research has not systematically examined this important phenomenon.
Marketing research on salesperson prospecting has developed along two major streams. One stream focuses on salesperson judgment of market opportunities, such as market demand for a new brand, customer needs, and expected performance (e.g., [27]; [31]; [66]). This research stream shows a linear positive relationship between customer demand judgment and salesperson performance. The other stream emphasizes the role of salespeople as decision makers in dealing with various types of uncertainty, such as salespeople's general risk aversion, context-specific uncertainty, or salesperson idiosyncratic characteristics (e.g., [ 2]; [10]; [41]; [55]; [60]). Although these research streams provide useful insights into the information salespeople use in their decision making, three important research gaps exist.
First, when prospecting, salespeople typically identify multiple potential opportunities but pursue only some of them. However, research is scant on the potentially curvilinear impact of salesperson judgment of opportunity magnitude—defined as a salesperson's judgment of the size of an opportunity—on sales performance. Although anecdotal evidence suggests that salespeople focus on large opportunities, other sources allude to major drawbacks in pursuing them. Opening this black box can be useful for improving companies' prospecting effectiveness. Second, there is a lack of understanding of how conversion uncertainty affects salespeople's decision making when prospecting. A focus on conversion uncertainty is important, because prospecting is costly for the firm and for salespeople. Third, little is known about how such decision making varies between salespeople and selling contexts. Knowledge of these contingencies help sales managers to effectively manage salesperson prospecting behavior.
To address these gaps, we seek answers to three key questions. First, what is salespeople's benefit–cost trade-off after they form an initial judgment of opportunity magnitude, and how does this affect their sales performance? The focus on initial judgments is based on prior research that underscores the importance of a primacy effect in both decision making and salesperson–customer interactions ([17]; [27]). Second, how does a salesperson's initial judgments of opportunity conversion uncertainty change the sales performance outcome of the benefit–cost analysis? Given that salespeople's compensation generally depends on conversion, understanding the effect of opportunity conversion uncertainty, or a salesperson's initial judgment of the likelihood to convert an opportunity into a deal, is important. Third, what are important boundary conditions of the effects of these initial judgments? We focus on two sets of moderators: ( 1) the selling context (i.e., product vs. solution selling) and ( 2) key salesperson characteristics (i.e., past performance success and salesperson experience). In doing so, we also explore the role of information level (i.e., prospect and portfolio levels) in salesperson decision making.
To answer our questions, we develop and test a contingency framework of salespeople's decision making when prospecting for market opportunities in a sequence of three studies. For theoretical foundation, we integrate research on contingency decision making ([46]) and conservation of resources (COR) ([30]). We augment these theories with field notes from in-depth interviews with sales professionals. While Studies 1 and 2 rely on multisource field data, Study 3 is a scenario-based experiment to provide evidence of the benefit–cost analysis as the underlying mechanism. Together, this multimethod approach allows us to rigorously triangulate the effects and unpack the theoretical mechanisms.
This research makes several contributions. First, we contribute to the emerging literature on salesperson judgment and decision making when prospecting by unpacking the underlying decision process. We provide theoretical arguments and strong empirical evidence that explains why salespeople avoid big-whale prospects. Specifically, we show that, based on their initial judgment of opportunity magnitude, salespeople conduct a benefit–cost analysis under resource constraints to decide which opportunity to pursue. This analysis results in an inverted U-shaped relationship between magnitude and performance. Spotlight analyses show that a one-standard-deviation (SD) increase in opportunity magnitude lowers salespeople's conversion rate by 10%.
Second, we provide insights into the effect of conversion uncertainty on the salesperson decision-making process when prospecting. The results show that when selling solutions, salespeople use a calibration decision-making strategy, in which the effects of opportunity magnitude are conditional on conversion uncertainty. However, this strategy occurs only at the portfolio level, underscoring the role of salesperson portfolio as a decision-making context. Ignoring the calibration effect in estimating performance outcomes may lead to under- or overestimation of conversion rates of up to 100%. When selling products, salespeople use a compensatory decision-making strategy that accounts for the effects of portfolio magnitude and conversion uncertainty in an additive manner. These findings extend prior work (e.g., [60]) on uncertainty in personal selling and salesperson decision making.
Third, we provide empirical evidence for how, in a solution-selling context, the calibration effect varies depending on salesperson past performance success and experience. The results suggest that when faced with high levels of conversion uncertainty, high performers and inexperienced salespeople perform much worse because their resource-conserving tendency makes them more sensitive to the cost increases associated with uncertainty. Simulations reveal that their quota attainment can suffer by as much as 50%. These insights extend prior research focusing on the salesperson–customer dyad in business-to-business (B2B) marketing and retail encounters (e.g., [27]; [43]).
Salespeople are generally assigned to a territory or a customer segment, and their sales opportunities can be self-generated or assigned ([48]). Within a given period, they move these sales opportunities through a funnel from prospects to closed sales deals. At any given time, salespeople form a judgment of the magnitude of specific sales prospects, with a certain level of conversion uncertainty. Prior research on salesperson prospecting provides useful insights into why salespeople fail to follow sales leads, how their judgment of opportunities linearly influences their performance, and how they deal with uncertainty. However, it has not examined why and when salespeople pursue or avoid big opportunities. To shed light on these issues, we view salesperson prospecting as decision making under resource constraints. In this section, we first briefly review the relevant literature and then present our conceptual framework.
Two major decision-making frameworks are the benefit–cost framework ([ 6]) and perceptual frameworks, such as prospect theory ([59]). In their review of these two frameworks, [46] posit that the former provides insights into rational decision making under multiple alternatives while the latter is useful in explaining cognitive biases and heuristics in decision making. In their review, they also underscore task and individual characteristics, such as willingness to bear uncertainty, as important contingencies of individual decision making.
Unlike the majority of general decision-making frameworks that assume away any resource constraint, COR theory emphasizes that people "strive to retain, protect, and build resources and that what is threatening to them is the potential or actual loss of these valued resources" ([30], p. 513). Furthermore, people must invest resources to gain resources, and those who experience a lack of resources attempt to conserve remaining resources ([26]). We argue that COR theory is particularly relevant in the context of B2B salespeople's prospecting for three reasons. First, unlike simple, low-effort choices between two lotteries, the pursuit of a prospect is costly—salespeople need to invest their time, effort, and resources in converting prospects into sales ([48]). Second, uncertainty in prospecting brings salespeople's resource constraints to the fore. Unlike gambling, which can be replayed, a forgone sales opportunity might be gone for good, and a failure to convert opportunities represents a loss of resources. Thus, salespeople need to balance between risk seeking and COR. Third, salespeople differ in terms of resource constraints ([48]).
We integrate decision-making frameworks with COR theory to propose a contingency framework of salesperson decision making when prospecting. Our framework focuses on the initial judgments of sales opportunities in terms of magnitude and conversion uncertainty. First, although salespeople encounter multiple market opportunities, they only invest their resources into converting some of them. The benefit–cost framework suggests that this decision is based on rational benefit–cost analyses of opportunity magnitude before action ([ 6]). COR theory offers a similar explanation that the pursuit of an opportunity is a trade-off between resource acquisition (e.g., the expected benefits of a sale) and resource conservation (e.g., the costs of resources expended on pursuing the opportunity). Second, salespeople make this decision under uncertainty. In line with contingency decision-making frameworks and COR theory, we expect that salespeople's benefit–cost analysis of opportunity magnitude is contingent on conversion uncertainty. This is because uncertainty prevents action by obfuscating "whether the potential reward of action is worth the potential costs" ([40], p. 139; see also [26]; [46]).
Third, task and personal factors represent additional contingencies that distort the rational benefit–cost analyses. We focus on two sets of contingencies. The first is the selling context (i.e., product vs. solution selling), in which a product denotes a physical object that can be sold in a transactional way (e.g., lamps) and a solution refers to a product-service system (e.g., smart lighting) that requires a relational process and tailoring. Solution selling represents a more uncertain task than product selling ([57]; [60]). Examining the selling context is important because many firms that shift from product to solution selling often struggle to cope with the inherent greater uncertainty (e.g., [18]; [60]). The second set includes two salesperson characteristics related to the propensity to conserve resources under uncertainty. Past performance success refers to salesperson quota attainment in the previous quota period. Salesperson experience refers to a salesperson's time in the sales territory, with the company, and in the sales profession ([ 2]). We focus on these two moderators because prior research suggests these factors are related to how salespeople deal with uncertainty and conserve resources ([26]; [30]; [48]).
To test our conceptual framework, we conducted three studies using multiple methods. We provide an overview of our conceptual framework, hypotheses, and the studies in Figure 1. Study 1 focuses on how salesperson initial judgments of opportunity magnitude determine the actual conversion of a prospect into a sale, which in the aggregate influences the salesperson conversion rate at the portfolio level. In doing so, we also investigate how salespeople calibrate opportunity magnitude for opportunity conversion uncertainty and whether such calibration differs between product and solution selling. In Study 2, we examine the heterogeneity of such calibration effect, with a focus on salespeople's past performance success and experience. The dependent variable in Study 2 is salesperson quota achievement, which is theoretically connected with the conversion rate examined in Study 1. In Study 3, a scenario-based experiment, we elucidate the underlying benefit–cost mechanism and the role of resource slack. Table 1 summarizes the key concepts in our framework and corresponding operational measures.
Graph: Figure 1. A contingency framework of salesperson decision making when prospecting and overview of three studies.
Graph
Table 1. Overview of Key Concepts and Operationalizations in Studies 1 and 2.
| Key Concepts | Description and Conceptual Meaning | Conceptual Foundations | Operationalization | Study 1 | Study 2 | Representative Studies |
|---|
| Performance Outcome | | | |
| Sales performance | Degree to which the salesperson obtains a desired outcome | Ahearne et al. (2010) | Prospect-level performance: A binary measure, where 0 reflects no deal and 1 reflects that a deal has been made (objective likelihood of conversion) | ✓ | | Smith, Gopalakrishna, and Chatterjee (2006); Mayberry, Boles, and Donthu (2018) |
| Portfolio-level performance: The ratio of prospects that are successfully turned into deals in a salesperson's portfolio (objective conversion rate) | ✓ | | Own operationalization |
| Salesperson performance: Percentage of quota attainment | | ✓ | Ahearne et al. (2010) |
| Initial Cues and Judgment Formation | | | |
| Opportunity magnitude | The size of a potential sales option | Kumar, Petersen, and Leone (2013) | Prospect magnitude: Log-transformed salesperson initial judgment of revenue for a prospect (i.e., deal magnitude in $) | ✓ | | Mayberry, Boles, and Donthu (2018) |
| Portfolio baseline magnitude: The mean of all prospects' magnitude in a salesperson' portfolio | ✓ | | Own operationalization |
| Portfolio magnitude: Reflective four-item construct capturing a solution-selling salesperson's initial estimates in terms of the size of order intake, sales volume, revenue, and profits of their entire portfolio. | | ✓ | Van der Borgh, De Jong, and Nijssen (2019) |
| Opportunity conversion uncertainty | The subjective likelihood of being able to convert an opportunity into a desired sales outcome | McMullen and Shepherd (2006) | Prospect conversion uncertainty: Categorical measure differentiating among high, medium, and low levels of likelihood to convert a prospect into a paying customer within six months (−1, 0, and 1), based on salesperson initial judgment. | ✓ | | Own operationalization |
| Portfolio baseline conversion uncertainty: The average of prospect conversion uncertainty across all the prospects in a salesperson's portfolio | ✓ | | Own operationalization |
| Portfolio conversion uncertainty: Reflective four-item construct capturing a salesperson's initial judgment of uncertainty for realizing anticipated outcomes for solution selling for their entire portfolio (in terms of size of order intake, sales volume, revenue, and profits) | | ✓ | Own operationalization |
| Contingencies | | | |
| Salesperson characteristics | Differences in motivation, attitude, or risk propensity that determine whether a salesperson is willing to bear uncertainty or not | McMullen and Shepherd (2006); Payne, Bettman, and Johnson (1992) | Past performance success: The percentage of quota achieved in the previous quota cycle | | ✓ | Mayberry, Boles, and Donthu (2018) |
| Salesperson experience: Composite measure consisting of three measures of sales experience (i.e., time in sales territory, time with the company, and time in the sales profession). We z-scored these scores and averaged them to form an overall experience index. | | ✓ | Ahearne et al. (2010) |
| Task characteristics | Various dimensions, descriptors, or attributes of a particular organizational position that determine task execution and/or outcomes | Payne, Bettman, and Johnson (1992) | Binary measure indicating whether a prospect requires product selling or solution selling | ✓ | | Mayberry, Boles, and Donthu (2018) |
| Information levels | Denotes the reference class of judgments, distinguishing between judgment about the specific case or the aggregate of multiple cases | Sniezek and Buckley (1995) | Data are separated into information at the single case level (prospect information) and aggregate level (portfolio baseline information) | ✓ | | Own operationalization |
Study 1 examines the interaction effect between opportunity magnitude and conversion uncertainty in product- and solution-selling contexts at both the prospect and portfolio levels. We supplement our theoretical development for this study with verbatim quotes from a qualitative study of seven salespeople (for a description of respondents, see Web Appendix W1).
We predict that the effect of opportunity magnitude on salesperson performance follows an inverted U-shaped relationship. This is due to two countervailing underlying mechanisms: a linear positive effect from potential benefits of pursuing an opportunity and a curvilinear effect from potential costs of such pursuit. We follow [24] recommendation to visually summarize this benefit–cost analysis in Figure 2, Panel A. On the one hand, the higher the magnitude of the opportunity, the greater the extrinsic and intrinsic benefits of pursuing a sizable opportunity. Extrinsic benefits take the form of potential compensation and recognition from the selling firm, whereas intrinsic benefits include the potential enjoyment in pursuing sizable opportunities and the learning associated with the pursuit ([ 7]; [15]).
Graph: Figure 2. Illustration of theoretical arguments for the inverted U-shaped effect and the calibration effect.
On the other hand, pursuing a sizable opportunity entails substantial explicit and implicit costs. Salespeople incur explicit costs when pursing an opportunity because they need to invest resources, such as time and effort ([48]). Implicit costs refer to the opportunity costs of such pursuit because when pursuing an opportunity, salespeople must forgo other opportunities ([54]). As the opportunity magnitude increases, both explicit and implicit costs accelerate significantly because salespeople are constrained by limited resources and information-processing capacity ([30]; [48]; [55]). One senior salesperson of a software company explained this issue succinctly:
Big opportunities? Um, the pros. The prospect of earning a load of money.... Second, obviously it's also more satisfying or fulfilling.... So, it's more complex, which is also like a nice challenge. Plus, you learn the most from complex deals and bigger deals.... But it costs a lot of time and whatever time you spend on one deal you cannot reinvest anymore in smaller deals. So, there's always a trade-off.
Therefore, when both potential benefits and costs are considered, the effect of opportunity magnitude on salesperson performance will incrementally increase at first, but after a threshold, the costs to act on moderate to large opportunities outweigh their benefits. Thus,
- H1: All else being equal, salesperson judgment of opportunity magnitude has a curvilinear, inverted U-shaped effect on salesperson performance.
By itself, conversion uncertainty can be a source of benefits for salespeople because uncertainty stimulates positive feelings and excitement ([50]). However, conversion uncertainty also increases costs of pursuing—both explicit costs (i.e., salespeople need to exert greater effort to convert highly uncertain opportunities) and implicit costs (i.e., highly uncertain opportunities carry higher opportunity costs). Under the compensatory decision-making strategy, salespeople assess the benefits and costs of opportunity magnitude and conversion uncertainty in an additive manner. However, our interviews suggest that salespeople at times calibrate for conversion uncertainty in a multiplicative rather than additive manner. For example, one solution-selling salesperson was very clear on his decision-making strategy to deal with conversion uncertainty:
When I take a look at a deal that has a very high certainty, so say you've a 90% closing chance, but it's very small in size, and you have a very big deal that has a small closing chance. Yeah, I can just multiply it [size by uncertainty] and see where I get the most buck for my uncertainty, so to speak. Even though it's very simple, it's pretty effective.
We refer to this multiplicative strategy as the calibration hypothesis, such that the inverted U-shaped relationship between opportunity magnitude and salesperson performance in H1 is contingent on conversion uncertainty ([40]). We again follow [24] recommendation to visually summarize how conversion uncertainty influences the benefit and cost functions in our arguments in Figure 2, Panel B.
In terms of the benefit function, the (previously noted) solution-selling salesperson's calculus is consistent with both expectancy theory and COR theory ([30]; [64]). These theories suggest that salespeople calculate the expected benefits of an action by multiplying its benefits by the success odds (i.e., expected benefits = magnitude-based benefits × conversion uncertainty). Thus, when conversion uncertainty is high, the expected benefits of pursuing a large opportunity are lower, making it less motivating to pursue. Importantly, this calculus is universal, without evoking any individual characteristics as contingencies. Therefore, conversion uncertainty weakens the slope of the benefit line, shifting the inverted U-shaped effect of opportunity magnitude on salesperson performance to the left.
How do salespeople calibrate for conversion uncertainty in assessing the costs of pursuing an opportunity? As mentioned previously, conversion uncertainty can be positively stimulating for risk seekers but harmfully costly for people who want to conserve resources. In this regard, prior research indicates that salespeople are heterogeneous in their risk-seeking behavior for various reasons, such as their past performance success and their capability (e.g., [42]). Given this heterogeneity, the cost function can swing in either direction, and thus we predict that, in the aggregate, opportunity conversion uncertainty may appear as not having an influence on the cost function. For the leftward shifting effect to occur, opportunity conversion uncertainty only needs to shift the benefit function downward and does not need to change the shape of the cost function ([24]). Thus,
- H2: Opportunity conversion uncertainty moderates the effect of opportunity magnitude on salesperson performance, such that it shifts the inverted U-shaped effect of opportunity magnitude on salesperson performance to the left.
Prior research on decision making suggests that, under high levels of situational uncertainty, people search for a relevant reference class to calibrate their judgments ([23]; [33]). However, [60] suggest that, beyond outcome uncertainty, such as conversion uncertainty, solution selling has a higher level of need and process uncertainty than product selling. We argue that it is this difference in overall situational uncertainty that causes salespeople to calibrate for conversion uncertainty differently when selling products versus solutions. Specifically, because need and process uncertainties are higher in solution selling, salespeople do not have a reliable frame of reference to count on. By contrast, because need and process uncertainties are lower in product selling, salespeople can confidently draw from their knowledge of customer needs and requirements, the sales process, and product configurations to deal with conversion uncertainty. Therefore, compared with product-selling salespeople, solution-selling salespeople are more sensitive to conversion uncertainty and tend to calibrate for this uncertainty more intensely. Thus, we expect the weakening effect of conversion uncertainty on the benefits function predicted in H2 to be stronger in solution- than product-selling contexts.
- H3: The leftward shifting effect of opportunity conversion uncertainty on the inverted U-shaped relationship between opportunity magnitude and salesperson performance is stronger in solution- than product-selling contexts.
We collected data from a Fortune Global 500 firm, a market leader in lighting products and solutions for enterprise customers; at the time of data collection, the company generated more than $25 billion annually in total revenue. The company provides a broad portfolio of lighting offerings, ranging from products (e.g., luminaires, lighting electronics, horticulture lighting) to system solutions (e.g., connected, smart luminaires; lighting management software). Customers come from various industries, such as food and fashion retail, health care, education, sports, municipalities, hospitality, infrastructure, and manufacturing. For its field-based sales approach, the company relies primarily on direct sales, and salespeople are subject to the same compensation and incentive scheme. Salespeople obtain a fixed yearly salary plus commission (maximum 30% of the fixed salary). To explore the impact of initial judgments of opportunity magnitude and conversion uncertainty, we gathered archival data from the company's sales force automation (SFA) system for all prospects within one market. For every prospect, we obtained transaction-level records from January 2016 to May 2017, including initial estimates of opportunity magnitude (i.e., prospect deal size) and conversion uncertainty, updated estimates, and the final sales outcome. The SFA data cover 12,988 B2B prospects, handled by 173 salespeople, who logged 110,278 events in total. We provide supplemental information about the research context and the SFA data in Web Appendix W2.
Because we are interested in the performance impact of a salesperson's initial judgments, we aggregated the event-level data to the prospect level. This approach allows us to estimate a two-level model in which prospect-level data (case-specific; within salesperson) are nested within portfolio-level data (baseline; between salesperson).
A unique feature of Study 1 is that we leverage the company's SFA data to operationalize the key variables. We measure opportunity magnitude as the salesperson's initial point estimate of a prospect in terms of revenue. Following [62], we log-transform the measure to correct for right-skewness. We measure opportunity conversion uncertainty as a categorical measure that captures the probability of converting a prospect into a deal within six months. We coded these categories as −1 (low), 0 (medium), and 1 (high) to facilitate interpretation and enhance model parsimony ([16]). We measure prospect-level performance as a binary measure that indicates the actual conversion of a prospect at the end of the sales cycle (0 = no deal, 1 = a deal). Portfolio-level performance indicates the salesperson's portfolio-level conversion rate, aggregated from their prospect-level actual conversion.
To obtain unbiased estimates, we control for the nonlinear effects of uncertainty by including a square term ([20]). To zero in on the effects of initial judgments, we control for several time-related dynamics. Specifically, we control for five (e.g., [ 1]). First, we control for duration of a sales cycle by including the sales cycle length ([39]). Second, we control for frequency of leads by including workload ([48]), measured as the total number of leads under a salesperson's wing during the assessment. Previous studies have shown that workload affects judgments (e.g., [22]). Third, we control for the frequency of uncertainty updates (i.e., process uncertainty), which reflects the total number of changes a salesperson has made after the initial uncertainty estimate. It reflects the doubt inherent in the sales process and filters out variation in the dependent variable after salespeople made their initial judgments, which is the focus of the article. Fourth, we control for accuracy and recency effects by including the difference between the initial and final estimates for opportunity magnitude and uncertainty (magnitude accuracy = [last estimate − first estimate]; conversion uncertainty accuracy = [first estimate − last estimate]). Fifth, we control for timing- and sequence-related dynamics by including time fixed effects. We use dummy variables to account for the prospect's industry (i.e., public, office, retail, and other). We also control for the potentially curvilinear effect of uncertainty because prior research suggests that people respond more rigorously to two ends of the uncertainty continuum than to moderate levels of uncertainty ([ 2]; [55]; [67]). Web Appendix W3 provides sample descriptives and the correlation matrix.
As is true in many B2B selling contexts, salespeople are responsible for a portfolio of prospects. In the "Study 1 Hypothesis Development" subsection, we use the term "opportunity" without specifying whether this opportunity is at the prospect or portfolio level. However, previous studies in decision making (e.g., [33]; [58], [59]) show that people ( 1) can leverage two levels of information (case-specific and base-rate) and ( 2) use a reference point. From a multilevel perspective, portfolio baseline judgments are essentially salesperson-level constructs that capture between-salesperson variation, serving the function of the base-rate information about the sales territory. Prospect-level judgments reflect within-salesperson judgment about specific prospects relative to each salesperson's portfolio baseline judgments ([12]). Because assuming that an effect existing at a higher level will generalize to a lower level (or vice versa) can be erroneous ([13]), we employ multilevel modeling techniques to estimate the impact of a salesperson's initial judgments of opportunity magnitude and conversion uncertainty on performance and test the effect at the prospect (case-specific) and portfolio (baseline) levels. Conceptually, people tend to rely heavily on the mean value in their decision making ([28])—akin to a salesperson's baseline. Therefore, in Study 1, we examine the portfolio average magnitude and average conversion uncertainty as the reference points at the portfolio level and refer to these as portfolio baseline magnitude and conversion uncertainty.
We specify a multilevel model using Mplus 8.3 ([44]). To allow for unbiased estimates at the between and within levels, we decompose the manifest variables into uncorrelated latent "between" and "within" components ([47]). Latent means of focal variables are estimated at the between level, while "pure" within-person effects are estimated in the within-level portion of the model. This specification is necessary for teasing apart prospect- and portfolio-level effects ([47]).
The complexity of the model and the use of a binary dependent variable did not allow use of robust maximum likelihood estimation techniques because of a lack of model convergence. As an alternative, we employed Bayesian estimation techniques and therefore specified a Bayesian multilevel probit model. We estimate two sets of models for the pooled (no separation of selling contexts), product-selling, and solution-selling data, respectively. Models 1–3 are main-effects-only models, and Models 4–6 are the interaction-effects models. All the manifest variables are standardized to aid in interpretation, with the exception of conversion uncertainty. We provide details on the model specification and estimation in Web Appendix W4.
The effect of opportunity magnitude and conversion uncertainty on performance may be endogenous, because common unobserved factors may influence both predictors and outcomes in our model (due to, e.g., simultaneity, measurement error, omitted variables). Following prior studies ([22]; [49]; [63]), we address endogeneity in three ways: ( 1) by adopting a rich data-modeling approach, ( 2) by controlling for endogeneity due to omitted variables, and ( 3) by checking for endogeneity due to selection bias. Because we consider a multilevel setting, we need to address endogeneity at each level ([37]). We correct for Level 1 endogeneity using a control function procedure ([49]) and check for robustness with an instrument-free Gaussian copula approach ([45]). We control for Level 2 (cross-level) endogeneity in our multilevel latent covariate model by allowing a correlation between random intercepts and slopes ([ 3]). A direct test of the random-effects assumption (Wald = 2.226, p = .136) indicates that Level 2 endogeneity is not a concern ([ 3]). Web Appendix W4 provides further details of model-free evidence of inverted U-shaped relationship, robustness checks of adding higher-order terms and seasonal variation to the empirical model, and endogeneity corrections.
We present the results of our tests for H1 and H2 for solution selling at the portfolio level before discussing the differences between solution selling and product selling (H3). In the "Discussion" section, we explore differences across portfolio and prospect levels.
Table 2 shows the results of the analyses. To test H1, we follow a rigorous three-step procedure ([24]). First, we find a significant, negative effect of (opportunity magnitude)2 on salesperson performance for solution selling (Model 3: γ02 = −.142, p < .001). We plot this effect in Panel A of Figure 3, which shows an inverted U-shape. Second, we formally test that the marginal effects on the left side of the turning point of the inverted U-shape are positive and significant and those on the right side of the turning point are negative and significant. Mathematically, we test whether γ01 + 2γ02XL is positive and significant and γ01 + 2γ02XH is negative and significant, where XL and XH represent low and high values of opportunity magnitude within the data range, respectively. For portfolio baseline magnitude, the results confirm this pattern (for details, see Web Appendix W5). Third, we examine whether the turning point (i.e., X) is located within the data range. Taking the first derivative of the Level 2 equation specified for Model 3 and setting it to zero yields a turning point X of −γ01/2γ02. We found that the turning point is 2.52 SD below the mean value and within the data range (Xsolution = −2.52 SD; 95% confidence interval [CI] = [−4.15, −1.38]). Overall, these results confirm an inverted U-shaped relationship between opportunity magnitude and salesperson performance for solution selling, in support of H1.
Graph: Figure 3. Study 1: inverted U-shape and the moderating effect of opportunity conversion uncertainty in solution selling.
Graph
Table 2. Study 1—Results of Multilevel Probit Analyses: Effect of Initial Judgment on Performance Outcomes.
| Step 1 | Step 2a | Hyp. |
|---|
| Model 1: Pooled | Model 2: Products | Model 3: Solutions | Model 4: Pooled | Model 5: Products | Model 6: Solutions |
|---|
| b | | SD | b | | SD | b | | SD | b | | SD | b | | SD | b | | SD |
|---|
| L2: DV = Portfolio-Level Performance |
| Magnitude (γ01) | −.536 | *** | .149 | −.301 | * | .162 | −.716 | ** | .232 | −.562 | *** | .178 | −.377 | * | .198 | −.240 | | .303 | |
| Magnitude2 (γ02) | −.264 | *** | .056 | −.225 | ** | .066 | −.142 | *** | .045 | −.261 | *** | .056 | −.234 | *** | .069 | −.161 | *** | .048 | H1 |
| Uncertainty (γ04) | −1.136 | *** | .249 | −1.026 | *** | .269 | −1.541 | ** | .481 | −1.248 | *** | .262 | −1.086 | *** | .279 | −1.556 | *** | .466 | |
| Magnitude × Uncertainty (γ03) | — | | — | — | | — | — | | — | .051 | | .158 | .108 | | .175 | −.663 | ** | .227 | H2/H3 |
| L1: DV = Prospect-Level Performance |
| Magnitude (β1j) | −.456 | *** | .075 | −.410 | *** | .062 | −.536 | *** | .155 | −.393 | *** | .052 | −.414 | *** | .065 | −.484 | *** | .159 | |
| Magnitude2 (β2j) | −.142 | *** | .026 | −.158 | *** | .022 | −.098 | ** | .038 | −.123 | *** | .019 | −.157 | *** | .022 | −.129 | *** | .043 | |
| Uncertainty (β4j) | −.932 | *** | .112 | −.877 | *** | .105 | −.571 | * | .258 | −.911 | *** | .094 | −.885 | *** | .104 | −.510 | *** | .259 | |
| Magnitude × Uncertainty (β3j) | — | | — | — | | — | — | | — | −.017 | | .040 | .011 | | .045 | −.053 | | .086 | |
| Controls |
| Uncertainty2 (L2) | .323 | | .275 | .249 | | .304 | 1.100 | * | .515 | .394 | *** | .273 | .251 | | .309 | .929 | * | .508 | |
| Uncertainty2 (L1) | −.093 | | .080 | −.129 | | .083 | −.167 | | .228 | −.058 | | .077 | −.126 | | .079 | −.235 | | .237 | |
| Process uncertainty | −.270 | *** | .016 | −.288 | *** | .017 | −.204 | *** | .040 | −.276 | *** | .016 | −.290 | *** | .017 | −.207 | *** | .040 | |
| Sales cycle length | .092 | *** | .021 | .088 | *** | .023 | .015 | | .142 | .087 | *** | .020 | .091 | *** | .022 | .069 | | .144 | |
| Magnitude accuracy | −.047 | *** | .012 | −.026 | * | .013 | −.095 | *** | .035 | −.047 | *** | .013 | −.027 | * | .013 | −.096 | ** | .035 | |
| Uncertainty accuracy | .820 | *** | .028 | .835 | *** | .031 | .675 | *** | .073 | .819 | *** | .028 | .834 | *** | .031 | .684 | *** | .073 | |
| Workload | −.004 | | .008 | .002 | | .008 | −.008 | | .019 | −.002 | | .006 | .003 | | .007 | −.010 | | .019 | |
| Magnitudeb | .020 | | .039 | −.006 | | .038 | .168 | | .119 | −.023 | | .032 | −.008 | | .039 | .114 | | .124 | |
| Magnitude × Magnitude | −.006 | | .015 | −.011 | * | .016 | .096 | * | .058 | −.038 | ** | .016 | −.012 | | .017 | .149 | ** | .066 | |
| Uncertainty | .062 | * | .030 | .015 | | .028 | −.095 | | .070 | .000 | | .027 | .015 | | .029 | −.112 | | .071 | |
| Uncertainty × Uncertainty | .109 | *** | .021 | .106 | * | .018 | .191 | *** | .047 | .095 | *** | .016 | .103 | *** | .020 | .194 | *** | .048 | |
| IMR | −1.291 | *** | .084 | −1.416 | *** | .088 | −2.292 | *** | .593 | −1.027 | *** | .079 | −1.423 | *** | .087 | −2.112 | *** | .592 | |
| Industry fixed effects | Yes | | | Yes | | | Yes | | | Yes | | | Yes | | | Yes | | | |
| Time fixed effects | Yes | | | Yes | | | Yes | | | Yes | | | Yes | | | Yes | | | |
| Constant (γ00) | 1.364 | *** | .202 | 1.282 | *** | .161 | 1.465 | | .833 | 1.199 | *** | .156 | 1.220 | *** | .173 | 1.101 | | .866 | |
| Pseudo-R2: L2/L1 | .795/.554 | .645/.545 | .805.716 | .801/.509 | .678/.543 | .831/.713 | |
| n (prospects)/N (salespeople) | 12,988/173 | 10,991/166 | 1,997/121 | 12,988/173 | 10,991/166 | 1,997/121 | |
1 **p < .01.
- 2 ***p < .001; unstandardized coefficients.
- 6 a To determine the extent of bias in our estimates ([24]), we also tested an extension of Models 4–6, in which we included Magnitude2 × Uncertainty at L2 (γ05) and L1 (β5j). The results for these models showed nonsignificant effects. The results show that neither interaction is significant ( = .064, p > .05; = .002, p > .05).
- 7 b We also tested the robustness of our findings by controlling for endogeneity using the instrument-free Gaussian copula approach. Findings are similar when compared with the control function approach reported in this table. For details of the results of these additional analyses, see Web Appendix W4.
- 8 Notes: L2 = portfolio level; L1 = prospect level. Magnitude = opportunity magnitude; Uncertainty = opportunity conversion uncertainty. IMR = inverse Mills ratio.
To test H2, we add the interaction term (magnitude × uncertainty)ij to the equation. The results of Model 6 in Table 2 confirm that the interaction is significant and negative in the solution-selling context (γ03 = −.663, p <.01). However, we cannot determine significance from the estimated interaction term alone in a nonlinear model (probit) with nonlinear interaction terms ([65]). Therefore, we formally test how the turning point changes as conversion uncertainty changes. To do so, we derive the turning point "magnitude*" (X*) by setting the first derivative of Model 6's equation with respect to X to zero. Then, we take the derivative of the turning point with respect to conversion uncertainty (Z) to show how the turning point changes as conversion uncertainty changes, yielding δX*/δZ = (−γ02 γ05)/[2(γ02)2]. We find that this term is significant and negative (b = −2.032, p <.01), in support of H2.
To test H3, we compare the results for the two selling tasks (see Table 2). At the portfolio level, in contrast with the significant and negative interaction in the solution-selling context (Model 6: γ03 = −.663, p <.01), we find no interaction effect of opportunity magnitude and conversion uncertainty on salesperson performance for product selling (Model 5: γ03 = .108, p >.10). Testing the difference between solution versus product selling reveals significant differences at the portfolio level (Δ[γ03_Solution; γ03_Product] = .771, p <.01), in support of H3.
Study 1 provides evidence of an inverted U-shaped relationship between opportunity magnitude and sales performance, across levels and selling context. However, we only find evidence of the calibration hypothesis for the solution-selling context at the portfolio level. This suggests that salespeople respond differently to opportunities of different magnitude, depending on the baseline conversion uncertainty of their portfolio of solution prospects. To illustrate the impact of this effect, we ran a simple counterfactual analysis and compared the calibration model with a compensatory model (i.e., fixing the interaction coefficient γ03 to zero). The results show that under certain conditions the compensatory model over- or underestimates conversion rates by almost 100% (e.g., predict 100% conversion, 0% "true" value). For large opportunity magnitude (>∼$26.750), the compensatory model overestimates conversion rates by up to 30% for high conversion uncertainty but underestimates conversion rates by up to 90% for low conversion uncertainty. Overall, these results highlight the importance of the calibration model.
We also found that the information level (i.e., prospect- and portfolio-level) matters. While prior research on decision making generally focuses on two simple choices, it might be cognitively impossible for salespeople to constantly make decisions at the prospect level while juggling a portfolio of prospects in their sales funnel. Consistent with this notion, our findings in Table 2 suggest that salespeople adopt a simpler compensatory decision-making strategy at the prospect level (i.e., that accounts for conversion uncertainty in an additive/subtractive manner) but rely on a more complex decision-making strategy at the portfolio level (i.e., a calibration strategy that accounts for conversion uncertainty in an interactive manner, using portfolio baseline information) for solution selling. In practice, salespeople generally have an idea about this so-called portfolio baseline in their assigned territory, such as the average magnitude and uncertainty of their portfolio. They then assess individual opportunities relative to this baseline and prioritize accordingly ([53]; [58]).
Although Study 1 shows that salespeople calibrate for opportunity conversion uncertainty when selling solutions at the portfolio level, it does not investigate how salespeople differ in their calibration. Study 2 focuses on salespeople's past performance success and experience as two key boundary conditions that bias their rational calibration of benefit–cost analyses. Empirically, a test of these contingencies is a three-way interaction test of the two-way interaction in H2.
We first focus on how past performance success influences the way salespeople calibrate their benefit function for conversion uncertainty differently. Compared with salespeople with low past performance, those with high past performance have a higher sense of competence. As a result, they are more risk-seeking and view uncertain opportunities as challenging and intrinsically motivating ([42]; [56]). To these high performers, the intrinsic benefits associated with high conversion uncertainty may outweigh the potential loss of extrinsic benefits. By contrast, salespeople with low past performance repulse opportunities that have high conversion uncertainty. This aversion arises because these highly uncertain opportunities not only threaten their potential extrinsic benefits (e.g., losing compensation and rewards) but also represent an unreliable path to achieve intrinsic benefits (e.g., bolstering their lack of competence; see [15]). Thus, the downward shift of the benefit function created by opportunity conversion uncertainty (predicted in H2) is weaker among salespeople whose past performance success is high (vs. low).
In terms of the cost function, being successful in the prior period induces high performers to be more sensitive to opportunity conversion uncertainty for two reasons. First, consistent with COR theory, they are likely to slow down to conserve their resources to reduce stress ([30]). Empirical evidence shows that people tend to hold back after achieving a goal before working on the next goal (e.g., a resetting period; [11]; [34]). Second, high performers are more sensitive to high implicit costs associated with high conversion uncertainty because opportunities that can be converted with certainty allow them to maintain their status (e.g., [38]). Therefore, for high performers, opportunity conversion uncertainty is likely to shift their cost function upward more strongly. This upward shift is especially strong when opportunity magnitude is large because large opportunities require them to invest much more resources. By contrast, while salespeople whose past performance was less successful are also sensitive to the opportunity costs associated with high conversion uncertainty, their main concern is to prove themselves to the firm. Therefore, these poor performers need to exert greater efforts and cannot afford to conserve their resources. As a result, poor performers' cost function shifts upward less strongly when opportunity conversion uncertainty is high.
Taken together, compared with salespeople who are low past performers, high performers view opportunities with high conversion uncertainty as more beneficial but also more costly. In prospecting, although all salespeople have limited resources ([48]), high performers are more inclined to conserve their resources than poor performers. Thus, for high past performers, we predict that the upward shifting effect of conversion uncertainty on the cost function will outweigh its downward yet weaker shifting effect on the benefit function. As a result, the expected net benefits of pursuing an opportunity will be lower when conversion uncertainty is high, causing the inverted U-shaped relationship between opportunity magnitude and salesperson performance to shift more strongly to the left ([24]).
- H4: The greater salesperson past performance success, the stronger is the leftward shifting effect of opportunity conversion uncertainty on the inverted U-shaped relationship between opportunity magnitude and salesperson performance.
We argue that because experienced salespeople differ from less experienced salespeople in terms of resources, they calibrate their benefits and costs under conversion uncertainty differently. First, they have better network-based resources in the form of relationships they have built over time. Second, they are more knowledgeable about various aspects of the sales process (e.g., customers, the market, the competition, the company), another resource critical for success in prospecting ([48]).
In terms of the benefit function, the resources accumulated over time make experienced salespeople believe they are more capable, resulting in more risk-seeking behavior ([42]; [56]). For them, the challenge associated with uncertain opportunities can be intrinsically motivating. Because experienced salespeople are more strongly motivated by intrinsic than extrinsic benefits (e.g., [14]), the intrinsic benefits associated with high conversion uncertainty may outweigh the potential loss of extrinsic benefits. By contrast, the lack of capability and resources makes inexperienced salespeople more concerned about potential losses of both intrinsic and extrinsic benefits at high levels of opportunity conversion uncertainty ([42]). Therefore, the downward shifting effect created by opportunity conversion uncertainty on the benefit function is stronger among inexperienced salespeople than experienced ones.
In terms of the cost function, the abundance of aforementioned resources makes experienced salespeople less concerned about COR when pursuing opportunities with high conversion uncertainty. Conversely, given their lack of resources, inexperienced salespeople are more concerned about conserving their limited resources and are more sensitive to the costs associated with high conversion uncertainty ([26]; [30]). Thus, opportunity conversion uncertainty is likely to create a weaker upward shift of the cost function among experienced than inexperienced salespeople. Taking the benefit and cost effects together, experienced salespeople expect greater net benefits when conversion uncertainty is high. For them, the inverted U-shaped relationship between opportunity magnitude and salesperson performance shifts less strongly to the left ([24]).
- H5: The greater salesperson experience, the weaker is the leftward shifting effect of opportunity conversion uncertainty on the inverted U-shaped relationship between opportunity magnitude and salesperson performance.
In Study 2, we corroborate Study 1's findings and examine the postulated boundary conditions of salespeople's calibration for conversion uncertainty in solution selling at the portfolio level. We collected data from the sales organization of a large firm ($16.4 billion in total revenue per year). The firm, which operates in the B2B market, provides information and technology solutions (e.g., workspace systems, data center solutions, managed services, security) to customers in industries such as finance, government, education, transport, service, retail, and media. Field-based salespeople are grouped according to the industries the firm serves, with each assigned a territory. All the salespeople are subject to the same compensation and incentive scheme and obtain a fixed yearly salary plus commission (with a progressive plan for all sales beyond quota). Using a survey instrument, we collected information about the salespeople's perceptions of their portfolios. Of the 248 salespeople, 211 completed the questionnaire (85% response rate). Consistent with the length of the average sales cycle, we collected objective salesperson performance from the firm's records six months after the survey.
In Study 2, we examine the portfolio magnitude and conversion uncertainty in the aggregate at the portfolio level. Thus, portfolio magnitude corresponds to opportunity magnitude, and portfolio conversion uncertainty corresponds to opportunity conversion uncertainty.
To measure portfolio magnitude, we use the expected customer demand scale from [61]. The scale has four items that cover salespeople's judgment of the opportunity magnitude in terms of order intake, sales volume, revenue, and profits for the solutions in their portfolio. To measure portfolio conversion uncertainty, we developed a new scale that asks salespeople to assess their degree of (un)certainty about the portfolio magnitude. We inversely coded the scores to obtain uncertainty scores. We obtained salesperson past performance success (in the previous quota cycle) and salesperson performance from company databases. We used the percentage of quota achievement, as previous studies indicate that it accurately captures measurable task performance output while accounting for situational factors ([ 2]). Following previous studies (e.g., [ 2]), we operationalize salesperson experience as a composite measure consisting of three separate measures of experience: time in sales territory, time with the company, and time in the sales profession.
We control for nonlinear effects of uncertainty by including a squared term ([20]). Dummy variables account for salespeople's industry. We also account for individual characteristics that may influence their judgments (i.e., age and workload). Finally, we control for salesperson trait competitiveness, measured with a scale from [ 9]. Web Appendix W6 provides measurement scales and descriptives of Study 2.
A confirmatory factor analysis of the measures indicated good model fit ( = 88.108, p < .01; comparative fit index = .950; Tucker–Lewis index = .933; root mean square error of approximation = .074; square root mean residual = .045; [ 5]). The scales achieved sufficient reliability, with composite reliabilities between.77 and.90 and average variances extracted exceeding.50 for all constructs, indicating reliability. The average variance extracted of each construct exceeds the average variance shared with any other construct, providing evidence of discriminant validity. In addition, all factor loadings are significant (p < .01) and have standardized values ranging from.65 to.91, thus demonstrating convergent validity of the constructs. To examine the effects of opportunity magnitude, conversion uncertainty, past performance success, and salesperson experience on salesperson performance, we specified a multilevel model in Mplus 8.3 ([44]) to control for the nesting of the data. We provide the model specification in Web Appendix W7.
The effect of salesperson opportunity magnitude and conversion uncertainty on sales performance may be spurious as a result of omitted variables (e.g., group-level factors) and correlation between independent variables and the error terms. For example, a sales manager's and coworkers' judgments may influence a salesperson's judgments and performance outcomes. To control for possible endogeneity in our analyses, we adopted [21] control function procedure. Web Appendix W7 provides further details.
We present the results of our retests of the main effects of the opportunity magnitude (H1) and calibration (H2) hypotheses for solution selling at the portfolio level. We then report the findings regarding the boundary conditions of past performance and salesperson experience (H4 and H5).
We report the results in Table 3. To retest H1 about the inverted U-shaped effect of opportunity magnitude on salesperson performance, we again rely on the three-step approach. First, in line with our results from Study 1, we find that opportunity magnitude2 has a significant, negative effect (Model 7: ζ2 = −.059, p < .05). Second, we formally test marginal effects. We find that the slope is positive and significant for low values of opportunity magnitude and negative and significant for high values (see Web Appendix W5). Third, we calculate the turning point. The estimated turning point is just above the average of the opportunity magnitude scale (i.e., mean + .36 = 3.55), with an estimated CI well within the data range (95% CIraw score = [3.02, 4.78]). These results confirm the inverted U-shaped relationship between opportunity magnitude and salesperson performance, corroborating H1.
Graph
Table 3. Study 2: Results (Solution-Selling Context).
| Portfolio | Robust Maximum Likelihood Estimates | Hyp. |
|---|
| Model 7 | Model 8 | Model 9 | Model 10 |
|---|
| b | | SD | b | | SD | b | | SD | b | | SD |
|---|
| Magnitude (ζ1) | .042 | | .043 | .040 | | .043 | .057 | | .041 | .098 | ** | .038 | |
| Magnitude2 (ζ2) | −.059 | * | .026 | −.074 | ** | .029 | −.075 | ** | .028 | −.116 | *** | .030 | H1 |
| Uncertainty (ζ3) | −.087 | ** | .037 | −.088 | * | .038 | −.121 | * | .056 | −.098 | | .059 | |
| Past performance success (ζ4) | .040 | | .027 | .042 | | .028 | .037 | | .039 | .034 | | .037 | |
| Salesperson experience (ζ5) | .015 | | .078 | .029 | | .078 | .071 | | .108 | .035 | | .109 | |
| Moderation Effects | | | | | | | | | | | | | |
| Magnitude × Uncertainty (ζ6) | — | | — | −.044 | *a | .025 | −.044 | | .027 | −.066 | * | .028 | H2 |
| Magnitude2 × Uncertainty (ζ7) | — | | — | — | | — | .026 | | .016 | .018 | | .013 | |
| Magnitude × Past perf. success (ζ8) | — | | — | — | | — | −.051 | | .037 | −.026 | | .034 | |
| Uncertainty × Past perf. success (ζ9) | — | | — | — | | — | −.054 | | .037 | −.012 | | .043 | |
| Magnitude2 × Past perf. success (ζ10) | — | | — | — | | — | .007 | | .026 | −.023 | | .032 | |
| Magn. × Uncert. × Past perf. success (ζ11) | — | | — | — | | — | — | | — | −.058 | * | .027 | H4 |
| Magn.2 × Uncert. × Past perf. success (ζ12) | — | | — | — | | — | — | | — | −.022 | | .016 | |
| Magnitude × Salesperson exp. (ζ13) | — | | — | — | | — | −.107 | * | .064 | −.142 | * | .059 | |
| Uncertainty × Salesperson exp. (ζ14) | — | | — | — | | — | .041 | | .039 | .150 | * | .079 | |
| Magnitude2 × Salesperson exp. (ζ15) | — | | — | — | | — | .024 | | .041 | .075 | * | .033 | |
| Magn. × Uncert. × Salesperson exp. (ζ16) | — | | — | — | | — | — | | — | .016 | | .044 | H5 |
| Magn.2 × Uncert. × Salesperson exp. (ζ17) | — | | — | — | | — | — | | — | −.070 | ** | .024 | |
| Controls | | | | | | | | | | | | | |
| Age | .019 | | .049 | .017 | | .048 | .015 | | .051 | .014 | | .056 | |
| Workload | −.123 | *** | .038 | −.119 | *** | .037 | −.113 | ** | .038 | −.101 | ** | .038 | |
| Trait competitiveness | .029 | | .041 | .037 | | .040 | .033 | | .040 | .054 | | .040 | |
| Dummy Govt. & Edu. | −.018 | | .147 | −.020 | | .148 | −.049 | | .123 | −.010 | | .142 | |
| Dummy Industry & Transport | −.074 | | .110 | −.075 | | .112 | −.134 | | .107 | −.102 | | .130 | |
| Dummy Services, Retail & Media | .014 | | .110 | .006 | | .112 | −.031 | | .105 | .044 | | .136 | |
| Uncertainty2 | .054 | ** | .020 | .041 | * | .020 | .035 | | .022 | .027 | | .022 | |
| Constant | 1.826 | *** | .087 | 1.841 | *** | .089 | 1.890 | *** | .079 | 1.876 | *** | .100 | |
| Pseudo-R2 | .161 | .171 | .208 | .250 | |
| Log-likelihood | −800.078 | −798.029 | −793.395 | −788.232 | |
| Log-likelihood χ2 diff. test (d.f.)b | — | 5.07(1)* | 11.33(7) | 344.20(4) ***c | |
| N (salespeople) | 211 | 211 | 211 | 211 | |
- 9 *p < .05.
- 3 **p < .01.
- 4 ***p < .001; unstandardized coefficients.
- 10 a To determine the extent of bias in our estimates ([24]), we also examined an extension of Model 8 in which we added Magnitude2 × Uncertainty (ζ7) to our equation. The results show that the coefficient of this interaction is not statistically different from zero (ζ7 = .018, p > .10) and does not improve model fit (χ21 = .660, p > .10). See also Web Appendix W7.
- 11 b When using the MLR estimator in Mplus, a log-likelihood difference test statistic is calculated using log-likelihoods and scaling correction factors for the null and alternative models.
- 12 c Model fit of Model 10 is also significantly better than that of Model 8 (χ21 = 41.97, p < .001).
- 13 Notes: Magnitude = opportunity magnitude; uncertainty = opportunity conversion uncertainty.
To retest H2, we add ζ6(magnitude × uncertainty)jh to the equation and test its significance. Model 8 in Table 3 shows that the interaction is significant and negative (ζ6 = −.044, p < .05). We estimate the change in the turning point using the same approach we reported in Study 1. The result again confirms H2, as the change in the turning point is negative and significant (b = −.301, p < .05).
To test the moderating role of past performance success and salesperson experience, we extend Model 8's equation and test ζ11 and ζ16. Model 10 in Table 3 shows that ζ11 is negative and significant (ζ11 = −.058, p < .05), in support of H4. By contrast, ζ16 is not significant (ζ16 = .016, p > .10) and thus does not support a shifting effect, as postulated in H5. Instead, we find a significant, negative curvilinear moderating effect of salesperson experience (ζ17 = −.070, p < .01), suggesting a flipping effect.
We performed several additional robustness checks of our results. First, we used the Wilcoxon rank-sum test (p = .270) to compare respondents and nonrespondents. The tests showed no significant differences between respondents and nonrespondents, alleviating concerns about self-selection bias in our sample. Second, results from Ramsey's regression error specification test (RESET) (χ2 = 1.14, p = .285) alleviate concerns about omitted variables. Third, the maximum variance inflation factors is 3.62, well below the threshold value of 10 ([25]), indicating no multicollinearity issues. Fourth, to test heteroskedasticity we conducted the Cameron–Trivedi test (p = .337) and Breusch–Pagan test (p = .269), neither of which was significant, thus alleviating concerns about heteroskedasticity in our results.
The results of Study 2 not only corroborate the key findings of Study 1 in a different context but also reveal the boundary conditions of salesperson calibration. Figure 4, Panel A, reveals that high past performance success triggers COR, whereas low past performance success provokes more risk-seeking behavior. Salespeople with high past performance success perform best under low uncertainty, whereas those with low past performance success do better under high uncertainty (see right-hand and left-hand sides, respectively). Figure 4, Panel B, shows that highly experienced salespeople perform best under the most challenging situations (low/moderate opportunity magnitude; high conversion uncertainty). However, the left-hand side shows that for inexperienced salespeople, high conversion uncertainty dampens quota achievement significantly. These findings suggest that less experienced salespeople tend to conserve resources under high conversion uncertainty, whereas highly experienced salespeople are more willing to bear uncertainty because they have more resources available.
Graph: Figure 4. Study 2: three-way moderating effects (opportunity magnitude × opportunity conversion uncertainty × salesperson characteristics).
Study 3, a scenario-based experiment, has three objectives. First, we replicate the inverted U-shaped effect of opportunity magnitude on sales performance in a controlled setting. Second, we unpack the underlying benefit–cost mechanism of this effect assumed in Studies 1 and 2. Third, we examine the role of resource slack to show the appropriateness of using COR theory.
According to COR theory, salespeople with limited resources are more likely to conserve them than those with abundant resources ([26]; [30]). For the former, small opportunities do not provide significant resources, whereas large opportunities are prohibitively resource straining. Thus, the indirect negative effect of opportunity magnitude on willingness to pursue an opportunity through costs is amplified when salesperson resource slack is limited. For these salespeople, the total indirect effect of opportunity magnitude through benefits and costs follows an inverted U-shape. By contrast, salespeople with high resource slack are motivated to pursue larger opportunities because they have the resources and, by mobilizing them, can gain even more resources ([26]). For these salespeople, the total indirect effect of opportunity magnitude through benefits and costs is convex. Thus,
- H6: Salesperson resource slack buffers the negative effect of costs on salesperson willingness to pursue an opportunity. As a corollary, the total indirect effect of opportunity magnitude on salesperson willingness to pursue an opportunity follows an inverted U-shape only when salesperson resource slack is low.
Given the consistent findings of an inverted U-shape across levels, in Study 3 we focus on the prospect level. We partnered with a prominent market research firm to access a diverse panel of salespeople from various industries. The research firm randomly recruited 216 experienced salespeople (64% 36–45 years of age, 62% male, 53% in the information technology industry) for our between-subjects experiment.
We then randomly assigned them to one of five scenarios. Each scenario informed participants that they were assigned a territory where the typical revenue of a prospect was $50,000. We included this portfolio baseline information to ensure the design matches with Study 1 and real-life selling contexts. They identified a new sales lead (Prospect A) with a specific opportunity magnitude. In line with data from Study 1, we set the opportunity magnitude at five levels: $1,000, $10,000, $50,000, $250,000, and $1,000,000. After participants read the scenario, we assessed their willingness to pursue Prospect A, anticipated costs, and anticipated benefits on a seven-point scale (1 = "strongly disagree," and 7 = "strongly agree") and their resource slack for prospecting activities. We used the natural variation of salespeople's resource slack in their jobs, as previous research shows that resource slack affects people's framing of costs and benefits in decision making ([68]). Post hoc tests indicated that resource slack was not differently distributed between treatment groups (F���= .30; p > .10), thereby providing evidence that the manipulation itself did not affect participants' perceptions of resource slack. We included the manipulation check, attention and realism checks and demographic questions.
Analysis of variance revealed significant between group differences in willingness to pursue the prospect (F( 4, 211) = 2.835, p = .025). Specifically, willingness to pursue is significantly greater (p < .05) in the $50,000 condition (5.50) than in the small ($1,000; 4.75) or large ($1,000,000; 4.96) conditions. Thus, we replicate the inverted U-shaped effect of opportunity magnitude found in Studies 1 and 2. We then specified the path model of the Study 3 panel in Figure 1. We find that the effect of costs on willingness to pursue is contingent on resource slack (b = .242, p < .01). To test the mediating benefit–cost mechanism and the COR effect, we examined the "instantaneous conditional indirect effect" of opportunity magnitude on willingness to pursue, with salesperson resource slack as the moderator ([29]). The results show that when resource slack is low, the total indirect effect of opportunity magnitude through the two mediators is only significant at moderate levels of opportunity magnitude (θopp.mag=2 = .111, p < .05). When resource slack is high, moderate to high levels of opportunity magnitude translate significantly into willingness to pursue (θopp.mag=5 = .202, p < .01). These results lend support to H6 and our contention that, under resource constraints, the inverted U-shaped effect of opportunity magnitude operates through the benefit–cost mechanism. For further details, see Web Appendix W8.
Integrating decision-making and COR theories, we develop and test a framework of salesperson decision making when prospecting in three multimethod studies. Together, the empirical evidence explains why salespeople avoid big-whale sales opportunities.
Our research stems from the idea that salespeople differ from participants in studies that focus on low-effort, constraint-free, and repeatable choices ([36]). First, the potential benefits—extrinsic and/or intrinsic—of salespeople's decisions are consequential rather than trivial. Second, given their resource constraints and the ephemeral nature of sales opportunities, their costs—explicit and/or implicit—are not negligible. Third, their decision-making context abounds with uncertainties. Our findings confirm and provide novel insights into the theoretical importance of these differences for research on salespeople's decision making, especially when prospecting.
We provide strong empirical evidence that in deciding on which opportunities to pursue, salespeople conduct a benefit–cost analysis based on their initial judgment of opportunity magnitude. We show that the relationship between initial judgment of opportunity magnitude and actual conversion follows an inverted U-shape, regardless of selling task (product vs. solution selling) and information level (prospect vs. portfolio). This finding debunks the intuition that salespeople gravitate toward big-whale opportunities, an insight that extends current understanding of salesperson prospecting behavior. Our result also confirms that salespeople's initial judgment of opportunity magnitude exerts a strong impact on their subsequent behavior and performance, even after controlling for transient phases. This finding complements prior research on salesperson intuition ([27]) and on primacy and anchoring effects ([58]).
We also found that solution-selling salespeople take into consideration opportunity conversion uncertainty in their benefit–cost analysis. Due to this calibration, the inverted U-shaped relationship between opportunity magnitude and performance shifts to the left. This shift implies that salespeople are generally more risk-seeking when opportunity magnitude ranges from small to moderate and risk-averse when opportunity magnitude is large. The counterfactual analyses we conducted show that the calibration effect reduces misspecification of conversion rates by up to 100%, when compared with the estimates from a compensatory decision strategy in which uncertainty is simply factored in as an extra cost. This finding provides a more nuanced understanding of the differences between salespeople's decision-making strategies (i.e., calibration vs. compensatory) when prospecting. Furthermore, it joins two separate streams of research on salesperson decision making, one focusing on salesperson judgment of demands and the other on uncertainty.
Compared with product selling, solution selling is full of uncertainties (e.g., need, process, outcome; [60]). Although these uncertainties are likely to influence salesperson behavior, salesperson behavior and decision making in solution selling has not received much academic research. We contribute to the literature by showing that salespeople indeed use different decision-making strategies in solution selling versus product selling. Specifically, they rely on a calibration decision-making strategy only in solution selling and only at the portfolio level. At the prospect level, regardless of the selling context, salespeople assess each individual opportunity relative to their portfolio baseline in terms of magnitude and conversion uncertainty using a compensatory strategy in which a large magnitude can make up for high uncertainty (and vice versa).
We find that salesperson past performance success and salesperson experience are important contingencies of salespeople's decision-making process under uncertainty (i.e., calibration). The interaction plots (Figure 4) suggest that salespeople who have achieved past performance success and/or have low experience tend to conserve their resources and become more risk averse when selling solutions. They perform better under low or average than high conversion uncertainty conditions. Experienced salespeople are able to overcome this cautious approach.
Our findings also address the contrast between the COR perspective ([30]) and the risk-seeking perspective based on research on risky choice, such as gambling ([56]). While the latter perspective is not specific to the selling context, the COR argument is uniquely relevant to the personal selling context, as it accounts for the notions that ( 1) salespeople are subject to resource constraints and their efforts are costly and ( 2) salespeople need to conserve resources to avoid stress in the long run ([48]). Therefore, researchers who apply decision-making theories to the context of personal selling will benefit from accounting for the uniqueness of salespeople as decision makers.
Our findings highlight a dual information-processing framework in salesperson decision making when prospecting for solution-selling opportunities ([51]; [58]). Under these conditions, we find that salesperson performance is a function of two processes. At the prospect level, salespeople rely on a simple compensatory model in their decision making, such that a large magnitude can make up for high uncertainty (and vice versa). At the portfolio level, they integrate information about both the magnitude and uncertainty of the prospects they targeted in a more complex calibration model. In this decision-making strategy, conversion uncertainty interacts with opportunity magnitude in driving salesperson portfolio performance. This insight is a meaningful step toward a better understanding of salespeople's prioritization of resources and the importance of considering salesperson characteristics in prospecting. It also sheds first light on potential differences between findings at the prospect level and those at the portfolio level (i.e., a lack of homology) and calls for additional multilevel research of this kind.
Our findings provide both managers and salespeople with several new insights into salesperson decision making when prospecting. We underscore key performance implications of our findings by simulating several what-if analyses using the parameters from our results.
The results from three studies consistently show that, all else being equal, salespeople are likely to gravitate toward medium-sized opportunities, leaving smaller and larger opportunities unattended. Using simulated data from Study 1, we find that a salesperson with a prospect whose magnitude equals their baseline opportunity magnitude of ∼$26,750 will have an 86.5% probability of successful conversion. Yet receiving a new prospect that is 1 SD larger in terms of magnitude (∼$143,250) will decrease the conversion odds by more than 15%. This effect is due to the propensity to conserve resources when there are constraints, as Study 3 further shows that the anticipated costs only affect salespeople's pursuit of large prospects when operating under resource constraints.
Thus, to assuage salespeople's avoidance of big-whale deals, managers can leverage their firms' CRM databases. Specifically, a manager can use historical CRM data to calculate the baseline estimates of opportunity magnitude (and conversion uncertainty) for each salesperson. Then, the manager can use this information to match marketing-generated prospects with a salesperson's portfolio baseline, because a large difference in opportunity magnitude between a new opportunity and the salesperson baseline is demotivating and decreases conversion success. Furthermore, when necessary, managers should alter salespeople's benefit–cost calculus when prospecting. For example, they should provide salespeople who work on relatively large opportunities with extra benefits (both extrinsic and intrinsic) and additional resources (to relax the resource constraints), thereby increasing the likelihood of conversion. In addition, managers could pair salespeople with peers with larger portfolio baselines to create ad hoc sales teams to follow up. Such a temporary arrangement can reduce the costs for the focal salespeople.
Our findings show that salespeople's decision-making strategy differs between solution and product selling. For product selling, salespeople rely on a compensatory decision-making strategy at both prospect and portfolio levels. For solution selling, however, they rely on a calibration decision-making strategy, which is noncompensatory in nature, with conversion uncertainty acting as the calibrator of opportunity magnitude at the portfolio level. This calibration effect underscores the important role of portfolio-level information, in terms of both magnitude and conversion uncertainty, in salesperson decision making for solutions. Thus, sales managers should pay close attention to this important "between-salespeople" difference when reallocating prospects for maximum effect. Continuing with the previous example from the simulated data, a sales manager could intuitively decide to allocate the solution-selling prospect of ∼$143,250 to a salesperson with a portfolio baseline magnitude of about the same size (i.e., ∼$143,250). However, if this salesperson's portfolio baseline conversion uncertainty is 1 SD higher, the probability of closing the deal decreases by 36.2%. To reduce conversion uncertainty, managers can play an active role by, for example, using more behavior-based control to curtail salespeople's pursuit of highly uncertain opportunities and providing them with more frequent feedback. Firms can also leverage advanced sales analytics capabilities to decrease uncertainty in opportunity costs and train salespeople how to use the information in their prospecting decisions.
Our results indicate that past performance success and experience can alter the way salespeople calibrate for conversion uncertainty. Thus, these two variables are important for managers as well as salespeople. In a post hoc analysis, we used Study 2's results to predict salesperson quota achievement under various combinations of levels of salespeople's past performance success and experience (±1 SD as high and low values). Drawing on the results summarized in Table 4, Panel A, we derive the most effective managerial actions for managing salesperson prospecting in Table 4, Panel B. Three insights are worth noting. First, regardless of past performance success, salespeople's quota attainment is the worst when they gravitate toward highly certain but small opportunities. Second, salespeople who performed well in the past are most likely to "hit" quota again when the portfolio opportunity is large and conversion uncertainty is low (96%). Nevertheless, these high performers become average performers when conversion uncertainty and portfolio magnitude are average (ranging from 50% to 79%). Therefore, an effective way to manage high performers' prospecting is to give them a large portfolio but also help them reduce conversion uncertainty. This combination allows them to conserve resources while also maintaining high levels of performance. By contrast, salespeople who performed poorly in the past can achieve a quota attainment as high as 83% when they have a large portfolio and conversion uncertainty is high. The increase in opportunity magnitude is more motivating to these poor performers—they are willing to exert greater efforts without conserving resources to prove themselves. Thus, an effective, but perhaps counterintuitive, way to manage poor performers' prospecting is to give them a larger, more uncertain portfolio to challenge them.
Graph
Table 4. Study 2: Managerial Insights into Boundary Conditions of Salesperson Prospecting in Solution Selling.
| A: Post Hoc Analysis for Study 2 |
|---|
| Moderating Role of Salesperson Past Performance Success (H4) |
|---|
| Low Past Performance Success (−1 SD) | High Past Performance Success (+1 SD) |
|---|
| Conversion Uncertainty | Conversion Uncertainty |
|---|
| Magnitude | 1 SD Lower | Average | 1 SD Higher | 1 SD Lower | Average | 1 SD Higher |
|---|
| 1 SD lower | 36% | 40% | 48% | 39% | 50% | 64% |
| Average | 68% | 67% | 71% | 84% | 81% | 79% |
| 1 SD higher | 67% | 76% | 83% | 96% | 70% | 50% |
Table 4. Study 2: Managerial Insights into Boundary Conditions of Salesperson Prospecting in Solution Selling.
| Moderating Role of Salesperson Experience (H5) |
|---|
| Low Salesperson Experience (−1 SD) | High Salesperson Experience (+1 SD) |
|---|
| Conversion Uncertainty | Conversion Uncertainty |
|---|
| Magnitude | 1 SD Lower | Average | 1 SD Higher | 1 SD Lower | Average | 1 SD Higher |
|---|
| 1 SD lower | 29% | 28% | 27% | 54% | 76% | 106% |
| Average | 106% | 75% | 53% | 53% | 75% | 106% |
| 1 SD higher | 118% | 84% | 60% | 60% | 62% | 64% |
Table 4. Study 2: Managerial Insights into Boundary Conditions of Salesperson Prospecting in Solution Selling.
| B: Managerial Takeaways |
|---|
| Observation from Data | Suggested Managerial Action |
|---|
| Regardless of past performance success, salespeople's quota attainment is worst when they gravitate toward highly certain but small opportunities. | Point out the importance of "bread-and-butter" prospects, as salespeople may ignore them while such prospects could be of strategic importance. |
| Salespeople with greater past performance success perform relatively well when portfolio opportunity is large and conversion uncertainty is low. | Give these salespeople a large portfolio but also help them reduce conversion uncertainty (to help them conserve resources). |
| Salespeople with lower past performance success tend to perform better for relatively larger and more uncertain portfolios. | Give these salespeople a larger, more uncertain portfolio to challenge them, while also providing opportunities to recuperate from poor past performance. |
| Inexperienced salespeople perform better for relatively larger, but certain portfolios. | Reduce conversion uncertainty (e.g., via information provision, training) and provide them with larger portfolios. |
| Experienced salespeople tend to perform especially well for relatively small to moderate portfolio magnitude with relatively high conversion uncertainty. | Challenge these salespeople with opportunities that have high conversion uncertainty, while ensuring the portfolio itself is not too large. |
14 Notes: 100% = on-target performance. Shaded boxes reflect higher levels of quota achievement.
Third, when conversion uncertainty is reduced, inexperienced salespeople who handle a large portfolio can go from zero to hero, as their quota attainment increases from 60% to 118%. Inexperienced salespeople also achieve low quota (under 30%) when their portfolio is small, regardless of conversion uncertainty. Therefore, an effective way to manage inexperienced salespeople's prospecting is to reduce conversion uncertainty and provide them with ample opportunities. By contrast, experienced salespeople do not perform well when their portfolio opportunity is large, regardless of conversion uncertainty (range: 60%–64%). However, they thrive under high conversion uncertainty and when their portfolio is moderate in size, with a quota attainment exceeding 100%. Thus, an effective way to manage experienced salespeople is to challenge them with opportunities that have high conversion uncertainty, while ensuring the portfolio opportunity itself is not too large.
What do these results mean for salespeople? Our results show that salespeople need to be cognizant of potential biases created by their past performance success and experience. This is because these biases can significantly improve or impair their sales performance, as indicated by the aforementioned potential gains and losses in quota attainment. By changing the benefit–cost analysis and reducing factors that drive conversion uncertainty (e.g., learn from peers, ask managers for support, form ad hoc teams), salespeople can become more effective in closing big-whale deals and hitting their targets despite conversion uncertainty.
While our research covers three empirical contexts and our data came from multiple sources, this article has several limitations. First, although opportunity magnitude and conversion uncertainty are two of the most important factors in salesperson decision making, they are by no means the only factors. In our research, we included several contingencies and control variables to account for heterogeneity. Nevertheless, we urge further research to consider other aspects as contingencies of salesperson calibration, such as salesperson perceptual accuracy in forming judgments of opportunity magnitude and uncertainty, customer characteristics, competition, and the source of the sales leads ([46]; [48]). Future research could also explore how managers can influence salesperson calibration (e.g., through incentives, by changing baseline conversion uncertainty via altering the composition of self-generated and assigned leads). Moreover, the nature of uncertainty itself and how it affects judgment and decision making could be further explored. Second, we focus on portfolio baseline magnitude and conversion uncertainty as the frames of reference for how salespeople form relative comparisons of prospects in their portfolios. Although this focus is both theoretically and empirically justified, further research could examine other reference points as suggested in the judgment–decision making literature (e.g., sales goals, status quo, minimum requirement). Other measures of magnitude, such as customer lifetime value, could also be examined.
Third, we control for several time-related effects in Study 2 but did not examine the dynamics. Although the first impression generally serves as the anchor point, people adjust their anchors as they receive new information ([58]). Further research could examine how salespeople update their judgment of uncertainty over time by exploring how this effect manifests itself in salesperson prospecting. Fourth, we focus on salesperson past performance success and salesperson experience as moderators, but other moderators may exist, such as control systems and training. Finally, although theoretical arguments exist in support of the moderating role of salespeople's past performance success and experience, future research could explicitly test how these contingencies influence the underlying benefit–costs analysis.
sj-pdf-1-jmx-10.1177_00222429211037336 - Supplemental material for Why Salespeople Avoid Big-Whale Sales Opportunities
Supplemental material, sj-pdf-1-jmx-10.1177_00222429211037336 for Why Salespeople Avoid Big-Whale Sales Opportunities by Juan Xu, Michel van der Borgh, Edwin J. Nijssen, and Son K. Lam in Journal of Marketing
Footnotes 1 Christian Homburg
2 The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
3 The author(s) received no financial support for the research, authorship and/or publication of this article.
4 Michel van der Borgh https://orcid.org/0000-0002-6266-2419 Son K. Lam https://orcid.org/0000-0001-9860-3149
References Aguinis Herman , Bakker Rene M.. (2021), " Time Is of the Essence: Improving the Conceptualization and Measurement of Time ," Human Resource Management Review , 31 (2), 100763.
Ahearne Michael , Rapp Adam , Hughes Douglas E. , Jindal Rupinder. (2010), " Managing Sales Force Product Perceptions and Control Systems in the Success of New Product Introductions ," Journal of Marketing Research , 47 (4), 764 – 76.
Antonakis John , Bastardoz Nicolas , Rönkkö Mikko. (2021), " On Ignoring the Random Effects Assumption in Multilevel Models: Review, Critique, and Recommendations ," Organizational Research Methods , 24 (2), 443 – 83.
Asparouhov Tihomir , Muthén Bengt. (2010), " Bayesian Analysis of Latent Variable Models Using Mplus ," (September 29) , https://www.statmodel.com/download/BayesAdvantages18.pdf.
5 Bagozzi Richard P. , Yi Youjae. (2012), " Specification, Evaluation, and Interpretation of Structural Equation Models ," Journal of the Academy of Marketing Science , 40 (1), 8 – 34.
6 Beach Lee Roy , Mitchell Terence R.. (1978), " A Contingency Model for the Selection of Decision Strategies ," Academy of Management Review , 3 (3), 439 – 49.
7 Bowman Douglas , Narayandas Das. (2004), " Linking Customer Management Effort to Customer Profitability in Business Markets ," Journal of Marketing Research , 41 (4), 433 – 47.
8 Brontén George. (2014), " Should B2B Salespeople Prospect ," Membrain (December 30) , https://www.membrain.com/blog/should-b2b-sales-people-prospect.
9 Brown Steven P. , Cron William L. , Slocum John W. Jr.. (1998), " Effects of Trait Competitiveness and Perceived Intraorganizational Competition on Salesperson Goal Setting and Performance ," Journal of Marketing , 62 (4), 88 – 98.
Brown Steven P. , Ganesan Shankar , Challagalla Goutam. (2001), " Self-Efficacy as a Moderator of Information-Seeking Effectiveness ," Journal of Applied Psychology , 86 (5), 1043 – 51.
Casas-Arce Pablo , Martinez-Jerez F. Asis. (2009), " Relative Performance Compensation, Contests, and Dynamic Incentives ," Management Science , 55 (8), 1306 – 20.
Chan David. (1998), " Functional Relations Among Constructs in the Same Content Domain at Different Levels of Analysis: A Typology of Composition Models ," Journal of Applied Psychology , 83 (2), 234 – 46.
Chen Gilad , Bliese Paul D. , Mathieu John E.. (2005), " Conceptual Framework and Statistical Procedures for Delineating and Testing Multilevel Theories of Homology ," Organizational Research Methods , 8 (4), 375 – 409.
Cron William L. , Dubinsky Alan J. , Michaels Ronald E.. (1988), " The Influence of Career Stages on Components of Salesperson Motivation ," Journal of Marketing , 52 (1), 78 – 92.
Deci Edward L. , Ryan Richard M.. (1985), " Cognitive Evaluation Theory ," in Intrinsic Motivation and Self-Determination in Human Behavior. Boston : Springer , 43 – 85.
De Haan Evert , Verhoef Peter C. , Wiesel Thorsten. (2015), " The Predictive Ability of Different Customer Feedback Metrics for Retention ," International Journal of Research in Marketing , 32 (2), 195 – 206.
Eisenkraft Noah. (2013), " Accurate by Way of Aggregation: Should You Trust Your Intuition-Based First Impressions ?" Journal of Experimental Social Psychology , 49 (2), 277 – 79.
Fang Eric , Palmatier Robert W. , Steenkamp Jan-Benedict E.M.. (2008), " Effect of Service Transition Strategies on Firm Value ," Journal of Marketing , 72 (5), 1 – 14.
Frost Aja. (2017), " 4 Techniques for Landing Your Sales Whale ," HubSpot (August 31) , https://blog.hubspot.com/sales/how-to-win-bigger-sales.
Ganzach Yoav. (1997), " Misleading Interaction and Curvilinear Terms ," Psychological Methods , 2 (3), 235 – 47.
Garen John. (1984), " The Returns to Schooling: A Selectivity Bias Approach with a Continuous Choice Variable ," Econometrica , 52 (5), 1199 –1 218.
Germann Frank , Ebbes Peter , Grewal Rajdeep. (2015), " The Chief Marketing Officer Matters ," Journal of Marketing , 79 (3), 1 – 22.
Gigerenzer Gerd , Hoffrage Ulrich , Kleinbölting Heinz. (1991), " Probabilistic Mental Models: A Brunswikian Theory of Confidence ," Psychological Review , 98 (4), 506 – 28.
Haans Richard F.J. , Pieters Constant , He Zi-Lin. (2016), " Thinking About U: Theorizing and Testing U- and Inverted U-Shaped Relationships in Strategy Research ," Strategic Management Journal , 37 (7), 1177 – 95.
Hair Joseph F. , Black William C. , Babin Barry J. , Anderson Rolph E.. (2003), Multivariate Data Analysis. Upper Saddle River, NJ : Pearson Higher Education.
Halbesleben Jonathon R.B. , Neveu Jean-Pierre , Paustian-Underdahl Samantha C. , Westman Mina. (2014), " Getting to the 'COR': Understanding the Role of Resources in Conservation of Resources Theory ," Journal of Management , 40 (5), 1334 – 64.
Hall Zachary R. , Ahearne Michael , Sujan Harish. (2015), " The Importance of Starting Right: The Influence of Accurate Intuition on Performance in Salesperson–Customer Interactions ," Journal of Marketing , 79 (3), 91 – 109.
Harrison J. Richard , March James G.. (1984), " Decision Making and Postdecision Surprises ," Administrative Science Quarterly , 29 (1), 26 – 42.
Hayes Andrew F. , Preacher Kristopher J.. (2010), " Quantifying and Testing Indirect Effects in Simple Mediation Models When the Constituent Paths Are Nonlinear ," Multivariate Behavioral Research , 45 (4), 627 – 60.
Hobfoll Stevan E.. (1989), " Conservation of Resources: A New Attempt at Conceptualizing Stress ," American Psychologist , 44 (3), 513 – 24.
Homburg Christian , Wieseke Jan , Bornemann Torsten. (2009), " Implementing the Marketing Concept at the Employee–Customer Interface: The Role of Customer Need Knowledge ," Journal of Marketing , 73 (4), 64 – 81.
Horstmann Josh. (2017), " Can a Big Deal Save Your Quarter? " Sales Benchmark Index (March 27) , https://salesbenchmarkindex.com/insights/can-a-big-deal-save-your-quarter/.
Jones Steven K. , Jones Kristine Taylor , Frisch Deborah. (1995), " Biases of Probability Assessment: A Comparison of Frequency and Single-Case Judgments ," Organizational Behavior and Human Decision Processes , 61 (2), 109 – 22.
Kivetz Ran , Urminsky Oleg , Zheng Yuhuang. (2006), " The Goal-Gradient Hypothesis Resurrected: Purchase Acceleration, Illusionary Goal Progress, and Customer Retention ," Journal of Marketing Research , 43 (1), 39 – 58.
Kumar Viswanathan , Petersen J. Andrew , Leone Robert P.. (2013), " Defining, Measuring, and Managing Business Reference Value ," Journal of Marketing , 77 (1), 68 – 86.
Lam Son K. , Van der Borgh Michel. (2021), " On Salesperson Judgment and Decision Making ," Journal of the Academy of Marketing Science , 49, 855–63.
Magnotta Sarah , Murtha Brian , Challagalla Goutam. (2020), " The Joint and Multilevel Effects of Training and Incentives From Upstream Manufacturers on Downstream Salespeople's Efforts ," Journal of Marketing Research , 57 (4), 695 – 716.
Marr Jennifer Carson , Thau Stefan. (2014), " Falling from Great (and Not-So-Great) Heights: How Initial Status Position Influences Performance After Status Loss ," Academy of Management Journal , 57 (1), 223 – 48.
Mayberry Robert , Boles James Sanders , Donthu Naveen. (2018), " An Escalation of Commitment Perspective on Allocation–of–Effort Decisions in Professional Selling ," Journal of the Academy of Marketing Science , 46 (5), 879 – 94.
McMullen Jeffery S. , Shepherd Dean A.. (2006), " Entrepreneurial Action and the Role of Uncertainty in the Theory of the Entrepreneur ," Academy of Management Review , 31 (1), 132 – 52.
Misra Sanjog , Coughlan Anne T. , Narasimhan Chakravarthi. (2005), " Salesforce Compensation: An Analytical and Empirical Examination of the Agency Theoretic Approach ," Quantitative Marketing and Economics , 3 (1), 5 – 39.
Mittal Vikas , Ross William T. , Tsiros Michael. (2002), " The Role of Issue Valence and Issue Capability in Determining Effort Investment ," Journal of Marketing Research , 39 (4), 455 – 68.
Mullins Ryan R. , Ahearne Michael , Lam Son K. , Hall Zachary R. , Boichuk Jeffrey P.. (2014), " Know Your Customer: How Salesperson Perceptions of Customer Relationship Quality Form and Influence Account Profitability ," Journal of Marketing , 78 (6), 38 – 58.
Muthén Linda K. , Muthén Bengt. (2017), Mplus User's Guide , 8th ed. Los Angeles : Muthén & Muthén.
Park Sungho , Gupta Sachin. (2012), " Handling Endogenous Regressors by Joint Estimation Using Copulas ," Marketing Science , 31 (4), 567 – 86.
Payne John W. , Bettman James R. , Johnson Eric J.. (1992), " Behavioral Decision Research: A Constructive Processing Perspective ," Annual Review of Psychology , 43 , 87 – 131.
Preacher Kristopher J. , Zhang Zhen , Zyphur Michael J.. (2016), " Multilevel Structural Equation Models for Assessing Moderation Within and Across Levels of Analysis ," Psychological Methods , 21 (2), 189 – 205.
Sabnis Gaurav , Chatterjee Sharmila C. , Grewal Rajdeep , Lilien Gary L.. (2013), " The Sales Lead Black Hole: On Sales Reps' Follow-Up of Marketing Leads ," Journal of Marketing , 77 (1), 52 – 67.
Sande Jon Bingen , Ghosh Mrinal. (2018), " Endogeneity in Survey Research ," International Journal of Research in Marketing , 35 (2), 185 – 204.
Shen Luxi , Fishbach Ayelet , Hsee Christopher K.. (2015), " The Motivating-Uncertainty Effect: Uncertainty Increases Resource Investment in the Process of Reward Pursuit ," Journal of Consumer Research , 41 (5), 1301 – 15.
Sherman Steven J. , Beike Denise R. , Ryalls Kenneth R.. (1999), " Dual-Processing Accounts of Inconsistencies in Responses to General Versus Specific Cases ," in Dual Process Theories in Social Psychology , Chaiken Shelly , Trope Yaakov , eds. New York : Guilford Press , 203 – 27.
Smith Timothy M. , Gopalakrishna Srinath , Chatterjee Rabikar. (2006), " A Three-Stage Model of Integrated Marketing Communications at the Marketing–Sales Interface ," Journal of Marketing Research , 43 (4), 564 – 79.
Sniezek Janet A. , Buckley Timothy. (1995), " Cueing and Cognitive Conflict in Judge-Advisor Decision Making ," Organizational Behavior and Human Decision Processes , 62 (2), 159 – 74.
Stone Daniel F.. (2015), " Clarifying (Opportunity) Costs ," American Economist , 60 (1), 20 –2 5.
Syam Niladri , Hess James D. , Yang Ying. (2016), " Can Sales Uncertainty Increase Firm Profits? " Journal of Marketing Research , 53 (2), 199 – 206.
Thaler Richard H. , Johnson Eric J.. (1990), " Gambling with the House Money and Trying to Break Even: The Effects of Prior Outcomes on Risky Choice ," Management Science , 36 (6), 643 – 60.
Tuli Kapil R. , Kohli Ajay K. , Bharadwaj Sundar G.. (2007), " Rethinking Customer Solutions: From Product Bundles to Relational Processes ," Journal of Marketing , 71 (4), 1 – 17.
Tversky Amos , Kahneman Daniel. (1974), " Judgment Under Uncertainty: Heuristics and Biases ," Science , 185 (4157), 1124 – 31.
Tversky Amos , Kahneman Daniel. (1981), " The Framing of Decisions and the Psychology of Choice ," Science , 211 (4481), 453 – 58.
Ulaga Wolfgang , Kohli Ajay K.. (2018), " The Role of a Solutions Salesperson: Reducing Uncertainty and Fostering Adaptiveness ," Industrial Marketing Management , 69 (2), 161 – 68.
Van der Borgh Michel , de Jong Ad , Nijssen Edwin J.. (2019), " Balancing Frontliners' Customer- and Coworker-Directed Behaviors When Serving Business Customers ," Journal of Service Research , 22 (3), 323 – 44.
Van Heerde Harald J. , Gijsbrechts Els , Pauwels Koen. (2008), " Winners and Losers in a Major Price War ," Journal of Marketing Research , 45 (5), 499 – 518.
Vomberg Arnd , Homburg Christian , Gwinner Olivia. (2020), " Tolerating and Managing Failure: An Organizational Perspective on Customer Reacquisition Management ," Journal of Marketing , 84 (5), 117 – 36.
Vroom Victor H.. (1964), Work and Motivation. New York : John Wiley & Sons.
Wiersema Margarethe F. , Bowen Harry P.. (2009), " The Use of Limited Dependent Variable Techniques in Strategy Research: Issues and Methods ," Strategic Management Journal , 30 (6), 679 – 92.
Wieseke Jan , Homburg Christian , Lee Nick. (2008), " Understanding the Adoption of New Brands Through Salespeople: A Multilevel Framework ," Journal of the Academy of Marketing Science , 36 (2), 278 – 91.
Wu George , Gonzalez Richard. (1996), " Curvature of the Probability Weighting Function ," Management Science , 42 (12), 1676 – 90.
Zauberman Gal , Lynch John G. Jr.. (2005), " Resource Slack and Propensity to Discount Delayed Investments of Time Versus Money ," Journal of Experimental Psychology: General , 134 (1), 23 – 37.
~~~~~~~~
By Juan Xu; Michel van der Borgh; Edwin J. Nijssen and Son K. Lam
Reported by Author; Author; Author; Author
Copyright of Journal of Marketing is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Record: 142- Working It: Managing Professional Brands in Prestigious Posts. By: Parmentier, Marie-Agnès; Fischer, Eileen. Journal of Marketing. Mar2021, Vol. 85 Issue 2, p110-128. 19p. 1 Chart. DOI: 10.1177/0022242920953818.
- Database:
- Business Source Complete